close

Вход

Забыли?

вход по аккаунту

?

Application of Microwave and Ultrasonic Techniques for Defect Detection in Pipes

код для вставкиСкачать
APPLICATION OF MICROWAVE AND ULTRASONIC TECHNIQUES
FOR DEFECT DETECTION IN PIPES
A Dissertation Submitted
to the Graduate School
University of Arkansas at Little Rock
in partial fulfillment of requirements
for the degree of
DOCTOR OF PHILOSOPHY
in Engineering Science and Systems
Mechanical and Materials Engineering
in the Department of Systems Engineering
of the Donaghey College of Engineering and Information
Technology
May 2017
Wissam M. Alobaidi
M.Sc. of Systems Engineering, University of Arkansas at Little Rock, 2013
Higher Diploma in Refrigeration and Air-Conditioning Engineering Technology, Middle
Technical University, 2006
B.Sc. of Mechanical Engineering, University of Technology, Iraq, 2002
ProQuest Number: 10636110
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
ProQuest 10636110
Published by ProQuest LLC (2018 ). Copyright of the Dissertation is held by the Author.
All rights reserved.
This work is protected against unauthorized copying under Title 17, United States Code
Microform Edition © ProQuest LLC.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
© Copyright by
Wissam M. Alobaidi
2017
This dissertation, “Application of Microwave and Ultrasonic Techniques for Defect
Detection in Pipes,” by Wissam M. Alobaidi, is approved by:
Dissertation Advisor:
Eric Sandgren
Professor of System Engineering
Dissertation Committee:
Alexandru S. Biris
Professor of System Engineering
Mariofanna Milanova
Professor of Computer Science
Srikanth Pidugu
Professor of Engineering Technology
Haitham Q. Nash
Professor and Senior System Engineering
Member, FMC Technologies
Program Coordinator:
Jing Zhang
Associate Professor of System Engineering
Interim Graduate Dean:
Abhijit Bhattacharyya
Professor of System Engineering
Fair Use
This dissertation is protected by the Copyright Laws of the United States (Public Law 94553, revised in 1976). Consistent with fair use as defined in the Copyright Laws, brief
quotations from this material are allowed with proper acknowledgement. Use of this material
for financial gain without the author’s express permission is not allowed.
Duplication
I authorize the Head of Interlibrary Loan or the Head of Archives at the Ottenheimer Library
at the University of Arkansas at Little Rock to arrange for duplication of this dissertation for
educational or scholarly purposes when so requested by a library user. The duplication will
be at the user’s expense.
APPLICATION OF MICROWAVE AND ULTRASONIC TECHNIQUES
FOR DEFECT DETECTION IN PIPES by Wissam M. Alobaidi, May 2017
ABSTRACT
Microwaves promulgate through a waveguide, reflecting off its inner surface, of course. It is
known that any irregularity (discontinuity) in the surface will alter the course of the reflected
signal, and that the reflected signal can be recaptured and processed algorithmically to
generate a waveform characteristic of the discontinuity. The work of this dissertation is based
in this phenomenon. The aim is to develop methods for detecting and then characterizing
pipe wall thinning (PWT) discontinuities in pipes. The work includes both simulations and
practical experiments. New possibilities were discovered as the work progressed. Prior work
revealed a promising initial setup and as the work proceeded, adaptations and refinements
optimized the results. The initial phases of research were conducted using Computer
Simulation Technologies (CST) software and a standard set of PWT specimens so that the
simulations could be calibrated. The modeled (and later the practical) setup consisted of a
vector network analyzer providing a sweeping frequency range of microwaves from 1.4 GHz
to 2.3 GHz. The frequency range was determined by the inner diameter of the pipe (6 in.;
152.4mm). The pipe length was 1m, with PWT specimens located along its length. All
theoretical pipes had 0.5in (12.7mm) wall thickness. The PWT specimens were set as
percentages of that thickness. The initial simulations were made with only microwave signals
used to detect and characterize the PWT specimens, which yielded inconclusive location
results. Still, the measurement errors were relatively consistent, so a second phase was added
to the protocol, to more accurately locate and characterize the PWT specimens. The more
successful technique is straight-beam ultrasonic (UT) probes guided to a starting position
based on the predictions of the microwave phase. During the practical experimental phase,
straight beam UT was used and could both accurately locate the PWT specimen, and draw
the profile of a rectangular or semicircular cross-section for characterizing the shape of the
PWT. In the process of determining how to apply UT to the system, angle-beam UT was
also tested, and an optimization technique was developed using CST simulations. The
predicted locations of the PWT derived from the microwave phase fell within a half-width
(W/2) ahead of and W/2 beyond the actual location. This led to the development of an
innovative sensor positioning equation to calculate the excitation points (EPns) to use when
positioning the UT probe. In addition, the EPns are passed to a genetic algorithm to fine-tune
the angles used for the UT probes. In practical application, these same manipulations worked
well with straight-beam UT probes.
While developing the protocol, it was realized that pattern recognition using correlation
analysis could be applied to the microwave waveforms in order to characterize
discontinuities. To verify this, 71 different CST simulations were used to develop a database
to which unknown PWT specimens could be compared. This process was successful both in
simulation and experimental identification of unknown specimens. The experimental
verification used only three positions, three profiles and three depths of PWT. A further
advance that grew out of this pattern recognition work was the development of a neural
network that can learn and build the pattern recognition database on its own. In experimental
terms, development of the database would be prohibitively expensive. The 108-sample
training dataset was developed using CST, was then learned by the neural network, which
itself optimized the weights and topology for the network. The network was then able to
predict the characteristics of unknown samples that were introduced to the system, with
acceptable accuracy. This advance is certainly worth further development work.
Dedication
Dedicated to....
My Parents.
My wife “Entidhar Alkuam”.
My daughter "Fatima Alobaidi".
My son "Abbas Alobaidi".
My daughter "Narjis Alobaidi".
My Family.
Also dedicated to My Adviser "Dr. Eric Sandgren".
Best of Luck and Thank You.
Wissam Muzher Alobaidi
May 2017
Acknowledgements
It seemed like this day would never get here! But it has finally. And I want to thank
some people who helped me more than I can express during the process of earning a Ph.D.
Doctor Eric Sandgren not only took me on as a doctoral candidate, he shared his
experience and knowledge with me in all circumstances. My English is not able to express
the many ways in which he helped me, challenged me and defended me during the whole
four-year long Ph.D. process. He never asked to be mentioned by name, but he didn’t hold
back in encouraging the many ideas that came to me as I worked on this project. And during
the whole time he was polite and understanding. There were no problems between us during
the whole time.
I must also thank my wife, Entidhar Alkuam, for her patience and support during
these four years. She was understanding and supportive from coming to the U.S., and
working through the Ph.D. She took care of the kids and the household without complaint,
and all the while she is working on her own Ph.D. project! Although I try to be a good
husband to her, I cannot help but think that she is a much better wife to me than I am
husband to her. And with other who have not treated her as well as they should have, during
her Ph.D study, she simply responds to them with understanding on their level.
My father “Muzher Alobaidi” passed away in 2008, but I want to think that he was
with me throughout the time I worked on this degree, and that his encouragement remains as
strong as it was when he was here with us. My mother, who is disabled, came with us to the
United States, and found it better to return home to Iraq a year ago, but she has helped us
with her prayers all along. Like any mother, she worried about what the hard work was doing
to my health, but when I told her that I had earned the Ph.D. she rejoiced with us.
During the process I did two US Provisional Patent, published 15 papers, and got
three awards. All along, I saw earning a Ph.D. as a practical thing to do, and aimed all my
research toward applications in the real world.
Wissam Muzher Alobaidi
May 2017
Table of Contents
Chapter 1 ................................................................................................................................ 1
A Survey on Benchmark Defects Encountered in the Oil Pipe Industries .............................. 1
1.1 Introduction..................................................................................................................... 2
1.2 Review of Weld Joint Types ............................................................................................ 4
1.3 Defects............................................................................................................................. 6
1.3.1 Porosity .................................................................................................................. 6
1.3.2 Inclusions ............................................................................................................... 7
1.3.3 Lack of fusion (LOF) ............................................................................................... 8
1.3.4 Lack of penetration (LOP) ...................................................................................... 8
1.3.5 Cracks .................................................................................................................... 8
1.3.6 Burn-Through ........................................................................................................ 8
1.3.7 Irregular Shapes..................................................................................................... 9
1.4 Summary of Survey ......................................................................................................... 9
1.5 Discussion ...................................................................................................................... 13
1.6 Conclusion ..................................................................................................................... 14
Chapter 2 .............................................................................................................................. 16
Experimental Application of Microwave Technology for Detecting Discontinuities in Pipes:
A Review ................................................................................................................................. 16
2.1 Introduction................................................................................................................... 16
2.2 Original Experimental Approaches................................................................................ 19
2.3 Wave Guide ................................................................................................................... 20
2.4 Parameters and Materials ............................................................................................. 20
2.4.1 Pipe Wall Thinning ............................................................................................... 21
2.4.2 Pipe Wall Thickening ........................................................................................... 23
ix
2.5 Modeling and Analysis .................................................................................................. 24
2.6 Modeling Results ........................................................................................................... 26
2.7 Discussion ...................................................................................................................... 28
2.8 Conclusions.................................................................................................................... 29
Chapter 3 .............................................................................................................................. 30
Applications of Ultrasonic Techniques in Oil and Gas Pipeline Industries: A Review ......... 30
3.1 Introduction................................................................................................................... 31
3.2 Fundamental Theory of Ultrasonic ............................................................................... 33
3.2.1 Ultrasound ........................................................................................................... 33
3.2.2 Waves .................................................................................................................. 34
3.3 Transducers ................................................................................................................... 37
3.4 Approaches.................................................................................................................... 38
3.4.1 Straight Beam Evaluation .................................................................................... 38
3.4.2 Angle Beam Evaluation........................................................................................ 41
3.5 Summary ....................................................................................................................... 46
3.6 Discussion ...................................................................................................................... 47
3.7 Conclusion ..................................................................................................................... 48
Chapter 4 .............................................................................................................................. 50
Enhancing Production Efficiency of Oil and Natural Gas Pipes Using Microwave Technology
................................................................................................................................................. 50
4.1 Introduction................................................................................................................... 51
4.2 Sequential Stations of Non-Destructive examination in spiral pipe plants .................. 53
4.3 Theoretical Analysis....................................................................................................... 56
4.3.1 Confirming the time inspection of the VI and RT stations for PP case ............... 56
4.3.2 Time and Speed analysis for inspect PP with in VI, MW and RT ......................... 60
4.3.3 Assess the Ratio of work done Through Proposed MW Technology .................. 61
4.4 Results and Discussion .................................................................................................. 62
x
4.5 Conclusion ..................................................................................................................... 64
Chapter 5 .............................................................................................................................. 65
NDT Applied to the Detection of Defects in Oil and Gas Pipes: A Simulation-Based Study 65
5.1 Introduction................................................................................................................... 65
5.2 Materials and Methods ................................................................................................. 68
5.3 Results and Discussion .................................................................................................. 69
5.4 Conclusions.................................................................................................................... 78
Chapter 6 .............................................................................................................................. 79
Localized Surface Plasmon-like Resonance Generated by Microwave Electromagnetic
Waves in Pipe Defects ............................................................................................................ 79
6.1 Introduction................................................................................................................... 79
6.2 Model Setup and Analysis ............................................................................................. 81
6.3 Results ........................................................................................................................... 84
6.4 Discussion ...................................................................................................................... 89
6.5 Conclusion ..................................................................................................................... 90
Chapter 7 .............................................................................................................................. 91
Detection of Defects in Spiral/Helical Pipes Using RF Technology....................................... 91
7.1 Introduction................................................................................................................... 91
7.2 Modeling and Numerical Simulation............................................................................. 94
7.2.1 Reference Waveguide ......................................................................................... 95
7.2.2 Modeled Smooth and Spiral Weld pipes with PWT ............................................ 97
7.3 Results and Discussion ................................................................................................ 101
7.3.1 Smooth Pipe and Spiral Pipe with PWT ............................................................. 101
7.3.2 Effect of Increasing Number of Helical Coils on Waveform .............................. 103
7.4 Interpretation and Conclusion .................................................................................... 105
xi
Chapter 8 ............................................................................................................................ 108
High-Efficiency Remote Measurement of Pipe Defect Using RF/UT Technologies: A
Theoretical Analysis Part One—Straight Beam UT ............................................................. 108
8.1 Introduction................................................................................................................. 108
8.2 Basic Principles of RF and UT Technologies ................................................................ 112
8.2.1 Electromagnetic Waves ..................................................................................... 112
8.2.2 Ultrasonic waves ............................................................................................... 112
8.3 Wave Propagation ....................................................................................................... 112
8.3.1 Microwaves ....................................................................................................... 113
8.3.2 Longitudinal waves ............................................................................................ 113
8.4 Theoretical Approaches .............................................................................................. 113
8.4.1 Microwave Approach ........................................................................................ 114
8.4.2 Ultrasonic Method (Straight Beam Approach) .................................................. 118
8.5 Analysis Results and Discussions ................................................................................. 122
8.5.1 Analysis of Microwave Signals in Time Domain ................................................ 122
8.5.2 Prediction and Detection of the PWT Location ................................................. 124
8.5.3 Detection of the PWT Depth ............................................................................. 127
8.6 Conclusions.................................................................................................................. 129
Chapter 9 ............................................................................................................................ 131
Two-stage Technical Protocol for Locating, Validating, and Characterizing Manufacturing
Defects Remotely using Hybrid System RF/UT: Part Two—Angle Beam UT ..................... 131
9.1 Introduction................................................................................................................. 132
9.2 Microwave Approach .................................................................................................. 134
9.3 Ultrasonic Method (Angle Beam Approach) ............................................................... 135
9.3.1 Ultrasonic Angle Beam Analysis ........................................................................ 135
9.3.2 Measurement and Numerical Configurations ................................................... 137
9.4 Hybrid RF/UT System .................................................................................................. 139
9.4.1 Combination System ......................................................................................... 139
9.4.2 Sensor Positioning ............................................................................................. 140
xii
9.4.3 Sensor Positioning, Results and Discussion ....................................................... 141
9.5 Modeling of the Ultrasonic Technique........................................................................ 146
9.6 Optimization of Sensor Location ................................................................................. 154
9.7 Genetic Optimization Algorithm ................................................................................. 155
9.8 Optimization Results and Discussion .......................................................................... 157
9.8.1 The Rectangular Defect, all Parameters Fixed, but Angle of Beam Varies ....... 157
9.8.2 Adjustments of Beam Positions and Consideration of Other Defect Profiles .. 161
9.9 Experimental Validation .............................................................................................. 165
9.10 Conclusion ................................................................................................................. 167
Chapter 10 ......................................................................................................................... 169
Experimental Evaluation of Novel Hybrid Microwave/Ultrasonic Technique to Locate and
Characterize Pipe Wall Thinning .......................................................................................... 169
10.1 Introduction............................................................................................................... 169
10.2 Specimens under Test ............................................................................................... 171
10.2.1 Pipe Sections ................................................................................................... 171
10.2.2 Ring Sections, Fabrication ............................................................................... 172
10.2.3 Cap Fabrication ................................................................................................ 173
10.3 Experimental Set-up .................................................................................................. 174
10.4 Theoretical Analysis................................................................................................... 176
10.4.1 Confirming Microwave Technique .................................................................. 176
10.4.2 Confirming Straight Beam UT Techniques ...................................................... 177
10.5 Calibration Conditions ............................................................................................... 178
10.5.1 Calibration of Group Waveform Velocity ........................................................ 178
10.5.2 Calibration of UT Probe ................................................................................... 179
10.6 Results ....................................................................................................................... 180
10.6.1 Microwave Testing Results.............................................................................. 180
10.6.2 Phase Two: UT Ultrasonic Straight Beam Evaluation ...................................... 191
10.7 Discussion of Results ................................................................................................. 194
xiii
10.7.1 Microwave Experimental Detection................................................................ 194
10.7.2 Ultrasonic Experimental Detection ................................................................. 196
10.8 Conclusion ................................................................................................................. 197
Chapter 11 ......................................................................................................................... 198
Classification of the Extent of Wall Thinning in Pipes Based on Simulations in the Time and
Frequency Domain ............................................................................................................... 198
11.1 Introduction............................................................................................................... 198
11.2 Materials and Methods ............................................................................................. 201
11.3 Result and Discussion ................................................................................................ 201
11.4 Conclusion ................................................................................................................. 208
Chapter 12 ......................................................................................................................... 209
Waveform Pattern Recognition Applied To Rapid Detection of Wall-Thinning In Pipes: A
Simulation-Based Case Study .............................................................................................. 209
12.1 Introduction............................................................................................................... 210
12.2 Fundamentals of Microwaves ................................................................................... 212
12.3 Pattern-Recognition Analytical Approaches ............................................................. 214
12.4 Microwave Pattern-Recognition Technique ............................................................. 215
12.4.1 Microwave Analysis ......................................................................................... 215
12.4.2 Correlation Analysis ......................................................................................... 218
12.5 Correlation Analysis Results and Discussion ............................................................. 218
12.5.1 Pattern Recognition to Evaluate PWT Location .............................................. 218
12.5.2 Pattern Recognition to Evaluate PWT Length ................................................. 220
12.5.3 Pattern Recognition to Evaluate PWT Depth .................................................. 223
12.5.4 Pattern Recognition to Evaluate PWT Profile ................................................. 225
12.6 Conclusion ................................................................................................................. 226
xiv
Chapter 13 ......................................................................................................................... 228
Experimental Validation for Waveform Pattern Recognition ............................................ 228
13.1 Methods and Materials ............................................................................................. 228
13.1.1 Specimens........................................................................................................ 228
13.1.2 Experimental Set-up ........................................................................................ 229
13.2 Results ....................................................................................................................... 230
13.3 Discussion and Conclusion ........................................................................................ 233
Chapter 14 ......................................................................................................................... 234
Development of an Optimized Neural Network for the Detection of Pipe Defects Using a
Microwave Signal ................................................................................................................. 234
14.1 Introduction............................................................................................................... 234
14.2 Neural Networks ....................................................................................................... 236
14.3 Development of a Neural Network ........................................................................... 237
14.4 Evolutionary Optimization ........................................................................................ 240
14.5 Microwave Modeling for Training Data .................................................................... 241
14.6 Creating a Neural Network for Evaluating Pipe Defects ........................................... 245
14.7 Optimizing the Neural Network ................................................................................ 248
14.8 Results ....................................................................................................................... 252
14.9 Summary and Conclusions ........................................................................................ 261
References ......................................................................................................................... 263
xv
List of Figures
Chapter 1
Figure 1.1: Schematic of the ultrasonic angle beam probe in use for pipeline ..................................... 2
Figure 1.2: Diagram of five butt weld filling types ................................................................................. 5
Figure 1.3: Diagram of five tee weld filling types................................................................................... 5
Figure 1.4: Diagram of three corner weld filling types .......................................................................... 5
Figure 1.5: Diagram of lap weld filling ................................................................................................... 5
Figure 1.6: Diagram of edge weld filling ................................................................................................ 5
Figure 1.7: Cross section of ideal contour for welding bead for oil and natural gas pipeline ............... 6
Figure 1.8: Cross-sectional illustration of cluster, linear and worm porosity discontinuities in pipeline
weld ........................................................................................................................................................ 7
Figure 1.9: Cross sectional illustration of slag inclusions in pipeline weld ............................................ 7
Figure 1.10: Cross sectional illustration of LOF discontinuity in pipeline weld ..................................... 7
Figure 1.11: Cross sectional illustration of LOP discontinuity in pipeline weld ..................................... 8
Figure 1.12: Cross sectional illustration of crack-type discontinuities in pipeline weld ........................ 8
Figure 1.13: Cross sectional illustration of burn-through discontinuity in pipeline weld ...................... 9
Figure 1.14: Cross sectional illustration of undercut, underfill, high weld and overlap defects in
pipeline welds......................................................................................................................................... 9
Figure 1.15: Relationship between surface and subsurface defects and the amount of labor and time
necessary for detection, by type of defect........................................................................................... 15
Chapter 2
Figure 2.1: This diagram shows the detailed configuration for a self-constructed coaxial line sensor.
With kind permission from the Japan Institute of Metals and Materials (Linsheng Liu et al., 2011).. 19
Figure 2.2: Picture of the experimental dual probe sensors connecting the coaxial line with the
waveguide. With kind permission from the American Society of Mechanical Engineers (Yang Ju,
2007)..................................................................................................................................................... 20
Figure 2.3: Picture of five joints and one cap used in the experiment to present the wall thinning and
the closed terminal. With kind permission from the American Society of Mechanical Engineers (Yang
Ju, 2007) ............................................................................................................................................... 21
xvi
Figure 2.4: Picture shows experimental closed terminal end of waveguide with joint and cap in
place. With kind permission from the American Society of Mechanical Engineers (Yang Ju, 2007) ... 21
Figure 2.5: Illustration of the pipe under test connecting to the dual probe sensors ......................... 21
Figure 2.6: This photo shows the system setup for the second PWT experiment. The enlarged insets
show details of, as labeled: the coaxial line sensor port, and the connectors of the PWT. With kind
permission from Elsevier (Linsheng Liu et al., 2013) ............................................................................ 22
Figure 2.7: Scheme for the pipe under test with two Pipe wall thinning are represented ................. 23
Figure 2.8: This diagram shows the detail of a setup with separate transmitting and receiving
sensors.................................................................................................................................................. 23
Figure 2.9: (a) VNA connected to the pipe under test, closed end condition. (b) Pipe wall thickening
(adhesive tape). (c) Dual probe connecting to the waveguide. With kind permission from Elsevier
(Nasser Saber et al., 2013) ................................................................................................................... 24
Figure 2.10: Modeling of experimental results for Yang Ju (2007). Shown is scattering parameter for
the outer radius of PWT ranging from 8.6mm to 9.4mm .................................................................... 25
Figure 2.11: Modeling of experimental results for Nasser Saber et al (2013). Shown is scattering
parameter for S11 for the 51mm length biofilm section....................................................................... 27
Figure 2.12: Modeling of experimental results for Nasser Saber et al. Scattering parameter for S11 for
the 34 mm length biofilm section ........................................................................................................ 27
Figure 2.13: Modeling of experimental results for Nasser Saber et al. Scattering parameter for S11
for the 17mm length biofilm section.................................................................................................... 28
Chapter 3
Figure 3.1: Graphical depiction of parallel motion response of material particles subjected to
longitudinal ultrasonic waves, showing compression and rarefaction regions ................................... 34
Figure 3.2: Graphical depiction of perpendicular motion response of material particles subjected to
shear ultrasonic waves, showing wavelength ...................................................................................... 35
Figure 3.3: Graphical depiction of limited detection area of Rayleigh waves, showing how they are
confined mostly to the surface of a material ....................................................................................... 36
Figure 3.4: Graphical depiction of ultrasonic Lamb waves (plate waves) showing how they move
through a test object of a certain thickness which is directly related to the wavelength ................... 36
xvii
Figure 3.5: Scheme for the test sample. Two cases are represented. Case 1 has S1 (straight beam
#1). Case 2 has S2 (straight beam #2) and S3 (straight beam #3). This figure represents calculation of
the value DD, using Equation (3.2). See Section 3.6 Discussion .......................................................... 39
Figure 3.6: A test sample with 8 cases of defects showing the reflection of test signals from straight
beam probes......................................................................................................................................... 40
Figure 3.7: The time domain for the 8 cases from Figure 3.6 are graphed for each of the 6 materials
in Table 3.1 ........................................................................................................................................... 41
Figure 3.8: Representation of Skip Distance through Three Legs, ½ Skip, Full Skip and 1½ Skip. D is
the material thickness and βR SP angle ................................................................................................ 42
Figure 3.9: Skip distance plotted against sound path (SP) angle, showing the SP distances achieved
with ½, full and 1½ SKD, for use by the operator in selecting the SP angle to be used during test
depending on the material ................................................................................................................... 43
Figure 3.10: Sound path length plotted against SP angle. This shows the exact length of penetration
of sonic waves according to the material for ½, full and 1½ SKD ........................................................ 44
Figure 3.11: Illustration of the test object and defect depth found by Angle Beam Transducer (defect
found by the 2nd leg) ............................................................................................................................. 45
Figure 3.12: Illustration of the test sample and discontinuity depth as located by angle beam
transducer (defect located by the 1st leg) ............................................................................................ 45
Figure 3.13: Pipe cross section schematic of discontinuity depth discovered on the 2nd leg by Angle
Beam Transducer.................................................................................................................................. 46
Chapter 4
Figure 4.1: Illustration of Oil and Natural Gas Pipe Manufacturing Process ....................................... 54
Figure 4.2: Illustration of a typical setup for microwave propagation for NDT of small diameter pipe
.............................................................................................................................................................. 55
Figure 4.3: Path Overview of NDT Stations, Used in the Evaluation .................................................... 57
Figure 4.4: Evaluated Times for VI station as per the G1 & G2, with pipe length 24384mm, number of
operator 3 and shift work 8hrs ............................................................................................................. 57
Figure 4.5: Inspection Times for RT station as per the G1, with long pipe 24384mm, NOP=1 and
various shift hours ................................................................................................................................ 58
xviii
Figure 4.6: Calibrate velocity as per the G1, with long pipes 15240mm, 18288mm, 21336mm and
24384mm, NOP=1 and different period shift hours ............................................................................ 59
Figure 4.7: Line efficiency result of microwave station as per the G1, with various long pipe, and
NOP=1................................................................................................................................................... 60
Figure 4.8: Maximum LO results after microwave station add it to inspect pipes G1, with different
length pipe and shift hours .................................................................................................................. 61
Chapter 5
Figure 5.1(a): Cross-section illustrates quarter-circumferential PWR in pipe model with single port
and open end........................................................................................................................................ 68
Figure 5.1(b): (Detail) quarter-circumferential PWR dimensions. Depth of defect = 6.35mm.
Thickness of pipe wall = 12.7mm. PWR width = 25.4mm .................................................................... 69
Figure 5.2: Geometry of full-circumferential PWR model ................................................................... 70
Figure 5.3: Magnitude of S11 relative to sweeping frequency for each PWR width in fullcircumferential model .......................................................................................................................... 70
Figure 5.4: Detail of sweeping frequency for the full-circumferential model ..................................... 71
Figure 5.5: Geometry of half-circumferential PWR model .................................................................. 72
Figure 5.6: Magnitude of S11 relative to sweeping frequency for each PWR width in halfcircumferential model .......................................................................................................................... 72
Figure 5.7: Detail of sweeping frequency for the half-circumferential model .................................... 73
Figure 5.8: Geometry of three-quarter-circumferential PWR model .................................................. 73
Figure 5.9: Magnitude of S11 relative to sweeping frequency for each PWR width in three-quartercircumferential model .......................................................................................................................... 74
Figure 5.10: Detail of sweeping frequency for the three-quarter-circumferential model .................. 74
Figure 5.11: Geometry of quarter-circumferential PWR model .......................................................... 75
Figure 5.12: Magnitude of S11 relative to sweeping frequency for each PWR width in quartercircumferential model .......................................................................................................................... 75
Figure 5.13: Detail of sweeping frequency for the quarter-circumferential model ............................ 76
Figure 5.14: Scattering parameter at Port-1 and resonance frequency for each PWR width ............. 76
Figure 5.15: Limitation of PWR width for full-circumferential model per resonance frequency ........ 77
xix
Chapter 6
Figure 6.1: Cross section schematic of eight pipe defects; in each type of defect (Wdefect = 25.4mm)
and (DD = 6.35mm) .............................................................................................................................. 83
Figure 6.2: Oil pipe model setup. Tetrahedral mesh technique was used for finite elements LSP
analysis. (a) 3D model setup, and (b) 2D model setup. Pipe inner diameter is 152.4mm, and outer
diameter is 177.8mm, pipe thickness is 127mm, and pipe length is 762mm, with defect site in the
middle of the pipe, i.e. 381mm from wave inlet port .......................................................................... 84
Figure 6.3: LSP-like and electric field at defect site during resonance frequency for eight different
defect shapes. Circles indicate points evaluated. Also in each case, the electromagnetic wave
disruption pattern is shown ................................................................................................................. 85
Figure 6.4: Electric field for each defect type corresponding to the resonance frequency ................ 86
Figure 6.5: Graphs of resonance frequency for the scattering parameter S11, for four types of defect
.............................................................................................................................................................. 87
Figure 6.6: Scattering parameter S21 with resonance frequency for each of four profiles of defect
.............................................................................................................................................................. 88
Chapter 7
Figure 7.1: Three dimensional cross-section of waveguide (reference pipe) free of defects ............. 95
Figure 7.2: Reference waveforms for microwave reflections in smooth pipe (waveguide). The
scattering parameter (S21) is at 0 dB, and scattering parameter (S11) is at approximately -95 dB. The
level waveforms tell us that there is no defect in the inner surface of the waveguide (pipe) ............ 95
Figure 7.3: 3D model of the spiral welded standard pipe, with the weld shown from ID (a) and OD (b)
.............................................................................................................................................................. 97
Figure 7.4: Reference signals of microwave noise reflections in spiral welded pipe (waveguide). The
S21 is at 0 dB, and S11 is approximately in dB level between -10 dB and -48 dB................................... 97
Figure 7.5: Illustration of the three types of defects (PWT): (a) rectangular, (b) toroidal, (c) one
rounded and one right-angle corner, with dimensions. The defect width (W) is 31.75 mm in all cases
.............................................................................................................................................................. 98
Figure 7.6: Side-view cross-section of model for waveguide containing (a) toroidal PWT in smooth
pipe, and (b) toroidal PWT in spiral-welded pipe................................................................................. 99
xx
Figure 7.7: shows a spiral welded waveguide and rectangular PWT with 6 coils in this helix.
Waveguide material is aluminum......................................................................................................... 99
Figure 7.8: The original helix is shown in yellow, new tighter helix in orange. The number of coils in
the helix is doubled from the original, making 12.............................................................................. 100
Figure 7.9: The new number of coils, 12, are now represented in yellow......................................... 100
Figure 7.10: Doubles the 12 coils (shown in yellow) to 24 shown in orange ..................................... 100
Figure 7.11: Shows 24 coils of the helix clearly, in yellow ................................................................. 101
Figure 7.12: Is the final duplication of coils to 48, shown here in orange, with the previous 24 coils
shown in yellow .................................................................................................................................. 101
Figure 7.13: Is the 48-coil helix shown clearly in yellow .................................................................... 101
Figure 7.15: PRF for rectangular PWT in the spiral welded pipe, showing S11 and S21 ...................... 102
Figure 7.14: Peak resonance frequency (PRF) for rectangular PWT in the smooth waveguide (pipe)
showing S11 and S21 ............................................................................................................................. 102
Figure 7.17: PRF for rounded-right angle profile PWT in the spiral welded pipe, showing S11 and S21
............................................................................................................................................................ 103
Figure 7.16: PRF for rounded-right angle profile PWT in the smooth waveguide (pipe), showing S11
and S21................................................................................................................................................. 103
Figure 7.19: PRF for toroidal profile PWT in the spiral welded pipe, showing S11 and S21 ................. 103
Figure 7.18: PRF for toroidal profile PWT in the smooth waveguide (pipe) showing S11 and S21 ...... 103
Figure 7.21: Shows the resonance frequency for a 12-coil helix, S11 and S21..................................... 104
Figure 7.20: Shows the resonance frequency for a six-coil helix, S11 and S21 .................................... 104
Figure 7.22: Shows the resonance frequency for a 24-coil helix, S11 and S21..................................... 104
Figure 7.23: Shows the resonance frequency for a 48-coil helix, S11 and S21..................................... 104
Figure 7.24: Illustrates PRF relative to the three PWT profiles modeled in this research. Both smooth
pipe and spiral welded pipe PRFs are shown for each profile ........................................................... 105
Figure 7.25: Shows PRF graphed against volume of the helical weld. Here it is easy to see that
increased bead solid volume shifts the resonance peak to a lower frequency ................................. 106
Figure 7.26: Shows PRF plotted against number of coils in the helix. Here it is easy to see that
increased coils forces the resonance peak to a lower frequency ...................................................... 107
xxi
Chapter 8
Figure 8.1: Model for case study for first step (microwave), 2-port configuration, with fullcircumferential rectangular defect profile. Di is inner diameter, DD is depth of defect, D is pipe wall
thickness, W is the width of PWT ....................................................................................................... 110
Figure 8.2: Diagram of ultrasonic processing of piece of pipe section to detect a PWT defect, using a
straight probe ..................................................................................................................................... 111
Figure 8.3: Geometry of PWT rectangular profile model, showing most distant incidence of PWT
located 711.2m from port one ........................................................................................................... 114
Figure 8.4: Cut-away section of a circular pipe showing location of PWT, to demonstrate that such a
pipe will act as a circular waveguide .................................................................................................. 116
Figure 8.5: Longitudinal section of modeled pipe used in simulation study, showing first two and
last two examples of PWT. Ld is the distance between PWTs on center; Lp is the length of the pipe
............................................................................................................................................................ 117
Figure 8.6: Schematic of model with PWT instances and UT straight-beam evaluation. (a) Shows
detectors placed directly above the PWT, measuring DDO. Ro is the external radius of the pipe. Ri is
the internal radius of the pipe. RPWT is the radius of the PWT. (b). Schematic of model with PWT
instances, showing three cases of placement of the UT straight-beam probe. Case 1 shows the probe
at a PWT edge. Case 2 shows the probe between PWT instances, measuring plate thickness. Case 3
shows the probe placed directly above one instance of PWT, and able to accurately measure DDO
............................................................................................................................................................ 120
Figure 8.7: Time domain modeling and analysis results for waveguide: (a) Calibration pipe (reference
waveguide) measured delay time (ΔTC) with open end condition in order to calculate calibration
velocity. (b) Graph of signal propagation in waveguide with a single PWT defect. Location is 254mm
from Port One to the center of PWT. (c) Graph of signal reflection from the same PWT defect
............................................................................................................................................................ 123
Figure 8.8: Evaluations of PWT with microwave signals, to determine LS and LE of each instance in
the model, measured from port 1. Shown are results from four selected cases. (a) PWT centered at
50.8mm, showing that the results for each of the four DDi fall within the width of the PWT. (b) PWT
centered at 203.2mm, showing shifting of the detection points toward the LE, compared to the first
case, so that one measurement falls beyond the actual LE. (c) PWT centered at 254.0mm, showing
detection points shifted nearer to the LS, compared to the first and second cases, but all within the
xxii
width of the PWT. (d) PWT centered at 558.8mm, showing all four detection points shifted beyond
the LE of that PWT instance ................................................................................................................ 127
Figure 8.9: Comparison of DDO determined using straight beam UT in two materials, with the four
DDi modeled. The shift in time delay is due to signal velocity differences in brass 91% compared to
aluminum, rolled ................................................................................................................................ 128
Chapter 9
Figure 9.1: Illustration of the ultrasonic angle beam transducer with representation of two nodes
through skip distance (SKD). D is the depth of material .................................................................... 136
Figure 9.2: Three dimensional of test object with PWT geometry (rectangular profile), shows zig-zag
scanning patterns the edge reflections. W is the width of PWT, DDO is depth to defect, DDi depth of
defect, Leg1 is the 1st leg and Leg2 is the 2nd leg ................................................................................. 136
Figure 9.3: Representation of full SKDArc for arc of the waveguide surface. SP1, SP2 and SP3 are the
sound paths for straight plate. SPC1 and SPC2 are sound paths for circular plate............................... 138
Figures 9.4 (a) and (b): Is a depiction of the combined hybrid microwave-UT defect detection system
described in this paper ....................................................................................................................... 139
Figure 9.5: Shows the predicted location from microwave detection technique and the EPn setup for
the combination system. LE is the ending point of the PWT, LS is the starting point of the PWT, and
(both measured form port 1) as written in equation below .............................................................. 141
Figure 9.6: Shows four depths of PWT possibly detected by the innovative UT sensor positioning
equation with a 45° angle beam. The microwave system uses two ports ......................................... 142
Figure 9.7: Shows five modeled PWT examples as noted in each figure, from the 14 locations
modeled. The UT beam angle is 45° for all cases. Circles indicate the microwave predictions for
distance from port 1 at four modeled depths. The yellow diamonds represent the EPns calculated for
each PWT example for the four depths ............................................................................................. 144
Figure 9.8: Shows the calculated Epns for UT angle beam probes for two instances of PWT, and four
selected depths. Circles represent the predicted locations from the microwave step. Red triangles
and blue squares represent the calculated EPns for each case. (a) PWT instance #13 tested with a 60°
angle beam (b) PWT instance #14 tested with a 70° angle beam ...................................................... 145
xxiii
Figures 9.9 (a) and (b): Global coordinates X and Y are plotted against the cross section of a pipe
wall segment. The segment contains a PWT instance defined by points A, B, C and D. Skip distance
SKD3 is defined by points E, F, G and H. Note that point H intersects the PWT edge at point D ....... 147
Figure 9.10: Intersection of UT angle beam SP and edge of PWT defect in pipe wall ....................... 148
Figure 9.11: Global coordinates X and Y are plotted against the cross section of a pipe wall segment.
The segment contains a PWT instance defined by points A, B, and C. Skip distance (SKD3) is defined
by points D, E, F, and G. Note that point G intersects the PWT edge at point C. Detail of triangle
shape shows the basis of error in the angle beam position ............................................................... 150
Figure 9.12: Global coordinates X and Y are plotted against the cross section of a pipe wall segment.
The PWT instance is defined as a semi-circle. xc and yc are the center coordinate of the defect. xd
and yd represent any point on the semicircular circumference of the defect. YC=0 because we
measure from the inside of the pipe wall .......................................................................................... 152
Figures 9.13(a) (b) and (c): The PWT model is defined as a semi-circle, as shown in Fig. 9.12. (a) The
UT beam fails to intersect with the defect, yielding a zero value, due to the reflected beam missing
the defect entirely. (b) Represents the UT pulse tangent to the circle defect. (c) Represents our
model with two values, one positive and the other negative. We are interested only in the positive
value ................................................................................................................................................... 154
Figures 9.14(a) and (b): The baseline represents UT at 45° angle. Left peak is minimization of total
error found during the simulations. Right peak is minimization of maximum error found in any single
simulation. (a). There is up to 35% error in the predicted depth. (b). There is up to 20% error in the
predicted width .................................................................................................................................. 158
Figures 9.15(a) and (b): Optimum beam angles represented are 32°, 37°, 46° (sensing locations to
right) 49°, 43°, and 39° (sensing locations to the left). Maximum error reduced due to optimization
of beam angles. (a). Maximum error in predicted depth is less than 2.5% (b). Maximum error in
predicted width is about 6%............................................................................................................... 159
Figures 9.16(a) and (b): Optimum beam angles represented are 51°, 46°, 42° (sensing locations to
right) 55°, 49°, 46° (sensing locations to the left). (a). Depth error increased slightly over
minimization. (b). Width error was reduced slightly.......................................................................... 160
Figure 9.17: Represents the error for the rectangular defect, based on the width and depth ......... 161
Figure 9.18: Shows the error for the triangular defect based on the width and depth .................... 162
Figure 9.19: Shows the error for semi-circular profile defects based on width and depth ............... 163
Figure 9.20: Represents the width error for all defect profiles (rectangle, triangle, semi-circle) ..... 163
xxiv
Figure 9.21: Shows depth error for all types of defect (rectangle, triangle, semi-circle) .................. 164
Figure 9.22: Evaluation of the simulation results for the PWT discontinuity locations (Comparison of
Result)................................................................................................................................................. 166
Chapter 10
Figure 10.1: Illustration of 6061-T6 Aluminum Alloy Pipe Used for the Study .................................. 172
Figure 10.2: Design Drawing of the Standard Ring Showing Dimensions and Structural Fitting ....... 173
Figure 10.3: Design drawings of the shapes of PWT Showing Shape and Dimension of the Groove
............................................................................................................................................................ 173
Figure 10.4: Design Drawings for the Cap Fabrication Showing All Dimensions ............................... 174
Figure 10.5: Experimental Set-up for Short-circuited Case Used to Calibrate the Group Velocity of the
Waveguide with Inner Diameter of 154.051 mm, and 1000 mm Length .......................................... 175
Figure 10.6: Experimental Set-up for the End Location for PWT Specimens. Here, the specimen ring
is located at approximately 5/6 the length of the reference pipe from the ellipsoidal cap Length
............................................................................................................................................................ 175
Figure 10.7: (a) shows the calibration of the UT straight beam probe to the 1” block (25.4 mm). (b)
Shows the calibration for the 0.2” block (5.08 mm). (c) Calibration block to cover the range of all
possible measurements in the experiment. (d) The Ultrasonic Couplant bottle ............................... 180
Figure 10.8: Results in Pipe Under Test (PUT) of 1m plus 0.0508m.................................................. 181
Figure 10.9: Experimental result for rectangular shape PWT specimens FR6, 4 & 2. Placed at the
front location of the pipe, with the group velocity method applied to the raw MW reflection data
............................................................................................................................................................ 183
Figure 10.10: Experimental data for the reflection MW signals from the probe and cap for specimens
FR 6, 4 & 2........................................................................................................................................... 183
Figure 10.11: The first three reflections according to the PWT specimens FR6, 4 & 2 peaks, with TOF
ranging from 4.83125 ns to 4.96875 ns ............................................................................................. 184
Figure 10.12: Experimental result for the PWT specimens FC6, 4 & 2, semi-circular shape at the front
location, after the group velocity method is applied to the MW reflection data .............................. 185
Figure 10.13: Reflected experimental MW signal data from the probe and cap for samples FC 6, 4 &
2 .......................................................................................................................................................... 186
xxv
Figure 10.14: Reflection peaks from the PWT samples FC6, 4 & 2, showing TOF ranging from 4.83125
ns to 5.0375 ns ................................................................................................................................... 186
Figure 10.15: Represents the experimental data for PWT samples MR6, 4 &2, with TOF range from
8.3375 ns to 8.475 ns. The waveforms for the three depths of PWT samples is shown very clearly in
circle D-6 ............................................................................................................................................. 187
Figure 10.16: Experimental result showing the disruption peaks in circles D-9 and D-10, caused
because PWT is near end of pipe. The rang of TOF is 13.35625 ns to 13.425 ns ............................... 188
Figure 10.17: Microwave detection location prediction data points are shown falling into sequence
for phase one. Shown are probe positions and relative movement, the UT techniques to be applied
in order to characterize the shape and size of the PWT, in phase two.............................................. 189
Figure 10.18: Microwave experimental detection results for six PWT specimens (rectangular and
semi-circular) located at the middle position in the pipe. The predicted locations are shifting from
the middle of the pipe to this range: 595 mm to 625 mm from the port .......................................... 190
Figure 10.19: Experimental prediction results for phase one. It is clear that there is disruption in the
sequence of predicted locations, due to the PWT samples being located near the end of the pipe
............................................................................................................................................................ 191
Figure 10.20: Experimental results of straight-beam ultrasonic using the dual probe with 15 mm
diam. crystal. This shows the detection of rectangular PWT at three depths: 5.08 mm, 10.16 mm,
and 15.24 mm. With comparison to the DDO (actual known depth to defect, from fabrication) .... 192
Figure 10.21: Experimental results from the same procedure as shown in fig.20. This is for semicircular PWT at the same three depths: 5.08 mm, 10.16 mm, and 15.24 mm. With comparison to the
DDO to five trial tests by dual straight beam UT probe. .................................................................... 192
Figure 10.22: (a) shows the rectangular PWT test 15.24 mm depth correctly, DDO = 2.97 mm. (b)
shows the same rectangular specimen with 10% variation. (c) shows testing of the middle of the
semi-circular PWT. (d) shows the probe placed left of middle on the same specimen ..................... 194
Chapter 11
Figure 11.1: Three types of PWT profiles modeled in CST to investigate the effect of PWT crosssection shapes on the S11 and S21 waveforms. D is wall thickness; DD is depth of PWT. W is width of
PWT .................................................................................................................................................... 202
xxvi
Figure 11.2: Resulting waveforms for three types of PWT profiles modeled in CST showing the S11
waveforms, each of which is measured for the three lengths of PWT (33%, 67% and 100% of
circumference). (a) shows results for the half-circle profile, (b) for the tapered profile, and (c) the
fillet and right corner profile .............................................................................................................. 203
Figure 11.3: With 100% circumference length, constant depth and constant width, shows the shift in
resonance frequency induced by changes in volume of the PWT. (a) S11, and (b) S21. ...................... 204
Figure 11.4: Three types of PWT depths modeled in CST to investigate the effect of PWT depths on
the S11 and S21 microwave signals ...................................................................................................... 205
Figure 11.5: Shows the waveforms generated for three depths for the three profile types. (a) is halfcircle, (b) is tapered, (c) is fillet and right angle ................................................................................. 207
Figure 11.6: shows the waveforms for the 11.43mm depth for each of the three profile types. This
graph demonstrates the property of the resonance frequency peaks for greater volumes of PWT to
fall to the left of the graph (lower GHz frequency). (a) S11, and (b) S21.............................................. 207
Chapter 12
Figure 12.1: Illustration of the orientations of electrical field and magnetic field intensities, with the
velocity of microwave signal propagating inside a waveguide .......................................................... 213
Figure 12.2: Illustration of an electromagnetic (EM) wave exhibiting linear polarization propagating
in empty space, where Z is the direction of travel of the microwave signal; E and H are the electrical
field and magnetic field amplitudes, respectively; λ is the wavelength ............................................ 213
Figure 12.3: Sectional view of modeled waveguide used in simulation case study, showing first two
and last two occurrences of PWT. Ld is the pitch of PWTs on center; Lp is the length of the
waveguide, DD is depth of defect, Do is the outer diameter ............................................................. 217
Figure 12.4: Cross-section of pipe showing standard PWT at positions 3, 4, and 5. In addition, two
unknown defects are shown, with their correlation to the standard PWT instance ......................... 219
Figure 12.5: Location shows resonance frequency peaks (RFP) for three standard PWT instances,
plus the peaks for PWTN2 and PWTN2. The peak position demonstrates that the RFP for each standard
shifts forward and decreases in height with greater distance from port 1 ....................................... 219
Figure 12.6: Cross section of waveguide showing percentage of circumference involved in the PWT
length standard examples (a)-(d). Example (a) shows the 25% circumference length, (b) is 50%, (c) is
75% and (d) is 100% (full-circumferential PWT). In addition, Fig. 12.5 shows example (e), a 62.5%
xxvii
length and its correlation to (b) and (c). Example (f) shows a 90% length, and its correlation to (c)
and (d) ................................................................................................................................................ 221
Figure 12.7: (a) the four standard peaks based on percentage of circumference (volume) of the
standards PWT defect (fixed width and type of profile). (b) the peaks represent the two standard
peaks between which the unknown case resonance frequency peak falls (for 90% circumference).
Note that the peak is located closer to the peak for 100% circumference. (c) the peaks represent the
standard resonance frequency peaks for 50% and 75 % circumference, with the unknown case peak
(62.5% circumference) falling midway between them ...................................................................... 223
Figure 12.8: Section of waveguide with radii of the standard PWT depths (78.74mm to 86.36mm)
shown in a standard width PWT instance (W=25.4mm). Unknown cases in red show two PWT cases
with radii of 82.55mm and 86.995mm. The correlation values for each unknown case are shown in
the discussion ..................................................................................................................................... 223
Figure 12.9: four standard resonance frequency peaks for the standard patterns for PWT radii. The
four standard peaks are shown that the greater radius is to the left. Two additional cases are shown
with their peaks superimposed .......................................................................................................... 224
Figure 12.10: shows the type of defect profiles possible for PWT .................................................... 225
Figure 12.11: profile demonstrates the waveform generated by each of the five profiles tested.
Further work on the variations for profiles remains to be done in the future .................................. 226
Chapter 13
Figure 13.1: Photograph of the experimental setup used for this experiment. The vector network
analyzer with the pipe connected to the cap and coaxial cable. The PWT specimen ring is in place at
the middle location ............................................................................................................................ 228
Figure 13.2: Cross section of ellipsoidal cap created from 6061-T6 aluminum alloy ....................... 229
Figure 13.3: Experimental result for semicircular profile in the middle of the pipe, with depth
varying from 5.08 mm 10.16 mm and 15.24 mm............................................................................... 230
Figure 13.4: Experimental result for semicircular profile, focusing on the sweeping frequency region
2.12 GHz to 2.22 GHz, in order to show that the maximum frequency peak correlates to maximum
volume ................................................................................................................................................ 231
xxviii
Figure 13.5: Experimental results with sweeping frequency 2.00 Ghz to 2.20 GHz, showing the slight
differences in the recognition waveforms based on the volume change as the profile shape changes.
All the PWT is at the same depth (5.08 mm) and location (middle of pipe) ...................................... 231
Figure 13.6: Experimental results focusing on the frequency range 2.30 GHz to 2.12 GHz, comparing
the results of the volume change as the PWT shape changes ........................................................... 232
Figure 13.7: Experimental results focusing on the frequency range 2.14 GHz to 2.2 GHz, showing
that the sequence of frequency peaks (lowest frequency is for highest volume) is not as expected
............................................................................................................................................................ 232
Chapter 14
Figure 14.1: Basic Layout of a Neural Network................................................................................. 238
Figure 14.2(a) and (b): Scheme for model pipe with two ports are represented. LP is pipe length. DD
is depth of full circumferential PWT. Di is inner diameter ................................................................. 243
Figure 14.3: Microwave signals for semi-circular shape, distance 381 mm from port one, depth is
11.43 mm, full circumferential length, with all the resulting parameters ......................................... 244
Figure 14.4: Microwave signals for 2-filet shape, distance 381 mm from port one, depth is 11.43
mm, 66% circumferential length, with all the resulting parameters ................................................. 245
Figure 14.5: Assumed Neural Network Topology ............................................................................. 247
Figure 14.6: Output Distribution for Defect Location for the First Output Level ............................. 250
Figure 14.7: Output Distribution for Defect Location for the Second Output Level ........................ 251
Figure 14.8: Output Distribution for Defect Location for the Third Output Level ............................ 251
Figure 14.9: Output Distribution for Defect Location for all Output Levels ..................................... 252
Figure 14.10: Optimal Neural Network Topology for Defect Depth Detection ................................ 256
Figure 14.11: Optimal Neural Network Topology for Defect Length Detection ............................... 257
Figure 14.12: Optimal Neural Network Topology for Defect Position Detection ............................. 258
Figure 14.13: Optimal Neural Network Topology for Defect Type Detection .................................. 259
Figure 14.14: Output Distribution for Defect Depth for all Output Levels ....................................... 260
Figure 14.15: Output Distribution for Defect Length for all Output Levels ...................................... 260
Figure 14.16: Output Distribution for Defect Type for all Output Levels ......................................... 261
xxix
List of Tables
Chapter 1
Table 1.1: Summary of benchmark defects and NDT detection techniques used in the oil pipe
industry................................................................................................................................................. 10
Chapter 3
Table 3.1: Longitudinal wave velocities shown for 6 different materials. ........................................... 40
Table 3.2: Summary and comparison of the SKD and SP lengths for 6 different beam angles ........... 43
Table 3.3: Selected prior research into uses of ultrasonic NDT to detect discontinuities in pipeline
materials ............................................................................................................................................... 46
Chapter 4
Table 4.1: Summary of the reduced velocity per quality control during different shift periods ......... 63
Chapter 5
Table 5.1: Wall Reduction Formula and Resulting Widths................................................................... 69
Chapter 6
Table 6.1: Maximum electric field at evaluation points as shown in Fig. 6.3 and the resonance
frequencies in each case, arranged from maximum to minimum ....................................................... 86
Chapter 9
Table 9.1: Design variable values at their optimal levels for all of the optimization formulations ... 164
Table 9.2: Parameters of Experimental PWT ..................................................................................... 166
Chapter 10
Table 10.1: Full Dimension and Quantity for Aluminum pipe sections used in the experiment ....... 172
xxx
Table 10.2: Dimensions of the Rings and Wall-thinning Specimens .................................................. 174
Table 10.3: Explanation of identification codes used in experimental graphs (10.9-10.21) ............. 182
Table 10.4: Showing the shape profile; the fabrication depth, which is the distance from the inner
pipe wall to the deepest point of the PWT specimen; the DDO, which is the known measurement
from the outer pipe wall to the highest point on the PWT specimen; and the UT detection trials,
which are the results of the measurements taken by straight-beam UT across the span of the PWT
specimen ............................................................................................................................................ 193
Chapter 14
Table 14.1: This table shows the results from the model with these parameters: the shape is
rectangular, distance 5 in. (127 mm) from port one to the center of the PWT, depth is 0.45 in. (11.43
mm), and the length is 100% circumferential .................................................................................... 244
Table 14.2: All the parameters for the unknown defects designed and modeled to make data entry
to test the neural network ................................................................................................................. 253
Table 14.3: The predicted outputs from the optimized neural networks ......................................... 253
xxxi
List of Acronyms
NDT Non-Destructive Testing
NDE Non- Destructive Testing
LOF Lack of Fusion
LOP
Lack of Penetration
UT
Ultrasonic Testing
RT
Radiography Testing
DXR Digital X-Ray Testing
ET
Electromagnetic Testing
MPT Magnetic Particle Testing
VI
Visual Inspection
ASME American Society of Mechanical Engineers
ASNT American Society for Nondestructive Testing
API
American Petroleum Institute
CST
Computer Simulation Technology
PWT Pipe Wall Thinning
TPW Thickening of the Pipe Wall
FAC Flow Accelerated Corrosion
LDI
Liquid Droplet Impingement
VNA Vector Network Analyzer
LW
Longitudinal Wave
ID
Inner Diameter
OD
Outer Diameter
MW
Microwave
xxxii
SD
Surface Distance
SKD
Skip Distance
SKDArc Full Skip Distance for Arc of the Pipe Surface
SP
Sound Path or Scrap Pipe
SP
Surface Plasmons
SPC1
Sound Path (First Leg Distance for Arc of the Pipe Surface)
PAT
Phased Array Technique
HHT Hilbert-Huang Transform
SW
Shear Waves
RW
Rayleigh waves
LMW Lamb waves
SAM Standard Allowed Minutes
PP
Perfect Pipe
RP
Repair Pipe
PWR Pipe Wall Reduction
COUT Coil Ultrasonic Testing
HS
Hydraulic System
SMUT Seam Ultrasonic Testing
MUT Manual Ultrasonic Testing
D-XR Digital X-Ray
HFW High-frequency Weld
ERW Electrical Resistance Weld
LE
Line Efficiency
TMP Total Minutes Produced
TMA Total Minutes Attended
xxxiii
ETOD Exact Time to Inspect Outside Diameter
ETID Exact Time to Inspect Inside Diameter
TET
Total Exact Time
LO
Line Output
NOP Number of Operators
WH
Working Hours
MWT Microwave Technique
PEC
Perfect Electric Conductor
LSP
Localized Surface Plasmon
FEA Finite Element Analysis
SD
Square Defect
FCD Fillet Corner Defect
FD
Fillet Defect
HCD Half Circle Defect
TCD Triangle Corner Defect
TD
Triangle Defect
ZD
Zigzag Defect
GD
Gear Defect
SERS Surface Enhanced Raman Spectroscopy
PUT
Pipe under Test
TM
Transverse Magnetic Mode
TE
Transverse Electrical Mode
EM
Electromagnetic
RF
Radio Frequency
PRF
Peak Resonance Frequency
xxxiv
RFP
Resonance Frequency Peak
CAM Computer Aided Manufacturing
CNC Computer Numerical Controlled
IFFT Inverse Fast Fourier Transform
TOF Time of Flight
Cij
Correlation Analysis
LSD
Least Square Distance Analysis
CSD
Cosh Spectral Distance Analysis
xxxv
List of Symbols
θ
Incident Angle on the Surface of the Metal (Defects Site)
θR
Angle of Wave Reflection
βR
Reflection Angle
θS
Spread Angle
λg
Wavelength
λ
Wavelength
λLW
Longitudinal Wavelength
r
Permittivity
μ
Permeability
Pnm
Two Mathematical Roots of the First-Kind Bessel Function
n
Electrical Mode of Wave Propagation
m
Magnetic mode of Wave Propagation
Jn(x) Bessel Function Root
TEnm Transverse Electric Mode
TMnm Transverse Magnetic Mode
S11
Scattering Parameter
S21
Scattering Parameter
VL
Velocity of LWs
TT
Transit Time
TD
Time of Transit to and from the Defect

Angular Frequency
p
Plasma Frequency
xxxvi
k
Wave Vector
Ɛ
Dielectric Function
f
Operating Frequency or Resonance Frequency
fTE11
Lowest Cut-off Frequency
fcTEnm Lowest Cutoff Frequency Mode
fc
Cut-off Frequency
fcTMnm Sweeping Frequency at TMnm Mode
fcTEnm Sweeping Frequency at TEnm Mode
Cd
Speed of Light
kLSP
Propagation Constant of LSP
k0
Wavevector (/c)
m
Metal Relative Permittivity

Relaxation Time
εd
Relative Permittivity of the Dielectric Medium
Em
Modulus of Elasticity
μm
Poisson's Ratio
ρ
Material Density
V
Ultrasonic Signal Velocity
VC
Calibration Velocity or Group Velocity
KCnm Cutoff Wave Number
Nb
Length of the Near Zone Beam
ΔTC
Delay Time
ΔTPWT Delay Time of PWT
Si(k) Vector value of the patterns i and j at point k
Sj(k) Vector value of the patterns i and j at point k
xxxvii
Si
Average Value of the Patterns i and j
Sj
Average Value of the Patterns i and j
EPn
Excitation Point
FR
Front Location of PWT (Rectangular Shape) Closest to Probe
FC
Front Location of PWT (Semi-circular Shape) Closest to Probe
MR
Middle Location of PWT (Rectangular Shape) Half Way down Pipe
MC
Middle Location of PWT (Semi-circular Shape) Half Way down Pipe
ER
End location of PWT (Rectangular Shape) farthest from probe
EC
End location of PWT (Semi-circular Shape) farthest from probe
X!
Cofactor Term for Hybrid Automated System
t
Wall Thickness of the Pipe
tb
Thicknesses of the Biofilm
T
Thickness of the Pipe
D
Material Thickness or Pipe Wall Thickness
Di
Inner Waveguide (Pipe) diameter
DD
Depth of Defect (PWT)
DDO
Depth to Defect
DDi
Depth of Defect
d
Diameter
d1
Inner Diameter
d0
Inner Cable Length
DC
Diameter of the Straight-Beam Crystal
DDES Estimated Depth of Defect
DRI
Inner Ring Diameter
DRO
Outer Ring Diameter
xxxviii
R
Radius
Ri
Inner Radius
Ro
External Radius of the Pipe
Ri
Internal Radius of the Pipe
RC
Radius of the Circle
RPWT Radius of the PWT
W
Defect (PWT) Width
Wdefect Width of the PWT
WES
Estimated Width of Defect
L1
Pipe Length
L2
Joint Length
L11
Pipe Segment Length
L12
Pipe Segment Length
L13
Pipe Segment Length
La
Length of the Coaxial Line Sensor
Lb
Lengths of the Biofilm Layer
LP
Total Length of the Modeled Pipe or Reference Pipe
LP1
Half-length of Reference Pipe (LP)
LP2
Approximately Five-Sixths Length of LP
LP3
Approximately One-Sixth Length of LP
LPWT Distance from the Port to the PWT Location (Predicted Position)
Ld
Distance between PWTs on Center
LS
Distance from the Port to Starting Point of PWT Location
LE
Distance from the Port to Ending Point of PWT Location
L(PWT)C Distance from the Port to the PWT Center
xxxix
L1
Segment of the SP 1
L2
Segment of the SP 2
L3
Segment of the SP 3
G1
First Group
G2
Second Group
P1
Pipe Segment Number 1
P2
Pipe Segment Number 2
P3
Pipe Segment Number 3
xl
Preface
This dissertation centers around the interesting phenomenon of microwave reflection
inside a waveguide, and how any discontinuity on the inner surface will cause a measurable
change in the reflected wave pattern. In the process of studying this phenomenon a number of
papers were generated as I homed in on what was to be the subject of the completed
dissertation.
As a preface to the work, I quote from the abstracts for the papers, both published and
unpublished which were completed during the time I worked on this problem. Primarily, I
wish to use the phenomenon of microwave reflection to detect the presence of pipe wall
thinning in pipes at the time of manufacture, or when being tested after being in service for
some period of time. The focus is on the manufacturing end, with an eye toward reducing the
time involved to run nondestructive testing for quality assurance purposes before the pipes
are shipped. Certain diameters of pipes cannot be easily inspected, because they are too small
for a person to enter. I eventually devised a hybrid technique using microwave emissions to
detect and predict the location of a defect, then providing that information to a computermediated system to position ultrasonic probes in order to accurately locate and characterize
the defects.
We have also developed a second technique which is microwave-based, using pattern
recognition to characterize the defects based on sample specimens in calibration pipes. Both
techniques have been found to be useful and functional in real world situations.
The pattern recognition system led to the development of a neural network that is
trained using the results of 108 modeled samples, and has been able to characterize unknown
modeled samples with promising accuracy. Continued work on this neural network should
result in a system that can be verified experimentally in the near future.
Although protocols varied as the work progressed, there are certain aspects of the
work that remained largely the same throughout the studies. All the microwave modeling and
experimental studies were conducted using a vector network analyzer generating a sweeping
frequency range. The sweeping frequencies in all the theoretical and simulation studies vary
between 0.7 GHz to 3.0 GHz. In the experimental study the range was guided by TM01 and
TM11, and ranged from 1.4 GHz to 2.3 GHz.
The theoretical and simulated pipe designs were 6 in (152.4mm) inner diameter, with
1 m length. The pipe wall thinning specimens (PWT) were consistently 1 in (25.4mm) in
width. The wall thickness of theoretical pipes was 0.5 in (12.7 mm) which allowed us to
model PWT depths from 0.25 in (6.35 mm) to 0.45 in (11.43 mm). The experimental pipe
wall thickness was 7.1755 mm. The PWT specimens were machined into moveable rings to
xli
present the PWT at different positions and depths. Each ring has a larger outer diameter,
0.7175 in (18.2245 mm), but the same inner diameter as the pipe.
In most studies the lengths of PWT around the pipe were one-quarter, one-half, threequarters and full-circumferential. Some research for building the database used 33%, 66%
and 100% circumferential lengths, which were then used to train the neural network in order
to characterize unknown PWT specimens where we used 30%, 40%, 70% or 90%
circumferential lengths.
The theoretical pipes were modeled using Computer Simulation Technology (CST)
software, with the exception of the localized surface plasmon paper.
During the course of research the possibility of using a localized surface plasmon-like
effect to characterize PWT led to a paper based on modeling with COMSOL Multiphysics TM
software. This was selected because of its specific characteristics that CST did not have. The
paper itself was based on recent research that showed strongly localized electron activity
could be stimulated by scoring a surface with a particular sub-wavelength sized groove.
When we introduced that to our COMSOL models we were able to recognize PWT examples
theoretically, and that led to a publishable paper.
This research led to the design of an experiment based on the CST modeling results.
We made three 1 meter aluminum pipes, and a set of machined PWT specimens that can be
moved from place to place. We used three depths of PWT. Two of the pipes were cut into
two sections, one at the midpoint, one at 1/6th its length. Thus, three possible locations were
set up. A special cap was machined to hold the coaxial probe from the vector network
analyzer.
After calibration of the reference pipe, the tests with the various combinations of
PWT specimens were undertaken. Our experimental results with microwaves were as
accurate in the 6 inch (152.4 mm) pipe as earlier researchers had gotten with 17mm pipe. The
dual straight beam ultrasonic probe was able to further characterize the known PWT
specimens. The success of the results was actually better than I hoped for. There is more
work to be done, so that the merger of the microwave Phase One and UT Phase Two into a
single, computer-mediated system can be achieved in the future.
xlii
Chapter one has been adopted from: Alobaidi, W. et al. A Survey on Benchmark Defects
Encountered in the Oil Pipe Industries. International Journal of Scientific &
Engineering Research, 844-853, (2015).
Chapter three has been adopted from: Alobaidi, W. M. et al. Applications of Ultrasonic
Techniques in Oil and Gas Pipeline Industries: A Review. American Journal of
Operations Research, 844-853, doi: 10.4236/ajor.2015.54021 (2015).
Chapter four has been adopted from: Alobaidi, W. M. et al. Enhancing Production Efficiency
of Oil and Natural Gas Pipes Using Microwave Technology. Energy and Power
Engineering, 440-450, doi: 10.4236/epe.2015.710043 (2015).
Chapter five has been adopted from: Alobaidi, W. M. et al. NDT Applied to the Detection of
Defects in Oil and Gas Pipes: A Simulation-Based Study. ASME 2015 International
Mechanical Engineering Congress and Exposition. American Society of Mechanical
Engineers, pp. V02BT02A008, doi: 10.1115/IMECE2015-50641 (2015).
Chapter six has been adopted from: Alobaidi, W. M. et al. Localised surface plasmon-like
resonance generated by microwave electromagnetic waves in pipe defects.
Nondestructive Testing and Evaluation, 109-118,
doi: 10.1080/10589759.2017.1311331 (2018).
Chapter seven has been adopted from: Alobaidi, W., & Sandgren, E. Detection of Defects in
Spiral/Helical Pipes Using RF Technology. ptc 2016 11th Pipeline Technology
Conference, pp. 22-33, (2016).
xliii
Chapter eight has been adopted from: Alobaidi, W. M., & Sandgren, E. High-Efficiency
Remote Measurement of Pipe Defect Using RF/UT Technologies: A Theoretical
Analysis Part One—Straight Beam UT. ASME 2016 Pressure Vessels and Piping
Conference. American Society of Mechanical Engineers, pp. V005T10A006,
doi: 10.1115/PVP2016-63624 (2016).
Chapter ten has been adopted from: Alobaidi, W. M. et al. Experimental Evaluation of Novel
Hybrid Microwave/Ultrasonic Technique to Locate and Characterize Pipe Wall
Thinning. Transactions of the ASME. Journal of Pressure Vessel Technology,
pp. 011501, doi: 10.1115/1.4038517 (2018).
Chapter eleven has been adopted from: Alobaidi, W. M., & Sandgren, E. Classification of the
Extent of Wall Thinning in Pipes Based on Simulations in the Time and Frequency
Domain. ASME 2016 Pressure Vessels and Piping Conference. American Society of
Mechanical Engineers, pp. V005T10A005, doi: 10.1115/PVP2016-63387 (2016).
Chapter twelve has been adopted from: Alobaidi, W. M. et al. Waveform Pattern
Recognition Applied to Rapid Detection of Wall-Thinning in Pipes: A SimulationBased Case Study. ASME 2016 11th International Pipeline Conference. American
Society of Mechanical Engineers, pp. V003T04A038, doi: 10.1115/IPC2016-64320
(2016).
Chapter fourteen has been adopted from: Alobaidi, W. M. et al. Development of an
Optimized Neural Network for the Detection of Pipe Defects Using a Microwave
Signal. Transactions of the ASME. Journal of Pressure Vessel Technology, (2018).
xliv
1
Chapter 1
A Survey on Benchmark Defects Encountered in the Oil Pipe Industries [1]
Abstract
Oil and natural gas have been transported by pipeline for over a century, yielding a large
amount of information about defects in manufacture and in service. Research has moved
toward early detection of defects in the body and welds of pipe during the manufacturing
process. The most common defects occur in the welds, and can be categorized into 7 basic
types: porosities, slag inclusions, lack of fusion, lack of penetration, cracks, burn-through and
irregular shapes. Any of these may occur in the five most common welding configurations
used in manufacturing. The five common joint types are: butt-weld joint, tee-weld joint,
corner-weld joint, lap-weld joint and edge-weld joint. The purpose is to aid in the elimination
of problems in the manufacturing process that lead to manufacturing defects, thus enhancing
product quality. The relationship between defect type and the Non-destructive testing (NDT)
methods which best detect each type, are summarized in tabular form. The table also relates
the location of the defect (whether surface or subsurface, or both) to the NDT techniques.
Illustrations of each type of defect are presented for reference. The relationship of defect type
and location to cost and labor needed to detect each type, is presented in graphic form. The
surface defects are easily detected with Visual Inspection, while subsurface defects can be
caught with Radiographic Testing when conducted at the recommended speed of 50mm/s,
but if they are missed they can be detected with Ultrasonic Testing, which is more laborintensive, and which must be verified with a second NDT technique, Digital X-ray Testing.
To enhance production efficiency, and the series of NDT stations needed for oil and natural
gas pipeline manufacturing, we must determine how to incorporate new techniques to cover
the shortcomings of present methods of detecting defects. This will reduce labor time and
increase throughput while maintaining the quality of the finished product.
2
1.1 Introduction
The most effective system for transferring oil and gas for great distances is the
pipeline. The existing pipeline infrastructure is aging, which causes a great deal of concern
over the remaining useful life of current pipelines. In order to protect both the environment
and the population of areas served by these pipelines, it has become increasingly important to
devise ways to monitor the integrity of these structures. Operators are concerned about
monitoring leaks, and detecting potential faults in older sections of the pipeline system [2].
Oil and natural gas have been transported by pipeline since the early 20th century. With over
a century of experience with the technology, the industry has kept careful records of bad
outcomes due to weld and material failures. In addition to endangering public safety and the
environment, pipeline failures can lead to personal injury. Records show that there were four
incidents of injury due to failure of pipeline material or welds in the years between 1996 and
2003 [3].
Non-destructive evaluation (NDE) for locating and identifying discontinuities in the
pipes themselves during both manufacturing and while in service, represents an effective
methodology for insuring the integrity of the current network [4] [5] [6] [7] [8] [9] [10] [11]
[12] [13] [14] [15] [16] [17]. Ultrasonic, angle beam probe an example for exams the section
of pipe body as illustrated in Fig. 1.1.
Figure 1.1: Schematic of the ultrasonic angle beam probe in use for pipeline
3
Where θR is the angle of wave reflection, T is the thickness of the pipe, 1st leg, 2nd leg, and
3rd leg show the configuration of the sound paths. Note that flaws in the pipe body are
usually laminations or inclusions [18], as shown in Fig. 1.1.
Over the century since oil and natural gas pipeline was first introduced there has been
great improvement in the methods used for manufacture and welding. Today’s steel contains
fewer impure inclusions, and welding is a more exact process [3] [19]. Pipe is tested under
pressure by the manufacturer before it leaves the factory. New welds are assessed with NDE
to be certain that they are discontinuity-free. Well-established and tested standards are
necessary during the manufacture of pipe in order to insure the continued integrity of the
pipeline network into the future [19]. Government bureaus, stakeholders, and landowners
demand that the installed pipelines are as free of defects as realistically possible [2] [3] [19].
Pipe sections must be transported from the factory to the field. Despite the care taken during
manufacture and installation, repeated flexing of pipe sections during transport can
sometimes cause cracks in the wall of the pipe. This transit fatigue cracking is normally
revealed when the installed pipeline is pressure tested before being put into service. Still,
some transit fatigue goes undetected by this pre-service testing, and can worsen while the
pipe is repeatedly pressurized during normal operation. Eventually, these undiscovered
discontinuities can lead to failure [3]. Due to the nature of materials, some pipe sections will
develop larger discontinuities during their service life. Such a section must be individually
tested for structural integrity, and an assessment must be made concerning its continued
fitness for service. A growing arsenal of testing methods has been developed over the past
several decades for accurately determining the magnitude of such discontinuities [19].
Recently, research has moved toward early detection of defects in the body and
welded areas of pipeline during manufacture. Much of the research is focused on detecting
reduction defects in pipe wall thickness, and coordinating this detection with existing NDE
technologies.
The lack of information about other types of defects which can occur during the
manufacture of oil and natural gas pipelines has motivated us to survey the range of surface
and subsurface discontinuities, and to present the information in one place. Most of the
4
information presented here relied on technical websites as the source. Some of the
information comes from the author’s personal experience working in pipe manufacture.
With all this information assembled into one paper it is easier for researchers to take
advantage of the untapped areas of study, and to find solutions to the various problems that
are available in this field. Moreover, in this paper, we present a thorough overview of the
types of weld joint discontinuities and their positions relative to the weld. The paper presents
the full range of weld and body discontinuities that may occur during manufacture both
before forming of the pipe, and after forming. We also discus discontinuities that may form
within the weld itself.
Our main purpose is to highlight all the potential defects that may occur during
manufacturing, and give a wide space for research into ways to reduce these defects through
the further development of existing NDE technologies available to the pipe manufacturing
plant.
1.2 Review of Weld Joint Types
Welding is the practice of joining components by using melting of a filler material to
fuse the constituent parts together [20]. There are five basic types of joints used in welding:
First, the butt weld joint. This is the way pipes are made. The pieces are of equal
thickness, and joined at the edge by V-shaped double and single filling, U-shaped double and
single filling, and square filled joints, as shown in Fig. 1.2, [21] [22].
Second, the tee weld joint. Tee joints are made with pieces intersecting at a 90°
angle, with the joint occurring in the middle of one part of the structure. The joints are bevelshaped, single or double filling, J-shaped single or double filling, and square filled joints, as
shown in Fig. 1.3, [21] [22].
Third, the corner weld joint. Corners are formed using the corner weld, another joint
whose components are joined at a 90° angle. The weld is located on the outside edge, and
may use V-shaped single or double filling, U-shaped single or double filling, or square filled
5
Figure 1.3: Diagram of five tee weld
filling types
Figure 1.2: Diagram of five butt
weld filling types
Figure 1.5: Diagram of lap weld
Figure 1.4: Diagram of three corner weld
filling
filling types
Figure 1.6: Diagram of edge weld filling
6
joints, as shown in Fig. 1.4, [21] [22].
Forth, lap weld joint. To create the lap weld, one piece is placed so that one overlaps
the other for a specific distance. They are joined at the parallel interface. The bead can be run
down one side only, or on both sides of the overlap, as shown in Fig. 1.5, [21] [22].
Fifth, edge weld joint. The edge weld is another type of weld where the components
are parallel at the weld. Two sheets are placed together, and the bead is placed along the
aligned edges of the parts. This can be used to join a J-shaped piece to a flat sheet or two flat
sheets, as shown in Fig. 1.6, [21] [22].
1.3 Defects
The steel manufacturing process can leave impurities in the finished steel. These
inclusions can cause the crystalline structure of the steel to be bonded in a weaker state,
resulting in discontinuities. More modern steel-making techniques have greatly reduced the
likelihood of inclusions, creating much higher quality steel. Yet, even the lower incidence of
inclusions can still lead to failure [3]. Fig. 1.7 shows the ideal formation of a welding bead,
in pipe cross-section.
Figure 1.7: Cross section of ideal contour for welding bead for oil and natural gas pipeline
1.3.1 Porosity
Porosity defects come from gas bubbles trapped in the metal filler as it solidifies.
There are many sources of these bubbles in the welding process, but porosity can most often
be avoided if the work pieces are completely cleaned before welding. Porosity can also be
reduced if the welding current is kept below excessive levels. Faster manufacturing speeds
7
are more likely to generate porosity defects [18] [23]. Three common types of porosity
defects are shown in Fig. 1.8.
Figure 1.8: Cross-sectional illustration of cluster, linear and worm porosity
discontinuities in pipeline weld
1.3.2 Inclusions
Non-metallic substances stuck within the weld metal, or between the bead and the
base are called slag inclusions [18] [23], as shown in Fig. 1.9.
Figure 1.9: Cross sectional illustration of slag inclusions in pipeline weld
Figure 1.10: Cross sectional illustration of LOF discontinuity in pipeline weld
8
1.3.3 Lack of fusion (LOF)
Contaminated surfaces can lead to lack of fusion (LOF) defects in welds. The name
explains how the defect occurs, the welding bead fails to adhere to, to fuse with, the base
metal, leaving a weak joint [18] [23], as shown in Fig. 1.10.
1.3.4 Lack of penetration (LOP)
Lack of penetration (LOP) describes the type of defect where the weld metal does not
fully fill the joint. The fill is smaller than it should be, leading to a stress point that can easily
give rise to a crack [18] [23], as shown in Fig. 1.11.
Figure 1.11: Cross sectional illustration of LOP discontinuity in pipeline weld
1.3.5 Cracks
If the weld metal shrinks as it solidifies, cracks may develop within the bead. This
weakens the weld, because the weld metal is no longer continuous [18] [23], as shown in
Fig. 1.12.
Figure 1.12: Cross sectional illustration of crack-type discontinuities in pipeline weld
1.3.6 Burn-Through
A burn-through defect happens due to excessive heat which actually burns the weld
metal, often creating globs of metal on the back side of the weld [18], as shown in Fig. 1.13.
9
Figure 1.13: Cross sectional illustration of burn-through discontinuity in pipeline weld
1.3.7 Irregular Shapes
Any difference from the ideal contour of the weld bead is called an irregular shape
defect. Each type of irregular shape has its particular cause, but they all result in stresssensitive joints that are subject to early failure [18] [23]. The various types of irregular shape
defects are shown in Fig. 1.14.
Figure 1.14: Cross sectional illustration of undercut, underfill, high weld and overlap
defects in pipeline welds
1.4 Summary of Survey
An overview of the characteristics of benchmark defects, and corresponding NDT
detection techniques used in oil pipe manufacturing is shown in Table 1.1 below.
10
Table 1.1: Summary of benchmark defects and NDT detection techniques used in the
oil pipe industry
Pipe Section shape
Defect
Type
Surface
Sub-
NDT
Surface
Techniques
Defects including in the weld
only
UT
Cluster
Porosity
RT
Strongly
Detectable
Detectable
DXR
UT
Linear
Porosity
RT
Strongly
Detectable
Detectable
DXR
UT
Worm
Porosity
RT
Strongly
Detectable
Detectable
DXR
UT
Strongly
Detectable
Slag
RT
DXR
Detectable
11
Lack of
UT
fusion
RT
Strongly
Detectable
Weakly
Detectable
DXR
UT
Lack of
penetrati
RT
on
Strongly
Detectable
Weakly
Detectable
DXR
UT
Crack
RT
Strongly
Detectable
Detectable
DXR
VI
Burn
Through
RT
VI
RT
Strongly
Detectable
Strongly
Detectable
Strongly
Detectable
Strongly
Detectable
12
VI
RT
VI
VI
Strongly
Detectable
Strongly
Detectable
Strongly
Detectable
Strongly
Detectable
The NDT methods used for quality assurance testing in pipe manufacturing.
Ultrasonic Testing (UT), Radiography Testing (RT), Digital X-Ray Testing (DXR),
Electromagnetic Testing (ET), Magnetic Particle Testing (MPT) and Visual Inspection (VI)
are common NDT techniques used in the manufacture of pipes.
Defects considered in the survey are summarized in Table 1.1 above. The purple
color means that the defect may appear on the surface, the yellow means that the defect may
appear sub-surface. Note that most of the defects have both colors in the table, meaning that
they can appear in either position, although the example cross-sectional illustrations show
only one type or the other. The green color indicates the necessity of verifying the detection
of a particular type of defect by UT, with a secondary DXR inspection.
13
1.5 Discussion
Researchers, mechanical and industrial engineers, and those who are responsible for
the design, or who must investigate a problem with pipeline manufacturing processes often
consult websites such as the American Society of Mechanical Engineers (ASME), American
Society for Nondestructive Testing (ASNT), NDT Resource Center, American Petroleum
Institute (API) and other technical sites, when investigating a problem, or referencing a
source. These standard references help them adapt a design to satisfy the customer’s needs, to
learn the solution for a specific problem, or to develop a process. On the other hand, these
organizations have a lot of information about many different pipeline components and
standard processes. This paper gathers information in two important areas: first, the five
most-common welding configurations used in the industry, and second, the most-commonly
found defects in the manufacturing of oil and natural gas pipes. The purpose is to aid in the
elimination of problems in the manufacturing process that lead to manufacturing defects, thus
enhancing product quality.
It should be noted that the Summary of the survey, which appears in Table 1.1, shows
us the wide range of defect types that appear in the welds used during pipe manufacture.
Many of these defects can occur as a surface defect, or a sub-surface one, notably, the
porosities, cracks and inclusions. Conversely, the remaining types of defects, such as
undercut, underfill, overlap, high weld, concavity and burn-through only appear as surface
defects.
Most of the surface defects are caught by VI, but the ability to detect defects in this
way is dependent on the size of the defect; some are too small to see. The smaller defects
may be revealed by RT, but this still commonly depends on the size of the defect. The way of
testing the pipe in RT stations is forward, helical movement with the speed being the same as
production speed, which depends on the customer’s needs, in the author’s experience. The
commonly recommended speed is 50mm/s, which presents a good opportunity for catching
smaller defects with RT. But many times production speed is used, in which case the RT
station will miss smaller defects (for example, due to an eye blink as the defect passes on the
monitor screen).
14
For this reason, the summary Table 1.1 above is based on the recommended speed of
50mm/s [24]. If it is classified as “strongly detectable,” just “detectable,” or just
“verification” in the table, as is the case with DXR, the UT station will need a lot of labor,
which is proportional to the time needed to detect the flaws, and then verify using DXR.
When we design a new way to build pipes, new problems are created in either the
body of the pipe or in the weld. So we need to follow up the new design with research into
possible defects that may arise in the modified manufacturing process, as well as determining
NDT techniques best for detection of these defects, in order to ensure safety and long life for
the products and to keep the environment safe.
1.6 Conclusion
◊
The six most common methods (VI, UT, RT, ET, MPT and DXR) of NDT technology
are extensively applied in the oil pipe industries. Although these techniques are not
profitable to use for small diameter pipes, 254 mm to 762 mm, because they require too
much time and labor for these smaller pipes. Subsurface defects are first detected by UT
and subsequently verified with DXR.
◊
V, U and square shaped welds are used in the oil pipe industries.
◊
Most pipe defects occur as a result of welding operations.
◊
Before the welding operation, the surface of the test object should be totally clean (free
from any substance that prevents fusion in the welding process, like rust and grease, etc).
◊
Burn through defects in the weld if not detected before the hydraulic system test, will
cause significant delay in production.
◊
Burn through, LOP, underfill and undercut defects if not detected during manufacturing,
shorten the life of the pipeline, and they are most likely to cause future leakage in the
pipe while in service. But burn through, if it occurs during pipe installation in the field
causes immediate leakage, if it is not detected.
◊
Multiple cracks will cause weakness in the structure of the weld.
15
Fig. 15 shows the relationship between the defect and its location, whether surface or
subsurface. The line represents the relative cost and labor involved in detecting the defect
type, showing that subsurface discontinuities are much more intensive in both cost and time,
when compared to surface defects. Because the detection of surface defects takes time, it
should be greater than zero on the Y axis, and most surface defects are detected by VI.
Subsurface defects require much more time to detect. They are usually detectable by RT at
the recommended speed; otherwise they must be caught by UT, and verified by DXR.
Figure 1.15: Relationship between surface and subsurface defects and the amount
of labor and time necessary for detection, by type of defect
16
Chapter 2
Experimental Application of Microwave Technology for Detecting
Discontinuities in Pipes: A Review
Abstract
This work reconstructs three experiments using Computer Simulation Technology (CST)
software. The original experimental designs are illustrated, then the modeled examples are
discussed. Two sweeping frequency ranges are incorporated: 20 GHz to 25 GHz, and 17 GHz
to 21 GHz. Two inner diameter pipes are modeled: 17 mm, and 17.03 mm, depending on the
original experimental designs. The pipes modeled are 2-port designs. The wall thickness is
1.0 mm, perfect conductor pipes. The original experimental works seek to detect wall
thinning, or wall thickening due to biofilms. For the pipe wall thinning (PWT) recreation,
additional modeled PWT depths are investigated. For the biofilm recreation, the original
experimental parameters are used.
2.1 Introduction
Metal surfaces exposed to moving fluids are subject to increases and decreases in the
thickness of the wall materials. These changes may be irreversible, depending upon the
system and application in which the pipe functions. Surface defects in pipeline transportation
systems represent one of the realistic problems addressed in Research and Studies [1] [25]
[26] [27] [28] [29] [30].
Pipe wall thinning (PWT) became such a problem in nuclear power plants that it
became necessary for management procedures to be established in many nuclear countries in
order to control PWT [31] [32] [33] [34].
When microorganisms grow on the inside surfaces of industrial pipelines the process
is known as biofouling, and the resultant biological layer is called a biofilm. This biofilm
layer can adversely affect the integrity of the pipes themselves, as well as interfering with
17
desirable properties of the pipes. The presence of the biofilm can lead to corrosion of the
inner surface of the pipe, causing the defect known as PWT [35]. PWT has two major
parameters: one is degree (how wide, how long) and the other is location (where it is within
the pipe) [32] [34]. The biofilm may be an insulating mass which can reduce the heat transfer
flow rate, which is a problem in the steam power, cooling tower and nuclear power plants,
for example [35] [36].
Biofilms are most likely to be produced in the case of water distribution pipelines of
all types. Crude oil being transported through pipelines often carries with it water droplets
entrained in the oil. The microorganisms can use water in either circumstance in order to
adhere to the pipe walls and to grow [35].
The presence of the biofilm leads to what can be called thickening of the pipe wall
(TPW), reducing the inner diameter (throttling) of the passage within the pipe. This, of
course, increases the backpressure in any pumps used in the system. This problem is reversed
when the corrosion of metabolic by-products eats away the inner wall of the pipe, leading to
the aforementioned PWT [35].
The technical rules and performance-based rules for targeting the causes of PWT
group the causes differently. However, the causes listed are the same. They are Flow
Accelerated Corrosion (FAC), and three types of erosion: liquid droplet impingement (LDI),
flashing, and cavitation [33].
FAC occurs when the protective oxide layer inside the pipe begins to dissolve at an
accelerated rate due to the flow of the fluid itself. This is especially a problem wherever
turbulence is elevated (for example, at T-junctions, or in areas just downstream from valves)
[33].
LDI erosion primarily affects pipes carrying steam. The liquid droplets travel at high
velocities, delivering a large amount of force to the pipe wall material, or to the oxide layer,
causing erosion of the pipe wall, resulting in thinning (for example, in pipe elbows) [33].
Cavitation erosion occurs when a change in static pressure results in formation of
bubbles in the fluid, which ultimately collapse. This also generates a large amount of force
18
against the pipe walls, which can lead to thinning of the material. It has been said that both
LDI and cavitation erosion are the result of “the local impulsive force” generated by the
droplets, and collapsing bubbles [33].
When PWT or biofilm growth happen in the field, non-destructive evaluation
techniques (NDE) will be used to enable the detection of such discontinuities. Microwave
NDE is a proposed method that has been evaluated in the laboratory, as well as by use of
numerical simulation modeling previously [37].
Microwave is also proposed as a remote NDT method for solving problems with both
PWT and biofilm accumulation, because most pipes naturally function as a circular
waveguide for microwaves. The propagation of signals along the waveguide of the pipe is
separated from the environment outside the pipe, because all the microwave energy is
contained within the pipe walls [37].
The aim of this research is to merge the experimental results and the abilities of
modeling, to determine and develop the boundary conditions and limitations of PWT and
biofilm parameters that will control the development of real-world techniques to be used in
the field, for detection and measurement of these discontinuities.
This research will take published experimental parameters, replicate these in
Computer Simulation Technology (CST) finite-difference time domain commercial software
models, and compare the real experimental results to the modeling results. This will show
that numerical simulation can exactly reproduce the results from these parameters, or that it
can very closely reproduce them. Ideally, this research will show that modeling can evaluate
the transmission of microwaves and reflection of the waveforms, according to the
discontinuity (PWT or biofilm), with the ultimate aim being to develop the modeling by
adding new boundary conditions and parameters to the simulation. We try to demonstrate
many realistic experimental problems with our models (CST). If shown to be effective, such
a simulation tool can assist in developing improved techniques for quantitatively detecting
and measuring the degree and location of pipeline discontinuities based on realistic
parameters.
19
2.2 Original Experimental Approaches
This study has been made to reconstruct three prior experiments using Computer
Simulation Technologies (CST) modeling. This was done in order to establish the CST
modeling technique as a legitimate method of producing results which would replicate real
world experiments. First, a technique to demonstrate the detection of wall thinning in the
internal pipe surface without exactly locating where this wall thinning is in the pipe [32].
Second, a remote technique for locating an instance of PWT in longer pipes without exactly
measuring the starting and ending points of the defect. PWT defect lengths are known to be
equal to or less than the inner diameter of the pipe in most cases. Third [35], because
biofilms are known to cause an opposite condition, wall thickening, an experiment to monitor
biofilm buildup in pipes was recreated with CST.
All these experimental approaches used a vector network analyzer (VNA) and coaxial
line cables with a sensor for transmitting and receiving the microwave signal. The First and
Third of the studies used two sensors, one of them as an emitter and the second as a receiver;
while the Second study used a single probe that both emits and senses microwave signals, as
shown in Fig. 2.1. In all these studies the sensors were designed by the experimenters. The
returning (reflected) microwave signals were analyzed in order to detect PWT or biofilm
buildup [32] [34] [35].
Figure 2.1: This diagram shows the detailed configuration for a self-constructed coaxial
line sensor. With kind permission from the Japan Institute of Metals and Materials
(Linsheng Liu et al., 2011)
20
2.3 Wave Guide
The microwave signal will propagate in the waveguide (pipe) only at certain
frequencies. For any defect that exists somewhere in the pipe, the depth of this defect
increases the inner diameter of the waveguide. This will cause a decrease in wavelength (λg)
at any of the resonant frequencies. In order to detect the increased diameter, we must apply a
higher frequency signal in order to create resonance. Because we cannot know the exact
resonant frequency of a defect (the presence of the defect is unknown) we sweep through a
range of frequencies in order to detect any possible inner diameter that may occur at some
location within the waveguide. Because of the relationship between resonance frequency and
inner diameter, the amount of reduction can be calculated based on the resonant frequency
[32]. Fig. 2.2 shows the dual probe used in the experiment with separate sending and
receiving sensors.
Figure 2.2: Picture of the experimental dual probe sensors connecting the coaxial line
with the waveguide. With kind permission from the American Society of Mechanical
Engineers (Yang Ju, 2007)
2.4 Parameters and Materials
The three experiments that are reproduced in this modeling research all share a large
number of parameters in common: the diameter of the pipe is very similar. The construction
of the pipe is similar. The VNA as the source of microwaves is common to all three
experiments. The parameters that vary are discussed in the individual discussions of each
experiment below.
21
2.4.1 Pipe Wall Thinning
2007 experiment (Yang Ju) [32]: The parameters of the waveguide used in this
experiment are: inner diameter d1 17mm, wall thickness 1 mm, length L1 900mm, made of
copper. To create thickness reduction, five pipe joints with inner diameters as a fraction of
the wall thickness (10%, 20%, 40%, 60%, and 80%) were created as shown in Fig. 2.3. Each
joint is 17mm in length L2, with inner depth increases of 0.1mm, 0.2 mm, 0.4mm, 0.6mm,
and 0.8mm, respectively. The system is sealed using a cap that closed the terminal of the pipe
(Closed End Condition) as shown in Fig. 2.4. The final length of the test system is 917mm.
Dual probe sensors used in the experiment as shown in Fig. 2.5.
Figure 2.3: Picture of five joints and one cap
Figure 2.4: Picture shows experimental
used in the experiment to present the wall
closed terminal end of waveguide with
thinning and the closed terminal. With kind
joint and cap in place. With kind
permission from the American Society of
permission from the American Society of
Mechanical Engineers (Yang Ju, 2007)
Mechanical Engineers (Yang Ju, 2007)
Figure 2.5: Illustration of the pipe under test connecting to the dual probe sensors
22
2011 experiment (Lenshing Liu, et al) [34]: The parameters of the pipe used in this
experimental paper are the same as those of the Ju experiment from 2007, except the length
of the pipe, because three pieces of pipe are used to create a long waveguide with wall
thinning (PWT) as shown in Fig. 2.6 [34], the pipe segment lengths are L11 453mm, L12
455mm, and L13 2000mm, P1, P2 legitimate and P3, respectively as shown in Fig. 2.7. The
five joints used represent PWT degrees of 5%t, 10%t, 20%t, 40%t, and 60%t, where t is the
wall thickness of the pipe. The length of each joint (L2) is 17.0mm, as aforementioned with
Yang Ju (2007). The tail end of the pipe in this experiment is uncapped (Open End
Condition). In addition, the probe used is developed by the experimenters. The length (La) of
the coaxial line sensor is 52.0mm and inner cable length (d0) is 6.5mm to ensure a strong
microwave signal within the waveguide, as shown in Fig. 2.2, [38].
Figure 2.6: This photo shows the system setup for the second PWT experiment. The
enlarged insets show details of, as labeled: the coaxial line sensor port, and the connectors
of the PWT. With kind permission from Elsevier (Linsheng Liu et al., 2013)
23
Figure 2.7: Scheme for the pipe under test with two Pipe wall thinning are represented
2.4.2 Pipe Wall Thickening
2013 experiment (Nasser Saber et al.) [35]: The pipe parameters used for this
experiment as shown in Fig. 2.8: inner diameter d1 17.03mm, wall thickness t 1mm, length L1
450mm. To simulate the wall thickness increase (biofilm layer), adhesive tape was used. An
example of this tape is shown in Fig. 2.8 The lengths of the biofilm layer Lb are 17mm,
34mm, and 52mm. The thicknesses of the biofilm tb are 0.21mm, 0.42mm and 0.63mm. The
VNA produced microwaves were conducted to the port of the waveguide through a probe
system designed by the experimenters as shown in Fig. 2.9. The end of the waveguide was
capped.
Figure 2.8: This diagram shows the detail of a setup with separate transmitting and
receiving sensors
24
Figure 2.9: (a) VNA connected to the pipe under test, closed end condition. (b) Pipe wall
thickening (adhesive tape). (c) Dual probe connecting to the waveguide. With kind
permission from Elsevier (Nasser Saber et al., 2013)
2.5 Modeling and Analysis
Because the pipe functions as a circular waveguide, of diameter d, with microwaves
at operating frequency f, the wavelength (λg) of the microwave signals disseminating through
the pipe can be expressed in this way [35] [39]:
 =
1
2
√   2 − [1.841]

(2.1)
25
where r and μ represent the permittivity and permeability (considered to be constants) of the
medium (air) in the waveguide. And the numerical value 1.841 represents Pnm as a first kind
Bessel function root, Jn(x) for TE11 wave.
For this model we created two ports with two sweeping frequencies. The first range
was used to confirm the research of Yang Ju, and Linsheng Liu et al.: 20 GHz-25GHz. The
second range was from 17 GHz to 21 GHz to confirm the work of Nasser Saber et al.
Computer Simulation Technology software has been used to recreate experimental
designs. We used as the Hexahedral (legacy) mesh. Two-port design. Two inner diameters
are considered: 17 mm and 17.03 mm. Perfect conductor pipes. Short-circuited model. 1.0
mm pipe wall thickness. All models are evacuated.
Recreating the 2007 research [32]: PWT is located at the end of the pipe. The design
was totally recreated, with two differences: we removed the cap and modeled 2 ports. We
also re-sequenced the PWT.
The original 2007 study used PWT at 10, 20, 40, 60, & 80% of wall thickness. For
our modeling research the radius of PWT was modeled with step-width 0.1 mm from 8.6 to
9.4 mm radius.
Figure 2.10: Modeling of experimental results for Yang Ju (2007). Shown is scattering
parameter for the outer radius of PWT ranging from 8.6mm to 9.4mm
26
Recreating the 2011 research [34]: PWT between two pipe sections. Will recreate the
two-pipe one coupler configuration. PWT at 5, 10, 20, 40, & 60% of wall thickness.
Recreating the 2013 experiment [35]: This research simulated a Biofilm at the end of
the pipe. The original used 17.03mm ID, which we recreated; length of wall thickening was
51mm, 34mm, 17mm.
We marked the wall-thickening examples in our results as 399 mm, 416 mm, and 433
mm, the distance from port 1 to the beginning point of the biofilm. The thickening increase,
with step width 0.21 mm: 0.21mm, 0.42mm, 0.63mm was recreated as well, for each
example length.
2.6 Modeling Results
First, we will examine the results of the modeled recreation of the work of Yang Ju
(2007). In Fig. 2.10, the shallowest PWT model produces no peak. But for the other eight
models, a peak is generated. As the radius of PWT increases (PWT deepens, volume
increases) the peak also increases in amplitude up to a certain point. Then, as the volume
continues to increase, the microwave (MW) signal is attenuated by the volume of the defect,
such that a great deal of energy is lost, resulting in a decreased amplitude, as shown in Fig.
2.10. The greatest volume (R=9.4 mm) generates a broad peak without much height. As
expected, the peaks migrate toward port one with increasing volume of the PWT model.
Second, we will look at the results of modeling the experiment conducted by Nasser
Saber et al in 2013. For the results with the thickening at 399mm from port 1, Fig. 2.11, at all
three modeled thicknesses, the waveform fails to differentiate the wall-thickening due to the
length of the biofilm.
This figure, Fig. 2.12, shows the modeled biofilms with starting point at 416mm from
port 1. Ri=8.515 mm with defect-free pipe walls (inner radius of the pipe). The radius of
thickening (biofilm) 7.885 mm, has a large peak because it represents the thickest biofilm
(0.63mm). The second one the biofilm radius 8.095 mm and 8.305 mm for thickness 0.42
27
mm and 0.21 mm respectively, has a small reflection, as shown in Fig. 2.12. Again, all these
results are for the biofilm sections with 34mm length.
Figure 2.11: Modeling of experimental results for Nasser Saber et al (2013). Shown is
scattering parameter for S11 for the 51mm length biofilm section
Figure 2.12: Modeling of experimental results for Nasser Saber et al. Scattering
parameter for S11 for the 34 mm length biofilm section
28
Figure 2.13: Modeling of experimental results for Nasser Saber et al. Scattering
parameter for S11 for the 17mm length biofilm section
In Fig. 2.13 we see the results of the modeled biofilms with 17mm length, located at
433mm from port 1. We have a large reflection from the R=7.885 mm sample, and the
smallest reflection from R=8.095 mm. Still, we have a small reflection from the radius 8.305
mm, because it is near to the signal for the defect-free pipe (R=8.515 mm).
2.7 Discussion
In Fig. 2.10, it is very clear that we have peaks from most of the PWT models,
beginning with R=8.7 mm. The peak height increases as the PWT volume increases until
peak maximum is reached with R=9.0 mm. The larger PWT models continue to generate
peaks, but they reduce in height due to attenuation of the energy (loss of energy in the PWT)
until we reach the peak for the greatest diameter (R=9.4 mm), and greatest volume of PWT.
In all three figures from the recreation of Saber et al, (Figs. 2.11, 2.12 and 2.13) we
have a peak when we model the thickest biofilm. The microwave signal is reflected and
attenuated at this thickness. When the thickness diminishes to be near the inner diameter of
29
the pipe, the peaks almost flatten out becoming very much like the peaks produced by the
defect-free inner pipe wall. Especially for Fig. 2.11, the peaks flatten out with the greatest
length (51mm) of the modeled biofilm. This is because of the wavelength used. We have a
small diameter pipe, and a small wavelength to target the greatest length of biofilm modeled.
2.8 Conclusions
CST can replicate real world conditions. The researchers tried to see if there was a
measurable effect on the microwave signals in the case of anything anomalous showing up
inside the pipe under test. They sought to understand the behavior of the microwave signal
inside the metal waveguide, and the influence on the reflection waveforms whether by
anomalous empty space (PWT) or areas filled with anomalous material deposits (biofilm).
From these researchers (the authors worked in PWT and biofilm detection), we saw
the results produced as they calibrated their microwave detection systems with the specimen
pipe empty inside (defect-free pipe). And then the subsequent experimental results as they
attempted to see if the microwave signal was affected as they changed the specimen
parameters: location and size of PWT or biofilm.
Lenshing Liu in 2011 tried to detect the defect (PWT) in a pipe using the group
velocity method of signal processing. There was difficulty in distinguishing the
characteristics of the defect. But it was clear that a measurable effect on the microwave
signal was generated by the presence of PWT specimens.
As we conclude from the work begun by these researchers we need a good procedure
to locate and to seek the classification of PWT. The aim is to eventually be able to
characterize the PWT defect by recognizing the depth, length around the pipe, width of, and
accurate location of the PWT from the reflected waveforms.
30
Chapter 3
Applications of Ultrasonic Techniques in Oil and Gas Pipeline Industries:
A Review [29]
Abstract
The diversity of ultrasound techniques used in oil and gas pipeline plants provides us with a
wealth of information on how to exploit this technology when combined with other
techniques, in order to improve the quality of analysis. The fundamental theory of ultrasonic
nondestructive evaluation (NDE) technology is offered, along with practical limitations as
related to two factors (wave types and transducers). The focus is limited to the two main
techniques used in pipe plants. First, straight beam evaluation and second, angle beam
evaluation. The depth of defect (DD) is calculated using straight beam ultrasonic in six
different materials according to their relative longitudinal wave (LW) velocities. The
materials and respective velocities of LW are: rolled aluminum (6420m/s), mild steel
(5960m/s), stainless steel-347 (5790m/s), rolled copper (5010m/s), annealed copper
(4760m/s), and brass (4700m/s). In each material eight defects were modeled; the first
represents l00% of the material thickness (D), 50.8mm. The other seven cases represent the
DD, as 87.5% of the material thickness, 75%, 62.5%, 50%, 37.5%, 25%, and 12.5% ,
respectively. Using angle beam evaluation, several parameters are calculated for six different
reflection angles (βR) (45°, 50°, 55°, 60°, 65° and 70°). The surface distance (SD), ½ skip
distance (SKD), full SKD, and 1½ SKD, ½ sound path (SP) length, full SP, and 1½ SP are
calculated for each βR. The relationship of SKD and SP to the βR is graphed. A chief
limitation is noted, that ultrasound testing is heavily dependent on the expertise of the
operator, and because the reading of the outcome is subjective, precision may be hard to
achieve. This review also clarifies and discusses the options used in solving the industrial
engineering problem, with a comprehensive historical summary of the information available
in the literature. Merging various NDE inspection techniques into the testing of objects is
discussed. Eventually, it is hoped to find a suitable technique combined with ultrasonic
inspection to deliver highly effective remote testing.
31
3.1 Introduction
Ultrasonic testing is one of the important techniques of nondestructive testing (NDT).
It uses ultra-high-frequency sonic energy to locate and identify discontinuities in materials
that are both on and below the surface of the material (such as metals or plastic, commonly
used to make pipes, depending on the application) [40] [41] [42] [43] [44].
In 2007 D. S. Caravaca et al [45] studied polyethylene pipe joints and detection of
improper preparation of the joints using a phased array technique (PAT). The problems that
arise with the electrofusion type of bonding used for the pipe are analogous to those that
occur in metal welds for steel pipe. This paper aims to offer an ultrasonic method for
evaluating polyethylene pipe welds, in pipeline used for gas and water distribution, along
with results from both laboratory and field experiments [45].
The sonic energy passes through the substrate. There is a reduction in energy
intensity, as well as reflection of the waves by the back wall of the material, and where
discontinuities are encountered. The returning signal is captured, mathematically analyzed
and presented on a screen, with the resultant waveform showing the location of defects on or
within the substrate [46] [47] [48] [49] [50].
In 2006, MAO Yi-mei and QUE Pei-wen investigated the possibility of using a thennew sonic signal processing method for inspection of oil pipelines. They compared
“ultrasonic signals reflected from defect-free pipelines and from pipelines with defects” [51]
and treated the recaptured waveforms with the Hilbert-Huang transform (HHT). The results
demonstrated the feasibility of using the technique to successfully locate and determine the
extent of discontinuities in oil pipes [51].
Reflected signals are attenuated to different degrees depending on the type of
interface they encounter. An interface between a metal and a liquid presents a reduced
reflection of the sonic energy, whereas an interface between a gas and a metal causes nearly
100% reflection of the sonic waves [46]. The actual percentage of reflection is dependent on
the ratio of the parameters of certain physical properties between the two types of material.
For example, the ratio between the metal and the liquid substance at the interface [52] [53]
[54] [55].
32
Cracks, holes, laminations, slags, cavities, porosities, bursts, lack of fusion, flakes,
lack of penetration and other discontinuities that produce sharp boundaries are easily
identified by ultrasonic testing. Other types of discontinuities that produce a more diffuse
boundary are still identifiable because they will disrupt the sonic waves in a detectable
manner [1] [46].
In 2015, Wissam Alobaidi et al surveyed seven types of defects commonly found in
pipe joint welds, and five often-used types of welds in manufacturing. The correlation
between each defect type and the NDT technologies which best reveal the defects are
presented in a table [1]. The ability and shortcomings of four NDT techniques commonly
used for testing pipe are compared, one of which is ultrasonic testing, and the table reports
whether each technique can detect surface, or subsurface flaws. The paper examines ways in
which new quality assurance techniques can be incorporated alongside the standard methods
in order to overcome the shortcomings of current methods, with the aim to reduce labor costs
and increase line output [1].
Because the sensing mode of ultrasonic evaluation is a mechanical process, the
frequency range is limited to avoid permanent damage to the targeted objects. Frequencies
used most often range from 0.1 MHz to 25 MHz. Although Ultrasonic testing (UT) is capable
of identifying surface defects, it is primarily used to detect and locate discontinuities that are
below the surface, especially in metal parts. UT is useful for other types of inspection,
including welds, wall thinning, and surface defects, as mentioned above [40] [46].
This review paper presents applications of, and limitations of some ultrasonic
techniques; we will demonstrate the fundamental theory of ultrasound and type of waves
used; we will thoroughly examine the inspection approaches of the contact and angle beam
techniques. Approaches discussed focus on the measurement of defects in oil and gas pipe
manufacturing. Primarily we are interested in determining the depth of the defect (DD) below
the material surface. We will also address the limitations of scanning pipe that depend on the
transducers used (contact for pipe body and angle beam for welds). Moreover, the paper
presents a literature survey for applications of ultrasonic techniques in the pipe industry. The
primary aim for this review is to investigate possible future coupling of one of the ultrasonic
techniques with other NDE techniques, to develop a hybrid detection system for
33
discontinuities. This research study is an accumulation of the practical experience of the
authors, as well as the practical experience represented by the reviewed papers.
3.2 Fundamental Theory of Ultrasonic
3.2.1 Ultrasound
Ultrasonic inspection uses sound as the source for testing the medium under
consideration. This is the same kind of sound that creates the motion of our eardrums and
allows us to hear. The vibrations used for Ultrasonic Testing (UT) are very much higher
frequency than what we can hear. But just like any sound wave that moves through the air,
the ultrasonic waves that are sent into metal will propagate through the solid medium. When
these vibrations encounter interfaces between discontinuous materials (which represent
defects in the materials and welds of pipes, for example) they will be reflected in predictable
ways. UT is a commonly used method in industries for quality control purposes. It is useful
for testing the integrity of metal parts, both before and after forming into pipes. The roll stock
can be tested for invisible defects using straight beam ultrasonic testing, allowing the
material to be categorized as acceptable, repairable, or scrap before it is incorporated into
pipes [1] [46] [56].
Because air does not transmit ultrasound waves as well as solids or gels the difficulty
of introducing the signals into metals is overcome by using water or grease as a conducting
medium between the transducer and the material to be tested [52].
Ultrasonic testing is used both during the manufacturing of pipe and for inspection of
in-service pipelines. The particulars of pipeline insulation may require ingenious ways of
getting the ultrasonic transducers into contact with the pipe to be tested. For example, in
2009, H. Lei et al [57] were interested in developing a device for assessment of the inner
walls of underwater oil pipeline. The device, called a pig, uses ultrasonic testing to discover
and determine the extent of corrosion within the pipe, while storing the data on a hard disk.
Following retrieval of the pig, the recorded data is analyzed in order to determine the
reduction in the oil pipe wall, and to identify areas of wall thinning which are then
34
categorized using the American Petroleum Institute (API) standards. The authors claim
“perfect performance” of the device [57].
3.2.2 Waves
Waves commonly used are:
Longitudinal Waves (LW): Another name for these waves is “compression waves”. LW is
the type of sound wave that we hear, and that is used in manual UT for testing the front end
and tail end of the pipe body, and in coil UT for testing the integrity of the plates before they
are formed into pipe. The LW pushes the molecules of the tested material in the same
direction as the movement of the wave, as shown in Fig. 3.1. The velocity of the ultrasonic
LWs is different in different metals; for example, their velocity through copper is roughly
4760 m/s, making them the most rapidly propagating ultrasonic waves used in NDT.
Through the analysis of wave velocity the depth of defect (DD) can be found [46] [58] [59].
This is what we demonstrated in Section (3.4.1).
Figure 3.1: Graphical depiction of parallel motion response of material particles subjected to
longitudinal ultrasonic waves, showing compression and rarefaction regions
Shear waves (SW): Shear waves, also known as transverse waves, propagate more slowly
and at shorter wavelengths than LWs at equal frequencies. The particle motion is at right
angles to the movement of the wave, as shown in Fig. 3.2. SW is usually used for angle beam
UT (for example, to detect discontinuities in both the inner diameter and outer diameter of
35
the weld in pipes). As with LW, the SW velocity varies with the type of metal. Some
example velocities and corresponding metal types are: Aluminum, roughly 3040 m/s; Steel,
347 Stainless, roughly 3100 m/s. When SWs reflect from an interface they sometimes
become LWs [46] [58] [60] [61]. This is demonstrated in Section (3.4.2).
Figure 3.2: Graphical depiction of perpendicular motion response of material particles subjected
to shear ultrasonic waves, showing wavelength
Rayleigh waves (RW): These waves, which penetrate the material only to the sub-surface
distance of one wavelength (at any given frequency), are also called Surface waves. RW
travels along the surface of the tested material at velocities equal to those of SWs. RWs are
useful for detecting cracks that break the surface of the tested part. They are also useful for
testing pieces with intricate rounded surface features. Any defect in zone α as shown in Fig.
3.3, would rest deeper than the wavelength (λ) of the test signal, and would likely not be
detectable by RW [46] [58] [60].
In 2014, N. P. Aleshin et al examined various methods and devices for proficiently
introducing ultrasonic signals into pipelines with thicknesses ranging between 6mm to
20mm. They covered the choice between Surface and Plate wave modes for conducting these
assessments [62].
Lamb waves (LMW): Lamb waves are vibrations that occur from the upper to the lower
surface (up to several wavelengths in thickness) of the tested material, usually a plate
36
Figure 3.3: Graphical depiction of limited detection area of Rayleigh waves, showing
how they are confined mostly to the surface of a material
Figure 3.4: Graphical depiction of ultrasonic Lamb waves (plate waves) showing how they
move through a test object of a certain thickness which is directly related to the wavelength
(composites or metals), so they are also called Plate waves. They propagate not only through
the full thickness of the tested material but are capable of propagating from a single point of
excitation over significant distances within the material, as shown in Fig. 3.4. Because the
LMWs travel through the solid in a way that is significantly like the behavior of
electromagnetic waves within a waveguide, the characteristics of transmission vary from
material to material. The velocity of LMWs is dependent on many factors, including density,
plate thickness, and the elastic properties of the material being tested [46] [63].
37
In 2005, Kevin R. Leonard et al explored use of “helically propagating Lamb waves”
transmitted and recaptured with longitudinal transducers. They describe a “meridional-array
scheme” [64] used to test the concept of using tomography to detect location and extent of
discontinuities in pipes. The research sample used wall-thinning as the defect to be detected.
The study also investigated improved reconstruction programs for assessment of helical array
signals where “the transmitters and receivers lie along circumferential parallel rings”,
confirming that frequency compounding reduces noise and artifacts, leading to clearer
imagery [64].
3.3 Transducers
There are five general categories of ultrasonic transducers used in NDE: straight
beam, angle beam, dual element, delay line and immersion transducers. Straight beam and
angle beam transducers are used in pipe manufacturing NDE procedures. Usually, the
standardized inspection codes will determine the type of transducer the manpower (operator)
uses for a particular test. In the case where there is no specification or customer
requirements, the operator will select a suitable transducer based on prior experience and
knowledge [40] [65] [66].
Many studies investigated ways to reduce the manpower needed for inspection of
pipes by using ultrasonic waves, and also ways to reduce the number of transducers required
for an inspection protocol, in order to save capital expense.
In 1998, M.J.S. Lowe et al [67] studied the use of a single transducer for pulse-echo
testing of in-service pipeline in order to reveal corrosion in insulated pipes, including oil
pipes. They removed only part of the pipe insulation, in order to reduce labor cost. The study
focused on selection of the most effective wave modes, and understanding the relationships
between the size of a flaw and the signal strength when reflected. The technique was in field
trials at the time of publication [67].
In 2004, M.H.S. Siqueira et al investigated the use of a single transducer to replace
groups of transducers that are used for rapidly inspecting pipe. The research assessed the use
38
of guided wave pulse-echo conformations with low ultrasonic signal to noise ratios (S/N),
together with processing via “frequency bandpass filters and wavelet analysis”. Their results
confirmed the practicality of the concept, showing up to 12 dB S/N enhancement of the
recaptured signal, allowing analysis even with otherwise unusably noisy signals [68].
In 2006, Younho Cho et al [31] carried out a feasibility study of using guided sonic
signals for remote monitoring of stainless steel pipe. They report that their experimental
approach, intended to allow them to optimize the guided wave mode resulted in the discovery
that “Predicted modes could be successfully generated by controlling frequency, receiver
angle and wavelength.” By analyzing scattering patterns mode by mode, they were able to
determine that “mode conversion characteristics are distinct depending on dispersive pattern
of modes” [31].
3.4 Approaches
3.4.1 Straight Beam Evaluation
The ultrasonic signal used for UT is not continuous. A brief pulse of ultrasound is
emitted into the test material from a transducer; the signal travels through the test piece
thickness and echoes from either the back wall of the piece, or from a discontinuity within
the piece. The echoing signal is captured by the transducer only a few microseconds after
being emitted. This gives the process the name pulse-echo [40].
The velocity of travel within the tested material must be known in order to calculate
both the presence and the depth of the defect. In a flawless test piece, the distance down and
back would each equal D, the full thickness of the metal. Thus, the transit time TT represents
the sonic waves propagating from probes S1 and S2 in both cases in Fig. 3.5 and the
reflection from the back wall of the piece, so for the full thickness the total travel will be 2D.
The factor (VL) is the velocity of LWs within the type of metal tested, as shown in Equation
(3.1). If the signal finds a defect, the transit time TD would represent the sonic waves
propagating from probe S3 in case 2 and the reflection from the interface of the defect, so the
distance traveled will be 2DD where DD is the distance from the front surface to the interface
39
with the defect as shown in Fig. 3.5, case 2. Thus, the velocity of LWs (VL) in this case is
used to determine DD, and TD is the time of transit to and from the defect, as shown in
Equation (3.2) [69].
Figure 3.5: Scheme for the test sample. Two cases are represented. Case 1 has S1 (straight
beam #1). Case 2 has S2 (straight beam #2) and S3 (straight beam #3). This figure represents
calculation of the value DD, using Equation (3.2). See Section 3.6 Discussion
 =
2

(3.1)
 =
2

(3.2)
Straight-beam testing is commonly used in pipe manufacturing to test the roll stock
which will be used for building the pipe body. Straight beam testing is effective for detecting
cracks that occur parallel to the surfaces of the tested material as well as discontinuities
within the body of the material, such as voids or areas of porosity, and inclusions [70].
Fig. 3.6, shows a test sample including eight cases, with the case number shown on
the probes from 1 to 8, respectively. This demonstrates the measurement of the various
40
distances calculated as DD according to Equation (3.1) and Equation (3.2) above, and as
graphed in Fig. 3.7.
The velocities of sonic waves penetrating different materials are compared in Table
3.1. These known constant values are necessary to calculate the distance to defect depending
on the material from which the tested object is made. The eight cases are calculated for each
of six materials, given different transit times for each, as shown in Fig. 3.7, which compares
the curves resulting from the eight defects in different materials.
Table 3.1: Longitudinal wave velocities shown for 6 different materials.
Materials
Aluminum,
Steel, mild
rolled
VL (mm/s)
6420000
5960000
Steel, 347 Copper,
Copper,
Stainless
rolled
annealed
5790000
5010000
4760000
Brass
4700000
Figure 3.6: A test sample with 8 cases of defects showing the reflection of test signals
from straight beam probes
In practical use researchers design mechanisms in order to test their ideas, In 2013,
Jin-sheng YANG et al [71] reported a system to measure variations in the wall thickness of
41
gas pipe using ultrasound based on the standard cleaning pig with some adaptations. To
remove the need of a couplant, the system incorporated a wheel made of an elastic substance,
which through mechanical tightness of fit conveyed the vibrations to the pipe wall from a
standard piezoelectric probe. The paper covers the operational principle and the system
design [71].
Figure 3.7: The time domain for the 8 cases from Figure 3.6 are graphed for each of the 6
materials in Table 3.1
3.4.2 Angle Beam Evaluation
Straight beam ultrasonic techniques are best for locating defects in plate-type
materials where defects are often parallel to the surface of the object, but they are ineffective
for testing welds. The discontinuities in welds are usually at an angle to the surface of the
sample under test. Beams approaching the weld interface at an angle are effective at
detecting discontinuities within the welding bead. The angle beam transducer is used to
generate the test signal in the majority of ultrasonic inspections. To assess discontinuities
using angle beam examination, skip distance (SKD) is used to describe the sound path (SP)
reflected from the back wall interface (1st leg) and going immediately to reflecting, where it
is again reflected from the front wall (2nd leg). SKD is formed by the full SP (1st leg and 2nd
42
leg), and is the distance from the point of excitation to the end of the second leg, as shown in
Fig. 3.8 [1] [40]. All the equations below are adapted from [40].
Figure 3.8: Representation of Skip Distance through Three Legs, ½ Skip, Full Skip and
1½ Skip. D is the material thickness and βR SP angle
 = 2 (β )
(3.3)
Surface distance (SD) is equal to half SKD.
 =  (β )
(3.4)
These calculations cannot be completed unless D and βR to the front surface are
known. If the length of the 1st leg is known, the SKD and SD can be computed as.
 = 2(1) (β )
(3.5)
 = 1 (β )
(3.6)
and
The segments of the SP numbered L1, L2 and L3 in Figure 3.8 are 1st leg (L1), 2nd
leg (L2) and 3rd leg (L3) respectively. And through the trigonometric functions SKD, SD,
L1, L2 and L3 can be calculated as shown in Table 3.2, by taking 70, 65, 60, 55, 50 and 45
degrees as the βR. The first leg and second leg are calculated as follows.
43
1 =

( )
(3.7)
and
1 + 2 =
2
(β )
(3.8)
Table 3.2: Summary and comparison of the SKD and SP lengths for 6 different beam angles
SKD (mm)
βR
SD (mm)
½ SKD
SP (mm)
Full
1(½)
SKD
SKD
½ SP
Full
1(½)
SP
SP
45
25.4
25.4
50.8
76.2
36
72
108
50
30.27
30.27
60.54
90.81
39.515
79.03
118.545
55
36.275
36.275
72.55
108.825
44.3
88.6
132.9
60
44
44
88
132
50.8
101.6
152.4
65
54.47
54.47
108.94
163.411
60.1
120.2
180.3
70
69.786
69.786
139.572
209.358
74.265
148.53
222.8
Figure 3.9: Skip distance plotted against sound path (SP) angle, showing the SP distances
achieved with ½, full and 1½ SKD, for use by the operator in selecting the SP angle to be
used during test depending on the material
44
According to the Table 3.2 above, we graphed the SP angle against the SKD. These
graphs show the limitations of the scan distance to the pipe weld depending on βR. Where,
SKD and SP length depends on the propagating sonic angle. That means, the larger the angle
is, the greater the scanning distance becomes, based on the (½, Full and 1½) for SKD and SP
as shown in Figures 3.9 and 3.10.
Figure 3.10: Sound path length plotted against SP angle. This shows the exact length of
penetration of sonic waves according to the material for ½, full and 1½ SKD
To calculate the depth of the discontinuity (DD), we measure vertically from the point
of reflection at the interface with the defect, up to the front surface of the test sample [69], as
shown in Figures 3.11 and 3.12. To calculate the DD’s, the values of D, SKD and SD must
be known, as shown below. Fig. 3.11 shows a defect interrupting the path of L1. Fig. 3.12
shows a defect interrupting the path of L2.
 = 1 (β )
 = 2 − [(β )(L1 + L2)]
(3.9)
(3.10)
45
Fig. 3.13 shows a schematic of a pipe cross section containing a cluster porosity
discontinuity in the pipe weld. This is an example of one type of discontinuity that is best
discovered by using an Angle Beam Transducer.
Figure 3.11: Illustration of the test object and defect depth found by Angle Beam
Transducer (defect found by the 2nd leg)
Figure 3.12: Illustration of the test sample and discontinuity depth as located by
angle beam transducer (defect located by the 1st leg)
46
Figure 3.13: Pipe cross section schematic of discontinuity depth discovered on the
2nd leg by Angle Beam Transducer
3.5 Summary
Ten analyses done by ultrasonic NDE specialists are listed in Table 3.3 below which
summarizes the problems studied by each paper, the name of the first author and the year of
publication.
Table 3.3: Selected prior research into uses of ultrasonic NDT to detect discontinuities in
pipeline materials
No.
Problems
First
Author Year
Name
1
Defect detection in pipes using guided waves
M.J.S. Lowe
2
The use of ultrasonic guided waves and wavelets M.H.S. Siqueira
1998
2004
analysis in pipe inspection
3
Lamb wave tomography of pipe-like structures
Kevin R. Leonard
2005
47
4
A wall thinning detection and quantification based on Younho Cho
2006
guided wave mode conversion features
5
Application of Hilbert-Huang signal processing to MAO Yi-mei
2006
ultrasonic non-destructive testing of oil pipelines
6
Ultrasonic phased array inspection of electrofusion joints
D. S. Caravaca
2007
in polyethylene pipes
7
Ultrasonic Pig for Submarine Oil Pipeline Corrosion H. Lei
2009
Inspection
8
A New Type of Wheeled Intelligent Ultrasonic Jin-sheng YANG
2013
Thickness Measurement System
9
Automatic Ultrasonic Inspection of Large-Diameter N. P. Aleshin
2014
Pipes
10
A Survey on benchmark defects encountered in the oil Wissam Alobaidi
2015
pipe industries
3.6 Discussion
To summarize the objectives of our research reported in this paper are?
◊
Extending the service life of pipes through early quality control testing by finding more
effective and appropriate ways to evaluate them.
◊
Demonstrating the limitations of two Ultrasonic testing techniques as covered in sections
above.
◊
Employing the technique to extend the effectiveness of this technology in determining the
location and depth of defects.
◊
Coupling ultrasonic technology with another NDT technique in an automated system to
quickly determine the exact location and degree of the discontinuity.
◊
After designing a new system merging the Ultrasonic technique with another NDT
technique, to build the actual system and employ it in real-world conditions to remotely
detect and evaluate discontinuities.
48
The focus here will be on approaching straight beam and angle beam testing, with
each considered as a different technique. The first is a longitudinal wave mode where the
wave direction is parallel to the SP, and the second one is a shear wave mode where the wave
direction is perpendicular to the SP. These techniques have different uses, as the straight
beam is best for the pipe body and the angle beam for the pipe weld. As the review shows,
the limitations for these techniques have been examined before. It should be mentioned that,
while preparing a newly designed system, the limitations and advantages of the techniques
must be considered. The resulting new system must be fit and appropriately in line with these
technical conditions.
As shown in Fig. 3.5, the straight beam probe transmits an ultrasonic wave into the
test object. In Case 1 there is no defect, thus the beam travels the full thickness of the
material. The peaks for Case 1 represent the origination of the sonic signal, and the reflection
of the signal from the back wall. In Case 2 probe S2 sends an unimpeded signal, which
behaves like that in Case 1. But S3 sends a signal which is reflected by a defect, so the signal
reflection arrives in less time, represented by the middle peak in the graph for Case 2. This
yields the distance to defect (DD).
A consideration for angle beam testing is that the attenuation of the sonic waves
varies depending on the material from which the object is made. As shown in Figures 3.9 and
3.10, βR is adjusted according to the degree of penetration possible with a given material. For
a more penetrable substance, βR is greater, allowing a greater SKD, while a less penetrable
substance requires a smaller βR, with a respective reduction in SKD.
3.7 Conclusion
It is usual practice to base an NDT research on the various materials used in order to
estimate and evaluate the suitability of ultrasonic NDE for testing the materials involved. The
next step is to develop the needed devices, then to investigate the problem of using the
ultrasonic NDE technique to assess the structural objects.
49
Through previous research, and the techniques that have been focused on, it can be
seen that the straight probe works up to a limited thickness (such as the thickness of a pipe
wall), and is employed to reveal subsurface defects, especially cracks that are parallel to the
front and back surfaces of the test object. The example given in this article demonstrates a
pipe wall with D=50.8mm, with eight cases of defect where DD ranges from 6.35mm to
44.45mm and presents the graphed results based on six different materials commonly used in
the pipe construction industry (shown in Fig. 3.7).
The angle beam, is used to reveal discontinuities in the weld regions, with the
limitation that the SKD cannot be increased unless the SP angle is increased. And that
increased distance attenuates the sonic energy. The examples included in this article
calculated the limitations of effective SD, SKD and SP using six reflection angles. Focusing
on the 45° and 70° angles at the extremes for our example of D=25.4mm, we show that the
45° angle yields a ½ SKD (25.4mm), full SKD (50.8mm) and 1½ SKD (76.2mm) with higher
energy if we compare to the same material but a different angle, such as 70° angle which a
greater test distance, but with more signal attenuation. Using the 70° angle we get ½ SKD
(69.786mm), full SKD (139.572mm), and 1½ SKD (209.358mm). The greater reach leads to
reduced sensitivity because of the signal attenuation; operators must balance these factors
when ultrasonic NDE is used. The general understanding is that 1½ SKD is the accepted
limit of successful range for angle beam testing, all that according to the previous studies and
technical websites. This requires continual attention by customer requirement and if there is
no requirement, by the operator when assessing the integrity of pipes during manufacture.
It is recommended that a second, powerful NDE technique capable of remote
detection would be useful alongside ultrasonic testing in order to direct the placement of the
probe.
50
Chapter 4
Enhancing Production Efficiency of Oil and Natural Gas Pipes Using
Microwave Technology [30]
Abstract
The research reported in this paper aims at developing means of Non Destructive testing
(NDT) to increase the line efficiency of pipe production in oil and natural gas pipe
manufacturing plants using the Standard Allowed Minutes (SAM) method. Existing line
production stations encounter difficulties in maintaining the recommended testing speed of
smaller diameter pipe, due to limitations in the Visual Inspection (VI) station. We propose to
implementing one additional technique which will prevent the decline of line efficiency in a
pipe production factory. The range of diameters identified as a problem in this research is
from 254mm to 762mm. Microwave techniques are expected to improve the line efficiency
by increasing the production of the plant. This happens as a consequence of maintaining the
production rates of the identified pipe diameters, so that they equal the production output of
the larger pipe diameters. We analyze the velocity traveled by the pipe through Radiographic
Testing (RT) according to the VI output (production). The RT velocity is decreased for the
diameters identified above, in order to maintain quality control and cover the shortcoming of
the VI. The number of pipes produced is computed during shift hours of the factory and pipe
lengths of the forming department are determined. We compare the output (production) of a
series of NDT line stations with and without the microwave technique for the first of the
three pipe cases considered in this study, classified as perfect pipe (PP), repair pipe (RP) and
scrap pipe (SP). The velocity of RT stations analyzed in the paper range from 50mm/s for
larger diameter pipe, and decline to 16.667mm/s for the identified diameters. The analytical
calculations of line output (production) and line efficiency demonstrate the solution of this
velocity problem after the microwave technique is introduced. It demonstrates that an
economical and precise methodology to extend the production capability of the pipe plant has
been determined.
51
4.1 Introduction
Oil pipes with diameters up to 500mm are used as the standard method of
transporting petroleum and derivatives or natural gas for great distances overland. Corrosion
of the inner pipe wall often occurs at the bottom of the pipe in the field because of the
presence of water in the pipelines [72]. This corrosion leads to removal of material, creating
a reduction in the wall of the pipe, called Pipe Wall Reduction (PWR) [32] [73]. Defects
develop even during the manufacturing process, but these are often caught in factory due to
standard testing guidelines. Such discontinuities can be assessed and repaired per the
standards and codes, before the pipe is shipped to the field. During its time in service, any
pipeline will develop discontinuities that are of a greater extent than those that show up in
manufacturing. Once again, there are standards that can be used to determine whether repair
or replacement is the proper action on a case by case basis [1] [19].
Seeking a solution for determination of whether a field-defective pipeline is still
suitable for service has inspired a great deal of research. A number of analytical approaches
have been tested for detection and sizing of such defects in situ [74].
NDT techniques have been developed during the past few decades to determine the
presence and extent of such discontinuities. Our previous survey paper reviewing the
application of several of these NDT methods, looked at digital x-ray, ultrasonic, visual
inspection, radiographic testing and so on [1]. Detection of surface defects requires a large
amount of time and labor for each of these techniques. For larger diameter pipes the
techniques can be used effectively, but smaller diameter pipelines prevent entry of human
inspectors to perform visual inspection, for example [1].
Previous studies have demonstrated that microwave NDT techniques can reliably
detect surface defects in pipes. The technique uses a microwave emitter/detector that
introduces a microwave signal into the waveguide (the pipe itself) and detects the reflected
waves [32] [75]. Normal pipe walls generate a constant resonant frequency (no loss in
energy) in the reflected waves. PWR creates variances in the waveforms produced, allowing
calculation of the location and extent of PWR defects. Inspecting the pipe inner walls can be
done without entering the pipe, and without moving the detector during the test, while
52
detection proceeds in real time. This has obvious advantages in the case of smaller diameter
pipeline [32].
After the advantages of microwave NDT techniques become widely known, we
foresee the inclusion of microwave testing stations among the series of NDT testing stations
in pipe manufacturing facilities.
Standard Allowed Minutes (SAM) was used to establish a general plan for the target
(line output and line efficiency standard) and was developed by analysis of the time required
to produce the product [76] [77].
Many of the techniques listed in our survey paper work well with larger diameter
pipes, but are difficult or impossible to use on smaller diameter pipes [1].
We present an analytical study of inspection routines including Visual Inspection (VI)
and Radiographic Testing (RT) with and without the microwave technique (which will be
placed between VI and RT) for smaller diameter pipe, and then study the case of the PP with
the SAM for the two routines.
It is worth mentioning that this paper essentially aims to enhance the production and
line efficiency of oil and gas pipes by using microwave technology. The observational study
employs microwave technology to carry out the testing of the inner surface of the pipe to
meet the VI requirement. The commonly used VI station can carry out the test for all
diameters except the group ranging in size from 254mm to 762mm, which cannot be
inspected from the inside with VI.
This limitation requires slowing the velocity of the test pipe in the RT station, as
mentioned in our previous study, from 50mm/s to 16.667mm/s [1]. Placing microwave
technology between the two stations has three advantages. First, to get the velocity of the RT
station up to the customer requirement. Second, to inspect the pipe from inside for those
smaller diameters listed above. Consequently, the pipe can be released to the RT station
without interruption, resulting in improved line efficiency and line output of the product.
Third, ensuring the high quality of the product during the test.
53
This research will confirm the times needed for examination of the pipe after its
formation for the diameter range listed above. Then we will compute the line production
values to yield line efficiency for each station, by analyzing the necessary decrease in speed
of the wagon that carries the pipe to conduct the ultrasonic test for the weld before the
radiography test. Calculations will be performed for various lengths of pipe: 24384mm,
21336mm, 18288mm and 15240mm respectively. Line production and line efficiency will be
computed for various shift lengths: 8hrs, 9hrs, 10hrs, 11hrs and 12hrs. After adding
microwave technology the computations will be repeated for the new setup. Results for the
previous sequencing of stations after VI will be compared to the new sequence including the
microwave technique which evaluates the pipe from inside.
4.2 Sequential Stations of Non-Destructive examination in spiral
pipe plants
◊
Coil Ultrasonic Testing (COUT): This station is responsible for detecting
discontinuities in the materials of the pipe body before it is spirally formed. The station
uses straight beam ultrasonic signals to test the roll stock, as shown in Fig. 4.1. COUT
allows identification of discontinuities with enough precision to allow the ones that
exceed the standards to be cut away before the pipe is formed. In most cases the
manufacturer specifications require a clean run of at least 40 feet up to 80 feet, for the roll
stock to be usable [29] [78].
◊
Visual Inspection (VI): After the formation of the pipe and the application of spiral
welds internally and externally, the pipe is sent to the VI station to evaluate the pipe body
and its welds for outer diameter (OD) and inner diameter (ID) [1] [79].
◊
Radiography Testing (RT): Inspection in this station is confined to the spiral weld only,
from the front end to the tail end of the pipe. This station contains two examination
methods, as shown in Fig 4.1. First is the Ultrasonic machine to give an indication of the
manpower needed.
The examination process follows the weld by spiral rotational
54
movement of the pipe, to ensure the probe is placed reliably to provide indication of any
discontinuity for the ID and OD (two separate groups of probes are used to test the ID
and OD of the weld simultaneously). The second is the RT machine which is used when
UT indicates that follow-up inspection is necessary [1] [80].
◊
Hydraulic System (HS): This is a system to inject water inside the pipe to create the
amount of internal pressure called for by the requirements and standards code. Pipes are
checked for any leakage [81].
◊
Seam Ultrasonic Testing (SMUT): This machine is limited to the main section of the
weld. Probes are used to test the ID and OD of the weld, beginning 200mm from the front
end, and ending at 200mm from the tail end (to protect the probes from damage due to
contact with the flanges of the pipe). The pipe is moved helically for scanning. This test
paints a red line on any section of the weld that shows a discontinuity. The red lines
indicate to the NDT person which areas must be tested at the next station [29] [82].
Figure 4.1: Illustration of Oil and Natural Gas Pipe Manufacturing Process
◊
Manual Ultrasonic Testing (MUT): This testing station includes two checks of the pipe.
First is the straight beam probe. This is used to check the body of the pipe
55
circumferentially from the outside within a 254mm linear distance from both front end
and tail end (to cover the pipe body in the areas not covered by SMUT). Second is the
angle beam probe. This checks the spiral weld at both front end and tail end, within a
254mm linear distance. The angle beam probe is then used to check any indication from
the SMUT (red painted areas) of a discontinuity along the spiral weld [29] [83].
◊
Digital X-Ray (D-XR): This last station is for verifying the places that have been pointed
out by MUT. After this test, pipes with defects needing repair (RP) are sent to the repair
station. Scrap pipe (SP) go to the scrap area. Perfect pipe (PP) goes to the beveling
station, then to final inspection [1] [84].
◊
Microwave Inspection: High-frequency Weld (HFW) or Electrical Resistance Weld
(ERW) small diameter steel pipe is sometimes tested with microwave signals. This kind
of pipe is useful in situations with liquids under reduced pressure, such as the transfer of
water, oil, and in equipment manufacturing [85] [86]. All of the research done on
microwave as an NDT technique was done with smooth inner wall pipe of this type [32]
[75]. We propose to incorporate the microwave technique to test spiral weld pipe to
reduce the test times spent in the stations following visual inspection (where the pipe
cannot be entered by manpower for direct visual inspection, due to small inner diameter).
Fig. 4.2 below shows the schematic of the microwave process.
Figure 4.2: Illustration of a typical setup for microwave propagation for NDT of
small diameter pipe
56
4.3 Theoretical Analysis
4.3.1 Confirming the time inspection of the VI and RT stations for PP case
In manufacturing plants is always considered the perfect product for setting up the
output line charts daily, weekly and monthly [87]. To customize the standard timing, when
pipes containing defects are found, the material is divided into three cases: Perfect Product
(PP), Repair Product (RP) and Scrap Product (SP). Furthermore, the correct timing of the
product is always based on the product free of defects. This concept is extensively used in
various industries [1]. As shown in Fig. 4.3, the path overview of the series of NDT stations
for evaluating pipes in manufacturing and testing commences with visual inspection and ends
at the production area. For this evaluation, we apply the concept of SAM to pipe
manufacturing [88] [89]. It is worth mentioning that, while the time the PP spends in the
NDT area is not exactly a SAM, it is considered as an exact time to conduct the partial test
and full test as shown in Fig. 4.4, in order to inspect the pipe in the VI station. The pipes are
split into two groups based on diameter: the first group (G1) consists of pipes ranging from
254 to 762 mm in diameter, and the second group (G2) ranging from 889mm to 1270mm in
diameter. Generally, these two groups will control the increase and decrease in the line
output for each station. Thereby, line efficiency (LE) is different for G1 and G2.
The three equations below are excerpted from [88].
  =
   ()
%
   ()
(4.1)
Exact time to inspect outside diameter (ETOD) = 4min, as well as exact time to
inspect inside diameter (ETID) = 11min. This term is total exact time (TET) = 15min for
each pipe. Moreover, the line output (LO) of this station ranges from 30 to 50 pieces/shift,
per the working conditions for the shift period of 8hrs. So, G1 matches to the ETOD and G2
to the TET.
 =  ∗ 
(4.2)
57
Because of the difficulty the operator has in testing and going through the G1 pipes
from the inside. At this point only the process called partial test is appropriate for G1.
Figure 4.3: Path Overview of NDT Stations, Used in the Evaluation
Figure 4.4: Evaluated Times for VI station as per the G1 & G2, with pipe length
24384mm, number of operator 3 and shift work 8hrs
Therefore, the full test relates to G2, because of the ease of inspecting the pipe from the
inside as well as outside. Once the full and partial testing modes of the VI station for PP are
58
quantified, in accordance to what has been mentioned above, the time for the full test is set
up to be higher than that for the partial test for pipe length 24384mm.
Further, the LE depends on the TMP, and is changed through equation (4.1),
depending on the constancy of TMA, which is described below.
 =  ∗  ∗ 60
(4.3)
Where NOP is number of operators and WH is working hours, both per shift.
Inspection time of the PP through the VI station will decrease because only the VI
partial test is done on the product (G1) to insure the quality of the weld only, not for the
body; but the partial VI test increases the time spent in RT. The PP case of various length and
diameters of pipe in G2 were successfully inspected and evaluated at the recommended
velocity of 50mm/s when the RT station comes after the VI station [1]. Whereas, in the case
of PP from G1, VI followed by microwave technique (MWT) could be used to decrease the
inspection time in the RT station.
Figure 4.5: Inspection Times for RT station as per the G1, with long pipe 24384mm,
NOP=1 and various shift hours
59
Figure 4.6: Calibrate velocity as per the G1, with long pipes 15240mm, 18288mm,
21336mm and 24384mm, NOP=1 and different period shift hours
The velocity of the wagon utilized to move pipe through the testing stations controls
the time spent in each station. The TET is relative to the length of the pipe. For PP we get the
minimum test processing (time interval) for any station. The velocity can be mathematically
computed through displacement function to the velocity and PP time interval. Descending
velocity occurs for G1 as shown in Fig. 4.5, which leads to increased time interval. The
velocities of the G1 PP inspection decrease, because quality control for the pipe weld can
then be given as the indication for decreasing the LO and LE.
When using G1 to calibrate the RT wagon velocity, the recommended velocity
decreases as demonstrated in Fig. 4.6, for different proposed shift periods and different pipe
lengths.
60
4.3.2 Time and Speed analysis for inspect PP with in VI, MW and RT
The microwave (MW) station has been proposed to work with G1 pipe. It is expected
that the MW station located after VI will increase both the LO and the LE of the PP to be
inspected by the RT station. With the MW station working, the partial test for a PP in Fig. 4.4
is not needed. As shown in Fig. 4.7, PP ranging from 30 to 50 pieces for each station (VI,
MWT & RT) per shift were suggested for different cases as a LO, and improvements in the
LE were obtained automatically. The TETs in the MW station are 2.5min, 3min, 3.5min and
4min with the tests applicable to the PP lengths 15240mm, 18288mm, 21336mm and
24384mm respectively. These lengths were evaluated with MWT and each length was run
with different shift hours in order to analyze the MWT efficiency at the recommended
velocity (50mm/s) for the RT station.
The computed results of the maximum LO for each pipe length and shift hours 8, 9,
10, 11 and 12hrs are shown in Fig. 4.8. It should be noted that we do not consider random
events or calibration errors, or necessary repairs to broken testing machinery that occur at
work and increase the time interval in the RT station beyond the standard time allowed. For
our analysis the recommended velocity of PP processing is assumed to be constant.
Figure 4.7: Line efficiency result of microwave station as per the G1, with various
long pipe, and NOP=1
61
Figure 4.8: Maximum LO results after microwave station add it to inspect pipes G1,
with different length pipe and shift hours
4.3.3 Assess the Ratio of work done Through Proposed MW Technology
To evaluate the defects qualitatively and quantitatively using VI for G1, the
microwave technique (MWT) was used as the input factor of the LO and LE theoretical
analysis, in order to complete the inspection for the defect locations quantitatively. SAM is
factored into the proposed method. During the analysis, the exact times to inspect the OD and
ID by the inspector were factored into LE calculations. Pipes are inspected in VI and
according to the exact time, the ratio of work done was calculated, to be 26.667% work done
by VI for G1. Hence, the rest of the work ratio is 73.334% done by MWT. Finally, the
percentage of work completed by the microwave (MW) station is a high, and this is
confirmed in Fig. 4.7 above. MWT kept the velocity of the wagon which carries the pipe
during the testing process in the RT station at the recommended velocity according to the
standards and codes, as mentioned in [1]. For G2, work done ratios are 100% for VI and RT
individually.
62
4.4 Results and Discussion
The new generation of line output and efficiency has been managed in view of the
complexity of the nondestructive evaluation sequence integrating the latest developments of
microwave technology-based inspection systems. MW can be used to control the declining
pipe process velocity through the RT station, and to overcome the deficiencies in the
aforementioned VI method for G1 pipes.
Fig. 4.4 shows the values of LE for VI station per the characteristics of standardized
evaluation of the pipes in this station. The values are divided into two parts, as shown in the
graph, partial test and full test according to the diameter groups, G1 & G2. In G1, the VI has
a physical limitation to test the pipes related to pipeline specifications. Such as dimension,
pipe formation, pipe material and purpose of the pipe. But those inspections must be made
because the specifications reflect standard codes. These codes specify to what level pipes are
examined in terms of acceptance of sizes of defects (whether a particular size of defect size is
acceptable or not) that exist in the body of the pipe or in welded areas. Naturally, the testing
standards for oil and gas pipeline vary from, for example a chemical pipeline, during the
manufacturing phase. Let us suggest without arguing that the efficiency and production are
based on pipe inspection time. And by analyzing the SAM, the LE is controlled by TMP
which changes according to the actual time expended to test the pipe. In G2, the pipes are
inspected both outside and inside, and thus satisfied the full test. But when we compute the
LE for G1 which is represented only by the partial test, the LE declines. The reason is TMP
includes only a partial test for VI which means less time spent to inspect only the outside
diameter, the time varies and TMA is constant, as we mentioned above. This is because it
depends on the number of operators which is constant for each station according to the
requirements, and the length of the work shift is constant according to the production time.
Because of the partial test in the VI station, the RT station must reduce the velocity of
the wagon that carries the pipe during the test from 50mm/s to 16.667mm/s, as shown in Fig.
4.5. LO for pipe length 24384mm falls from 60 to 18 pieces, and from 89 to 27 pieces per
shift worked 8hrs and 12hrs, respectively. Similarly, Fig. 4.6 shows that reducing the
inspection velocity to keep up with quality control requirements causes the LO to decline as
shown in Table 4.1 below.
63
Table 4.1: Summary of the reduced velocity per quality control during different shift periods
Pipe Length (mm)
24384
21336
18288
15240
Maximum
Minimum Production Period Shift (hr)
Production (Pieces)
(Pieces)
89
27
12
60
18
8
102
31
12
68
21
8
119
36
12
79
24
8
142
44
12
95
29
8
Fig. 4.6 and the Table 4.1 above show that the shorter the length of the pipe, the more
pieces are produced within a shift.
Fig. 4.7 shows the increase in LE as per LO associated with using the microwave
station, where improved TET were achieved. More specifically, the curves in Fig. 4.7
demonstrate that the TET’s (2.5min, 3min, 3.5min and 4min) are directly correlated to the
length of pipe (15240mm, 18288mm, 21336mm and 24384mm) respectively.
Finally, Fig. 4.8 shows the measured maximum LO related to the various shift lengths
(8hrs, 9hrs, 10hrs, 11hrs and 12hrs) obtained for four PP lengths as mentioned above.
Velocity for the RT wagon improved to 50mm/s for all four PP pipe lengths as shown in Fig.
4.8 with time intervals 487.68s, 426.72s, 365.76s and 304.8s, starting from the longer to
shorter PP's. Thus, the final result is that pipe velocity in the RT station was maintained at the
recommended velocity for all pipe lengths.
64
4.5 Conclusion
To analyze the line output and line efficiency of a pipe manufacturing plant at
arbitrary and random production capacity due to a partial VI test, we propose to add a
microwave station for a particular pipe diameter for the analysis of the PP.
Microwave technology can be situated in diverse production system line
configurations in manufacturing plants per the stakeholder and customer needs. In our study,
microwave kept the recommended velocity of the RT inspection process synchronized for the
PP case, and handled a 73.334% as a ratio of work done for sequence system in the plant, and
covered the shortcoming of the VI.
Also MWT kept the pipes at a high level of quality assurance. Because if it is
compared to the previous status, before and after the microwave technique, significant
changes may be observed by tracking curves as mentioned above. Even with reduction in
velocities of the wagon carrying the pipe out to be examined during the RT station. As
mentioned in section 2, it is noted that the inspection is only for the weld areas.
Consequently, if there is any defect in the inner wall of the pipe body, it will not be detected.
Categorically, the microwave is shown to be a promising technology since it allows the RT
station to keep the velocities as recommended. It can detect defects in both the weld and the
body of the pipes easily after calibration and without human effort, thus preserving the
quality of the product and is expected to enhance production efficiency for oil and natural gas
pipe production.
65
Chapter 5
NDT Applied to the Detection of Defects in Oil and Gas Pipes: A
Simulation-Based Study [90]
Abstract
This research investigates application of microwave nondestructive testing (NDT) to oil and
gas pipe wall reductions (PWR) in manufacturing that are less than full-circumferential in
extent. Pipes were modeled using Computer Simulation Technology (CST) simulation
software, holding pipe length, wall thickness, depth of PWR and configuration constant. The
study looks at 32 models in order to determine sweeping frequency limitations for fullcircumferential,
half-circumferential,
three-quarter-circumferential
and
quarter-
circumferential extents of PWR with 8 widths ranging from 6.35mm to 50.8mm. The
limitations were determined to be between the extremes. The 6.35mm width did not cause a
resonance peak even with a full-circumferential PWR. The 50.8mm length produced a
detectible resonance peak, but it is within the range established by the six middle widths.
5.1 Introduction
Reduction of pipe wall thickness is among the most hazardous defects in the metal
pipes that are extensively used in the petroleum industry (oil and natural gas transportation,
for example), in many aspects of the chemical industry, and in fossil power plants [25] [32].
When pipes are in service it is very important to be able to detect and evaluate any
thickness reduction in the pipes. This is for purposes of safety as well as assessing the service
lifetime of the pipes [25]. Archives show several injuries because of the failure of installed
pipes in the years between 1996 and 2003 [1].
Much of the research that is done on Non-destructive Testing (NDT) techniques is
focused on locating and measuring pipe wall reduction (PWR) [25] [32] [75]. This is because
PWR is an important consideration in the field, where water condensing at the base of oil or
66
natural gas pipelines can lead to such corrosion [72]. The issue of detecting defects early in
the manufacturing of oil and natural gas pipeline is extremely important in the expansion of
the pipe industry to meet the main requirements of oil and natural gas transportation.
Earlier NDT series were designed and used to inspect the pipes; currently, improved
techniques are used. There are six major NDT technologies that are in current use for
determination of manufacturing defects. These include visual inspection, ultrasonic testing,
radiographic testing, digital-x-ray testing, magnetic particle testing and eddy current testing,
which are discussed in our previous study, and in other studies [1] [29] [62].
Most of these methods require human effort, which exposes the parts under test to
human error. Ultimately, the human error causes what amounts to wasted time and increased
expense, because the quality of the pipe must be assured through testing. Even though these
stations have shortcomings, we cannot dispense with them, as mentioned by Wissam
Alobaidi et al [1].
Wall thinning can occur during correction of defects in the manufacture of the metal
plate used to fabricate the pipe. The grinding process is a normal aspect of metal parts
manufacture (plates, pipes, bearings, shafts, for example). Thus, the material to be formed
into pipes may already have thinning present. The formation of the plate material into pipe
may introduce further defects that are grind down during visual inspection. The degree of
thinning must be carefully measured during manufacturing to be certain that it is within the
tolerances allowed by the intended customer [91] [92]. Ultrasonic, Radiographic and X-Ray
testing cannot economically detect the full extent of thinning [1].
In 2007 Yang Ju demonstrated a technique which allowed remote detection of wall
thinning reductions from 10% to 80% of the 1mm wall thickness. This method was
developed using extremely small inner diameter pipe (17mm). The measurement was
accomplished by analyzing the resonance frequencies of microwaves, using the pipe itself as
a circular wave guide [32].
Linsheng Liu et all in 2011 were able to use the time of flight of the microwave signal
to determine where the wall thinning was located longitudinally, but could not resolve the
67
reflected signals enough to determine the size of the defects because the starting and ending
points of the defect created overlapping peaks [34].
In 2012 Yasutomo Sakai et al, modeled simulated pipes in order to clarify the
relationship between signal peaks and the profiles of defects, as well as how microwave
signals act relative to the actual defects within pipes. Simulation modeling allowed the pipe
length and the flaw location to be known ahead of time [93].
Microwave technology is one of the promising techniques, and it has been recently
well researched. Until now, though, only full-circumferential samples have been used in
microwave NDT [25] [32] [93].
Our concern in this paper is with application of microwave NDT techniques to the
manufacturing side of the industry. However, it should be noted that in both the field and
manufacturing processes, three-quarter-circumferential, half-circumferential and quartercircumferential defects are found.
In this paper our scheme simulates microwaves that disseminate along the pipe. The
circular interior of the pipe acts as a waveguide. We have used the finite-difference time
domain commercial software package of Computer Simulation Technology (CST) to conduct
extensive simulation studies of full-circumferential, three quarter circumferential, halfcircumferential and quarter-circumferential models to both locate and measure the PWR. A
series of 32 perfect conductor pipes were modeled, with these variables held constant: the
length of the pipes is 508mm. The external diameter is 254mm. The pipe wall thickness is
12.7mm. The depth of the defect (PWR) is half the thickness of the pipe wall. The variable
we looked at in this study is the area of the PWR. There are 8 widths of PWR modeled:
6.35mm, 12.7mm, 19.05mm, 25.4mm, 31.75mm, 38.1mm, 44.45mm, and 50.8mm. For each
width
of
PWR,
four
extents
were
modeled:
Full-circumferential,
three-quarter
circumferential, half-circumferential and quarter-circumferential.
We expand the size of pipe modeled for this type of research according to the
standards of size used for oil and natural gas pipeline, the PWR width, and the percentage of
circumference involved in the PWR. The microwave signal is introduced at port 1, yielding
negative peaks based on the scattering parameter (S11), the sweeping frequency.
68
5.2 Materials and Methods
The simulations will be have been conducted a pipe model with full-, three-quarter-,
half- and quarter-circumferential PWR. Our CST simulation uses a pipe model with a single
port. Port 1 is excited by a TE11 mode at a given bandwidth, and short circuited at the other
end which provides a clear reflection from the end of the pipe. The pipe wall was modeled as
perfect electric conductor (PEC), and the inner volume was filled with air as the medium
within the pipe. Different widths of PWR were considered in the middle of the pipe: 6.35mm,
12.7mm, 19.05mm, 25.4mm, 31.75mm, 38.1mm, 44.45mm, and 50.8mm as listed in Table
5.1 below. The pipe outer radius is 127mm; the inner radius is 114.3mm. The wall thickness
is 12.7mm. The wall reduction is 6.35mm in depth. The total length of the modeled pipe (LP)
is 508mm. For a pipe with an inner radius of 114.3mm, the cutoff frequency starts at 0.77
GHz to ensure the TE11 mode, the sweeping frequency start at 0.7 GHz. The same
simulations were also performed on the three-quarter-, half- and quarter-circumferential
PWR models, with the experiments being guided by the results from the prior fullcircumferential simulation studies [93]. Fig. 5.1(a) illustrates one of our modeled
circumferential wall reductions (Quarter Circumferential PWR). Fig. 5.1(b) is a detail with
dimensions.
Figure 5.1(a): Cross-section illustrates quarter-circumferential PWR in pipe model
with single port and open end
69
Figure 5.1(b): (Detail) quarter-circumferential PWR dimensions. Depth of defect =
6.35mm. Thickness of pipe wall = 12.7mm. PWR width = 25.4mm
Table 5.1: Wall Reduction Formula and Resulting Widths
Model Number
PWR Ratio (Lp/10) % PWR Width (mm)
1
12.5
6.35
2
25
12.7
3
37.5
19.05
4
50
25.4
5
62.5
31.75
6
75
38.1
7
87.5
44.45
8
100
50.8
5.3 Results and Discussion
One limitation of the full-circumferential model is that the simulations were unable to
detect widths of PWR that are smaller than 6.35mm. However, for 12.7mm, 19.05mm,
25.4mm, and 31.75mm widths, the simulations produced sharp negative peaks in the
reflected signal.
70
When the magnitudes of the scattering parameter minimum (S11 – which means the
signal was introduced at port 1 and measured at port 1) are plotted against the frequency, the
reflection remains at zero if there are no discontinuities. The presence of PWR generates a
reflection which is measured by the S11, which reveals information about the defects.
Figure 5.2: Geometry of full-circumferential PWR model
Figure 5.3: Magnitude of S11 relative to sweeping frequency for each PWR width in
full-circumferential model
71
When defects are implemented in our simulations, as shown in Fig. 5.2, which shows
the pipe cross section for the simulation with full circumferential PWR, we observed
reflections generated by microwaves in the range between 4.0 GHz and 4.25 GHz as shown
in Fig. 5.3, which shows the negative reflections generated, for PWR with widths of 12.7mm,
19.05mm, 25.4mm, and 31.75mm. The reflections were observed for PWR starting at
12.7mm, moreover, the peak obviously decreased in depth (meaning that reflection
increased) for larger widths of PWR. For the 38.1mm width, the peaks begin to diminish,
with 44.45mm having a further diminished peak. For 50.8mm there is only a very small peak.
Fig. 5.4 shows the resonance frequency (frequency at which negative peak reflection
is observed) clearly for eight curves, each of which represents one PWR width modeled. We
show the frequency range from 4.1 GHz to 4.15 GHz.
Figure 5.4: Detail of sweeping frequency for the full-circumferential model
As shown in Fig. 5.4 the PWR width 6.35mm yields zero peaks, the lower limitation
for our research. The negative peak first appears for PWR width 12.7mm; and as PWR width
increases the negative peaks for the sweeping frequencies begin to decrease (reflection is
increased), and to shift from approximately 4.15 GHz toward 4.1 GHz. The peaks indicate
maximum reflection at PWR width 31.75mm, and as we move toward 50.8mm the negative
peaks rise to almost zero. The simulations revealed that as the width of the PWR increases,
72
the peak of S11 increases (reflection decreases) this gives an indication of the size of the
PWR.
Figure 5.5: Geometry of half-circumferential PWR model
Figure 5.6: Magnitude of S11 relative to sweeping frequency for each PWR width in
half-circumferential model
Fig. 5.5 shows our half-circumferential model, which generates the peaks shown in
Fig. 5.6 and Fig. 5.7. The reflection of the half-circumferential PWR defect produces graphs
similar to those from the full-circumferential model. That is because the PWR is symmetrical
73
in both cases and gave us symmetrical shape reflection, full PWR and half PWR (equal
amounts of the circumference have PWR, and have no PWR). But the half-circumferential
case shows an increase in the resonance frequency (decrease in reflection) for each curve,
when the two cases are compared. This is expected, because in the case of the halfcircumferential pipe, half of the waves introduced into the waveguide contact no PWR,
instead propagating along the defect-free half of the inner pipe wall.
Figure 5.7: Detail of sweeping frequency for the half-circumferential model
Figure 5.8: Geometry of three-quarter-circumferential PWR model
74
In Fig. 5.8 we modeled the pipe with three-quarter-circumferential PWR; the
reflection graph appears different from those of the full-circumferential PWR and halfcircumferential PWR pipes. In Fig. 5.9 and Fig. 5.10 we see different peak shapes, and we
can explain these differences by looking to the surface area of the PWR. The surface area of
the defect is circumferentially greater than the respective defect-free surface area (a quarter
has no defect while three-quarters has a defect).
Figure 5.9: Magnitude of S11 relative to sweeping frequency for each PWR width in
three-quarter-circumferential model
Figure 5.10: Detail of sweeping frequency for the three-quarter-circumferential model
75
In Fig. 5.11 we show the pipe modeled with a quarter-circumferential defect. The
graphs in Fig. 5.12 and Fig. 5.13 show us the similarity of these curves to those we see in the
figures for the three-quarter-circumferential PWR. Once again, this is the plotted data for
increase in S11 (reflection decreased).
Figure 5.11: Geometry of quarter-circumferential PWR model
Figure 5.12: Magnitude of S11 relative to sweeping frequency for each PWR width
in quarter-circumferential model
76
Figure 5.13: Detail of sweeping frequency for the quarter-circumferential model
Figure 5.14: Scattering parameter at Port-1 and resonance frequency for each PWR width
Note that the reflections in Fig. 5.12 and Fig. 5.13 are small (scattering parameter less
than -0.5), which means that it is hard to detect. Compare this to Fig. 5.14, which shows the
much larger reflection for a full-circumferential defect. But still if we consider our calibration
pipe as full-circumferential, and then decrease the length to quarter-circumferential, we still
find a small peak for the reflected signal. This indicates the presence of a defect.
77
Fig. 5.14 is a plot of the resonance peaks for the full- and the three partialcircumferential PWR models. The plotted points represent the relationship of the S11
parameter to resonance frequency for each of the PWR widths (related to the area of PWR).
Note that plotting with increasing S11 scattering yields a smooth curve, but causes the
frequencies to appear out of numerical order. This is because the peak reflection of each
PWR width occurs within a limited frequency range, and the actual peak frequency moves
non-uniformly from one PWR width to the next.
We can see that it is easy to detect the full-circumferential defect. And if we consider
that length as a calibration case, it is clear that the partial circumferential defects can be
detected in relation to the calibration case. It is worth mentioning that if the depth of defect
decreases or increases for the three partial circumferential cases the defect may show an
increased reflection.
Figure 5.15: Limitation of PWR width for full-circumferential model per
resonance frequency
Fig. 5.15 is a plot of the PWR width against the resonance frequencies for the fullcircumferential model. The widths from 12.7mm to 44.45mm show exact results (new
limitation), representing the sensitivity limits for accurate detection of the PWR width. These
78
peaks set the frequency range limits for this technique. The point for the 50.8mm width falls
within the range set by the other widths, thus, the 50.8mm resonance peak is detectable, but it
is outside the limitations for this technique. The 6.35mm width shows a zero reflection in
Fig. 5.14, thus it was not plotted in Fig. 5.15.
Fig. 5.14 shows the limitations of the length of the circumferential defect. And Fig.
5.15 shows the limitations of the width of the defect.
5.4 Conclusions
These studies have established the limitations of the PWR widths that can be detected
within this range of sweeping frequencies for a pipe of 114.3mm inner radius. For this study,
the limitations of the PWR widths are: (i) 508mm/40 = 12.7mm minimum and 508mm/10 =
50.8mm maximum, for full circumferential PWR. Hint: the length of the pipe is 508mm. (ii)
508mm/26.667 = 19.05mm minimum and 508mm/10 = 50.8mm maximum, for three-quartercircumferential and half circumferential PWRs. (iii) 508mm/20 = 25.4mm minimum and
508mm/11.428 = 44.45mm maximum, for quarter-circumferential PWR. The fullcircumferential defect is easy to detect and can be used as a reference point for the partial
circumferential defects. We are planning work on larger sizes of pipes, in order to determine
the limitations of PWR width and depth (degree of discontinuities) detectable within those
inner radii.
79
Chapter 6
Localized Surface Plasmon-like Resonance Generated by Microwave
Electromagnetic Waves in Pipe Defects [94]
Abstract
Localized surface plasmon (LSP)-like resonance phenomena were simulated in COMSOL
Multiphysics™, and the electric field enhancement was evaluated in eight pipe defects using
the microwave band from 1.80GHz to 3.00GHz and analyzed by finite element analysis
(FEA). The simulation was carried out, in each defect case, on a pipe that has 762mm length
and 152.4mm inner diameter, and 12.7mm pipe wall thickness. Defects were positioned in
the middle of the pipe and were named as follows; SD: Square Defect, FCD: Fillet Corner
Defect, FD: Fillet Defect, HCD: Half Circle Defect, TCD: Triangle Corner Defect, TD:
Triangle Defect, ZD: Zigzag Defect, GD: Gear Defect. The LSP electric field, and scattering
parametric (S21, and S11) waves were evaluated in all cases and found to be strongly
dependent on the size and the shape of the defect rather than the pipe and or the medium
materials.
6.1 Introduction
Surface plasmons (SP) are a very well-known phenomenon that occurs when
electromagnetic waves propagate through a metal-dielectric interface. At resonance
frequency, metal surface electrons start oscillating vigorously, generating an intense new
localized electric field, usually known as confined or trapped electromagnetic fields [95]
[96] [97] [98] [99].
According to H. Reather [100], the electromagnetic radiation propagates over the
metal surface with multiple Eigen frequencies ranging from  =  to  =  /√ depending
mainly on wave vector k. Surface plasmons have longer wave vector than the propagated
electromagnetic light waves of the same energy. Therefore, surface plasmons describe the
80
fluctuation of surface electron density which decreases exponentially in space vertical to the
metal surface.
Propagation of an electromagnetic wave through a surface of metal usually associated with
interaction with the outer metal electrons such a phenomenon resembles the behavior of
plasma with dielectric function Ɛ=1-(p2/2), [100].
Dielectric function is negative below p. Therefore, metal surface starts supporting
the oscillation of the outer electrons and maintains small electric field that called surface
plasmons [100]. There are many applications of surface plasmon such as Surface enhanced
Raman spectroscopy (SERS) [101], Raman signal might amplified by surface plasmons to
106-1012 [102]. Surface plasmon polariton waveguides were inspected previously in band gap
structures [103], metallic strips [104], optics [105], and many others [106]. Metals with
different dielectric functions can provide different electromagnetic fields, perfect conductors
are exception. However, even perfect conductors can produce surface plasmons when an
array of holes created in the surface. The only valid rule to induce these electric fields
(surface plasmons) is that these geometric holes must be smaller than the wavelength of the
propagated electromagnetic wave. Generally, electric fields generated from structured
surface were usually associated with microwave frequencies [107].
In UV-Visible frequencies (terahertz; THz), LSPs are easily formed with metals that
have negative values of relative permittivity. However, in microwave frequency bands
(gigahertz; GHz), the metal becomes a perfect electric conductor, and all surface electrons
are highly delocalized. This causes conditions that do not support formation of the LSP
phenomenon. Recently, it has been reported that decorating the metal surface with special
grooves that have dimensions less than one-quarter of the wavelength (subwavelength) of the
incident wave does strongly localize the electrons, and generates an LSP-like resonance
[108] [109].
Therefore, choosing a specific metal has no effect on producing a particular LSP-like
response in microwave other than the shape and dimensions of grooves on the metal surface.
Radar engineers demonstrated long ago that if the metal surface has a specific design such as
a vertical narrow groove, it will generate resonance with incident radiation, transforming the
81
short circuit at the closed side to an open circuit on the open side. Under these circumstances,
metal surface impedance changes to vast and imaginary values, and these conditions
completely block electromagnetic wave propagation, and reflect incident radiation without
phase shift [110].
In this work, Finite element numerical analysis (FEA) was used to model and
simulate the microwave band (1.80GHz to 3.00GHz) (wavelength 166.55mm - 100mm)
propagation through an oil pipe to detect surface defects via LSP like resonance behavior.
Different shapes of pipe defects were modeled to fulfill the groove (defect) dimensions that
are required for generating very active LSP like resonance using a microwave signal.
Furthermore, electric field was evaluated for each kind of defect.
If the techniques in this work are applicable as non-destructive testing (NDT), they
could save much time and expense in locating and characterizing pipe wall thinning (PWT)
defects, which are common, especially in the pipe manufacturing process. The importance of
NDT methods for pipe manufacture and maintenance cannot be overstated. Many pipes are
of sufficient diameter to be examined by physically entering the pipe section for visual
inspection. But many pipes are also too small for a person to get inside. In those cases some
method is required for determining whether defects in the inner surface are present, and how
severe they are, as well as determining the location of those defects. Microwave propagation
overcomes the small diameter problem and reduces the time and cost of quality assurance
operations. Swift and accurate identification and characterization of PWT could mean the
difference between needlessly discarding a salvageable section of pipe, and performing the
corrective actions which would make the pipe serviceable, and within specifications [30].
6.2 Model Setup and Analysis
Cylindrical waveguides used in many processes, control the flow of circularly
polarized waves. Pipes, especially metal ones, can function as circular waveguides. The
circular waveguide has some remarkable modes, such as TE01 modes, which are useful
because diminution of the signal is very low at high frequencies [111].
82
This research uses Finite Element Analysis in order to evaluate LSP for several PWT
profiles. The waveguide is modeled with two ports, and an inner diameter of 152.4mm, that
yields a cut-off frequency (1.154GHz at fTE11) as shown in Equation (6.1) [39].
It is worth noting that the physical parameters of the pipe under test (PUT) directly
determine the cutoff frequency, which is the frequency at which the signal begins to
attenuate, or to be reflected within the waveguide [39] [111].
 =
 

(6.1)
Where Cd is the speed of light in the pipe under test, Pnm represents two mathematical roots
of the first-kind Bessel function as shown in equation (6.2). Di is the inner waveguide
diameter. The values of n and m are strictly related to the transverse electrical and magnetic
modes of wave propagation. The waveguide works with bandwidth 1.30GHz as shown below
and expressed in Ref. [111].
To insure inclusion of this bandwidth, sweeping frequency range starts at 1.80GHz to
3.00GHz.
ℎ =

(5.1356 − 3.0542)

(6.2)
One of the problems with using microwave technology to evaluate PWT is that the
techniques developed so far cannot accurately determine the location, but could evaluate the
width and length of the thinning area (what percentage of the circumference is affected by
the PWT case, and can we detect it?) [90] [112] [113]. This leaves the field open to testing of
new ways to make use of microwave signals, including new effects, such as LSP-like
phenomena, which may be related to a type or feature of PWT. For example, if an LSP-like
signal could indicate whether the disclosed defect was triangular or semi-circular in crosssection, it would be a step forward in analysis of PWT by microwaves. In general, LSP is
thought less likely to occur at the frequencies in the microwave band. Yet some recently
discovered effects suggest that with the right preparation of the pipe under test, LSP-like
conditions could exist [26] [90].
83
Eight kinds of pipe defects were modeled and tested to assess the formation and
strength of LSP-like resonance frequencies, see Fig. 6.1. Transverse magnetic mode (TM)
was polarized parallel to the xz-plane, while Electrical mode (EM) was polarized with
incident angle θ on the surface of the metal (defects site). All modeling was performed using
COMSOL Multiphysics™, the pipe metal surface was chosen to be pure copper (perfect
electric conductor, PEC) and the medium in which the wave propagates, was air. Both ends
of the pipe were set up to be inlet and outlet wave ports. Finite elements meshes were chosen
in each case, so there is no difference in effect on the results, see Fig. 6.2.
Figure 6.1: Cross section schematic of eight pipe defects; in each type of defect
(Wdefect = 25.4mm) and (DD = 6.35mm)
84
Figure 6.2: Oil pipe model setup. Tetrahedral mesh technique was used for finite
elements LSP analysis. (a) 3D model setup, and (b) 2D model setup. Pipe inner diameter
is 152.4mm, and outer diameter is 177.8mm, pipe thickness is 127mm, and pipe length
is 762mm, with defect site in the middle of the pipe, i.e. 381mm from wave inlet port
6.3 Results
Fig. 6.3 shows the results of FEA for eight different shapes of defects. The LSP-like
electric field, and the resonance frequency of each defect was evaluated at a specific point
(circled point) in each case.
Maximum electric field at LSP points were plotted against sweeping frequency as
shown in Fig. 6.4 and the calculated values were tabulated in Table 6.1 for each defect case.
The coincidence of resonance frequencies seen in Table 6.1 and Fig. 6.4 can be
explained because of the way in which we modeled the defects used in the experiment. The
width, length, depth and placement of the various defects are the same. The volume varies
only as a result of slight changes in shape, the contour of the defect.
85
Figure 6.3: LSP-like and electric field at defect site during resonance frequency
for eight different defect shapes. Circles indicate points evaluated. Also in each
case, the electromagnetic wave disruption pattern is shown
86
Figure 6.4: Electric field for each defect type corresponding to the
resonance frequency
Table 6.1: Maximum electric field at evaluation points as shown in Fig. 6.3 and the
resonance frequencies in each case, arranged from maximum to minimum
Resonance frequency
LSP
f (GHz)
Field (V/m)
HCD
2.77
857.70
FCD
2.505
499.59
SD
2.505
423.73
TCD
2.2
349.49
TD
2.2
316.28
FD
2.505
288.70
ZD
2.775
280.50
GD
2.51
189.12
Defects Type
Maximum
Electric
87
Thus, because volume is roughly equal for the SD, FCD, and FD defects, they
generate resonance frequencies at the same point in the frequency sweep. Volume of defect
controls the resonance frequency. The defects with greater variation in contour have
resonance frequencies at different points in the sweep, because they have greater variances in
volume.
Figure 6.5: Graphs of resonance frequency for the scattering parameter
S11, for four types of defect
According to previous work, the volume and location of the defect controls the
resonance frequency, including its signal strength. For TCD, TD, ZD and GD that range,
from -0.5 to -1.4 dB, is considered shallow when compared to the signal strength for the
other control defects [113] [114]. Not all previous studies of this type of investigation have
differentiated between S11 and S21 peaks in their reported results. But Linsheng Liu has
88
demonstrated differentiation in S21 peak strengths in experiments using smaller diameter
pipes [37].
Figure 6.6: Scattering parameter S21 with resonance frequency for each of
four profiles of defect
In this case, the scattering parameter represents the signal reflected from the defect.
We can distinguish the S11 for example, from the reflection from any groove (PWT)
according to the volume created by the parameters of the PWT. As Pendry mentioned, the
structural hole causes a resonance frequency [108]. Our defect is a hole, so it changes the
surface structure within the pipe, and the resultant scattering parameter identifies the
presence of a defect in the pipe wall. This is the link between Pendry and us.
89
6.4 Discussion
LSP phenomenon was explained in detail using Drude’s model [115] [116], which
assumes that the metal is a gas of electrons, and frequency is a function of metal relative
permittivity (m):
2
2 
 = 1 −
+
1 + 2 2
(1 + 2  2)
(6.3)
where p is the plasma frequency,  is the relaxation time, and  is the angular frequency.
However, the propagation constant of LSP kLSP can be estimated from Maxwell’s equations
according to the following equation.
 = 0 √
 
 + 
(6.4)
where εd is the relative permittivity of the dielectric medium, and k0 is the wavevector (/c).
However, at microwave frequencies ( << p), τ << 1. The metal relative permittivity
values become large and imaginary. Under these conditions equation (6.4) can be re-write
into.
 ≈ 0
(6.5)
According to equation (6.4) LSP-like effects can form using microwave frequency since the
difference between kLSP and k0 is tiny.
Our results were in complete agreement with what was recently reported: that
decorating the metal surface with special grooves that have a dimension less than one-quarter
of the wavelength (subwavelength) of the incident wave would strongly localize the electrons
and generate LSP-like resonance [108] [109]. According to the results, it has been clearly
shown that LSP is formed in each type of oil pipe defect irradiated with microwaves that
have a wavelength four times longer than the defect length. Under these subwavelength
conditions, LSP formed at the sharp edges of the defect, producing electric field
enhancement that varied from 189.12-857.70 (V/m) according to the size and the geometry
of the defect, as shown in Fig. 6.4 and Table 6.1.
90
Formation of LSP at the defect site, in each case, caused a huge disruption to the
propagated wave through the pipe, which in turn produced intense S21 and S11 waves that can
be measured to specify the resonance frequency, see Figs. 6.5 and 6.6.
Results also show that GD and ZD have relatively low-intensity LSP compared with
the other types of defects, which is clearly due to the small size of the defect (around 5.08mm
to 12.7mm width) when compared with other cases that have a full 25.4mm width defect.
6.5 Conclusion
FEA modeling and simulation were used to prove our hypothesis that the size and the
shape of a defect is the main parameter in formation of LSP. Several pipe defects were tested
using a narrow band of microwave frequency 1.80GHz to 3.00GHz. In each case the
propagated incident waves were disrupted strongly via LSP at the defect size. Simulation was
performed on one kind of oil pipe, used for each case, that has specific dimensions, and the
defect location was at the center of the pipe. S11, S21 waves and electric field energy were
estimated for each defect. Defect geometry and size is the main parameter in generating and
holding an intense LSP, rather than the metal materials.
91
Chapter 7
Detection of Defects in Spiral/Helical Pipes Using RF Technology [112]
Abstract
Because pipelines are important to commerce, and most pipes used for transportation of large
volumes of liquids are made with spiral welds, there is some concern that the welds
themselves may reduce the usefulness of microwave signals as a way to nondestructively test
the pipe for pipe wall thinning (PWT). Computer Simulation Technology (CST) software is
used to research detection of defects with microwave signals. Smooth wall and spiral welded
pipes are modeled and compared. Both pipes are composed of aluminum; 1016 mm in
length; outer diameter 203.2 mm; inner diameter 152.4 mm; wall thickness 25.4 mm. To both
types we add fully circumferential wall-thinning defects, depth 19.05 mm. The thinning
defects have three profiles: rectangular, toroidal, and with one rounded and one right-angle
corner. The PWT width is 31.75 mm in all cases. The microwave testing is carried out with
sweeping frequency from 1.507 GHz to 2.4 GHz. The results of modeling discredit one
original hypothesis concerning the effect of the weld bead on microwave behavior. A
correlation is seen between PWT volume and the point at which the peak resonance
frequency (PRF) occurs for each of the three PWT profiles. A correlation is also found
between the PRF and the solid volume of the weld bead. The effect of number of coils in the
helix on PRF is also examined (6, 12, 24 and 48 coils), and is shown to correlate highly with
the weld bead volume in its effect. This research establishes the usefulness of pattern
recognition algorithms to analyze unknown pipes, and identify characteristics, such as
presence and extent of PWT to allow exact discretization of an unknown case.
7.1 Introduction
Piping systems are a major part of the infrastructure in many industries. Because
defects can cause pipe failure, maintenance of the pipe requires periodic inspection in order
to locate potential defects. Nondestructive testing techniques are necessary to carry out such
92
maintenance inspections. Techniques used for this task include ultrasonic and radiographic
testing, but such techniques are labor intensive [117].
Rolled riveted metal pipes were first manufactured in the United States for
transportation of water, beginning in the 1850s. By the 1930s lock-bar pipe, which used no
rivets, was the most commonly used metal water pipe. Welding of the seams of rolled pipe
began to predominate by the early 1930s due to automatic welding techniques and greater
economy of manufacturing [118].
A technique of forming welded pipe that was known for gun manufacture since the
middle ages was used to manufacture pipe on an industrial scale in the U.S. in the 1880s.
There are many advantages to the spiral-weld method. (i) The same width stock can be
adapted to build pipes of various lengths and diameters. (ii) Creation of large-diameter pipe
is much easier, and requires the same number of seams as small diameters. (iii) Thin-walled
pipe becomes a manageable possibility. (iv) There is an overall reduction in the investment
required to set up a pipe mill. But the technology to produce spiral-weld pipe suitable for
high-pressure uses was developed only in the 1960s [119].
A method for manufacturing helically formed and welded pipe was patented in 1940.
Initial diameters ranged from 100 mm to 875 mm. After World War II, German machines
were imported to the U.S., and these enabled construction of pipes with diameters as great as
3600 mm. The specifications and data for noteworthy steel pipelines have been published by
Cates (1971) and Hinds (1954). The specifications for many other pipelines have been
standardized by organizations such as the American Society for Nondestructive Testing
(ASNT), and the American Petroleum Institute (API,) and published in journals, such as
AWWA [118].
Beginning in the 1960s metal pipes have been used under pressure in a broad range of
industries, including the transportation of oil and natural gas. Certain types of defects can
lead to burst pipes due to compromised pressure load bearing capability. As the in-service
periods of pipes now stretch to several decades, nondestructive means of testing the installed
pipe have been developed. There are two aspects of defects to determine: first is the degree
of the defect, its physical size. Second is the location of the defect, where it is in the pipe
93
[38]. It may be a surface defect, or it may be subsurface. During the manufacturing process,
quality assurance procedures using nondestructive testing are employed to locate such
potentially fatal defects before installation. Most often ultrasonic techniques are used, often
in tandem with other nondestructive techniques [1] [29].
In order to locate defects in spiral/helical pipe, a remote sensing technique would be
advantageous after visual inspection of the outer surfaces, especially in pipes with smaller
inner diameters. A technique that is sufficiently rapid could also be scaled for use in larger
diameter pipes. The speed of the simulated technique would enable swift approximate
location of defects in a helical pipe in the longitudinal dimension. Once the defect was
located between two coils of the helical weld, the indications would be released to the next
sequence of nondestructive testing.
Use of microwave signals to locate defects has been researched for basic techniques
[38] [90]. No papers currently report testing on real-world geometries. For example, the
practical and simulated use of microwaves to detect presence of defects, and location of
defects have used very regular, smooth wall pipes of small diameter (for example, 17mm)
and to detect wall thinning or thickening [35] [38]. The defect has most frequently been full
circumferential. The research shows that the technique should be viable. Our work results
from simulations of helical welds in spiral welded pipe remotely sensed by microwave
propagation from port one of a 152.4mm inner diameter pipe.
Recently, spiral welded pipes have become more widely used in industry [30] [120].
The existing research done on use of microwave technology to detect defects on the inner
wall of pipes has been done with smooth pipe acting as the waveguide [35] [38] [90].
Calibration of the effects that a spiral weld may have on the reflected waveform within a
spiral welded waveguide, has not been performed, until now.
To determine the effect of real-world geometry on the proven techniques, we use
Computer Simulation Technology (CST) to research microwave defect detection. We model
and compare two types of pipe: smooth wall and spiral welded. The common specifications
of both pipes are: composed of aluminum; length 1016 mm; inner diameter 152.4 mm; wall
thickness 25.4 mm. To both types we add fully circumferential wall-thinning defects, depth
94
19.05 mm. The thinning defects will have three profiles: rectangular, toroidal, one rounded
and one right-angle corner. The defect width is 31.75 mm in all cases.
We see two significant effects: a presence of noise in the reflections in the spiral
waveguide, and a change in dB levels between the smooth and spiral pipes. The waveform
contour remains the same. The amplitude of peaks is only slightly affected.
The simulation incorporates the fact that the inner surface of the pipe body is regular,
while the welds themselves are discontinuous, but evenly spaced. However, the microwaves
do not return regular peaks for each coil of the helical weld, which was expected before the
investigation. As it turns out, the microwaves do not sense the presence of the weld at all.
There is a semi-regular waveform in a defect-free pipe with helical weld, but this is mere
noise, not sensing of the weld itself. Instead, actual defects which appear along the weld, or
in the pipe body, return a resonance peak, indicating a surface defect (PWT) of the body
material. The only notable change caused by the presence of the weld is a shifting of the peak
maximum to a slightly higher frequency when compared to the peak maximum for an
identical defect modeled in smooth pipe.
7.2 Modeling and Numerical Simulation
For smooth pipe and spiral welded pipe two mathematical Bessel function roots are
set up as (P01=2.4048, P11=3.8317). The lowest cut-off frequency will be (fcTE11 = 1.154
GHz) at mode (TE11), so the frequency bandwidth (0.893 GHz) for these waveguides will be
presented in the equations below [39] [111].
01 =
2.4048

1.8412 11
(7.1)
11 =
3.8317

1.8412 11
(7.2)
95
7.2.1 Reference Waveguide
7.2.1.1 Smooth Pipe (Waveguide)
Pipes with smooth inner walls are used for the transportation of clean water (in
homes, schools, clinics), and for transport of wastewater to treatment plants. One type of
smooth pipe construction is done with high-frequency welding (HFW), and the other types is
done with Electrical Resistance Welding (ERW) [30]. This kind of pipe is used for cooling
systems in nuclear power plants. Such pipes develop major problems with wall thinning
when in service for extended periods [35] [38] [90].
We modeled pipe with thickness 25.4mm, and inner diameter 152.4mm, free of
defects and evacuated, to generate a standard waveform for comparison to the waveforms
generated by testing unknown cases (as shown in Fig. 7.2.). In Fig. 7.1. we see a cross
section of such a pipe.
Figure 7.1: Three dimensional cross-section
Figure 7.2: Reference waveforms for microwave
of waveguide (reference pipe) free of
reflections in smooth pipe (waveguide). The
defects
scattering parameter (S21) is at 0 dB, and scattering
parameter (S11) is at approximately -95 dB. The
level waveforms tell us that there is no defect in
the inner surface of the waveguide (pipe)
96
7. 2.1.2 Pipe with Spiral Weld
Spiral (helical) welded pipes are used for transporting of petroleum products (oil and
natural gas). Spiral welded pipe construction is done with electrical welding. This kind of
pipe has inner diameters (ID) and outer diameters (OD) that range from smaller to larger in
various usages [30]. This type of construction is employed in making pipes used as gathering
pipelines (101.6mm to 304.8mm, OD), feeder pipelines (from the minimum to the maximum
standard OD), transmission pipelines (101.6mm to 1212mm, OD), and distribution pipelines
(12.7mm to 152.4mm, OD) [121].
The problem in quality assurance for manufacturing spiral welded pipe, as the author
experienced, occurs for the visual inspection of the pipe inner surface prior to radiographic
testing, for pipes with smaller inner diameters between 254mm and 762mm [30]. Some
defects removed by grinding the unformed plate metal before it is formed into a pipe, result
in PWT that cannot be seen in these smaller pipes [90] [91] [92]. So, we propose to test such
pipes with microwave emissions (replacing visual inspection of the inner surface) for small
pipe diameter. The spiral weld itself is not detected by the microwaves (the defect-free inner
surface produces a waveform with the characteristics of a smooth pipe), so we should be able
to tell if there is any defect (PWT) from the reflected waveform.
Figs. 7.3. (a) and (b) show the geometry of the spiral-welded pipe from ID (a) and
OD (b). This model is identical in specifications to the smooth pipe model, but with the
addition of the spiral weld.
As we mentioned before, we have modeled a waveguide with ID of 152.4mm and the
radius of the spiral weld bead is 6.35mm.
Fig. 7.4. is the noise waveform created by the spiral weld in the reference pipe. The
change in dB for the S11 is from approximately -10 dB to -48 dB. The waveforms (noise) tell
us that the pipe is free of defects (PWT). See the Discussion section for an explanation why
the waveform seems to have a regular pattern.
97
Figure 7.3: 3D model of the spiral welded standard pipe, with the weld shown from
ID (a) and OD (b)
Figure 7.4: Reference signals of microwave noise reflections in spiral
welded pipe (waveguide). The S21 is at 0 dB, and S11 is approximately in dB
level between -10 dB and -48 dB
7.2.2 Modeled Smooth and Spiral Weld pipes with PWT
To make this process practical, simulations must be done using real-world pipe
geometry. The tested structure is as follows: The simulated microwave emissions are
generated by a vector network analyzer (VNA). In order to produce readable reflections, the
98
microwave frequency employed sweeps through a range. In this process, the range of the
microwave frequency sweep depends on the inner diameter of the pipe (circular waveguide).
The pipe simulations for the test of the process have inner diameters of 152.4mm. The pipe
outer diameter is 203.2mm. The ports are sealed in the simulation.
7.2.2.1 Types of Defects Modeled
We modeled three types of defect (PWT) profiles: rectangular cross-section; toroidal
cross-section; and one rounded corner with one right angle corner, in cross-section. The
defects were fully circumferential in all cases. Fig. 7.5 shows the three types of PWT crosssections, with the dimensions of each part. The defect was located at 711.2mm on center
from port 1, for each PWT case as shown in Figs. 7.6 (a) and (b).
Figure 7.5: Illustration of the three types of defects (PWT): (a) rectangular,
(b) toroidal, (c) one rounded and one right-angle corner, with dimensions.
The defect width (W) is 31.75 mm in all cases
99
Figure 7.6: Side-view cross-section of model for waveguide containing
(a) toroidal PWT in smooth pipe, and (b) toroidal PWT in spiral-welded pipe
7.2.2.2 Number of Helical Coils in Model
For rolled pipe that has a longitudinal weld, there is one line of stress along the pipe
where it is welded. Helical-formed and spiral-welded pipes are important because the stock is
uniform all around the circumference of the pipe. The weld runs all around the pipe, and the
pipe wall material also runs all around the pipe. The stress is pretty much equal everywhere.
This allows various widths of material to be used to make the same diameter of pipe [30]
[120]. The defect modeled for these cases is the rectangular cross-section PWT in a pipe with
a spiral weld. Fig. 7.7. shows a spiral welded pipe made from aluminum with the number of
helices determined by the width of the plate stock at standard requirements. There are 6 coils
in this helix.
Figure 7.7: shows a spiral welded waveguide and rectangular PWT with 6 coils
in this helix. Waveguide material is aluminum
100
With narrow material the number of helical coils increases, but the pipe diameter is
independent of the width of the material. The figures will focus on the spiral weld on the ID,
showing an increasing number of coils, while pipe diameter is held at 152.4mm. The PWT
with rectangular cross section is located at 711.2mm on center, from port 1, in each case.
Four models with increasingly narrow stock are created. Each model is tested with
microwave emissions to determine the effect on the waveform caused by the growing
number of coils in the helical weld. Figures 7.8-7.13 show the models used, with
comparisons of the number of coils as the stock width decreases, ranging from 12 coils in the
helix up to 48 coils.
Figure 7.8: The original helix is shown in yellow, new tighter helix in orange.
The number of coils in the helix is doubled from the original, making 12
Figure 7.9: The new number of coils, 12, are now represented in yellow
Figure 7.10: Doubles the 12 coils (shown in yellow) to 24 shown in orange
101
Figure 7.11: Shows 24 coils of the helix clearly, in yellow
Figure 7.12: Is the final duplication of coils to 48, shown here in orange, with
the previous 24 coils shown in yellow
Figure 7.13: Is the 48-coil helix shown clearly in yellow
7.3 Results and Discussion
7.3.1 Smooth Pipe and Spiral Pipe with PWT
We see the results for the rectangular PWT in smooth pipe. The S 11 and S21 peak
resonance frequency (PRF) is 2.185 GHz. (See Fig. 7.14).
102
In the spiral welded pipe (see Fig. 7.15.) we see the S11 and S21 peak resonance
frequency (PRF) is 2.184 GHz. The difference between the two is 0.001 GHz.
In Fig. 7.16. We see the result for the defect profile with one rounded and one rightangle corner in a smooth pipe. The S11 and S21 peak resonance frequency is 2.225 GHz. Fig.
7.17. Shows the same PWT profile in a spiral welded pipe, where the peak resonance
frequency is 2.229 Ghz.
Before going to the defect profile for the toroidal PWT we will discuss the differences
between the rectangular and the rounded and one right-angle corner profiles. It is clear that
the rectangular profile has the larger volume, and it is shifted to a lower GHz resonance
frequency.
In Fig. 7.18, with the toroidal profile in a smooth pipe, the resonance frequency for
S11 and S21 is 2.254 GHz. The same profile in the spiral-welded pipe Fig. 7.19, yields S11 and
S21 resonance frequency peaks at 2.255 GHz. Note that the peak has shifted to a higher
frequency, related to the lower volume of the toroidal cross-section PWT.
The results show only very small changes in resonance frequency and in dB. But the
important point is the waveform shape distinguishes the smooth pipe from the spiral welded
pipe.
Figure 7.14: Peak resonance frequency (PRF)
Figure 7.15: PRF for rectangular PWT in the
for rectangular PWT in the smooth waveguide
spiral welded pipe, showing S11 and S21
(pipe) showing S11 and S21
103
Figure 7.16: PRF for rounded-right angle
Figure 7.17: PRF for rounded-right angle
profile PWT in the smooth waveguide
profile PWT in the spiral welded pipe,
(pipe), showing S11 and S21
showing S11 and S21
Figure 7.18: PRF for toroidal profile
Figure 7.19: PRF for toroidal profile
PWT in the smooth waveguide (pipe)
PWT in the spiral welded pipe, showing
showing S11 and S21
S11 and S21
7.3.2 Effect of Increasing Number of Helical Coils on Waveform
To examine this effect, we duplicated the original 6 coils to 12, the 12 coils to 24, and
the 24 coils to 48. The results for the S11 and S21 peak resonance frequencies, as shown
below, yields large shifts to lower frequencies as the number of coils increases. This means
104
that by increasing the volume of the weld, the resonance frequency is “backed up” to a lower
frequency. Fig. 7.20. Shows the resonance frequency for S11 and S21 at 2.183 GHz.
Figure 7.20: Shows the resonance
Figure 7.21: Shows the resonance
frequency for a six-coil helix, S11 and S21
frequency for a 12-coil helix, S11 and S21
Figure 7.22: Shows the resonance
Figure 7.23: Shows the resonance
frequency for a 24-coil helix, S11 and S21
frequency for a 48-coil helix, S11 and S21
In Fig. 7.21. The number of coils has increased to 12, and that causes a change in
resonance frequency for S11 and S21 to a lower 2.165 GHz.
In Fig. 7.22. The helix with 24 coils shows a shift in S11 and S21 to 2.096 GHz. Once
again, a shift to a lower frequency.
105
Now a big difference results when the number of coils becomes 48. That gives a
sharp decrease in the frequency for S11 and S21 to 1.613 GHz, as shown in Fig. 7.23. As we
create more coils in the helix that effectively decrease the inner diameter of the waveguide.
7.4 Interpretation and Conclusion
By calculating the exact volume for each modeled PWT profile (100% of pipe
circumference in length) we are able to compare the volume to peak resonance frequency
(PRF) in a graphic presentation. Fig. 7.24. tells us that for two cases (toroidal, and one
rounded one right angle profiles) the PRF for the smooth pipe falls at a lower frequency than
the PRF for the same PWT profile in the spiral welded pipe. However, the difference is very
small in GHz. Still, we cannot declare this as a standard effect. Notice that the opposite
occurs with the rectangular profile: the PRF for the smooth pipe occurs at a higher frequency
(in GHz) than that of the identical PWT in the spiral welded pipe. From this we can conclude
that the spiral weld generally falls at a higher frequency if compared with the smooth pipe
with six coils in the pipe.
Figure 7.24: Illustrates PRF relative to the three PWT profiles modeled in this
research. Both smooth pipe and spiral welded pipe PRFs are shown for each profile
106
The volume of PWT is a negative volume, in other words material is missing from the
pipe wall. The helical weld bead also has a volume, but it is calculated as the volume of a
solid. If we increase the number of coils in the helix, we decrease the pitch. This increases
the solid volume of the weld itself. We wanted to determine if that change has an effect on
the PRF in the spiral welded pipe. Fig. 7.25 shows the volume of the spiral bead (mm3)
plotted against the PRF for a single rectangular profile PWT case located at 711.2mm from
port 1. With a six-coil helix, the PRF falls at 2.182 GHz. Doubling the number of coils to 12,
causes the PRF to shift to 2.165 GHz, a lower frequency. The 24-coil helix generates a PRF
at 2.096 GHz. The extremely low-pitch helix with 48 coils produces the PRF at 1.613 GHz.
Increasing the coils as in Fig 7.26 shows that the PRF falls to a lower frequency as the
number of coils doubles from model to model.
Figure 7.25: Shows PRF graphed against volume of the helical weld. Here it is easy to
see that increased bead solid volume shifts the resonance peak to a lower frequency
The final conclusion is that the number of spirals affects the waveform of the
microwave signal reflection and the peak resonance frequency. The 24-coil and 48-coil helix
models were developed simply to gauge the effect on PRF; such pipes would never be built
in the real world.
107
Figure 7.26: Shows PRF plotted against number of coils in the helix. Here it is easy to
see that increased coils forces the resonance peak to a lower frequency
Frankly, the distinguishable differences in the waveform shape between smooth pipe
and spiral-welded pipe demonstrates that buried pipe, the construction of which is unknown,
could be categorized to either smooth construction or spiral weld construction on the basis of
these waveforms.
By modeling the different characteristics that affect PRF, we change the PRF a little
bit in such a way that the waveform is distinguishable case to case. In future work we are
about to build a comprehensive database of the possible waveforms created by all sorts of
inner wall conditions and PWT characteristics. This will allow use of pattern recognition to
nondestructively identify and pinpoint the categorization of the pipe, its structure, and inner
condition.
108
Chapter 8
High-Efficiency Remote Measurement of Pipe Defect Using RF/UT
Technologies: A Theoretical Analysis Part One—Straight Beam UT [122]
Abstract
This analysis has established a new hybrid RF/UT system for non-destructive testing of pipe
walls for pipe wall thinning (PWT) in order to predict location, and enable measurement of
the depth of defect by combining the group velocity method and calibration condition. A
simulation of microwave (MW) behavior in a 91% brass waveguide (762mm pipe, Young’s
Modulus 102KN/mm2) was developed using Computer Simulation Technology (CST). The
model included a frequency band of 1.283GHz for the TMnm mode (TM01 and TM21), with a
sweeping frequency from 0.70GHZ to 2.00GHz. The model includes 14 instances of fullcircumferential PWT, regularly spaced along the length of the waveguide with step-width of
50.8mm on center. For each we have modeled four cases of increasing PWT (5.08mm,
10.16mm, 15.24mm and 20.32mm). Considering the measurement with MW as a prediction
of the location of the PWT, rather than a measurement, we can guide a straight-beam UT
probe to the position predicted by MW, and use the appropriate signal velocity ultrasound to
accurately measure the depth to defect from the outer surface of the pipe. The straight beam
UT is found to be no better at determining the geometry of the defect than MW, but the
accurate depth to defect (DDO) measurement would allow estimation of the volume of the
PWT.
8.1 Introduction
Major industries make extensive use of pipes for transporting oil and gas, for moving
water to end users, for getting chemicals from storage to point of use in manufacturing
plants, and to provide heating and cooling within power generation plants. A large proportion
of these pipes have been in service for decades. Older pipes are more susceptible to failure,
which can lead to injury as well as damage of the surrounding structures in complex piping
109
systems. To avoid catastrophic failure, industries have a vested interest in developing ways to
detect defects, such as pipe wall thinning (PWT) which are among the major causes of pipe
failure. Nondestructive methods of detection and evaluation (NDE) of the extent of damage
are necessary, since the pipes must remain in service. The more cost-effective techniques of
NDE allow industries to avoid life-threatening mishaps, to effectively maintain pipes, to
predict the useful lifetime of these pipes, and to provide a guarantee of safety for the people
who work around them [37] [90].
In addition to field defects, there are defects in both the plate material and the formed
pipes during manufacture. Internal defects (subsurface) may be treated by grinding to remove
them. In some cases the amount of material removed by grinding can result in a thinning of
the pipe wall that is outside tolerances [90] [91] [92].
NDE during manufacturing allows the facility to quantitatively manage this kind of
defect. Techniques in broad use are optical inspection, radiographic testing (RT) [30],
ultrasonic testing (UT) and so on. These are dynamic processes in that either the product
under test is moved, or the probe is moved [1]. There are other techniques that show promise
such as MW energy, but these cannot be used on their own. However, these promising
techniques might be quite useful in a hybrid system with another NDE method [37] [90].
Wissam Alobaidi et al. suggest that “It is recommended that a second, powerful NDE
technique capable of remote detection would be useful alongside UT in order to direct the
placement of the probe” [29], in a UT techniques review paper.
A remote NDE method based on microwave (MW) technology has been previously
researched, and presented as a possible way to resolve PWT location problems [38]. Fig. 8.1,
shows the first technique used in our approach: a microwave signal propagating inside the
pipe, which acts as a circular wave guide, is used in order to locate the area of PWT [90].
Microwave signals can travel through metal pipes without losing much of their
energy. This means that even very long pipes can be inspected rapidly with MW technology
[37] [90].
110
Ultrasonic Testing (UT) of metals is useful for detecting and measuring both surface
and subsurface discontinuities. UT has many other uses, including measurement of the extent
and thickness of defects [29]. The most often used frequencies range from 0.1 to 25 MHz.
Achievement can be accomplished at stress levels far below that which would cause
permanent alterations of the metal parts [46]. Fig. 8.2, shows the second technique: straight
beam UT is used to determine the depth of PWT and determine the circumferential length of
the PWT [29].
Figure 8.1: Model for case study for first step (microwave), 2-port configuration,
with full-circumferential rectangular defect profile. Di is inner diameter, DD is depth
of defect, D is pipe wall thickness, W is the width of PWT
In this paper, the fundamental Principles of RF and UT Technologies have been
demonstrated. Historically, both technologies have been used in practice and in
experimentation to evaluate PWT in pipes. The aim is to use the knowledge gained by use of
UT (longitudinal waves) and RF (microwaves) to both locate and characterize the structural
tolerances of PWT. The main goal of this research is to join the radio-frequency (RF)
techniques with ultrasonic (UT) straight beam probe techniques in a hybrid system. Detection
would occur in two phases. The first phase would use MW signals to detect and generally
locate PWT defects in the pipe wall. The information generated by the RF phase would then
be used to remotely guide a straight-beam UT probe to the area, where it would be used to
determine the specific boundaries of the PWT.
111
The pipe parameters are: outer diameter 254mm, inner diameter 203.2mm, length of
pipe calibration 762mm. The pipe material used is brass 91%. PWT parameters are: width
and wall thickness are 25.4mm each, and the diameter of defect (PWT) varies as 213.36mm,
223.52mm, 233.68mm and 243.84mm. For this paper 14 different locations were simulated
for detect the edges of PWT.
Figure 8.2: Diagram of ultrasonic processing of piece of pipe section to
detect a PWT defect, using a straight probe
The model included Transverse Magnetic (TMnm) modes (TM01 and TM21) with
frequency range (Bandwidth) 1.283GHz. Where n and m mathematical roots for Bessel
functions [123].
112
8.2 Basic Principles of RF and UT Technologies
8.2.1 Electromagnetic Waves
From the equations devised by Maxwell we can learn about the major facets of
electromagnetics as a field of study: light itself is a form of electromagnetic energy; magnetic
fields are created by electric fields that vary over time (and vice versa); electric currents can
be created in wires. But, Maxwell’s equations also demonstrate that electromagnetic energy
can travel over distance with or without a medium (such as MW energy in air-filled and
evacuated pipes). Electromagnetic energy moves as waves, which allows action at a distance
from the source. Thus, forces can be brought to bear far from the point of origination (port)
of the electromagnetic waves, and can be delayed in time [123].
8.2.2 Ultrasonic waves
The mechanical energy of ultrasonic signals performs substantially the same as the
sound waves that we can hear. In important ways, they also behave in a manner similar to
light waves. The ultrasonic waves can travel through any phase of matter, with the rate of
travel dependent upon the properties of the substance being penetrated, rather than the wave
properties of the ultrasonic beam. This rate of travel, or velocity, varies as the waves move
through various homogeneous media, but a beam of ultrasound will interact with interfaces in
ways similar to light beams. Ultrasound waves can be reflected from test objects, they refract
at the boundary between dissimilar materials (each of which has its own wave velocity
characteristics), and they are diffracted when the beam reaches an edge surface. An irregular
object surface will scatter the ultrasonic waves, as will particulates, reducing the energy
carried by the beam. Unlike light waves, which can move through a vacuum, ultrasound
cannot propagate in a vacuum [46].
8.3 Wave Propagation
Two types of wave propagation are used in the system we propose in this paper.
113
8.3.1 Microwaves
Electromagnetic waves in the frequency range from 0.3 GHz to 325 GHz are called
microwaves. The corresponding wavelengths are 10,000 mm to 1mm, respectively [46].
Microwaves have been known since the days of Maxwell, but technology did not
produce generators or receivers suitable for evaluating materials with microwaves, until the
invention of radar in the 1940s. Early in the application of microwaves for NDE they were
used to inspect parts, including antennas, waveguides and the radomes for radar apparatus
[46].
8.3.2 Longitudinal waves
Of the various types of ultrasonic waves, Longitudinal waves (LW) are the most
commonly used when inspecting metals. These waves create vibrations in the particles of the
substance under test, with the vibrations moving forward and backward parallel to the
direction the waves travel [46].
Ultrasonic LW have different velocities in different materials. One example is that
LW moving through (penetrating) rolled copper travel at about 5010m/s [29].
 (1 −  )
 = √
(1 +  )(1 − 2 )
(8.1)
The LW velocity (VL) formula above depends on the material properties. Em is
Modulus of Elasticity, μm is Poisson's Ratio and ρ is material density [124].
8.4 Theoretical Approaches
The Calibration of Group Velocities method has been used to determine the starting
and ending points of PWT regions. But the peak of microwave (MW) reflections in time
domain measurement is not sharp enough to determine the actual starting and ending limits
of the defects [34]. Even the best optimizations for development of the waveform peaks fail
to target the exact starting and ending points, and still only show the general location of the
114
defect. The peaks cannot be accurately determined to disclose the actual boundaries of the
defect. The peaks may even appear ahead of or beyond the true limiting locations of the
discontinuity [38].
The equation below defines the group velocity of MWs in a circular pipe which are
within a sweeping frequency range (explained later in the subsection “Modeling Analysis and
Numerical Simulation.”) [34]:
 2
√
 =  . 1 − ( )

(8.2)
Where f is the operating frequency, for VC which is calibration velocity, and fc is the
cut-off frequency, Cd speed of light inside the wave guides for free space. There is difficulty
in working with a single frequency calculated from the equation. Thus it is necessary to
determine the group velocity specifically for the condition within the pipe, determined by a
calibration method explained later in the subsection “Analysis of Microwave Signals in Time
Domain.” [34] [39].
8.4.1 Microwave Approach
8.4.1.1 Geometry and Boundary Conditions
One cross-sectional profile of defect has been considered (rectangular shape). For
defect length full-circumferential wall thinning has been used, as shown in Fig. 8.3.
Figure 8.3: Geometry of PWT rectangular profile model, showing most
distant incidence of PWT located 711.2m from port one
115
The width parameter of the PWT is 25.4mm, wall thickness 25.4mm, and the depth of
defect (DDi) parameter varies as 5.08mm, 10.16mm, 15.24mm and 20.32mm. The pipe metal
thickness (D) represents a pipe free of defects (100% smooth). The four cases of DDi used
represent 20% of the metal thickness, 40%, 60%, and 80%, respectively). The pipe
parameters are: inner diameter (Di) 203.2mm, length (Lp) 762mm. Some compositions of
pipes will attenuate the signal by absorbing a portion of the MW. In this case, to make a
calibration (baseline) pipe, we use a brass conductor (the pipe material used is brass 91%).
 =  − 
(8.3)
Where DDi is the depth of defect measured from the inner surface and DDO is the
depth to defect from the excitation point of the UT straight beam.
A single port is used to calibrate the MW velocity by creating a reflection from the
closed end of the waveguide [34].The medium within the smooth pipe is air. The system is
short-circuited. Port 1 is excited by a TE11 mode across a specified bandwidth, 1.283 GHz.
And then, two ports are used to reveal and locate the defect. The modeling mesh type used in
CST simulation is Hexahedral (Legacy).
8.4.1.2 Modeling Analysis and Numerical Simulation
A pipe circular in cross-section acts as a circular waveguide (shown in Fig. 8.4),
which works with both TE and TM modes. The TE mode is the first to disseminate
(propagate) as an electrical field within the circular waveguide, and it is the dominating
mode, thus, most often used [39].
The cutoff wave number of TE11 mode is expressed as below [39].
 =
2 

(8.4)
fcTEnm is the lowest cutoff frequency mode shown in the system. Also, n and m correspond to
the mathematical roots which depend on the modes of the waveguide, Cd speed of light inside
the wave guides for free space.
116
Figure 8.4: Cut-away section of a circular pipe showing location of PWT,
to demonstrate that such a pipe will act as a circular waveguide
 =
 =  = 3 ∗ 108



√
(8.5)
. If the pipe is filled with air, the permittivity, εr = 1.
The cutoff wave number is defined by the roots of Bessel function Jn(x), x is Pnm; so Jn(Pnm)
= 0, for TEnm or TMnm with Pnm mathematical roots. So, Kc is expressed as follows.
 =
2

(8.6)
Di is the inner diameter of the wave guide. From equations (8.4) and (8.6), the cutoff
frequency could be written as follows below [39].
2  2
=


 =
 

(8.7)
(8.8)
117
The lowest cut-off frequency created by Port 1 and Port 2 as TE11 is 0.8653 GHz. The
Bessel function for the frequency range is 1.283 GHz during the TMnm mode (TM01 and
TM21), (P01=2.4048, P21=5.1356). To ensure inclusion of the TM01 we will set the sweeping
frequency from 0.7 GHz to 2.0 GHz. The simulated waveforms must be given something to
detect, of course, and for that we constructed a regular series of PWT into the model, as
described below.
Multiple occurrences of PWT will be located beginning at the port 1 end of the pipe,
with 14 incremental positions moving stepwise toward the far end of the pipe. The 14
positions of PWT have been simulated in CST, incrementing at 50.8mm on center. The
center of the first PWT instance is located at 50.8mm from the point of origin of the
microwave signals. The 14 instances range from 50.8mm at center, to 711.2mm at center
within the pipe length (Lp). The pipe model configurations are presented in Fig. 8.5.
This step, using regular, known PWT occurrences will allow us to formulate the
equations for positions all along the pipe length that can be used to develop the programming
system.
Figure 8.5: Longitudinal section of modeled pipe used in simulation study,
showing first two and last two examples of PWT. Ld is the distance
between PWTs on center; Lp is the length of the pipe
118
By changing the location through a series of PWT cases, we seek to establish that
MW cannot precisely locate the PWT instance by itself.
8.4.2 Ultrasonic Method (Straight Beam Approach)
8.4.2.1 Modeling analysis and Boundary Conditions
Step two moves from electromagnetic waves as the medium of detection, to a
mechanical process that vibrates the material itself in order to accurately determine the
boundaries of the defect, and its volume [125].
The rate of travel of sonic waves through different pipe metals varies, according to
the type of metal. The fundamental velocity law says [125]:
 =  ∗ 
(8.9)
Where V is the ultrasonic signal velocity, λLW is the LW length and f is the frequency.
When LW move through any material, such as the metal of a pipe wall, they cause the
molecules of the material to vibrate in the direction of travel of the waves. The ultrasound
waves can move through both solids and liquids. Thus, liquid (water) or another fluid
(grease) are used as a medium to insure contact between the probe surface and the surface
under test [125].
Ultrasonic straight-beam probes used to detect PWT do not operate continuously.
Instead, they emit brief pulses of ultrasonic LW, which are produced by a piezoelectric
crystal and transmitted through a disc. “The probe emits a wave packet which has a dominant
frequency equal to the natural frequency of the crystal,” according to R. Halmshaw [125],
This is necessary in order for a resonant condition to exist, causing the probe disc to vibrate,
and creating LW that can travel through the pipe metal under test. The disc is controlled by a
mechanical or electrical damper so that the individual wave pulses have a fast decay profile
[125].
119
Ultrasonic LW propagate along a more or less straight path because of their minute
wavelength, but the beam diffuses somewhat. The formula for spread angle, θS is as follows
[125]:

 1.22
=
2

(8.10)
where λLW is the longitudinal wavelength, DC is the diameter of the straight-beam crystal.
High frequency LW diffuse less.
An ultrasonic LW beam has a “near zone” (roughly parallel beam) and a “far zone”
(more diffuse beam). The length of the near zone beam (Nb) can be calculated as [125]:
 =
( )2
4
(8.11)
There are considerations due to fluctuation of the beam intensity that are particular to
the near zone, which must be incorporated into the test mode. In other words, it is difficult to
reliably measure the size of flaws if they appear in this zone. The intensity of the beam’s far
zone can be calculated using the spread angle formula [125].
8.4.2.2 Ultrasonic Straight Beam Measurement
Two ultrasonic testing techniques are important for measuring defects in pipes in
general, and in oil and natural gas pipelines in particular, either during manufacture or inservice: they are Straight Beam and Angle Beam. Both these techniques depend upon the
type of signal used to penetrate the test object (pipe, vehicle, shuttle, concrete and so on). The
UT technique is selected to serve the problem case (degree of defect), and the transducer
used is fabricated to fit and solve that problem. However, both these techniques have
restricted boundary and configuration conditions. Straight beam employs the LW ultrasonic
concept, and is best used in order to measure the DDO in a pipe, rather than the structural
boundary of the discontinuity (PWT) as shown in Fig. 8.6(a).
120
Figure 8.6: Schematic of model with PWT instances and UT straight-beam evaluation.
(a) Shows detectors placed directly above the PWT, measuring DDO. Ro is the external
radius of the pipe. Ri is the internal radius of the pipe. RPWT is the radius of the PWT.
(b). Schematic of model with PWT instances, showing three cases of placement of the
UT straight-beam probe. Case 1 shows the probe at a PWT edge. Case 2 shows the
probe between PWT instances, measuring plate thickness. Case 3 shows the probe
placed directly above one instance of PWT, and able to accurately measure DDO
121
We consider three cases classified according to where the UT straight-beam probe is
sent, and the reflected readings it will generate, as shown in Fig. 8.6(b). The UT probe is sent
to a location to test, based on the distance from port 1 which is estimated by the first
technique (microwave technique). According to previous research, microwaves detect the
presence of PWT, but measure the PWT location within a range. The measured distance may
fall within the boundary area of PWT, or outside it but still near it (either before or after the
PWT).
The direction of the LW excited in the metal by the UT probe is perpendicular to the
surface of the plate. When the diameter of the UT beam, which depends on the diameter of
the straight-beam crystal (DC), is larger than the parallel PWT surface, or is completely
within the boundaries of the PWT surface, the reading returned will show accurate DDO.
As the UT probe moves backward and forward it may return three possible distances
based on the time delay of the returned signals. When near the PWT, but outside it, the
reflected signals will show distance D (case 2 in Fig. 8.6(b)). When directly above the PWT
(case 3), it will return DDO. The first case shows the third possibility produced as the probe
moves along the outer surface: the straight beam signal will disperse in propagation as
mentioned before, and the presence of PWT will be confirmed because of the capability of
this technique to detect defects which have a surface parallel to the plate surface. But, in this
case, because the beam is not wider than the PWT surface, it reflects from both the back wall
of the plate and the top edge of the PWT. It will show a distance greater than the actual depth
of the PWT (DDO), but less than the full thickness of the plate (D). The straight beam cannot
determine that the probe is centered above the perpendicular edge of the PWT. Straight beam
UT accurately measures the DDO but cannot recognize the starting and ending point of the
PWT because the edge is not parallel to the excitation point of the straight-beam probe on the
pipe surface.
Finally, the UT probe will use straight beam longitudinal waves, LW (compression
waves). It will be guided to the general region of the defect by the data produced by the RF
detection. The UT probe will scan for the starting and ending points of the discontinuity by
using penetrating sonic waves and the wave average velocity. The LW velocity for brass
91% is 4,700,000mm/sec approximately.
122
8.5 Analysis Results and Discussions
8.5.1 Analysis of Microwave Signals in Time Domain
According to the inner waveguide diameter (Di) 203.2mm, and waveguide length (Lp)
762mm, the sweeping frequencies are fcTM01 to fcTM21 (1.13GHz to 2.413GHz respectively).
As mentioned above the set-up is 0.7GHz to 2.0GHz to insure the TM01 mode inclusion. The
average velocity for RF should be calibrated for a pipe free of defects (PWT), and then the
same calibration velocity will be used for the pipe with a defect. Fig. 8.7 shows the result of
microwave signals propagated in the reference waveguide and reflected from the closed end
of the waveguide.
The waveguide calibration pipe has a single port, is free of defects, and has a
reflection from the end of the pipe (close end condition). According to equation (8.14) below,
Lp is 762mm (reference pipe length), the delay time (ΔTC) in seconds, as shown below. And
in our simulation we consider 14 instances of PWT within the same waveguide. The example
below considers a PWT at the distance 254mm from port 1 to the center of that PWT. The
delay time of PWT (ΔTPWT) will be calculated as shown below.
123
Figure 8.7: Time domain modeling and analysis results for waveguide: (a) Calibration
pipe (reference waveguide) measured delay time (ΔTC) with open end condition in
order to calculate calibration velocity. (b) Graph of signal propagation in waveguide
with a single PWT defect. Location is 254mm from Port One to the center of PWT.
(c) Graph of signal reflection from the same PWT defect
124
Group velocity and calibration condition are calculated as shown below. From Fig.
8.7.(a) the propagation and reflection signal peaks calculated as the ΔTC. From Fig. 8.7.(b)
and (c) the propagation and reflection signal peaks calculated as the ΔTPWT from which we
can calculate the distance to PWT. Arrows in Fig. 8.7(b) and (c) indicate the times in peak
and trough used in the calculations below.
Δ
1 = 2.9508 ∗ 10−9
1,1 = 9.7646 ∗ 10−9
] = 6.8138 ∗ 10−9 
(8.12)
1 = 2.9520 ∗ 10−9
1,1 = 5.1559 ∗ 10−9
] = 2.2039 ∗ 10−9 
(8.13)
{
Δ {
To apply the velocity calibration method for the microwave technique we use the
equation below.
 =
2
Δ
(8.14)
Lp is the length of the calibration pipe free of defects, in meters (0.762) with a delay time of
ΔTC seconds along the waveguide; as calculated above, the calibration velocity is
VC = 0.224 x 109 m/s.
8.5.2 Prediction and Detection of the PWT Location
The modeling simulation predicts the PWT location. We would expect the predicted
location to fall between the starting and ending points of the structural PWT created in the
model. In reality, the predicted location may be earlier, or later, or within the boundaries of
the modeled PWT, although the predicted location is always close to the actual location for
the PWT. As mentioned above, 14 PWT instances are simulated along the waveguide. From
Equation (8.14), the approximate distance to PWT in meters can be calculated and written as:
 =
Δ ∗ 
2
(8.15)
125
Where LPWT is the distance from the port to the PWT location, in meters (0.246), with a delay
time of ΔTPWT (2.2039 10-9) seconds, as calculated above. For this PWT location, the
distance to starting point (LS) and the distance to ending point (LE) should be 241.3mm and
266.7mm from port one, respectively. The calculated result is nearer to the LS rather than LE.
These two exact points are undetected by the calibration velocity method, while the
predicted location has been detected within the boundary and configuration of the pipe wall
reduction (PWR). In another instance the location was calculated to be near the PWR, but
ahead of it.
Fig. 8.8 below shows the results for four modeled locations of PWT detected among
the 14 instances, and clarifies the differences among them. All the graphs below (a, b, c and
d) represent the distances which are simulated from the excitation port (port 1) of the
microwave signals, to the pipe wall thinning center (L(PWT)C). The four points in each graph
represent the four depths of defect, although we are concerned only with the distance to
defect L(PWT)C in this case. In Fig. 8.8(a), the L(PWT)C detected for each DDi falls within the
starting and ending points of the PWT region. In this case, the detected L(PWT)C points shift
closer to LE, proportional to the depth of defect (DDi). Next, in Fig. 8.8(b) the detected
L(PWT)C points are comparatively much closer to LE, and the last point at depth DDi =
20.32mm is beyond it (out of the defect region). In other cases, the detected L(PWT)C points
are all near the LS, as shown in Fig 8.8(c). Additionally, in Fig. 8.8(d) all the depths returned
detected L(PWT)C points that are outside the defect (beyond the PWT).
126
127
Figure 8.8: Evaluations of PWT with microwave signals, to determine LS and LE of
each instance in the model, measured from port 1. Shown are results from four
selected cases. (a) PWT centered at 50.8mm, showing that the results for each of the
four DDi fall within the width of the PWT. (b) PWT centered at 203.2mm, showing
shifting of the detection points toward the LE, compared to the first case, so that one
measurement falls beyond the actual LE. (c) PWT centered at 254.0mm, showing
detection points shifted nearer to the LS, compared to the first and second cases, but all
within the width of the PWT. (d) PWT centered at 558.8mm, showing all four
detection points shifted beyond the LE of that PWT instance
8.5.3 Detection of the PWT Depth
The UT detects the depth to defect (DDO) according to the velocity of the waves
penetrating the material under test. Pipes modeled as 91% brass material with four DDO of
5.08mm, 10.16mm, 15.24mm and 20.32mm would be read with more or less the same group
velocities, independent of the thickness of the material. The concept is the same for other
materials, such as aluminum, rolled, where the signal velocities would be different but the
principles of measurement would remain the same.
MW technology determines the general location of any defects within the standard
limitation. By using an Ultrasonic Method (Straight Beam Approach), the DDO has been
calculated accurately. If the LW velocity VL is constant then.
 =
2 2
=


(8.16)
Where VL is the LW velocity. When an ultrasonic signal travels through a pipe of thickness D
millimeters in a delay time of TD nano-seconds, if the LW velocity VL is constant then 2D is
the distance from the excitation point of the ultrasonic signal port across the pipe thickness.
The ultrasonic signal will reflect from the back wall of the pipe surface or the PWT surface.
Also, TDDo is the delay time of travel to and from the PWT. From equations (8.1) & (8.15).
128
 ( )2 (1 −  )
 = √
(4)(1 +  )(1 − 2 )
(8.17)
DDO measurements determined after MW signal predicted the PWT region by using a
straight-beam ultrasonic probe are represented in Fig. 8.9 for two materials that have been
modeled. The first is Brass 91% and the second is Aluminum, rolled. The aluminum, rolled is
modeled only to compare the relative ultrasonic signal velocities of the two materials, not for
the MW step. The time domain differences shown here are shifted according to the relative
signal velocity of ultrasound waves as they penetrate each material, as mentioned in equation
(17) above.
Figure 8.9: Comparison of DDO determined using straight beam UT in two
materials, with the four DDi modeled. The shift in time delay is due to signal
velocity differences in brass 91% compared to aluminum, rolled
As we mentioned in the “modeling analysis and boundary condition” of the straight
beam approach section above, due to fluctuations in the intensity of the beam in the near
zone, measurement of distance to the PWT surface and the back wall surface of the pipe can
vary. This can make such measurements unreliable, especially at the starting and ending
129
edges of the PWT. In the far zone the wave intensity can be calculated using the spread angle
formula.
8.6 Conclusions
This paper has established a new hybrid technique pairing RF/UT techniques for a
two-step process with which PWT location can be generally predicted (microwave), and then
measured (UT). Newly established microwave technologies used for inspection of pipe
metals, have evaluated several linear and nonlinear calibration techniques such as the group
velocity method (calibration conditions). The microwave simulation used a CST model with
sweeping frequency from 0.70GHz to 2.00GHz in a waveguide (pipe) 762mm in length, with
inner radius of 101.6mm, and full-circumferential, rectangular profile section PWT. It is
usual practice to study the various materials used, to develop the needed devices, and then to
investigate the practicality of using an NDT technique for examination.
The techniques will be complementary. The RF will contact the inner surface of the
pipe, and then the UT will penetrate the pipe wall from the exterior to the interior. This will
allow measurement of the DDO within the pipe wall. We must acknowledge that the straightbeam UT technique is better at detecting the DDO than it is at detecting the edges of the
PWT, when compared with other transducers, such as angle-beam UT probes.
Finally, microwave techniques (group velocity method and calibration conditions) are
predictive techniques useful for evaluating the region of the PWT. The four cases selected in
different places along the pipe (at 50.8mm, 203.2mm, 254.0mm, and 558.8mm on center
from port 1), taken from among the 14 total cases, show the limitations of MW when
predicting the location of PWT in a waveguide (pipe). There are clearly variations in the
detected position of PWT that are dependent on the depth of defect. The interaction between
distance and depth makes the MW technique good at.
1) Determining that PWT is present, and.
2) Estimating the distance it is from the port. But it cannot pinpoint the boundary structural
geometry of the type of defect.
130
UT has an advantage in detecting the DDO of the PWT, thus can be useful in
estimating the volume of the PWT. But the straight beam UT probe is also incapable of
determining precisely where LS and LE are.
Because MW cannot determine accurately the LS and LE of PWT, and straight beam
UT cannot accurately determine the LS and LE, either, but is good at determining DDO,
another technique is needed to home in on the edges of defects. There is a third technique,
angle beam UT, that might be used with MW to determine the structural geometry quite
accurately. Its strengths and how it could be paired with RF technology should be
investigated.
131
Chapter 9
Two-stage Technical Protocol for Locating, Validating, and Characterizing
Manufacturing Defects Remotely using Hybrid System RF/UT: Part
Two—Angle Beam UT [126]
Abstract
A novel hybrid RF/UT nondestructive testing (NDT) system is researched using Computer
Simulation Technology (CST) modeling. The system uses a two-step process. The first step
is predictive measurement of pipe wall thinning (PWT) defect locations using microwave
signals in a sweeping frequency range (0.70GHz to 2.00GHz). The predicted locations are
found to be situated in a region bounded by half width, W/2 ahead of, and W/2 following the
actual location of the PWT defect. The predicted locations are transferred to optimization
modeling which uses our innovative sensor positioning equation to calculate a series of
excitation points (EPns) for placement of angle-beam ultrasonic (UT) probes in step two. The
EPns are then passed to a genetic algorithm which determines the best beam angles (βR) for
each EPn in order to effect precise measurement of the defect edges and depth. The probes
are located in three carefully determined positions relative to the predicted location from step
one. Each of these placements of the UT angle beam allows for three legs of the sound path
to be available for detection, for 9 legs total. The resulting UT data is used to draw an
accurate picture of the PWT location and depth. Initially, the UT probes used for calibration
of the innovative location equation were set at 45° and were used for rectangular profile
defects. The % error (up to 20% for width and up to 35% for depth) was calculated for each
calibration test. Adjustments were made to the beam angle for each EPn until the depth and
width measurement errors were minimized (less than 2.5% for depth, and 6% for width). It
was finally resolved that in order to assure the most accurate measurements, six angles are
needed for six EPns, located three ahead of the predicted location and three beyond. The
angles determined for the six positions are 32°, 37°, 46° (sensing locations beyond predicted)
49°, 43°, and 39° (sensing locations ahead of predicted). Further optimization was
undertaken to allow characterization of known defects with three profiles (rectangular,
triangular and semi-circular). In this phase it was clear that systematic repositioning of the
132
UT beam would greatly increase accuracy of location and characterization of profile. The
best angles and placements for the UT probes were determined, and are presented in tabular
form. The model itself is further validated by simulated reconstruction of a prior published
RF-based PWT location experiment, which produces comparable results to those published,
with a small, but consistent offset. The simulation even reproduces an observed indistinct
peak noticed when the defect width is small, making accurate location of the defect
impossible with Radio Frequency (RF). This is the reason the UT position shifting step is
added to the EPn technique. In the final optimization stage the adjustments for positioning of
the UT beam for all defects combined, are at the minimum -2mm, and at maximum -11mm.
9.1 Introduction
Metals to be tested are examined with a number of nondestructive testing (NDT)
methods. This nondestructive evaluation (NDE) is used in many industries where
sheets/plates of metal are the focus of inspection (ships, trains, aerospace, aircraft, ducting,
and pipeline) [27]. Among the methods commonly used to test sheets of metal there are a
number of techniques, including eddy current, liquid penetrant, magnetic particle, acoustic
emanation, X-ray, ultrasonic (UT) and radiofrequency (RF) waves [1] [27] [37] [90]. Each
method has its particular strengths and weaknesses. As mentioned in Part One “Highefficiency Remote Measurement of Pipe Defect Using RF/UT Technologies—a Theoretical
Analysis (Part One: Straight Beam UT)” Wissam Alobaidi [122], microwaves (MW) have a
long reach, but are not good at discretization of defects, such as pipe wall thinning (PWT)
[34] [38] [112] [113]. UT can determine thickness, and starting and ending points (extent)
with great accuracy, but the UT probes must be very close to the defect in order to measure it
[29] [30]. Part One deals with use of microwaves coupled with straight beam probe UT
testing [122]. There is another UT technique that we want to examine in this part: the anglebeam probe approach.
Ultrasonic shear waves (SW) used in the angle beam probe approach, penetrate more
slowly, and use shorter wavelengths in comparison to ultrasonic longitudinal waves (LWs)
[29] [114].
133
We first use microwave (MW) to determine the degree of PWT (location and
volume). The main focus of the follow up with UT is on measuring the defect, determining
its extent and thickness. We use angle beam probes with capability to serve both straight and
circular plates, because of the arc of the pipe surface, to detect the edges of PWT. Modeling
of microwave was carried out with sweeping frequency 0.70GHz to 2.00GHz.
Fourteen (14) instances of PWT were modeled along the waveguide. The width (W)
was standardized at 25.4mm. The fourteen positions begin 50.8mm on center from port 1,
and proceed through the length of the waveguide on 50.8mm centers. Four depths of defect
(DD) were modeled for each location (5.08mm, 10.16mm, 15.24mm and 20.32mm).
The positions predicted by microwave analysis for the PWT instances ranged from
W/2 before the actual starting point of the modeled instances of PWT to W/2 after the actual
ending point of PWT.
These technologies will work together in order to size the defect. The microwave
signal will evaluate the inner surface area of the pipe, while the angle-beam UT will
penetrate the pipe wall from outside, determining the PWT location according to the angle of
reflection, which creates the skip distance (SKD) of the sound path (SP).
Initially, the UT probes used for optimization of the location were set at 45°. The %
error was calculated for these test runs, and the depth and width measurement errors were
minimized as adjustments were made to the beam angle. It was ultimately determined that in
order to bring precision to the measurements, six angles are needed for six excitation points
(EPn) three ahead of the predicted location and three beyond.
The first PWT examples tested were all rectangular in profile. The most effective UT
beam angles and positions were determined. The test results were then applied to known
PWT cases of known profile (rectangular, triangular and semi-circular) in order to assess any
errors that might arise due to variation in cross-section. To accurately locate and characterize
the defects, the EPns were passed along to a genetic optimization algorithm which controlled
the angle of the beam, as well as its placement. The beam angles and probe locations were
refined for each of the cross-sectional profiles.
134
For validation of the CST simulations a prior RF-based PWT defect location
experiment was reconstructed with the aim of comparing the experimental results to those of
the simulation. Comparable results would reassure us of the efficacy of using the simulation
technique to develop the RF/UT procedure. The parameters of the original experiment were
reproduced for one case in a 2011 publication, and produced the desired comparable results.
Although there was an offset from the original measurements with the simulation, the results
were consistent in direction and degree. This suggests that the CST simulation is a reliable
development tool.
9.2 Microwave Approach
For the RF technology two roots are set up for the Bessel function as (P01=2.4048,
P21=5.1356). The bandwidth is 1.283GHz bounded by cut-off frequencies fTM01 (1.13GHz)
and fTM21 (2.413GHz) which are created by Port 1 and Port 2, as expressed in the equations
below [123]. To ensure they are included we will set the sweeping frequency as 0.70GHz to
2.00GHz, as shown in the equations below. The waveguide is filled with air as a medium.
We recall the lowest cutoff frequency (fcTE11 = 0.8653 GHz) at TE11 mode from (Part
One: Straight Beam UT) for same pipe parameter and boundary configuration that have been
mentioned previously [122].
01 =
2.4048

1.8412 11
(9.1)
21 =
5.1356

1.8412 11
(9.2)
25.4mm is the pipe wall thinning (PWT) width. The pipe is modeled as: inner
diameter 152.4mm, wall thickness 12.7mm, length 762mm. The pipe is considered to be
brass material. As in Part One, a single cross-sectional profile is considered for the
discontinuity (wall-thinning rectangular shape, fully circumferential). 14 instances of defect
are modeled along the reference pipe. The width (W) is constant at 25.4mm. The fourteen
locations start at 50.8mm on center from port 1, and follow through the length of the
135
reference pipe on 50.8mm centers. Four depths of defect (DD) are modeled for each position:
5.08mm, 10.16mm, 15.24mm and 20.32mm.
Group velocity for microwave should be calibrated for a defect-free pipe. The same
calibration velocity must be used for the pipe with discontinuity.
The microwave modeling and analysis for velocity calibration method used in this
hybrid system is the same as explained in Hybrid System Part One [122].
9.3 Ultrasonic Method (Angle Beam Approach)
9.3.1 Ultrasonic Angle Beam Analysis
Ultrasonic hand scanners have been used for testing metal plate and pipe walls for
some time. The addition of automated equipment to carry out the UT scanning has increased
the ability to repeat scans reliably, and allows making a permanent record of the scan results
[46].
Angle-beam UT is most useful when testing pipe and in other circumstances where
the straight beam probe cannot make complete contact with the outer surface of the part
under test. Flaws can be located with angle-beam methods whenever there is no back
reflection, which allows angle-beam techniques to be used to find edges of flaws. By using
zig-zag scanning patterns the edge reflections can be further refined. Remember that straightbeam UT is able to determine the depth to defect, and angle-beam UT is very good at finding
edges [46].
In Fig. 9.1, an angle-beam shear wave is shown passing through plate metal. Note that
the path of the beam reflects from the back surface and continues in the same direction, now
angled toward the upper surface. Wherever the beam contacts the surface, this is called a
node. The distance between two nodes on the same surface is called “skip distance,” and is
useful for determining the path the transducer should take when testing the part for hidden
defects [46].
136
If the UT probe is moved in a regular zig-zag pattern with the points of the triangles
at one skip distance apart, UT can be used to determine the size and shape of the flaw. Fig.
9.2, shows the technique in use to determine and refine the accuracy of measurements when
seeking the edges of an instance of PWT [46].
Figure 9.1: Illustration of the ultrasonic angle beam transducer with representation of
two nodes through skip distance (SKD). D is the depth of material
Figure 9.2: Three dimensional of test object with PWT geometry (rectangular
profile), shows zig-zag scanning patterns the edge reflections. W is the width of
PWT, DDO is depth to defect, DDi depth of defect, Leg1 is the 1st leg and Leg2
is the 2nd leg
137
9.3.2 Measurement and Numerical Configurations
The UT probe will use angle beam shear waves, SW (transverse waves). It will refine
the general location of the defect produced by the RF detection. The angle probe will scan
according to the skip distance (SKD) for the starting and ending points of the PWT with
penetrating sound waves. The SW velocity for brass is 2,110,000mm/sec [29].
Using angle beam evaluation, several parameters are computed for two different
reflection angles (βR) (30°, 35°, 40°, 45°, 50°, 55°, 60°, 65°, and 70°). For optimization
purposes we have named three simplified skip distance (SKD) and sound path (SP)
combinations: ½ SKD is (SKD1), full SKD is (SKD2), and 1½ SKD is (SKD3), ½ SP is (SP1),
full SP is (SP2), and 1½ SP is (SP3), as expressed in the equations below [29]. These are
calculated for each βR.
In our previous work, the review paper for ultrasonic techniques [29], we recall the
equations as expressed below with βR that mentioned above, and the pipe wall thickness
25.4mm. SKD and sound path (SP), as expressed in the equations below. SP is equal to Leg1.
 = 2 ( )
(9.3)

( )
(9.4)
 =
We must add curved section correction for the probe reading perpendicular to the
pipe, as shown in Fig. 9.3. While the other readings may be read directly, as they work
parallel to the length of the pipe. The UT beams will read the outer edges of the defect,
adding shape information to volume and location information. Full SKDArc for arc of the pipe
surface, is given by the following formula [127] [128]. Ri the pipe inner radius, and Ro the
pipe outer radius.
  =
R 

[Sin−1 ( Sin  ) −  ]
90

(9.5)
138
To identify the edge (starting point) of PWT, as shown in Fig. 9.3, we measure the
first leg distance for arc of the wave guide surface (SPC1) as follows [127].

1 2 =  2 +  2 − 2  Cos [Sin−1 ( Sin  ) −  ]

(9.6)
To calculate the βR, if the SPC1 is known [127].
 [Sin (Cos−1
 = Sin−1
1 2 +  2 −  2
)]
2 1 

{
(9.7)
}
Figure 9.3: Representation of full SKDArc for arc of the waveguide surface. SP1,
SP2 and SP3 are the sound paths for straight plate. SPC1 and SPC2 are sound paths
for circular plate
139
9.4 Hybrid RF/UT System
9.4.1 Combination System
In Fig. 9.4(a) and (b) the innovated operational system is shown. The vector network
analyzer (VNA) propagates the microwave signal to detect the PWT edges, and after our
model we will consider the result of the simulation that W/2 before the actual starting edge of
the modeled instances of PWT to W/2 after the actual ending edge contains the actual
location of the PWT.
Figures 9.4 (a) and (b): Is a depiction of the combined hybrid microwave-UT
defect detection system described in this paper
140
Within this range we will optimize the best angle that should be selected for each
position predicted by microwave.
1. Vector Network Analyzer (VNA).
2. Coaxial cable.
3. Port 1.
4. Pipe under test (PUT).
5. UT angle beam probe.
6. U-beam slides the angle beam probe and contains four shower heads to provide a
compliant medium for the probe.
7. Rollers to rotate the PUT.
8. Water supply line.
9. Ground plate.
10. Water reservoir.
11. Water pump.
9.4.2 Sensor Positioning
A hybridized RF/UT technique is described by this paper to determine more accurate
beginning and ending point measurements for a defect, as well as its depth. A similar
technique was described by Part One [122]. This part deals with use of angle-beam UT,
rather than straight beam UT for the second step. For the investigation we used finite
difference time domain software by Computer Simulation Technologies (CST).
As shown in Fig 9.5. The predicted position (LPWT) is measured from port 1 to the
PWT. We then used the concept of manual UT to calculate the positions of the excitation
points (EPn) both backward and forward from the predicted position, as mentioned in our
innovative sensor positioning equation below.
=
Epn = | ∓ ∑  |
=1
(9.8)
141
EP is the excitation point position for each beam angle, n is the number of EP position, LPWT
is the predicted position which was created by microwave. SKDi is the number of skipping
distance for three legs, as we mentioned in section 3 above.
For each leg we program the position of the probe to propagate 3 legs, for a total of 9 legs.
Figure 9.5: Shows the predicted location from microwave detection technique
and the EPn setup for the combination system. LE is the ending point of the
PWT, LS is the starting point of the PWT, and (both measured form port 1) as
written in equation below
 = (() −  ) + ( − () )
(9.9)
LPWT is the predicted length to the defect from port 1, L(PWT)C is the distance from port1 to the
center of the PWT. EP1, EP2 & EP3 are EPn from Equation (9.8).
9.4.3 Sensor Positioning, Results and Discussion
Fig. 9.6 shows the results from three EPn of the probe. EP1 detects the PWT edge in
the first leg (Leg1). EP2 detects the edge in the second leg (Leg2). While EP3 detects the edge
in the third leg (Leg3). The four depths are set up in order to determine the detection
capabilities of the UT system component for detecting shallow to deeper DD.
142
Figure 9.6: Shows four depths of PWT possibly detected by the innovative UT sensor
positioning equation with a 45° angle beam. The microwave system uses two ports
In Figs. 9.7 (a), (b), (c), (d), and (e) show the location of defect (PWT) which are
simulated as locations 07, 08, 09, 10, and 12, selected from the 14 locations modeled. This
presents five examples of the predicted distance from port 1 (circles represent data from the
microwave step), and the optimized EPns used for measuring the location and depth
precisely, with angle beam UT probes (45° angle, represented by the yellow diamonds).
These EPn locations are calculated based on the data from the first step. The calculated EPns
guide the probes into position for the UT test. From these examples it can be determined that
for all four depths, the microwave results are located within detection limits bounded by W/2
ahead of or W/2 behind the actual location of the PWT. All five examples show the potential
of setting the EPns either before or after the predicted location.
In Figs. 9.8 (a), and (b) show the position of PWT which are simulated as positions
13, and 14, selected from the 14 positions modeled. This presents two instances of the
predicted distance from port 1 (circles symbolize data from the microwave step), and the
optimized EPns used for identifying the position and depth precisely, with angle beam UT
transducers (60°, and 70° angle, red triangles and blue squares represent the computed EPns
for each instance).
143
144
Figure 9.7: Shows five modeled PWT examples as noted in each figure, from
the 14 locations modeled. The UT beam angle is 45° for all cases. Circles
indicate the microwave predictions for distance from port 1 at four modeled
depths. The yellow diamonds represent the EPns calculated for each PWT
example for the four depths
145
Figure 9.8: Shows the calculated Epns for UT angle beam probes for two
instances of PWT, and four selected depths. Circles represent the predicted
locations from the microwave step. Red triangles and blue squares represent
the calculated EPns for each case. (a) PWT instance #13 tested with a 60°
angle beam (b) PWT instance #14 tested with a 70° angle beam
146
9.5 Modeling of the Ultrasonic Technique
Fig. 9.9(a) shows the system model for the pipe wall with a PWT section of width
(W). The defect in this case is modeled as a rectangular shaped defect, but the analysis could
be applied to other defect profiles as well. The location of the left and right edges of the
PWT section are denoted by LS and LE respectively and are measured from the Port 1
position for the microwave testing which may be thought of as the left edge of the pipe
section. The exact location and depth of the PWT are not known at this time, as they are
only estimated by the microwave analysis. Here, depth will be measured from the pipe wall
inner surface (without consideration of the defect). The predicted position determined by the
microwave analysis will be denoted by LPWT. From this position, it is desired to place an
ultrasonic transducer at some location EP1 where it will be looking at an angle βR from the
direction perpendicular to the pipe wall. Three skip distances (SKD1, SKD2 and SKD3) are
modeled, as shown in Fig. 9.9(b), where a half skip distance is defined as the distance along
the pipe before the opposite wall of the pipe is hit.
Other sensor locations may be added as well, and in Fig. 9.9(a), three total sensor
locations are considered. As the ultrasonic pulse leaves the sensor, the model must determine
whether the pulse intersects any of the edges of the defect. If an edge is in the pulse path, the
distance is determined from the total sound path (SP), taking into account the number of
reflections encountered from the pipe wall inner and outer surfaces.
Fig. 9.9(b) represents the ultrasonic path for a single ultrasonic sensor location. The
defect is constructed by three line segments. The left edge is modeled as line segment AB,
the top edge by line segment BC and the right edge by line segment CD. The path of the
ultrasonic pulse is divided into three line segments EF, FG and GH. For a specified angle,
βR, the distance from the sensor to each wall intersection point is given by the formula below
[29].
1 =  ( )
(9.10)
Here D is the pipe wall thickness. The calculation of an intersection, then, is simply
determining if the associated line segments intersect. The case of two line segments
147
intersecting is shown in Fig.9.10 the parametric representations of these line segments are. Xa
, Xb , Ya , Yb are the coordinates of the intersection of the defect (PWT) edges.
Figures 9.9 (a) and (b): Global coordinates X and Y are plotted against the
cross section of a pipe wall segment. The segment contains a PWT instance
defined by points A, B, C and D. Skip distance SKD3 is defined by points E,
F, G and H. Note that point H intersects the PWT edge at point D
 = 1 + (2 − 1 )
(9.11)
 = 1 + (2 − 1 )
(9.12)
And
148
 = 3 + (4 − 3 )
(9.13)
 = 3 + (4 − 3 )
(9.14)
Figure 9.10: Intersection of UT angle beam SP and edge of PWT
defect in pipe wall
In these equations U and V represent the line parameter which ranges from 0 to 1 as
the line segment progresses from one point to the other. Values of U and V which are either
negative or greater than one, still lie on the line through the segment, but do not lie within the
segment itself. To test whether the segments intersect, the equations for the X and Y
components are equated and the parametric values representing the intersection are
calculated. Equating Xa and Xb equations produces.
=
(3 − 1 )
(4 − 3 )
+
(2 − 1 )
(2 − 1 )
(9.15)
This equation is valid, except for the case when (X2-X1) equals zero. If points 1 and 2
are used to represent the ultrasonic beam path, this situation will never be encountered. From
the expression of U in terms of the parametric parameter V, the value from the above
equation may be back substituted into the Y parametric line equations and solved for the
value of V, this yields.
149
( − 1 )
(3 − 1 ) − [(3 − 1 ) ( 2
)]
(2 − 1 )
=
( − 1 )
[(4 − 3 ) ( 2
)] − (4 − 3 )
(2 − 1 )
(9.16)
Again, if the calculated values of U and V are both in the range of zero to 1, there is
an intersection. If this is not the case, there is no intersection of the segments.
Using the intersection equations, line segment EF is tested for an intersection with
pipe wall thinning segments CD, CB and AB, in that order. If an intersection is found, the
remaining segments of the pipe wall thinning section are ignored. So, each sensor location
EPi, will return either a single intersection, or no intersection. The combined information
from a number of sensor locations will form a number of intersections with the pipe wall
thinning segments, but other than the location of the intersections, nothing is known as to
which part of the defect was found. By sorting the intersections in both X and Y coordinates,
however, an estimated depth and defect width can be calculated. This allows for an
estimation of the defect volume as well. The only modification if the sensor position is to the
left of the defect is that the left edge is tested first, and then the top edge, and then the right
edge.
The modeling of the ultrasonic sensing array locating a rectangular shaped defect is
just one of an infinite number of possibilities. The important point to note is that all of these
possibilities can be modeled in a similar fashion. In order to demonstrate this point, two
additional defect profiles are considered, a triangular shaped defect and a semicircular shaped
defect. The geometric details of these defect profiles are shown in Figures 9.11 and 9.12. For
the triangular defect, the process is closely aligned with that for the rectangular defect. The
main difference is that the model of the triangular defect consists of only two line segments,
with endpoints denoted by points A, B and C. Following the ultrasonic beam path, the line
segment representing the beam path is checked for an intersection with the line segments AB
and BC representing the triangular defect. These intersections are calculated using Equations
(9.11)-(9.16) and the analysis process beyond calculating all of the intersections between the
ultrasonic sensor beams and subsequent reflections is analogous to that for the rectangular
defect model. The error is anticipated to be greater for the triangular case as in the case of the
rectangular defect model, only a single intersection in the right, lower and left edges of the
150
defect are required in order to precisely determine the dimensions of the defect. In the
triangular case, points as near as possible to the base and the apex need to be determined in
order to keep the predicted width and depth of the defect to a minimum.
Figure 9.11: Global coordinates X and Y are plotted against the cross section
of a pipe wall segment. The segment contains a PWT instance defined by
points A, B, and C. Skip distance (SKD3) is defined by points D, E, F, and G.
Note that point G intersects the PWT edge at point C. Detail of triangle shape
shows the basis of error in the angle beam position
The semicircular defect model relies on determining the intersection between the
various ultrasonic sensor beam angles and the equation of the semicircular arc as shown in
Fig.9.12. The equation of the circle which forms the semicircular defect is given by:
( −  )2 + ( −  )2 =  2
(9.17)
Where XC and YC represent the center of the circle and RC represents the radius of the
circle, as well as the radius of the defect. Letting the equation of the line segment
representing the beam path be represented in parametric form as given in Equations (9.11)
and (9.12), any intersection between the sensor path and the circumference of the circle can
151
be found by equating the equations of the circle and the line segment. After some
manipulation, the results are given by:
Xa and Ya here represent the coordinates for any point along the semi-circle
representing the defect. By substituting the equations (9.11) and (9.12) for the line segment
we simplify the equations above. The equation calculates the coordinates for any point of
intersection between the line segment of the UT beam, and the semi-circle of the defect.
From equations (9.11), (9.12) and (9.17) we derive equation (9.18).
(1 + (2 − 1 ) −  )2 + (1 + (2 − 1 ) −  )2 =  2
(9.18)
To simplify equation (9.18):
1 = 2 − 1
Let
2 = 1 − 
3 = 2 − 1
4 = 1 −  , this yields the equation:
(2 2 + 22 1  + 1 2  2 ) + (4 2 + 24 3  + 3 2  2 ) =  2
2 2 + 22 1  + 1 2  2 + 4 2 + 24 3  + 3 2  2 =  2
(1 2 + 3 2 ) 2 + (22 1 + 24 3 ) + (2 2 + 4 2 −  2 ) = 0
(9.19)
For the quadratic equation we must collect the terms, which yields the equation
below, to solve for U:
− ± √ 2 − 4
=
2
Where
 = 1 2 + 3 2
 = 22 1 + 24 3
and
(9.20)
152
 = 2 2 + 4 2 −  2
Figure 9.12: Global coordinates X and Y are plotted against the cross
section of a pipe wall segment. The PWT instance is defined as a semicircle. xc and yc are the center coordinate of the defect. xd and yd
represent any point on the semicircular circumference of the defect. YC=0
because we measure from the inside of the pipe wall
If the term under the square root is real valued, there will be one or two intersections
located. If the parametric value of u is between zero and one, the intersection is valid. We
need to solve for u (0, 1, 2). Zero (0) indicates no intersection, as shown in Fig. 9.13(a),
where the UT pulse misses the defect. One (1) indicates that the UT pulse is tangent to the
circle, as shown in Fig. 9.13(b). And if the root is positive, that means two (2) intersections,
positive and negative, as shown in Fig. 9.13(c) and we are only interested if the values have
positive yd components. If two intersections are located, the one with the smallest parametric
value, u, is the desired intersection as the second one represents an intersection with the
portion of the circle which is not part of the semicircular defect. The remainder of the model
calculations are handled much the same as for the rectangular and triangular defect. The
intersections are sorted and the maximum and minimum x and y intersection coordinates
form the estimate of the width and the depth of the defect (Semi-Circle).
153
154
Figures 9.13(a) (b) and (c): The PWT model is defined as a semi-circle, as shown
in Fig. 9.12. (a) The UT beam fails to intersect with the defect, yielding a zero
value, due to the reflected beam missing the defect entirely. (b) Represents the
UT pulse tangent to the circle defect. (c) Represents our model with two values,
one positive and the other negative. We are interested only in the positive value
9.6 Optimization of Sensor Location
With the model developed for the ultrasonic sensing, the question becomes where to
position the sensor measurements, and at what angle should the beam be fixed for each
sensor location so that the defect will be found precisely? The sensors are positioned based
on data collected from the microwave step, and passed to the ultrasonic probe, creating
multiple test positions for higher precision. For each location selected, we get data from an
additional three legs of UT signals. Thus, the sensors are specified to be at three locations
determined by our innovative equation (9.8).
With the positions specified, all that remains is the number of sensor locations to
utilize. For this case, three locations are fixed to the right of the defect position predicted by
the microwave testing, and three to the left. This leaves the determination of six beam angles
to be computed. In order to find the angles which will discover the majority of PWT defects,
the ultrasonic modeling was coupled with a genetic optimization algorithm. The genetic
algorithm was chosen in order to avoid local optima and find the global solution. Three
different depths of defect were simulated along a hundred equally placed positions of the
predicted defect location from the microwave testing. If the modeled defect can be located
100% of the time, the optimal beam angles can be utilized in actual testing. Even if the vast
majority of the defects are located within a millimeter or two, the result will be adequate.
The predicted defect position from the microwave testing is known to be within ½ of the
defect width (W/2) on either side of the defect, so this is the simulated range. The objective
function is then to minimize.
155
( ) = ∑|() |
() = |( − ) + ( − )|
(9.21)
(9.22)
Where
(Error)ij = (estimated width of defect– actual width) + (estimated depth of defect– actual
depth)
This error is summed for each of the three depths of defect considered, and at 100
locations in the range of the microwave estimate. Thus, given each of the six beam angles, it
is a straightforward objective function to compute. Here, the subscript i represents the defect
depths tested, and the subscript j represents the 100 positions for the predicted defect position
from the microwave analysis. The design variables are the six beam angles utilized for the
ultrasonic measurement positions. The optimization was tested on defects ranging from a
depth of 10 to 20 mm and a defect width of 25 mm. Variable bounds were established in
order to keep the beam angles within certain bounds. These constraints are.
( ) > 30o ;  = 1, 2, … ,6
and
( ) > 55o ;  = 1, 2, … ,6
The results of the optimization yielded a set of angles of (32o, 37o, 46o, 49o, 43o, 39o)
for rectangular shaped defect. The first three angles are used for the sensors to the right of the
predicted defect position, and the last three are used for the sensors to the left of the predicted
defect position. The defect was located to within one millimeter for all cases for this optimal
design (as shown in discussion figures 9.14-9.16).
9.7 Genetic Optimization Algorithm
The optimization algorithm employed is similar in detail to that presented by
Goldberg [129]. Genetic optimization algorithms mimic nature by selecting the fittest
156
members in a population of alternative designs for survival. A single design is represented by
an encoding, which in this case consists of the six beam angles for the ultrasonic sensor
scans. An initial population is generated randomly, with each value of design variable being
selected as a random integer between the prescribed upper and lower limits. This initial
population of designs is then acted upon, producing new generations of designs through
operators including parent selection, crossover and mutation. The expectation is that as the
generations progress, both the average objective function value, or fitness, will improve as
will the best solution in the population. The process is repeated for a fixed number of
generations at which time the process is terminated and the best design located to that point is
deemed the solution.
The advantage of the genetic algorithm is that, by utilizing a population of designs, it
searches the design space on a global level, making it less likely to terminate at local minima.
It is well suited for representing design variables when each variable can assume one of a set
number of alternatives. Some experimentation is needed in order to determine the population
size and the number of generations. In this particular case, a population of 1000 designs was
utilized, and a total of 100 generations was executed. Runs were made from several randomly
generated starting points, and the error contained in the solution was extremely small in
every case.
In this particular problem, the variable bounds were included in the design variable
selection. If additional constraints were added, a penalty function would be employed in
order to satisfy the constraints gradually as the search progressed. Different defect profiles
are easily handled by altering the model to include the new profile. As long as the defect
edges can be modeled by a series of line segments, the same approach can be easily utilized.
Extensions, including varying the sensor locations within the optimization, could also be
considered, but since a usable solution was generated with the locations fixed (depending
only upon the angle selected), this was not necessary. The genetic algorithm performed well
and virtually no fine tuning of the algorithm parameters was required.
The next step is to consider the small adjustment of each sensor location as additional
variables in the optimization for the rectangular as well as other defect profiles.
157
9.8 Optimization Results and Discussion
9.8.1 The Rectangular Defect, all Parameters Fixed, but Angle of Beam
Varies
From the calculated error in depth and width measurement of the defect, shown
respectively in Figures 9.14(a) and (b), it is clear that even with all ultrasonic beam angles set
at 45 degrees, the PWT defect can be located reliably, but there is up to a 20 percent error in
the predicted width, and up to a 35 percent error in the predicted depth of the PWT. Even
with a fixed angle of 45 degrees, the ultrasonic sensor locations specified by the sensor
positioning equation perform quite well, but the question remains whether or not these results
can be improved upon. To address this question, two separate optimization results are
presented. The first is a minimization of the total error found during the simulations, and the
second is a minimization of the maximum error found during any of the single simulations.
The error in depth and width for the defect from the minimization of the total error is
presented in Figures 9.15(a) and (b), respectively. Note the significant errors found in the
fixed angle simulation have now been largely eliminated. The maximum error in the
predicted depth is now below 2.5 percent and the error in the width has been reduced to
approximately 6 percent. In addition, these errors occur in a very small percentage of the
total simulations evaluated. The optimum beam angles for this solution are 32, 37, and 46
degrees for the sensor positions to the right of the defect position predicted by the microwave
simulation, and 49, 43, and 39 degrees for the sensing positions to the left. With few errors
and with the small magnitude of the errors, the technique is demonstrated to be quite accurate
in detecting this type of wall thinning defect.
In order to see if the error could be reduced even further, a second optimization was
conducted with the objective to minimize the maximum error found in any of the
simulations. The error in the depth and the width of the defect are presented in Figures
9.16(a) and (b), respectively. Here, the results show that the depth error increased slightly
over the minimization of the total error solution, but the width error was reduced. The
optimal beam angles were found to be quite different, however, with values of 51, 46, and 42
158
degrees for the right sensor locations and 55, 49, and 46 degrees for the left sensor locations.
The large differences in the two solutions demonstrate that there are a number of quite good
solutions, as long as the individual angles are selected appropriately. In any case, the viability
of the approach has been demonstrated.
Figures 9.14(a) and (b): The baseline represents UT at 45° angle. Left peak is
minimization of total error found during the simulations. Right peak is minimization
of maximum error found in any single simulation. (a). There is up to 35% error in
the predicted depth. (b). There is up to 20% error in the predicted width
159
Figures 9.15(a) and (b): Optimum beam angles represented are 32°, 37°, 46°
(sensing locations to right) 49°, 43°, and 39° (sensing locations to the left).
Maximum error reduced due to optimization of beam angles. (a). Maximum
error in predicted depth is less than 2.5% (b). Maximum error in predicted
width is about 6%
160
Figures 9.16(a) and (b): Optimum beam angles represented are 51°, 46°, 42°
(sensing locations to right) 55°, 49°, 46° (sensing locations to the left).
(a). Depth error increased slightly over minimization. (b). Width error
was reduced slightly
161
9.8.2 Adjustments of Beam Positions and Consideration of Other Defect
Profiles
The optimization to this point only considered the beam angles, and the type of defect
was limited to a rectangular profile. Equation 8 allows for the positioning of each sensor, but
as the type of defect varies along with the width and depth of the defect, it is convenient to
allow for a small shift in the position of each sensor location. This is easily accomplished by
including a shift variable for each of the six sensor locations in the genetic algorithm. The
results for the inclusion of this shift are presented in Figures 9.17-9.21.
In Fig. 9.17, the results are documented for the rectangular defect with the shift
variables included. The minor errors in detecting the width and depth of the defect present in
the previous optimization are seen to be completely eliminated by relatively small shifts in
the sensor positions. It should be noted that many similar solutions are available from the
genetic algorithm with different angle and sensor positions for the ultrasonic sensors.
Figure 9.17: Represents the error for the rectangular defect, based on the
width and depth
The rectangular defect profile is much easier to detect than a defect profile of another
type, such as a triangular defect or a semi-circular defect. In order to test the ability of the
162
ultrasonic sensors to locate these defect profiles, optimization runs were conducted replacing
the rectangular defect with a triangular and then a semicircular defect. The results for these
two optimizations are documented in Figures 9.18 and 9.19. It can be seen in these figures
that the errors in predicting the width and depth of the defect are greater than that for the
rectangular defect. This is due to the precision required in locating the top and bottom of the
defect, even as it moves in location and depth. The fact that the defect is consistently located
and the estimate of depth and width are relatively low is another indication of the value of
this approach. The errors are 30 to 40 percent in very limited regions, but for the most part,
they are quite accurate. Fig. 9.11 triangle shows the source of error in the angle beam
location technique for triangular profile defects. The three points in the detail of triangle
shape (one point in the left and two points in the right) intersect the walls of the triangular
defect, but these three points can also be read as a rectangular defect signature, as shown by
the hidden line. Note that the apexes of the triangle are the source of larger errors. It is
difficult to locate the vertical apex of a triangular defect. The primary advantage of this
approach is that the defect can be located, although in the case of a triangular defect profile it
cannot be characterized accurately.
Figure 9.18: Shows the error for the triangular defect based on the width and
depth
Finally, a general defect is considered. In all three of the above cases, the form of the
defect profile was known and the optimization was conducted with this knowledge. To
163
further test the approach, a final optimization was executed without knowledge of the defect
type (rectangular, triangular or semi-circular). For this case, all three defect types were
included and the total error was again minimized for various defect positions and depths.
The results from this optimization are detailed in Figures 9.20 and 9.21. The results are
again quite good and certainly comparable to the optimization results for the individual
defect types.
Figure 9.19: Shows the error for semi-circular profile defects based on width and
depth
Figure 9.20: Represents the width error for all defect profiles
(rectangle, triangle, semi-circle)
164
Figure 9.21: Shows depth error for all types of defect (rectangle, triangle, semi-circle)
Table 9.1: Design variable values at their optimal levels for all of the optimization
formulations
Design Variable
Right
Left
Rectangular
Triangular
Semi-Circular
All Combined
βR1 43°
41°
48°
49°
βR2 32°
35°
45°
30°
βR3 53°
36°
46°
47°
βR4 33°
36°
44°
34°
βR5 47°
31°
42°
31°
βR6 47°
34°
43°
35°
Shift1
-06 mm
+08 mm
+01 mm
-02 mm
Shift2
-12 mm
-09 mm
-11 mm
-11 mm
Shift3
+04 mm
-07 mm
-07 mm
-03 mm
Shift4
+02 mm
-06 mm
-05 mm
-08 mm
Shift5
+00 mm
+08 mm
+09 mm
+07 mm
Shift6
+06 mm
+04 mm
+09 mm
+03 mm
165
Table 9.1 lists all of the design variable values at their optimal levels for all of the
optimization formulations. From this table, it can be seen that the optimal angle and position
shift values for each of the cases have differences, but there are some distinct similarities.
This is especially true for the cases of the triangular defect detection and the unknown defect
profile detection cases. This is expected as these are the most challenging cases for defect
detection.
In the Table 9.1 the beam angles (B1 to B6, designated as Left or Right positions
relative to the defect) and shift variables (shift 1-6) for each of the six sensor locations in the
genetic algorithm. The results for the beam angles and of the positional shifts are presented in
the table for each of the three defect profiles and the combined results.
9.9 Experimental Validation
A study by Linsheng Liu et al (2011) [34] examines the propagation of microwave
signals inside a pipe, a study which we confirmed using CST. Fig. 9.22 shows the
configuration used in the numerical simulations. The CST simulations used a mesh type
hexahedral (legacy) model of a pipe with PWT going the full circumference of the pipe and
located at the approximate middle of the experimental pipe. In the cited study two different
PWT cases were used. Both cases included a rectangular cross-section with a depth H and
length W. We verified one case. The dimensions used in their experimental studies are
summarized in Table 9.2. In our simulation we modeled pipe with an inner radius of
R=8.5mm, which produced the lowest and highest cut off frequencies of 13 GHz and 21 GHz
respectively, using equations (9.1) and (9.2). The lengths of two pipe sections, P1 and P2,
were set to 453 mm and 455 mm, with wall thickness t=1mm. The two pipe sections were
joined by five joint pieces with W=17mm and varied inner diameters. P1, P2 and the five
connectors were combined to create a series of experimental pipes. These simulated
components allow for many joint thicknesses, with variable locations and amounts of PWT to
be analyzed. The parameters of the various components are laid out in Table 9.2. According
to Fig. 9.22 the author was unable to separate the PWT start point and end point at the time
of his experimental study. The signal resolution yielded overlapping signals of reflection,
166
with indistinct peaks when the start and end points of the PWT were close to one another.
Pipes with inner diameters from 12.7 mm to 152.4 mm are used in distribution pipelines. The
17mm inner diameter pipe used in our study falls at the lower end of the distribution pipeline
range.
In our future studies we will use pipes with inner diameters of 127 mm to 254 mm.
The pipe materials to be used will be selected at the time of the experimental study. The
options are aluminum, brass or copper.
Figure 9.22: Evaluation of the simulation results for the PWT discontinuity locations
(Comparison of Result)
Table 9.2: Parameters of Experimental PWT
Joint’s Number
1
2
3
4
5
Inner Radius, R (mm)
8.55
8.6
8.7
8.9
9.1
H (mm)
0.05
0.1
0.2
0.4
0.6
From Fig. 9.22 we see that the computational results agree well with the experimental
data, but it is understood that the computational approach will have some error in predicting
167
the actual defect location. This error is seen as a shift between the experimental and
computational results. While the trend is predicted quite well, the precise location is off by
some small value. The optimization of the ultrasonic sensor locations allows the sensors to be
placed so that the error in predicting the exact defect location is reduced by realizing this
shift and accounting for it in the best possible manner. This essentially means that our
expression for the sensor locations given in Equation (9.23) is actually in the form of.
=
Epn = | ∓ ∑  | + (ℎ )
(9.23)
=1
By considering this form where the shift is minimized by the optimal placement of
the sensors, the proposed sensor placement equation is validated.
9.10 Conclusion
The angle beam techniques (manual and automatic) concept for UT are used to
calibrate the microwave detection point in order to find the structural boundaries (edges) of
the PWT.
In Part One (using straight beam UT) it was difficult to detect the edge of PWT
precisely. In Part Two, using angle beam UT, by modeling the UT using a genetic algorithm,
we find the edges with a low percentage of error (2.5% error for depth, and 6% error for
width) a very precise measurement.
Use of UT for precise measurement is complex but straightforward. Optimization of
the EPn for detection of PWT edges is critical to the success of the hybrid system.
Positioning of the sensor probe is most effective if the angle of the beam can be varied, while
the excitation positions ahead of and following the predicted location (from the microwave
step) are determined using a genetic optimization algorithm. This is all computing, but allows
for minimal measurement error in the second (UT) step. The three legs for each EPn (9 legs
total) allow accurate measurement of both defect edge location, and depth (DD)
determination.
168
The two-step hybrid system, combining microwave detection and prediction with
ultrasonic angle beam measurement allows for the system to be fully automatic, although
under control of a human operator. The results of the test would be compiled and reported
quickly in order to determine the disposition of the PUT (in or out of spec).
The system developed is very flexible, allowing for beam angle optimization, and
optimal positioning of the UT probe in step 2. By automatically adjusting beam angle and
EPns according to the modeled results, the UT step can determine location and depth with
high accuracy, and can characterize the general profile of the defect from outside the pipe
wall. To validate the innovative sensor positioning equation (the hybrid technique, Equation
(9.8)) according to the individual defect shapes we have maximum and minimum adjustment
shifting terms. Rectangular: maximum 0mm, minimum -12mm. Triangular: maximum
+4mm, minimum -9mm. Semi-circular: maximum +9mm, minimum -11mm. In reality we
cannot find only rectangular, triangular or semi-circular defect shapes. There are more
options. Therefore the combined maxima and minima are more useful. They are combined
maximum +7mm, combined minimum -11mm. This is why the shifting term is added to
Equation (9.8), to be as an equation EPn.
169
Chapter 10
Experimental
Evaluation
of
Novel
Hybrid
Microwave/Ultrasonic
Technique to Locate and Characterize Pipe Wall Thinning [130]
Abstract
Research using microwaves to detect pipe wall thinning (PWT) distinguishes the presence of
wall thinning, but does not accurately locate the discontinuities. Ultrasonic testing is capable
of accurately locating the PWT defect, but cannot do so without time-consuming linear
scanning. This novel work combines the microwave technique as a way to predict the
location of a series of PWT specimens, and the UT technique as a way to characterize the
PWT specimens in terms of location, depth and profile shape. The UT probe is guided to the
predicted location derived from the Phase One microwave results, generating the Phase Two
results to determine accurate location, depth measurement and profile shape detection. The
work uses the previously successful experimental setup for testing of aluminum pipe with
154.051 mm ID pipe of 1 m length. A vector network analyzer generates a sweeping
frequency range of 1.4 to 2.3 GHz. This signal is propagated within reference pipes with both
open end and short-circuit configurations for calibration of the system. The calibrated system
is used to detect the presence and location of six PWT specimens, with two profile shapes, at
three depths of thinning and three locations along the pipe. The predicted locations from
Phase One are then used to guide a calibrated, manually-guided straight beam UT probe to
the predicted position. From that point, the UT probe is used in order to accurately localize,
and determine the depth and shape profile of the specimens.
10.1 Introduction
Nondestructive testing (NDT) techniques using microwaves have been proposed and
experimentally tested with the aim being rapid and remote inspection of the inner walls of
tubular pipes to detect degradation of the surface [25] [131] [132] [133] [134]. Basically, the
test propagates microwaves through the pipe, and captures the reflected microwave energy. A
170
smooth inner wall reflects the energy with very little attenuation. Flaws in the pipe wall, in
this case pipe wall thinning (PWT) generate specific reflections which indicate anomalies in
the inner surface of the pipe. There is no need to move the probe emitting the microwaves; it
can remain stationary throughout the test. This allows for a very rapid non-destructive
inspection of the interior surfaces of long tubes or pipes [134].
In practical use, most tubes or pipes are under pressure. Therefore, thinning of the
pipe wall creates areas of reduced strength, where the weakened pipe wall might fail.
Detection of wall reduction before it becomes unacceptably thin is a very important goal for
non-destructive testing of pipes both during manufacture and in service [37] [90] [113]. Most
of the experimental studies of microwave detection of pipe wall thinning have focused on
tubes, confirming the principle idea that the technique can detect the presence of such flaws
in small inner diameter pipes [32] [37].
Confirming the location of the PWT flaws with microwaves has also been tested, with
variable results [25] [75] [93] [135]. The microwave reflections can locate the instances of
PWT in a general sense, but without specificity. The reflection may indicate that the location
is ahead of or behind the actual position of the PWT, as well as indicating an accurate
location in some circumstances [34]. Accurate localization of a PWT instance is necessary
for NDT to be useful. Another type of information that is desirable is to know the depth,
length, width and shape of the defect, but to date, with microwave detection there are
problems with characterizing the flaw.
Non-destructive testing (NDT) techniques are often paired in order to take advantage
of the strengths of each technique. In this case, we pair microwave testing for detection and
general localization of PWT defects with Ultrasonic testing (UT) for precise location and
better characterization.
UT is an accepted and commonly-used type of NDT. The strength of the UT methods
is that unseen flaws can be detected by the ultrasonic signal. Frequencies ranging between
0.1 and 25 MHz are sufficient for detection of hidden flaws. UT can be used to measure the
thickness of a pipe wall, which makes it a candidate for locating PWT defects. However, the
probes must be moved across the external surface of the pipe under test, which is time-
171
consuming. Pairing another, rapid localizing technique with UT allows quick positioning of
the UT probe to a known location, reducing the overall time required for the test [29] [46].
In our previous work Computer Simulation Technology (CST) software was used to
model the microwave technique to localize the PWT. The model included a sweeping
frequency range from 0.7 GHz to 2.0 GHz, to ensure that the frequency necessary for
generating reflections from any PWT defects, was introduced. Then, a theoretical analysis of
UT straight beam was modeled to simulate further localization and characterization of the
PWT defects. These simulations were successful in detecting and locating the modeled PWT
defects using both types of NDT in concert [122].
Previous experimental work was successful in detecting PWT in tubes of smaller
diameter [37] [75] [134] [135]. This experimental study is the first known to be carried out
with pipes having standard diameters used in industry.
This study addresses two components of what is intended to be a hybrid NDT system
for pipes, with two phases of detection and locating defects in pipe walls. The individual
components, a microwave test, and an ultrasonic test, have been subjected to practical testing
in separate investigations. The current work is a report of the results of practical
experimentation based on the results of the modeled studies. The practical experiments
confirm the simulation results.
10.2 Specimens under Test
10.2.1 Pipe Sections
Five 6061-T6 aluminum alloy pipe sections were prepared from 6.065 in. (154.051
mm) inner diameter (DI) stock as shown in Fig. 10.1, with outer diameter (DO) 6.63 in.
(168.402 mm), a readily available, standard size pipe. The sections were machined in
Imperial units, but we have converted the actual lengths to mm for the paper. The five
sections used were: LP (reference pipe) 39.37 in. (1000 mm) length; LP1 (Half-length) 19.68
in. (500 mm); LP2 (approximately five-sixths length) 32.7 in. (830.58 mm); LP3
172
(approximately one-sixth length) 6.8 in. (172.72 mm), each used in the quantities shown in
Table 10.1. These were used to construct the pipes under test for this study.
Figure 10.1: Illustration of 6061-T6 Aluminum Alloy Pipe Used for the Study
Table 10.1: Full Dimension and Quantity for Aluminum pipe sections used in the experiment
Designation
LP
LP1
LP2
LP3
Length in. (mm)
39.37 (1000)
19.68 (500)
32.7 (830.58)
6.8 (172.72)
Quantity
1
2
1
1
10.2.2 Ring Sections, Fabrication
The rings represent the pipe wall thinning (PWT) examples for this experiment.
Computer Aided Manufacturing (CAM) was used to create six specimens from 6061-T6
Aluminum Alloy solid cylindrical bar stock, to very close tolerances, permitting friction
fitting when building the pipes under test (PUTs). The nine specimens were designed to
present pipe wall reductions with known characteristics and cross-sections to test the
microwave and ultrasonic systems scaled-up to real-world dimensions.
Two shapes of defect were fabricated: rectangular, and semi-circular, as shown in
Fig. 10.2. Each shape was machined at three systematically increasing depths: 0.2 in. (5.08
mm), 0.4 in. (10.16 mm), 0.6 in. (15.24 mm), as shown in Table 10.2. Each was embedded in
standard rings with inner ring diameter (DRI) 6.065 in. (154.051 mm); outer ring diameter
(DRO) 7.5 in. (190.5 mm). The width of all the rings is 3.0 in (76.2 mm). The width of the
wall thinning for all specimens is 1.0 in (25.4 mm).
173
Figure 10.2: Design Drawing of the Standard Ring Showing Dimensions and
Structural Fitting
Figure 10.3: Design drawings of the shapes of PWT Showing Shape and Dimension
of the Groove
10.2.3 Cap Fabrication
An aluminum cap was created from 6061-T6 Aluminum Alloy solid cylindrical bar
stock, using Computer Numerical Controlled (CNC) machining, to hold the probe at the
input port. The cap is ellipsoidal in cross-section as shown in Fig. 10.4. Which shows the
details of the cap construction. The length of the cap (LCAP) is 5.4 in. (137.7 mm).
174
Table 10.2: Dimensions of the Rings and Wall-thinning Specimens
Depth of Defect (DDI) Shapes and Quantities
in. (mm)
Rectangular
Semi-circular
0.6 in. (15.24 mm)
1
1
0.4 in. (10.16 mm)
1
1
0.2 in. (5.08 mm)
1
1
Figure 10.4: Design Drawings for the Cap Fabrication Showing All Dimensions
10.3 Experimental Set-up
In Fig. 10.5 the Vector network analyzer (VNA) is shown connected to the shortcircuited reference pipe for short-circuited calibration. At the left end is a 12”x12” (304.8
mm x 304.8 mm) gold-plated conductive plate, used because gold has very high conductivity,
creating the closest approximation to a true short-circuit. The ellipsoidal cap is connected to
the right end of the pipe, and holds the “pigtail test probe cable” that is connected to the
VNA. The jacket diameter is 1.19 mm, and the material is copper, tan. The solid inner
175
conductor is a copper clad steel wire, silver plated. The probe conductor protrudes 30 mm
into the ellipsoidal space inside the cap.
Figure 10.5: Experimental Set-up for Short-circuited Case Used to Calibrate the
Group Velocity of the Waveguide with Inner Diameter of 154.051 mm, and 1000 mm
Length
Figure 10.6: Experimental Set-up for the End Location for PWT Specimens. Here, the
specimen ring is located at approximately 5/6 the length of the reference pipe from
the ellipsoidal cap Length
176
After using the reference pipe LP with the plate at the end, to make the short circuit
calibration. We combined two pieces for each test specimen (location of PWT), and we used
six samples of PWT for each open-ended test. For the first location, we connected LP3 to LP2
to create the location Front (F) which is approximately 1/6 the reference pipe length from the
cap. We then reversed the pipe relative to the VNA, to be LP2 to LP3, to make the End (E)
configuration, where the PWT is located approximately 5/6 the reference pipe length from
the cap, as shown in Fig. 10.6.
10.4 Theoretical Analysis
10.4.1 Confirming Microwave Technique
10.4.1.1 Sweeping Frequency Bandwidth Measurements
Measurements were performed using an HP 8510C network analyzer equipped with a
26.5 GHz S-parameters test set. A short coaxial cable was used to connect the test set to the
calibration standards and probe. The network analyzer was set to perform a frequency sweep
for each measurement, spanning from 1.4 GHz to 2.3 GHz.
This corresponds with the bandwidth between the first two transverse-magnetic (TM)
propagation modes which occur within the waveguide, TM01 and TM11.
 =
 

(10.1)
fcTMnm is the cut-off frequency for the range of sweeping frequencies TM 01 and TM11. Di is
the inside diameter of the aluminum pipe. Cd is the speed of light within the free space inside
the waveguide, Cd = C = 3x108, if the pipe is occupied by an air column (permittivity=1). Pnm
is the mathematical root for mode TM01 and TM11, (P01=2.4048, and P11=3.8317).
For a radially symmetric coaxial probe centered within a circular waveguide,
transverse-electric (TE) modes do not propagate [34] [136] [137] [138]. The network
analyzer measured 801 discrete frequency points within this region, the maximum number
177
allowed by the network analyzer. This was chosen to maximize the resolution of the
measurements in both the time domain and frequency domain.
To minimize the effects of random errors and transient noise, the measurement at
each frequency point was averaged over 16 readings. A full 1-port calibration was done to
compensate for systematic errors within the system. Three calibration standards were used
for this procedure:
open-circuit, short-circuit and matched-load. Each standard was
connected to the end of the short coaxial cable, then measured by the network analyzer. Once
all three standards were measured, the network analyzer corrects for any contributions to the
measured scattering parameters by the analyzer itself or the coaxial cable. The resulting
measurements capture only the response due to the coaxial probe itself and the pipe.
10.4.1.2 Time Domain Analysis with Inverse Fast Fourier Transform
(IFFT)
As network analyzers are only capable of performing measurements in the frequency
domain, a transform such as the Inverse Fast Fourier Transform (IFFT) must be used to
obtain time-domain information [34].
To obtain the results presented here, the network analyzer’s internal time-domain
transform was used. The transform was performed over the interval from 5 ns before the first
strong reflection on to 50 ns after it. Since the transformation is performed over a finite
frequency span with a finite number of discrete data points, some spurious effects will
always appear in the time-domain data. A Kaiser windowing function (window beta = 6) and
time-domain gating over the same time interval was applied to the data to reduce these
effects to 20 dB to 30 dB below the desired time-of-flight information.
10.4.2 Confirming Straight Beam UT Techniques
The beam from a UT probe is emitted by a transducer, travels from the outer surface
of the material under test, and then reflects from the back wall of the material to the detection
178
probe. If there is a discontinuity present, the reflection will be from a shorter distance,
because the discontinuity will be within the wall itself. The velocity of the UT signal in the
material is dependent on the material being tested. Therefore, in order to correctly measure
distances traveled by the UT signal, the probe we used had to be set for and calibrated to the
specific characteristics of the aluminum material we used for our test pipes [29].
The UT Longitudinal waves (LW) travels a relatively short distance, but it still
diffuses slightly within the material. The spread angle of the beam can be calculated for any
material using Equation (10.2).
sin
θS 1.22λLW
=
2
DC
(10.2)
“Where λLW is the longitudinal wavelength, DC is the diameter of the straight-beam crystal.
High frequency LW diffuse less.” Wissam M. Alobaidi et al. [122].
10.5 Calibration Conditions
10.5.1 Calibration of Group Waveform Velocity
The operating frequencies (f) of microwaves in an air-filled pipe (circular waveguide)
are related to the working mode. The group velocity (VC) is a function of this relationship.
The cutoff frequency (fcTMnm) as mentioned in Eq. 1 is dominant TM01 (P01=2.4048), the
wavelength of this cutoff frequency is λg [34].
As mentioned by Lenshing Liu et al [34] the location of a discontinuity (PWT) in a
pipe cannot be determined by microwave energy, without knowing the group velocity, which
is calculated using the equation:
01 2
 =  . √1 − (

1
) ∝ (λ )

(10.3)
Equation (10.3) is insufficient to determine the group velocity for an entire sweeping
range, because it is meant for a single frequency. In this technique a calibration method must
179
be used to confirm the group velocity, because a sweeping frequency range is necessary to
insure high resolution in the time domain [34].
The group velocity can be determined as shown in the Equation (10.4):
 =
2

(10.4)
See the detail in the results sub-section 10.6.1.1.
This technique, determined by Linsheng Liu et al [34], is adapted for use in the
microwave component of this study.
The study specimens consist of five trial pipes dependent on the location of the PWT:
a control pipe (reference pipe), open-end; a control pipe, short-circuit end; defect at
approximately one-sixth of pipe length from port 1; defect at half pipe length from port 1;
defect at approximately five-sixths of pipe length from port 1.
Each trial pipe also incorporates two specimens of full-circumferential shape defects,
at three depths. This yields a total of 20 datasets, including the two reference pipes. By
considering the reference pipe with short-circuit, we calibrate the group velocity. We set this
as the group velocity to calculate the PWT for the 18 trials.
10.5.2 Calibration of UT Probe
Here we begin calibrating the aluminum material according to the block shown in
Fig. 10.7. The block consists of two parts. The dimensions of the thicker section are 25.4 mm
high, 50.8 mm width and depth. The thinner section is 5.08 mm height, 50.8 mm width and
depth. The UT device should be calibrated to these thicknesses to cover all the measurements
we need to make. This calibration is a standard procedure from ASME and ASNT. To
calibrate the straight beam UT dual probe we use “NDT Systems Ultrasonic Couplant”
UTGII02, as shown in the Fig. 10.7 below. The UT device used is a manually operated dual
probe as shown in the photo below. The device is set for aluminum material. This is a simple
procedure “manual UT” just to calibrate the thickness of the material.
180
Figure 10.7: (a) shows the calibration of the UT straight beam probe to the 1” block
(25.4 mm). (b) Shows the calibration for the 0.2” block (5.08 mm). (c) Calibration
block to cover the range of all possible measurements in the experiment. (d) The
Ultrasonic Couplant bottle
10.6 Results
10.6.1 Microwave Testing Results
10.6.1.1 Calibrating the Group Velocity Using Time Domain Analysis
According to Equation (10.4), and previous work, Lp is 1.0 m, which must be
accounted for twice, since the signal must travel to the end of the pipe, and then reflect back
to the transmitter port.
181
Lenshing Liu et al. [34] explained the particulars of this property, but his reference
pipe was only the first segment of the combined pipes they tested.
For the current work, the reference pipe chosen is the full length of the pipes under
test (PUTs). The total length of the partial pipes plus the defect rings used as specimens is
always the same length plus 5.08 cm. The additional length results from adding the ring to a
1m pipe that has been bisected at 1/6 or ½ the length, and then adding a defect specimen to
the resulting sections. For example, as shown in the Fig. 10.8 below, a second 1m pipe was
fabricated, and cut for the 1/6 length section. This test pipe was reversed to give two specimen
positions: at 5/6 Lp, and at 1/6 Lp from the port.
The delay time (time of flight) of ΔTC is 11.07 nanoseconds (ns) along the pipe; and
the calibration velocity is VC = 0.1807 x 109 m/s.
Figure 10.8: Results in Pipe Under Test (PUT) of 1m plus 0.0508m
10.6.1.2 Estimate Location of PWT Using Microwave Signal
As we mentioned we calibrated the pipe in two conditions. First, REF1: Reference
pipe without metal reflecting plate. Second, REF2: Reference pipe with metal reflecting
182
plate flush against the open end of the pipe. We considered three locations of the defect. At
1
/6 the length of the pipe from the transmitting port is location F: defect near front of pipe
(closest to probe). At ½ the length of the pipe from the transmitting port is location M: defect
in middle of pipe. At 5/6 the length of the pipe from the transmitting port is location E: defect
near end of pipe (farthest from probe).
As we worked with three locations, we worked with two defect shapes and three
depths of defect. In Fig. 10.9 the legend is set up as FR6, FR4 and FR2. F is the front
location, R is the rectangular defect, and the number (6, 4, 2) refers to the depth, which is
actually in imperial measure, thus inches. In Table 10.3 all the information pertains to the
PWT itself. The measurements are converted to metric measurements.
Table 10.3: Explanation of identification codes used in experimental graphs (10.9-10.21)
No.
Code
Description
Function
1
F
Front location closest to probe
Location of PWT
2
M
Middle location, half way down pipe
Location of PWT
3
E
End location farthest from probe
Location of PWT
4
R
Rectangular
Shape
5
C
Semi-circular
Shape
6
6
0.6 in. (15.24 mm)
Depth
7
4
0.4 in. (10.16 mm)
Depth
8
2
0.2 in. (5.08 mm)
Depth
In Fig. 10.9 there are three humps. The first as circled in D-1, is caused by the probe
and cap, and the one beside it varies according to the depth of the PWT. The third hump as
circled in D-2, is caused by the end of the open pipe under test (PUT).
Focusing on the area inside circle D-1 (see Fig. 10.10) from the first curves,
generated by the probe and cap, it becomes clear that the height of the curve is influenced by
the presence of PWT, and this effect is proportional to the depth of the defect, even though
183
the defect is not generating these curves. This effect is not large either in amplitude or in the
time domain, but it is measurable. For example, the time of flight (TOF) varies from 1.80625
ns to 1.875 ns.
Figure 10.9: Experimental result for rectangular shape PWT specimens
FR6, 4 & 2. Placed at the front location of the pipe, with the group velocity
method applied to the raw MW reflection data
Figure 10.10: Experimental data for the reflection MW signals from the
probe and cap for specimens FR 6, 4 & 2
184
Figure 10.11: The first three reflections according to the PWT specimens
FR6, 4 & 2 peaks, with TOF ranging from 4.83125 ns to 4.96875 ns
Fig. 10.11, details the second humps inside D-1, where the humps of three curves
represent the actual case study, which are specimens of rectangular PWT at three depths in
the front location. It is very clear from the magnitude of the curves that we have for FR2 dB
is -20.153, and FR4 the dB is -14.6289, and finally, for FR6 the dB is -10.2313. The TOF
varies as 4.96875 ns, 4.9 ns, and 4.83125 ns, respectively. It is clear from the graphics and
the numbers that there is difference according to the volume of the PWT.
10.6.1.3 Determine Effects of PWT Shape on the Microwave Signal TOF
Still considering the same calibration pipe, the group velocity will now be used to
characterize TOF effects caused by the shape of the PWT. The only portion of the pipe that is
changed is the ring containing the PWT specimen. With this specimen, we change the
contour of the PWT to semi-circular, rather than rectangular as in the previous discussion.
These examples are presented in order to demonstrate the effects that the shape of defect has
on the microwave curves. This specimen retains the same location and same three depths. We
use FC2, 4 and 6 in this section.
185
In Fig. 10.12 the first hump inside circle D-3 is generated by the probe and cap
reflections as in the previous example. The three curves adjacent to this represent the semicircular PWT defect. The waveform shape is very close if you compare the FR reflections to
the FC reflections. The only two things that vary in this hump are the dB and TOF numbers.
Fig. 10.13 shows the reflections from the probe and cap. The dB in this case varies
from -12.6297 to -12.8809. And, the time of flight (TOF) again for the three depths FC2,
FC4 and FC6 are 1.80625 ns, 1.80625 ns, and 1.875 ns. This proves to us that changes in
volume of the PWT specimen cause only slight changes in the parameters of the reflection
from the probe and cap.
Fig. 10.14 details the curves generated by the defect specimen, in this case a
semicircular PWT at three depths. As before, we have clear differences in the amplitude of
the waveforms: for FC2 dB is -22.4017, and FC4 the dB is -16.9595, and finally, for FC6 the
dB is -12.619. The TOF varies as 5.0375 ns, 4.9 ns, and 4.83125 ns, respectively. It is clear
from the TOF numbers that there is a change from the FR specimens, only for the FC2
specimen. The change in PWT shape is obvious from the changes in dB amplitudes. After all
this, by using Equation (10.4), we find out the detection.
Figure 10.12: Experimental result for the PWT specimens FC6, 4 & 2,
semi-circular shape at the front location, after the group velocity method is
applied to the MW reflection data
186
Figure 10.13: Reflected experimental MW signal data from the probe and
cap for samples FC 6, 4 & 2
Figure 10.14: Reflection peaks from the PWT samples FC6, 4 & 2,
showing TOF ranging from 4.83125 ns to 5.0375 ns
187
10.6.1.4 Effect of PWT Location on TOF
As shown in Fig. 10.15, the waveforms inside circle D-5 represent the probe and cap
reflection. D-6 with the various waveforms, represents the rectangular PWT in the middle of
the pipe. The last hump, in D-7, represents the reflection from the end of the pipe. Notice that
the TOF of the humps within circle D-6, when compared with FR, causes the waveforms to
appear in a different location along the X-axis. The TOF for MR2, MR4 and MR6 are 8.475
ns, 8.40625 ns and 8.3375 ns. This represents the detection of PWT in the middle of the pipe.
Notice that these MR TOF values, when compared with the FR TOFs, are much larger. This
makes sense, because the microwaves have spent much more time inside the pipe.
Notice that in Fig. 10.16, the waveform reflections from the excitation point appear in
circle D-8. Circle D-9 contains the waveforms generated by the PWT at the far end of the
pipe. Circle D-10 contains waveforms caused by reflection from the end of the pipe. The
amplitude of the waveform is reduced due to attenuation of the signal because of the reduced
distance between the ER PWT specimen and the end of the pipe.
Figure 10.15: Represents the experimental data for PWT samples MR6, 4
&2, with TOF range from 8.3375 ns to 8.475 ns. The waveforms for the three
depths of PWT samples is shown very clearly in circle D-6
188
Figure 10.16: Experimental result showing the disruption peaks in circles D-9
and D-10, caused because PWT is near end of pipe. The rang of TOF is
13.35625 ns to 13.425 ns
10.6.1.5 Phase One: Microwave Detection Results
Fig. 10.17 represents the six samples for rectangular and semicircular specimens in
the front location (nearer the excitation port). The X-axis represents the pipe length in mm,
the Y-axis represents the PWT radius. As we mentioned before, three depths yield three
radius measurements for the PWT. Keep in mind that the different depths of PWT specimens
are not intended to yield accurate detection of depth by the microwave technique. In fact, the
TOF information does not provide PWT depth information directly. Instead, we seek to
discover what effect the various depths of PWT (2, 4 or 6) have on the predicted location, as
determined by the microwave technique. In the figure the focus is on the predicted location
measurements, rather than the depth measurements. The pairs of FC and FR location
predictions are placed on the Y-axis at the known PWT depths, for demonstration purposes
only. The points could appear at any point along a vertical line passing through the center of
each depth point. The lines through the data points represent the microwave prediction of the
PWT location at the particular depth represented by the specimen. Note that all three
189
specimens were physically located at the same front location, but do not generate identical
predictions of location.
Figure 10.17: Microwave detection location prediction data points are shown
falling into sequence for phase one. Shown are probe positions and relative
movement, the UT techniques to be applied in order to characterize the shape
and size of the PWT, in phase two
The locations of the UT probe are included in this figure to demonstrate how we will
achieve accurate PWT location measurement using the UT technique in the second phase.
The microwave measurement has predicted the location of the PWT specimen (represented
by the vertical lines through the data points), but these predictions vary with the depth of
PWT. For example, the FR2/FC2 location predictions are farther from the excitation point
than the specimen, as are the other points. But, the predicted locations for the other two PWT
depths drift toward the excitation point. The UT probe will be placed according to the
prediction (UT-1), and then dragged toward the excitation point until the actual PWT is
located by the straight beam probe (UT-2).
Fig. 10.18 shows the location predictions for all 6 specimens of PWT, in the middle
location. These are the same six specimens. The Y-axis placement is once again at the known
PWT radius. The top-level detections between 90-95 mm on the Y-axis represent MR6 and
190
MC6. The middle, between 85-90 mm represent MR4 and MC4. And, finally, the bottom
pair between 80-85 mm represent the MR2 and MC2. That tells us that we still have a
sequence in detection as seen in Fig. 10.17.
Figure 10.18: Microwave experimental detection results for six PWT
specimens (rectangular and semi-circular) located at the middle position in
the pipe. The predicted locations are shifting from the middle of the pipe to
this range: 595 mm to 625 mm from the port
Finally, in Fig. 10.19, we see the location predictions for all samples of PWT at the
end position, farthest from the excitation point. As we said before, in Fig. 10.16, the peaks
within the circled detail 10 have been attenuated because the PWT specimen is located very
near the end of the pipe. So, the detected location prediction data points are not in a
predictable sequence.
Notice that the points for ER6 and ER4 predict the same location for the PWT, when
we would expect the predicted location for ER6 to be closer to the excitation point. The same
effect is seen with the points for EC4 and EC2, where we would expect the EC4 prediction to
be closer to the excitation point. Finally, the data points for EC6 and ER2 predict the same
location, as well.
191
Figure 10.19: Experimental prediction results for phase one. It is clear that
there is disruption in the sequence of predicted locations, due to the PWT
samples being located near the end of the pipe
10.6.2 Phase Two: UT Ultrasonic Straight Beam Evaluation
Fig. 10.20 shows the PWT rectangular specimen, with the red hidden lines
representing the three depths of PWT (6, 4, and 2, top to bottom). Remember that the UT
measurements are from the outside wall of the pipe, inward. Thus, depth to defect (DDO)
represents the actual machined distance from the outer pipe wall to the inner wall of the
PWT. Table 10.4 shows the parameters of the ring specimens, including shape, fabrication
depth, DDO, UT detection, and Ring width.
In Fig. 10.20 the three specimens are represented by similar data points. The X is the
DDO, the known depth of defect which is shown for comparison to the actual UT straight
beam measurements. The four data points at each depth represent readings taken with the
dual-probe straight beam device. For example, the 0.4” depth is represented by the yellow
squares. The X on that line is for the DDO of 8.0645 mm. The four data points represent the
four readings taken with the UT probe across the 25.4 mm width of the rectangular PWT
192
specimen. As you can see in Table 10.4, there is not much variation in the UT readings. The
fact that they all cluster around 8mm depth suggests that the shape of the PWT is rectangular.
Figure 10.20: Experimental results of straight-beam ultrasonic using the dual
probe with 15 mm diam. crystal. This shows the detection of rectangular PWT at
three depths: 5.08 mm, 10.16 mm, and 15.24 mm. With comparison to the DDO
(actual known depth to defect, from fabrication)
Figure 10.21: Experimental results from the same procedure as shown in fig.20. This
is for semi-circular PWT at the same three depths: 5.08 mm, 10.16 mm, and 15.24
mm. With comparison to the DDO to five trial tests by dual straight beam UT probe.
193
In Fig. 10.21 the X is again the DDO marker, in the middle of the PWT specimen, but
there is a fifth actual UT measurement. One of the 5 measurements we took for each
specimen often coincides closely with the DDO marker. The red Xs represent the three
depths of PWT (6, 4, and 2, top to bottom). The line for each 5 points gives us an estimation
of the shape of the PWT. The placement suggests the semi-circular shape. However, the
shallow 5.08mm depth data points might also suggest a triangular shape.
Table 10.4: Showing the shape profile; the fabrication depth, which is the distance from the
inner pipe wall to the deepest point of the PWT specimen; the DDO, which is the known
measurement from the outer pipe wall to the highest point on the PWT specimen; and the UT
detection trials, which are the results of the measurements taken by straight-beam UT across
the span of the PWT specimen
Shape
Fabrication
DDO Measurement UT Detection Trails (mm)
Depth in.(mm)
(mm)
Rectangular
0.2 (5.08)
13.1445
13.03, 13.01, 13.12, and 13.14
Rectangular
0.4 (10.16)
8.0645
8.04, 8.06, and 8.05
Rectangular
0.6 (15.24)
2.9845
2.97, 2.86, and 2.88
Semi-circular
0.2 (5.08)
13.1445
13.07, 13.05, 13.12, and 13.04
Semi-circular
0.1 (2.54)
15.6845
(14.85, 14.60, and 14.49) Left
(14.80, 14.73, 14.53 and 14.77)
Right
Semi-circular
0.4 (10.16)
8.0645
8.05,
Semi-circular
0.2 (5.08)
13.1445
(9.08, 9.63, and 10.13) Left
(9.94, 10.09, and 10.37) Right
Semi-circular
0.6 (15.24)
2.9845
3.04
Semi-circular
0.3 (7.62)
10.6045
(5.77, 5.62, and 5.61) Left
(5.59, 5.82, and 5.68) Right
In Fig. 10.22 we see the straight beam UT testing of two specimen rings. Each shows
two trials. A through D show two trials of 0.6” (15.24 mm) for different spot placements of
194
the probe to see the possibility for accurately detecting the correct DDO with both profile
shapes (rectangular and semi-circular). (a) and (b) show testing of the rectangular profile
specimen. (c) and (d) show testing of the semi-circular profile specimen.
Figure 10.22: (a) shows the rectangular PWT test 15.24 mm depth correctly, DDO =
2.97 mm. (b) shows the same rectangular specimen with 10% variation. (c) shows
testing of the middle of the semi-circular PWT. (d) shows the probe placed left of
middle on the same specimen
10.7 Discussion of Results
10.7.1 Microwave Experimental Detection
In Fig. 10.9 it is very clear from the first humps inside circle D-1 that the variation in
depths of the curves is related to the depth of the PWT. And in circle D-2 there is a slight
195
difference, but the reflection came from the end of the pipe. If we change the shape of the
PWT, this affects all the data results from the microwave technique, as can be clearly seen in
Fig. 10.12.
From these figures we can see that changes in the PWT specimen will cause changes
in all the parameters of the generated waveforms. The group velocity method of detecting
the PWT defect has been used before. The prior experiments have not provided completely
accurate location results.
In our experiment, there is a pleasant property to the predicted locations, which is that
the predicted locations are sequential, in most cases. The greater the volume of the PWT, the
closer the predicted location becomes to the actual location of the defect. This allows us to
formulate our innovative sensor positioning equation for the UT transducer in the second
phase of the process. If we examine Fig. 10.17, showing predicted location results for FC and
FR specimens (at the PWT location nearest to the MW excitation probe) with each shape at
three depths, we see that the data point for the greater volume PWT specimen (FR
specimens) is located nearer the excitation point, and also nearer the actual location of the
PWT specimen. We see the same effect in the specimens placed in the middle of the pipe
(MC and MR in Fig. 10.18). In this figure the volume of the PWT for MR2 and MC2
generate data points that fall in the same order. But MR6 and MR4, which have larger
volumes invert the positioning of the data points (MC2 is nearer the excitation point than
MR2). Still, overall, the six data points show a sequence of placement for the predicted
locations.
There is a disruption of this sequence when the PWT specimens are placed near the
end of the pipe (Fig. 10.19). That is because the PWT reflects the signals, and immediately
afterward the energy is reflected again from the end of the pipe. Thus, the predicted location
data points show that the greatest disruption is to the ER6 and EC6 specimens, those with the
greatest volume.
196
10.7.2 Ultrasonic Experimental Detection
In Fig. 10.20, we see the results of a straightforward procedure to detect the depth of
PWT after phase one (microwave) for all three depths: 5.08 mm, 10.16 mm, and 15.24 mm.
In reality, if we test the seamless pipe in manufacturing the rectangular shape will not be
created as we fabricated for our specimens. The shape will not be as smooth or regular as a
rectangular shape.
For Fig. 10.21 again we detect the semi-circle and the estimation was slightly
different from what we fabricated for the semicircle. Each point for the 15.24 mm and 10.16
mm specimens suggests the semicircular shape of the PWT. However, the triangular result
for 5.08 mm. All the time we have some irregularities with the shallow PWT in both the
microwave phase and the UT phase. Here, the irregularity is seen as the near-triangular shape
suggested by the data points for the UT detection.
The cofactor term (X!) in our innovative sensor positioning equation for the
microwave and straight-beam techniques, is very important and depends on the microwave
detection range. So, this one will tell our automated system to send the probe forward or
backward after MW detection according to the ID, thickness of pipe wall, length of the pipe.
All that is variable because the MW detection depends on these variables. For this reason,
each pipe under test represents its own unique case.
 = | ∓
!
|
2
(10.5)
The EPN represents the predicted position from the microwave and the excitation
point for the UT straight-beam. And LPWT is the predicted location from the front end of the
pipe (transmission port) to the predicted location from phase one. The X! cofactor depends
on the system calibration for Phase One (microwave testing).
For example, for the rectangular profile PWT in the middle position at the shallowest
depth, MC2, the predicted location is 625 mm from port one. This is past the PWT by 70.2
mm. We set the cofactor X! to 5! = 120 mm. The standard testing step for straight beam is
200mm. So, for this example we have to go back one step.
197
The automated system should depend on the ASME and ASNT standards for UT testing.
10.8 Conclusion
The microwave technique works as reliably for our larger inner diameter pipes as it
does for 17 mm ID tubing. The addition of UT technology in a second phase of the operation
allows more specific characterization for the PWT specimen, including PWT location and
determination of shape and depth.
In order to make the two-phase system work most efficiently, we need to build an
automated system with program software that shapes the Phase One microwave detection and
the Phase Two UT characterization into one package. Even with automation there will be a
need to calibrate the cap and probe every six hours. Once these critical parts are calibrated,
the system would be calibrated to known PWT specimens before the system was mated to
unknown pipes for quality assurance testing. It is common knowledge that the UT component
of the hybrid system will need calibration as well. But for the automated system, both
calibrations would be carried out at one time.
Finally, for the two-phase system we have devised the innovative sensor positioning
equation. This consists of two terms. The first term LPWT from Phase One gives us the
limitations for the second term, which is the co-factor, X!.
When constructing the automated system, we will need to select the reliable dual
probe for the UT measurements, and we must program the cofactor to move the UT probe by
steps to the real PWT position. The cofactor depends on the shallowest depth of the PWT
location predictions from Phase One, the microwave test. When the UT probe is sent to the
predicted location from Phase One, it will first be sent to the shallowest predicted location.
By steps, it will then move back toward the cap based on the cofactor calculated by the
software, proceeding toward the deepest predicted location. It will then be able to quickly
determine the actual PWT location.
198
Chapter 11
Classification of the Extent of Wall Thinning in Pipes Based on
Simulations in the Time and Frequency Domain [113]
Abstract
Numerical simulations created with Computer Simulation Technology (CST) modeling are
used to classify and evaluate the microwave signal (waveform) in order to determine whether
various parameters of pipe wall thinning (PWT) can be identified. Microwaves are used to
carry out the measurement of PWT in 9 CST simulations using three cross-sectional profiles,
three lengths of PWT, and three depths of PWT, to determine whether there are differences
in waveforms that can be used to distinguish the parameters of unknown discontinuities
(PWT). The modeled system uses the pipe as a circular waveguide (bandwidth 0.486GHz)
with sweeping frequencies from 1.914GHz to 2.4GHz. Waveforms are found to be
distinguishable based on the three parameters modeled in the study. This research establishes
the possibility of creating a nondestructive testing system with which PWT discontinuities
might be characterized using microwaves in the S11 and S21 scattering parametrics. The
waveforms generated for known series of PWT cases, once cataloged, can be used in the
future to identify discontinuities in test pipes, and to determine their degree of similarity to
the standardized waveforms via pattern recognition algorithms.
11.1 Introduction
Pipe wall thinning (PWT) is a major consideration when determining the safety of
pipes used for pressurized transport of materials in the field. During manufacture it is
sometimes necessary to remove inclusions by grinding. This imposes a kind of PWT. Any
kind of PWT is potentially dangerous, and may be costly to detect. The extent of PWT
includes area and depth, so the volume of the affected area is what we seek to determine.
Standards established by the pipeline industry tell us just how thin the wall can
become before it is no longer safe to use. An accurate nondestructive method to classify the
199
extent of wall thinning would be quite valuable for cost reduction in both manufacturing and
in situ testing [1] [30] [90]. In its best form pattern recognition analysis will provide the
ability to both detect and characterize PWT cases in unknown pipes.
For safety reasons it is imperative that any PWT in pipes be detected, especially when
the pipes are aging and in service. The records from 1996 to 2003 show that injuries have
occurred due to failure that can be traced to PWT. Assessment of PWT also affects the
estimate of useful life for the installed pipes [1].
A considerable amount of Non-destructive Testing (NDT) research has been done on
detection of PWT [32] [37] [139]. PWT becomes a great concern in the field, often
developing where water condensation can collect at the lower ranges of pipes, causing
corrosion. Detection of defects during manufacturing remains a very important area in NDT
[72].
Several main NDT techniques are used to detect manufacturing defects: visual
inspection, magnetic particle testing, liquid penetrant test, ultrasonic testing, x-ray testing,
and acoustic emanation testing, all discussed in previous studies [1] [29] [30] [62]. These
NDT techniques have improved in usefulness, in part due to experience with the techniques,
and in part because of changes in the method of transporting materials in pipes (speed,
viscosity, etc.) and architecture of pipelines (addition of elbows, increased number of pumps,
and so forth) [75] [90] [135].
Although systems are automated, there are always humans involved in running the
tests, which leads to human error. This increases wasted time, which means greater overhead.
But we cannot eliminate these stations altogether [1]. Prior to formation of the pipe body the
plate raw material must be tested for defects. Removal of these defects can lead to PWT due
to grinding. Because of this, the plate which is to be formed into pipes may already contain
instances of PWT. Grinding is a necessary part of the manufacturing process, and cannot be
eliminated altogether. Forming of pipe can also introduce further defects, which may be
remedied by grinding, leading to more PWT. Standard tolerances have been established, as
well as certain tolerances specific to the customer. Careful measurement of the degree of
200
PWT must be undertaken during manufacture, to ensure that the resulting pipe wall are
within tolerances [91] [92].
In 2008 and 2009 K. Abbasi, et al. started using microwave signals generated by a
vector network analyzer (VNA) to attempt to locate circumferential cracks in stainless steel
piping. He was successful, using the time domain and velocity calibration methods of the
micro-waveform, in detecting two positions of cracks in the pipe (which acted as an efficient
waveguide). Abbasi performed both theoretical and practical experiments, successfully
confirming the theory with the practical experiment. The 2008 experiment used a straight
section of pipe. In 2009 Abbasi repeated the experiment with a u-shaped bend in the pipe,
and a crack of 10mm length [75] [135].
In 2011 Linsheng Liu, et al, attempted to detect the edges of defects that are
analogous to PWT with the same velocity group method, for pipe with inner diameter 17mm
with various pipe lengths. The experiment was able to locate the longitudinal placement of
the defect, but could not resolve the leading and trailing edges of the defect (due to
overlapping peaks) [34].
By undertaking the classification of the extent of wall thinning, we will begin
calibrating the standard waveforms for both this case and future cases. With pattern
recognition analysis such standard waveforms will enable rapid detection of the type and
location of the defect (degree of defect) in the future. Three kinds of profile (tapered crosssection; one rounded and one right angle corner; half-circle) are modeled. Each profile is
modeled for three lengths (33.333%, 66.667%, and 100%) as percentage of circumference.
The depth will be fixed at 6.35mm for the PWT instances in the calibration models. The
width of PWT is fixed at 25.4mm in the calibration models. The test instances of PWT will
all be located in the middle of the pipe (381mm from port 1).
Three depths are measured for each instance (6.35mm, 8.89mm, and 11.43mm), for a
full-circumferential PWT, for all three profiles.
The main goal of this paper is to generate standard waveform patterns to be used for
pattern recognition analysis of unknown cases in the future.
201
By changing these parameters, we will deduce the relationship between the volume
and profile, and the resonance frequency.
11.2 Materials and Methods
The investigation is undertaken using Computer Simulation Technologies (CST)
finite difference time domain software.
Microwaves will propagate inside the pipe with the lowest cut-off frequency for
TM01 at 1.507GHz. The Bessel function for bandwidth frequency 0.893GHz is created by
Port 1 and Port 2 during the TMmn mode (TM01-TM11), with mathematical roots (P01=2.4048,
P11=3.8317). By considering the sweeping frequency range from 1.914GHz to 2.4GHz the
bandwidth will be 0.486GHz for simulation purposes, which are nearest to the TE21 and TE02
for this wave guide inner diameter. The first root and second root of the Bessel function for
these are (P21=3.0542, P02=3.8317) [90]. The pipe is evacuated.
PWT will be 33.333%, 66.667%, and 100% of the pipe inner circumference for three
profiles. One profile for the PWT is tapered in cross-section, another has one rounded and
one right-angle corner in cross-section, a third is half-circular in cross-section. All the lengths
and percentages above are modeled for each profile, with the PWT instance located in the
middle of the pipe (381mm from port 1) in order to deduce the difference in waveform and
resonance frequency produce by each combination. The width of the wall thinning is fixed at
25.4mm. The pipe inner diameter is 152.4mm. The outer diameter is 177.8mm. Length
762mm. Wall thickness is 12.7mm. Depths of PWT are 6.35mm, 8.89mm, and 11.43mm,
three depths are determined for each PWT profile for a full-circumferential PWT.
11.3 Result and Discussion
The tests made for effect on resonance frequency and peak definition are conducted
from both port 1 and port 2, so that we can investigate waveforms on both S11 and S21 axes to
see if there is any differentiation in the scattering parameter (S11, S21) based on the profile of
the PWT. Such differences, where they exist, can be used to distinguish the cross-sectional
profile in pattern recognition analysis.
202
In this modeling we have three kinds of PWT according to Fig. 11.1(a) half-circle, (b)
tapered, (c) one rounded and one corner (fillet and right angle). For each type we simulate the
lengths as 33%, 67% and 100% of circumference.
`
Figure 11.1: Three types of PWT profiles modeled in CST to investigate the effect of
PWT cross-section shapes on the S11 and S21 waveforms. D is wall thickness; DD is
depth of PWT. W is width of PWT
203
Figure 11.2: Resulting waveforms for three types of PWT profiles modeled in CST
showing the S11 waveforms, each of which is measured for the three lengths of PWT
(33%, 67% and 100% of circumference). (a) shows results for the half-circle profile,
(b) for the tapered profile, and (c) the fillet and right corner profile
By measuring S11 for three lengths of PWT half-circle, as shown in Fig. 11.2(a) we
see different peaks according to the volume of the PWT. The peak for the greater volume is
shifted to the left (lower frequency in GHz). There is greater definition of the S11 peak with
204
increasing volume of PWT. The same differences are seen with profiles (b) and (c) as shown
in the respective figures.
Figure 11.3: With 100% circumference length, constant depth and constant width,
shows the shift in resonance frequency induced by changes in volume of the PWT.
(a) S11, and (b) S21.
Fig. 11.3(a) full shows the three types of profile with 100% circumferential length.
There is a shift of the peak to the left based on volume in this graph, too. The fillet and right
205
angle resonance frequency peak occurs at a lower frequency (2.3809 GHz) than the
waveform peaks for tapered (2.3824 GHz), and half-circle (2.3866 GHz). The smallest
volume profile (half-circle) resonates at a higher frequency (appearing farther to the right in
the graph). The fillet and right angle profile also generates a higher peak in S 11 than the
tapered and half-circle profiles. The same results are seen in Fig. 11.3(b) S21 with the
resonance peaks appearing at the same GHz markers.
As shown in Fig. 11.4. Three depths are modeled and simulated to distinguish the
waveform peaks according to the depth (6.35mm, 8.89mm, and 11.43mm). Fig. 11.5(a) (b)
(c) show the half-circle resonance frequency peaks (2.3866GHz, 2.3782GHz, and 2.3653),
with the same effect seen for tapered resonance frequency peaks (2.3824GHz, 2.3731GHz,
and 2.3602), and for fillet and right angle resonance frequency peaks (2.3809GHz,
2.3677GHz, and 2.3515).
Figure 11.4: Three types of PWT depths modeled in CST to investigate the effect of
PWT depths on the S11 and S21 microwave signals
In Fig. 11.6(a) and (b), the 11.43mm depth generates a resonance frequency peak for
each type of defect profile. The greater volume of the fillet and right angle profile appears to
the left of the other two, lower volume types. It is worth noting that the relationship of
volume to resonance frequency holds regardless of the parameter being investigated.
206
207
Figure 11.5: Shows the waveforms generated for three depths for the three profile
types. (a) is half-circle, (b) is tapered, (c) is fillet and right angle
Figure 11.6: shows the waveforms for the 11.43mm depth for each of the three
profile types. This graph demonstrates the property of the resonance frequency peaks
for greater volumes of PWT to fall to the left of the graph (lower GHz frequency).
(a) S11, and (b) S21
208
11.4 Conclusion
The CST models are effective for simulating real-world conditions with which we can
generate standard pattern waveforms to be used in pattern recognition analysis. There are
three possible methods of analysis: the first is Correlation Analysis (Cij), the second is Least
Square Distance analysis (LSD), and the third is the Cosh Spectral Distance analysis (CSD).
It is yet to be determined which analysis technique is the best fit for pattern recognition.
According to the concept of pattern recognition it should be possible to identify the
volume (width, depth, and length), location, and profile type of PWT. With sufficiently
detailed standard patterns it is possible that unknown pipes can be evaluated using
microwave signals for analysis. Investigation of these possibilities is future work for this
researcher.
209
Chapter 12
Waveform Pattern
Recognition
Applied
To
Rapid
Detection
of
Wall-Thinning In Pipes: A Simulation-Based Case Study [114]
Abstract
Pattern recognition using correlation analysis (Cij) method is useful for non-destructive
testing of physical objects, including pipes. An evaluation of the technique based on
Computer Simulation Technology (CST) models has demonstrated the advantages of using
the technique to detect and classify pipe wall thinning (PWT) in pipes. Given enough
increments, the technique can be refined to detect any possible combination of PWT
attributes. For this research 71 different simulations were modeled for purposes of calibration
of the system, based on five varied properties of the modeled PWT instances. These
properties include: location (29 simulations based on distance from origin and two lengths of
PWT, for a total of 58 simulations), width (standardized at 25.4mm), depth (four simulations
as radius of PWT at 78.74mm, 81.28mm, 83.82mm, and 86.36mm), length (four simulations
as percentage of circumference: 25%, 50%, 75% and 100% circumferential PWT) and type
of defect (five simulations based on five discrete profiles).
Microwaves were simulated from port 1 and port 2, with a sweeping frequency range (0.5
GHz bandwidth), analyzed as S11 and S21 for measuring and calibrating the response to the
standards. The resulting waveforms became the standard patterns against which 11 unknown
simulations were compared, sometimes using S11waveforms for comparison, and at other
times S21.
The correlation analysis technique was able to distinguish parameters for the unknown test
cases. The technique is able to determine the correlation between the resonance frequency
peak (RFP) and waveform for an unknown case, and those of nearby calibration models, via
pattern recognition. For example, 0.847 and 0.872 correlations to two standard patterns for an
unknown RFP which appears midway between two standard RFPs, produces a peak for the
unknown that is equidistant from the RFPs for the standards.
210
12.1 Introduction
Pipe wall-thinning (PWT) is an important consideration in manufacturing and field
operations involving pipes. Grinding out of defects during manufacture can lead to wall
thickness reductions that exceed safe limits. Standards exist to determine safe limits, and
these are based on long-term accumulated practical knowledge. Installed pipe is also subject
to pipe-wall thinning, which might lead to pipe failure. Any failure of pipeline infrastructure
in populated areas could be catastrophic. Thus, there is a demonstrated need for methods of
testing in situ pipes to detect corrosion damage and other types of distress before the
discontinuities can lead to failure. Such techniques would allow determination of just how
close a given section of pipe might be to the limits [1] [38] [90] [140].
In pipe manufacturing there are several nondestructive testing (NDT) techniques,
such as ultrasonic, radiographic, and so on. These techniques require movement of the pipe
or the probe during testing [30].
There are two testing methods that involve radio frequency or sonic wave signals.
These have application during manufacture, but are especially useful in testing in service,
where the pipes cannot easily be moved [29] [37].
Microwave is a promising new static testing technique, and it has been wellresearched [135] [139]. Ultrasonic waves have been used to test metal structures for decades
[69] [141]. Both lend themselves to pattern recognition analysis.
In terms of ultrasonic waves there are four types of ultrasonic waves that can be used
to test metal pipes, and plate materials. All require a medium for propagation. The first is
Longitudinal Waves (LW, compression waves) which excite the particles of any material that
they pass through. The LW generate vibrations in the material that occur in the same
direction as the waves are moving. The second is Shear Waves (SW, transverse waves)
which propagate in materials causing particle motion that is perpendicular to the direction the
waves are moving. Rayleigh Waves (RW, surface waves) travel no deeper than a single
wavelength below the surface of the material under test. Lamb Waves (LMW, plate waves)
can move through the full thickness of the plate, and can move a considerable distance from
the point of excitation as well. Each type of wave moves at a different velocity in different
211
material, for example 91% brass and aluminum, rolled would have different signal velocities
when excited by the same ultrasonic waves [29] [59] [60] [61] [63] [142].
From material to material the absorption and attenuation of the ultrasonic signal
varies. Thus, the signal is reflected or refracted when it passes from one material to another
(via the material interface). In fact, the ultrasonic waves cannot propagate without a medium,
and must be conducted from the transducer to the material under test via water or grease. An
air gap will prevent the UT from working [29] [46].
Microwaves travel through fluids, or vacuum, requiring no medium. Even though
microwaves, being electromagnetic energy, behave in the same way as visible light (traveling
at the speed of light, being reflected or refracted from materials that they contact), this
provides an advantage for testing of air-filled pipes [46].
Frequently Used Techniques of Microwave Nondestructive Testing
“Swept-frequency
Continuous-wave
Transmission,
Fixed-frequency
Continuous-wave
Transmission, Swept-frequency Continuous-wave Reflection, Fixed-frequency Continuouswave Reflection.” [46].
Defects, including voids, accumulations (thickening), cracks, and wall-thinning tend
to either reflect or scatter the microwaves as they travel through the waveguide. There is
usually a rather abrupt boundary between the wall material and the material or void of the
defect, which results in a change in velocity of propagation for the waves. This induces a
change in behavior at these interfaces. The waves are either reflected, refracted, or scattered
at the boundary, which can be useful in locating such defects as patterns in the reflected
signals [1] [46].
All waves vary in measurable ways, according to the time domain and the frequency
domain of the waveforms produced as they travel through air, liquids, or solids. There are a
number of types of signals that have long been categorized, each of which might have
specific applications. The reflected waveforms can be analyzed for patterns generated when
the waves encounter different types of defects, and the patterns generated in a calibration
condition can be used to analyze test waveforms for pattern recognition. With pattern
212
recognition the volume (width, depth, and length), location, and type of defect can be
determined using wave signals for analysis.
The existing research done on full-circumferential samples using microwave NDT
has revealed the presence of wall-thinning, but has difficulty in determining the precise
location and volume of the defect along the pipe wall.
Pattern recognition allows comparison of test pipes with unknown defects to the
waveforms for the calibration pipes. The degree of correlation between unknown and
calibration will provide valuable information concerning the potential discontinuity in the
unknown pipe. This promising technique should become the basis for establishment of the
range of detection possible with microwave pattern recognition. With additional refinement
in future of the algorithms used, the technique will be able to precisely analyze PWT in
pipes.
This research examines the feasibility of using pattern recognition to detect and
characterize PWT, with high discretization, in pipes. To test the concept, models were built
to contain a broad array of PWT parameters from which standard patterns could be
generated, in order to calibrate the system. For five parameters individual simulations were
created and characterized. For the location of PWT, 29 simulations were modeled based on
distance from the waveguide origin, including two sets with varied lengths of PWT, for a
total of 58 simulations. The width parameter was standardized at 25.4mm. To characterize
the depth parameter, four simulations (as PWT radius) at 78.74mm, 81.28mm, 83.82mm, and
86.36mm were created. For the length of PWT, four simulations as percentage of
circumference: 25%, 50%, 75% and 100% were created. And for the type of defect
parameter, five simulations based on five discrete profiles were modeled and characterized.
12.2 Fundamentals of Microwaves
When traveling through a vacuum (free of medium) an electromagnetic (EM) wave
travels as a sine wave, with the vibration at right angles to the direction of wave travel. This
vibrating motion is relative to a fixed point in space, of course. Fig. 12.1 shows a side view
213
section of the waveguide with the electrical field, magnetic field, and velocity of the
microwave travel (as an electromagnetic wave with linear polarization).The sine wave
characteristic is illustrated in Fig. 12.2. Note that the magnetic field and electrical field
vibrate perpendicular to each other [46].
Figure 12.1: Illustration of the orientations of electrical field and magnetic field
intensities, with the velocity of microwave signal propagating inside a waveguide
Figure 12.2: Illustration of an electromagnetic (EM) wave exhibiting linear
polarization propagating in empty space, where Z is the direction of travel of the
microwave signal; E and H are the electrical field and magnetic field amplitudes,
respectively; λ is the wavelength
214
Microwave behavior is moderated by the wavelength of the signal. A shorter
wavelength (λ) produces a higher frequency signal (f), for example. When the wavelength is
small compared to the surface with which the signal interacts, the signal will be reflected
from the surface, and will retain the quality of a single wave. When the wavelength is larger
than the surfaces with which the signal interacts, the reflected waves have a spectrum of
wave properties: amplitude, phase and direction of travel. This is called scattering, and is
used to detect discontinuities in the surfaces under test. The effect is intensified when the
wavelength is similar in dimension to the irregular surface. For this reason, a sweeping range
of frequencies is often used to test for surface defects using microwaves [46].
12.3 Pattern-Recognition Analytical Approaches
It is an attribute of human beings that we are able to recognize patterns. We are also
able to develop means by which technical identification can be achieved by developing
patterns which the devices can recognize and report to us.
For example, a pipe is made of a material, is of a certain size, and contains known
defects. From this “standard” pipe a method of detecting defects may be developed. The
main impetus is to develop a plan to locate and identify the unknown defects in another pipe,
and to do so with low cost and low labor.
Correlation analysis compares the degree of correlation between two patterns. By
calculating Cij for both patterns, and finding how closely they match, a correlation value is
derived. A value of 1 shows identical or patterns, the value –1 shows a condition of
opposites, while a correlation of 0 shows total difference between the patterns. The nearer the
correlation value approaches 1, the higher the degree of match between the patterns [143]
[144].
 =
∑=1( () − ̅ )( () − ̅ )
[∑=1( ()
−
1
̅ )2 ]2
[∑=1( ()
− ̅
1
2
) ]2
(12.1)
215
Least Square Distance (LSD) analysis has been used in modeling of system and bio
recognition. It is calculated using. In this analysis the lesser value indicates a nearer match, a
greater value indicates a more distant correlation [143].

2
 = √(∑ ( () −  ()) )
(12.2)
=1
The Cosh Spectral Distance analysis (CSD) reveals the overall difference between
two pattern approaches. It is calculated using [143] [145] [146] [147].

S ()
1
S ()
S () S ()
 =
∑(
− log
+
− log
− 2)
2
S ()
S () S ()
S ()
(12.3)
=1
Long Qiao explains the variables in these equations quite well, “where n is the
number of vector points in the pattern; Si(k) and Sj(k) are the vector values of the patterns i
and j at point k; and Si and Sj are the average values of the patterns i and j, respectively.”
[143].
For this research the technique chosen is Correlation Analysis. In Correlation
Analysis the calculated correlation value between two patterns derived from defect-free
waveguides would be constant, but with the introduction of PWT, the reflection values
change, causing a change in the degree of correlation [144].
12.4 Microwave Pattern-Recognition Technique
12.4.1 Microwave Analysis
12.4.1.1 Modeling
The models used in the studies of pipe wall thinning are created with Computer
Simulation Technology (CST) Microwave Studio - Time Domain. The mesh type is
hexahedral with full-circumferential PWT. The simulation uses two ports. Port 1 and port 2
216
create bandwidth frequency (f) microwave at 0.5 GHz. The magnitudes of the scattering
parameter (S11 and S21) for the first port produce peaks according to the PWT.
Pnm is the mathematical roots for Bessel function are set up as expressed in the equations
below [123].
11 =
 

(12.4)
For air the permittivity, εr = 1, and Cd speed of light inside the pipe for free space is 3x108
m/s. fcTEnm is the lowest cutoff frequency at TE11 mode, Di is pipe inner diameter.
21 =
3.0542

1.8412 11
(12.5)
11 =
3.8317

1.8412 11
(12.6)
12.4.1.2 Waveguide Parametric, Boundary Conditions, and Numerical
Simulation
The pipes are modeled as perfect conductor (PEC) surfaces, and the inner volume is
evacuated. The occurrences of PWT are standardized at 29 locations along the pipe length of
762 mm. Inner diameter (Di) is 152.4 mm. Wall thickness (D) is 12.7 mm.
Parameters of PWT are: width, depth, length, location, and type of defect.
The location parameter (Fig. 12.3) is relative to port 1, expressed as distance from the
probe, numbered from the front end (at port 1) to the tail end of the waveguide. The 29
instances are positioned step-wise 25.4mm on center with a standard 25.4mm width. This
means that the ending point of instance 01 is the beginning point of instance 02, and this
relationship continues for all 29 instances of PWT. The cross-sectional profile used for
calibration of location is rectangular with PWT depth of 6.35mm. There are two lengths
assessed: 50% circumference, and 100% circumference.
217
Figure 12.3: Sectional view of modeled waveguide used in simulation case
study, showing first two and last two occurrences of PWT. Ld is the pitch of
PWTs on center; Lp is the length of the waveguide, DD is depth of defect, Do
is the outer diameter
The depth parameter is expressed as the radius of the PWT, and varies in the fillet
and right angle cross-section profile as: 78.74mm, 81.28mm, 83.82mm, and 86.36mm.
The width (W) of PWT is fixed at 25.4mm for all 29 instances.
The length parameter is modeled as four percentages of circumference: 25%; 50%;
75% and 100%, using the fillet and right angle cross-section profile, and with a fixed depth
of PWT at 6.35mm.
The type of defect parameter is the cross-sectional profile of the PWT. There are five
profiles: (a) rectangular, (b) one fillet and one right angle corner, (c) two fillets, (d) tapered,
and (e) triangular.
218
12.4.2 Correlation Analysis
Once all the factors have been set up for Microwave (MW) NDT technique, various
methods of analysis are reviewed to match the case study. The best method of pattern
recognition is then selected. In this case, the most effective method is Correlation Analysis.
After the analysis MW phase which generates the standard patterns (which can be
called set j) to be used for recognition, correlation analysis is used to determine the degree of
match between the patterns, and data collected from an unknown test case (which can be
called set i). This ambitious technique is a calculated method that computes the correlations
for all locations, depths, lengths and type of PWT in the pipe, to a reference point (many
calibration points, many precise analyses, for many parameters of a defect, which is set j),
which would be the standard patterns. The main goal is to quantify the vector values from
test data (set i) to those that have been modeled by the MW technique (set j) to find the
similarity in correlation.
12.5 Correlation Analysis Results and Discussion
12.5.1 Pattern Recognition to Evaluate PWT Location
Fig. 12.4 depicts the way in which the pattern recognition system detects the location
of an unknown PWT case based on correlation between the data from the test waveguide and
the patterns generated by the calibration set j. PWT3, PWT4 and PWT5 represent the 3rd, 4th,
and 5th locations of calibration PWT instances, a subset of (PWT)j. Two additional PWT
instances are shown, PWTN1 and PWTN2. These represent two locations of PWT in an
unknown location (test pipe). PWTN1 is located midway between PWT3 and PWT4, as
shown in Fig. 12.4. The second additional instance, PWTN2, is located in both PWT4 and
PWT5. The correlation analysis of PWTN1 shows that it has a 0.847 and 0.872 correlations to
each of the standard patterns for PWT3 and PWT4 respectively. The correlation of PWTN2 is
0.737 to PWT4 and 0.937 to PWT5, which indicates that it is located closer to the position of
PWT5. Rectangular cross-section profile for all locations.
219
Figure 12.4: Cross-section of pipe showing standard PWT at positions 3, 4,
and 5. In addition, two unknown defects are shown, with their correlation to
the standard PWT instance
Figure 12.5: Location shows resonance frequency peaks (RFP) for three
standard PWT instances, plus the peaks for PWTN2 and PWTN2. The peak
position demonstrates that the RFP for each standard shifts forward and
decreases in height with greater distance from port 1
220
In Fig. 12.5 the resonance frequency peaks for the 3rd, 4th and 5th calibration locations
of PWT are shown. Also, shows the resonance frequency peak for PWTN1, located halfway
between the standard peaks. The pattern recognition correlation analysis shows that the peak
should be halfway between the peaks for 3rd and 4th, because the correlation values to both
3rd and 4th standards are same approximately 0.847 and 0.872 for PWT3 and PWT4
respectively. In addition, Fig. 12.5 shows the peak for PWTN2, demonstrating that the
resonance frequency peak for that case falls closer to the peak for the 5th standard PWT, as
predicted by the correlation values of 0.25 to PWT4 and 0.75 to PWT5, as discussed above.
Also, the relative position of the unknown case RFP’s corresponds to the correlation values
for the surrounding standard peaks.
12.5.2 Pattern Recognition to Evaluate PWT Length
Fig. 12.6 shows PWT at four percentages of circumference. (a) is 25%; (b) is 50%;
(c) is 75% and (d) is 100% circumferential PWT. Fillet and right corner cross-section profile
for all lengths, in order to generate recognition patterns for analysis of unknown test cases.
The cases PWTL1 and PWTL2 demonstrate the use of correlation analysis to detect the length
of the PWT. PWTL1 has a 0.838 and 0.852 correlations to both the 50% and 75%
respectively, recognition patterns, indicating that the length of the PWT is 62.5% of
circumference. PWTL2 has a 0.799 correlation to the 75% recognition pattern and 0.97
correlation to the 100% recognition pattern, indicating that the length is near 100%.
When the volume of the PWT decreases (smaller percentage circumference length)
the resonance peak shifts to the right, and decreases in height, as shown in Fig. 12.7(a). The
two unknown cases PWTL1 and PWTL2 are shown in Fig. 12.7(b) and (c). Note that the
peaks fall relative to the correlation values discussed above, in each of these figures. All the
standards and unknown cases are modeled in the middle of the pipe.
221
Figure 12.6: Cross section of waveguide showing percentage of circumference
involved in the PWT length standard examples (a)-(d). Example (a) shows the 25%
circumference length, (b) is 50%, (c) is 75% and (d) is 100% (full-circumferential
PWT). In addition, Fig. 12.5 shows example (e), a 62.5% length and its correlation to
(b) and (c). Example (f) shows a 90% length, and its correlation to (c) and (d)
222
223
Figure 12.7: (a) the four standard peaks based on percentage of circumference
(volume) of the standards PWT defect (fixed width and type of profile). (b) the
peaks represent the two standard peaks between which the unknown case resonance
frequency peak falls (for 90% circumference). Note that the peak is located closer
to the peak for 100% circumference. (c) the peaks represent the standard resonance
frequency peaks for 50% and 75 % circumference, with the unknown case peak
(62.5% circumference) falling midway between them
12.5.3 Pattern Recognition to Evaluate PWT Depth
Fig. 12.8 shows PWT at four depths, expressed as the radius of the PWT: 78.74mm,
81.28mm, 83.82mm, and 86.36mm. Fillet and right corner cross-section profile for all
depths. Case PWTR1 has a radius of 82.55mm that is halfway between 81.28mm and
83.82mm, and the correlation for this case will be 0.829 and 0.914 respectively. Case PWTR2
has a radius of 86.995mm that is between 86.36mm and the outer radius of the pipe
88.90mm, with a correlation of 0.923 to the 86.36mm. The result says that the 86.995mm is
near to the 86.36mm recognition pattern.
Figure 12.8: Section of waveguide with radii of the standard PWT depths (78.74mm
to 86.36mm) shown in a standard width PWT instance (W=25.4mm). Unknown
cases in red show two PWT cases with radii of 82.55mm and 86.995mm. The
correlation values for each unknown case are shown in the discussion
224
Note that the resonance frequency peak shifts backward (to a lower value) when the
radius of PWT increases. The four reference peaks are shown in Fig. 12.9, with the
shallowest PWT to the right. The other two peaks represent case PWTR1 and PWTR2. We
can see that PWTR1 causes a peak located halfway between the peaks for 81.28mm and
83.82mm, and have a correlation of 0.829 and 0.914 respectively. And thus, the resulting
peak is located halfway between the reference peaks approximately.
Figure 12.9: four standard resonance frequency peaks for the standard patterns for
PWT radii. The four standard peaks are shown that the greater radius is to the left.
Two additional cases are shown with their peaks superimposed
Likewise, PWTR2 returns a peak to the left of the 86.36mm standard reference peak,
indicating that the 86.995mm radius is greater than our deepest standard PWT. Thus, it is
located between the outer surface and the deepest standard PWT, but is located just beyond
the deepest standard depth, with a correlation of 0.923. The array of resonance peaks
demonstrates that the relationship between known and unknown patterns is predictable, and
regular. Depths are modeled in the middle of the waveguide.
225
12.5.4 Pattern Recognition to Evaluate PWT Profile
Fig. 12.10 shows PWT in five cross-sectional profiles: rectangular, one fillet and one
right angle corner, two fillets, tapered, and triangular. Each one has a different S21 peak due
to inherent shifts in resonance frequency caused by the profile shape of the PWT as shown in
Fig. 12.11 Whenever we measure the degree of PWT (location, volume) the closer the defect
is located to the origin port of the signal, and the greater the volume of the PWT, the higher
the peak will appear on the S11 axis at the resonance frequency. That provides a greater
opportunity to recognize the MW signal.
For calibration, all profile examples are located in the mid-pipe region (381mm on
center from port 1) with the standard 25.4mm width, with 100% circumference as length.
Figure 12.10: shows the type of defect profiles possible for PWT
226
Figure 12.11: profile demonstrates the waveform generated by each of the
five profiles tested. Further work on the variations for profiles remains to be
done in the future
12.6 Conclusion
For pattern recognition when we are able to model the smallest increments for each
parameter, we will find that the unknown cases will always produce a correlation of 1.0 to
one of the patterns, yielding very precise results. The method is ambitious, aiming to build a
future algorithm with thousands of variations in each measured parameter against which
cases of unknown pipe can be compared.
The microwave technique requires standardization of resonance frequencies used to
scan unknown cases of pipe. Once these are established for each parameter, the algorithm
could be used to detect any aspect of the PWT: including width, depth, length, location, area,
volume and type of defect, accurately. As long as the sets of standard patterns exist for the
resonance frequency used, the pattern recognition technique will return identifying results.
227
Even irregularly shaped defects in manufacturing can be added to the algorithm, broadening
its usefulness.
In manufacturing, microwave NDT would need re-calibration at least twice a day, to
compensate for changes in the environment. This is especially true for pipes tested in openend condition. Changes in temperature, or the electric potential of the air could affect the
microwave behavior. Re-calibration would keep the pattern recognition system as accurate as
it could possibly be.
In the future, the correlation analysis-based pattern recognition can be applied to
other non-destructive testing (NDT) techniques, as long as it is possible to build a
comprehensive set of patterns for the most frequently-seen defects. Detection by ultrasound,
or acoustic emission testing could be enhanced by adding pattern recognition to the standard
resources.
228
Chapter 13
Experimental Validation for Waveform Pattern Recognition
13.1 Methods and Materials
13.1.1 Specimens
One meter length pipe has been used in this experiment, as shown in the photograph
below (Fig. 13.1). Nine specimens rectangular, semi-circular, right-triangular have been
created, at three depths, each. The specimen depths are 5.08 mm, 10.16 mm, and 15.24 mm.
The PWT specimens are located in the middle of the pipe. The pipe inner diameter is 154.05
mm. The outer diameter is 168.402 mm. The cap has been fabricated 137.7 mm length, with
ellipsoidal longitudinal cross section, as shown in the Fig. 13.2 below. All materials were
created from 6061-T6 aluminum alloy. The inner diameter of the cap exactly matches the
inner diameter of the pipe.
Figure 13.1: Photograph of the experimental setup used for this experiment.
The vector network analyzer with the pipe connected to the cap and coaxial
cable. The PWT specimen ring is in place at the middle location
229
Figure 13.2: Cross section of ellipsoidal cap created from 6061-T6
aluminum alloy
13.1.2 Experimental Set-up
We first calibrated the vector network analyzer (VNA) to standard loads for a oneport setup: a short-circuit, an open circuit and a matched load. The VNA was set for a
sweeping frequency range of 1.4-2.3 GHz. We achieved the expected flat reading at 0 dB in
all calibration conditions.
The VNA was set for a slow sweep to suppress transient noise, and averaged 16
readings at each discrete frequency point in the range. For open and short circuit conditions
the gamma should be 1 (for open circuit +1, for short circuit -1) due to complete reflection of
the signal. For the matched load the gamma should approach negative infinity, because
almost all the energy is absorbed. After achieving these readings, the VNA was calibrated for
our reference pipe.
230
The coaxial probe was connected to a specially-designed cup that fits onto the
reference pipe at port one. The first test was for the open condition. A set of calibration peaks
was generated that represents the frequencies that are more strongly absorbed by this
material. This was followed by a test at the short circuit condition, provided by a 12”x12”
(304.8mm x 304.8mm) gold-coated plate placed flush across the open end of the reference
pipe. It is commonly known that the high conductivity of gold provides the closest
approximation to a true short circuit that can be achieved. From this measurement, we
extracted time of flight (TOF) data from the time domain signal to be the correction factor for
the group velocity.
13.2 Results
In Fig. 13.3 we can see that within the frequency range from 2.04 GHz to 2.12 GHz
there are recognizable curves. The peak amplitudes vary according to the depth of the semicircular shape PWT specimens. The other frequency range from 2.12 to 2.22 GHz as shown
in Fig. 13.4, varies according to the same parameters. It is very clear that there is a different
maximum frequency peak according to the volume. The lowest frequency peak represents the
greatest volume of PWT.
Figure 13.3: Experimental result for semicircular profile in the middle of the
pipe, with depth varying from 5.08 mm 10.16 mm and 15.24 mm
231
Figure 13.4: Experimental result for semicircular profile, focusing on the
sweeping frequency region 2.12 GHz to 2.22 GHz, in order to show that the
maximum frequency peak correlates to maximum volume
Figure 13.5: Experimental results with sweeping frequency 2.00 Ghz to 2.20
GHz, showing the slight differences in the recognition waveforms based on
the volume change as the profile shape changes. All the PWT is at the same
depth (5.08 mm) and location (middle of pipe)
232
Figure 13.6: Experimental results focusing on the frequency range 2.30
GHz to 2.12 GHz, comparing the results of the volume change as the
PWT shape changes
Figure 13.7: Experimental results focusing on the frequency range 2.14
GHz to 2.2 GHz, showing that the sequence of frequency peaks (lowest
frequency is for highest volume) is not as expected
233
In Fig. 13.5 all the curves for the 5.08 mm depth are included in one graph for all
three PWT profile shapes. We can see from this figure that there is a slight recognition,
because the shape change creates a difference in volume, and the curves reflect the change,
as shown in Fig. 13.6. The range from 2.12 GHz to 2.2 GHz as shown in Fig. 13.7 maximizes
the slight changes in dB according to the microwave reflections. The slight change related to
the change in shape shifts the frequency at maximum peak, because the volume has been
changed slightly according to the shape. Interestingly, the resulting peak locations are not in
the same sequence that we see from changing the depth from 5.08 mm to 10.16 mm for the
same profile shape in the same location.
13.3 Discussion and Conclusion
When we change the depth of PWT in the same location, with the same defect profile,
we get results with a lower frequency at the maximum peak for the greatest depth specimen.
That is according to Figs. 13.3 and 13.4, for the semicircular profile.
In Fig. 13.5, for the frequency range 2.0 to 2.2 GHz, there is a slight difference in the
waveform pattern. This is in accordance with the slight change in volume due to the profile
shape changes. Here we used full-circumferential PWT with the same width, and same depth,
but with a change in the PWT shape. The sequence of the maximum peak frequencies has
been disrupted. The peaks are not in the sequence we see for volume changes caused by
changes in depth, when the profile remains the same. The frequency range from 2.04 to 2.12
GHz shows us that the maximum peak frequency for the rectangular profile shape almost fits
with that for the semicircular one. But for this region the sequence based on volume remains
as we have seen before. In the range 2.14 to 2.2 GHz, the sequence is altered, and the
triangular shape has both the lowest volume and the lowest peak frequency.
234
Chapter 14
Development of an Optimized Neural Network for the Detection of
Pipe Defects Using a Microwave Signal [148]
Abstract
Neural network technology is applied to the detection of a pipe wall thinning (PWT) in a pipe
using a microwave signal reflection as an input. The location, depth, length and profile
geometry of the PWT are predicted by the neural network from input parameters taken from
the resonance frequency plots for training data generated through computer simulation. The
network is optimized using an evolutionary optimization routine, using the 108 training data
samples to minimize the errors produced by the neural network model. The optimizer not
only specified the optimal weights for the network links, but also the optimal topology for the
network itself. The results demonstrate the potential of the approach in that when data files
were input that were not part of the training data set, fairly accurate predictions were made
by the network. The results from the initial network models can be utilized to improve the
future performance of the network.
14.1 Introduction
Defects in pipe manufacturing can result in serious problems if they are not detected.
A defect located during the manufacturing process can often be remedied, but even if the
pipe must scrapped, it is better than the alternative to have it utilized in a pipeline and fail at
some point in the future. This can raise the cost from several thousand dollars to repair a
defect or scrap a section of pipe to millions of dollars of damage to the environment in a
leakage situation in the field. Since production speed is important in the pipe manufacturing
process, many nondestructive defect location techniques have been investigated. Most
production systems utilize ultrasonic and radiographic testing, but these systems require
significant human intervention in the interpretation of the results. Any time humans are
involved; there exists a chance that a defect could remain undetected and a potentially
235
defective pipe could be shipped to the field for installation. There are many types of potential
defects in pipes, including porosity, inclusions, undercutting, off-seam welds and pipe wall
thinning (PWT). This particular study investigates the possibility of detecting PWT through
the utilization of microwave technology as an input to a neural network [29] [30] [32] [94]
[112] [149] [150].
Previous studies by Alobaidi, et. al. [90] [113] [114] [130] have documented that the
scattering observed in the resonance frequency curve from the microwave signal demonstrate
distinctive patterns for various shapes, depths, lengths and locations of defects in the pipe.
Simulations were conducted utilizing computer simulation technology (CST) software. While
distinctive patterns can be observed when one parameter of the defect is altered at each run, it
becomes much more difficult to evaluate a signal from a defect which is markedly different
than those used in the parametric studies. A significant impact on the returned microwave
signal is observed due to the volume of the defect, but each measured microwave signal
contains unique characteristics that can potentially identify the defect location as well as
provide an estimate of the size, depth and shape of the defect. The possibility of this
determination is the objective of this research. Since changes in the microwave signal can be
observed by the human eye, the question becomes whether or not artificial intelligence
methods such as a neural networks could be employed to automate the detection of defects.
A neural network is designed with the intention of determining the location, length,
depth and shape of the PWT. The output from a large number of test cases generated through
computational simulation serve as the training data for the neural network. Both the
configuration of the neural network and the weighting factors utilized in the neural network
are determined through the application of an evolutionary optimization algorithm. The
optimization was performed since it was not obvious what the configuration of the neural
network and the microwave signals generated in the test cases made it difficult to determine
appropriate weights for the connections between the nodes of the network. The resulting
network is then tested on a series of unknown defects. The goal was not to generate a neural
network that accurately predicted all of the defects, but to demonstrate the potential of the
approach.
236
14.2 Neural Networks
A neural network is essentially a computational model based loosely on the behavior
of a biological brain. The model consists of neurons which are interconnected and can
influence and propagate information to adjoining neurons. With this simplistic model, inputs
can be operated upon by the neurons to generate outputs which mimic a rudimentary form of
intelligence. The specific number and layout of the neurons, as well as how they are
connected is problem dependent and not easily determined. The weighting of the information
passed between neurons is determined through the use of training data. The training data is
selected based upon the fact that both the inputs and outputs are known, so that the weights
can be optimized. The optimization or refinement of the various weights can be performed
through a computational optimization, or through forward and backward propagation. The
process seeks to minimize the errors generated by the network in predicting the correct
output values for the training data. In this way, the weights are selected in a manner that
allows the neural network to correctly predict the outputs for the training data. Then the
network can be utilized to estimate the outputs for data that was not part of the training
process. A wide range of applications have been implemented, particularly in the computer
vision and speech recognition areas.
The history of the neural network can be traced back to the pioneering work of
McCulloch and Pits [151] in the early 1940s. They created a computational model for a
neural network based on simple logic operations. This work led to applications in the area of
artificial intelligence by a number of investigators. The work, however, was hampered by the
lack of computational power until the mid-1970s. The concept of back propagation was
proposed by Werbos [152] in 1974 which allowed for the rapid training of a network model.
This opened the technology to a wide range of applications which started in the 1980s and
1990s and continues today. Through this period, the size and complexity of the neural
networks expanded considerably, along with the difficulty of the application. In previous
work the area of the pipe defect location and classification are somewhat limited, but do
exist. In 2001, Simone et al. [153] applied a neural network do detect various classes of
defects in water pipes using ultrasonic sensors. In 2006 Carvello et al. [2] utilized a neural
network to detect and classify pipe weld defects. This work was based on processing
237
magnetic flux leakage signals. In 2008, Yang and Su [154] applied machine learning to
diagnose faults in sewer pipes based on image processing. In 2014, Chen, Huang and Zhao
[155] looked at three dimensional defect reconstruction with a neural network based on
processing a magnetic flux leakage signal.
In this current work, a rather simplistic neural network was applied to a microwave
signal in order to determine the location, depth, length and type classification of a defect in a
smooth pipe section. The goal is not to construct a sophisticated, large scale neural network
for precise detection and classification of the defect, but rather to test the possibility of
generating a simple neural network model that is capable of predicting defect characteristics
with some reasonable degree of accuracy. In addition, the exact network structure, as well as
the weights attached to each propagated signal from neuron to neuron, is determined through
the application of an evolutionary genetic algorithm. For this model, the complex microwave
signal was preprocessed to form a limited set of inputs for the model, and a single hidden
layer was utilized. The model can easily be extended to include additional complexity, but
this is not the goal of the present work.
14.3 Development of a Neural Network
The basic structure of a neural network is pictured in Fig. 14.1. The network consists
of a number of nodes representing neurons which are connected by links which represent
synapses in the biological equivalent. The neurons on the far left are the inputs to the neural
network model. These are somewhat similar to design variables in an optimization problem.
The inputs are passed to the next layer, known as the hidden layer. As each value is passed
from neuron to neuron, the values currently held in all connected neurons are operated upon
by a weighting factor in order to place a new value in the connected node. The simplest form
would be to simply sum the newly weighted input values passed to the node. The sum may
then be further modified by an operation known as an activation function. The goal of the
activation function is to better separate values in the receiving nodes, by pushing them
toward upper and lower limits. The column of nodes which operate on the input values is
termed the hidden layer because it is generally not visible by the user. There may be several
238
columns of hidden layers, particularly is the application represents a complex system. When
many hidden layers are utilized, the term “deep learning” is associated with the network.
After being operated upon by the hidden layers, the resulting node values are passed to the
output neurons, where the final values for the outputs from the network are calculated using
the same logic as that implemented within the hidden layers.
Figure 14.1: Basic Layout of a Neural Network
The activation function may take on several forms, but the most common are the
linear function, the sigmoid function and the hyperbolic tangent. These three functions are
given as:
=
=
1
(1 +  − )
( )
(14.1)
( )
( 14.2)
( )
(14.3)
and
=
(1 −  −2 )
(1 +  2 )
239
In equations (14.1)-(14.3), the y value is the output from the activation function,
given the value x as the input. The main issue in obtaining valid results from the neural
network model from a given set of inputs, is in the selection of the set of weights associated
with each connecting link between the nodes or neurons so that the inputs are processed in a
way that the correct output is determined. This is accomplished through the use of training
data. Training data refers to sets of input values for which the associated outputs are known.
By running the training data through the network model, with any arbitrary weights assigned
to the node connections, the weights can be modified so that the outputs from the model are
close to the expected values from the training data. The improvement of the weights may be
accomplished by a process known as backward and forward propagation, or by utilizing an
optimization algorithm which selects the weights in the model that produce the minimum
error between the known output values for the training data and the outputs produced by the
model.
The weights may initially be selected randomly. A process known as back
propagation can then be utilized to adjust the weights in order to improve the performance of
the network. In this process, the amount of error in the training data is calculated. The error is
simply the difference between the output value predicted by the network and the actual
output value from the training data. The process moves from the output layer, backward
through the hidden layers and each weight is modified in order to reduce the error. The exact
process of backward propagation is detailed in a number of references [2]. The ultimate
weights used in a neural network will be a compromise. That is, the errors will be reduced
over the training set as a whole, but errors are still present which can lead to incorrect
evaluation of new data fed into the neural network. Part of the problem is a result of having
to specify the topology of the neural network. We generally know how many inputs we need
and what specific outputs are desired from the network. The difficulty is that there is no easy
way to determine how many nodes or layers to utilize in the hidden layers. It is understood
that more layers and the more nodes may allow for better accuracy in the outputs, but this
leads to significantly more weights to assign and may require a large training data set to
accurately determine the weight values. Several alternative topologies for the network may
be required for testing the network in order to obtain good correlation between the outputs
predicted by the neural network and the outputs associated with the training data.
240
Once a network has been trained, it may be applied to data which was not part of the
training data set, and outputs can be estimated. If the performance is judged to be adequate,
the network can remain as configured. If errors still persist in predicting the outputs for new
data, then the network must be modified. This could require additional modification of the
weights, the application of additional training data, or the alteration of the network topology.
By any manual means, or even with backward propagation, the ultimate selection of the
network topology is not an easy process, or one that is guaranteed to produce good results.
The concept here is to assume a generic network configuration and then to let an
evolutionary optimization algorithm refine the topology of the network as well as specify the
appropriate weights utilized in the network. In this manner, the effort involved in creating the
network is significantly reduced.
14.4 Evolutionary Optimization
An evolutionary or genetic optimization algorithm seeks to mimic the process of
biological evolution. Just as natural species adapt to their changing environment, the current
state of a neural network can be improved utilizing the same approach. The nature driven
process promotes survival of the fittest which over many generations produces organisms
which are better able to survive in their surroundings. This natural evolutionary process can
be implemented computationally in order to optimize the weights in a neural network, as well
as alter the topology by removing unneeded nodes. It is also within reason to add additional
nodes in the hidden layers, although some topological framework is required. The use of an
evolutionary algorithm enables a global search for the optimum, which is required on this
class of problems where many local optimums will be present. The goal of an evolutionary
algorithm is to discover the fundamental characteristics which contribute to a good solution
and to exploit these characteristics in such a way as to produce the best possible solution. The
process relies on the randomness present in nature, but quickly exploits information
accumulated during the search to produce an effective solution.
An evolutionary or genetic optimization algorithm seeks to mimic the process of
biological evolution. Just as natural species adapt to their changing environment, the current
241
state of a neural network can be improved utilizing the same approach. The nature driven
process promotes survival of the fittest which over many generations produces organisms
which are better able to survive in their surroundings. This natural evolutionary process can
be implemented computationally in order to optimize the weights in a neural network, as well
as alter the topology by removing unneeded nodes. It is also within reason to add additional
nodes in the hidden layers, although some topological framework is required. The use of an
evolutionary algorithm enables a global search for the optimum, which is required on this
class of problems where many local optimums will be present. The goal of an evolutionary
algorithm is to discover the fundamental characteristics which contribute to a good solution
and to exploit these characteristics in such a way as to produce the best possible solution. The
process relies on the randomness present in nature, but quickly exploits information
accumulated during the search to produce an effective solution.
Evolutionary optimization methods differ from conventional optimization in a
number of other ways beyond the encoding. An evolutionary algorithm operates on a set or
population of potential solutions, rather than on a single point. The chromosomes of selected
members, or parents, from one generation are selected and are combined in order to produce
offspring. The rules utilized in the transition from one generation to the next are probabilistic
in nature, rather than deterministic. This is one means of introducing randomness into the
solution process, although the probability of being selected as a parent is closely tied to the
performance or fitness of the parent relative to the other members of the population. The
chromosomes of the parents are combined using various genetic operators, and mutation
occasionally allowed guaranteeing that no important genetic information is lost during the
progression. These differences may not seem that dramatic, but they produce a search
algorithm which exhibits aspects of intelligence.
14.5 Microwave Modeling for Training Data
This study used computer simulation technology (CST) to set up the modeling for this
research with sweeping frequency bandwidth of 0.5 GHz, determined according to the
equations below. Models and designs have proposed two ports (port 1 and port 2) appropriate
242
to produce that bandwidth inside the pipe. Two amplitudes of the scattering parameter are S 11
and S21 for port 1 yield humps as per the PWT [114]. Fig. 14.2 is a scheme of pipe modeled
in this study for the full circumferential PWT models. In previously published work these
equations were presented:
“Pnm is the mathematical roots for Bessel function are set up as expressed in the equations
below.”, Alobaidi, et. al. [114].
11 =
 

(14.4)
“For air the permittivity, εr = 1, and Cd speed of light inside the pipe for free space is 3x108
m/s. fcTEnm is the lowest cutoff frequency at TE11 mode, Di is pipe inner diameter.”, Alobaidi,
et. al. [114].
21 =
3.0542

1.8412 11
(14.5)
11 =
3.8317

1.8412 11
(14.6)
In this model we used background properties for the material type perfect conductor
(PEC). The mesh type for the model is hexahedral (legacy).
The parameters used for the modeled pipes are: four shapes, three locations, three
depths of defect, and three lengths around the pipe. The shape profiles used are rectangular, 2
filet, semi-circular, and triangular. The locations are measured from port one: 127 mm, 381
mm and 635 mm. The depths are 6.35 mm, 8.89 mm and 11.43 mm. The lengths are 33%,
66% and 100% of circumference.
243
Figure 14.2(a) and (b): Scheme for model pipe with two ports are represented.
LP is pipe length. DD is depth of full circumferential PWT. Di is inner diameter
The results of this modeling are presented in the two graphs, in Figures 14.3 and 14.4.
Fig. 14.3, with one peak, and Fig. 14.4 with two peaks. The modeling result data was used to
build the neural network for this work. According to the two graphs below, we studied the
starting dB, ending dB, dB at peak, frequency at peak, in GHz. We moved down from the
peak -5 dB, -10 dB, and -15 dB, to measure the width of the peak and bounding frequencies.
We studied all the intersections between the horizontal lines of -5 dB, -10 dB, and -15 dB
244
and the intersections to the waveform shape, as shown in the Table 14.1 below. These results
trained the neural network.
Table 14.1: This table shows the results from the model with these parameters: the shape is
rectangular, distance 5 in. (127 mm) from port one to the center of the PWT, depth is 0.45
in. (11.43 mm), and the length is 100% circumferential
Starting dB
Ending dB
dB @ Peak
Frequency @ Peak
(GHz)
-18.394389
-18.153811
-0.007506868
2.3358998
-5 dB
1 Left & 1 Right,
Intersections
Frequency
(GHz)
dB
Left
Right
-5 2.3252 2.345
Width of Peak
-10 dB
1 Left & 1 Right,
Intersections
Frequency
(GHz)
dB
Left
Right
-10 2.312 2.3526
-15 dB
1 Left & 1 Right,
Intersections
dB
-15
Frequency (GHz)
Left
2.282
Right
2.3619
Figure 14.3: Microwave signals for semi-circular shape, distance 381 mm from
port one, depth is 11.43 mm, full circumferential length, with all the resulting
parameters
245
Figure 14.4: Microwave signals for 2-filet shape, distance 381 mm from port
one, depth is 11.43 mm, 66% circumferential length, with all the resulting
parameters
14.6 Creating a Neural Network for Evaluating Pipe Defects
Resonance frequency data from 2.2 GHz to 2.5 GHz was collected for 108 test cases.
These cases utilized various combinations of length, depth, location and profile form for the
defect. These samples became the training data for the neural network. The resonance
frequency curve contains a significant amount of information, much of which is very difficult
to correlate to the defect condition in the pipe. The input parameters to the neural network
were selected as those characteristics which help define the first peak resonance peak as well
as the shape of this peak and the energy level contained in the signal. Additional parameters
could be selected, but through careful observation of the differences in the resonance
frequency curves, these parameters seemed to correlate well to the changes in the test cases
considered. In case additional inputs become necessary to fine tune the neural network, the
actual output signals for the test cases have been preserved.
The location of the defect is by far the most important output from the model.
Previous studies by Alobaidi [113] [114] have shown that the resonance frequency for the
246
first peak contains sufficient information to interpolate the rough location of the defect.
Unfortunately, the depth, length and profile of the defect also influence the curve in a ways
that confound a simple correlation. The question becomes one of whether or not a neural
network can decompose the specific impact of the input parameters which represent the
salient features of the resonance curve. The ultimate test of such a system model is the ability
to accurately predict the outputs for cases other than it was trained on.
The topology selected for the neural network for defect detection in a pipe is shown
in Fig. 14.5. The first nine inputs are input as three groups of three parameters which
represent the width and general shape characteristics of the first resonance peak. The three
parameters are the width of the first resonance peak, the width of the associated double hump
(as shown in Fig. 14.4), if one exists, and whether or not the hump is to the left or right of the
main peak. These three parameters are collected for the peak at levels 5, 10 and 15 db down
from the peak value of the signal. Taken as a whole, these inputs define the resonance peak
in magnitude and form. They are fed directly into the hidden layer nodes as combined inputs
without consideration of the values of the lower inputs. These inputs include the initial dB
level of the signal (at 2.2 GHz), the final Db level of the signal (at 2.5 GHz), the peak dB
level of the first resonance peak and the frequency at which this peak occurs. These four
inputs are fed into the hidden layer and finally combined with the first nine values when sent
to the output node. In the initial model all links are considered to be active, but as the
network is optimized, unneeded links are removed.
All values from the hidden layer are summed together with the associated link
weights used to multiply each value transferred from one node to another. This produces the
final output value, which should correspond closely to the actual measured outputs for the
training data. The weighting factor for each connection must be selected in order to make this
happen. The current network topology contains 50 links, so a total of 50 weights must be
determined. The activation function for each node was selected as a linear function with the
hope that if the output values could be determined accurately enough, interpolation would
allow the neural network to consider cases which were not closely represented by the training
data. All input values were scaled to values between zero and one. Four separate networks,
all in the general form of that shown in Fig. 14.5 were implemented. This allowed a separate
247
combination of the input parameters to be utilized for each of the four output characteristics
of the defect. Additional connections between nodes or additional layers of hidden nodes
could be added in order to increase the potential effectiveness of the neural network. This
exercise is left for future research. The current network configuration topology is deemed to
be sufficient in order to test the effectiveness of a neural network approach and to help
identify areas where improvement is necessary
Figure 14.5: Assumed Neural Network Topology
248
14.7 Optimizing the Neural Network
With the neural network topology specified, the next step in the process is to
determine the weights for the network that produce the best results in predicting the actual
outputs in the training data. In addition, it would be valuable to eliminate the connections
between nodes, or even the nodes themselves if they are not needed. In order to determine the
best, or optimal weights, and to modify the topology of the initial network model, an
evolutionary optimization algorithm was applied. In order to apply the optimization
algorithm, an encoding which represents the network model must be developed as well as an
appropriate objective function. The objective function should be constructed in order to
minimize the error that the network makes in predicting the outputs for the training data.
The encoding for the neural network model consists of two parts. The first is a
representation that allows for the determination of the weights associated with the links
between the nodes. The second represents the links themselves. That is whether or not the
link can be removed without hurting the predictive ability of the network. By removing the
unneeded links and nodes, a less complex network is constructed and the important
relationships between the input parameters may be easily identified. The weights for the
selected model amount to 50 separate values. These values were allowed to take on values in
the range from -1 to 1. Four hundred discrete values in this range were allowed. The next 50
positions in the chromosome represented whether or not a particular link was to be included
in the evaluation of the performance of the particular weight values contained in the
chromosome. These values were restricted to the binary values of zero and one. The value of
zero removes that particular link, while the value of one maintains it. Using this encoding
scheme, a population of 5000 sets of weights and link inclusions were utilized in the
optimization process.
Several forms for the objective function were utilized. The first objective function
form simply minimized the overlap between the various output levels. The output levels were
specified as integers between one and the maximum number of levels utilized in the training
data (3 for position, depth and length and 4 for defect profile type). The predicted outputs
from the model for the training data were grouped by type and the overlap among each range
of values for each type were evaluated for overlap. If a solution could be generated which
249
produced zero overlap, then the network would be able to accurately predict the outputs. If
this were possible, an additional optimization could be used to maximize the distance
between the ranges of output values representing each output type. Unfortunately, the
optimization was unable to produce a resulting network model which did not include some
level of overlap among the output types. This result is not unexpected as many of the input
parameters have similar impact on the output values.
The problem with the first objective function was that the training data produced
several outliers for each of the output levels. These outliers tended to dominate the
optimization which hurt the overall effectiveness of the optimization approach. The next
objective function tried was simply to sum the squared error between the predicted output
value and that predicted by the network model. This objective function was represented as:
 () = ∑(   –   )2
(7)
Here the summation is over the entire training data set of 108 individual data sets.
The result from this optimization was better, but several outliers still tended to dominate the
results, which again hurt the overall accuracy of the network model. Finally, an objective
function was utilized which simply summed the number of mistakes made in predicting the
outputs from the training data. For each mistake, the mistake counter was incremented and
the optimization sought to minimize the number of mistakes. For this objective function
evaluation, the value predicted from the neural network was rounded to the nearest integer
value. This objective function produced the best overall results, although the number of
mistakes made was not zero for the prediction of any of the four output parameters and their
individual levels. At least in this case, most of the mistakes were for cases were the predicted
value was still close to the actual value, although there were still a few outliers. With this
formulation, however, the outliers did not dominate the solution. Once again, this is a result
of the confounding impact of the individual input values on the outputs. Perhaps a more
complex neural network topology would help, but even for the initial topology, many links
were found to be unnecessary.
A typical result for the output of the neural network is shown below for the defect
location. Fig. 14.6 shows the output distribution from the optimized network for the first
250
defect location. Figures 14.7 and 14.8 show the output distribution for the second and third
defect location and Fig. 14.9 shows all of the training data results for the defect location.
Ideally, all defects at location one should have values from .5 to 1.5, while defects at location
two should range from 1.5 to 2.5 and defects at location 3 should range from 2.5 to 3.5. Most
of the data points fall into the correct range, with the exception of a group of points for defect
positions one and three. For these cases, there are a group of defects that register as the next
higher or lower defect location. This situation is simply an indication that some of the various
inputs, have similar, or confounding effects on the resonance curve and therefore on the
relationship between the inputs and outputs. This can be seen clearly in Fig. 14.8 where the
training data predicts the defect location fairly well overall, but the outliers are definitely
present. Perhaps the outliers can be utilized, along with an understanding of the optimal
topology for the network in order to generate a deeper version of the neural network.
Figure 14.6: Output Distribution for Defect Location for the First
Output Level
251
Figure 14.7: Output Distribution for Defect Location for the Second
Output Level
Figure 14.8: Output Distribution for Defect Location for the Third
Output Level
252
Figure 14.9: Output Distribution for Defect Location for all Output Levels
14.8 Results
In order to test the capability of the optimized neural network in identifying the four
outputs for unknown cases, a test case consisting of twelve additional pipe defect simulations
were executed and the resulting inputs were fed into the network models. The twelve test
cases are listed in Table 14.2 as to the location, depth, length, and profile type of the defect
modeled. All samples of unknown PWT are modeled and simulated with the parameters as
listed below. For example, Shape: right-angle triangle, location: 127 mm distance from port
one to the center of PWT, depth: 11.43 mm, 100% circumferential. The input data was fed
into the networks by an individual who had no knowledge of the true output values. The
resulting outputs from the neural networks are documented in Table 14.3. The results are not
perfect, but they are encouraging.
Overall the agreement with the actual output values is
good. There are, however, several outputs that the neural network made. This is not
surprising as this is a very complex problem and the interplay between the input values is not
easily modeled. Each of the four output categories will be considered separately and then the
optimal topology for each of the neural networks is considered.
253
Table 14.2: All the parameters for the unknown defects designed and modeled to make
data entry to test the neural network
Number
1
Number
2
1
3
24
35
46
7
5
8
69
710
11
8
12
9
Shape
Right-triangle
Shape
1 filet 1 taper
Right-triangle
1 filet 1 right angle
12filet
filet1 taper
12filet
filet1 right angle
filet
22filet
Rectangular
2 filet
Rectangular
2Rectangular
filet
Triangular
Rectangular
2 filet
Rectangular
Rectangular
Rectangular
Location (mm)
127
Location
(mm)
381
127
635
381
127
635
381
635
127
127
381
381
635
635
101.6
127
406.4
381
609.6
635
Depth (mm)
11.43(mm)
Depth
11.43
11.43
11.43
11.43
11.43
11.43
11.43
11.43
11.43
11.049
11.43
8.509
11.43
5.969
11.43
11.049
11.43
8.509
11.43
5.969
Length (%)
100 (%)
Length
100
100
100
100
90
100
60
40
90
100
60
100
40
100
100
100
100
100
100
100
10
Triangular
101.6
11.43
100
11
2 filet
406.4
11.43
100
Table
predicted outputs from
neural networks
12 14.3: The
Rectangular
609.6the optimized 11.43
100
Table 14.2. All the parameters for the unknown defects designed and modeled to make
Number Shape
Location (mm)
Depth (mm)
Length (%)
data entry to test the neural network
1
Semi-Circular
127
8.89
100
2
Rectangular
381
11.43
100
3
Rectangular
11.43
100
381
4
Rectangular
11.43
100
381
5
2 filet
381
6.35
66
6
2 filet
635
11.43
66
7
Rectangular
127
11.43
100
8
2 filet
381
8.89
100
9
2 filet
635
8.89
100
10
Semi-Circular
127
8.89
100
11
Rectangular
381
11.43
100
12
Rectangular
11.43
100
381
254
The shape prediction was extremely good. This is somewhat surprising as the neural
network for this case had significant compounding among the input variables for this output.
In other words, the results from neural network were better than expected due to the
difficulty separating the cases in the test set. Additionally, several geometries for the defect
that were not utilized in the training data were included in the test set. Only in one case, did
the network miss the true defect profile. This occurred in test samples one and 10 where a
triangular defect was recognized as a semi-circular defect. On a defect volume measure, this
is not a serious mistake. It must be remembered that the output values are extremely
dependent upon the volume of the defect. For several of the other profiles, a fillet was
predicted as a sharp corner, but for all of these cases the predicted defect profile is very close
geometrically to the actual profile.
The defect location prediction was also fairly good. Only three errors were made by
the network. These errors were for test cases three, four and twelve. In each of these cases,
the defect position was predicted to be in the middle location. In two of the cases, the defect
was actually located in the third location and once in the initial location. So even though the
exact position was missed, at least the network predicted a location that was next to the
actual location. The fact that the prediction errors produced by the neural network were
consistent allows for corrections in the network topology and weights to rectify this situation.
This is important as the defect location is by far the most important of the output parameters.
The depth of the defect was also predicted fairly well. Four cases were predicted
incorrectly by the model. These cases were test cases one, five, nine and ten. Several of the
test cases used locations not included in the training data, and for each of these cases, the
nearest correct location was predicted. The errors made, were again relegated to missing the
true depth by the depth nearest to the actual value. It is expected that a shallow defect would
be the hardest to determine, and this was the case for the neural network model. Once again,
the results may be utilized to improve the performance of the network. This would probably
require a more complex network topology.
255
The final output parameter is the length of the defect. This output was predicted
almost perfectly by the neural network model. The only discrepancy was for the cases that
included lengths that were not in the training data. For these cases, however, the closest
length from the training data was predicted. The use of output value in the test set that were
not included in the training data was based on being able to interpolate between values in the
training data, but the current state of the network is not quite accurate enough to allow this to
happen.
The optimal topologies for each of the four output parameters are shown in Figures
14.10, 14.11, 14.12 and 14.13. Note that each has a different topology. This is a direct result
of the inclusion of topological variable in the optimization process. The optimal network for
the depth of the defect, as shown in Fig. 14.10, eliminated two of the nodes in the hidden
layer and a majority of the links between the final four inputs. These four inputs are all
related to the dB levels of the signal and the first resonance peak. It appears that the depth is
a strong function of the width of the signal at the first resonance peak. The optimal topology
for the defect length, as shown in Fig. 14.11, maintains all of the hidden layer nodes as well
as most of the links connecting the nodes. Thus, the prediction of the defect length can be
seen to require information of all of the inputs and combinations of inputs. The defect
position network, shown in Fig. 14.12, is similar to that for the depth in that it represents a
simplified topology. Two of the hidden layer nodes from the initial topology were removed
and the relationships among the input variables were simplified through the removal of many
of the connecting links. Finally, the network for the defect profile is shown in Fig. 14.13.
This model was considerable simplified as three of the hidden layer nodes were removed and
the majority of the links connecting nodes were removed as well. The fact that the optimized
network predicted the correct profile for the majority of test cases is encouraging, but the
simplified network may be an indication of the lack of correlation between the input
parameters selected and the profile of the defect.
256
Figure 14.10: Optimal Neural Network Topology for Defect
Depth Detection
257
Figure 14.11: Optimal Neural Network Topology for Defect
Length Detection
258
Figure 14.12: Optimal Neural Network Topology for Defect
Position Detection
259
Figure 14.13: Optimal Neural Network Topology for Defect Type
Detection
260
Figure 14.14: Output Distribution for Defect Depth for all Output Levels
Figure 14.15: Output Distribution for Defect Length for all Output Levels
261
Figure 14.16: Output Distribution for Defect Type for all Output Levels
The confounding among the different output levels for the depth, length and type of
defect are shown in Figures 14.14, 14.15 and 14.16. The predicted level for each output is
plotted for each of the training data files. The output value levels are separated for the data
set and each is plotted as a separate curve in a separate color. In each of these cases, there is a
fair amount of separation among the predicted output levels. Unfortunately, there is still a
significant number of outliers in these curves. The positive aspect is that the outlying points
are clearly identified and the input data sets for these specific cases can now be reviewed,
along with the optimal network topologies to refine the network.
14.9 Summary and Conclusions
The application of a neural network to the prediction of the location, depth, length
and profile of a pipe defect based on an electromagnetic reflected waveform was attempted.
A series of 108 training cases were generated through a computational modeling technique.
An initial network configuration was created and an evolutionary optimization technique was
utilized to select the best weight for each link in the network as well as the optimal topology
262
of the network itself. The ability to alter the topology is clearly an important concept as there
is no easy way to select an appropriate topology for a specific problem. As the number of
hidden layers increases, the ability to alter the topology becomes even more important.
Four separate neural networks were optimized, one for each of the output parameters
for the defect. A series of twelve test cases that contained combinations not included in the
training data were generated and tested on the optimized networks. The optimized networks
performed well on the test cases, but not perfectly. This is to be expected on an initial model.
The optimal topologies for the four outputs were considerably different, which allows the
network to be modified based upon the cases that the network predicted the output
incorrectly. Even when the model did make a mistake on the actual output value; it predicted
a value close to the actual. This is a clear demonstration that the network is performing well.
The output value predictions for each level of the output variable produced by the
optimization process demonstrate overlap among the output values. The fact that the
optimizer was not able to clearly separate each output for all of the others is an indicator of
the difficulty of the problem.
Only thirteen input values were selected to represent the important features of a very
complex frequency resonance plot of the reflected microwave signal. Additional inputs could
potentially increase the accuracy of the model. The same is true for the complexity of the
neural network. By adding additional hidden layers, improvements in the performance may
be possible. This could be accomplished by enlarging the network and once again optimizing
the topology through the use of the evolutionary optimization. These are areas for future
research. The goal of this research was to demonstrate the potential of the application of a
neural network to this problem class using microwave technology. This goal has been
reached.
263
References
[1]
W. Alobaidi, E. Sandgren, and H. Al-Rizzo, “A Survey on Benchmark Defects
Encountered in the Oil Pipe Industries,” International Journal of Scientific &
Engineering Research, vol. 6, no. 2, pp. 844–853, 2015.
[2]
A. A. Carvalho, J. M. A. Rebello, L. V. S. Sagrilo, C. S. Camerini, and I. V. J.
Miranda, “MFL Signals and Artificial Neural Networks Applied to Detection and
Classification of Pipe Weld Defects,” NDT and E International, vol. 39, pp. 661–667,
2006.
[3]
“U.S. Department of Transportation / Pipeline & Hazardous Materials Safety
Administration.” [Online]. Available:
http://primis.phmsa.dot.gov/comm/FactSheets/FSMaterialWeldFailure.htm.
[Accessed: 02-Feb-2015].
[4]
J. Kollar and K. Lager, “Nondestructive Optical Technique for Electrically Testing
Insulated-Gate Integrated Circuits,” Electronics Letters, vol. 7, no. 15, pp. 432–433,
1971.
[5]
J. R. Wait, “Review of Electromagnetic Methods in Nondestructive Testing,”
PROCEEDINGS OF THE IEEE, vol. 67, no. 6, pp. 892–903, 1979.
[6]
H. Hoshikawa, H. Saiton, J. Koido, and Y. Ishibashi, “Energy Flow in Remote Field
Eddy Current Nondestructive Testing,” IEEE Transactions on Magnetcs, vol. 26, no.
2, pp. 885–888, 1990.
[7]
X. W. Dai, R. Ludwig, and R. Palanisamy, “Numerical Simulation of Pulsed EddyCurrent Nondestructive Testing Phenomena,” IEEE Transactions on Magnetics, vol.
26, no. 6, pp. 3089–3096, 1990.
[8]
O. Michelsson and F. H. Uhlmann, “On the Use of the 3D H-&-Formulation for the
Forward Solution for Eddy Current Nondestructive Testing,” IEEE Transactions on
Magnetics, vol. 34, no. 5, pp. 2672–2675, 1998.
264
[9]
R. V. Sabariego and P. Dular, “A Perturbation Approach for the Modeling of Eddy
Current Nondestructive Testing Problems With Differential Probes,” IEEE
Transactions on Magnetics, vol. 43, no. 4, pp. 1289–1292, 2007.
[10] S. Caorsi, A. Massa, and M. Pastorino, “A Crack Identification Microwave Procedure
Based on a Genetic Algorithm for Nondestructive Testing,” IEEE Transactions on
Antennas and Propagation, vol. 49, no. 12, pp. 1812–1820, 2001.
[11] E. Kerbrat, C. Prada, D. Cassereau, and M. Fink, “Ultrasonic Nondestructive Testing
of Scattering Media Using the Decomposition of the Time-Reversal Operator,” IEEE
Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 49, no. 8,
pp. 1103–1113, 2002.
[12] J. R. Claycomb, A. Brazdeikis, M. Le, R. A. Yarbrough, G. Gogoshin, and J. H.
Miller, “Nondestructive Testing of PEM Fuel Cells,” IEEE Transactions on Applied
Superconductivity, vol. 13, no. 2, pp. 211–214, 2003.
[13] A. Bernieri, G. Betta, L. Ferrigno, and M. Laracca, “A Biaxial Probe for
Nondestructive Testing on Conductive Materials,” IEEE Transactions on
Instrumentation and Measurement, vol. 53, no. 3, pp. 678–684, 2004.
[14] D. Vasić, V. Bilas, and D. Ambruš, “Pulsed Eddy-Current Nondestructive Testing of
Ferromagnetic Tubes,” IEEE Transactions on Instrumentation and Measurement, vol.
53, no. 4, pp. 1289–1294, 2004.
[15] H.-C. Wu, N. Gupta, and P. S. Mylavarapu, “Blind Multiridge Detection for
Automatic Nondestructive Testing Using Ultrasonic Signals,” IEEE transactions on
ultrasonics, ferroelectrics, and frequency control, vol. 53, no. 10, pp. 1902–1911,
2006.
[16] N. N. Qaddoumi, W. M. Saleh, and M. Abou-Khousa, “Innovative Near-Field
Microwave Nondestructive Testing of Corroded Metallic Structures Utilizing OpenEnded Rectangular Waveguide Probes,” IEEE Transactions on Instrumentation and
Measurement, vol. 56, no. 5, pp. 1961–1966, 2007.
265
[17] Q. Huang, Y. Wu, J. Baruch, P. Jiang, and Y. Peng, “A Template Model for Defect
Simulation for Evaluating Nondestructive Testing in X-Radiography,” IEEE
Transactions on Systems, Man, and Cybernetics-Part A:Systems and Humans, vol. 39,
no. 2, pp. 466–475, 2009.
[18] “Radiograph Interpretation - Welds.” [Online]. Available: https://www.ndeed.org/EducationResources/CommunityCollege/Radiography/TechCalibrations/Radio
graphInterp.htm. [Accessed: 02-Feb-2015].
[19] A. Cosham and P. Hopkins, “The Effect of Dents in Pipelines-Guidance in the
Pipeline Defect Assessment Manual,” International Journal of Pressure Vessels and
Piping, vol. 81, pp. 127–139, 2004.
[20] “OLYMPUS.” [Online]. Available: http://www.olympus-ims.com/en/ndttutorials/flaw-detection/weld-overview/. [Accessed: 02-Feb-2015].
[21] “ATELIERS B.G.INC.” [Online]. Available: http://www.ateliersbg.com/news/the-5types-of-basic-welding-joints.aspx. [Accessed: 02-Feb-2015].
[22] “IHS Engineering 360.” [Online]. Available:
http://www.globalspec.com/learnmore/manufacturing_process_equipment/welding_eq
uipment_supplies/welding_brazing_soldering_equipment. [Accessed: 02-Feb-2015].
[23] “Welding Defects.” [Online]. Available:
http://nptel.ac.in/courses/112107144/welding/lecture13.htm. [Accessed: 02-Feb-2015].
[24] R. Drai, A. Benchaala, M. Zergoug, A. Badidi Bouda, Y. Kabir, and J. Hadj Djilani,
Development of protocols for corrosion and deposits evaluation in pipes by
radiography, no. April. 2005.
[25] Y. Ju, L. Liu, and M. Ishikawa, “Quantitative Evaluation of Wall Thinning of Metal
Pipes by Microwaves,” Materials Science Forum, vol. 614, pp. 111–116, 2009.
[26] L. Linsheng, “Remote Detection and Quantitative Evaluation of Wall Thinning
Volumes in a Metal Pipe,” E-Journal of Advanced Maintenance, vol. 2, pp. 101–109,
2010.
266
[27] H. Zhang, B. Gao, G. Y. Tian, W. L. Woo, and L. Bai, “Metal defects sizing and
detection under thick coating using microwave NDT,” NDT & E International, vol.
60, pp. 52–61, Dec. 2013.
[28] S.-H. Yang, K.-B. Kim, and J.-S. Kang, “Detection of surface crack in film-coated
metals using an open-ended coaxial line sensor and dual microwave frequencies,”
NDT & E International, vol. 54, pp. 91–95, Mar. 2013.
[29] W. M. Alobaidi, E. A. Alkuam, H. M. Al-Rizzo, and E. Sandgren, “Applications of
Ultrasonic Techniques in Oil and Gas Pipeline Industries: A Review,” American
Journal of Operations Research, vol. 5, no. 4, pp. 274–287, 2015.
[30] W. M. Alobaidi, E. A. Alkuam, E. Sandgren, and H. M. Al-rizzo, “Enhancing
Production Efficiency of Oil and Natural Gas Pipes Using Microwave Technology,”
Energy and Power Engineering, vol. 7, pp. 440–450, 2015.
[31] Y. Cho, W. D. Oh, and J. H. Lee, “A Wall Thinning Detection and Quantification
Based on Guided Wave Mode Conversion Features,” Key Engineering Materials, vol.
321–323, pp. 795–798, 2006.
[32] Y. Ju, “Remote Measurement of the Pipe Thickness Reduction by Microwaves,” in
ASME 2007 Pressure Vessels and Piping Conference. American Society of
Mechanical Engineers, 2007, pp. 177–179.
[33] F. Inada, Y. Watanabe, T. Ohira, and A. Nakamura, “State of the Art Study for the
Basis of Pipe Wall Thinning Managmenet in Japan,” in 2013 21st International
Conference on Nuclear Engineering. American Society of Mechanical Engineers
(ASME), 2013, p. V001T01A012.
[34] L. Liu, Y. Ju, M. Chen, and D. Fang, “Application of Microwaves for Nondestructive
and High-Efficiency Detection of Wall Thinning Locations in a Long-Distance Metal
Pipe,” Materials Transactions, vol. 52, no. 11, pp. 2091–2097, 2011.
[35] N. Saber, Y. Ju, H. Y. Hsu, and S. H. Lee, “A feasibility study on the application of
microwaves for online biofilm monitoring in the pipelines,” International Journal of
267
Pressure Vessels and Piping, vol. 111–112, no. September 2015, pp. 99–105, 2013.
[36] P. Sriyutha Murthy, R. Venkatesan, K. V. K. Nair, D. Inbakandan, S. Syed Jahan, D.
Magesh Peter, and M. Ravindran, “Evaluation of sodium hypochlorite for fouling
control in plate heat exchangers for seawater application,” International
Biodeterioration and Biodegradation, vol. 55, no. 3, pp. 161–170, 2005.
[37] L. Liu, “Application of Microwave for Remote NDT and Distinction of Biofouling
and Wall Thinning Defects Inside a Metal Pipe,” Journal of Nondestructive
Evaluation, vol. 34, no. 4, pp. 1–8, 2015.
[38] L. Liu, Y. Ju, and M. Chen, “Optimizing the frequency range of microwaves for highresolution evaluation of wall thinning locations in a long-distance metal pipe,” NDT &
E International, vol. 57, pp. 52–57, Jul. 2013.
[39] D. Pozar, Microwave Engineering, Fourth Edi. 2005.
[40] “NDT Resource Center, The Collaboration for NDT Education, Iowa State University
(2001-2014) Introduction to Ultrasonic Testing (Basic Principles of Ultrasonic
Testing, Transducer Types, Normal Beam Inspection, Angle Beams I, Angle Beams II,
Angle Beam Ins.” [Online]. Available: https://www.ndeed.org/GeneralResources/Formula/AngleBeamFormula/AngleBeamTrig.htm.
[Accessed: 05-May-2015].
[41] M. D. Beard and M. J. S. Lowe, “Non-destructive testing of rock bolts using guided
ultrasonic waves,” International Journal of Rock Mechanics and Mining Sciences, vol.
40, no. 4, pp. 527–536, Jun. 2003.
[42] S. Chaki and G. Bourse, “Guided ultrasonic waves for non-destructive monitoring of
the stress levels in prestressed steel strands.,” Ultrasonics, vol. 49, no. 2, pp. 162–171,
Feb. 2009.
[43] B. W. Drinkwater and P. D. Wilcox, “Ultrasonic arrays for non-destructive evaluation:
A review,” NDT & E International, vol. 39, no. 7, pp. 525–541, Oct. 2006.
[44] B. Dutton, A. R. Clough, M. H. Rosli, and R. S. Edwards, “Non-contact ultrasonic
268
detection of angled surface defects,” NDT & E International, vol. 44, no. 4, pp. 353–
360, Jul. 2011.
[45] D. S. Caravaca, C. Bird, and D. Kleiner, “Ultrasonic phased array inspection of
electrofusion joints in polyethylene pipes,” Insight: Non-Destructive Testing and
Condition Monitoring, vol. 49, no. 2, pp. 83–86, 2007.
[46] J. R. Davis, ASME Handbook, Nondestructive Evaluation and Quality Control, 9th
Editio. 1992.
[47] V. Matz, M. Kreidl, and R. Šmíd, “Classification of Ultrasonic Signals,” in The 8th
International Conference of the Slovenian Society for Non-Destructive Testing.
Application of Contemporary Non-Destructive Testing in Engineering, 2005, pp. 27–
33.
[48] L. Mazeika, R. Kazys, R. Raisutis, and R. Sliteris, “Ultrasonic guided wave
tomography for the inspection of the fuel tanks floor,” International Journal of
Materials and Product Technology, vol. 41, no. 1/2/3/4, pp. 128–139, 2011.
[49] A. McNab and H. S. Young, “Knowledge-based approach to the formulation of
ultrasonic nondestructive testing procedures,” IEE Proceedings, Science, vol. 136, no.
3, pp. 134–140, 1989.
[50] A. K. Nandi, D. Mampel, and B. Roscher, “Blind deconvolution of ultrasonic signals
in nondestructive testing applications,” IEEE Transactions on Signal Processing, vol.
45, no. 5, pp. 1382–1390, May 1997.
[51] Y. Mao and P. Que, “Application of Hilbert-Huang signal processing to ultrasonic
non-destructive testing of oil pipelines,” Journal of Zhejiang University SCIENCE A,
vol. 7, no. 2, pp. 130–134, Feb. 2006.
[52] M. Ricci, L. Senni, and P. Burrascano, “Exploiting Pseudorandom Sequences to
Enhance Noise Immunity for Air-Coupled Ultrasonic Nondestructive Testing,” IEEE
Transactions on Instrumentation and Measurement, vol. 61, no. 11, pp. 2905–2915,
2012.
269
[53] A. J. Salazar and E. C. Rodríguez, “Studies of the effect of surface roughness in the
behaviour of ultrasonic signals in AISI-SAE-4340 steel: spectral and wavelets
analysis,” International journal of microstructure and materials properties, vol. 6, no.
3/4, pp. 224–235, 2011.
[54] C. E. R. Silva, A. V. Alvarenga, and R. P. B. Costa-Felix, “Nondestructive Testing
Ultrasonic Immersion Probe Assessment and Uncertainty Evaluation According to EN
12668-2:2010,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency
Control, vol. 59, no. 10, pp. 2338–2346, 2012.
[55] O. Zahran, S. Shihab, and W. Al-Nuaimy, “Recent Developments in Ultrasonic
Techniques For Rail-Track Inspection,” in Northampton British Institute of NonDestructive Testing, 2002, pp. 55–60.
[56] “OLYMPUS, Ultrasonic Flaw Detection Tutorial (What is Ultrasound).” [Online].
Available: http://www.olympus-ims.com/en/ndt-tutorials/flaw-detection/ultrasound/.
[Accessed: 05-May-2015].
[57] H. Lei, Z. Huang, W. Liang, Y. Mao, and P. W. Que, “Ultrasonic Pig for Submarine
Oil Pipeline Corrosion Inspection,” Russian Journal of Nondestructive Testing, vol.
45, no. 4, pp. 285–291, 2009.
[58] “OLYMPUS, Ultrasonic Flaw Detection Tutorial (Wave Propagation).” [Online].
Available: http://www.olympus-ims.com/en/ndt-tutorials/flaw-detection/wavepropagation/. [Accessed: 05-May-2015].
[59] J. Lucassen and M. Tempel, “Longitudinal Waves on Visco-Elastic Surfaces,” Journal
of Colloid and Interface Science, vol. 41, no. 3, pp. 491–498, 1972.
[60] P. Gabriels, R. Snieder, and G. Nolet, “In Situ Measurements of Shear-Wave Velocity
in Sediments with Higher-Mode Rayleigh Waves,” Geophysical Prospecting, vol. 35,
pp. 187–196, 1987.
[61] J. Lysmer and G. Waas, “Shear Waves in Plane Infinite Structures,” Journal of the
Engineering Mechanics Division, vol. 98, no. 1, pp. 85–105, 1972.
270
[62] N. P. Aleshin, Y. L. Gobov, A. V Mikhailov, Y. G. Smorodinskii, and M. M. Syrkin,
“Automatic Ultrasonic Inspection of Large Diameter Pipes,” Russian Journal of
Nondestructive Testing, vol. 50, no. 3, pp. 133–140, 2014.
[63] D. N. Alleyne and P. Cawley, “The Interaction of Lamb Waves with Defects,” IEEE
transactions on ultrasonics, ferroelectrics, and frequency control, vol. 39, no. 3, pp.
381–397, 1992.
[64] K. R. Leonard and M. K. Hinders, “Lamb Wave Tomography of Pipe-Like
Structures,” Ultrasonics, vol. 43, no. 7, pp. 574–583, 2005.
[65] “OLYMPUS, Ultrasonic Flaw Detection Tutorial (General Concepts of Transducer
Selection).” [Online]. Available: http://www.olympus-ims.com/en/ndt-tutorials/flawdetection/general-concepts/. [Accessed: 05-May-2015].
[66] M. Lach, M. Platte, and A. Ries, “Piezoelectric Materials for Ultrasonic Probes,” NDT
Net, vol. 1, no. 9, 1996.
[67] M. J. S. Lowe, D. N. Alleyne, and P. Cawley, “Defect Detection in Pipes Using
Guided Waves,” Ultrasonics, vol. 36, no. 1–5, pp. 147–154, 1998.
[68] M. H. S. Siqueira, C. E. N. Gatts, R. R. Da Silva, and J. M. A. Rebello, “The Use of
Ultrasonic Guided Waves and Wavelets Analysis in Pipe Inspection,” Ultrasonics,
vol. 41, no. 10, pp. 785–797, 2004.
[69] D. Cleveland, A. R. Barron, and A. N. Mucciardi, “Methods For Determining The
Depth Of Near-Surface Defects,” Journal of Nondestructive Evaluation, vol. 1, no. 1,
pp. 21–36, 1980.
[70] “OLYMPUS, Ultrasonic Flaw Detection Tutorial, Straight Beam Tests (Plates, Bars,
Forgings, Castings, etc.).” [Online]. Available: http://www.olympus-ims.com/en/ndttutorials/flaw-detection/straight-beam-tests/. [Accessed: 05-May-2015].
[71] J.-S. Yang, X.-G. Zhao, Y.-J. Zhang, and Z. Cao, “A New Type of Wheeled Intelligent
Ultrasonic Thickness Measurement System,” in 2013 Symposium on Piezoelectricity,
Acoustic Waves and Device Applications (SPAWDA), Oct.25-27, Changsha, Hunan,
271
China, 2013, pp. 1–4.
[72] J. Okamoto, J. C. Adamowski, M. S. G. Tsuzuki, F. Buiochi, and C. S. Camerini,
“Autonomous system for oil pipelines inspection,” Mechatronics, vol. 9, pp. 731–743,
1999.
[73] A. Contreras, S. L. Hernández, R. Orozco-Cruz, and R. Galvan-Martínez, “Mechanical
and environmental effects on stress corrosion cracking of low carbon pipeline steel in
a soil solution,” Materials and Design, vol. 35, pp. 281–289, 2012.
[74] M. Ahammed and R. E. Melchers, “Reliability estimation of pressurised pipelines
subject to localised corrosion defects,” International Journal of Pressure Vessels and
Piping, vol. 69, no. 3, pp. 267–272, 1996.
[75] K. Abbasi, S. Ito, and H. Hashizume, “Prove the ability of microwave nondestructive
method combined with signal processing to determine the position of a circumferential
crack in pipes,” International Journal of Applied Electromagnetics and Mechanics,
vol. 28, pp. 429–439, 2008.
[76] P. Sarkar, “OCS Online Clothing Study (2015) What is the Meaning of SAM in
Garment Industry?” [Online]. Available:
http://www.onlineclothingstudy.com/2013/12/what-is-meaning-of-sam-ingarment.html. [Accessed: 07-Jul-2015].
[77] “EDUCATION YEAR (2015) Calculate SAM of a Garment.” [Online]. Available:
http://www.onlineclothingstudy.com/2013/12/what-is-meaning-of-sam-ingarment.html. [Accessed: 07-Jul-2015].
[78] “BLUE STAR, Professional Electronics & Industrial Systems Division (2015)
Material Testing-Non Destructive, Automation Project Capabilities (Ultrasonic
Testing System).” [Online]. Available:
http://electronics.bluestarindia.com/MatTestNonDest/Auto Proj.htm. [Accessed: 07Jul-2015].
[79] “AMERICAN, THE RIGHT WAY (2015) AMERICAN Steel Pipe Manufacturing
272
Process.” [Online]. Available: http://www.american-usa.com/resources/americansteel-pipe-resources/american-steel-pipe-manufacturing-process. [Accessed: 07-Jul2015].
[80] “YXLON, Technology with Passion (2015) Inspection of pipe segments with
longitudinal and spiral welds at pipe mills.” [Online]. Available:
http://www.yxlon.com/Applications/Welds/Pipe-mills. [Accessed: 07-Jul-2015].
[81] “HUNGER, Hydraulik (2015) Hydraulic Equipment for the Mechanical Engineering.”
[Online]. Available: http://hunger-hydraulik.de/Mechanical-engineering.html.
[Accessed: 07-Jul-2015].
[82] W. Deutsch, M. Gessinger, and M. Joswig, “ECHOGRAPH Ultrasonic Testing of
Helical Submerged Arc-Welded ( HSAW ) Pipes,” in 18th World Conference on
Nondestructive Testing, 16-20 April 2012, Durban, South Africa, pp. 1–12.
[83] “GE Measurement & Control (2015) Krautkramer USN 60/60L, Portable Ultrasonic
Flaw Detectors.” [Online]. Available:
http://www.gemeasurement.com/sites/gemc.dev/files/usn_60_60l_brochure.pdf.
[Accessed: 07-Jul-2015].
[84] “YXLON, Technology with Passion (2015) Y. Pipe Solutions – X-ray systems for
pipe inspection.” [Online]. Available: http://www.yxlon.com/Products/Pipeinspection-systems. [Accessed: 07-Jul-2015].
[85] “U. S. Steel Tubular Products (2015) High Frequency Electric Weld (ERW) Line Pipe
and Standard Products.” [Online]. Available: http://usstubular.com/standard-and-linesteel-pipe/high-frequency-electric-weld-(erw)-line-pipe-and-s. [Accessed: 08-Aug2015].
[86] “U.S. Department of Transportation, Pipeline & Hazardous Materials Safety
Administration (2015) Fact Sheet: Pipe Manufacturing Process.” [Online]. Available:
https://primis.phmsa.dot.gov/comm/FactSheets/FSPipeManufacturingProcess.htm.
[Accessed: 08-Aug-2015].
273
[87] A. J. Ujam, P. O. Ekere, and T. O. Chime, “Performance Evaluation of a Gas Turbine
Power Plant by the application of Compressor Off-Line and On-Line Water Washing
Techniques. (A Case Study of 450MW Sapele Power Station in Delta State, Nigeria),”
IOSR Journal of Engineering, vol. 3, no. 11, pp. 29–41, 2013.
[88] P. Sarkar, “OCS Online Clothing Study (2015) How to Calculate Efficiency of a
Production Batch or Line?” [Online]. Available:
http://www.onlineclothingstudy.com/2011/09/4-how-to-calculate-efficiency-of.html.
[Accessed: 07-Jul-2015].
[89] B. L. Song, W. K. Wong, J. T. Fan, and S. F. Chan, “A recursive operator allocation
approach for assembly line-balancing optimization problem with the consideration of
operator efficiency,” Computers and Industrial Engineering, vol. 51, no. 4, pp. 585–
608, 2006.
[90] W. M. Alobaidi, H. M. Al-Rizzo, and E. Sandgren, “NDT Applied to the Detection of
Defects in Oil and Gas Pipes: A Simulation-Based Study,” in ASME 2015
International Mechanical Engineering Congress and Exposition. American Society of
Mechanical Engineers, 2015, p. V02BT02A008.
[91] K. M. Tiitto and R. J. Pro, “Detection of Heat Treat Defects and Grinding Burns by
Measurement of Barkhausen Noise,” in Proceedings of the Second International
Symposium on Nondestructive Characterization of Materials, 1986, pp. 481–489.
[92] J. F. Bussière, J.-P. Monchalin, C. O. Ruud, and R. E. Green, Nondestructive
characterization of materials II, 1st ed. New York and London, 1987.
[93] Y. Sakai, N. Yusa, S. Ito, and H. Hashizume, “Numerical Analysis of Microwave
NDT Applied to Piping Inspection,” Materials Transactions, vol. 53, no. 4, pp. 627–
630, 2012.
[94] W. M. Alobaidi, Z. A. Nima, and E. Sandgren, “Localised surface plasmon-like
resonance generated by microwave electromagnetic waves in pipe defects,”
Nondestructive Testing and Evaluation, no. April, pp. 1–10, 2017.
274
[95] X. Shen and T. J. Cui, “Ultrathin plasmonic metamaterial for spoof localized surface
plasmons,” Laser and Photonics Reviews, vol. 8, no. 1, pp. 137–145, 2014.
[96] P. A. Huidobro, X. Shen, J. Cuerda, E. Moreno, L. Martin-Moreno, F. J. Garcia-Vidal,
T. J. Cui, and J. B. Pendry, “Magnetic localized surface plasmons,” Physical Review
X, vol. 4, no. 2, pp. 1–7, 2014.
[97] Z. Liao, X. Shen, B. C. Pan, J. Zhao, Y. Luo, and T. J. Cui, “Combined system for
efficient excitation and capture of LSP resonances and flexible control of SPP
transmissions,” ACS Photonics, vol. 2, no. 6, pp. 738–743, 2015.
[98] Z. Liao, Y. Luo, A. I. Fernández-Domínguez, X. Shen, S. a Maier, and T. J. Cui,
“High-order localized spoof surface plasmon resonances and experimental
verifications.,” Scientific reports, vol. 5, 2015.
[99] D. Bao, K. Z. Rajab, W. X. Jiang, Q. Cheng, Z. Liao, and T. J. Cui, “Experimental
demonstration of compact spoof localized surface plasmons,” Optics Letters, vol. 41,
no. 23, pp. 5418–5421, 2016.
[100] R. H. Ritchie, “Plasma Losses by Fast Electrons in Thin Films,” Physical Review, vol.
106, no. 5, pp. 874–881, 1957.
[101] Z. A. Nima, A. Biswas, I. S. Bayer, F. D. Hardcastle, D. Perry, A. Ghosh, E. Dervishi,
and A. S. Biris, “Applications of surface-enhanced Raman scattering in advanced biomedical technologies and diagnostics*,” Drug Metabolism Reviews, vol. 46, no. 2, pp.
155–175, 2014.
[102] Z. A. Nima, M. Mahmood, Y. Xu, T. Mustafa, F. Watanabe, D. A. Nedosekin, M. A.
Juratli, T. Fahmi, E. I. Galanzha, J. P. Nolan, A. G. Basnakian, V. P. Zharov, and A. S.
Biris, “Circulating tumor cell identification by functionalized silver-gold nanorods
with multicolor, super-enhanced SERS and photothermal resonances,” Scientific
reports, vol. 4, p. 4752(1-8), 2014.
[103] S. I. Bozhevolnyi, J. Erland, K. Leosson, P. M. W. Skovgaard, and J. M. Hvam,
“Waveguiding in Surface Plasmon Polariton Band Gap Structures,” Physical Review
275
Letters, vol. 86, no. 14, pp. 3008–3011, Apr. 2001.
[104] J.-C. Weeber, A. Dereux, C. Girard, J. R. Krenn, and J.-P. Goudonnet, “Plasmon
polaritons of metallic nanowires for controlling submicron propagation of light,”
Physical Review B, vol. 60, no. 12, pp. 9061–9068, Sep. 1999.
[105] E. Takeda, T. Nakamura, M. Fujii, S. Miura, and S. Hayashi, “Surface plasmon
polariton mediated photoluminescence from excitons in silicon nanocrystals,” Applied
Physics Letters, vol. 89, no. 10, p. 101907(1-3), 2006.
[106] W. L. Barnes, A. Dereux, and T. W. Ebbesen, “Surface plasmon subwavelength
optics,” Nature, vol. 424, no. August, pp. 824–830, 2003.
[107] T. W. Ebbesen, H. J. Lezec, H. F. Ghaemi, T. Thio, and P. A. Wolff, “Extraordinary
optical transmission through sub-wavelength hole arrays,” Nature, vol. 391, no. 6668,
pp. 667–669, Feb. 1998.
[108] J. B. Pendry, L. Martin-Moreno, and F. J. Garcia-Vidal, “Mimicking Surface
Plasmons with Structured Surfaces,” Science, vol. 305, no. 5685, pp. 847–848, 2004.
[109] F. J. Garcia-Vidal, L. Martin-Moreno, and J. B. Pendry, “Surfaces with holes in them:
new plasmonic metamaterials,” Journal of Optics A: Pure and Applied Optics, vol. 7,
no. 2, pp. S97–S101, 2005.
[110] A. P. Hibbins, B. R. Evans, and J. R. Sambles, “Experimental Verification of Designer
Surface Plasmons,” Science, vol. 308, no. 5722, pp. 670–672, 2005.
[111] U. S. Inan and A. S. Inan, Electromagnetic Waves. 2000.
[112] W. Alobaidi and E. Sandgren, “Detection of Defects in Spiral/Helical Pipes Using RF
Technology,” in 11th Pipeline Technology Conference, 2016, no. May, pp. 22–33.
[113] W. M. Alobaidi and E. Sandgren, “Classification of the Extent of Wall Thinning in
Pipes Based on Simulations in the Time and Frequency Domain,” in ASME 2016
Pressure Vessels and Piping Conference. American Society of Mechanical Engineers,
2016, p. V005T10A005.
276
[114] W. M. Alobaidi, E. Sandgren, and H. M. Al-Rizzo, “Waveform Pattern Recognition
Applied to Rapid Detection of Wall-Thinning in Pipes: A Simulation-Based Case
Study,” in 2016 11th International Pipeline Conference. American Society of
Mechanical Engineers (ASME), 2016, p. V003T04A038.
[115] Y. G. Ma and C. K. Ong, “Generation of surface-plasmon-polariton like resonance
mode on microwave metallic gratings,” New Journal of Physics, vol. 10, pp. 1–14,
2008.
[116] S. S. Akarca-Biyikli, I. Bulu, and E. Ozbay, “Resonant excitation of surface plasmons
in one-dimensional metallic grating structures at microwave frequencies,” Journal of
Optics A: Pure and Applied Optics, vol. 7, no. 2, pp. S159–S164, 2005.
[117] K. Sasaki, L. Liu, N. Yusa, and H. Hashizume, “Optimized microwave excitation
probe for general application in NDT of wall thinning in metal pipes of arbitrary
diameter,” NDT and E International, vol. 70, pp. 53–59, 2015.
[118] “History, Uses, and Physical Characteristics of Steel Pipe,” in AWWA MANUAL M11,
Tom Robbins, 2004, pp. 1–19.
[119] W. Schmidt, “Why Spiral Weld Pipe?,” World Petroleum, vol. 43, no. 4, pp. 28–31,
1972.
[120] M. R. Forouzan, S. M. Mirfalah Nasiri, A. Mokhtari, A. Heidari, and S. J. Golestaneh,
“Residual stress prediction in submerged arc welded spiral pipes,” Materials and
Design, vol. 33, no. 1, pp. 384–394, 2012.
[121] “CEPA (Canadian Energy Pipeline Association), Types of Pipelines, Canada’s
Underground Pipeline Network,” CEPA. [Online]. Available:
http://www.cepa.com/about-pipelines/types-of-pipelines. [Accessed: 02-Feb-2016].
[122] W. M. Alobaidi and E. Sandgren, “High-Efficiency Remote Measurement of Pipe
Defect Using RF/UT Technologies: A Theoretical Analysis Part One—Straight Beam
UT,” in ASME 2016 Pressure Vessels and Piping Conference. American Society of
Mechanical Engineers, 2016, p. V005T10A006.
277
[123] U. S. Inan and A. S. Inan, Electromagnetic Waves. Tom Robbins, 2000.
[124] “NDT Resource center, Ultrasonic Formula, NDT Education Resource Center, 20012014, The Collaboration for NDT Education, Iowa State University.” [Online].
Available: https://www.ndeed.org/GeneralResources/Formula/UTFormula/ultrasonicFormula.htm. [Accessed: 01Jan-2016].
[125] R. Halmshaw, Introduction to the Non-Destructive Testing of Welded Joints, Second
Edi. Cambridge, England: Woodhead Publishing Limited, 1996.
[126] W. Alobaidi, E. Alkuam, and E. Sandgren, “Two-stage Technical Protocol for
Locating, Validating, and Characterizing Manufacturing Defects Remotely using
Hybrid System RF/UT: Part Two—Angle Beam UT,” International Journal of
Pressure Vessels and Piping.
[127] B. Clark and R. Stanley, “Ultrason ic testing applications in the chemical industries,”
in Nondestructive Testing Handbook, Second Edi., 1991, pp. 703–721.
[128] B. Lancon, “Circumferential Scanning in Ultrasonic Inspection,” The NDT
Technician, The American Society foe Nondestructive Testing, vol. 11, no. 1, pp. 1–12,
2012.
[129] D. E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning.
Addison-Wesley, 1989.
[130] W. M. Alobaidi, C. E. Kintner, E. A. Alkuam, K. Sasaki, N. Yusa, H. Hashizume, and
E. Sandgren, “Experimental Evaluation of Novel Hybrid Microwave/Ultrasonic
Technique to Locate and Characterize Pipe Wall Thinning,” Journal of Pressure
Vessel Technology, vol. 140, no. 1, p. 11501, 2018.
[131] K. Sugawara, H. Hashizume, and S. Kitajima, “Development of NDT method using
electromagnetic waves,” JSAEM Studies in Applied Electromagnetic and Mechanics,
vol. 10, pp. 313–316, 2001.
[132] H. Hashizume, T. Shibata, and K. Yuki, “Crack detection method using
278
electromagnetic waves,” International Journal of Applied Electromagnetics and
Mechanics, vol. 20, no. 3, 4, pp. 171–178, 2004.
[133] R. E. Jones, F. Simonetti, M. J. S. Lowe, and I. P. Bradley, “Use of microwaves for
the detection of water as a cause of corrosion under insulation,” Journal of
Nondestructive Evaluation, vol. 31, no. 1, pp. 65–76, 2012.
[134] K. Sasaki, T. Katagiri, N. Yusa, and H. Hashizume, “Demonstration of the
Applicability of Nondestructive Microwave Testing to the Long-Range Inspection of
Inner-Surface Cracks in Tubes,” Materials Transactions, vol. 58, no. 4, pp. 692–696,
2017.
[135] K. Abbasi, N. H. Motlagh, M. R. Neamatollahi, and H. Hashizume, “Detection of
axial crack in the bend region of a pipe by high frequency electromagnetic waves,”
International Journal of Pressure Vessels and Piping, vol. 86, no. 11, pp. 764–768,
2009.
[136] J. K. Piotrowski, “Analysis of a Coaxial line - circular waveguide junction used for
measurement of materials,” in 13th International Conference on Microwaves, Radar
and Wireless Communications. MIKON - 2000, 2000, pp. 333–336.
[137] B. Gimeno and M. Guglielmi, “Multimode equivalent network representation for
planar junctions involving elliptical waveguides,” International Journal of RF and
Microwave Computer‐Aided Engineering, vol. 7, no. 2, pp. 180–194, 1997.
[138] M. Adous, P. Quéffélec, and L. Laguerre, “Coaxial/cylindrical transition line for
broadband permittivity measurement of civil engineering materials,” Measurement
Science and Technology, vol. 17, pp. 2241–2246, 2006.
[139] R. Stakenborghs and J. Little, “Microwave based NDE inspection of HDPE pipe
welds,” in 17th International Conference on Nuclear Engineering. American Society
of Mechanical Engineers, 2009.
[140] Y. Ju, L. Liu, and M. Ishikawa, “Quantitative evaluation of wall thinning of metal
pipes by microwaves,” Materials Science Forum, vol. 614, pp. 111–116, 2009.
279
[141] P. A. Doyle and C. M. Scala, “Crack Depth Measurement by Ultrasonics: A Review,”
Ultrasonics, vol. 16, no. 4, pp. 164–170, 1978.
[142] J. Xia, R. D. Miller, and C. B. Park, “Estimation of near-surface shear-wave velocity
by inversion of Rayleigh waves,” Geophysics, vol. 64, no. 3, pp. 691–700, 1999.
[143] L. Qiao, A. Esmaeily, and H. G. Melhem, “Signal Pattern Recognition for Damage
Diagnosis in Structures,” Computer-Aided Civil and Infrastructure Engineering, vol.
27, no. 9, pp. 699–710, 2012.
[144] D. Posenato, F. Lanata, D. Inaudi, and I. F. C. Smith, “Model-free data interpretation
for continuous monitoring of complex structures,” Advanced Engineering Informatics,
vol. 22, no. 1, pp. 135–144, 2008.
[145] I. Trendafilova, “Pattern recognition methods for damage diagnosis in structures from
vibration measurements,” Key Engineering Materials, vol. 204–205, pp. 85–94, 2001.
[146] J. S. Owen, “Monitoring the degradation of reinforced concrete beams under cyclic
loading,” Key Engineering Materials, vol. 245–246, pp. 307–314, 2003.
[147] N. Haritos and J. S. Owen, “The Use of Vibration Data for Damage Detection in
Bridges: A Comparison of System Identification and Pattern Recognition
Approaches,” Structural Health Monitoring, vol. 3, no. 2, pp. 141–163, 2004.
[148] W. M. Alobaidi, E. A. Alkuam, and E. Sandgren, “Development of an Optimized
Neural Network for the Detection of Pipe Defects Using a Microwave Signal,”
Journal of Pressure Vessel Technology, vol. 140, no. 3, 2018.
[149] Z. Chen, W. Zhu, Q. Di, and W. Wang, “Prediction of Burst Pressure of Pipes With
Geometric Eccentricity,” Journal of Pressure Vessel Technology, vol. 137, no. 6, p.
61201, 2015.
[150] Z. Chen, H. Ye, S. Yan, X. Shen, and Z. Jin, “The Failure Window Method and Its
Application in Pipeline Burst,” Journal of Pressure Vessel Technology, vol. 139, no. 5,
p. 51702, 2017.
280
[151] W. McCulloch and W. Pitts, “Representation of Events in Nerve Nets and Finite
Automata,” In: Automata Studies, ed. By C.E. Shannon and J. McCarthy, Annals of
Mathematical Studies, No. 34, Princeton University Press, 1956.
[152] P. Werbos, “Beyond Regression: New Tools for Prediction and Analysis in the
Behavioral Sciences,” PhD thesis, Harvard University, 1974.
[153] G. Simone, F. Morabito, R. Polikar, P. Ramuhalli, L. Udpa, and S. Udpa, “Feature
extraction techniques for ultrasonic signal classification,” International Journal of
Applied Electromagnetics and Mechanics, vol. 15, pp. 291–294, 2001.
[154] M.-D. Yang and T. C. Su, “Automated diagnosis of sewer pipe defects based on
machine learning approaches,” Expert Systems with Applications, vol. 35, no. 3, pp.
1327–1337, 2008.
[155] J. Chen, S. Huang, and W. Zhao, “Three-dimensional defect inversion from magnetic
flux leakage signals using iterative neural network,” IET Science, Measurement &
Technology, vol. 9, no. August 2014, pp. 418–426, 2015.
Документ
Категория
Без категории
Просмотров
0
Размер файла
23 823 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа