close

Вход

Забыли?

вход по аккаунту

?

s00170-017-0743-x

код для вставкиСкачать
Int J Adv Manuf Technol
DOI 10.1007/s00170-017-0743-x
ORIGINAL ARTICLE
In-situ and in-process monitoring of optical glass grinding process
based on image processing technique
Yong Jie Zhao 1,2 & Hao Nan Li 1 & Ke Chen Song 1 & Yun Hui Yan 1
Received: 30 October 2016 / Accepted: 26 June 2017
# Springer-Verlag London Ltd. 2017
Abstract Optical glass K9 is a key kind of materials in many
industries, and grinding process of it is usually employed as
the main rough machining technique. However, most previous
observations have been performed either ex-situ after grinding
operations or in-situ but by human raw eyes, therefore cannot
satisfy the new paradigm of Industry 4.0. In this paper, an insitu and in-process observation and evaluation methodology
of machined surfaces in K9 grinding is attempted to be proposed, which is based on image processing techniques and
therefore enables the automation of the observation and evaluation processes. Grinding trials proved that the method could
output accurate evaluation results, and the method performance is stable even for the ground K9 surface images with
wide ranges of characteristics. Because the method could insitu and in-process observe a fixed spot on the ground surfaces, more in-depth understandings of K9 grinding mechanism are gained. More importantly, the method could quantify
the ductile/brittle region area and area proportions, based on
which, the proposed method could be utilized not only to
automatically in-situ and in-process monitor the grinding performance but also to optimize or provide suggestions for the
future intelligent or smart manufacturing of K9.
Keywords Optical glass . Grinding . In-situ . In-process .
Image process technique
* Yong Jie Zhao
yjzhaomm@163.com
1
School of Mechanical Engineering and Automation, Northeastern
University, Shenyang, China
2
Department of Computer Science, Loughborough University,
Loughborough, UK
1 Introduction
With good optical performance and high strength, optical
glass K9 could be considered as one kind of the most widely
used optical materials in many industries including metrology,
electronics, and optics [1]. Grinding process is usually
employed as the main rough machining technique for optical
glass thanks to low cost and high machining efficiency [2].
However, due to the high hardness of K9, grinding process
easily results in brittle fractured surface and subsurface damage, degrading the machined surface quality and mechanical
strengths of products. To this end, many studies have been
reported on the grinding process of optical glass.
Bifano et al. [3, 4] predicted material removal modes for
brittle materials based on the analytical modeling of the energy required for plastic deformation and brittle crack propagation. Based on the model, the authors expected that: (i) when
the depth of cut was smaller than a certain threshold, brittle
materials like optical glass K9 would be grinded and removed
by ductile mode, which was similar to the grinding process of
metals; (ii) when the cut depth was bigger than the threshold,
brittle materials would be removed mainly by brittle mode,
where fractured cracks including median and lateral and radial
cracks of the classic fracture system [5, 6] would emerge upon
and beneath the ground surfaces. Zhao et al. [7, 8] observed
the ground K9 surface by using atomic force microscope
(AFM) and scanning electron microscope (SEM). The results
showed the damage-free ground surfaces with regular grinding marks when using a 6–12 μm grain sized diamond grinding wheel with the depth of cut of 3 μm, validating the conclusions given by Bifano et al. [3, 4].
Gu and Yao [9, 10] also researched material removal modes
for optical glass K9 by performing scratch tests. Three types
of surface cracks, namely lateral cracks, radial cracks, and
cracks in front of the moving cutter, were observed in the
Int J Adv Manuf Technol
experiments. Lateral cracks were anticipated to be the main
cracking type in the K9 grinding process because of large
damage size and low crack initiation load. Further study given
by Gu et al. [11] investigated material removal modes of optical glass K9 by grinding experiments. The observations
demonstrated that ductile and brittle modes usually not separately existed but emerged together with each other. Based on
the ground surface morphology after etched by hydrofluoric
acid, the authors categorized K9 grinding process into four
modes i.e., ductile, semi-ductile, semi-brittle, and brittle
modes. Similar observations of the ductile-brittle-mixed
ground surface were also presented in the systematic K9
grinding trials performed by Li et al. [2, 12], and the authors
suggested three modes (i.e., ductile, ductile-brittle, and brittle
modes), because from the authors’ point of view, it would be
hard to clearly distinguish the semi-ductile and semi-brittle
modes proposed in Ref. [11]. The threshold depth of cut in
K9 grinding process was reported to be around 0.2–0.3 μm in
both numerical and experimental studies given in Ref. [13].
More importantly, the authors claimed that [13] pure ductile
and brittle modes would be hard to be observed, and in most
cases, the ground surface would be of ductile-brittle-mixed
characteristics.
It could find from the above studies that, although both
ductile and brittle evidences have been found on the ground
K9 surface, most observations of the ground K9 surfaces have
been performed by human eyes. However, the subjectivity and
the speed of raw eyes cannot meet the required high degree of
automation in the new paradigm of Industry 4.0 [14].
In fact, the development trend of modern grinding technique also shows urgent needs for the automatic detection
methodology to facilitate development towards digital and
intelligent manufacturing [15], but only few preliminary and
foundational studies have been conducted so far.
Zhang et al. [16] proposed a new multi-class classification system that could identify the defects on the machined
surface after grinding and polishing. More effective feature
descriptors were employed in the proposed system.
Validation experiments showed that the system can categorize all the defects on the machined surface into 15
predefined classes with the correction rate of approximate
82%. Chen et al. [17] detected the grinding-induced damages in the ceramics grinding process by the proposed strategy including image preprocessing, defect image extraction,
and decision tress classifier. The proposed method was anticipated to be more powerful in the future applications like
the automatic quality evaluation of the ground ceramics surface. Besides, image-processing-based methods were
employed to measure the ground surface 2d [18], and 3d
roughnesses [19], evaluate the surface conditioning and
dressing of the worn grinding wheel [20], monitor the grinding wheel surface wear flats [21], and inspect the grain wear
and pull-out during the grinding process [22].
It could find from above that, nearly all the previous
methods have been employed to monitor and evaluate process
details after grinding operation, rather than in-process, which
might highly limit the automation degree of grinding
monitoring.
Considering the great industrial importance of optical glass
K9 and the limited existing monitoring techniques [12, 23], an
in-situ and in-process observation and evaluation methodology of machined surfaces in K9 grinding is attempted to be
proposed in this paper. To validate the method, grinding trials
are performed and the results are analyzed. Different K9
grinding mechanisms in ductile and brittle modes and empirical formulas of ductile and brittle region ratios based on the
in-situ and in-process observation are also provided to show
the advance of the proposed method, which is believed to be
very meaningful to facilitate the development of digital or
intelligent manufacturing in the paradigm of Industry 4.0.
2 Method description
2.1 Basic principle and flowchart
The basic principle of the proposed methodology is
distinguishing ductile and brittle regions based on the different
reflectivity of the topography produced in ductile and brittle
grinding modes, and then quantifying and statistically analyzing the detected ductile and brittle regions to evaluate the
grinding modes and mechanisms.
As seen in Fig. 1a, in brittle grinding mode, the ground
workpiece is fractured, leading to rough machined surfaces,
diffuse reflection therefore happens, forming the image in the
camera; in the ductile grinding mode (see Fig. 1b), on the
other hand, it has high probability that specular reflection happens due to the relative smooth machined surfaces, i.e., incident rays are reflected at the same angle. Therefore, it could
probably conclude that the luminance information of the capture images would be different for ductile and brittle surfaces
if the camera and light source are the same, and according to
these differences, ductile and brittle surfaces might probably
be recognized.
The flowchart of the proposed methodology is shown in
Fig. 2, including the following three steps: (i) segmentation of
the whole image into superpixels; (ii) mergence of the
superpixels that have similar visual characteristics into ductile
and brittle regions; and (iii) quantitative analysis and area
measurement of the merged ductile and brittle regions.
2.2 Image segmentation into superpixels
The key of the evaluation of grinding modes is to recognize
ductile and brittle regions on the ground surface. However,
ductile and brittle regions with various sizes are, in most cases,
Int J Adv Manuf Technol
Fig. 1 Schematics of the basic principle of the proposed methodology: using the different reflectivity of the topography produced in the a brittle and b
ductile grinding modes
mixed with each other in captured images (see captured image
in Fig. 2); therefore, it would be tricky to accurately distinguish ductile and brittle regions by one step.
To solve this, the first step of the proposed method is to
segment the captured images into small regions by using
Entropy Rate Superpixel Segmentation (ERSS) method [24]
considering its superiority in high-precision adherence of zigzag boundaries and low computation efforts thanks to the
entropy-rate-based objective functions and the proposed
greedy optimal solution [24]. The generated small regions
are termed as superpixels (see the superpixel example in step
(i) in Fig. 2), within which it has similar features in terms of
intensity, texture, and color.
skewed shape (see Fig. 3b), which is utilized as one principle
in the superpixel mergence.
To automatically determine whether the intensity histogram distribution of a certain superpixel is left-skewed or
right-skewed, two parameters are defined here, which are the
following:
2.3 Similar superpixel mergence
M¼
By step (i) (Section 2.2), a continuous ductile/brittle region
might probably be segmented into several superpixels due to
the high distinguishability of ERSS method (see the
superpixels “a, b, c” in step (i) in Fig. 2). Therefore, the second
step of the proposed method is to merge superpixels with
similar visual characteristics. Here, superpixel similarities
are quantified by intensity histogram distributions, because
they can, to a large extent, present the probability density
(one kind of statistic characteristics) of the intensity level of
the superpixels in a simple and effective way [26], especially
when region luminance is the main difference between the
diffuse reflection of brittle surfaces and the specular reflection
of ductile surfaces (see Fig. 1).
In order to summarize the characteristics of ductile- and
brittle region superpixel histograms, the statistical analysis of
intensity histogram distributions is performed as seen in Fig.
3. It could find that the typical histogram distribution of brittle
region superpixels presents a left-skewed shape (see Fig. 3a),
while the one of ductile-region superpixels shows a right-
where p represents the intensity of the histogram.
&
&
Middle superpixel intensity M, the average value of the
maximum and minimum intensities of the histogram,
which could represent the median value of the histogram
intensity and could be calculated by Eq. (1);
arg½maxjhistðpÞj þ arg½minjhistjðpÞ
2
ð1Þ
Average superpixel intensity A, the arithmetic mean value
of the histogram intensity defined as Eq. (2), which is
close to the concentrated intensities, because A is the average value considering the intensities of all the pixels
within a certain superpixel.
A¼
p1 ⋅histðp1 Þ þ p2 ⋅histðp2 Þ þ ⋯ þ pn ⋅histðpn Þ
histðp1 Þ þ histðp2 Þ þ ⋯ þ histðpn Þ
ð2Þ
where n is the number of histogram intensity levels.
2.3.1 Superpixel mergence principle (i)
As shown as principle (i) in Fig. 4, based on the
definitions of A and M, superpixels satisfying the
Int J Adv Manuf Technol
Fig. 2 Flowchart of the proposed methodology including three steps: (i)
segmentation of the whole image into superpixels; (ii) mergence of the
superpixels having similar visual characteristics into ductile and brittle
regions; and (iii) quantitative analysis and area measurement of the
merged ductile and brittle regions
histogram relationship of A < M are left-skewed (see
Fig. 4a) therefore should be merged into brittle region,
while superpixels satisfying A > M are right-skewed
(see Fig. 4b) which should be merged into ductile
region.
2.3.2 Superpixel mergence principle (ii)
The above mergence principle (i) is effective for most
superpixels based on large numbers of experiment results,
but for the superpixels satisfying the relationship jA−M j≤ 5
Int J Adv Manuf Technol
Fig. 3 Typical histogram examples of superpixels belonging to a brittle region (showing left-skewed shape) and b ductile region (showing right-skewed
shape)
(see Fig. 4c), the mergence errors are largely increasing, which
might because when A ≈ M, certain extreme intensity values
influence the evaluation of the general features of the
superpixel.
To fix this, the mergence principle (ii) is suggested
(see principle (ii) in Fig. 4), which is quantifying similarities by using the histogram intersection distance [27]
rather than the comparison between A and M used in the
mergence principle (i) (see Section 2.3.1). According to
Ref. [27], the intersection distance D could present the
cumulative similarities between two different histograms
by considering the full-ranged intensity level. In this paper, the intersection distance D(H i , H b ) between the
median distribution Hi and the brittle region histogram
distribution Hb could be expressed as
n
∑ min½hist i ðpk Þ; hist b ðpk Þ
DðH i ; H b Þ ¼
k¼1
n
∑ hist i ðpk Þ
ð3Þ
k¼1
Similarly, the intersection distance D(Hi, Hd) between the
median distribution Hi and the ductile region histogram distribution Hd could be expressed as
n
∑ min½histi ðpk Þ; histd ðpk Þ
k¼1
ð4Þ
DðH i ; H d Þ ¼
n
∑ histi ðpk Þ
k¼1
Fig. 4 Schematics of the similar superpixel mergence principles if the superpixel histograms satisfying a A < M, b A > M, and c A ≈ M
Int J Adv Manuf Technol
Table 1
Material properties of the employed optical glass K9
Material
Young’s
modulus
(GPa)
Poisson’s
ratio
Hardness
(GPa)
Yield stress
(MPa)
(20 °C)
K9
82
0.203
7.7
1900
where Hb , d are the average intensity histogram of brittle/
ductile region superpixels and could be expressed by
As seen in step (iii) in Fig. 2, the actual length of 2 μm is
expressed by 65 pixels; therefore, the actual area containing N
pixels could be calculated by:
S ¼ ð2 N Þ=65
The proportions of the brittle region ζb and ductile region ζd
could be quantified as
ζb ¼
n
∑ hist j ðpk Þ
H b;d ðpk Þ ¼
j¼1
ð5Þ
N
where histj is the histogram of the jth superpixel merged in
the brittle/ductile region, pk is the intensity value of the
histogram, and N is the superpixel number in the brittle/
ductile region.
Hence, as seen in principle (ii) in Fig. 4, the superpixel
with median histogram would be merged into the brittle
region if D(H i , H b ) > D(H i , H d ), while the superpixel
would be merged into the ductile region if D(Hi, Hb) <
D(Hi, Hd).
2.4 Area measurement of merged brittle and ductile
regions
After the availability of merged ductile and brittle regions,
area measurement is conducted to obtain the total ductile
and brittle area as well as the area proportions so that
automatic quantification of grinding modes could be
achieved.
Since one pixel could be considered as the basic unit area in
the captured image, the region area could be measured by the
region pixel number. Binary maps are used to reduce computation efforts, in which the intensities of the pixels in brittle
regions are set to be zeros (in black) (see schematics in step
(iii) in Fig. 2), while those in ductile regions are set to be ones
(in white) (see schematics in step (iii) in Fig. 2). Hence, the
brittle region area is equal to the sum pixel number of black
region, while the ductile region area is the sum pixel number
of white region.
Table 2
Grinding
wheel
Dresser
wheel
a
ð6Þ
Sb
and ζ d ¼ 100%−ζ b
S all
ð7Þ
where Sb and Sall respectively represent the actual area of
brittle regions and the whole image area.
3 Experimental methodology for method validation
To validate the feasibility and the accuracy of the proposed
method, grinding trials of optical glass K9 are performed, and
the proposed observation system is employed.
3.1 Workpiece and grinding wheels
Optical glass K9 blocks (Glass Dynamics Ltd.) with the
dimension of 50 mm (length) × 15 mm (width) × 10 mm
(height) are employed as the specimens in the trials and
the corresponding properties are presented in Table 1.
Before the trials, workpiece surfaces are carefully
polished by using the abrasive slurry with the abrasive
sizes of #220, #400, #600, #1000, and #1200 until the
surface roughness Ra reaches the value of 5 μm, so that
the influence of any potential scratches induced by previous manufacturing processes could be minimized.
All the grinding trials are conducted by diamond grinding wheels (JR Diamond Ltd.) considering its wide applications in the modern grinding technique [1], and the
wheel details are presented in Table 2. Dressing operations
by a dresser wheel (JR Diamond Ltd., see Table 2) are
performed prior to the trials to restore the cutting ability
of the grinding wheel. The dressing depth is 20 μm (by 20
times), and the dressing ratio is −0.7.
Details of the diamond grinding wheel used in the trials and of the dresser wheel
Shape
Size (mm)
Abrasive
type
Abrasive
size
Hardness
Structure
Bond
type
Diamond
concentration (%)a
Plate
Ø115 × 10
Diamond
150
M
6
Vitrified
100
Plate
Ø50 × 1.5
Diamond
60
W
5
Metal
150
Based on Ref. [25], diamond concentration refers to the weight ratio between the abrasive grains and the bond mixture during the wheel fabrication
Int J Adv Manuf Technol
Fig. 5 The experiment setup and image acquisition system utilized in this study
3.2 Experimental setup
As seen in Fig. 5, grinding trials are performed on a multifunction surface grinding machine tool (Jones & Shipman
Ltd). The specimens are fixed on the jig, which is then
mounted on the worktable of the machine tool. During the
trials, no cutting fluids are employed, because they might
probably interfere with the view of the image acquisition
system.
Given that the aim of the validation experiments is not to
optimize grinding parameters but to verify the feasibility and
accuracy of the proposed in-process and in-situ observation
system, different depths of cut are employed in each trial while
the wheel and workpiece speeds are kept constant, because
depths of cut play the dominant role of material removal
modes during the brittle material grinding process [12, 24].
To obtain ductile, ductile-brittle-mixed, and brittle ground surfaces, relatively wide ranges of cut depths are used as seen in
Table 3.
3.3 Image acquisition system
As seen in Fig. 5, the image acquisition system is consisted of
a high-speed camera (Y4-S2, IDT Vision Ltd.) equipped with
a high-performance telephoto lens (Tamron 180 mm f3.5 macro, Nikon Ltd.) and a high-frequency lighting system
Table 3
The grinding parameters employed in the trials
Value
Depth of cut ap (μm)
Wheel speed vs (m/s)
Workpiece speed vw (m/min)
0.1, 0.2, 0.5, 1.0, 1.2, 1.5, 2.0,
3.0, 4.0, 5.5, 10, 15, 20, 30
16.5
1.5
(Constellation 120E, IDT Vision Ltd.). The utilized capture
frame rate and lighting frequency is 50 fps with the exposure
time of 1 ms (2000 ISO color). Before the experiments, the
camera view focus is manually kept on a fixed spot on the
ground workpiece surface so that the surface morphologies
could be clearly captured. During the trials, the camera is
not moved, and the images captured during the workpiece in
the camera view are used as the input images for the proposed
methodology. All the captured images are in JPG format with
24-bit pixel depth, 19.68 × 13.68 μm pixel size, and 640 × 480
image resolution.
Here, the detailed explanation of the camera configuration
needs to be given. Ideally, the camera should be allocated on
the top of the workpiece surface (see camera position 2 in Fig.
6a) so that the ground surface could be perpendicular to the
capture direction. However, in industrial grinding process, the
grinding wheel is commonly protected by a wheel cover to
avoid the splashed chip sparks (see Fig. 5). Therefore, as illustrated in Fig. 6a, the wheel cover would probably largely
reduce the observable area of the workpiece surface (see L1 in
Fig. 6a) if the camera is arranged at the position 2. Moreover,
more complex preparation work should also be performed to
establish a special camera support frame.
In the contrary, the camera position 1 enables more effective and feasible image capture. The observable area is obviously larger than that from the position 2 (L2 > L1), and the
camera could be simply supported by a tripod. More importantly, the aim of this paper is to capture the brittle and ductile
region ratios on the ground surface, and these ratios would not
be changed when different capture directions are used. As
seen in Fig. 6b, the theoretical brittle and ductile region ratios
are separately a ∗ b/(L3 ∗ L4) and c ∗ d/(L3 ∗ L4). When the
camera position 1 is used, these ratios are separately a ∗
sinβ ∗ b/(L3 ∗ sin β ∗ L4) and c ∗ sinβ ∗ d/(L3 ∗ sin β ∗ L4),
which are the same as the theoretical ones. Therefore, the
camera position 1 is employed in this study, where the
Int J Adv Manuf Technol
Fig. 6 a Illustration of the camera configuration and b illustration of the independent relationship between the ductile and brittle region ratios and the
capture directions
horizontal and vertical distances between the camera lens and
the ground workpiece surface are separately 1.1 and 0.4 m,
and the capture direction is in a 20 deg. tilted angle relative to
the horizontal direction. The captured images could be seen in
the first column of Fig. 9.
As seen in Fig. 7, ηarea and ηboundary are separately defined
as the ratios of the ductile/brittle region area and boundaries
that are correctly detected by the proposed method (separately
denoted as Sproposed , Bproposed) to the ones that are obtained by
raw eyes (separately denoted as Sraw , Braw), i.e.,
ηarea ¼
3.4 Methodology for model validation
To quantify the performance of the proposed method, quantitative validation is conducted from the following three aspects:
(i). the evaluation accuracy of the ductile/brittle region area
(denoted as ηarea);
(ii). the evaluation accuracy of ductile/brittle region boundaries (denoted as ηboundary);
(iii). the evaluation accuracy of the workpiece removal
modes (or grinding modes) (denoted as ηmode).
S proposed ∩S raw
100%
S raw
ηboundary ¼
Bproposed ∩Braw
100%
Braw
ð8Þ
ð9Þ
ηmode is defined as the ratio of the number of the captured
images in which the grinding modes are correctly detected by
the proposed method (denoted as Ncorrect) to the total image
number that is utilized in the trials (denoted as Nall), i.e.,
ηmode ¼
N correct
100%
N all
Fig. 7 Schematics of the evaluation accuracy of the ductile/brittle region area ηarea and boundaries ηboundary
ð10Þ
Int J Adv Manuf Technol
Fig. 8 Schematics of the difference between the parameter ηarea and ηboundary, and the difference between the parameter ηarea|ductile and ηarea|brittle
Please note that as seen in Fig. 8, the parameters ηarea and
ηboundary are different, indicating the evaluation accuracy of
the proposed methodology in terms of region area and boundaries. ηarea|brittle might be close to 100%, while ηboundary might
be only around 43.6% (see example in Fig. 8).
Another note is that, the parameter ηarea might be different
when separately discussing ductile and brittle regions. For
example, as seen in Fig. 8, ηarea for ductile regions is smaller
than 100%, while ηarea for brittle regions is equal to 100%.
Therefore, in the following validation, the term ηarea is specified as ηarea|ductile and ηarea|brittle, separately referring to area
evaluation accuracy for ductile and brittle regions.
4 Results and validations
To validate the feasibility and accuracy of the proposed methodology, large numbers of experimental results are compared
with the standard results obtained by raw eyes.
4.1 Visual comparisons
Visual results are presented in Fig. 9 where it could be generally observed that the ductile and brittle regions could be
properly recognized, even though these two regions are intermingle with each other via zigzag boundaries. The detection
results are close to the results detected by raw eyes. However,
by using the proposed method, the detailed edges are preserved (see label “a” in Fig. 9b, c), and small-sized ductile/
brittle regions could also be recognized (see label “b” in Fig.
9b, c).
It should note that false detection cases are found by using
the proposed method e.g., label “c” in Fig. 9b, c should be
brittle regions, but they are detected as ductile region.
However, it should also note that small-sized “unclear” regions (e.g., label “c” in Fig. 9b, c) are also tricky to be detected
even by raw eyes, and the number of false detection cases is
also limited. Considering the advances including automation,
high speed, relatively good detection accuracy, and nonhuman cost, the proposed methodology is anticipated to be
promising to in-situ and in-process observe K9 grinding
process.
4.2 Quantitative comparisons
To further validate the method, quantitative comparisons of
ductile and brittle regions detected by raw eyes and the proposed methodology are performed in this section from the
following three aspects: (i) evaluation accuracy of ductile/
brittle region area, (ii) evaluation accuracy of ductile/brittle
region boundaries, and (iii) evaluation accuracy of the workpiece removal modes (more explanations could be found in
Section 3.4).
4.2.1 Evaluation accuracy of ductile/brittle region area
As seen in Fig. 10, the evaluation accuracy of the ductile/
brittle region area in the images of Fig. 9 proves the high
detection precision of the proposed methodology. The minimum accuracy is more than 84.9% for ductile regions and
86.2% for brittle regions, while the maximum even reaches
95.1% for ductile regions and 94.8% for brittle regions, although there are some slight accuracy variations for different
images. The high accuracy may because the superpixel sizes
are small; therefore, it would not influence the evaluation of
ductile/brittle region area, even though certain superpixels are
merged into wrong regions.
It could also find from the standard error of the average
accuracy (2.7% for ductile region and 3.1% for brittle region)
in Fig. 10 that the proposed methodology performance is stable and could be applicable to the ground surfaces with various characteristics, proving the feasibility of the proposed
intensity-histogram-based superpixel mergence principles
(Section 2.3).
Int J Adv Manuf Technol
Fig. 9 The validation comparisons of the ductile and brittle regions detected by raw eyes and the proposed methodology: a captured images, b edge
results detected by raw eyes, c edge, and d region results detected by the proposed methodology
Int J Adv Manuf Technol
Fig. 10 Evaluation accuracy of
ductile/brittle region area of the
proposed methodology for
images with different
characteristics
4.2.2 Evaluation accuracy of ductile/brittle region boundaries
The results of evaluation accuracy of ductile/brittle region boundaries are shown in Fig. 11. Although the
boundary accuracy results are not as high as the ones
of region area in Fig. 10, it is still considered to be
encouraging: the maximum accuracy could achieve
86.7% and the minimum is 78.2% (see Fig. 11). The
reason why boundary accuracy is lower may because
the proposed method is able to recognize zigzag and
detailed boundaries, which, on the other hand, are difficult to be properly observed by raw eyes. It could be
also observed that the standard error of the average
accuracy is 2.3%, which indicates the high stability
of the proposed methodology.
ground surface ex-situ observed in Refs. [12] and
[10, 11], two application examples of the proposed
method are validated as seen in Fig. 12. It could find
that the material removal modes in Refs. [12] and [10,
11] could be correctly determined without any human
intervention, even though the possible grinding modes
defined in Refs. [12] and [10, 11] are different from
each other.
In fact, the availability of the quantified ductile and
brittle region proportions is more important than the
defined ranges of each grinding modes, because the defined ranges might be varied by different scholars’ definitions. The proposed method therefore provides a new
way to quantitatively study K9 grinding mechanism,
rather than qualitatively like previous studies did.
4.2.3 Evaluation accuracy of the workpiece removal modes
Because the proposed method is able to automatically
obtain the brittle/ductile region proportions, the K9
grinding modes could probably be in-situ and inprocess gained, if the region proportion ranges for different grinding modes are known. Based on the K9
Fig. 11 Evaluation accuracy of
ductile/brittle region boundaries
of the proposed methodology for
images with different
characteristics
5 Further discussions—examples of the method
applications
To show the advance, two examples of the proposed method
applications are further discussed in this section.
Int J Adv Manuf Technol
Fig. 12 Two application
examples of the proposed method
based on the K9 ground surface
ex-situ observed in Refs. [12]
(example (i)) and [10, 11]
(example (ii))
5.1 K9 grinding mechanism
Although the grinding mechanism of brittle materials has
been experimentally studied before (e.g., in Refs. [7, 11,
12]), nearly all the previous observations have been performed ex-situ and after grinding operations. Therefore,
it would be difficult to observe one same spot on the
ground surface at each observation, which might probably limit the credibility of the obtained understandings of
grinding mechanisms. The proposed methodology, on the
other hand, is in-situ and in-process, which enables the
automatic observations of specific spots on the ground
surfaces even right after the wheel passes the workpiece
surface. Given that cut depths have the largest impacts
on the brittle material grinding modes [12], different
levels of cut depths are employed in the following discussions, and wheel and workpiece speeds are kept
constant.
Based on the proposed in-situ and in-process observation
of a fixed spot on the ground K9 surface, some new findings
could be summarized as follows:
& From the macro-view, the grinding mechanism of optical glass K9 is consistent with the one of classic brittle
materials, i.e., a slight variance of cut depths might
completely change the material removal modes, and
ground surface topography and morphology:
When the cut depth is smaller than the threshold
ductile/brittle transition value of around 0.4 μm [12],
smooth and high-quality surfaces are produced without
any brittle cracks (see Fig. 13a). Shallow grooves and
Int J Adv Manuf Technol
Fig. 13. In-situ and in-process observation of the ground K9 surface by the proposed methodology when the wheel and workpiece speed are kept
constant (vs = 16.5 m/s, vw = 1.5 m/min) and different cut depths are employed (ap = 0.1, 0.5, 1, 1.5, 3, 5.5 μm)
&
regular pile-ups resulted from grain-workpiece plowing
are presented along the grinding direction, which is
similar to the ground metal surface.
When the cut depth is beyond the threshold value, smallscale cracks start to emerge (see Fig. 13b). These small
cracks grow to be large-scale (or cloud-like) cracks when
the depth of cuts is increased, resulting in large ruptured
fragments and rough ground surfaces (see Fig. 13c–f).
From the micro-view, for a same spot, the grinding mode
might also be possible to be changed from brittle mode to
ductile mode even though the cut depths are increased (see
spot A in Fig. 13b, c). This means that increased cut
Fig. 14 Empirical formulas
between ductile region
proportions ζd and depths of cut
ap in K9 grinding process
depths at macro-scale might increase the total brittle region area, for a specific local spot, however, machined
spot morphology is determined by grit penetration depths
at micro-scale, which are related with both micro-scale
grain protrusions from the wheel surface and macroscale cut depths.
To the best knowledge of the authors, this conclusion
has not been reported before, which might because previous studies were based on ex-situ observations of the
ground surfaces; therefore, it would be hard to observe a
fixed spot. This also indicates the importance of the in-situ
and in-process observation of brittle material grinding
Int J Adv Manuf Technol
process, especially when grinding mechanism is anticipated to be gained based on these observations.
&
It could also observe from Fig. 13 that the brittle area
proportion ζb varies within a wide range from 0.3 to
81.1% when the cut depths are increased from 0.1 to
5.5 μm. This implies that by choosing appropriate
grinding parameters, ground surface morphology/
quality theoretically could probably be properly controlled to achieve the balance between high machining
efficiency and good surface quality (more details will
be provided in Section 5.2).
In comparison with the artificial observations in previous studies [28, 29], the proposed study is also the
first time that brittle and ductile region area on the
ground surface is quantified, and grinding removal
modes are automatically determined, which could
probably reduce the human cost, shorten the evaluation
time, and improve the evaluation accuracy.
specific material removal rates in Fig. 14 when cut depths
are increased, it could find that, the cut depth of 2.591 μm
might be the reasonable value, which could simultaneously
achieve the relative smooth surface (with ductile region proportion of 40.13%) and high machining efficiency (with specific material removal rate of 3.89 mm3/(mm × s)).
6 Conclusions
In this paper, an observation and evaluation methodology of
machined surfaces in K9 grinding process is attempted to be
proposed. Based on the study, some conclusions might be
drawn as follows:
&
Unlike any previous methods that relied on raw eyes, the
proposed methodology is based on image process techniques; therefore, it enables the automation of the observation and evaluation in K9 grinding process, which not
only largely reduces human cost and shortens evaluation
time but also probably facilitate the development of smart
or intelligent manufacturing of optical glass;
Grinding trials proved the proposed method has not only
high monitoring accuracy (more than 84.9% for regions
and more than 78.2% for edges) but also highperformance stability (standard error of less than 3.1%
for regions and less than 2.3% for edges);
Another advantage of the proposed method is it could insitu and in-process observe a fixed spot on the ground
surfaces, which allows more in-depth understandings of
K9 grinding mechanism (see Section 5.1);
The empirical formulas of ground surface quality based on
the in-situ and in-process observation are also provided to
show the advance of the proposed method, which is believed to be very meaningful to facilitate the development
of digital or intelligent manufacturing in the paradigm of
Industry 4.0;
Due to the high compatibility, the proposed methodology
could also be transferable to other machining processes.
&
5.2 Empirical formulas in K9 grinding
Because the proposed methodology could quantify the brittle/
ductile regions on the ground surfaces, empirical formulas in
K9 grinding are established in this section, which are expected
to be meaningful to guide industrial smart manufacture of
optical glass K9 [30, 31].
Based on the observation of the scatter points obtained by
the proposed method (see blue circle symbols in Fig. 14) and
trials of many function forms, the basic form of formulas is set
to be the power function, i.e., ζd(ap) = A ⋅ (ap − B)C. Linear
least square method [32, 33] is employed to determine the
coefficients A, B, and C, and the obtained fitting empirical
formulas are expressed as:
1:562
ζ d ¼ 17:198 ap −1:29eð−5Þ
1:562 ζ d unit : %; ap unit : μm
ζ b ¼ 100−17:198 ap −1:29eð−5Þ
&
&
&
ð11Þ
where the fitting errors are presented in Fig. 14. To validate the
obtained formulas, more grinding trials are conducted, and the
captured data are plotted in the figure (see red cross symbols in
Fig. 14). It could find that experimental data are consistent
with the empirical formulas; therefore, the formulas could
probably be used to predict grinding surfaces based on the
employed grinding parameters. To the best knowledge of the
authors, this is also the first empirical formulas that provide
the relationship between ductile/brittle region proportions ζd , b
and depths of cut ap in K9 grinding process.
The obtained formulas might also be useful in the optimization of K9 grinding process. For example, by plotting
Acknowledgements The authors acknowledge the support from the
National Natural Science Foundation of China in undertaking this research work under grant reference number 51374063 and the
Fundamental Research Funds for the Central Universities, the numbers
N140303008, N141008001 and N150308001.
References
1.
2.
Malkin S, Hwang T (1996) Grinding mechanisms for ceramics.
CIRP Ann Manuf Technol 45(2):569–580
Li HN, Yu TB, Da Zhu L, Wang WS (2016) Evaluation of grindinginduced subsurface damage in optical glass BK7. J Mater Process
Technol 229:785–794
Int J Adv Manuf Technol
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Bifano TG, Dow TA, Scattergood RO (1989) Ductile-regime grinding of brittle materials: experimental results and the development of
a model. In: 32nd Annual Technical Symposium, pp 108–115
Bifano TG, Dow T, Scattergood R (1991) Ductile-regime grinding:
a new technology for machining brittle materials. J Eng Ind 113(2):
184–189
Lawn B, Wilshaw R (1975) Indentation fracture: principles and
applications. J Mater Sci 10(6):1049–1081
Huerta M, Malkin S (1976) Grinding of glass: the mechanics of the
process. J Eng Ind 98(2):459–467
Zhao Q, Liang Y, Stephenson D, Corbett J (2007) Surface and
subsurface integrity in diamond grinding of optical glasses on
Tetraform ‘C’. Int J Mach Tools Manuf 47(14):2091–2097
Zhao Q, Zhao L, Guo B, Stephensin D, Corbett J (2012)
Deformation analysis of micro/nano indentation and diamond
grinding on optical glasses. Chin J Mech Eng 25(3):411–418
Gu W, Yao Z (2011) Evaluation of surface cracking in micron and
sub-micron scale scratch tests for optical glass BK7. J Mech Sci
Technol 25(5):1167–1174
Yao Z, Gu W, Li K (2012) Relationship between surface roughness
and subsurface crack depth during grinding of optical glass BK7. J
Mater Process Technol 212(4):969–976
Gu W, Yao Z, Li H (2011) Investigation of grinding modes in
horizontal surface grinding of optical glass BK7. J Mater Process
Technol 211(10):1629–1636
Yu T, Li H, Wang W (2016) Experimental investigation on grinding
characteristics of optical glass BK7: with special emphasis on the
effects of machining parameters. Int J Adv Manuf Technol 82(5–8):
1405–1419
Guo X, Wei Y, Jin Z, Guo D, Maosen W (2013) A numerical model
for optical glass cutting based on SPH method. Int J Adv Manuf
Technol 68(5–8):1277–1283
Lindeberg T (1998) Feature detection with automatic scale selection. Int J Comput Vis 30(2):79–116
Wegener K, Hoffmeister H, Karpuschewski B, Kuster F, Hahmann
W, Rabiey M (2011) Conditioning and monitoring of grinding
wheels. CIRP Ann Manuf Technol 60(2):757–777
Zhang X, Krewet C, Kuhlenkötter B (2006) Automatic classification of defects on the product surface in grinding and polishing. Int J
Mach Tools Manuf 46(1):59–69
Chen S, Lin B, Han X, Liang X (2013) Automated inspection of
engineering ceramic grinding surface damage based on image recognition. Int J Adv Manuf Technol 66(1–4):431–443
Dhanasekar B, Mohan NK, Bhaduri B, Ramamoorthy B (2008)
Evaluation of surface roughness based on monochromatic speckle
correlation using image processing. Precis Eng 32(3):196–206
19.
Hu Z, Zhu L, Teng J, Ma X, Shi X (2009) Evaluation of threedimensional surface roughness parameters based on digital image
processing. Int J Adv Manuf Technol 40(3–4):342–348
20. Lachance S, Bauer R, Warkentin A (2004) Application of region
growing method to evaluate the surface condition of grinding
wheels. Int J Mach Tools Manuf 44(7):823–829
21. LaChance S, Warkentin A, Bauer R (2003) Development of an
automated system for measuring grinding wheel wear flats. J
Manuf Syst 22(2):130
22. Feng Z, Chen X (2007) Image processing of the grinding wheel
surface. Int J Adv Manuf Technol 32(5–6):452–458
23. Wu C, Li B, Yang J, Liang SY (2016) Prediction of grinding force
for brittle materials considering co-existing of ductility and brittleness. Int J Adv Manuf Technol 87(5):1967–1975
24. Liu M, Tuzel O, Ramalingam S, Chellappa R (2011) Entropy rate
superpixel segmentation. In: IEEE Conference on Computer Vision
and Pattern Recognition (CVPR) pp 2097–2104
25. Malkin S, Guo C (2008) Grinding technology: theory and application of machining with abrasive. McGraw-Hill, USA
26. Gonzalez RC, Richard E (2002) Digital image processing. Prentice
Hall Press
27. Androutsos D, Plataniotis K, Venetsanopoulos A (1999) A novel
vector-based approach to color image retrieval using a vector
angular-based distance measure. Comput Vis Image Underst
75(1):46–58
28. Li HN, Yu TB, Da Zhu L, Wang WS (2017) Analytical modeling of
ground surface topography in monocrystalline silicon grinding considering the ductile-regime effect. Arch Civ Mech Eng 17(4):880–
893
29. Li HN, Yu TB, Da Zhu L, Wang WS (2017) Analytical modeling of
grinding-induced subsurface damage in monocrystalline silicon.
Mater Des 130(15):250–262
30. Li HN, Yu TB, Zhu LD, Wang WS (2015) Analysis of loads on
grinding wheel binder in grinding process: insights from
discontinuum-hypothesis-based grinding simulation. Int J Adv
Manuf Technol 78(9–12):1943–1960
31. Li HN, Yu TB, Zhu LD, Wang WS (2015) Modeling and simulation
of grinding wheel by discrete element method and experimental
validation. Int J Adv Manuf Technol 81(9–12):1921–1938
32. Marquardt D (1963) An algorithm for least-squares estimation of
nonlinear parameters. J Soc Ind Appl Math 11(2):431–441
33. Li HN, Yu TB, Wang ZX, Da Zhu L, Wang WS (2016) Detailed
modeling of cutting forces in grinding process considering variable
stages of grain-workpiece micro interactions. Int J Mech Sci 126:
319–339
Документ
Категория
Без категории
Просмотров
0
Размер файла
16 613 Кб
Теги
0743, 017, s00170
1/--страниц
Пожаловаться на содержимое документа