close

Вход

Забыли?

вход по аккаунту

?

Development of a grape harvester yield monitoring system forapplication in Precision Viticulture systems

код для вставкиСкачать
DEVELOPMENT OF A GRAPE HARVESTER YIELD MONITORING SYSTEM FOR
APPLICATION IN PRECISION VITICULTURE SYSTEMS
A Thesis
Presented to
The School of Engineering
of
The University of Guelph
by
ALBERT JAMES BROOKS
In partial fulfillment of the requirements
For the degree of
Master of Applied Science
September, 2009
© Albert Brooks, 2009
Bibliothèque et
Library and Archives
?F?
Canada
Archives Canada
Published Heritage
Direction du
Branch
Patrimoine de l'édition
395 Wellington Street
395, rue Wellington
OttawaONK1A0N4
Ottawa ON K1A0N4
Canada
Canada
Your file Votre référence
ISBN: 978-0-494-68599-0
Our file Notre référence
ISBN: 978-0-494-68599-0
NOTICE:
AVIS:
The author has granted a non-
L'auteur a accordé une licence non exclusive
exclusive license allowing Library and
Archives Canada to reproduce,
publish, archive, preserve, conserve,
communicate to the public by
telecommunication or on the Internet,
loan, distribute and sell theses
worldwide, for commercial or non-
commercial purposes, in microform,
paper, electronic and/or any other
permettant à la Bibliothèque et Archives
Canada de reproduire, publier, archiver,
sauvegarder, conserver, transmettre au public
par télécommunication ou par l'Internet, prêter,
distribuer et vendre des thèses partout dans le
monde, à des fins commerciales ou autres, sur
support microforme, papier, électronique et/ou
autres formats.
formats.
L'auteur conserve la propriété du droit d'auteur
The author retains copyright
ownership and moral rights in this
thesis. Neither the thesis nor
substantial extracts from it may be
et des droits moraux qui protège cette thèse. Ni
la thèse ni des extraits substantiels de celle-ci
ne doivent être imprimés ou autrement
printed or otherwise reproduced
reproduits sans son autorisation.
In compliance with the Canadian
Privacy Act some supporting forms
may have been removed from this
Conformément à la loi canadienne sur la
without the author's permission.
protection de la vie privée, quelques
formulaires secondaires ont été enlevés de
thesis.
cette thèse.
While these forms may be included
in the document page count, their
removal does not represent any loss
Bien que ces formulaires aient inclus dans
la pagination, il n'y aura aucun contenu
of content from the thesis.
¦+¦
Canada
manquant.
ABSTRACT
DEVELOPMENT OF A GRAPE HARVESTER YIELD MONITORING SYSTEM
FOR PRECISION VITICULTURE
Albert Brooks
Advisor:
University of Guelph
Professor R.B. Brown
Precision Viticulture is a relatively new management method which tailors vineyard
inputs according to site specific conditions. This thesis describes a grape harvester yield
monitoring system that was developed to measure the instantaneous and total weight
of grape material as it flowed through a grape harvester. An in-lab test-bed was
developed to provide dynamic calibration and analyze the yield monitor prototype in a
controlled and repeatable environment. In-lab testing resulted in an average error in
measurement of -2.4% between actual and yield monitor weights. Field testing
indicated an average error of 7.1% between total harvested weight and yield monitor
weight. Results showed that the method used to measure the mass flow rate of grape
material is capable of meeting the demanding requirements of a precision yield
monitoring system.
Acknowledgments
I have had the distinct pleasure of working under the supervision of Dr. Ralph Brown
during the course of this thesis project. He encouraged me to maintain a clear objective,
and seemed understanding of my methods of learning, as well as my slightly off canter
way of looking at the world. I would also like to thank him for providing a detailed
introduction to instrumentation during my undergraduate education. I would also like to
thank Dr. Gordon Hayward, a man of countless talents and entertaining stories. He was
instrumental in helping me develop my instrumentation, while increasing my knowledge
of random facts, will help me in trivia some day.
I would also like to acknowledge all of the assistance that Moatasim Sidahmed has
provided me over the course of this project. Micha Wallace was beyond helpful, and
aided me in many ways. I would also like to thank my parents, Joseph and Mary Brooks
for all of their help over the years. My father, who is an incredibly talented mechanic
and inventor, was the original source of my mechanical curiosity.
This project would not have been possible without the assistance of Lakeview Vineyard
Equipment Inc. and William Falk Farms Ltd. In particular, I would like to thank Bill and
Trevor Falk, and Joe Pillitteri, who provided the main drive behind this project, as well as
providing access to harvesters, parts and any information that was needed. Their
assistance was invaluable and greatly appreciated. It was a pleasure to work in
conjunction with industry partners such as themselves.
i
Table of Contents
Acknowledgments
I
Table of Contents
'¦
List of Tables
?
List of Figures
vii
Nomenclature
·
Chapter 1
1.0
*¦
1
Introduction
1
1.2
Precision Viticulture
1
1.3
The Role of Yield Monitoring in Precision Viticulture
3
Chapter 2
5
2.0
5
2.1
Current Vineyard and Mass Flow Yield Monitoring Systems
Overview of Current Yield Monitoring Methods
2.1.1
2.1.2
2.1.3
2.1.4
HarvestMaster™HM-500Yield Monitor
Farmscan™ Canlink 3000 GRM
HarvestMaster™HM-570Yield Profile Sensor
Yield Monitoring Systems under Development
.'.
Chapter 3
3.0 Grégoire™ Grape Harvesters
3.1 Harvesting Module
3.2 Harvesting Components
3.2.1
3.2.2
3.2.3
3.2.4
3.2.6
5
5
6
8
9
12
12
12
15
Picking Head
Catcher Tray
Main Conveyor Belt
Stainless Steel Drop Chute and Cross Conveyor Arrangements
Collection Bins and Discharge Conveyor
15
17
18
20
21
Chapter 4
24
4.0
24
Project Objectives
4.1 Grape Harvester Yield Monitoring Prototype
4.2 Static and Dynamic Calibration of Yield Monitoring Prototype
4.3 Field Test Yield Monitoring Prototype
24
24
25
4.4 Provide Design Recommendations for Future Development
25
Chapters
26
5.0 Yield Monitor Design Requirements
5.1 Constraints
5.2 Criteria
26
26
27
·
5.3 Design Requirements
5.3.1
Main Conveyor Belt Mass Flow Rate
5.3.2
Harvested Wine Grape Material Properties
5.4 Yield Monitor Instrumentation Sites
5.4.1
Grape Storage Bin
5.4.2
Cross Conveyor Unloading Site
5.4.3
Main Conveyor Belt Tension Load Cell
5.4.4
Main Conveyor Discharge Installation Site
5.5 Installation Site Infrastructure Limitations and Design
5.5.1
Cross Conveyor Mounting
27
·
27
29
31
31
32
33
34
36
36
Ü
5.5.2
Yield Monitor Power Source
5.5.3
5.5.4
Yield Monitor Conveyor Assembly and Discharge Rate
Sanitary Design Considerations
...38
38
39
Chapter 6
6.0 Mechanical, Instrumentation and Experimentation Design
6.1 Suspended Conveyor Assembly
6.2 Suspended Conveyor Hopper Assembly
6.3 Mechanically Isolated and Suspended Conveyor Configuration...:
6.3.1
41
41
41
43
44
Static Loading Conditions and Load Cell Loading Range
45
6.4 High-speed and Static Weighing using Load Cells
6.4.1
6.4.2
6.4.2
49
Dynamic Response Characteristics of a Load Cell Force Transducer
Impact Weighing and Momentum Transfer
Projected Grape Material Impact Loading
51
53
56
6.5 Constructed Prototype
6.6 Instrumentation Design
58
59
6.6.1 Omega Engineering LCAE- 20Kg Load Cell
6.6.2
Load Cell Circuitry
6.6.3
Amplifier Circuitry and Offset Removal Circuitry
6.6.4
Accelerometer Circuitry
6.6.5
Conveyor Drive Motor Circuitry
6.6.6
Instrumentation Power Supply System and Regulation
6.6.7
Instrumentation Installation and Cable Routing.:
59
61
62
64
66
67
68
6.7 Control and Data Acquisition Interface Methodology
6.7.1
70
Data Acquisition System
70
6.8 Raw Signal to Mass Measurement Conversion
6.9 Yield Monitor Test-bed Design
6.9.1
6.10
6.11
6.12
71
74
Conveyor Belt Speed Control
75
Test-bed Performance and Calibration
Test-bed Dynamic Testing Methodology
Field Testing Methodology
77
79
80
Chapter 7
7.0
7.1
-81
Results and Discussion
Yield Monitor Static Calibration and Performance
7.1.0
Static Calibration Methodology
7.1.2
Static Calibration Equations
7.2
Yield Monitor Dynamic Calibration and Performance..
·
··¦····—
81
81
.....81
82
84
7.2.1
7.2.2
Load Cell Signal Results
Frequency Analysis of Load Cell Signals
.86
87
7.2.3
Filtered and Conditioned Load Cell Signals
90
7.2.4
7.2.5
7.2.6
7.2.7
7.2.8
Load Cell Periodic Waveform Analysis
Vibration and Acceleration Analysis
Accelerometer Frequency Analysis
Dynamic Calibration Uncorrected Results
Dynamic Calibration Corrected Results
98
100
104
105
109
7.3
Yield Monitor Field Testing
7.3.1
7.3.2
. 7.3.3
7.3.4
7.3.5
112
Field Testing Load Cell Analysis
First Order Dynamic Response Characteristics
Frequency Analysis of In-field Testing Load Cell Signals
Vibration and Acceleration Analysis
Field Testing Accuracy Results
7.3.6 Experimental Yield Plots
7.4 Yield Monitor Mechanical Performance Evaluation
Chapter 8
113
116
1 19
121
123
125
127
130
¡ii
8.0 Post Experimental Tear-down and Evaluation
130
Chapter 9
133
9.0
133
Conclusions
9.2 Evaluation of Dynamic Testing of the Yield Monitor Prototype
9.3 Evaluation of Field Testing of the Yield Monitor Prototype
9.4 Evaluation of Instrumentation and Signal Filtering Techniques
Chapter 10
137
10.0
Recommendations
10.1
Further Production Recommendations
10.1.1
10.1.2
10.1.3
133
134
755
137
137
Overview of Yield Monitoring System
Supervisory Control and Human Interface System
Power Supply Regulation and Control
137
139
139
10.1.4 Instrumentation and Yield Monitor Control (CANBUS System)
140
10.1.5
10.1.6
10.1.7
10.1.8
140
141
142
143
10.2
10.3
Instrumentation Amplifier and Offset Removal Circuitry
Hydraulic Drive System and Servo Control
Hardware Design
Control Methodology
Performance Testing Recommendations
Signal Processing and Filtering
143
143
11.0
References
............................................................................................................................................................ 145
Appendix A: Butterworth Filtering Parameters
Appendix B: MatLab Fast Fourier Transform (FFT) Code
Appendix C: Unit Analysis of Mass Measurement
149
151
152
Appendix D: Dynamic Calibration Results
Appendix E: Field Testing Results
Appendix F: Engineering Drawings
154
160
162
IV
List of Tables
Table 1 : Grape yield, vineyard and harvester productivity, trellis spacing and conversion
FACTORS
.'
Table 2: Operating specifications for the Omega Engineering LCAE-20kg load cell
28
60
Table 3: DVE switching power supply model number DSP-1454P output voltages and current
ratings
67
Table 4: Auger calibration trials illustrating the mass of grape material, duration, and
average mass flow rate
78
Table 5: Dynamic calibration trials results indicating the mass of grape material for each
trial, duration and calculated averaged mass flow rate
106
Table 6: Test results from the test-bed dynamic calibration testing utilizing Load Cell C
LINEAR CALIBRATION EQUATION
106
Table 7: Test results from the test-bed dynamic calibration testing utilizing Load Cell B
linear calibration equation
107
Table 8: Test results from the test-bed dynamic calibration testing utilizing the combined
signals from the load cell b and c linear calibration equations
108
Table 9: Test results from the test-bed dynamic calibration utilizing the combined signals
from the Load Cell B and C linear calibration equations, but assuming an even
distribution of loading between load cells. therefore the full load is half the
measured amount
109
Table 10: Results from the test-bed dynamic calibration testing utilizing the corrected Load
cell c linear calibration equation
110
Table 1 1 : Test results from the test-bed dynamic calibration testing utilizing the corrected
Load Cell B linear calibration equation
Ill
Table 12: Test results from the test-bed dynamic calibration testing utilizing the combined
signals from the corrected Load Cell B and C linear calibration equations, and
corrected offset ;
Ill
Table 13: Results from the in-field testing utilizing the corrected Load Cell C signal
123
Table 14: Results from field testing using the uncorrected static calibration equation as
applied to Load Cell C signal
124
Table 15: Results from the in-field testing utilizing summed Load Cell A and C signals, and
assuming an even distrd3ution of mass
124
Tablé 16: Auger calibration trials illustrating the mass of grape material, the duration, and
the calculated average mass flow rate
154
Table 17: Auger calibration error analysis descriptive statistics
154
Table 18: Dynamic calibration trials results indicating the mass of grape material for each
trial, duration and calculated averaged mass flow rate
154
Table 19: Dynamic calibration trial error analysis descriptive statistics for the average
mass flow rate of each trial
155
Table 20: Test results from the test-bed dynamic calibration testing utilizing Load Cell C
SIGNAL AND LINEAR CALIBRATION EQUATION
155
SIGNAL CALIBRATION EQUATION
156
Table 21 : Load Cell C error analysis descriptive statistics
155
Table 22: Test results from the test-bed dynamic calibration testing utilizing Load Cell B
Table 23: Load Cell B error analysis descriptive statistics
156
Table 24: Test results from the test-bed dynamic calibration testing utilizing the combined
signals from the Load Cell B and C calibration equations
Table 25: Combined Load Cell B and C signal error analysis descriptive statistics
156
157
Table 26: Test results from the test-bed dynamic calibration utilizing the combined signals
from the Load Cell B and C linear calibration equations, but assuming an even
distribution of loading between load cells. Therefore the full load is half the
measured amount
157
Table 27: Combined Load Cell B and C error analysis descriptive statistics
157
Table 28: Results from the test-bed dynamic calibration testing utilizing the corrected Load
Cell C linear calibration equation
158
V
Table 29: Corrected Load Cell C error analysis descriptive statistics
158
Table 30: Test results from the test-bed dynamic calibration testing utilizing the corrected
Load Cell B calibration equation
Table 31: Corrected Load Cell B error analysis descriptive statistics
158
159
Table 32: Test results from the test-bed dynamic calibration testing utilizing the combined
signals from the corrected load cell b and c calibration equations, and corrected
OFFSET
.'
Table 33: Combined and corrected Load Cell B and C error analysis descriptive statistics.
Table 34: Results from the in-field testing utilizing the corrected Load Cell C signal
Table 35: Load Cell C in-field testing error analysis descriptive statistics
Table 36: Results from field testing uncorrected Load Cell C signal
Table 37: Uncorrected Load Cell C in-field testing error analysis descriptive statistics
159
159
160
160
160
160
Table 38: Results from the in-field testing utilizing summed Load Cell A and C signals, and
ASSUMING AN EVEN DISTRIBUTION OF MASS
161
Table 39: Combined Load Cell A and C in-field testing error analysis descriptive statistics.
............ ;
161
vi
List of Figures
Figure 1: HarvestMaster™ HM-500 yield monitor installation (Jardine & Muffoletto, 2001)a.5
Figure 2: Farmscan™ Canlink 3000 GRM yield monitor load cell (top), and installed unit
(bottom) (Farmscan, 2005)
7
Figure 3: HarvestMaster™ HM-570 installation (left) and generated yield distribution map
(right) (bramley & williams, 2001)
8
Figure 4: Schematic representation of a Grégoire™ harvester with the detachable
harvesting module(a) as well as the mounting points onto the frame (b and c)
(Kverneland Group, 2006)
13
Figure 5 : Flow diagram illustrating the movement of product from site of harvest to
collection bins. Illustrated are the cross conveyor (A), picking head assembly (B) and
catcher tray (C) components (Kverneland Group, 2006)
Figure 6: ARC system picking head assembly (Kverneland Group, 2006)
14
16
Figure 7: Rear view of the picking head (A), as well as catcher tray (B) and main conveyor
belts (C)
17
Figure 8: Image showing the stainless steel drop chute, upper pulley and main conveyor belt
(Kverneland Group, 2006)
19
Figure 9: Horizontal cross conveyor arrangement. This is the most commonly seen model of
harvester in the Niagara region
20
Figure 10: Collection bins with upper extraction fans mounted on the saddle bins (left), and
central collection bin arrangement (right) (kverneland group, 2006)
21
Figure 1 1 : Side discharge conveyor on a Grégoire™ G108 harvester (Kverneland Group, 2006).
............................................................................................................................................. 22
Figure 12: Harvested grape material included leaves (A), stems (B), free running juice (C) and
SINGULAR GRAPES (D). CHARDONNAY GRAPES HARVESTED FROM WILLIAM FALK FARMS IN
September of 2008
30
Figure 13: Main conveyor belt drive and tension spring assembly. Note the installation of a
pancake load cell under the drive belt tension spring (g)
33
Figure 14: Site of the cross conveyor unit and the installation site for the yield monitor. This
location was the only common area to all gregoire™ harvester models with adequate
although limited space. indicated were the general dimensions of the cross conveyor
UNIT
·
35
Figure 15: Illustration of the suspended conveyor prototype. Top image illustrates the
numerous internal components. bottom image illustrates the fully assembled
prototype
41
Figure 16: Direction of grape material flow over the prototype yield monitoring system
42
Figure 17: Grape harvester yield monitor prototype hopper arrangement. The metal frame (A)
and flexible shielding (B) are illustrated. The hopper assembly bolts onto the outer
frame of the yield monitor prototype
:
44
Figure 18: Illustration of active and inactive components of the weighing system. Left:
Bottom of the prototype illustrating active area (B), inactive area (C) and load cells
(A). Right: top view of the active components that would be subject to grape material
FLOW
45
Figure 19: Identified are the main components of the monitoring system, and the loading points
of Load Cell A (A), Load Cell B (B), Load Cell C (C), and the centre of mass (D)
46
Figure 20: Static distribution of weight over the load cells when the conveyor is unloaded.
Note that Load Cell B would experience a higher static loading then the others
Figure 21 : Free- body diagram of the live end of the load cell (Gilman & Bailey 2005)
Figure 22: Collision between the live end of the load cell and the equivalent mass, M.
Illustration of the approach and separation of the objects (F & Bailey 2005)
48
52
54
Figure 23 : Grape yield monitor prototype installed on a Grégoire™ harvester. This
configuration was used for final field testing and dynamic calibration testing
59
Figure 24: Omega Engineering LCAE-20 kg load cell. Note the tapped mounting holes and
WATER RESISTANT EPOXY COATING
-60
VU
1
Figure 25: Wheatstone bridge strain gauge configuration used in the Omega Engineering
LCAE- 20kg load cell (Omega Engineering, 2008)
62
Figure 26: Instrumentation amplifiers and offset null ciRCurr schematic. Simple circuit
ALLOWED FOR LOAD CELL SIGNAL AMPLIFICATION AS WELL AS OFFSET NULL BY ALTERING THE
VOLTAGE ON THE POSHTVE INSTRUMENTATION AMPLIFIER INPUT PIN VIA A TRIM POTENTIOMETER 62
Figure 27 : Performance characteristics of the LT 1 1 68 instrumentation amplifier integrated
circuits. Left: Common Mode Rejection Ratio (CMRR) vs. Frequency plot illustrating
signal attenuation. Right: Gain vs. Frequency plot illustrating signal frequency
attenuation as gain levels increase, effectively limiting bandwidth (linear technology,
2008)
Figure 28: Accelerometer response to a perturbation and subsequent system response.
64
65
Figure 29: Dimension Engineering Buffered 3g 3-axis Accelerometer breakout board with
voltage regulation, reverse voltage protection and output short protection. This
breakout board was designed to fit into a 16 pin DIP socket for soldering onto boards
(Dimensionengineering, 2008)
66
Figure 30: Installed amplifiers (A), three axis accelerometer (B), and offset removal (C)
circuitry inside of the monitor frame. Note the use of strain relief connectors and
grommets to protect the load cell and signal cables from environmental hazards,
accelerometer axes are labeled as well
68
Figure 3 1 : Signal conditioning board layout with instrumentation amplifiers (A), offset tare
circuitry (d), and 3-axis accelerometer (c) (dimension engineering de-accm3d
accelerometer which incorporates an adxl330 mems accelerometer)
69
Figure 32: Instrumentation and control flow diagram for the yield monitor prototype.
71
Figure 33: Visual concept of Simpson's Rule of integration approximation. The shaded region
corresponded to the area under the curve created by the approximation function p(x)
(Simpson's Rule for Integration, 2008)
72
Figure 34: Selected period over which the load cell signal was integrated to produce the
total mass upheld on the conveyor belt. figure illustrates the force (n), as well as the
mass (kg) on the system, and illustrates the removal of gravity from the system
74
Figure 35: Left: Parts diagram illustrating the general layout of a Grégoire™ harvester
main conveyor assembly. right: constructed evaluation test bed for dynamic
calibrationoftheyieldmonnor
75
Figure 36: Gear head drwe motor and speed controller for the dynamic test-bed. Linear belt
speed can be dialed into correct levels for testing
76
Figure 37: Metering auger for grapes to ensure a more uniform discharge of grape material
onto the main belt. speed of the auger was controlled by a motor with a rheostat
77
Figure 38: Discharge from the test-bed onto the top surface of the weighing conveyor (Left).
Grégoire™ grape harvester discharging into the cross conveyor (Right)
Figure 39: Final prototype installed into a Grégoire™ harvester for field testing trials
79
80
Figure 40: Basic layout of the yield monitor. The weight illustrates the location and method
of static loading for static calibration purposes
82
Figure 41 : Static calibration plot illustrating the response of all three load cells to a .
known centrally placed weight. this calibration was obtained wtth the conveyor belt
installed and tensioned to operating levels. shown are the calibration equations, as
well as their associated coefficients of determination values
83
Figure 42: Grape harvester yield monitor installed into the testing apparatus. Grapes fall
from the top of the pulley, conveyor belt area. grapes were discharged toward the
front for collection and bulk weighing for calibration purposes
85
Figure 43: Load cell voltage outputs when subjected to dynamic loading scenarios identical
to field conditions
87
Figure 44: Fast Fourier Transform (FFT) of the unfiltered test-bed signal (left) for Load Cell
C. LOW PASS FILTERED SIGNAL (RIGHT)
,
88
Figure 45: Fast Fourier Transform (FFT) of the unfiltered test-bed calibration for Load Cell
B(LEFT). LOW PASS FILTERED SIGNAL (RIGHT)
89
Figure 46: Fast Fourier Transform (FFT) of the unfiltered test-bed signal (left) for Load Cell
A. LOW PASS FILTERED SIGNAL (RIGHT)
89
VIII
Figure 47: Load cell signal (voltage) output when subjected to dynamic loading scenarios
identical to the expected field conditions. indicated are when the prototype conveyor
was turned on and when grape material impacted and was discharged from the surface.
................................................................................................................................................. 91
Figure 48: Load cell (voltage) output when subjected to dynamic loading scenarios identical
to the expected fteld conditions. only the offloading (load cell c) signal is
illustrated
Figure 49: Calibrated Load Cell C signal
Figure 50: Calibrated Load Cell B signal
Figure 51: Calibrated Load Cell A signal
92
93
94
95
'.
Figure 52: Resulting yield mass flow rate when the signals from Load Cell B and C are
combined in an additive method
96
Figure 53: Resulting yield mass flow rate when all load cell signals are combined .
97
Figure 54: Periodic waveform illustrating.the difference (error) in the shifted signals from
the original Load Cell C signal...
98
Figure 55: Periodic waveform illustrating the difference (error) tn the shifted signals from
the original Load Cell B signal
100
Figure 56: Test-bed results of the 3-axis accelerometer sjgnal. Illustrated is the full time of
a loading experiment, including run up and down time. Signal offset was the result of
inclination error from the placement of the 3 AXIS ACCELEROMETER within the
instrumentation box
Figure 57: Test-bed results of the 3-axis accelerometer signal, illustrating the output
CHARACTERISTICSOFTHESIGNALS5ASWELLASTHEMEANOFTHESIGNAL
101
103
Figure 58: Test-bed results of the 3-axis accelerometer signal. A 15 second interval is
illustrating the characteristics output of the signals. Signals with inclination offset
removed
104
Figure 59: Fast Fourier Transform (FFT) of the unfiltered test-bed calibration trial for
accelerometer axis 1 signal (top left), Axis 2 (top right), and Axis 3 (bottom)
105
Figure 60: 1 Hz low pass filtered load cell signals from a typical in field trial. (Vineyard Row
3 at William Falk Farms)
114
Figure 61 : Grape material loading characteristics of the yield monitor during field testing.
Active measurement points of Load Cells A, B and C is indicated as well as main area of
loading(d)
115
Figure 62: 1 Hz low pass filtered load cell signals (calibrated) from a typical in field test.
Load cell signals were combined in each possible configuration for evaluation purposes.
Vineyard Row 3 at William Falk Farm
Figure 63: First order response characteristics of the field testing
116
117
Figure 64: Fast Fourier Transforms (FFT) of the unfiltered field testing signals. Load Cell C
(top left), Load Cell B (top right), and Load Cell A (bottom).
120
Figure 65: Field testing results of the 3-axis accelerometer signal. Illustrated is the full
harvesting time of a vine trellis
122
Figure 66: First three rows of measured field yield information. Yield distribution map
produced using surfer8 mapping software. the kriging interpolation method used to
generate interlining data points
126
- Figure 67 : Last three rows of measured field yield information. Yield distribution map
PRODUCED USING SURFER8 MAPPING SOFTWARE. THE KRIGING INTERPOLATION METHOD USED TO
GENERATE INTERLINING DATA POINTS
Figure 68: Installed yield monitor (A), area for discharge (B), vertical extraction fan
126
HOUSING (C) AND PLASTIC FLAP REMOVED (D)
127
system illustrating the buildup of material (right)
130
Figure 69: Identified design flaw in the system. A build up of material (left) caused by the
limited discharge area under the vertical vacuum extraction fan (right)
128
Figure 70: Suspended conveyor yield monitoring system inside the test-bed (left). Removed
Figure 7 1 : Conveyor unit with the belting lace removed, and illustrating the buildup of
MATERIAL ONLY ALONG THE OUTER EDGES (LEFT). CLOSE UP VIEW OF THIS BUILD UP MATERIAL
(RIGHT)
'¦
IX
131
Figure 72: Underside of the Conveyor unit illustrating the lack of liquid or material
contamination (left) and the instrumentation box completely unscathed (right)
132
Figure 73: Instrumentation and control flow diagram to meet the requirements of having two
yield monitors installed on the harvester, one for each main conveyor
138
Figure 74: Comparison between raw Load Cell C output signal and different filtering
parameters
150
Figure 75 : Accumulated mass of static material as well as impulse force from falling
material was measured by the load cells. the output signal was an additive sum of
these forces
,
152
Figure 76: Illustration of the area under the load cell signal, which was a combination of the
impulse grape impact loading, and the held mass on the yield monitor conveyor
153
Figure 77: Engineering drawing of the reference dimensions of the assembled yield monitor
prototype
.....162
Figure 78: Engineering drawing of the top outer frame assembly illustrating the main
components of the design
163
Figure 79: Engineering drawing of the motor mounts end plate. Included are detailed sections
of the design
164
Figure 80: Engineering drawing of the small end plate. Included are detailed sections of the
end plate assembly
,
165
Figure 81: Engineering drawing of the tension and bearing support assembly
Figure 82: Engineering drawing of a conveyor roller tube section
166
167
Figure 83: Engineering drawing of the V-grove roller insert that allows the conveyor belt
to track correctly
167
Figure 84: Engineering drawing of the Omega Engineering LACE-20kg aluminum cantilever
loadcell
168
Figure 85: Top UHMW polyethylene plate of the yield monitor conveyor
169
X
Nomenclature
Analog-to-Digital Converter (ADC)
Common Mode Rejection Ratio (CMRR)
Dual Inline Package (DIP)
Electromagnetic Interference (EMI)
Fast- Fourier Transform (FFT)
Global Positioning System (GPS)
Matter Other than Grapes (MOG)
Precision Agriculture (PA)
Precision Viticulture (PV)
Xl
Chapter 1
1.0
Introduction
1.2
Precision Viticulture
"Precision Viticulture (PV) is an all encompassing term given to the use of a range of
information technologies that enable grape growers and winemakers to better see and
understand variability in their production systems, and to use this understanding to
better match the input to production to desired or expected outputs (Bramley, 2003) ".
The concept is based on new tools and information sources provided by modern
technologies, such as yield monitoring devices, soil, plant and pest sensors and remote
sensing (Moráis et al, 2008). The adoption and implementation of PV to vineyard
management is a continual cyclical process that begins with observation of vineyard
performance and associated vineyard attributes, followed by interpretation and
evaluation of the collected data, leading to implementation of targeted management
(Bramley, 2003; Sethuramasamyrija et al, 2007; Lamb et al, 2004).
Targeted management can mean the timing and application rate of water, fertilizer or
spray, or the use of machinery and labour for operations such as harvesting, pruning or
just about any aspect of vineyard management (Bramley, 2003). The key technologies
involved in PV are the global positioning system (GPS) and a grape yield monitor, which
may be considered as a primary observation tool.
1
The use of variable rate equipment and knowledge of vineyard performance promotes
the most efficient use of inputs to the production system, leading to improved cost
effectiveness, and enhanced sustainability (Bramley, 2003). PV also allows the ability to
separate the crop at harvest and to harvest according to quality specifications, which
can have a drastic effect on industry profitability [Bramley, 2003, (Lamb et al, 2004) ].
Winemakers strive to produce higher valued wines using better quality grapes, but the
inclusion of sub-optimal fruit can drastically reduce the overall value. Therefore the
ability to ensure that only the best quality fruit is used holds considerable appeal to
winemakers (Bramley, 2003). A case study by Dr. Bramley, (2003) determined that it is
possible to use yield monitoring and other PV derived information to selectively harvest
and segregate grapes based upon quality indicators (Bramley, 2003). This study
indicated that by applying selective harvesting techniques a total gross retail value of
production using targeted harvesting strategy could result in an income increase of up
to $2661/t of fruit harvested, or $30,790/ha (Bramley, 2003).
There are other potential benefits of incorporating PV practices into modern vineyard
management strategies, including the ability to better manage and schedule harvesting
timelines and strategies. Crop separation affords possibilities with respect to tailoring
harvesting according to market demand, expectations and winery storage capacity
(Bramley, 2003). The ability to ensure that the best wines are given optimal time in the
fermentation tanks has become a critical issue in winery management. Knowledge of
vineyard variability and yield monitoring information should lead to an improved
2
capacity for accurate crop forecasting. Such information could allow greater
information for vineyard management decisions, such as scheduling of harvest, and midseason operations relating to yield or quality control such as thinning or leaf plucking.
1.3
The Role of Yield Monitoring in Precision Viticulture
Yield monitoring is the process of documenting spatial and temporal yield variability
throughout a field; and is often the first step in developing a precision agriculture
program (Bramley & Hamilton, 2004; Bora et al, 2006). Yield information can be tagged
to Global Positioning System (GPS) coordinates to provide accurate information for use
in yield distribution field maps and by use in other precision application processes
(Bramley, 2003). Information obtained by mounting a yield monitoring system onto a
mechanical harvester can give accurate information about the distribution and variation
of yield acrossa vineyard of interest. Based on this information; vineyard management
practices can be tailor made to the variability of the vineyard (Bramley, 2003; Lamb et
al, 2004; Bora étal, 2006). The acquisition of grape yield information could result
(under proper management) in an increase in the efficiency of vineyard operations
(Bramley, 2003). It is possible to determine the temporal stability of the vineyard
variation and enact target harvesting strategies for the regions of interest. The direct
measurement of yield distribution information by a yield monitoring system provides
the most usable and accurate information about vineyard growing characteristics when
compared to other vineyard measurements. The ease at which this information can be
obtained and the correlation between grape yield and grape quality indicate that yield
3
monitoring using a harvester mounted yield monitoring system can provide the most
important measurement needed for precision viticulture management practices.
Chapter 2
2.0 Current Vineyard and Mass Flow Yield Monitoring Systems
The following sections detail current yield monitoring strategies. Those designed
specifically for grape harvesters, as well as other systems that incorporate features or
illustrate issues that would be applicable to a grape harvester yield monitoring system
were introduced.
2.1
Overview of Current Yield Monitoring Methods
2.1.1 HarvestMaster™ HM-500 Yield Monitor
The HarvestMaster™ HM-5001 yield monitor relied on two conveyor belt idlers
instrumented with load cells to record the weight of product as it moves across a belt
(Jardine & Muffoletto, 2001). Figure 1 illustrates of how the HM-500 system is installed
onto a conveyor belt system, and as can be seen, the load cell is oriented between the
harvester frame bolts and the idler wheel. This cantilever arrangement improves the
sensitivity of the system, but may result in vibration and noise interference.
?
/
\\ \
ppp
/
Scafii·
Hameáter r name
/
!/4"NeItEh
Loaded!
(a)
(b)
Figure 1: HarvestMaster™ HM-500 yield monitor installation (Jardine & Muffoletto,
2001)a.
Juniper Systems Incorporated. North Logon, Utah.
5
Figure 1 above illustrates where this type of unit may be installed on a conveyor belt (a)
as well as an actual load cell and idler configuration (b) (Jardine & Muffoletto, 2001a).
This type of monitor can only be installed on harvesters with a side discharge conveyor,
or a similar section of long conveyor belting.
Pelletier et al. (2001) noted that the most serious operational challenge to beet
weighing systems was the dynamic,, off-road environment with changing conveyor
angles. Errors in belt weighing methods may be limited to 0.25-0.5% provided a number
of criteria were met. A stable, non-mobile platform with long belt runs of over 30cm
and less then 12° of inclination is required. A system mounted on a grape harvester
would be subjected to strong vibrations from the picking head, as well as the engine,
hydraulic motors and the uneven ground conditions of a vineyard.
2.1.2 Farmscan™ Canlink 3000 GRM
The system designed by Farmscan™'2 was very similar to the HarvestMaster™ system.
The Canlink 3000 GRM system utilized cantilever idlers with load cells to weigh product
as it moved along a conveyor belt (Figure 2) (Farmscan, 2005). Unlike the
HarvestMaster™ system, this system contained plates before and after the load cell, as
can be seen in Figure 2. This reduces many of the dynamic characteristics of the belt;
such as wave-like motions and excess vibration. The technical challenges involved in
producing an accurate conveyor weigh system previously mentioned also apply to this
configuration.
2 Farmscan, Perth, Australia.
6
Figure 2: Farmscan™ Canlink 3000 GRM yield monitor load cell (top), and installed unit
(bottom) (Farmscan, 2005).
These types of systems would often be installed in a long discharge conveyor that
unloads into a trailing hopper. These long discharge conveyors can be hydraulically
lifted and lowered, so an inclination sensor was required to correct for the downward
acceleration due to gravity.
The research conducted by James Taylor at the University of Sidney (Taylor, 2002)
indicated that the Farmscan™ Canlink system provided far more accurate results then
the HarvestMaster™ counterpart. He also indicated that the HarvestMaster™ system
was often turned off due to abnormally high or low readings.
7
2.1.3 HarvestMaster™ HM-570 Yield Profile Sensor
The HarvestMaster™ HM-5703 profile yield monitoring system was an ultrasound
system which records the height of material as it moved along a conveyor belt, and
applied a density to the recorded height to obtain a measurement of the mass of
material (Jardine & Muffoletto, 2001b). This profile unit had a scan frequency of 16.67
samples per second, and if the conveyor were traveling at an average discharge rate of 1
meter per second, this resulted in a height profile measurement every 6cm (Jardine &
Muffoletto, 2001b). This system must also be placed at a set height from a flat conveyor
belt, as seen in Figure 3. Any raised buckets or ribs on the belt would be counted as
product with this sensor measurement system. The transducer array itself contained
either 3 or 5 piezoelectric elements, depending on the model (Jardine & Muffoletto,
2001b).
Yield (t/ha)
yrgg-2
8-10
----- 10-12
Figure 3: HarvestMaster™ HM-570 installation (left) and generated yield distribution
map (right) (Bramley & Williams, 2001).
Bramley and Hamilton (2004), indicated that the HarvestMaster™ system calibration
remained stable over any harvesting operation, and due to this stability the data can be
post processed after the harvest on the basis of total tonnage delivered to the winery.
'juniper Systems Incorporated. North Logon, Utah.
8
Taylor, (2002) compared different monitoring systems, including the HarvestMaster
HM-570 profile yield sensor, as well as the load cell type (HM-500). The accuracy and
precision of this system was determined to be poor, and regarded as unsatisfactory
(Jardine & Muffoletto, 2001b). The manufacturing company no longer supplies this
device.
2.1.4 Yield Monitoring Systems under Development
Several types of yield monitoring systems have been under development within
academia for many years for numerous types of applications such as sugar cane, sugar
beet, tomato and potato crops. The sugar cane yield monitor developed by Magalhaes
and Cerri (2007) was an impact type yield monitor similar to systems developed for
cereal grain combines. This system was based on load cells acting as a billet weighing
instrument, and was located on the upper part of the harvester side conveyor. The
weighing plate was supported by two GL-30 load cells with 30 kg nominal capacity. The
output signal from this flat plate load impact sensor was filtered using an analogue low
pass filter to remove mechanical vibrations of the harvester, and accelerometers were
placed on several points of the elevator for analysis by fast Fourier transform. This
design was able to measure total weight of the sugar cane product to within an average
of 4.3% error.
Another system developed by Walter and Backer (2003), was based on a
HarvestMaster™ HM-500 yield monitoring system that was reconfigured for
continuously weighing sugar beets on a conveyor chain system. The authors of this
9
study compared two versions of their system, the factory standard HarvestMaster
HM-500 system type that consisted of an idler sprocket supporting the conveyor chain,
and another that incorporated a polyethylene slide to remove any artifacts caused by a
grabbing drive chain. The performance of both systems was very dependent on the
installation of the system into the pre-existing harvester infrastructure. If the chain
tension was changed or the orientation of the idler wheels was not tangent to the chain,
the weigh-sensing elements would read incorrect or variable loadings. This system
could only be installed in machines with long sections of conveyor chains or belting and
in areas where there was at least one idler between the sensing elements and a drive
sprocket or chain tension mechanism.
This system was able to maintain an average accuracy of 2.5 - 3.5% of actual weight for
the modified polyethylene slide and idler arrangement, respectively (Walter & Backer,
2003). This load cell arrangement for continuously weighing product produced adequate
accuracy and repeatability, but was highly dependent on both harvester infrastructure
and installation variability. This system also requires a long run of horizontal conveyor
for installation, limiting applicability to other systems.
Numerous studies and yield monitor prototypes have focused on weighing the
accumulated mass of product as it was loading into a storage bin. This mass
accumulation technique was used on a peanut yield monitoring system, as well as a
silage harvester. In these systems, the storage bins were mechanically isolated from the
10
rest of the system by load cells, which measured the accumulated mass of the
suspended components. The bin must be isolated from the rest of the machine to
prevent interference from hydraulic cylinders or binding between mechanical parts
(Durrence ef al, 1999). Both peanut harvesters and silage wagons are typically easy
systems to mechanically isolate and the harvested material are mass flow rates are
typically high as well, making them ideal candidates for this method of bulk weighing.
As noted by Durrence et al, (1999), accelerations induced by rough terrain can cause
large mass fluctuations to be recorded by the load cells, due to the inertia of the large
amount of suspended mass. Since the load cells must be scaled for full load capacity,
small changes in continuous loading can be lost due to limited sensitivity in the load
cells, as well as noise from ground and machinery induced accelerations. This limits the
applicability of this type of system to instantaneous yield monitoring, but provides very
accurate totalized yield information.
11
Chapter 3
3.0 Grégoire™ Grape Harvesters
The following is a general description of how the different components of a Grégoire™
grape harvester function, as well as a description of the individual components within
each assembly. There are numerous different mechanical harvesters currently on the
market, such as Pellenc, Korvan™, Braud and Grégoire™. All function in a similar
manner with oscillating beater bars to remove grapes, and process harvested material in
similar manner. Grégoire™ grape harvesters are by far the most common type of
harvester platform used in the Niagara grape growing region, with an approximate
market share of 95%. Lakeview Vineyard Equipment Incorporated was the only and
service distributor for grape harvesters within the region, and was kind enough to act as
collaborators.
3.1
Harvesting Module
Mechanically harvesting grapes presents numerous technical challenges. Not only must
the machines produce top quality grapes, but they must do this cost effectively, as well
as accurately to reduce the chances of lost product in a very demanding operating
environment. Self propelled harvesters must straddle a grape trellis evenly, even over
steep slopes, terracing and different types of trellising. These harvester platforms have
become very well adapted for grape production, but can be reconfigured to perform
numerous tasks such as spraying, tillage, trimming and hedging operations.
12
O)
SJL.
©-
''"->/
DI
Í/VM
\or_S,! » W ; ¿*
Figure 4: Schematic representation of a Grégoire™ harvester with the detachable
harvesting module(A) as well as the mounting points onto the frame (B and C)
(Kverneland Group, 2006).
Figure 4 represents the harvest module (A) on a Grégoire™ harvester. This module is
attached by pivot points (B) and (C), which allows the picking head to float in the center
of the machine. This pivoting motion prevents damage to the trellis when operating on
steep gradients (Kverneland Group, 2006). All harvesting functions are performed
within this easily detachable module, but the module itself is a sophisticated piece of
agricultural machinery.
The basic operation of these machines can be simplified to three processes.
1.
Remove the grapes from the vine
2.
Remove all matter other than grapes (MOG) to improve quality
3.
Transport the grapes from the site of harvest to storage bins.
The heart of the harvesting module is the picking head, located (Figure 5) where the
vine trellis moves through the machine. Within this chamber the vines are gently
shaken to remove the fruit, which falls downward to the catcher tray. This catcher tray
13
was designed to catch falling fruit while allowing for the vine to move through the
picking head. From here the fallen and collected fruit move on a long conveyor belt
along the bottom of the picking head, through the vertical extractors and discharged
into a stainless steel drop chute. Cross conveyors move the fruit to the collection bins.
The removal of MOG occurs at two locations, along the catcher tray and conveyor belt,
and along the upper cross conveyor. The flow of product through a Grégoire
harvester can be seen below in Figure 5.
®
A-"
?
fr
ß
&
i
\
>J-
^
>^
/
/
«%%4 Hs
s^¿
&
t^SÎP
/
B
/i
.U
?
nr
/ î
p
r
fr re
3
d
^^P^^^^^PÎ?!?f^
kiS>ÎSSS
¦mCi- ¡«güT-agf: -&-¦¦·-&---&
?,
^a
à
©J
Figure 5: Flow diagram illustrating the movement of product from site of harvest to
collection bins. Illustrated are the cross conveyor (A), picking head assembly (B) and
catcher tray (C) components (Kverneland Group, 2006).
All Grégoire™ harvester models have a similar method of harvesting and conveying fruit
up to the top of the machine. There are many different models and possible options,
such as a side discharge conveyor and de-stemmers. It should be noted that for a
14
collection bin model, both the right and left sides are symmetrical, meaning there are
two main conveyors, drop chutes, cross conveyors and collection bins.
3.2
Harvesting Components
The following components operate together to perform the three main tasks outlined
above. Each description explains how each part functions as well as how they integrate
into the entire system. Any method of monitoring yield was designed to be integrated
into this pre-existing infrastructure.
3.2.1 Picking Head
The picking head is responsible for removing the fruit from the vine while not adversely
harming the vine or trellis. The structure seen in Figure 6 is the ARC system developed
by Grégoire™ and is common to all 100 series machines (Kverneland Group, 2006). This
type of harvesting system uses flexible rods to flex against the base of the vine gently
enough not to cause harm to the foliage or graft. A reciprocating motion is created by a
set of cams mounted on the top of the chassis. These cams are powered by a variable
flow hydraulic motor to provide variable speed control.
15
Figure 6: ARC system picking head assembly (Kverneland Group, 2006).
The labeled components in the diagram correspond to the following:
(E)
(F)
Shock absorbing alignment wishbones
Main chassis
(G)
Mounting chassis for drive cams and hydraulic drive motor
(H)
(I)
(J)
Cam connecting rods
Picking rod holders
Adjustable picking rods
Nearly every feature of a picking head is adjustable to allow for a variety of picking
conditions. The variable delivery hydraulic system can adjust the rod speeds from 0 to
600 rpm on most models. The distance between the picking rods can be changed to suit
the type of vine or trellis as well. Some harvesters also feature a 'pinch' which is used to
adjust the width of the picking head. The pinch allows for the picking head to open
when passing a post (Kverneland Group, 2006).
16
Figure 7 ¡s a rear view of the picking head, as well as the catcher tray and main conveyor
belts. This image allows for a better understanding of how all of the picking head
components integrate into the harvesting module.
Figure 7: Rear view of the picking head (A), as well as catcher tray (B) and main conveyor
belts (C).
3.2.2 Catcher Tray
The catcher tray is oriented directly below the picking head, and is comprised of flexible
black plastic 'flaps' that allow for the trellis to move through the picking head without
damaging vines. As the name implies, the catcher tray prevents product from falling on
the ground. For this reason it needs to be properly adjusted and in good working order.
The tray itself is not a single piece of material, but an overlapping series of flexible
material angled to direct the fallen material to the main conveyor belt. Figure 7 shows
17
the black catcher tray (B). The manufacturer notes that the catcher tray will be
watertight only when the machine is aligned with the vine row. Therefore the way the
machine is driven has a major impact on wastage (Kverneland Group, 2006). The
configuration of this catcher tray assembly means that approximately half of the
harvested material flows to each side of the machine, to be carried away by the
conveyor belt.
3.2.3 Main Conveyor Belt
The main conveyor traverses the entire circumference of the harvesting module, and is
one of the most important components of the system. The routing of the conveyor can
be seen in Figure 5, and it can be seen clearly in Figure 7. Grapes are collected from the
catcher tray and carried by the molded plastic buckets upwards through the rear
elevators. After the conveyor has rounded the upper pulley the contents of the
conveyor are dropped, and the conveyor continues onward to the bottom of the catcher
tray.
The speed of the belt can be hydraulically varied, as well as reversed. These belts can
become jammed, and as such there is a safety system in place that detects conveyor
stoppage or abnormal slowdown (Kverneland Group, 2006). Figure 8 shows the upper
drive and tensioning pulley at the top of the harvesting module.
18
Figure 8: Image showing the stainless steel drop chute, upper pulley and main conveyor
belt (Kverneland Group, 2006).
The main pulley (A) drives the conveyor belt via the V-grove in the conveyor belt. In
order to keep material out of the V-grove scrapers (B) are installed. Tension is
maintained in the belt by adjusting the position of the tensioning arm (C), and spring
(D). Material discharged from the main conveyor passes into the drop chute (E).
Harvester operators have noted serious deficiencies in this system. Most notable was
the tendency for this long belt to slip, consequently the belt was often over tensioned to
as a preventative measure. Over tensioning the belts may result in added stretch, wear,
and a greater chance of binding failure. The flexible rubber buckets were also
susceptible to wear and failure. These reinforced rubber belts were said to be highly
susceptible to clogging and slip caused by grape material.
19
3.2.4 Stainless Steel Drop Chute and Cross Conveyor Arrangements
The drop chute is located beneath the main conveyor and directs the falling material
into the cross conveyor. The drop chute can be seen in Figure 8 where it attaches by a
bolt to the main drive pulley housing. It should be noted that models with upper fans
have a different discharge configuration then that shown in Figure 8.
Dual cross conveyors are only found on models that use either central, or side bins for
storage. The configuration of these conveyors depends on whether upper fans are in
place, and what styles of storage bins were employed, and if a destemmer system was
installed on the harvester.
Another, more common configuration incorporates a central bin with a single vertical
extraction fan placed between both elevator conveyors. Under the discharge of the
main conveyor belt is a flat, horizontal conveyor belt that moves material towards the
middle, and under the extraction fan, as can be seen in Figure 9 below.
Figure 9: Horizontal cross conveyor arrangement. This is the most commonly seen
model of harvester in the Niagara region.
20
This location contained the largest amount of available space for the installation of any
aftermarket additions.
3.2.6 Collection Bins and Discharge Conveyor
Collection bins are responsible for collecting the cleaned grapes while harvesting. These
large bins are mounted on either side of the picking head, as seen in Figure 10 below,
and this configuration of harvesting module has upper extraction fans mounted on the
bins. Not seen are the cross augers used to evenly distribute the grapes throughout the
collection bins. This allows for a greater volume to be loaded before necessitating
unloading.
Illustrated in Figure 10 are the most common types of collection bins present in the
Niagara wine region; the saddle bins (left) and the central collection bin arrangement
(right). To unload, the bin is pushed out by a hydraulic cylinder past the side of the
harvester, where the material is pushed out by a chain conveyor (Figure 10). The central
collection bin model was the most common type used in Ontario.
JWJf
W
f***r\
a'
^
Figure 10: Collection bins with upper extraction fans mounted on the saddle bins (left),
and central collection bin arrangement (right) (Kverneland Group, 2006).
21
To unloading a saddle bin, a hydraulic cylinder pushes upwards on the bin, which pivots
around the rear mounting pin. When filling the bin in the field, it rests on the rear pivot
hinge, as well as the hydraulic cylinder and a front rest. To ensure the bins do not
return too quickly a 2 mm orifice is installed in the hydraulic coupling feeding the
cylinder. The central collection bin arrangements dump the harvested material by
sliding the box assembly out the side, where a chain pushes the material out (right
image of Figure 10).
On some models a side discharge conveyor may be used to unload the grapes while in
motion. When the side discharge conveyor is not unloading, all grapes are shifted to a
side collection bin. These conveyors use a belt similar to the cross conveyors seen in
Figure 9, although much longer. Conveyor speed may be hydraulically varied, as well as
reversed to meet harvesting conditions, as well as to remove blockages.
e
R
è
?«.
TM
Figure 11: Side discharge conveyor on a Grégoire G108 harvester (Kverneland Group,
2006).
This long section of conveyor belting (Figure 11) was the installation location for
HarvestMaster™ and Farmscan™ yield monitors. These yield monitors required the
22
long length of unsupported belting that could only be found of this model of harvester,
which is not commonly found in the Niagara wine region.
23
Chapter 4
4.0 Project Objectives
4.1
Grape Harvester Yield Monitoring Prototype
The purpose was to develop and test a functional grape harvester yield monitor
prototype. This prototype was developed for a Grégoire™ G170 grape harvester picking
head featuring the central bin collection arrangement. This machine and configuration is
the most popular in the Niagara wine region. The prototype was designed to be easily
and quickly constructed, at a minimum of development cost for retrofitting to existing
harvesters. Custom designed and hand build circuitry was developed, however easily
obtainable and cost effective components and materials were used when possible for
construction of the prototype.
The scope of this project did not include the interfacing of the yield monitor prototype
with a GPS system. As such, yield maps created with GPS coordinates will not be
created, but estimates based on average harvester speed and row length were used to
generate yield maps.
4.2
Static and Dynamic Calibration of Yield Monitoring Prototype
The yield monitor prototype was evaluated by testing it in static conditions, as well as
through dynamic calibration trials. This testing was performed to determine how
sensitive the system was to the desired mass flow rate information.
24
Likewise, the quality of the signal was analyzed to determine if it can be improved
through signal processing techniques, as well as through design iterations. The static
and dynamic calibrations were applied to field testing data.
4.3 Field Test Yield Monitoring Prototype
The prototype was tested on a Grégoire™ harvester in field conditions to evaluate
numerous criteria including accuracy, mechanical performance and foreign material
buildup.
4.4 Provide Design Recommendations for Future Development
Recommendations for further development were presented based on the results
obtained through construction, calibration, field trials and subsequent analysis. These
recommendations indicate the main limitations of the system, and briefly present
solutions that address these limitations.
25
Chapters
5.0 Yield Monitor Design Requirements
The following constraints were placed on the yield monitor design and limited the
application of some measurement methods and designs. These design expectations,
constraints and criteria provided direction while designing the prototype yield monitor.
5.1
Constraints
It was decided that to provide the most accurate and usable yield monitor prototype for
use in the Niagara region viticulture industry, the system must meet the constraints as
identified below:
? Yield monitor must be easily installed within the existing infrastructure without
permanently changing the harvester
? Yield monitor must be interchangeable between different models of Grégoire™
harvesters commonly found in the Niagara region with a minimum of
modifications
? Yield monitor prototype must function and not impede harvester operation in
any typical field conditions
? Sensors and yield monitor design must be robust, and provide accurate and
sensitive mass measurement data for further processing
? System must be able to provide instantaneous and totalized yield information
? System must function for an extended period of time in a harsh operating
environment.
26
5.2 Criteria
The following criteria were used to create a design as applicable as possible. The
decision was made to design the yield monitor to increase marketability of the design by
limiting the installation complexity and finding cost effective solutions.
? Yield monitor design should be as cost effective as possible
?
Provide as clean and sensitive measurement of massas possible
? The more easily installed and removed from the harvester the better
? Use simple and effective instrumentation and control systems
? Allow for easy cleaning while providing protection to sensitive components.
5.3 Design Requirements
>
The yield monitor was expected to be exposed to harsh operating conditions and, as
such, must be a reliable and robust design that could be easily cleaned and serviced. The
expected grape material flow rate flowing through the harvester was predicted to be
approximately 3.4 kg/s, or 1.7 kg/s carried by each main conveyor belt (see section
5.3.1). The more accurate the measurements, the better the quality of the continuous
yield information. The accuracy of the accumulated mass measurement utilizes the
continuous yield measurements and therefore, obtaining the greatest accuracy was the
paramount design requirement.
5.3.1 Main Conveyor Belt Mass Flow Rate
The instrumentation used for measuring the mass flow rate of the grape material was
required to be sensitive enough to accurately measure the changing forces resulting
from variations in yield within the vineyard, these present themselves as variations in
27
this flow rate. The maximum mass flow rate of grapes was used to determine the
expected loading range that the system would experience, in order for the
instrumentation to be scaled to maximize sensitivity.
An estimation of the maximum mass flow rate of material was made based on known
factors, such as maximum yield and average harvesting ground speed. The following
calculations were preformed based on the information presented in Table 1 below.
Table 1: Grape yield, vineyard and harvester productivity, trellis spacing and conversion
factors.
Maximum Estimated Yield per Acre
9000kg/acre
Harvester Ground Speed
1 .4mph
Trellis Row Spacing
Acre Conversion
8 feet
1 acre = 43560 square feet
Mile Conversion
1 mile = 5280 feet
Driving length to harvest 1 acre:
(Square Feet
Length =
I
/ acre
Width of Row
Equation 1
(43560sq.fi/
[
/ 1 acre ,
Length = ±
8ft
L = 5445ft
Harvesting Speed:
,
Harvesting Speed = (Harvesting Ground Speed)?(Length of Mile) x\ \hr/^QQ5
Harvesting Speed = (l.4mph)x(52S0ft / mile)x\
28
Equation 2
Ihr,
'360Os = 2.05./? /sec
Time required for harvesting this area:
Time(s) -
Length of Row
Harvesting Speed
Equation 3
Time{s) =
— = 265 Is
2.05ft /sec
Mass flow rate based on the above information:
Flow Rate =
Total Harvested Mass
Time
Equation 4
FlowRate = 2651.785
900Q^ = 3A0kg/s
Each harvester features two main conveyor belts to carry the fruit into the storage bin,
and as such each handles approximately 1.70 kg/s.
Linear speed of the main conveyor belt:
LinearBelt Speed = (Pulley Circumference)?(Pulley RPM)x{^m^y^s)
Equation 5
The normal rotational speed of the drive pulley is 125 rpm, with a diameter of 25.4cm.
LinearBeltSpeed = (0.7986m / rev)x(125revImin)?(lmi%¡0sec) = 1-66m/s
The estimated linear speed of the main conveyor and the material was 1.66 m/s.
5.3.2 Harvested Wine Grape Material Properties
Figure 12 illustrates the physical characteristics of the harvested material after passing
through the harvester. The free running juice is often the highest valued component of
the harvested material, and the harvesters were designed to collect and hold as much as
possible to limit losses. Depending on the condition of the grapes, free running juice
can be a significant percentage of the harvested material. Besides juice (C), other
29
materials often pass through the harvester including leaves (A), stems (B) and vines. On
average, the majority of the discharged material was individual grapes that had been
removed from the stems on the vines (D).
**'·*?^'1
Figure 12: Harvested grape material included leaves (A), stems (B), free running juice (C)
and singular grapes (D). Chardonnay grapes harvested from William Falk Farms in
September of 2008.
When dried, the juice from the grapes causes significant friction and sticking between
components of the harvester, such as pulleys and belts. The grape material can be
caustic as well and, after prolonged exposure, unprotected materials degrade. Any yield
monitor prototype that came into contact with the harvester grape material needed to
30
be designed to function despite this harsh operating environment, where sticky grape
material can cause mechanical clogging or binding.
5.4
Yield Monitor Instrumentation Sites
Self propelled grape harvesters are very compactly built machines designed specifically
to function in the tight confines of a vineyard. Locations where the flow of harvested
grape material could be intercepted or diverted to be weighed were limited due to the
lack of available space. Four locations were identified; the discharge area of the main
conveyor belt, grape storage bin, main conveyor drive system, and horizontal cross
conveyor unloading area.
5.4.1 Grape Storage Bin
Obtaining the weight of a storage bin while being continuously filled would require that
the harvester be mechanically isolated from the bin. The harvester and bin would only
be connected by load cells measuring the empty weight of the bin, as well as any
material in it. The Grégoire™ grape harvesters in the Niagara region were not
configured for this type of weighing system because of the complexity of the bins. To
aid in unloading, chain augers and false bottoms are used to push material out while
dumping, a process which involved sliding the whole assembly out to the side (Figure
10). Mechanically isolating this system would have involved significant, irreversible
changes to the harvester infrastructure that would not be applicable to other picking
head assemblies.
31
Although this method could have provided accurate measurements of totalized yield,
instantaneous yield measurements would have been of poor quality and subject to a
large amount of error. The load cells weighing the storage bin would be scaled for full
load conditions with a factor of safety to prevent overloading due to inertial loading and
over ranging. The sensitivity to small continuous changes in mass would be reduced,
while minor acceleration changes due to rough terrain and vibration would induce
fluctuating and erroneous mass measurements.
5.4.2 Cross Conveyor Unloading Site
The grape material discharge of the horizontal cross conveyors was a potential location
for mass measurement because of the clean and predictable movement of material.
Although the material flowing through this location would have been beneficial for mass
measurement, there were numerous limitations to the applicability of this location.
There was little space underneath the discharge due to augers and grape spreaders that
distribute material throughout the bin. Different harvester picking head models also
had drastically different horizontal cross conveyor arrangements, this would require
greater modification of each system to meet the requirements of specific models.
The discharge velocity of the grape material from the horizontal cross conveyor was
substantially greater than the release velocity of the main conveyor discharge. The
placement of an impulse measurement plate would have had adversely affected the
quality of the material due to the impact between the plate and the soft material of the
32
grapes. Accumulation and liquid mass measurement from juice would not have been
possible, potentially resulting in a significant reduction in yield measurement accuracy.
5.4.3 Main Conveyor Belt Tension Load Cell
Figure 13 below illustrates the drive and belt tension mechanism for the main conveyor
belt. To maintain proper tension to drive the belt, there was a spring tension
mechanism (F) mounted below the large chain driven sprocket (B and C). As the belt
became more heavily loaded with grapes, the main pulley assembly pivoted on a bolt
behind the back of the main conveyor chute (D). This increases the force being exerted
on the spring and measured by a pancake load cell (G).
-'fri fit IH
Figure 13: Main conveyor belt drive and tension spring assembly. Note the installation
of a pancake load cell under the drive belt tension spring (G).
(A)
Main Conveyor Belt
(E)
33
Conveyor Drive Motor
(B)
(C)
Belt Drive Pulley
Drive Chain
(D)
Tension Pivot Point
(F)
(G)
Tension Spring
Pancake Load Cell
Preliminary testing was performed on this pre-tensioned load cell configuration during
the ice-wine harvest at William Falk Farms in the winter of 2008. The pancake style load
cell mounted between the spring and the mounting bracket (G) measured the force
exerted by the pre-tension spring, and any increase in force caused by changing loading
conditions. The large pre-tension load applied by the spring to allow the pulley to drive
the belt without slippage required the use of a load cell with a 225kg capacity to prevent
overloading and failure.
These tests indicated that the sensitively of the load cell was too low to accurately
measure the grape material as it passed along the conveyor, little change from the pretensioned load was seen. This could have been due to the low yields associated with
ice-wine harvest, but primarily due to the mechanical configuration of the main drive
assembly, pivot point and where the grape material was carried and discharged from
the main conveyor belt.
5.4.4 Main Conveyor Discharge Installation Site
After careful analysis of all of the available locations with a universal installation in
mind, the area below the discharge of the main conveyor belt proved to be the best
location for the installation of an aftermarket yield monitoring system (Figure 14).
34
¦a
25cm
1
27cm
62.5cm
56.5cm
15cm
51cm
X
^
I - .'
. -
39k.
:
~?
"?
¿sr
J-
—^JMMi
Figure 14: Site of the cross conveyor unit and the installation site for the yield monitor.
This location was the only common area to all Grégoire™ harvester models with
adequate although limited space. Indicated were the general dimensions of the cross
conveyor unit.
This location provided the most space available for the installation of a yield monitor
prototype, as well as having direct access to the product flow. Although this location
presented the greatest potential, there were still serious design and application
limitations present. The grape material was normally. discharged as a wide stream of
material from the main conveyor belt, and flung onto the smaller cross conveyor
featured in the above photograph (Figure 14). Although this area provided the most
space and the best accesses to material flowing through the system, the dimensions
presented in Figure 14 illustrate that any yield monitor installed in this location would
have to fit within an area only 51 cm long by approximately 50 cm wide and as vertically
35
'
compact as possible to prevent impacting of the main conveyor belt buckets on any
aftermarket additions.
To ensure that the flow of material was maintained in a continuous fashion, a design
that did not compartmentalize the grape material was deemed necessary.
Measurement systems that stop the flow of material often result in clogging and poor
mechanical performance in environments such as a grape harvester. An open system
which allowed for an overflow of material was necessity to ensure uninterrupted
performance even at the expense of mass flow measurement accuracy. The yield
monitoring system was required to maintain harvester functionality at all times.
From an aftermarket installation point of view, this location proved to be ideal. The
main conveyor belts on all styles of Grégoire™ harvesters discharge at the same
location and in very similar manners. Since this location was standard, any drop-in after
market yield monitor should be easily reconfigurable to work in central bin, side bin,
discharge conveyor or any other picking head model types.
5.5 Installation Site Infrastructure Limitations and Design
5.5.1 Cross Conveyor Mounting
The location and design of the cross conveyor frame was common to all central bin
"Grégoire™ harvester modules, and very similar to other configurations. It was
determined to be advantageous from an aftermarket product point of view that the
yield monitor prototype be easily installed into the harvester and removed if necessary,
36
without permanently changing the harvester. The hopper frame for the cross conveyor
allowed the system to be clamped onto the outer rims, providing easy drop in place
installation. Therefore, the system was designed to fit within the space of the hopper,
while mounting onto the top rim of the hopper frame (Figure 14).
The requirement for an open system that was low profile, and compact enough to be
installed into the hopper limited potential designs. The use of refined types of
technology that had been used successfully on grape harvesters was also desired. A
mechanically suspended low profile conveyor belt suspended on three load cells was
conceived. The inner frame of the yield monitor was designed to attach to the cross
conveyor frame and to the stationary end of load cells. The outer frame with all of the
suspended conveyor components attached to the active measurement end of the load
cells. These load cells were used to record the weight of everything attached to them,
including the suspended conveyor belt system, as well as any grape material on it.
Due to the weight of the mechanically isolated conveyor on the load cells, a static initial
loading was applied. In order to increase sensitivity and reduce inertial loading from this
mass, the yield monitor was required to be as light as possible while being robust and
functional. The instrumentation for the load cells was also designed to tare out as
much of this initial static loading in order to increase the sensitivity of the system to
mechanical loading.
37
5.5.2 Yield Monitor Power Source
For preliminary testing it was decided to use an easily obtainable 12V dc gear head
motor to power the conveyor belt on the yield monitor prototype. The harvester
electrical system was a 12V dc negative ground system that featured auxiliary power
plugs for added aftermarket equipment. This power supply was readily available, and
provided up to 15A of fuse protected power. The use of a 12V gear head motor allowed
for easy installation and development of the drive mechanics and electrical system of
the yield monitor prototype, which was much easier than using a hydraulic system, at a
fraction of the cost. The selected gear head motor, commonly seen in windshield wiper
applications, featured two separate power lines for two different speed settings,
allowing more operational variability during testing and development.
5.5.3 Yield Monitor Conveyor Assembly and Discharge Rate
It was decided to construct the yield monitor assembly out of light weight thin gauge
steel for manufacturing and cost reasons. The purpose of the first prototype was for
preliminary testing, so the use of corrosion resistant materials was unnecessary. The
yield monitor conveyor belt assembly was designed to be easily constructed using a
minimum of machining and complex tooling, to aid in manufacturability and repair. The
yield monitor prototype was designed to use common components such as flange
bearings, load cells, belting and drive sprockets.
The belting used for the yield monitor conveyor was selected to be a food grade silicone
coated belt with a stainless steel alligator lace for splicing. This standardized, readily
38
available belting was selected because it allowed for easy removal, and good drive and
tracking performance. The conveyor belt had a molded A- series V-groove in the middle
for better belt tracking and driving, which meshed with A- series V-groove pulley inserts
in the drive and idler rollers.
The discharge rate of the yield monitor was designed to be higher than the rate at which
it was discharged. If this criterion were not met, the discharged material would build up
and cause clogging. If the conveyor discharged the loaded material too quickly, the
amount of accumulated mass would have decreased and the sensitivity and accuracy of
the measurement would have been reduced due to a lower magnitude of signal and a
less efficient use of the full span of the measurement instrumentation. Careful
consideration towards expected yield and discharge rate of the yield monitor prototype
were given to maximize sensitivity while trying to prevent clogging. A discharge rate of
approximately 2.0 kg/s was selected to allow a factor of safety of 200%, based on the
maximum predicted flow rate of material through the harvester. Further calculations
can be seen in section 6.3.1 Static Loading Conditions and Load Cell Loading Range'.
5.5.4 Sanitary Design Considerations
The buildup of grape material can have many negative effects on the device from a food
safety perspective. The buildup of material can increase the potential for mold, mildew
and fungus growth (Sanitary Design Organization, 2008). As such, the buildup of this
material can cause health concerns to un-cleaned material. The buildup of material can
also have detrimental effects of the mechanical performance of a system. There are
39
numerous design features that should be incorporated into the design and
manufacturing of the system to prevent this buildup of material. This would improve
the overall mechanical performance and measurement accuracy of the system.
The following guidelines were obtained from the Sanitary Design in Food Equipment
organization, an organization that works to educate equipment designers about sanitary
design for food processing environments and serves as a guideline when designing
equipment (Sanitary Design Organization, 2008).
1. Cleanable to a microbiological level
2. Made of compatible materials
3. Accessible for inspection, maintenance, cleaning and sanitation
4. No product or liquid collection
5. Hollow areas should be hermetically sealed
6. No niches
7. Sanitary operational performance
8. Hygienic design of maintenance enclosure
9. Hygienic compatibility with other plant systems
10. Validate cleaning and sanitizing protocols.
40
Chapter 6
6.0 Mechanical, Instrumentation and Experimentation Design
6.1
Suspended Conveyor Assembly
The design presented below (Figure 15) illustrates the layout of the prototype for
measuring the mass flow rate of grape material as it continuously flows through the
grape harvester. The belt has been omitted to show inner components.
B
t.
L·
¡e>
T
H
Figure 15: Illustration of the suspended conveyor prototype. Top image illustrates the
numerous internal components. Bottom image illustrates the fully assembled prototype.
(A)
(B)
(C)
Gear Head Drive Motor
Outer Conveyor Frame
Inner Conveyor Frame
(E)
(F)
(G)
Conveyor Belt Roller
Top Loading Plates
Belt Tension Mechanism
(D)
Load Cell
(H)
Roller Chain Drive Sprockets
41
A fully suspended conveyor belt arrangement was conceived after much design work to
fit within the very limited confines of the preexisting harvester infrastructure. This
system was robust, reliable, easily reconfigurable to different size limitations or
locations, as well as being less prone to the buildup of material and clogging. The
configuration seen above was designed to fit within the horizontal cross conveyor chute
and intercept the flow of grape material, while maintaining flow in the same direction as
the machine was already designed.
The suspended weighing conveyor functions by intercepting the flow of material off of
the main conveyor belt as it would be normally discharged (Figure 16). The material
was supported on the conveyor belt encircling the entire unit and frame, and rotated by
the gear head motor to continuously discharge material, as indicated below.
Main Conveyor Discharge
F:
Yield Monitor Discharge
Figure 16: Direction of grape material flow over the prototype yield monitoring system.
42
The load cells mounted within the frames measure the mechanical loading in the
vertical direction, and therefore they measure a component of the impact force of the
grapes, as well as the static weight of the held material prior to being continuously
discharged.
6.2 Suspended Conveyor Hopper Assembly »
To ensure the proper flow of the harvester material as it impacted and was discharged
from the conveyor belt, a hopper was installed to confine the material and direct flow.
The hopper illustrated in Figure 17 was designed to direct the flow of material to
improve the measurement and material flow performance of the yield monitoring
system. This simple hopper prevented material from flowing off the back end as it was
redirected perpendicular to its original unloading direction.
The hopper was designed with flexible molded polyurethane sides that were used to
prevent damage to the harvester if any component were torn free. This was necessary
due to the very limited space available for installation, and the proximity of the main
conveyor belt to the hopper.
43
(I)
Figure 17: Grape harvester yield monitor prototype hopper arrangement. The metal
frame (A) and flexible shielding (B) are illustrated. The hopper assembly bolts onto the
outer frame of the yield monitor prototype.
6.3
Mechanically Isolated and Suspended Conveyor Configuration
In order to measure the mass of material on the suspended conveyor at any given time,
the system must be fully suspended and mechanically isolated from the inner frame,
and any portion of the harvester to which it would be attached. Figure 18 below
illustrates the active, suspended components of the design, as well as the inactive main
inner frame. The image on the left is a bottom view of the suspended conveyor, with
the three active load cells in red. The image on the right is a top view of the conveyor,
and the area that is loaded by the falling grape material. Therefore, any grape material
that may fall onto the areas indicated in blue were be recorded by the load cells
mounted within the conveyor frame.
44
A three load cell configuration was selected over other configurations to provide the
fully suspended configuration, while not over constraining the system and inducing
forces through improper mechanical connections and twisting from installation, while
reducing instrumentation cost and complexity.
1
¦
©
Figure 18: Illustration of active and inactive components of the weighing system. Left:
Bottom of the prototype illustrating active area (B), inactive area (C) and load cells (A).
Right: top view of the active components that would be subject to grape material flow.
6.3.1 Static Loading Conditions and Load Cell Loading Range
Since the weighing conveyor was completely suspended on three load cells, these load
cells would experience an initial static load even when there was no product flow. This
static loading between the cells was uneven as well, primarily due to the location of the
gear head drive motor. Knowing the initial loadings on each load cell was necessary to
re-balance the bridge and increase the readable range of forces, as well as to increase
the sensitivity of the readings by properly scaling the output signals to the analog to
digital converter input range.
45
All components with the exception of the drive assembly had a centre of mass at
approximately the centre of the conveyor, and weighed approximately 10 kg. The drive
assembly weighed approximately 3 kg, and was offset by the amounts indicated in
Figure 19. Total weight of the suspended conveyor platform was estimated to be 13 kg.
?
a
m
B5
®
i
i
tí
C
©H1 6.5cm
21cm
7cm
3 12cm
?
O
4cm
®
tt
©
Figure 19: Identified are the main components of the monitoring system, and the
loading points of Load Cell A (A), Load Cell B (B), Load Cell C (C), and the centre of mass
(D).
Figure 19 is a top view of the entire conveyor system, and indicates the location of the
three cantilever load cells, conveyor belt rollers, mounting brackets, the centre of mass
for the system, as well the point of load cell force measurement.
46
The centre of mass was therefore offset more over one load cell then the others,
indicating that this load cell would experience significantly different mechanical
loadings. The actual distribution of forces was calculated using the information in Figure
19. The list of moment equations, as well as the resulting calibration equation follows.
^Mc y =0 = Fc(0cm)+FB(2\ + I6.5cm) + FA(2\-4cm)-(m.5N)(2lcm)
Equation 6
£ MFrty = 0 = -Fc (2 1cm) +FB (1 6.5cm) - FA (4cm)
Equation 7
S MAx = 0 = -Fc (7 + 12cm) +FB (7 + 12cm) + (127.57\T)(12cm)
Equation 8
By substituting the above equations into a matrix and solving, the following calibration
matrix was obtained.
^C
^B
^A
^Load
^C
^B
?A
^Load
0
37.5
15
-2678.13
1
0
0
-30.428ÌV
-21
16.5
-4
0
0
1
0
-50.1 177V
-19
-19
0
1530.36
0
0
1
-46.9857V
Equation 9
FA, Fb, and Fc were load cell A, B, and C, respectively, and FL0ad represents the unloaded
static weight of the suspended conveyor system, and Mc,y, MFiV and MA,X are moments
caused by the applied forces about each identified location.
The distribution of these forces can be seen in Figure 20 below. The load cell (Fc) closest
to the centre of mass experiences a higher static loading then the rest of the load cells in
47
the system. Due to the disproportionate loading between the load cells, Load Cell C
with a static loading of 5.1 kg will be considered the worst case static loading scenario.
Load Cell (A)
3.1kg
J-
Load Cell ©
4.78kg
Load Cell (B)
5.1kg
Figure 20: Static distribution of weight over the load cells when the conveyor is
unloaded. Note that Load Cell B would experience a higher static loading then the
others.
The unloaded static loading of 5.1 kg, as well as a maximum predicted loading of 3.0kg/s
of grape material flowing through the system, when assuming a uniform, continuous
discharge rate, any load cell with at least this capacity would provide adequate
measurement range and overload protection.
In order to adequately discharge the grape material from the surface of the conveyor,
the drive motor on the conveyor was set to a constant speed of 15 rpm, but geared in a
5/6 ratio, corresponding to a roller speed of 12 rpm. The roller diameter was 7.62 cm,
resulting in a linear belt speed of 5.0 cm/s. Assuming that all grape material fell on
average at the middle of the belt, it would have to travel 13 cm to be discharged. The
average hold up time for the grape material on the conveyor would be approximately
48
2.6 seconds.
The two main conveyor belts carry the fruit at a flow rate of approximately 3.40 kg/s, or
approximately 1.7 kg/s of material per conveyor. With a hold up time of 2.6 seconds and
a flow rate of 1.7 kg/s, the average loading on the yield monitor conveyor belt would be
approximately 4.42 kg. The inertial loading from falling grape material impacting on the
top surface would amount to approximately 61% increase in mass (section 6.4.2
Projected Grape Material Impact Loading'). The yield monitor would be
subjected to an approximately loading of 7.1 kg. Given these estimates, Load Cell B
would experience a projected total loading of 12.2 kg.
Due to the predicted loading of approximately 12 kg on the load cells from both the
static weight of the yield monitor conveyor and frame, as well as grape material, load
cells with at least this load measurement range were necessary. A cantilever 20kg load
cell met this loading criterion, allowing for a significant factor of safety (166% overload
protection) while providing the necessary measurement sensitivity.
6.4 High-speed and Static Weighing using Load Cells
__
Harvesting equipment for non-grain crops often lack a convenient location for installing
a sensor. Other factors that contribute to the absence of alternative crop yield monitors
include large variation in crop size or shape, and harsh conditions within the harvester.
One proposed solution was to use load cells to weigh the collected crop during harvest,
49
but excessive noise and inadequate sensitivity have prevented the realization of this
technique to date (Durrence eí al, 1999).
Load cell yield monitoring systems utilize load cells in primarily two ways, as a static low
speed weighing process, and as a high-speed weighing method using the impact of the
material. Current grape yield monitoring systems rely on the steady state method of
measurement, where the applied load represents a step response of the force
transducer to determine the weight of the material on the belt surface. This static step
response method does not provide the most accurate measurement of mass when the
material being weighed is not static, but in motion.
Other yield monitoring systems, particularly those developed for cereal grains rely on
the impact impulse force of the material being weighed to provide the necessary
information about yield. All systems that use force transducers must contend with
dynamic loading conditions if the material is measured in a dynamic, continuous fashion
(Gilman & Bailey, 2005; Piskorowski & Barcinski, 2008).
Force transducers typically have long settling times and are prone to vibration created
by conveyor belting or anything to which they are attached (Gilman & Bailey, 2005;
Piskorowski & Barcinski, 2008; Woods & Lawrence, 1997). It is possible to use adaptive
filtering to decrease the effect of external vibration, but there are benefits in some
systems to eliminate this vibration by making the weighing table motionless. If this is
50
the case, loading and unloading of items to record the step response becomes
problematic. This was due to the increase in mechanical complexity required to weigh a
continuous flow of material into discrete measurements. This increased the risk of
adverse mechanical performance, complexity, cost and space requirements.
6.4.1 Dynamic Response Characteristics of a Load Cell Force Transducer
A load cell serves as the reaction for the applied load and focuses the effect of the load
into an isolated, uniform strain-field (Gilman et al. 2005). In a cantilever load cell, the
spring element is a part of the cantilever itself; where the beam stores elastic potential
energy as the live end is deflected from the equilibrium position.
Since the deflection is small, the live end of the load cell can be assumed to travel
linearly, perpendicular to the measurement axis of the cantilever load cell, causing a
vertical deflection. The simple equation defined by Hooke's law applies for static loading
conditions while at equilibrium. The mathematical derivations pertaining to this analysis
were derived by Gilman & Bailey. (2005). Nomenclature is as follows:
M
m
ks
kD
Suspended Mass
Impacting Mass
Spring Constant
Damping Constant
?
J
i
fD
fs
??
vM
Spring Force
f(t) External Excitation Force
Dampening Ratio
?
Natural Dampening Frequency
Velocity of Suspended Mass vm Velocity of Impacting Material
e
d
Coefficient of Restitution
Height of Falling Object
g
Acceleration due to Gravity
vm
tT
Displacement
Impulse
Change in Displacement
Dampening Force
Material Separation Velocity
Duration of Object Freefall
51
fs = -KX
Equation 10
Dynamic loading conditions fundamentally change the loading characteristics, and the
damping of the load cell must be considered, as seen below (Equation 11).
Jd ~ kDsx
Equation 11
Using Newton's 2nd law of motion, the equation representing the movement of the live
end of the load cell in response to an external excitation force f(t) (Equation 12).
Mx(t) = /'(/) - ksx(t) - kDx(t)
Equation 12
applied load
f(t)
M
kDx
ksx
damping
spring
Figure 21: Free- body diagram of the live end of the load cell (Gilman & Bailey 2005).
Where the under damped natural frequency, and the damping ratio is as follows in
Equation 13 and Equation 14 below.
µ _ ^D
hL —
IM
Equation 13
? =
\ZS___ZD_
M
AM2
Equation 14
52
6.4.2 Impact Weighing and Momentum Transfer
The previous section describes how the weight of an object placed onto the load cell
would be measured by evaluating the step response or the difference between before
and after placing the object. After the transient responses have decayed, the step
response output is proportional to the input load.
In some industrial applications and indeed for this yield monitoring project, a situation
exists where the object being weighed is dropped onto the load cell. An impact occurs
when the object comes onto contact with the live end of the load cell, where the load
cell and the object exert relatively large forces on each other, but only for a short time.
Although complete understanding of the impact requires an extensive analysis of the
contact forces, the stresses, and deformations of the colliding bodies, the principle of
conservation of momentum can be used to obtain a good approximation to the impact
event.
A number of simplifying assumptions can be made to keep the analysis straightforward,
including that the object is assumed to rebound and travel in the opposite direction
after the collision with the load cell. The velocity at separation is also assumed to be a
fixed fraction of the velocity of the approach, and the initial velocity of the load cell is
assumed to be negligible.
53
Approach
Separation
04?
(?t)??„
M
i*
M
An,
Figure 22: Collision between the live end of the load cell and the equivalent mass, M.
Illustration of the approach and separation of the objects (Gilman & Bailey 2005).
The coefficient of restitution, e, (Equation 15) represents the fractional value
representing the ratios of the velocities before and after the impact. This coefficient is a
function of the final velocity after impact of both objects, and is a material property of
the falling object.
vM -vm =e(ym-0)
Equation 15
In order to simply the system, the impact interval was assumed to be sufficiently short,
and due to this short duration, the effect of gravity of the velocity of the object
impacting the load cell can be assumed negligible. By assuming the effect of restoring
the damping forces on the velocity of the load cell were negligible, and the position or
orientation of the object or the load cell has no significant change during the impact, the
following calculations apply.
As the system is isolated and closed, the conservation of the linear momentum gives the
following:
mv„ + 0 = mv„ + Mv M
Equation 16
54
Where m is the mass of the falling object, vm is the velocity of that object, and M is the
suspended mass of the system.
Solving for the above equations for the final velocities gives the following equations:
?
m
(m-eM)-v
= -
TL r
m+M
m
Equation 17
m(\ + e)
m+M
Equation 18
Therefore, the impulse, J, due to the impact force acting on the live end of the load cell
by an object is equal to the change in momentum of the load cell:
mMvm ,„
J = MvM= m + M
f-QL + e)
Equation 19
The coefficient of restitution, e , is an index of the degree to which object and load cell
recovers from deformation due to a collision. It is frequently considered a constant for
given geometries and a given combination of contacting materials, but in reality, it also
depends on the impact velocity, as it approaches unity when the impact velocity
approaches zero. However, if the impact velocity is above a certain threshold, the
restitution coefficient can be considered to be independent of it.
Rearranging Equation 19 to give the mass of the object being measured as a function of
the impulse J and the impact velocity vm gives the following (Equation 20):
m(J,vm)
MJ
M(l + e)vm-J
Equation 20
55
A number of approaches can be used to determine the velocity of a falling object, and
the simplest method is to measure the time it takes for an object to fall a known
distance prior to impact. Assuming an object is being accelerated solely by the force of
gravity, the impact velocity becomes a function of tT and d, as indicated in Equation 21
below.
,
^
d
l
Vm(tT,d) = y + -gtT
Equation 21
These derivations are mainly valid for situations where the impact loading far exceeds
the static loading of the material on the load cell. This type of impulse mass
measurement is used on most commercial cereal grain yield monitors, where the
release velocity and rate are known, quantifiable variables. Impulse forces that are
overlaid on primarily static systems are typically eliminated by applying linear offset
corrections.
6.4.2 Projected Grape Material Impact Loading
Based on the equations presented in section 5.3.1 Main Conveyor Belt Mass Flow Rate
the impulse, J1 was calculated as follows. The mass impacting the top surface of the
conveyor was calculated below (Equation 22) as the discrete amount of grapes that
were discharged from the main conveyor belts molded buckets given the linear speed of
the belt, bucket spacing and flow rate of material. It was assumed that the discharge
would be in a pulsatile form, given the arrangement of the main conveyor belt buckets
and the assumption that this material would be striking the surface of the yield monitor
as one mass.
56
Mass
(Belt Bucket Spacing)(Flow Rate)
{Belt Speed)
Equation 22
(0.2 w)(l
.7 ^) ._„,
- — = 0.205£g
Mass = -
(1.66-/)
J = MVM=^(l +e)
m+M
j = MVm = (0-205^)(13^)(1.66^)
(0.205 ¿g) + (13%)
335
It can be assumed that the coefficient of restitution for grapes approaches zero, in much
the same way the coefficient of restitution for tomatoes was found to have little effect
on the impact force of a tomato yield monitor developed by Upadhyaya et ai. (2002). In
the case of a grape yield monitor, the coefficient of restitution is not just a material
property, but is a function of the initial impact direction. Since the discharge impact was
positioned at an angle to the vertical measurement direction of the load cells, this
impact impulse force is further reduced. The impacting grape material would also be
cushioned by material already present on the surface of the yield monitor conveyor
belt, further reducing the impact force. A suspended mass, M, of 13 kg was assumed for
the yield monitor apparatus attached to the load cells. Utilizing Equation 22, this
corresponds to an approximate loading of 1.61 times the true mass of the weighed
material due to the momentum of the grape material being transferred to the weighing
apparatus.
57
Since the momentum of the grape material increases the mass of the system by only a
projected 61% and as such represents a relatively small increase in the overall mass of
the system. A static loading step response method of material weighing appeared to be
more applicable then an impulse mass measurement method. This said, corrections for
the impulse energy transfer had to be made.
6.5
Constructed Prototype
Figure 23 is a photograph of the manufactured version of the grape harvester yield
monitor installed into the Grégoire grape harvester. This system was constructed in the
same configuration as was presented in previous sections. The total weight of the
installed system was 18.45 kg, with a total suspended mass of 13.65 kg. This was similar
to the predicted weight of 13 kg, indicating that the calculations performed above were
valid.
58
^
fc;'aJ^^B
em
m»
ß,
^¡ÉWm
rj!
ri
^
m
TM
Figure 23: Grape yield monitor prototype installed on a Grégoire harvester. This
configuration was used for final field testing and dynamic calibration testing.
6.6 Instrumentation Design
The following sections detail the design of the instrumentation that was used in the
prototype. Much of it was custom designed and constructed to meet the demands of
the system, as well as to serve the demands of the rapid and cost effective prototyping.
6.6.1 Omega Engineering LCAE- 20Kg Load Cell
Based upon the previous mechanical loading calculations, it was determined that 20kg
load cells would provide sufficient loading range, as well as overload protection. Three
Omega Engineering LCAE 20kg4 range aluminum cantilever load cells were selected for
this prototype (Omega Engineering, 2008). These inexpensive load cells provide up to
4 Omega Engineering. Omega Engineering LCAE-20kg load cell. Laval, QUE, CAN.
59
20kg of loading, while having an ultimate overload protection of 200% (Omega
Engineering, 2008). The load cell, illustrated below in Figure 24 contains four tapped
holes for mounting to external frames. These load cells come with a four conductor
cable with two excitation wires and two signal wires, as well as a metal shielding wire
(Omega Engineering, 2008).
Figure 24: Omega Engineering LCAE-20 kg load cell. Note the tapped mounting holes
and water resistant epoxy coating.
These load cells incorporate a full Wheatstone bridge arrangement with a nominal
resistance of 350 ohms, and a maximum excitation voltage of lOVdc. The following are
the specifications for each load cell, as provided by the manufacturer (Table 2) (Omega
Engineering, 2008).
Table 2: Operating specifications for the Omega Engineering LCAE-20kg load cell.
Full Scale Output: 2mV/V +/- 10% Zero Balance: +/- 5%FS
Excitation: 10Vdc (15Vdc max)
Creep/Creep Recovery: 0.02%FS
Non-Linearity: +/- 0.01 5%FS
Operating Temperature: -10 to 5OC
Hysteresis: +/- 0.01 5%FS
Non-Repeatability: +/- 0.02%FS
Compensated Temperature: -10 to 5OC
60
A standard load cell signal processing methodology was used to amplify the Wheatstone
bridge signal generated by the load cell. The load cell signal was amplified via an
instrumentation amplifier, which fed directly to an analog to digital converter. Simplicity
in this design was selected over more complicated methods due to reliability, cost, and
the fact that this method simply worked the best, along with allowing for post
processing filtering. The following sections detail each component of the design.
6.6.2 Load Cell Circuitry
The Omega Engineering LCAE-20 load cell has a maximum excitation voltage of 15 volts.
The -5V and +5V power supply was connected to the red and white power supply wires
of a standard switching mode power supply, DVE model numberDSP-1454P5 (see
Section 6.6.6 Instrumentation Power Supply System and Regulation). Figure 25 is a
schematic representation of the strain gauges mounted inside of the cantilever load cell.
The two output wires, Green (Signal +), and Blue (Signal -) were connected to the
amplification board and to each instrumentation amplifier. The ground shield wire was
connected to the box to prevent ground loops and to short electromagnetic
interference (EMI) to ground (Webster, 1999).
; Dee Van Enterprise Co, LTD. Fremont CA, USA. DSP-1454P Switching Mode Power Supply.
61
R- AR
R+AR
R+AR
R- AR
Figure 25: Wheatstone bridge strain gauge configuration used in the Omega Engineering
LCAE- 20kg load cell (Omega Engineering, 2008).
6.6.3 Amplifier Circuitry and Offset Removal Circuitry
The load cell signal was amplified by a Linear Technology LT1168 instrumentation
amplified on the load cell amplifier board. The schematic for this circuit can be seen in
Figure 26 below. Each amplifier is presented as Instrumentation Amplifier A, B, and C,
one for each load cell.
-* 5V Rail
-> ?
Offset
NuIIA
Load QeII A
InpJjt *-
^ Load CbII C
*L, Load CeI I B
Input
Offset
NuIIC
Offset
NuIIB
Input
'
^
_ -5V Rail
"~
Instrumentation
Instrumentation
Instrumentation
Amplifier A
Amplifier B
AmDlifierC
-¦ Load Cell A Signal
-· Load Cell B Signal
· Load Cell C Signal
Figure 26: Instrumentation amplifiers and offset null circuit schematic. Simple circuit
allowed for load cell signal amplification as well as offset null by altering the voltage on
the positive instrumentation amplifier input pin via a trim potentiometer.
The positive and negative power rails are indicated, as well as the output signals. To
provide manual control over the offset measurement of the load cell, trim
potentiometers were included to vary the voltage on the non-inverting input pin of the
instrumentation amplifier, and the positive signal wire of the load cell. By manually
62
varying this voltage offset, it was possible to accurately remove the initial static loading
from each load cell signal (Hayward, 2008).
Although other ways for removing this offset exist, this method allowed for manual
control over the input range, while proving reliable and repeatable. This method was
simple to construct and adjust, proving ideal for the rapid development period of the
prototype (Linear Technology, 2008). Other methods were investigated and discarded,
including amplifier and ADC offset adjustment, and Wheatstone bridge arm adjustment
methods.
It was necessary to be able to control the offset measured by the load cells for a variety
of reasons. The ability to remove the offset voltage caused by the static mass of the
empty conveyor allows for the signal to be amplified more, and at the same time
maintain proper scaling of the output signal to the input voltage range of the analog to
digital converter (ADC) (Linear Technology, 2008). This system removed the tare weight
effect of the conveyor belt, and allowed the amplifier gain to be set to allow for a full
scale swing of the amplifier (voltage rails) to correspond to the full scale span of the
projected loading range caused by product flow.
The Linear Technology LT1168 instrumentation amplifier6 used is a resistor set gain
amplifier that can allow for amplification levels of up to 10000 (Linear Technology,
6 Linear Technology. Milpitas, CA, USA. Linear Technology LT1168 Instrumentation Amplifier.
63
2008). The gain was set to approximately 1000 for the load cells used. Figure 27
illustrates the tradeoffs between common mode rejection ratio (CMRR), frequency
response and gain.
Common Mode Rejection Ratio vs
Frequency (1k Source Imbalance)
Gain vs Frequency
? ilium ?
G = 1000
G = 100O
G = 100
g=iod
G = 10
o 100
G = 10
Q
60
VS=±15V
TA = 25'C
Ik SOURCE IMBALANCE
0.1
1
10
100
1k
10k
100k
FREQIJENCY(Hi)
0.01
0.1
1
10
100
1000
FREQUENCY (kHz)
Figure 27: Performance characteristics of the LT1168 instrumentation amplifier
integrated circuits. Left: Common Mode Rejection Ratio (CMRR) vs. Frequency plot
illustrating signal attenuation. Right: Gain vs. Frequency plot illustrating signal
frequency attenuation as gain levels increase, effectively limiting bandwidth (Linear
Technology, 2008).
At the higher levels of gain used to amplify the load cell signals, attenuation may occur
at frequencies as low as 10 hertz. The load cell signals of interest were anticipated to be
very low frequency in nature, and this was confirmed in subsequent sections (7.2.2
Frequency Analysis of Load Cell Signals') with frequency spectrums of interest in
the range of IHz.
6.6.4 Accelerometer Circuitry
There were many sources of harvester vibration and impact forces that could have been
picked up by the cantilever load cells. A grape harvester contains many different
vibration sources, such as the harvester picking head, conveyor belts, engine, drive
motors and suspension lift cylinders (Burks et al (2002), Dales et al (N. D)). Significant
64
amounts of vibration can also occur from traversing rough terrain, where sharp vertical
accelerations may cause larger variations in load cells measurements due to the inertia
of the suspended conveyor. This noise was harder to eliminate from the signal due to
the longer, lower frequencies of the signal as it overlapped load cell yield information
(Dales et al (N. D); Durrence et al, 2000; Pelletier, 2001).
By monitoring an accelerometer in the vertical direction, the impulse accelerations can
be determined and provide an indication of when there was a perturbation, as well as
the magnitude of it. This information could be actively integrated into the force
calculations to indicate what information to ignore, and over what time period the
accelerations are within tolerable limits (Durrence et al, 2000). An example of this kind
of sensor feedback system is detailed below.
3g
—^ Data Ignored
Acceleration Tolerance
Og
riiriiWKseVLini
',,"(IIIII'iKajLiaïétt.LAi
¦»HfTBft <<1»,^t>*frJ WrfWfl· HW i MMium,»..».
-3g
Acceleration
Time
Figure 28: Accelerometer response to a perturbation and subsequent system response.
This system would function by identifying the impact from the rapid impulse
acceleration (Figure 28), and depending on the severity of the impact, stop recording
the signal until it stabilizes (Durrence et al, 1999). Figure 28 illustrates this active
feedback system, where the accelerometer signal and the signal tolerances govern the
65
response of the control system. When the accelerometer signal exceeds the acceptable
range as illustrated above, all data would be ignored until the system returns to an
acceptable steady state level. To compensate for the gap of information from the load
cells while the data was being ignored, the previous yield values would be applied until
data was of sufficient quality, using a linear interpolation approach.
The Dimension Engineering7 DE-ACCM3D2 accelerometer breakout and power
regulation board that was mounted inside the instrumentation box can be seen in
Figure 29. This sensor board incorporates a voltage regulator, as well as preamplifiers
and filters for the three different axes of acceleration measurement
(DimensionEngineering, 2008). Output was a 0 to 5 volt analog signal. The breakout
board was designed to fit into a standard 16 pin DIP socket, allowing for easy integration
onto the amplifier printed circuit board.
eieund
Ves
(3StOlSVl
ss
Vaut
1IK^
î
CM
Z3taS
ZQuI
X oui
I
ÍWÍS33S
T
T 3MS
??p?p?
«cgyísier
X «WS
8mm
Figure 29: Dimension Engineering Buffered 3g 3-axis Accelerometer breakout board with
voltage regulation, reverse voltage protection and output short protection. This
breakout board was designed to fit into a 16 pin DIP socket for soldering onto boards
(DimensionEngineering, 2008).
6.6.5 Conveyor Drive Motor Circuitry
For simplicity, a DC gear head drive motor was connected to a 12V, 15A DC power
supply. The motor had two different power supply lines, one for high speed and the
7 Dimension Engineering. DE-ACCM3D2 Buffered Accelerometer Breakout Board. Akron, OH, USA.
66
other for a lower speed setting. The higher speed setting of 15rpm was used to ensure
an adequate rate of material discharge. For field testing, the gear head motor was
connected to an external accessories plug on the harvester, which provided the same
12V power required.
6.6.6 Instrumentation Power Supply System and Regulation
A standard switching mode power supply, DVE model number DSP-1454P, commonly
found in computer power supplies was used for the positive and negative 5V power
supply rails, as well as the common ground. Modern switching mode power supplies
feature good ripple removal, are stable voltage sources and proved to be adequate for
this system. This 200W power supply provided the following voltage output voltages
and amperages Table 3.
Table 3: DVE switching power supply model number DSP-1454P output voltages and
current ratings.
Voltage
Amperage
45V
+12V
-12V
-5V
2OA
7.5A
0.5A
0.5A
The power supply wires were hard wired into the amplification circuitry, with no
consideration for on board power supply regulation. This meant that if a different
power supplies were used, offsets in the voltage rails would cause all calibrations to be
different. For this reason, the power supply was kept the same for all testing, but this
should be addressed in future versions.
67
6.6.7 Instrumentation Installation and Cable Routing
The excitation and signal cables provided with each load cell were only 65 cm in length.
To improve the quality of the signal, the instrumentation amplifiers were installed
underneath the suspended conveyor belt within the inner frame (Figure 30). This
reduced the cable length that would otherwise be required, and thus reduced the
tendency for the wires to pick up stray electromagnetic noise.
Figure 30: Installed amplifiers (A), three axis accelerometer (B), and offset removal (C)
circuitry inside of the monitor frame. Note the use of strain relief connectors and
grommets to protect the load cell and signal cables from environmental hazards.
Accelerometer axes are labeled as well.
The underside of the conveyor belt frame also provides a convenient location for the
instrumentation box, as seen in Figure 30. The frame of the conveyor provides
mechanical protection and prevented the buildup of any organic residues. As can be
seen, all cables used fittings that provided a water tight seal as well as wire strain relief.
As indicated in Figure 30, the instrumentation box contained the amplifier printed
68
circuit board, as well as the 3 axis accelerometer break out board. This package was
designed to fit within the size constraints of the conveyor frame (box dimensions of 99
mm ? 51 mm ? 20 mm).
The layout of each individual component can be seen in Figure 31, the offset tare
potentiometers are the components indicated by (B), the instrumentation amplifiers (A)
and the 3-axis accelerometer breakout board (C). This simple board performs all
necessary functions required for the instrumentation of the conveyor system, providing
6 channels of information.
Figure 31: Signal conditioning board layout with instrumentation amplifiers (A), offset
tare circuitry (D), and 3-axis accelerometer (C) (Dimension Engineering DE-ACCM3D
accelerometer which incorporates an ADXL330 MEMS accelerometer).
This board was designed using Eagle Layout Editor 4.16r2 software, and the printed
circuit board layout above (Figure 31) was transferred to a copper clad printed circuit
board using the toner transfer method. The board was then etched in a solution of
Copper-ll chloride. Through holes for components were drilled and the different
components then mounted and hand soldered in place. The instrumentation amplifiers
69
and accelerometer breakout boards were placed in dual inline package (DIP) sockets to
allow for easy removal of the chips and to prevent heat damage from soldering.
6. 7 Control and Data Acquisition Interface Methodology
6.7.1 Data Acquisition System
For rapid development purposes, National Instruments™ LabView 8.68 Data acquisition
and graphical programming software was used to acquire and process the raw analog
signals from the conditioning board. This program allowed for an easy and quick
connection of the output signal wires to the National Instruments™ USB-6210 data
acquisition card. The 6 channels of analog voltage provided by the load cells and
accelerometer were converted by the National Instruments™ A/D converter, and
LabView 8.6 was used to store the acquired information as a ".txt" file for later use.
Each channel was sampled at 100Hz during testing, using a Pentium IV computer with
Windows XP operating system. Figure 32 illustrates the instrumentation and
information flow diagram of the acquired signals. The A/D converter as well as the data
acquisition computer was not mounted on the prototype, but as a harvester mounted
piece of equipment, as indicated.
The NI USB-6210 DAQ ADC used for testing features 16 single ended 16 bit input
channels, with a conversion rate of up to 250,000 samples per second. Although the
USB card features 16 single ended input channels, it was configured for 6 differential
National Instruments. Vaudreuil-Dorion, Quebec, CAN.
70
input channels to provide for better floating ground noise reduction, as well as an
increase in signal resolution.
Load CeE
Amplifier/ JJ
Offset NuU ??
Load Cell
:
I
!
Amplifier/
Offset Null
Load Cell
Amplifier/
Offset Null
Data Acquisition
Converter
Computer
Ac celerometer
Axis 1
Storage
Accelerometer
Axis 2
Accelerometer
Axis 3
Yield Monitor
Mounted
Harvester Mounted
Equipment
Figure 32: Instrumentation and control flow diagram for the yield monitor prototype.
6.8
Raw Signal to Mass Measurement Conversion
The following process was used to determine the total amount of grapes that passed
over the weighing conveyor, as well as the method for determining instantaneous yield.
Simpson's' rule for integration (below in Figure 33) was applied to create a better and
more accurate approximation to the total area under the measured and calibrated
signals.
71
P(XP
M
K
f(x)
Figure 33: Visual concept of Simpson's Rule of integration approximation. The shaded
region corresponded to the area under the curve created by the approximation function
P(x) (Simpson's Rule for Integration, 2008).
The Simpson's rule of integration equation for the area under the curve seen above can
be seen below (Equation 23) (Simpson's Rule for Integration, 2008). Simpson's rule is
more accurate then linear interpolation or other interpolation methods because it
applies a quadratic function to better approximate the curve (Simpson's Rule for
Integration, 2008).
¡f(x)dx * (^) /(fl)+4/(^)+/(¿)
Equation 23
The total area under the curve was added together by taking the period of b- to- a of
each sampling interval. Therefore, all of these segments would be added together to
obtain the total area under the curve of any time interval as seen in Equation 24.
bx-ax
Ra)x+AfXfLiA) + ^)1 + AzJh Ka]2+ 4/(^LÍA) + f{b)i
+ ... + (^LZf=L) /(a)„_1+4/(fl-^"-') + /(6)„_1
Equation 24
72
The load cells within the yield monitor prototype record the impulse, J, of the grape
material striking and being held on the conveyor. The area under the impulse-time
curve corresponds to units of Newton*seconds (Ns). The unit analysis of the system
(Appendix D: Unit Analysis of Mass Measurement) simplified by removing the constant
acceleration due to gravity, equating the system to units of kgm/s. It was assumed that
the majority of the loading was the result of the held mass and not impulse loading
(section 6.4.2 Projected Grape Material Impact Loading'). When assumed constant, this
impulse loading cam removed by applying a linear shift correction. As can be seen in
the figure below, the range over which the integration was performed was visually
selected from the raw calibrated data, and constitutes the beginning and end of the
material flowing onto, and off the prototype. The offset after the integration period in
Figure 34 was the result of grape material building up on the active components of the
prototype. After reaching steady state, this offset neither increased nor decreased.
73
1 Hz Low Pass Filtered Load Cell Signal (Calibrated)
5000
50
4500
45
40
35
30
25 S
3 2500
F
(J
20 £
15
10
5
0
1ItO]
-5
Time (seconds)
-~Mass (g) Plot -i-Force (N) Plot
Figure 34: Selected period over which the load cell signal was integrated to produce the
total mass upheld on the conveyor belt. Figure illustrates the force (N), as well as the
mass (kg) on the system, and illustrates the removal of gravity from the system.
This type of linear offset correction effectively eliminates the offset caused by the initial
velocity and momentum of the grapes, and approximates the system as a static loading
scenario. This method of correcting for inertial forces as was done in similar yield
monitoring systems, such as the sugar beet harvester system designed by Hennens ef ai
(2003).
6.9
Yield Monitor Test-bed Design
Due to limitations with access to a fully functional Grégoire™ harvester, as well as for
convenience, a lab scale test-bed was designed and constructed to aid in the
development and testing of the prototype. The test-bed design provided an almost
identical method of loading as the actual harvester, in a controlled environment.
74
Great effort was taken to match yield monitor loading conditions to the actual harvester
by using the same conveyor belt to elevate the grape material as well as maintaining the
same drop distances and angles. A variable speed gear head motor was used to
maintain the proper linear belt speed.
Figure 35 illustrates the exploded parts diagram of the main conveyor belt (left), as well
as the test-bed used for dynamic calibration and general testing and evaluation. The
following sections will detail the individual components of the test bed design.
I
s
i
^
%<&
£?
ì
H
i
^
?
i
ì
i
il
L^«-:
*£·*" *-'?
I
Í
^l
¿ß-,1
Î2
?.
^
1
?
1.-5
^
¡?
ti
]&¦^
--^
TM
Figure 35: Left: Parts diagram illustrating the general layout of a Grégoire harvester
main conveyor assembly. Right: Constructed evaluation test bed for dynamic calibration
of the yield monitor.
6.9.1 Conveyor Belt Speed Control
The linear speed of the conveyor belt is controlled by the variable speed main drive
motor, as seen in Figure 36 below.
75
W'M'J
Figure 36: Gear head drive motor and speed controller for the dynamic test-bed. Linear
belt speed can be dialed into correct levels for testing.
(A)
(B)
Main conveyor belt
Prototype yield monitor
(C)
(D)
Gear head drive motor
Drive motor speed controller
Figure 37 illustrates the metering auger and hopper arrangement that was used to
discharge grapes in a controlled fashion onto the main conveyor belt below. The
metering auger speed was variable due to the addition of a triac controlled motor,
allowing for fine tuning of the conveyor belt speed.
76
m
mm
/
?
ì
i
/
«£.
?
?
S
-JlER1
s
?
Jf
&
1
©
3uk!F
^
I^ m^
£,¦'**
»
^
*»1«
$£*
«I m
«f%
**^
JC
(Hf
Iv
-i
ft
.<
«r
à
^s
i
>
>
"S
5C
á*
^
î
JJ
4^·*·
JSs
(^J"
V
•y.
i
?
^
V
s,íÍSi
~>
-*-_
í^ksa
4^
"-s
^.
¿?
^:<y
®
"
®...
i
Figure 37: Metering auger for grapes to ensure a more uniform discharge of grape
material onto the main belt. Speed of the auger was controlled by a motor with a
rheostat.
(A)
(B)
Main conveyor belt
Polyethylene conveyor belt slides
(C)
Grape hopper
(D)
(E)
Metering auger drive pulley
Variable speed drive motor
6.10 Test-bed Performance and Calibration
The linear belt speed of the main conveyor belt was to be maintained at approximately
1.66 m/s, identical to that of the harvester during picking. Unfortunately, due to the
limited power rating of the electric drive motor, the belt speed could not always be
maintained at a uniform rate throughout the trial. As the belt became heavily loaded,
the increase in drag and weight reduced the speed of the belt by approximately 0.66
77
m/s. This reduced the consistency and repeatability of the loading trials, but adequate
loading characteristics and rates were obtained.
Table 4 illustrates the mass of grape material, duration and the calculated average mass
flow rate during a trial run to determine the best belt and metering auger speeds to
maintain the proper mass flow rate of material.
Table 4: Auger calibration trials illustrating the mass of grape material, duration, and
average mass flow rate.
Trial
Mass (kg) Duration (s) Average Mass Flow Rate (kg/s)
1
2
3
13.65
13.43
13.39
9.84
8.53
7.97
1.38
1.57
1.68
4
13.1
8.16
1.60
5
12.78
8.15
1_56
These trials yielded an average mass flow rate of 1.56 kg/s, with a standard deviation of
0.11 kg/s over the 95% confidence level. This flow rate corresponds to half the
maximum expected flow rate (3 kg/s) calculated earlier (Equation 4).
During this initial testing and set up of the system, it was noted that there was very little
lag time between the unloading of the metering auger, transport and loading of the
prototype. It can be assumed that there was no slip occurring between the fruit and the
main conveyor belt due to the cups formed into the belt.
78
6.11 Test-bed Dynamic Testing Methodology
Grapes were hand harvested at the University of Guelph Horticultural Research Station
vineyard near Vineland, Ontario. These grapes were kept in cool storage at 50C until
needed for a test. To ensure accurate and repeatable test conditions, an initial known
mass of grapes was fed uniformly into the metering auger hopper. The grapes traveled
along the conveyor to be discharged onto the prototype in the same fashion as the
actual harvester (Figure 38). All weighed material that was discharged from the
prototype was collected and subsequently weighed again to provide a gross total of the
experimental mass. The second weight was necessary because some juice and grapes
were thrown off of the conveyor belt during a trial.
r
?
\¿. i
\
I
Figure 38: Discharge from the test-bed onto the top surface of the weighing conveyor
(Left). Grégoire™ grape harvester discharging into the cross conveyor (Right).
Field conditions were also mimicked by preparing the grapes to match those of
harvested grapes by removing some stems and adding leaf material. This produced
greater amounts of free running juice, which was indicative of actual in field testing and
harvesting conditions.
79
6.12 Field Testing Methodology
Field testing consisted of installing the prototype onto a Grégoire™ harvester and
running it while harvesting numerous vineyard trellis rows. At the end of each trellis
row the central hopper was dumped into tote boxes and weighed. This total weight
represents the total amount of fruit that was collected by both conveyors. As such,
roughly half of this total weight should correspond to the measured mass of grape
material that passed over the prototype. Figure 39 illustrates the final prototype
installed into the Grégoire™ harvester prior to field testing.
t
/.
J
/¦
/
/
¿.
"N.
/
ifcfe
Figure 39: Final prototype installed into a Grégoire harvester for field testing trials.
80
Chapter 7
7.0 Results and Discussion
The following sections detail the static and dynamic calibrations of the prototype
weighing conveyor design. Included is the methodology used as well as the results of
said testing.
7.1
Yield Monitor Static Calibration and Performance
7.1.0 Static Calibration Methodology
In order to accurately determine the response of the mounted load cells to mechanical
loading in a controlled environment, standard weights were applied at the centre of the
weighing platform. The image below in Figure 40 illustrates the partially assembled
prototype, without the belt, with a standard weight placed in the exact middle of the
active loading area.
81
Figure 40: Basic layout of the yield monitor. The weight illustrates the location and
method of static loading for static calibration purposes.
In the actual static calibration, the conveyor belting was installed and tensioned to
proper operating levels. A series of weights was then applied, in the same fashion as
illustrated above, ranging from the empty, static weight of the conveyor, to 12 kg. The
manufacturers' specifications for the Omega Engineering LCAE-20kg load cells indicated
a hysteresis error of only +/- 0.015% of the full scale output, and creep error of only +/0.02%. This indicated that hysteresis testing would yield negligible results.
7.1.2 Static Calibration Equations
Figure 41 details the static calibration results. As can be seen, three distinct load cell
signals were obtained while loading the centre of the prototype. It should be noted that
82
two of the load cells (designated Load Cell A and Load Cell C) yielded a negative slope,
meaning the more heavily loaded, the more the voltage decreased. The load cell
designated Load Cell B had a positive slope, meaning increased mechanical loading
increased the output signal voltage. The calibrations for each load cell were different
occurred because the load cells were resisting different moments due to the location of
the centre of mass of the weighing belt, and the active end of the load cell.
Final Static Calibration with Conveyor Belt Installed and Tensioned
14???
y = 261 1.9x +7567.5
R2 = 0.9981
Load Cell C
17????
1 698. 2x +5633.5
R2 = 0.9956
1????
Load Ce A
mnn
y = -2858.8x + 13711
R2 = 0.9873
at
%
4QQO
Load Cell B
2QQÛ
Q
I
4.00
3.00
2.00
I
0.DO
1.00
1.00
?
2.00
3.00
4.00
2QQQ
Sensor Output Voltage (V)
? Load Cell A ? Load Cell B
? Load Cell C
Linear (Load Cell B)
Linear (Load Cell A)
Linear (Load Cell C)
Figure 41: Static calibration plot illustrating the response of all three load cells to a
known centrally placed weight. This calibration was obtained with the conveyor belt
installed and tensioned to operating levels. Shown are the calibration equations, as
well as their associated coefficients of determination values.
The following linear least squares regression lines and equations were obtained, and
their associated coefficients of determination (R2) values are provided.
83
5.00
Load Cell A:
Mass(g) = -2S5S.S*(VLoadCdl) + mU
R2= 0.9873
Equation 25
Load Cell B:
Massig) = 261 1 .9 * (VLoadCeU ) + 7567.5
R2= 0.9981
Equation 26
Load Cell C:
Mass(g) = -169Z.2*(VLoadCell) + 5633.5
R2= 0.9956
Equation 27
As can be seen, each load cell signal had a good fit with the linear static calibration
equations. These static calibrations were applied to the load cell signals obtained during
dynamic calibration, as well as the field trials. Dynamic testing was plotted as mass
versus time, when in reality this mass was an impact and holdup force, with the
gravitational constant omitted for simplicity.
7.2
Yield Monitor Dynamic Calibration and Performance
The dynamic calibration was performed by feeding a quantity of grape material similar
to field conditions through the test rig, allowing it to be weighed by the monitor
prototype, then collected and weighed independently.
The feed rate from the metering auger, as well as the belt speed of the main conveyor
was maintained at as uniform a speed as possible. Grapes were hand harvested at the
University of Guelph Horticultural Research Station vineyard near Vineland, Ontario.
These grapes were kept in cool storage at 50C until they were placed in the metering
auger. Following testing as much material as possible that passed over the yield
monitor was collected in an attempt to minimize experimental error.
84
Figure 42 illustrates the yield monitor installed into the test rig for dynamic testing and
calibration. A large (173L) plastic storage tote was positioned under the discharge end of
the monitor belt to catch the grapes and juice as a dynamic trial was run. Grape material
collection proved difficult, and some mass was lost as whole grapes and juice. Likewise,
repeated handling of the grapes caused greater amounts of juice to be released, and
some material may have missed the prototype conveyor belt entirely.
Figure 42: Grape harvester yield monitor installed into the testing apparatus. Grapes fall
from the top of the pulley, conveyor belt area. Grapes were discharged toward the front
for collection and bulk weighing for calibration purposes.
It should also be noted that there was unavoidable variation in the rate at which grapes
were fed into the metering conveyor. Multiple baskets were dumped by hand into the
metering conveyor one at a time, and although the metering conveyors purpose was to
85
remove variations in loading, it was not always effective. Therefore, there were trials
where an uneven rate of loading was detected.
7.2.1 Load Cell Signal Results
The following plots are a subset of the results obtained during the dynamic calibration
trials. These graphs were obtained from the load cell signals produced during the same
trial, and are representative of all of the testing conducted.
The raw load cell signal illustrated in Figure 43 was the characteristic response obtained
from Load Cell C in dynamic trial 3. This signal is un-calibrated and unfiltered, and
illustrates the noise nature of the raw load cell information. As noted in earlier
discussions, the output voltage from this load cell decreased as mass on the platform
increased.
86
Raw Load Cell Signal (Uncalibrated and Unfiltered)
¦ ii^iàiàailkik.^
160
Time (seconds)
Figure 43: Load cell voltage outputs when subjected to dynamic loading scenarios
identical to field conditions.
Filtering was used to remove the majority of the noise present within the signals, as
seen in the following section.
7.2.2 Frequency Analysis of Load Cell Signals
The frequency spectra of the previous raw data for Load Cell A, B and C are presented in
Figure 44, Figure 45, and Figure 46, respectively. The use of a Fast- Fourier Transform
(FFT) allowed for the different frequency components of the load cell signal to be
identified (Haykins & Van Veen, 2003).
It should be noted that the magnitude versus frequency of the signals are displayed on a
decibel (dB) scale, to allow a clearer indication of noise levels (Haykins & Van Veen,
87
2003). Since the load cell signals were recorded at a sampling rate of 100Hz, the
frequency plots only illustrate frequencies of up to 50Hz, in accordance with the Nyquist
sampling theory (Haykins & Van Veen, 2003).
j£j$jie-Sided Amplitude Spectrum of Filtered Load Cell Signal C
? lO^ingle-Sided Amplitude Spectrum of Load Cell Signal C
Frequency (Hz)
Frequency (Hz)
Figure 44: Fast Fourier Transform (FFT) of the unfiltered test-bed signal (left) for Load
Cell C. Low pass filtered signal (right).
The above frequency plots (Figure 44) for Load Cell C (left) indicated that noise
frequencies peaked at approximately 12, 20, and 40 Hz, respectively. Due to the nature
of the frequency distributions, a digital 4th order Butterworth filter with a corner
frequency of IHz was applied to remove these unwanted frequencies. The results can
be seen in the frequency plot on the right in Figure 44.
88
]f)5Single-Sided
,«Single-Sided Amplitude Spectrum of Load Cell Signal B
» igSingle-Sided Amplitude Spectrum of Filtered Load Cell Signal B
s?
R
ÍN
8
10
10
15
12
14
5
.10
15
20
? ¦
25
.
30
?
¦
35
.
40
I
45
_2ni
50
'
0
1
5
1
10
'
15
'
X
Frequency (Hz)
25
'
30
>
35
¦
40
'
45
'
50
Frequency (Mz)
Figure 45: Fast Fourier Transform (FFT) of the unfiltered test-bed calibration for Load
Cell B (left). Low pass filtered signal (right).
The frequency plots for Load Cell A, and Load Cell B indicated similar quantities, and
frequencies of noise, as can be seen in Figure 45 above and Figure 46 below,
respectively.
K 105Single-Sided Amplitude Spectrum of Load Cell Signal A
? Jingle-Sided Amplitude Spectrum of Fiitered Load Cell Signal A
1
C
R
R io
-ß
10
15
12
4
¦I
?
5
10
15
20 25 30
Frequency (Hz)
?
35
?
40
?
45
I
.201
50
0
'
5
'
10
'
15
'
'
'
20 25 30
Frequency (Hz)
¦—:—'
35
40
'
45
'
50
Figure 46: Fast Fourier Transform (FFT) of the unfiltered test-bed signal (left) for Load
Cell A. Low pass filtered signal (right).
There were many sources of noise in the testing environment. The main source was
from the drive system of the large conveyor belt, but other sources existed as well, such
as the cleats and buckets on the belt, pulley rotation and imbalance, weighing belt drive
motor, and the loading auger; as well as others. The source of more widespread
89
frequencies of noise could arise from the grape material impacting onto the yield
monitor belt surface.
7.2.3 Filtered and Conditioned Load Cell Signals
th
When the raw output signal seen in Figure 43 was filtered using the 4 order
Butterworth filter with a 1 Hz corner frequency, the un-calibrated result in Figure 47 was
obtained.
The quality of the signal had been improved, although there were still periodic
waveforms present. The different stages in the dynamic calibration can also be seen. At
approximately the 5 second mark the conveyor belt on the yield monitor was turned on,
illustrated by the start of the periodic waveform. Likewise, the start of the loading by
the grape material occurred approximately 50 seconds into the trial, and ended after
approximately 85 seconds. Interestingly, the starting offset and final offset weights are
not identical, indicating that there was some material remaining on the prototype after
the test, indicating incomplete discharge of the grape material (berries and juice).
90
1 Hz Low Pass Filtered Load Cell Signal (Uncalibrated)
0.5
20
40
60
80
100
120
140
160
Time (seconds)
Figure 47: Load cell signal (voltage) output when subjected to dynamic loading scenarios
identical to the expected field conditions. Indicated are when the prototype conveyor
was turned on and when grape material impacted and was discharged from the surface.
Figure 48 is the resulting Load Cell C signal after the static load cell calibration equation
had been applied. The starting and ending offsets were more prominent, as well as and
magnitude of the periodic waveform while the yield monitor is running unloaded.
91
1 Hz Low Pass Filtered Load Cell Signal (Calibrated)
5000
4000
3000
2000
Offset after test
Offset while running empty
1000
A^ AtLi^AtIiI
tm * -,lit EtI t.
? Ali
140
150
160
?ttptp^
-1000
Time (seconds)
Figure 48: Load cell (voltage) output when subjected to dynamic loading scenarios
identical to the expected field conditions. Only the offloading (Load Cell C) signal is
illustrated.
The following three plots illustrate the individual Load Cell A, Load Cell B, and Load Cell
C signal responses for a typical calibration trial. As can be seen, the amplitude of Load
Cell B signal was greater than the amplitude of Load Cell C, indicating a greater
sensitivity to mechanical loading. Unfortunately, there were also greater amounts of
noise in the Load Cell B signal while unloaded.
This was because Load Cell C was located at the discharge point of the conveyor belt,
away from conveyor drive motor vibration and torque action. Load cell B had to
support the majority of the motor, whereas Load Cell C was supporting only the
92
discharge. Grape material flowed off of Load Cell C, meaning that material was
constantly moving away from Load Cell B, with all material passing over this load cell.
Load Cell C Signal (Calibrated)
14000
2000
10000
8000
(?
S
6000
4000
2000
60
70
80
90
2000
Time (second)
Figure 49: Calibrated Load Cell C signal.
93
100
110
120
130
140
150
160
Load Cell B Signal (Calibrated)
14000
12000
10000
8000
6000
4000
2000
60
70
80
F#*+F#*%f#**?<
90
100
110
120
130
140
150
160
-2000
Time (second)
Figure 50: Calibrated Load Cell B signal.
The response characteristics of the Load Cell A signal (Figure 51) indicated that this
particular load cell was not very sensitive to the applied mechanical loading, and was
not worth further study for the lab calibration trials. This was due to the location of the
load cell in relation to the location of loading, and the direction of product flow. Most
of the material discharged onto the yield monitor moved towards Load Cell B and C,
leaving Load Cell A carrying little loading. The flow of material was off of Load Cell C,
away from the already lightly loaded Load Cell A.
94
Load Cell A Signal (Calibrated)
14000
12000
10000
"3
SOOO
6000
4000
2000
dlattaLiiiA tl MU Il Mi Mi k MM
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
-2000
Time (second)
Figure 51: Calibrated Load Cell A signal.
The combined load cell signals seen in Figure 52 and Figure 53 are of greater overall
amplitude then the individual load cells, as a direct result of the summation of
calibrated signals. The calibration equations automatically compensate for the negative
output voltages from Load Cells A and C under load.
95
Load Cell B and C Signals (Calibrated and Summed)
14000
12000
10000
8000
_
6000
4000
2000
-2000
I W^^f^WP
60
70
80
90
1 00
110
1 20
1 30
1 40
1 50
1 60
•4000
Time (second)
Figure 52: Resulting yield mass flow rate when the signals from Load Cell B and C are
combined in an additive method.
As can be observed in the increase in overall noise when comparing the resulting signals
in Figure 52, in Figure 53, the addition of the Load Cell A output resulted in a lower
quality result.
96
Fully Suspended Mass Measurement (All Load Cell Signals Summed)
14000
2000
10000
8000
6000
M
4000
2000
EO
70
80
90
100
110
20
130
140
150
60
2000
4000
Time (second)
Figure 53: Resulting yield mass flow rate when all load cell signals are combined
Many conclusions may be drawn from the results presented above. Most importantly,
the quality and sensitivity for each load cell was different, Load Cell A being ineffective
at producing a usable signal for reasons previously discussed. All load cell signals were
added together in a linear fashion, under the assumption that load cell loading would be
equal, given an equal distribution of grape material on the conveyor belt.
Load Cell B and C presented very good signal responses to the applied mechanical
loading. These signals can be summed together to provide a signal with greater overall
sensitivity to mechanical loading. Unfortunately, this compounds the error that might
be associated with each signal. Interestingly, each load cell had a unique periodic
97
waveform present when the yield monitor belting was run without any loading, as seen
in the following section.
7.2.4 Load Cell Periodic Waveform Analysis
The low pass filtered signal obtained from Load Cell C seen in Figure 54 contains the
original, unloaded signal (blue trace) as well as this signal shifted a period of 1 and 2
wavelet wavelengths, indicated by the green, and red traces, respectively. The orange
and dark grey traces are the difference between the original signal (blue trace) and the
shifted periodic waveform.
1 Hz Low Pass Filtered Load Cell Signal (Illustrating Phase Shifts of 1 and 2 Periods)
1000
2 Period Shift
1 Period Shift
-H
800
BOO
¡Î
8
4GO
200
i
m
M
35
5
h=
200
X-,»
L
1 Period Shift
Ù
40
¦
/!
??-?
EO
45
Time (seconds)
Original Waveform
1 Period Phase Shift
2 Period Phase Shift
Deviation (1 Period)
Diviation (2 Periods)
Figure 54: Periodic waveform illustrating the difference (error) in the shifted signals
from the original Load Cell C signal.
As can be seen, this periodic and very repeatable waveform had a period of
approximately 18 seconds, peak to peak for the signal wavelet. This corresponded to
98
65
the time it took for the conveyor belt to take one full revolution of the conveyor
monitor. The highest peak (Figure 55) occurred when the stainless steel belt lace
traveled around the tensioning pulley at the site of Load Cell C. The steel belt lace
crossed this point every 8.31 seconds, which was a result of the motor speed (15
rotations/minute), drive roller diameter, and belt length of 90 cm.
If the phase of the periodic waveform was known, then this resulting signal may be
subtracted from the original, effectively removing this periodic waveform. This method
would have to be applied to remove this unwanted signal because it is overlaid on the
desired signal produced by the grape material. If the corner frequency of the
Butterworth filters were changed to remove this periodic waveform, drastic amounts of
the desired signal would be removed as well due to the lower frequency of this periodic
signal.
Load Cell B also contained a periodic waveform, but with much higher frequency
oscillations, as seen in Figure 55 below. The green trace represents the phase shift of
the original signal by 1 period, and the red trace a shift of two periods, for the purpose
of illustrating the repeatability of the signal. The orange and gray traces are the
differences between the original signal, and the phase shifted.
99
1 Hz Low Pass Filtered Load Cell Signal (Calibrated) Illustrating Periodic Waveform
9DO
7ÜD
500
S
300
in
%
^
10
15
f
I
a
100
20
?
\
>
25
35
30)
m
40
45
50
00
55
OJ
\
-.IJ
?
Up5
70
300
500
Time (seconds)
Original Waveform
1 Period Phase Shift
2 Period Phase Shift
Deviation (1 Period)
Deviation ß Periods)
Figure 55: Periodic waveform illustrating the difference (error) in the shifted signals
from the original Load Cell B signal.
As can be seen, the deviation between the original and phase shifted signals was much
greater for Load Cell B then for Load Cell C, primarily due to the increased frequency
and reduced repeatability of the signal. Therefore, the signal from Load Cell C would
produce the more accurate results after post processing using the subtractive method.
7.2.5 Vibration and Acceleration Analysis
The 3- axis accelerometer placed within the prototype measured the different levels of
accelerations experienced by the device, and consequently the load cells, over all three
directions (x, y and z). The three traces in Figure 56 illustrate the different levels of
acceleration over the duration of a dynamic calibration test trial.
100
Test-Bed Results: Unfiltered 3-Axis Accelerometer Signals
1
1 E-OBx + 0.0935
t^m.fc.t.
Il .1.1. tLttktitJr
tlL.rtk.lJ
R2 = 7E-06
t
p*??**.? »m»1 «?
y = 1E-OBx -0.0788
R2 = 2E-05
<
t
tA
iC^ö O-J'^-^AjV
1
t
tï_JLî
? /^v=
«?-
'V
sA
•siï1—XJ
???
???^??^?'????
/^|rn^"T^|^
*r**%M ?
fgv*~t>
WtJ
?
î
¦^J^GSSäLisO
S;
V ,a
1.5
ÎL
&
ÜL^^jVj
^-^¦!^¡G-'?G^?-w^^
Tl
y = 2E-07x- 1.1071
R2=1E-07
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
Time (seconds)
—¡—Accelerometer Axis 1
—-Linear (Accelerometer Axis 2)
=s—Accelerometer Axis 2
Linear (Accelerometer Axis 1)
—^-Accelerometer Axis 3
Linear (Accelerometer Axis 3)
Figure 56: Test-bed results of the 3-axis accelerometer signal. Illustrated is the full time
of a loading experiment, including run up and down time. Signal offset was the result of
inclination error from the placement of the 3 axis accelerometer within the
instrumentation box.
As can be seen, there were three different offset levels measured, and this corresponds
to the angle of the accelerometer to the vertical direction. The offset was caused by the
different angle between each accelerometer to the downward force of gravity.
Accelerometer Axis 2 measured the vertical acceleration components, and was inverted
when placed inside the instrumentation box inside the yield monitor resulting in the
negative acceleration offset. Accelerometer Axis 1 and 3 correspond to the transverse
(yield monitor discharge direction) and lengthwise accelerations (perpendicular to the
direction of yield monitor discharge) of the yield monitor, respectively.
101
The plot ¡? Figure 56 illustrates three different stages of low level accelerations and
vibrations present for each accelerometer axis measured. For approximately the first 50
seconds low levels of vibration were present due to the conveyor belt and motor of
both the prototype, as well as the main loading conveyor belt and motor. Higher levels
of low amplitude accelerations occurred when the monitor was loaded by grapes being
discharged from the loading conveyor belt onto the yield monitor prototype. The
magnitude of the grape induced vibration was a function of the mass flow rate, release
velocity and drop height of the falling grape material. The grapes impacting along the
lengthwise axis induced greater accelerations in this direction then in the yield monitor
discharge direction. This was due to the impact angle of the grapes onto the yield
monitor conveyor belt. The vibrations themselves traveled through the suspended
conveyor to be recorded by the accelerometers mounted internally.
Figure 57 illustrates a 15 second segment of the signals seen above in Figure 56.
Although the accelerometer data was only sampled at 100Hz, low levels of vibratory
accelerations can be seen. Since larger levels of vibration, indicating impacts that may
affect the integrity of the load cell signal were not detected during in lab testing, the
intention of using accelerometer signals to block load cells signals under a sharp impact
condition was not required.
102
Test-Bed Results: Unfiltered 3-Axis Accelerometer Signals (15 Second Segment)
0.5
y = -19-06x + 0.0935
*>t Ljitit*
¡. L·,!.* L ??*?*? utÀta tL*.t ¿kjitfii /1 tit titMs^tiMttjAit^k
-- 7E-06
Ti'r^'^"-'te'J-'''"w"'*^"^
'
y = 1 E-OBx - 0.0788
R2 = 2E-05
5 -0.5
i
í!
'-·* & bt<sn\ß " tw· G'?'?'?' '
G*
'4*f^>^
?» ìti lo- w,"^F-:^^ ^.
! f
1J VuJ1T
11 -1W fe " G ¦'*«'
? «lili1-
*- ? .11 ir L V
f¥f?fÄ
Fl
-1.5
y = 2?-07?- 1.1071
R2 = 1E-07
65
66
57
68
69
70
71
72
73
74
75
76
77
78
79
80
Time (seconds)
—s—Accelerometer Axis 1
------ Linear (Accelerometer Axis 2)
—»-Accelerometer Axis 2
Linear (Accelerometer Axis 1)
—•—Accelerometer Axis 3
Linear (Accelerometer Axis 3).
Figure 57: Test-bed results of the 3-axis accelerometer signal, illustrating the output
characteristics of the signals, as well as the mean of the signal.
When the inclination error and acceleration due to gravity was removed (Figure 58), the
vibratory nature of the resulting signal indicated that the planned method of using the
accelerometer to give an indication of load cell signal quality was not a viable, nor
necessary method of improving signal quality.
103
Test-Bed Results: Unfiltered 3-Axis Accelerometer Signals (15 Second Segment)
D.8
0.6
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
Time (seconds)
- Acceleromter Axis 1 —o— Accelerometer Axis 2 —>—Accelerometer Axis 3
Figure 58: Test-bed results of the 3-axis accelerometer signal. A 15 second interval is
illustrating the characteristics output of the signals. Signals with inclination offset
removed.
Fortunately, the accelerometer data also provided an indication of the vibration
frequencies present within the system, as well as what specific frequencies should be
filtered out of the signal.
7.2.6 Accelerometer Frequency Analysis
A Fast Fourier Transform (FFT) of the above signal resulted in the frequency spectrum of
the accelerometer signals, as seen below in Figure 59. This spectrum indicates that
there are noise frequency spikes at approximately the 20 and 40 hertz ranges, similar to
those in Figure 44 from the yield monitor load cells.
104
in
Singl&Sided Amplitude Spectrum ofAccelsroniier Axis Í
10
Single-Sided amplitude Speciium of Acceierorrster Asís 2
0
a
S)
-o
S 10
R
?
?
14
?
IB
8
10
15
20
40
35
30
45
B
50
10
15
20
Frequency (Hz)
25
30
35
40
45
50
Frequency (Hz)
ig5
Sôngle«Sidêd Amplitude Spectrum of Acceleramter Asris 3
-8
S
|l
Q
'
5
L-
10
1
15
,J
20
!
25
I
2)
I
35
1
40
i "
45
?
50
Físquencj (Hz)
Figure 59: Fast Fourier Transform (FFT) of the unfiltered test-bed calibration trial for
accelerometer axis 1 signal (top left), Axis 2 (top right), and Axis 3 (bottom).
Interestingly, the FFTs that were performed on the raw load cell signals provided a
better indication of the frequencies of the noise present in the system then the
accelerometers. This indicates that the 3-axis accelerometer would not be necessary,
and can be omitted from future designs, primarily due to the lack of large level impact
type accelerations present during in-lab testing.
7.2.7 Dynamic Calibration Uncorrected Results
Nine dynamic tests were performed on the yield monitor prototype in the lab using the
test rig. Table 5 lists the average mass flow rate of grape material for each of the
performed trials.
105
Table 5: Dynamic calibration trials results indicating the mass of grape material for each
trial, duration and calculated averaged mass flow rate.
Trial Mass (kg) Duration (s) Average Mass Flow Rate (kg/s)
1
2
3
4
5
6
7
26.62
20.94
20.34
19.19
21.97
20.78
18.06
27.00
20.47
17.31
16.75
22.63
24.75
21.12
0.986
1.023
1.175
1.146
0.971
0.840
0.855
8
16.65
16.56
1TJ06
An overall average mass flow rate of 0.966 kg/s was obtained, with a standard deviation
of 0.15 kg/s, over a 95% confidence interval of 0.12 kg/s.
After each trial, the load cell signals were analyzed to create a total weight of grapes
that had passed over the prototype. Table 6 illustrates the actual mass, the measured
mass and the percentage difference (error) between the actual mass, and the mass
produced by the prototype. The results below were obtained by applying the static
calibration equation for Load Cell C, and using Simpson's rule of integration to calculate
the impulse, J, of the grapes striking, and the held-up mass on the yield monitor.
Table 6: Test results from the test-bed dynamic calibration testing utilizing Load Cell C
linear calibration equation.
Test
Actual Mass (g) Measured Mass (g)
Error (%)
1
2
3
4
26624.00
20936.00
20336.00
19188.00
20490.64
17551.51
17201.98
18032.96
-29.93
-19.28
-18.22
-6.41
5
21974.00
17525.44
-25.38
6
7
20781.00
18056.00
17166.54
14220.53
-21.06
-26.97
8
16652.00
13433.12
-23.96
106
The mean error for Load Cell C trial between the measured and actual mass was -21.4%,
with a confidence level (95%) 6.0% error. The bulk of the error stems from the forces
imparted by the grapes onto the suspended conveyor belt due to the vertical drop, and
the initial velocity and momentum of the grape material. The loss of grape material
during each trial also accounted for a portion of the constantly negative error
measurements.
The results presented in Table 7 illustrate the percentage of error between the actual
mass, and the mass measured by using only the Load Cell B. As can be seen, there was a
great deal of variability in the accuracy of the signal, ranging from 7.9% to 33.4%. The
mean error for this load cell trial was 19.56%, with a 95th percentile confidence interval
of 8.8%.
Table 7: Test results from the test-bed dynamic calibration testing utilizing Load Cell B
linear calibration equation.
Test
Actual Mass (q) Measured Mass (q)
Error (%)
1
2
3
4
5
6
7
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
30624.08
25072.86
22267.66
20833.33
32972.80
30235.57
26062.14
13.06
16.50
8.67
7.90
33.36
31.27
30.72
8
16652.00
19644.80
15.23
Due to the low quality of the signal from Load Cell A, it was not possible to discern any
variation in loading. For this reason, Load Cell A was omitted from further evaluations.
107
When Load Cell B and C signals were combined to produce a larger amplitude signal, the
error naturally got larger, as did the variation and distribution of error.
Table 8: Test results from the test-bed dynamic calibration testing utilizing the
combined signals from the Load Cell B and C linear calibration equationsTest
Actual Mass (q) Measured Mass (g)
Error (%)
1
2
3
4
5
6
7
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
50815.12
42414.11
39334.42
38730.76
44211.57
55451.52
39338.48
47.61
50.64
48.30
50.46
50.30
62.52
54.10
8
16652.00
40285.44
58.66
It should be noted that to a certain degree, a lot of the error can be corrected for by
applying calibration offsets to obtain the actual desired accuracy, provided that the
distribution and range of the error is small. It is not possible to correct for great
amounts of variation, since it is no longer a linear offset, but a random distribution of
errors. The above trial obtained a mean error of 52.8%, and confidence level (95%) of
4.4%.
Since the prototype conveyor is suspended on three load cells, it can be assumed that
the total load is distributed evenly amongst them, given an even distribution of loading.
The total summed masses obtained by the load cells can be evenly divided. The result of
this analysis can be seen in Table 9 below.
108
Table 9: Test results from the test-bed dynamic calibration utilizing the combined signals
from the Load Cell B and C linear calibration equations, but assuming an even
distribution of loading between load cells. Therefore the full load is half the measured
amount.
Test
1
2
3
4
5
6
7
8
Actual Mass (g) Measured Mass (g)
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
16652.00
25407.56
21207.05
19667.21
19365.38
22105.79
27725.76
19669.24
20142.72
Error (%)
4.57
-1.29
3.29
-0.92
-0.60
-33.42
-8.93
-20.96
This resulted in a mean error of -7.3%, with a confidence level (95%) of 11.2%. It should
be noted that if trial 6 and 8 were erroneous in nature, the average error would be only
0.77%, with a confidence level (95%) of 4.9% error. More analysis would be beneficial.
7.2.8 Dynamic Calibration Corrected Results
It was possible correct any offset errors by applying an offset correction to the original
static calibration equations as seen in Equation 28.
Mass{g) = -1698.2 * (VLoadCell ) + 5633.5 + MCorrection
Equation 28
This Mcorrection for each load cell signal is a unique value that corrects for the impulse
force generated by the falling grapes, as well as any linear offsets caused by the static
calibration equations. This MC0rœction value effectively equates the impulse measured
through the integration of the signal, to a mass (g) value. These MCo/Tecf,on values are
empirically derived, and are the result of experimentally varying the offset bias to
eliminate the difference between the actual mass and the measured mass. The
equation used to correct the Load Cell C signal can be seen below in Equation 29.
109
Massig) = -1698.2 * (VLoadCel, ) + 5633.5 + 366
Equation 29
This correction resulted in the removal of the error offset as can be seen below in Table
10. This resulted in a mean error of -2.7%, with a 95% confidence level of only 3.9%
error.
Table 10: Results from the test-bed dynamic calibration testing utilizing the corrected
Load Cell C linear calibration equation.
Test
Actual Mass (q) Measured Mass (g)
1
2
26624.00
20936.00
3
4
5
6
7
8
Error (%)
24148.20
20601.51
-10.25
-1.62
20336.00
19640.76
-3.54
19188.00
21974.00
20781.00
20472.96
21184.22
20215.32
6.28
-3.73
-2.80
18056.00
16652.00
17879.31
15871.90
-0.99
-4.91
The same static calibration equation offset correction procedures were applied to the
Load Cell B signal. The original calibration equation was altered to take the form of
Equation 30.
Massig) = 261 1.9 * iVLoadCell) + 7567.5 - 55?'
Equation 30
The equation resulted in eliminating much of the offset in the signal, but could not
correct for the large distribution of error, as can be seen below in Table 11 below (error
of 2.3%, confidence level (95%) of 10.8%).
110
Table 11: Test results from the test-bed dynamic calibration testing utilizing the
corrected Load Cell B linear calibration equation.
Test
1
2
3
4
5
6
7
8
Actual Mass (g) Measured Mass (g)
·
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
16652.00
25127.75
20487.69
18600.99
17164.83
27302.33
26018.90
20519.51
15978.14
Error (%)
-5.95
-2.19
-9.33
-11.79
19.52
20.13
12.01
-4.22
Due to the variation in Load Cell B and C signals, when the load cell signals were
combined, the error in both load cells was added together. This resulted in a lower
quality signal after compensation, but still a more realistic value then the uncorrected
signal. The combined compensated load cell calibration equations can be seen in
Equation 31, where MCorrect¡on, universal corrects for any offset caused by the addition of the
two load cell signals.
Table 12: Test results from the test-bed dynamic calibration testing utilizing the
combined signals from the corrected Load Cell B and C linear calibration equations, and
corrected offset.
Test
1
2
3
4
5
6
7
8
Actual Mass (g) Measured Mass (g)
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
16652.00
24992.35
20875.18
20306.09
21498.86
28608.85
26066.74
15452.48
15724.03
111
Error (%)
-6.53
-0.29
-0.15
10.75
23, 19
20.28
-16.85
-5.90
As indicated, there was a mean error 3.1%, with a 95th percentile confidence interval of
11.6%, indicating a lower resolution with a greater amount of distribution about the
mean.
Massig) = [- 1698.2 * (VLoadCellc ) + 5633.5 + MCorrection, LoadCellc\
+ [261 1.9 VloadCellB) + '567.3 — MCorrection^ LoadCellBi+ ^ Correction,Universal
Equation 31
Massig) = [-1698.2 * iVLoadCellc) + 5633.5 + 366]
+ [2611.9*(rWCWi) + 7567.5-550]-2400
Equation 32
Based on the potential for post processing, and the ability to save cost in
instrumentation, the use of a single load cell would be preferred over the fully
suspended conveyor system. Removal of the periodic waveform in Load Cell C would
yield a more accurate result, and negate the need for other load cells. This would reduce
the number of load cells from three down to one, reducing cost, complexity and signal
processing requirements. The location of Load Cell C was better suited for mass
measurement due to its location at the point of discharge where all material was forced
to pass, and away from the drive motor ensure greater sensitivity and reduced signal
noise.
7.3
Yield Monitor Field Testing
Field testing of the yield monitor represented uncontrolled real world testing, and
represented the most rigorous form of testing possible. This information aided in
determining how applicable the current design was to real conditions. Chardonnay
grapes at William Falk Farms were harvested in September of 2008. A total of 6
vineyard trellis rows were harvested, each approximately 79 meters in length at an
112
average ground speed of 2.2 km/hr, harvester main conveyor belt speed of 110 rpm, a
picking head speed of 420 rpm, bottom and top fan speed of 1830 rpm and 1478 rpm,
respectively. The following information presented was indicative of each trellis row that
was harvested, and used to convey the average response characteristics of the yield
monitor prototype to field conditions.
7.3.1 Field Testing Load Cell Analysis
Figure 60 illustrates the filtered and calibrated results obtained from the prototype over
the course of harvesting one trellis row. Each individual trace represents a single load
cell. There are numerous interesting features present in this signal, such as the initial
spike in the signal, which corresponds to the main conveyor belt turning on and
discharging held material onto the belt.
Figure 60 below illustrates the load cell signal responses that were obtained as the yield
monitor weighed the grape material. Illustrated are the three load cell signals (Load Cell
A, B and C). Interestingly, the Load Cell A and Load Cell C signals featured the greatest
sensitivity to mechanical loading, whereas Load Cell B signal had little sensitivity. This
was the opposite of the loading scenario seen during lab testing, where Load Cell A
illustrated little change to flow rate loadings. This was likely the result of the large
buildup of material on the conveyor belt, levels of which were not obtained in lab
testing. The material collected towards the site of Load Cell A, with less material
building up at Load Cell B. All material was discharged over the measurement site of
Load Cell C (Figure 61) as indicated. The location of the buildup of grape material led to
113
large amplitude signals from Load Cell A and C, with mess material in the vicinity of Load
Cell B, as was indicated by the results in Figure 60 and Figure 61.
In Field Testing Results: 1 Hz Low Pass Filtered Load Cell Signals (Calibrated)
14000
12000
10000
B00Ü
S
¡8
6000
4000
2000
i
oí
&
30^40
\?
^W
IT
50
BO 70
80
?
¡SpffiWä
«
^5 -^?
M
90 100 110 120 130 140 150 160 170 180 190 200 210-^0^230
240
G? is
^?
lí^cáEbíw P
j? '9ft.
2000
Time (seconds)
Load Cell B
Load Ce C
Load Ce A
Figure 60: 1 Hz low pass filtered load cell signals from a typical in field trial. (Vineyard
Row 3 at William Falk Farms).
114
B
?
ffclrï
Figure 61: Grape material loading characteristics of the yield monitor during field
testing. Active measurement points of Load Cells A, B and C is indicated as well as main
area of loading (D).
Figure 62 was the result of combining the load cell signals in each possible configuration.
From a signal analysis and instrumentation point of view, and based upon the data
obtained from the in-lab testing, the most accurate signal was obtained from Load Cell
C. No further analysis was completed on the combination of these load cells besides
what is presented in Figure 62.
115
In Field Testing Results: 1 Hz Low Pass Filtered Load Cell Signals (Calibrated)
1800D
1B000
14000
12000
10000
S
8000
2000 -
10
20
30
40
50
60
70
90
100 110 120 130 140 150 160 170 160 130 200 210 220 230 240
-2000
Time (seconds)
Load Cell B and C
Load Cell A and C
-All Load Cells
Load Cell A and B
Figure 62: IHz low pass filtered load cell signals (calibrated) from a typical in field test.
Load cell signals were combined in each possible configuration for evaluation purposes.
Vineyard Row 3 at William Falk Farm.
7.3.2 First Order Dynamic Response Characteristics
The dynamic, in-lab testing responded as a simple time delay between the start of the
test and when the first material reached the prototype. The loading characteristics can
be approximated as a step loading situation due to the lack of material slip on the main
conveyor belt. The observed dead time was a function of the linear speed of the belt,
and the time required for the loaded conveyor cups to discharge at the site of the yield
monitor prototype.
It was observed during field testing that there was a short first order transportation
delay from when the harvester entered the row and when grape material first started to
116
appear at the site of measurement. This lag was the result of numerous variables, such
as the picking head speed, the mechanical attachment of the grapes with the vine, main
conveyor belt speed, harvester ground speed, and yield. Figure 63 is the calibrated load
cell signal obtained from field testing, and illustrates this first order lag as the harvester
first entered the vineyard trellis. The initial spike in mass seen in the graph was the
result of the main conveyor belt being engaged and discharging held material onto the
prototype.
In Field Testing Results: 1 Hz Low Pass Filtered Load Cell Signals (Calibrated)
10000
9000
8000
7000
Main Harvester Conveyer Engaged
6000
V)
S
5000
4000
3000
2000
Harvester Entered Trellis Row
1000
I
0
I
I
15
20
25
I
I
30
35
I
I
60
85
I
40
45
50
55
I
70
30
85
90
?
95
100
Time (seconds)
Green Strip Load Cell
Figure 63: First order response characteristics of the field testing.
Unfortunately, it was not possible to determine the dead time of the system because no
positional and time stamp data was collected as to when the harvester entered the row.
The above plot (Figure 63) does illustrate some first order response characteristics.
117
This form of loading can be considered a step-function input, where initial loading can
be considered the initial, steady state condition, and the final value representing a
steady state final value after the step input (Brock et al, 2001). The first time constant of
the system is 63.2% of the rise time, and is derived from Equation 33 below (Brock et al,
2001). This is the generic first order step input response equation for some initial state
xis, and a final state xFS (Brock et al 2001).
-?/
jc(0 = xFS -(Xp3 -xIS)e /t (Brock & Richardson, 2001)
Equation 33
The value of interest was the time constant of the system, to determine the average
delay between the start of harvest and when 63.2% of the full flow rate was obtained.
This information could be used to correct for the delay and shift in yield distribution
maps upon the entering a vineyard row, provided that the dead time of the system was
known as well.
This analysis yielded a time constant of 2.7 seconds for the above trial. The other field
trials produced similarly small time constants, which indicated that there was a low first
order lag time between the start of material flow and full, steady state unloading
conditions. Such a low time delay was the result of the straight forward design of the
harvester, in that harvesting was completed in a single quick step, and most of the delay
was caused by conveying the harvested material to the point of measurement. Due to
the low ground speed during harvesting and a short time delay, there would be little
need to correct for the first order response of the system; unlike in other more
118
complicated harvesting systems such as cereal grain combines (Arslan et al, 2002; Burks
et al, 2002).
7.3.3 Frequency Analysis of In-field Testing Load Cell Signals
The following Fourier transforms (Figure 64) illustrate the frequency responses for all
three load cell signals from a field test (William Falk Farm vineyard row 3). When
compared to the frequency responses of the dynamic in-lab calibration, it is apparent
that field trials contain significantly greater amounts of environmental noise, as seen in
the larger variations, and greater frequency response peaks in the load cell signals. This
was to be expected with an actual harvester that was moving in the field, with an
assortment of mechanical components such as trash blower fans, hydraulic pumps and
motors all operating at once.
Possible sources of noise occurred from the engine, hydraulic systems, picking head
oscillations, leaf extraction fans, grape impact forces and ground conditions. Likewise,
the specific frequencies of each of these sources was variable, due to changing
operating conditions, as well as the complex interactions between the machine and
vibration, such as damping and resonance. For example the large, low pressure tires
and the low ground speed of the harvester may dampen out many of the sharp impulses
caused by the uneven terrain.
119
rgSingJe-Sided Amplitude Spectrum of Load Cell C
10
15
20
25
30
Frequency (Hz)
35
c ig5 Single-Sided Amplitude Specttutn of Load Celt B
40
45
10
50
15
20
25
3D
35
40
45
SO
Frequency (Hz)
, I05 Single-Sided Amplitude Spectrum of Load CeIIA
Frequency (Hs)
Figure 64: Fast Fourier Transforms (FFT) of the unfiltered field testing signals. Load Cell C
(top left), Load Cell B (top right), and Load Cell A (bottom).
There were significantly greater amounts of noise throughout the 50 Hz frequency band
seen above. More specifically, there were frequency spikes at approximately 4, 10, 17
and 25 Hz for Load Cell C, although these 'spikes' represent a large bandwidth of noise.
Load Cell A and B also contained significant amounts of noise. There was a greater
distribution of noise over wider frequency ranges then were seen in the lab testing. This
was most likely the result of both larger amplitudes of vibration, but also a much greater
range of frequencies from the increased number of vibration sources present on the
harvester.
120
7.3.4 Vibration and Acceleration Analysis
The unfiltered 3- axis accelerometer signals are presented below in Figure 65. This plot
illustrates the entire harvest duration for a single vineyard trellis. If these signals were
compared to the in-lab results presented in Figure 56, it is apparent that there were
significantly greater amounts of vibratory noise, which was especially present in the
signal from the accelerometer axis 3, which corresponded to the lengthwise direction of
accelerations. Interestingly, the majority of this vibratory noise was the result of the
grape material impacting onto the top surface of the prototype in the lengthwise
direction. This vibration then transferred through the system to the accelerometer
mounted within the inner frame. The different levels of vibratory noise can be seen in
Figure 65 at the beginning and end of the test, when there was no grape material falling
onto the yield monitor. This lack of falling material occurred approximately 215 seconds
from the start of the test, and drastically reduced the magnitude of the vibratory
oscillations.
121
In Field Testing Results: Unfiltered 3- Axis Accelerometer Signal
1.5
y = -6E-05x +0.1087
R2 = 0.001
0.5 4t-t
LLJi
I
LinJin J Lj «Jil ?.Jr LU
lltltll .i-.i.H.Jt.injtlll
Ai. Uli.
t .ili hi.. ...liLj.lt ,LJi
F
TWVIHW t?t??G??tG???'" ???"G?t"»t?t'?""'?'»??'?'»?G^? p*?t-?»-w^mi>·»{· -*R»r
1
y = -9?-05? - 0.0929
R2 = 0.0062
?
1
1.5
1?-05?- .0992
R2 = 9?-05
0
10
20
30
40
50
60
70
BO
90
100 110 120 130 140 150 160 170 180 190 200 210 220
Time (seconds)
Accelerometer Axis 1
Accelerometer Axis 2
Accelerometer Axis 3
Linear (Accelerometer Axis 3)
Linear (Accelerometer Axis 1)
Linear (Accelerometer Axis 2)
Figure 65: Field testing results of the 3-axis accelerometer signal. Illustrated is the full
harvesting time of a vine trellis.
The tendency for the majority of the accelerometer signals to result from the impacting
vibrations caused by the grape material was the same as was seen during the in-lab
dynamic calibration trials, only with a greater amplitude. This indicated that the grape
material was falling with greater energy, indicating either a greater quantity or the
release velocity of the grapes from the main conveyor belt was higher then during in-lab
testing. This could have resulted from the limited power output of the in-lab test rig
conveyor motor. Other features of the accelerometer signals remained the same as
during dynamic calibration trials, such as the inclination error (offset) caused by the
positioning of the accelerometer within the frame of the conveyor.
122
7.3.5 Field Testing Accuracy Results
After each vineyard row had been harvested, the collected material was weighed for
comparison to the mass measurements obtained from the prototype. Six vineyard rows
were harvested, and of these only three total weights were recorded for comparison
because one pass included up and back on two vineyard rows. The mass of the odd
vineyard rows are marked as not applicable (N.A) in the following results. Table 13
illustrates the results from the in -field testing, including the total weights recorded for
the rows that were measured, as well as the measured mass obtained from the yield
monitor prototype. The last column of the table is the mean error between the actual
mass obtained by weighing the collected material on scales, and the measured mass
recorded by the monitor prototype.
Table 13: Results from the in-field testing utilizing the corrected Load Cell C signal.
Row
Actual Mass (g) Measured Mass (q)
Error
1
2
3
4
5
N.A
482845.00
N.A
402565.00
N.A
213704.51
472862.00
467871.20
451160.80
468787.90
N.A
-2.11
N.A
10.77
N.A
6
423655.00
484349.80
12.53
The mean error for these three trials was 7.1%, using the calibration methodology
outlined. Due to the limited number of trials that were performed, significant
conclusions about overall field accuracy cannot be obtained.
123
Table 14 indicates the results of applying the uncorrected static calibration equation to
the Load Cell C signal obtained during field testing data. The mean error was reduced to
2.4% from the actual mass measured at the end of each row.
Table 14: Results from field testing using the uncorrected static calibration equation as
applied to Load Cell C signal.
Test
Actual Mass (g) Measured Mass (g)
Error (%)
1
2
3
4
5
N.A
482845.00
N.A
402565.00
N.A
199431.00
450293.20
440301.70
428592.00
446219.10
NA
-7.23
N.A
6.07
N.A
6
423655.00
461781.00
,8.26
Table 15 illustrates the mean error analysis for the summed Load Cell A and B signals
calibration method. A mean error of -6.9% was obtained.
Table 15: Results from the in-field testing utilizing summed Load Cell A and C signals,
and assuming an even distribution of mass.
Test
Actual Mass (q) Measured Mass (g)
Error (%)
1
2
3
4
5
N.A
482845.00
N.A
402565.00
N.A
207604.51
436785.60
371760.06
383747.10
393984.50
N.A
-10.55
N.A
-4.90
N.A
6
423655.00
402029.60
-5.38
Although these values appeared to be accurate, the lack of a large sample set reduces
the ability to conclusively provide an indication of accuracy. The standard deviation for
the above calibration trials was s= 8.0%, s= 8.4%, and s= 3.1%, respectively. The 95th
percentile confidence intervals were 20%, 21%, and 8% respectively.
124
7.3.6 Experimental Yield Plots
The following three dimensional yield plots were generated using the Load Cell C signal
and corrected calibration method for each of the harvested vineyard rows. The image in
Figure 66 was the result of the first three rows of the trial, and Figure 67 represents the
last three rows. These trials were not combined into a single yield map because the two
sets of three rows were separated by multiple rows. This was due to the way the
harvester moved up and down the rows to allow for turning and unloading.
The vertical, axis illustrates the mass of grape material in grams that was weighed by the
yield monitor. The horizontal axis was the width of the measured area (in meters), and
the longitudinal axis corresponds to the normalized time when the material was
harvested (in seconds). This normalization was necessary due to the lack of positional
information and the accurate and instantaneous measure of ground speed. The limited
number of rows, as well as the lack of geo-positional data limits the usefulness of the
yield maps. This said, it does illustrate one method of graphically illustrating this yield
distribution.
125
-11000
-10500
"10000
-9500
-9000
-3500
-8000
-7500
-7000
8500
6000
5500
5000
4500
4000
3500
3000
2500
2000
1500
1000
500
Figure 66: First three rows of measured field yield information. Yield distribution map
produced using Surfer8 mapping software. The Kriging interpolation method used to
generate interlining data points.
1-11500
11000
10500
10000
9500
9000
8500
8000
7500
7000
6500
6000
5500
5000
^\
4500
µ 4000
3500
3000
2500
Figure 67: Last three rows of measured field yield information. Yield distribution map
produced using Surfer8 mapping software. The Kriging interpolation method used to
generate interlining data points.
126
7.4
Yield Monitor Mechanical Performance Evaluation
The main purpose of the field testing was to evaluate how well the yield monitor
functioned in an uncontrolled environment, in conditions where it would be expected to
perform to satisfactory levels in industry. Although the system was designed to fit within
the tight tolerances of the preexisting harvester infrastructure, the mechanical
performance of the system was not perfect. While harvesting it was observed that the
grape material would occasionally build up on the yield monitor conveyor due to a
failure to discharge all of the material. This occurred because of a restriction in the
discharge from the vertical extraction fan housing (Figure 68). To allow grape material
to pass under the fan housing, a plastic flap was removed to increase the throat area for
discharged material (Figure 68). The gap between the fan housing and the end of the
yield monitor was approximately 6.5 cm, with tapered ends (Figure 68). This gap could
be increased by further lowering the yield monitor.
ß
®
^^¦^
Figure 68: Installed yield monitor (A), area for discharge (B), vertical extraction fan
housing (C) and plastic flap removed (D)
127
The ¡mages below in Figure 69 were taken while the harvester was processing the grape
material, and indicates the buildup of material at the opening below the vertical
extraction fan. After material started to accumulate it continued to build up until the
yield monitor was clogged and the harvester main conveyor plugged with grape
material. As the current system stands the field performance was fundamentally
flawed, and alterations to the placement or configuration of the prototype would be
needed to correct this deficiency.
?
«
#êf
Figure 69: Identified design flaw in the system. A build up of material (left) caused by
the limited discharge area under the vertical vacuum extraction fan (right)
A primary reason for the buildup of material was the lack of space through which the
material was discharged. Other factors contributed, such as the smooth surface of the
128
conveyor belt; the lack of ribbing on the belt prevented it from pushing the material
through, which would have reduced this clogging tendency. The mass flow rate of grape
material was projected to be 1.70 kg/s (section 5.3.1
Main Conveyor Belt Mass
Flow Rate'), substantially less than the 2.3 kg/s flow rate of material that was
experienced during harvesting of the Chardonnay grapes at William Falk Farms. The
slow discharge rate of the conveyor under the unexpectedly heavy yield conditions was
seen to be major design flaw, although easily correctable.
129
Chapter 8
8.0 Post Experimental Tear-down and Evaluation
Over the course of evaluating the mechanical performance of the yield monitor
prototype, grape material was allowed to accumulate on the device to simulate a lack of
maintenance and cleaning, as well as to investigate how much foreign material built up
and to determine how much contamination occurred within the system.
The images below in Figure 70, Figure 71, and Figure 80 detail the buildup of foreign
matter when the yield monitor was installed within the system as well in detail during
deconstruction of the yield monitor prototype (Figure 71 and Figure 72). As can be seen
there was relatively significant amounts of accumulated material around the vicinity of
the hopper, as well as the end of the monitor in the gap between the suspended frame
and the inner frame, as detailed in Figure 71 (right image).
ESWl
KU
Figure 70: Suspended conveyor yield monitoring system inside the test-bed (left).
Removed system illustrating the buildup of material (right).
130
The gap between the inner frame and the main conveyor frame was the main location
for matter accumulation. There was a limited amount of material located underneath
the conveyor belt, as indicated by the image on the left of Figure 71.
t
1
Figure 71: Conveyor unit with the belting lace removed, and illustrating the buildup of
material only along the outer edges (left). Close up view of this build up material (right).
There was essentially no foreign matter buildup located within the underside of the
main conveyor frame where the instrumentation box was located, as seen in Figure 72
below. The lack of contamination indicated that this location provided adequate
protection from mechanical forces, as well as foreign matter contamination. The lack of
drift or signal artifacts during testing indicated that the buildup of material that did
occur had little to no affect on the overall performance of the system.
131
Figure 72: Underside of the Conveyor unit illustrating the lack of liquid or material
contamination (left) and the instrumentation box completely unscathed (right).
132
Chapter 9
9.0 Conclusions
The grape harvester yield monitoring prototype developed can satisfy the requirements
for a precision viticulture application, as indicated by the preliminary results presented.
Through the course of evaluating the mechanical performance and accuracy of the
system it was determined that the prototype presented good accuracy during dynamic
calibration (average error of -2.7%) and field testing (7.1% average error).
9.2
Evaluation of Dynamic Testing of the Yield Monitor Prototype
Through dynamic testing using the in-lab test rig, it was determined that the load cell
located at the discharge end of the yield monitor not only provided the most accurate
and repeatable results, but the output signals were also ideal for signal processing
filtering techniques to remove unwanted belt lace induced artifacts. The use of the test
rig was instrumental in calibration, and allowed for thorough in-lab testing.
Although the in lab test rig provided invaluable information, lab conditions were not
identical to field testing conditions. Grape material flow rate was limited by the power
requirements of the motor and drive system of the test rig, and consistent, non variable
loading proved hard to obtain.
The test rig was helpful in proving that the conveyor belt system induces a time delay
into the yield monitoring system, but due to the molded cups within the belt and the
method of loading the system did not incorporate any first order response
133
characteristics as were seen in the field testing results. This indicated that the first order
response characteristics could be traced back to the mechanical attachment of the
grape to the vines, picking head operating speed and performance characteristics, as
well as the catching tray response. After the grape material reached the conveyor belt it
could be considered a zero slip condition and add a simple time delay into the first order
system.
9.3
Evaluation of Field Testing of the Yield Monitor Prototype
Although the number of field trials was limited, many conclusions can be drawn. The
current yield monitor prototype installation did not perform adequately in field
conditions due to the tendency for it to clog when exposed to the unexpectedly high
flow rates of grape material. This was mainly the result of necking of product outflow
due to a vacuum extraction fan housing located in close proximity to the yield monitor
prototype. The drive speed of the 12V gear head motor was too low for the loading
rates experienced during in-field testing, and a higher speed would have aided by
increasing the discharge rate of the yield monitor. Although there were limitations in
the mechanical performance of the system, these limitations could be easily corrected
in subsequent prototypes with little sacrifice in the output quality of the load cell
signals. Overall, the performance described in pervious sections indicated that this type
of measurement method and location of installation provided adequate accuracy and
repeatability despite the harsh operating conditions of a grape harvester.
134
It was also determined through analysis of the field trials that there was very little dead
time and harvester lag within the system from when the harvester entered and existed
a vineyard row. The harvesters' dynamic response characteristics indicate that this
system would be ideal for the creation of yield distribution maps, due to the lack of long
dead times and first order delays.
9.4
Evaluation of Instrumentation and Signal Filtering Techniques
Throughout in-lab and field testing, the load cell amplification and null circuitry
functioned flawlessly. No instrumentation failures were encountered throughout
testing, and this includes load cells and amplification circuitry, as well as the 3-axis
accelerometer. No load cell signal drift or abnormal signals were seen over the course
of testing, despite the rigorous testing and rough treatment of the yield monitor
prototype.
The amplifier and null circuit introduced in section 6.6.3 performed without incident,
allowing the static initial loading to be easily removed. This offset was stable and did
not change over the course of testing, leaving load cell calibrations unaffected and
accurate. It should be noted that if a different power supply were used with small
offsets in voltages, all offsets, amplification levels and power supplies would have
changed a small amount. The circuit design used lacked on board voltage regulators
and represents a serious design limitation that should be corrected.
135
Post processing the acquired load cell signals by applying a IHz, 4th order low pass
Butterworth filter proved adequate at removing vibratory noise and improving the
overall quality of the signal. The quality of the filtered load cell signals indicated that
the 3-axis accelerometer would not be required for future designs. Impulse
accelerations were removed from the load cell signals via the Butterworth filter, without
having to resort to identifying accelerations or impacts with the 3-axis accelerometer.
For a given steady state loading of approximately 6000 grams during field testing and
with a linear offset correction of 366 grams for Load Cell C, this impulse loading only
amount to approximately 6.1% of the total yield monitor load cell signal. This was
significantly less than predicted in section 6.4.2
Impact Weighing and Momentum
Transfer'. The differences between the predicted and actual impulse loading was
primarily the result of the assumption that the mass impacting the surface of the
conveyor in solid lumps at the same time. The cushioning effect of a layer of grape
material already present on the belt was neglected. The combination of these factors
resulted in the large difference in predicted and actual loadings. Therefore, the early
design decision to weigh the grape material on the basis of mass instead of momentum
transfer was valid, and provided better results then would have been possible with this
other measurement method.
136
Chapter 10
10.0 Recommendations
10.1 Further Production Recommendations
The following recommendations should be incorporated into any next generation grape
harvester yield monitor prototype/and these are based on the results obtained through
the mechanical and instrumentation analysis performed while testing the prototype.
10.1.1 Overview of Yield Monitoring System
In order to provide the most accurate measurement of material flow rate as possible,
two yield monitoring units would need to be mounted on each harvester to collect and
weigh the material flowing from each conveyor. Likewise, to aid in incorporating this
system into a grape harvester, industry standard communication and measurement
methods should be utilized. Universal Serial Bus (USB) communication was used for the
yield monitor prototype due to availability of components and for convenience. A
CANBUS system should be used to transmit sensor information from the yield
monitoring unit to the control system, as well as to any control devices. The flow
diagram of this system can be seen in Figure 73, and illustrates the general control
methodology and the individual components of the system, as well as their installation
location.
Major components include the instrumentation located within the yield monitors
themselves, the digitization and CANBUS network controllers, and the yield monitor
137
computer and human interface. Each portion of this system will be briefly described in
the following recommendation sections.
Load Cell
Amplifier/
A/D
Offset Null
Converter
Belt Speed
CANBUS
Network
Encoder
Motor Speed
Encoder
Motor Speed
Encoder
CANBUS
Belt Speed
Network
Encoder
Load Cell
Amplifier/
A/D
Converter
Offset Null
DGPS
Receiver
Servo-hydraulic
Regulator Valve
Yield Monitor
Operator
Computer
Interface
Servo-hydraulic
Regulator Valve
Yield Monitor
Mounted
Load Cell Digitizer
Digital Low
Pass Filtering
Calibration
and CANBUS
Network Controller
Harvester Mounted
Equipment
Yield Map
Generation
Storage
Figure 73: Instrumentation and control flow diagram to meet the requirements of
having two yield monitors installed on the harvester, one for each main conveyor.
138
10.1.2 Supervisory Control and Human Interface System
The yield monitor computer can be based on a commercial high performance control
computer (e.g. MicroClient JrDX9) fan less computer. These rugged, versatile and
inexpensive computers feature full PC capacities in a fraction of the cost, size and power
consumption of a traditional computer system. They also have the ability to run
numerous operating systems including DOS, Linux, WinCE and WinXP, and are able to
interface with full USB 2.0 capability (NorhTec, 2008).
This field computer would have to be interfaced with the user, while being able to
control the CANBUS system and interface with the DGPS receiver. Custom graphical
user interface (GUI) software would need to be developed to incorporate all systems
together, while allowing the user to easily access information and control settings.
10.1.3 Power Supply Regulation and Control
Due to the lack of adequate regulation and power supplies on board the yield monitor,
an intelligent automotive style of power supply would need to be integrated into the
system. A DC-DC converter (e.g Opus OPS 120W DC-DC ATX/BTX), that incorporates
such features as regulated power supply lines, startup and shut down controllers and
engine cranking protection would be necessary (Opus Solutions, 2008). This unit would
serve to protect the yield monitor computer, as well as the entire electrical systems of
both the harvester and the yield monitor.
9 MicroClient. MicroClient JrDX fan less computer.
139
10.1.4 Instrumentation and Yield Monitor Control (CANBUS System)
There are numerous changes that are recommended in regards to instrumentation and
control, mainly the use of a CANBUS system. There are aftermarket products such as a
Mantracourt Load Cell Digitizer Module and DSC10 card that digitize load cell signals and
stream them via the CANBUS system to the supervisory controller system, such as a
yield monitor computer (Mantracourt, 2008). Multiple units can be strung together and
provide the data acquisition system for a harvester mounted system.
The most drastic change to the system would be the use of only a single load cell
located at the discharge of the suspended yield monitor conveyor. This would reduce
the cost and complexity of the system, while maintaining higher levels of accuracy. The
addition of a Hall Effect encoder embedded in the conveyor belt on the yield monitor to
record the rotation rate would prove invaluable while filtering belt induced artifacts
from the load cell signal. A Hall Effect encoder incorporated into the drive motor would
also be beneficial in both the feedback control system, but also for determining belt slip.
A digital output Hall Effect sensor signal would be ideal for these applications, and
would not require a dedicated A/C converter channel, further simplifying the
instrumentation requirements of the system.
10.1.5 Instrumentation Amplifier and Offset Removal Circuitry.
The offset removal and amplification method performed without fault throughout all of
the lab and field testing. Slight changes may be required due to the addition of the
CANBUS system and limitations in input voltage to the DSC digitizer module. The most
10 Mantracourt Load Cell Digitizer Module and DSC card.
140
significant change would be the addition of on an on-board voltage regulator to remove
the dependence of the load cell on the voltage levels available from a particular power
supply. This would allow for switching power supplies in the event of a failure without
changing the calibrations on the load cells.
10.1.6 Hydraulic Drive System and Servo Control
A significant limitation in the current prototype was the limited power output and
rotational speed of the 12V gear head drive motor. Clogging of the conveyor belt may
have been avoided if the conveyor belt speed were variable and discharged at a higher
speed. The current motor also put a significant drain on the electrical system of the
harvester, and might have presented a problem if more than one was running.
Hydraulic drive motors feature large power per weight ratios, and would be ideal for
this application due to their power and speed output, as well as the preexisting
hydraulic system on a harvester. A small and lightweight unit would meet projected
power and speed requirements (e.g. a Sauer Danfoss OM M811 hydraulic drive motor),
while an electro-hydraulic flow control valve would be needed to maintain a steady
state rotational speed, despite fluctuations in the harvesters' hydraulic system flow
rates and pressure (Sauer Danfoss, 2008).
The addition of a variable speed drive system opens up greater control possibilities, such
as optimizing the holdup time of the conveyor belt to field conditions. This would
Sauer Danfoss OMM8 hydraulic drive motor.
141
provide the maximum measurement accuracy while preventing excessive buildup and
clogging.
10.1.7 Hardware Design
The mechanical design of the suspended conveyor mechanism can be improved in many
areas, mainly to improve the manufacturability of the system, as well as correcting for
flaws in the current prototype. The main mechanical changes would be to prevent the
buildup of foreign material, increasing the discharge efficiency, and reducing the
material, manufacturing and instrumentation costs. The use of a single load cell, located
at the discharge of the conveyor belt would save significant instrumentation costs, as
well as simplifying construction by no longer requiring the mounting bracket assemblies
for the other load cells. Changing the mounting method of the load cell to an adjustable
connector would reduce the tolerances required for the components and may reduce
manufacturing time and cost.
The hopper should be redesigned to cover the gap between the inner and outer frames
to prevent the buildup of foreign material at this location. This would allow for easier
cleaning of the components, while improving the mechanical performance of the system
without sacrificing measurement accuracy and sensitivity. All major components should
be constructed of stainless steel to resist corrosion and allow for easier cleaning and a
longer life. The belt should be changed from a flat belt to one with ribs molded into the
surface to help push material off of the conveyor mechanism.
142
10.1.8 Control Methodology
The raw signal data should be filtered using a low pass filter, as well as other active
filtering techniques. This signal should then be converted into a mass measurement by
applying a predetermined calibration equation. Periodic zeroing should be preformed
when the yield monitor is unloaded to correct for any material that may be stuck onto
the sensor. If multiple yield monitor belt speeds are used, different calibration or
filtering protocols may be required if periodic waveforms change.
10.2 Performance Testing Recommendations
It would be advantageous to develop a more uniform method of discharging grape
material onto the prototype while testing in the lab. A more uniform flow rate of
material would provide more accurate dynamic calibrations, and help to pinpoint other
operational abnormalities. Likewise, larger and longer periods of loading would help to
better mimic the field conditions.
Obviously, a great deal of field testing would be required to provide a true idea of the
systems in-field performance and overall yield monitoring accuracy. Greater testing into
the effects of rough terrain and vibration on the accuracy of the system would also be
beneficial.
10.3 Signal Processing and Filtering
Incorporating a subtractive filtering methodology on the load cell signal to remove
repeating periodic waveforms would likely improve the quality of the signals. Other
more sophisticated methods of filtering and data analysis and calibration methodologies
143
would likely yield greater sensitivity and measurement accuracies. A minimal amount of
analysis and filtering was done due to the proof- of- concept nature of this project, but
the added benefits of such filtering and post processing warrants further future
investigation.
144
H.OReferenœs
Arslan S; Colvin T (2002). "Grain yield mapping: Yield sensing, yield reconstruction, and
errors". Precision Agriculture. Vol. 3, pp. 135-154.
Bora G; Ehsani R; Lee K; Lee W (2006). "Development of a test rig for evaluating a yield
monitoring system for citrus mechanical harvesters" Congress on Computers in
Agriculture and Natural Resources.
Bramley R (2003). "Being profitable precisely- a case study of precision viticulture from
Margaret River". The Australian & New Zealand Grapegrower & Winemaker. Pp. 84-87.
Bramley R (2001). "Progress in the development of precision viticulture- variations in
yield, quality and soil properties in contrasting Australian vineyards". CSIRO Land and
Water and Cooperative Research Centre for Viticulture.
Bramley R; Hamilton R (2004). "Understanding variability in winegrape production
systems". Australian Journal of Grape and Wine Research. Vol. 10, pp. 32-45.
Bramley R; Williams S (2001). "A Protocol for the construction of yield maps from data
collected using commercially available grape yield monitors". Cooperative Research
Centre for Viticulture, Adelaide.
Brock F; Richardson S (2001). Meteorological Measurement Systems. Oxford University
Press.
Burks T; Fulton J; Shearer S; Sobolik C (2002). "Effects of time-varying inflow rates on
combine yield monitor accuracy". Applied Engineering in Agriculture.
Burks T; Lee W; Schueller J. Wagon-based silage yield mapping system.
Dales D; Hart E; Hioll R; Knowlton A; Perry C; Vellidis G; Wells N. (N. D)"lnstantaneous
accuracy of cotton yield monitors". University of Georgia.
DimensionEngineering. DE-ACCM Buffered ±5g Accelerometer. Accessed April, 2008.
<http://www.robotshop.ca/PDF/DE-ACCM5Gbuffered-accelerometer.pdf>
Durrence G; Hamrita T; Kvien T; Perry C; Vellidis G (2000). Noise réduction in a load cell
based peanut yield monitor using digital signal processing techniques. IEEE. Vol. 2, pp.
1008-1015.
Durrence G; Hamrita T; Vellidis G (1999). "A load cell based yield monitor for peanut
feasibility study". American Society of Agricultural Engineers. Vol. 1, pp. 301-317.
145
Farmscan Canlink 3000 GRM Operators Manual. 2005.
Funk P; Hart W; Moody F; Wilkerson J (2001). "Design and evaluation of a cotton flow
rate sensor". American Society of Agricultural Engineers. Vol. 44(6) pp. 1415-1420.
Gilman A; Bailey D (2005). "High-speed weighing using impact on load cells" IEEE.
Haykin S; Van Veen B (2003). Signals and Systems- Second Edition. John Wiley & Sons,
Inc.
Hayward G. Personal Communication. 2008. University of Guelph, Ontario.
Hennens D; Baert J; Broos B; Ramon H; De Baerdemaeker J (2003). "Development of a
flow model for the design of a momentum type beet mass flow sensor" Biosystems
Engineering. Vol. 85, pp. 425-436.
Jardine J; Muffoletto C, eds. (2001)a. HarvestMaster field data collection tools.
HarvestMaster HM-500 Service Manual.
Jardine J; Muffoletto C, eds. (2001)b. HarvestMaster Field Data Collection Tools.
HarvestMaster HM-570 Service Manual.
Kverneland Group. Grégoire Technical Training (Part 1) (2006-2007). Ref. nb: ANG
001420 11/06.
Lamb D.W; Bramley R; Hall A (2004). "Precision Viticulture- an Australian Perspective".
Viticulture-Living with Limitations.
Linear Technology. LT1168 Low Power, Single Resistor Grain Programmable, Precision
Instrumentation Amplifier. Accessed April, 2008.
<http://www.linear.com/pc/downloadDocument.do?navld=H0,Cl,C1154,C1009,C1045,
P1870,D1608>
Magalhaes P, Cerri D (2007). "Yield monitoring of sugarcane," Biosystems Engineering,
Vol. 96, pp 1-7.
Mantracourt. Load Cell Digitizer and DSC Card. Accessed 2008.
<http://www.mantracourt.co.uk/products/Amplifiers_Digitisers/DSJl>
Moráis R; Fernandes M; Matos S; Serodio C; Rerreira P; Reis M (2008). "A ZigBee multipowered wireless acquisition device for remote sensing applications in precision
viticulture" Computers and Electronics in Agriculture. Vol. 62. pp.94-106.
146
NorhTec. Networking Out of the Box. MicroCNent Jr Dx. Accessed 2008.
<http://www.norhtec.com/products/mcjrdx/index.html>
Omega Engineering (2008). LCAE-20kg Series Single-Point Load Cell Specifications Sheet.
<www.omega.com>
Opus Solutions. Opus 12V DC-DC converter. Accessed 2008.
<http://www.opussolutions.com/index.php?p=product&id=2>
Pelletier G (2001). "Adaptive signal processing for removal of impulse noise from yield
monitor signals". The Journal of Cotton Science. Vol.5, pp. 224-233.
Pelletier G; Shrini K; Upadhyay B (N.D). "Development of a tomato load/yield monitor".
Computers and Electronics in Agriculture. Vol. 23, pp. 103-117.
Piskorowski J; Barcinski T (2008). "Dynamic compensation of load cell response: A time-
varying approach". Mechanical Systems and Signal Processing. Vol. 22 pp. 1694-1704.
Sanitary Design Organization. Design Principles. Accessed 2008.
<www.sanitarydesign.org>
Sauer Danfoss.OMM8 Hydraulic Drive Motor. Accessed 2008.
<http://domweb.sauerdanfoss.com/domdb/SauerLit.nsf/1526f939eb2e202286256c5b0
062ff21/0C9E7F564D09B42BC1256A7F00459E56/$file/520L0346_OML%20OMM%20M
otors%20ORB_TI_Aug-2005_REV%20C.pdf>
Sethuramasamyrija B; Sachidhanantham S; Yen M; Wample R (2007). "Interpolation of
wine grape quality indicators (anthocyanin and brix) and development of differential
harvest attachment". American Society of Agricultural and Biological Engineers.
Simpson's Rule for Integration (2008)
http://thesaurus.maths.org/mmkb/entry.html;jsessionid=EE89B5A6A8C887687DBC936
75A2727B8?action=entryByConcept&id=421&expand=0&msglang=da
Taylor J. (2002). Digital terroirs and precision viticulture: Investigations into the
application of information technology in Australian vineyards. Doctor of Philosophy
Thesis Submission. Australian Centre for Precision Agriculture.
Upadhyaya S; Shafii M; Garciano L..(N. D) "Development of an impact type electronic
weighing system for processing tomatoes". American Society of Agricultural and
Biological Engineers. An ASABE Meeting Presentation. Paper Number 061190.
147
Walter J; Backer L (2003). "Sugarbeet yield monitoring for site-specific farming part 1-
laboratory test and preliminary field". Precision Agriculture. Vol. 4, pp. 421-431.
Webster J, ed. Measurement, Instrumentation, and Sensors Handbook. CRC Press. 1999.
Woods R; Lawrence K (1997). Modeling and Simulation of Dynamic Systems. Prentice
Hall, Inc.
148
Appendix A: Butterworth Filtering Parameters
A Butterworth filtering topography was selected over different filtering techniques for
multiple reasons. The Butterworth filter topography allows for a maximally flat band
pass region. Therefore, the load cell signal frequencies would not have been artificially
amplified or reduced below the cutoff frequency selected. A fourth order filter was
selected because if featured satisfactory levels of frequency attenuation above the
cutoff frequency, while easily built into future analog circuits.
Numerous filtering cutoff frequencies were trial on the raw load cell signals, including
10, 1, 0.5 and 0.1Hz, as illustrated below in Figure 74. Each trial frequency was selected
based on the results of a Fast Fourier Transform (FFT). FFT were preformed on the
filtered results to ensure the effectiveness of each in the frequency domain. Each
filtering cutoff frequency was compared to one another, as illustrated in Figure 74.
The lower cutoff frequencies (0.5Hz and 0.1Hz) removed too much of the variation
within the signal, whereas the higher cutoff frequencies allowed too much signal noise
to pass through. A cutoff frequency of IHz was seen as the best compromise between
removing satisfactory levels of noise while not removing excessive amounts of the
desired load cell signal.
149
Uncalibrated Load Cell C Output Signal using Different Filtering Cutoff Fréquentes
*Ä
?,
i
\i
/
m
> 1
^w
1
?
4D
45
I
1
50
55
60
65
70
75
35
90
95
Time (seconds)
Unfiltered Signal
OHz
Hz
0.5Hz
0.1Hz
Figure 74: Comparison between raw Load Cell C output signal and different filtering
parameters.
Overaggressive filtering, as seen in the 0.1Hz trace in Figure 74 resulted in the near
removal of the variability in the load cell signal. This was unsatisfactory since the
purpose of the yield monitor was to measure these smaller variations in yield.
150
00
Appendix B: MatLab Fast Fourier Transform (FFT)
Code
The following Matlab code was written to provide the frequency spectrum of the
acquired load cell signals.
Matlab Fast Fourier Transform Program Coding
Fs = IOO;
% Sampling frequency
T=l/Fs;
% Sample time
L = 22600;
% Length of signal
t = (0:L-1)*T;
% Time vector
NFFT = 2Anextpow2(L);
% Next power of 2 from length of y
Y = fft(Acc3,NFFT)/L;
f=Fs/2*linspace(0,l,NFFT/2+l);
% Plot single-sided amplitude spectrum.
plot(f,20*log(2*abs(Y(l:NFFT/2+l)))/0.00015258789)
% Amplitude on decibel scale
title('Single-Sided Amplitude Spectrum of Accelerometer Axis 3')
xlabel('Frequency (Hz)')
ylabel('20log|Y(f)r)
These Fourier transform plots were plotted on a decibel scale, as seen in Equation 34
below.
Amplitude = 20 log(F / Vred )
Equation 34
The reference voltage, Vref was the voltage corresponding to one bit increment used by
the analog to digital converter. In other words 10V/216 divisions = 0.00015258789
Volts/division.
Volts per Division
Voltage Range (ADC)
Number of Divisions
Equation 35
151
Appendix C: UnitAnalysis of Mass Measurement
The load cells within the yield monitor prototype record the impulse, J, of the grape
material striking the top surface of the yield monitor, and the force of the held material
on the conveyor, as illustrated in Figure 75.
Figure 75: Accumulated mass of static material as well as impulse force from falling
material was measured by the load cells. The output signal was an additive sum of
these forces.
The area under the impulse-time curve corresponded to units of Newton*seconds, as
seen in Figure 76. The calibration of the system to units of mass flow rate were
simplified by removing the constant acceleration due to gravity, equating the system to
units of mass*velocity. It was assumed that the majority of the loading was the result of
the held material mass and not impulse loading (section 6.4.2
Material Impact Loading').
152
Projected Grape
P(xP
K
M
f(x)
Figure 76: Illustration of the area under the load cell signal, which was a combination of
the impulse grape impact loading, and the held mass on the yield monitor conveyor.
Impulse force is a combination of the mass of the object impacting the yield monitor
prototype, as well as the initial velocity of the grape material. Due to the stability of the
linear belt velocity of the main conveyor and because of the molded buckets, the
release velocity of the grape material was assumed to be the same as the linear velocity
of the belt, and therefore a constant value.
Load cells only measure an applied force acting on them, which in the case of the yield
monitor prototype was a combination of the held material, and the material falling onto
the conveyor, which broken down looks like Equation 36.
G = mGrapea + mGrapeVimpact
Equation 36
The majority of the impacting mass fell in the form of individual grapes, where each
grape impacts with the same velocity, and it was assumed that there was little variation
in grape size, mass, density, and impact coefficient of restitution. This mass velocity
impulse impact can be considered as a steady state offset for a given flow rate of
material.
153
Appendix D: Dynamic Calibration Results
Table 16: Auger calibration trials illustrating the mass of grape material, the duration,
and the calculated average mass flow rate.
Trial
Mass (kg) Duration (s) Average Mass Flow Rate (kg/s)
13.65
13.43
13.39
13.1
12.78
9.84
8.53
7.97
8.1E
8.15
1.387
1.574
1.680
1.605
1.568
Table 17: Auger calibration error analysis descriptive statistics.
_________Error Analysis
Mean
Standard Error
Median
Standard Deviation
1 .563036
0.04825
1.574443
0.107891
Sample Variance
Range
0.011641
0.292855
Minimum
Maximum
Sum
Count
1.387195
1.68005
7.815179
5
Confidence Leve I (95.0%)
0.133965
Table 18: Dynamic calibration trials results indicating the mass of grape material for
each trial, duration and calculated averaged mass flow rate.
Trial
1
2
3
4
5
6
7
8
Mass (kg) Duration (s) Average Mass Flow Rate (kg/s)
26.62
20.94
20.34
19.19
21.97
20.78
18.06
16.65
27.00
20.47
17.31
16.75
22.63
24.75
21.12
16.56
0.986
1.023
1.175
1.146
0.971
0.840
0.855
1.006
154
Table 19: Dynamic calibration trial error analysis descriptive statistics for the average
mass flow rate of each trial.
_________Error Analysis
Mean
Standard Error
Median
Standard Deviation
1.000041
0.042277
0.995815
0.119578
Sample Variance
0.014299
Range
0.335176
Minimum
Maximum
Sum
Count
0.839636
1.174812
8.000332
8
Confidence Level(95.0%)
0.099969
Table 20: Test results from the test-bed dynamic calibration testing utilizing Load Cell C
signal and linear calibration equation.
Test
1
2
3
4
5
6
7
8
Actual Mass (g) Measured Mass (g)
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
16652.00
20490.64
17551.51
17201.98
18032.96
17525.44
17166.54
14220.53
13433.12
Table 21: Load Cell C error analysis descriptive statistics.
Error Analysis
Mean
Standard Error
Median
Standard Deviation
-21.4015
2.555949
-22.5087
7.229315
Sample Variance
Range
52.263
23.52735
Minimum
Maximum
Sum
Count
-29.9325
-6.40515
-171.212
8
Confidence Level(95.0%)
6.043859
155
Error (%)
-29.93
-19.28
-18.22
-6.41
-25.38
-21 .06
-26.97
-23.96
Table 22: Test results from the test-bed dynamic calibration testing utilizing Load Cell B
signal calibration equation.
Test
1
2
3
4
5
6
7
8
Actual Mass (g) Measured Mass (q)
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
16652.00
30624.08
25072.86
22267.66
20833.33
32972.80
30235.57
26062.14
19644.80
Error (%)
13.06
16.50
8.67
7.90
33.36
31.27
30.72
15.23
Table 23: Load Cell B error analysis descriptive statistics.
Error Analysis
Mean
Standard Error
Median
Standard Deviation
19.5893
3.724458
15.86696
10.53436
Sample Variance
Range
110.9727
25.45961
Minimum
Maximum
Sum
Count
7.897572
33.35718
156.7144
8
Confidence Leve I (95.0%)
8.806943
Table 24: Test results from the test-bed dynamic calibration testing utilizing the
combined signals from the Load Cell B and C calibration equations.
Test
Actual Mass (g) Measured Mass (g)
50815.12
42414.11
39334.42
38730.76
44211.57
55451.52
39338.48
40285.44
26624.00
20936.00
20336.00
19188.00
21974.00
20781 .00
18056.00
16652.00
156
Error (%)
47.61
50.64
48.30
50.46
.50.30
62.52
54.10
58.66
Table 25: Combined Load Cell B and C signal error analysis descriptive statistics.
_________Error Analysis
Mean
Standard Error
Median
Standard Deviation
52.82386
1 .862626
50.54852
5.268301
Sample Variance
27.755
Range
14.91787
Minimum
Maximum
Sum
Count
47.60615
62.52402
422.5909
8
Confidence Level(95.0%)
4.40441
Table 26: Test results from the test-bed dynamic calibration utilizing the combined
signals from the Load Cell B and C linear calibration equations, but assuming an even
distribution of loading between load cells. Therefore the full load is half the measured
amount.
Test
Actual Mass (g) Measured Mass (g)
Ratio (g/g)
Error (%)
1
2
3
4
5
6
7
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
25407.56
21207.05
19667.21
19365.38
22105.79
27725.76
19669.24
0.95
1.01
0.97
1.01
1.01
1.33
1.09
4.57
-1.29
3.29
-0.92
-0.60
-33.42
-8.93
8
16652.00
20142.72
1.21
-20.96
Table 27: Combined Load Cell B and C error analysis descriptive statistics.
Error Analysis
Mean
Standard Error
Median
Standard Deviation
-7.28467
4.717237
-1.10956
13.34236
Sample Variance
Range
178.0186
37.98775
Minimum
Maximum
Sum
Count
-33.4188
4.568959
-58.2774
8
Confidence Level(95.0%)
11.1 5449
157
Table 28: Results from the test-bed dynamic calibration testing utilizing the corrected
Load Cell C linear calibration equation.
Test
1
2
3
4
5
6
7
8
,
Actual Mass (g) Measured Mass (g)
24148.20
20601.51
19640.76
20472.96
21184.22
20215.32
17879.31
15871.90
26624.00
20936.00
20336.00
19188.00
21974.00
20781 .00
18056.00
16652.00
Error (%)
-10.25
-1.62
-3.54
6.28
-3.73
-2.80
-0.99
-4.91
Table 29: Corrected Load Cell C error analysis descriptive statistics.
________Error Analysis
Mean
Standard Error
Median
Standard Deviation
-2.69614
1 .627787
-3.16901
4.604078
Sample Variance
21.19753
Range
16.52891
Minimum
Maximum
Sum
Count
-10.2525
6.276388
-21.5691
8
Confidence LevelQ5.0%)
3.849106
Table 30: Test results from the test-bed dynamic calibration testing utilizing the
corrected Load Cell B calibration equation.
Test
1
2
3
4
5
6
7
8
Actual Mass (g) Measured Mass (g)
26624.00
20936.00
20336.00
19188.00
21974.00
20781 .00
18056.00
16652.00
25127.75
20487.69
18600.99
17164.83
27302.33
26018.90
20519.51
15978.14
158
Error (%)
-5.95
-2.19
-9.33
-11.79
19.52
20.13
12.01
-4.22
Table 31: Corrected Load Cell B error analysis descriptive statistics.
Error Analysis
Mean
Standard Error
Median
Standard Deviation
2.272304
4.576588
-3.2028
12.94455
Sample Variance
Range
167.5613
31.91787
Minimum
Maximum
Sum
Count
-11.7867
20.13114
18.17843
8
Confidence Level(95.0%)
10.82191
Table 32: Test results from the test-bed dynamic calibration testing utilizing the
combined signals from the corrected Load Cell B and C calibration equations, and
corrected offset.
'
Test
Actual Mass (g) Measured Mass (g)
Error (%)
1
2
3
4
5
6
7
26624.00
20936.00
20336.00
19188.00
21974.00
20781.00
18056.00
24992.35
20875.18
20306.09
21498.86
28608.85
26066.74
15452.48
-6.53
-0.29
-0.15
10.75
23.19
20.28
-16.85
8
16652.00
15724.03
-5.90
Table 33: Combined and corrected Load Cell B and C error analysis descriptive statistics.
_________Error Analysis
Mean
Standard Error
3.062584
4.911664
Median
Standard Deviation
-0.21932
13.89228
Sample Variance
Range
192.9955
40.04016
Minimum
Maximum
Sum
Count
Confidence Leve I (95.0%)
159
-16.8486
23.19161
24.50067
8
11.61424
Appendix E: Field Testing Results
Table 34: Results from the in-field testing utilizing the corrected Load Cell C signal.
Test
1
2
3
4
5
6
Actual Mass (g) Measured Mass (g)
N. A
482845.00
N.A
402565.00
N.A
423655.00
213704.51
472862.00
467871.20
451160.80
468787.90
484349.80
Error
N. A
-2.11
N.A
10.77
N.A
12.53
Table 35: Load Cell C in-field testing error analysis descriptive statistics.
_________Error Analysis
Mean
Standard Error
Median
Standard Deviation
7.063762
4.61552
10.77128
7.994316
Sample Variance
63.90908
Range
14.64238
Minimum
Maximum
Sum
Count
-2.11119
12.53119
21.19129
3
Confidence Level(95.0%)
19.85898
Table 36: Results from field testing uncorrected Load Cell C signal.
Test
Actual Mass (g) Measured Mass (g)
1
2
3
4
5
N.A
482845.00
N.A
402565.00
N.A
199431.00
450293.20
440301.70
428592.00
446219.10
6
423655.00
461781.00
.
Error (%)
N. A
-7.23
N.A
6.07
N.A
8.26
Table 37: Uncorrected Load Cell C in-field testing error analysis descriptive statistics.
_________Error Analysis
Mean
2.366649
Standard Error
4.839068
Median
Standard Deviation
6.072675
8.381512
Sample Variance
Range
70.24974
15.48532
Minimum
Maximum
Sum
Count
Confidence Level(95.0%)
160
-7.22902
8.256295
7.099947
3
20.82083
Table 38: Results from the ¡?-field testing utilizing summed Load Cell A and C signals,
and assuming an even distribution of mass.
Test
1
2
3
4
5
6
Actual Mass (g) Measured Mass (g)
N.A
482845.00
N.A
402565.00
N.A
423655.00
Error (%)
N.A
-10.55
N.A
-4.90
N.A
-5.38
207604.51
436785.60
371760.06
383747.10
393984.50
402029.60
Table 39: Combined Load Cell A and C in-field testing error analysis descriptive statistics.
^_
Error Analysis
Mean
Standard Error
Median
Standard Deviation
Sample Variance
Range
Minimum
Maximum
Sum
Count
-6.94262
1.806449
-5.37906
3.128862
9.789779
5.641358
-10.5451
-4.90372
-20.8279
3
Confidence Level(95.0%)
161
7.772525
Appendix F: Engineering Drawings
I=^T
~49. ?? 10. [D-t53Li.2 MJ. LO-
485.6' :0.1a
'¦El. 2 «?.??
533.2 10.10
Figure 77: Engineering drawing of the reference dimensions of the assembled yield
monitor prototype.
162
=t±
¿'?. ü tc.jü2?. Q ±5- io
_i
,a *<>..
-,7CB1Q i3. LE-Í1G-0 i3-12
Il
II
_á.O to. JQ-
32.0 te, :o
t
r
I
?
52 1 0 î 1^ ¦ ^ ü
2S.0 to. jo 3¿»C io. JO
: . § «i. ip36,5 iO.:o
- i 'J 2.(3 *ü.iü- ¿?? ,Q iiì.in-22"E „5 id. ig-
-ZZO. ^ to.-o-
-±93.5 iÛ.:Q60-0 io. tB35-0 to. (C5,5 to. lo—
5, Q »ß.?ß: 70n £D.]ß-i
*—?—*—?—?—F
FJ.3Ü.Ü i«. ID
+
220.0 *?.?a
4Ì0.O 10. IO
-33 = 0 io. lo
+
3ZE-. O to. ic
tzt
-é*^
Pt . U io. LO
Figure 78: Engineering drawing of the top outer frame assembly illustrating the main
components of the design.
163
?0 . 0 *e
L
G
32.0 su
I 26.3
431 .B tP.ia
-5 1 ! .G ±ß. so
2^8.9 to.jo
2Q5-0 ??.10
2*7.5 iO.io
22Z.7 ío.íO
179, Z to. to
i ¡5.7 lO.iQ
56. 7 to.. o
¦1! . / *o..
^
1S?
S5
7"X"
^>4
l·
#=
7^
\
fe^
-RS. O tO. ?e
-ZE. 5 to.LO
Figure 79: Engineering drawing of the motor mounts end plate. Included are detailed
sections of the design.
164
\
^>c-
/
\
i
\
V
\
\
;/-
«
/
X
WE
\j ^ ¿?
-rVk-T-t
/
-A
¥
X
X
m
^=Wl
^L
W
I/
;
/1
K
Figure 80: Engineering drawing of the small end plate. Included are detailed sections of
the end plate assembly.
165
S3. S ±a.
4L i^iO-Jü—
2^.5 ?a. :o29.0 ±o.:o-
Figure 81: Engineering drawing of the tension and bearing support assembly.
166
i SC .O ta. io
Figure 82: Engineering drawing of a conveyor roller tube section.
e 73. O ±0.10
-&-
32.7 io.ïo
??1
Za 7 IO. ID ["
il
Figure 83: Engineering drawing of the V-grove roller insert that allows the conveyor belt
to track correctly.
167
] OVj1, C *o.¡g—
-21Z -Ct .O.iû—
N
-0-1.3 O.LI!
^-'í'l.Ol
Figure 84: Engineering drawing of the Omega Engineering LACE-20kg aluminum
cantilever load cell.
168
M
Figure 85: Top UHMW polyethylene plate of the yield monitor conveyor.
169
Документ
Категория
Без категории
Просмотров
0
Размер файла
5 917 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа