Upload
liam-walls
View
57
Download
0
Embed Size (px)
DESCRIPTION
A recent innovation in underground mining operations is the use of automated Load-Haul-Dump (LHD) vehicles. These vehicles are able to transport ore from a stope toan ore-pass and return without the direct in involvement of a human operator. Aninteresting, and unexpected, problem that has arisen with the introduction of thesevehicles is that the haul road on which they operate can become heavily degradeddue to a lack of feedback on road condition. This results in increased maintenance and downtime. On non-automated LHD vehicles, the problem is circumvented by the on-board operator who senses road condition by ride quality and reports back to his or her supervisor when the road becomes unacceptably degraded.This thesis develops a simple tool, called the Road Surface Quality (RSQ) monitor,that aims to report road condition to the automated vehicle’s supervisory controlsystem. The monitor is intended to provide feedback on road condition analogousto that provided by an on-board operator. This feedback can in turn be used toguide decisions such as the operating speed over different sections of road and when to perform road-maintenance such as re-grading.The methodology used to determine road condition is to capture the vertical accel-eration of the vehicle chassis and derive a measure of the “bounce energy” of thevehicle as it traverses the road to serve as a proxy for the “quality of ride”. Theapproach exploits the fact that automated LHD vehicles make repeatable passes from run-to-run. In particular, the wheel tracks and speed are consistent duringeach transit from the stope to the ore-pass and back again. This makes possible thedirect comparison of signals measured from run-to-run.The main conclusion of the thesis is that it is possible to differentiate betweenroads of different condition using measurements of vertical acceleration on automated LHDs. This conclusion is supported by several case studies based on installation of the monitor on the Dynamic Automation Systems (DAS) Autotram LHD in operation at Olympic Dam Mine. Data collected from the monitor shows the system is also capable of detecting abnormally rough operation of the vehicle, e.g. unnecessarily heavy collisions of the bucket with the environment. A detailed plan for further testing of the monitor is also proposed.
Citation preview
The University of Queensland
Quantifying Road Surface QualityFor Underground Haul Roads
By
Liam D. Walls
B.E. (Mechanical, Hons)
A thesis submitted for the degree of
Master of Philosophy
Principal Supervisor: Assoc. Professor P. R. McAree
Associate Supervisor: Professor H. Gurgenci
Division of Mechanical Engineering,
School of Engineering,
The University of Queensland,
Australia.
July 2006
c© Copyright 2006
by
Liam D. Walls
i
I hereby declare that this submission is my own work and to
the best of my knowledge it contains no material previously
published or written by another person, nor material which
to a substantial extent has been accepted for the award of
any other degree or diploma at UQ or any other educational
institution, except where due acknowledgement is made in the
thesis. Any contribution made to the research by colleagues,
with whom I have worked at UQ or elsewhere, during my
candidature, is fully acknowledged.
I also declare that the intellectual content of this thesis is the
product of my own work, except to the extent that assistance
from others in the project’s design and conception or in style,
presentation and linguistic expression is acknowledged.
Liam D. Walls
ii
Abstract
A recent innovation in underground mining operations is the use of automated Load-
Haul-Dump (LHD) vehicles. These vehicles are able to transport ore from a stope to
an ore-pass and return without the direct in involvement of a human operator. An
interesting, and unexpected, problem that has arisen with the introduction of these
vehicles is that the haul road on which they operate can become heavily degraded
due to a lack of feedback on road condition. This results in increased maintenance
and downtime. On non-automated LHD vehicles, the problem is circumvented by
the on-board operator who senses road condition by ride quality and reports back
to his or her supervisor when the road becomes unacceptably degraded.
This thesis develops a simple tool, called the Road Surface Quality (RSQ) monitor,
that aims to report road condition to the automated vehicle’s supervisory control
system. The monitor is intended to provide feedback on road condition analogous
to that provided by an on-board operator. This feedback can in turn be used to
guide decisions such as the operating speed over different sections of road and when
to perform road-maintenance such as re-grading.
The methodology used to determine road condition is to capture the vertical accel-
eration of the vehicle chassis and derive a measure of the “bounce energy” of the
vehicle as it traverses the road to serve as a proxy for the “quality of ride”. The
approach exploits the fact that automated LHD vehicles make repeatable passes
iii
from run-to-run. In particular, the wheel tracks and speed are consistent during
each transit from the stope to the ore-pass and back again. This makes possible the
direct comparison of signals measured from run-to-run.
The main conclusion of the thesis is that it is possible to differentiate between
roads of different condition using measurements of vertical acceleration on auto-
mated LHDs. This conclusion is supported by several case studies based on instal-
lation of the monitor on the Dynamic Automation Systems (DAS) Autotram LHD
in operation at Olympic Dam Mine. Data collected from the monitor shows the
system is also capable of detecting abnormally rough operation of the vehicle, e.g.
unnecessarily heavy collisions of the bucket with the environment. A detailed plan
for further testing of the monitor is also proposed.
iv
Acknowledgements
I would like to thank my supervisor Ross McAree for his limitless support and
infinite wisdom and for pulling me out of the gutter, handing me a hot cup of coffee,
a pencil stub and a piece of torn napkin whilst muttering the wise words “Better
get started. You’ve got a ways to go”.
This project was conducted through CRCMining who provided me with a Postgradu-
ate Scholarship and an outstanding environment in which to undertake Postgraduate
study.
The project would not have been possible without the financial and practical support
provided by Charles McHugh during his Tenure with WMC. A big thank you to all
the people at Olympic Dam, including those from WMC, BHP and DAS who were
a great help during the experimental trials.
I would also like to thank my girlfriend Jess for reading and rereading a thesis
that she doesn’t understand nor care to understand and for providing me with
nourishment as required; and my parents for living 1600 kilometres away and only
visiting occasionally.
Thanks also to Michael Little for his help with troubleshooting hardware and for
giving up precious time for preliminary testing; Mark Calder for his seemingly lim-
itless knowledge of electrical systems and for passing that knowledge onto me; and
Rowan Gollan for the latex template on which this document is based.
v
Contents
1 The rationale for monitoring underground road surface quality 5
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Olympic Dam Mine (ODM) . . . . . . . . . . . . . . . . . . . 7
1.2.2 The RH2900 Load Haul Dump (LHD) vehicle . . . . . . . . . 8
1.2.3 The Dynamic Automation Systems (DAS) Autotram system . 8
1.3 Motivation for monitoring road surface condition . . . . . . . . . . . 11
1.3.1 LHD productivity . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Abnormal event detection such as collisions . . . . . . . . . . 15
1.3.3 Grader scheduling . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2 Methods for measuring and interpreting road roughness 17
2.1 Road roughness measurement . . . . . . . . . . . . . . . . . . . . . . 17
vi
2.2 Road roughness measurement standardisation and the International
Roughness Index (IRI) . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 Definition of the IRI . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Calculation of the IRI . . . . . . . . . . . . . . . . . . . . . . 21
2.2.3 Summary of the IRI . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Power Spectral Density (PSD) methods . . . . . . . . . . . . . . . . . 23
2.3.1 Hypotheses for simplified road roughness description . . . . . 23
2.3.2 Homogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.3 Isotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.4 Transient identification . . . . . . . . . . . . . . . . . . . . . . 25
2.3.5 The Gaussian assumption and problems associated with non-
stationary PSDs . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.6 PSD shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.8 Response-type versus profilometric methods for underground
haul roads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Design and development of the road surface quality monitor 31
3.1 An overview of the Autotram control system . . . . . . . . . . . . . . 31
3.2 Integration of the RSQ monitor with the Autotram system . . . . . . 33
3.3 RSQ monitor hardware . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.1 Accelerometers . . . . . . . . . . . . . . . . . . . . . . . . . . 38
vii
3.3.2 Anti-aliasing filter . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.3 Analog-to-digital converter . . . . . . . . . . . . . . . . . . . . 39
3.3.4 The delay-boxes . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3.5 Compact flash data storage . . . . . . . . . . . . . . . . . . . 41
3.3.6 Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 RSQ monitor software . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4.1 RSQ monitor software architecture . . . . . . . . . . . . . . . 45
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4 Monitoring road surface quality 47
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Definition of the RSQ . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3 Choice of data for computing the RSQ index . . . . . . . . . . . . . . 50
4.3.1 Which accelerometer? . . . . . . . . . . . . . . . . . . . . . . 52
4.3.2 Which tramming direction? . . . . . . . . . . . . . . . . . . . 54
4.4 The tramming route . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5 Road surface quality along the tramming route . . . . . . . . . . . . . 61
4.6 Road degradation with number of runs . . . . . . . . . . . . . . . . . 63
4.7 The RMS-RSQ: A broad measure of road surface quality . . . . . . . 66
4.8 Case studies in the use of the RSQ index . . . . . . . . . . . . . . . . 68
4.8.1 Case study 1: Day-shift of October 4, 2004 . . . . . . . . . . . 68
viii
4.8.2 Case study 2: Day-shift of October 18, 2004 . . . . . . . . . . 70
4.8.3 Case study 3: Day-shift of October 19, 2004 . . . . . . . . . . 73
4.9 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . 75
5 The RSQ monitor as a collision detection tool 77
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Operator feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3 LHD UL038 - 15/16th of June, 2005 . . . . . . . . . . . . . . . . . . 79
5.3.1 Background information . . . . . . . . . . . . . . . . . . . . . 81
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.4.1 Identifying cause of collisions . . . . . . . . . . . . . . . . . . 85
5.4.2 Estimating Collision Intensity . . . . . . . . . . . . . . . . . . 90
5.5 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . 93
6 Conclusions and recommendations for future work 94
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.1.1 Aims of field trials . . . . . . . . . . . . . . . . . . . . . . . . 95
6.1.2 Experiments addressing longitudinal study aims . . . . . . . . 98
6.2 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Bibliography 105
A Data analysis methodology 106
ix
List of Tables
2.1 A comparison of response-type and profilometric methods for mea-
suring road roughness *[1]. . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 The nine parameters used to completely describe a road surface [2]. . 28
x
List of Figures
1.1 RH2900 autonomous LHD at Olympic Dam Operations. . . . . . . . 6
1.2 Stope mining process. (Figures taken from http://www.uraniumsa.org) 9
1.3 The RH2900 LHD. Top view shows articulation motion. Side view
shows various bucket positions. Figure taken from the CAT RH2900G
data sheet without permission. . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Time spent in each gear vs. number of buckets/hr. Buckets/hr is
shown for current gear configuration. . . . . . . . . . . . . . . . . . . 14
2.1 The Golden Car model [3]. . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Comparison between measured coherency functions and those calcu-
lated assuming isotropy. (A: Motorway, B: Minor road, C: Paving)
[4] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Road elevation domain vs. road spatial acceleration domain (raw
data and statistical distribution) [5]. . . . . . . . . . . . . . . . . . . 26
2.4 Statistical distributions of spatial acceleration and transient ampli-
tudes [2]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Four PSDs for roads of different roughnesses [6]. . . . . . . . . . . . . 29
xi
3.1 A schematic representation of the Autotram network with the RSQ
monitor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 The Autotram control room supervisory control system (SCS) work-
station. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 The relationship between the Autotram OBC and SCS and the RSQ
monitor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 The RSQ index as displayed on the Autotram operator control console. 35
3.5 The RSQ monitor hardware. . . . . . . . . . . . . . . . . . . . . . . . 36
3.6 The LHD with accelerometer and RSQ monitor mounting locations
shown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.7 The method of mounting the accelerometers to the LHD. A lid is then
used to seal the protective case and prevent internal corrosion. . . . . 38
3.8 Signal processing in the RSQ monitor. . . . . . . . . . . . . . . . . . 40
3.9 Power flow wiring diagram for the RSQ monitor. . . . . . . . . . . . . 43
3.10 A diagrammatical representation of the finite state machine RSQ
monitor software. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1 An example of smooth and rough underground haul roads. The light
source is displaced vertically downwards to allow road surface irreg-
ularities to present more clearly. . . . . . . . . . . . . . . . . . . . . . 48
4.2 Bode plot showing the frequency response of the RSQ filter applied
on squared acceleration measurements. . . . . . . . . . . . . . . . . . 50
xii
4.3 A plan view of the tunnels that make up the 58 Orange 57 haul
route with draw-point and ore-pass marked. The path taken by the
Autotram is shown by the blue dashed line. The portion of the haul
route between the green triangle markers is referred to as the ‘main’
section of the tramming route. . . . . . . . . . . . . . . . . . . . . . . 51
4.4 Front vs rear accelerometer frequency information. Empty-tram data. 53
4.5 Front vs rear accelerometer frequency information. Full-tram data. . . 56
4.6 A plan view of the path the LHD takes along the 58 Orange 57 haul
route (see Fig. 4.3 for a map of the route itself). O designates the
coordinate origin and dotted horizontal lines designate sections of
curved road. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.7 Data from the LHD during the empty-tram runs on the night-shift of
the 16th/17th October 2004. Two separate y-axes are shown for each
figure. Vertical lines represent important features. The solid lines
represent the boundaries of the area of road that is used to obtain
results, the dashed lines represent the zones where gear changes occur
and the dotted lines represent the segments of curved road. . . . . . . 59
4.8 A 3D realisation of the RSQ measure (from the night-shift of the
16th/17th October 2004) mapped against the haul route. The x and
y-axes are the x and y haul route map coordinates respectively, and
the z-axis is the RSQ corresponding to the x-y coordinate. . . . . . . 62
4.9 Waterfall plots of the RSQ versus sample number with increasing runs
for the entire night-shift on the 16th/17th of October 2004. Each plot
is an average of five runs; the first plot represents runs 1 to 5 etc. Both
plots are presented because they clearly show different aspects of the
RSQ trend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
xiii
4.10 A waterfall plot of the differential RSQ (minus the initial RSQ) versus
sample number for the main section of road (between the vertical solid
black markers in Fig. 4.7) with increasing runs for the entire night-
shift of the 16th/17th of October 2004. Each plot is an average of
five runs; the first plot represents runs 1 to 5 etc. . . . . . . . . . . . 65
4.11 The RMS of the RSQ vs run number for every run of the night-shift
on the 16th/17th of October 2004. A moving average curve is plotted
in red to help identify the trend in the data. . . . . . . . . . . . . . . 66
4.12 The haul route road at the end of the day shift on the 4th of October
2004. The light source is displaced vertically downwards to allow road
surface irregularities to present more clearly. . . . . . . . . . . . . . . 68
4.13 A waterfall plot of the differential RSQ over the entire day shift on
the 4th of October 2004. . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.14 The RMS-RSQ for every run over the shift on October 4, 2004. . . . 70
4.15 A 2D waterfall plot of the RSQ over the entire day shift on the 18th
of October 2004. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.16 A plot of the rear accelerometer data (converted to g) from run 60 on
the day shift of October 18, 2004 at the time of the large RSQ spike. 72
4.17 2D waterfall plots of the RSQ data obtained on the 19th of October
2004. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.18 The RMS-RSQ of the RSQ data obtained during the day shift of the
19th of October, 2004. . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.1 Damage to transmission cover plate. . . . . . . . . . . . . . . . . . . 80
5.2 Damage to DAS remote control logic box and main steering hose. . . 80
xiv
5.3 Grizzly with rails to prevent LHD damage. . . . . . . . . . . . . . . . 82
5.4 LHD and grizzly. Detail shows how bucket tip collides with edge of
Grizzly. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.5 Longitudinal LHD acceleration vs time (hrs) for the 15th and 16th of
June. Points of high acceleration are shown: green = > 10g, yellow
= > 15g, red = > 20g, where 1g is approximately 9.81m/s2. The X
is the location of the grizzly center and the traced circular line has
a radius equivalent to the distance from the LHD’s coordinate centre
to the grizzly when the LHD bucket is touching the grizzly edge. The
acceleration spikes marked with a cross are spikes deemed to be a
result of something other than grizzly collision. . . . . . . . . . . . . . 87
5.6 Longitudinal acceleration vs Time of UL038 over two days. Vertical
dividers represent shift change and ’S’ represents a LHD service. Time
is in 24 hour time in the format HH:MM. . . . . . . . . . . . . . . . . 89
5.7 A time-scaled plot of y-acceleration Vs. time for one of the col-
lisions. Sample points, represented by hollow circles, are 1/200sec
apart. Time units are HH:MM:SS:SSSS in 24 hour time. . . . . . . . 92
5.8 A time-scaled plot of y-acceleration Vs. time for the positive accelera-
tion spike. Sample points, represented by hollow circles, are 1/200sec
apart. Time units are HH:MM:SS:SSSS in 24 hour time. . . . . . . . 92
xv
Nomenclature
α A unit-less scaling factor used to scale the RSQ to a more palatable magnitude
z̈k The measured acceleration at sample k
Dex The distance travelled in x gear during the empty-tram
Dfx The distance travelled in x gear during the full-tram
N The moving average window size
RSQj The RSQ at sample j
T The total time to transport one bucket of ore from the draw-point to the
ore-pass
Tb The time taken to bog the ore
Td The time taken to dump the ore
Te The time taken to tram back to the draw-point
Tf The time taken to tram the ore to the ore-pass
Tex The time spent in x gear during the empty-tram
Tfx The time spent in x gear during the full-tram
xvi
Tmin The minimum possible time to transport one bucket of ore from the draw-
point to the ore-pass.
Vex The LHD speed in x gear during the empty-tram
Vfx The LHD speed in x gear during the full-tram
xvii
Abbreviations
ARS average rectified slope
AD analogue-to-digital
BHPB BHP Billiton
CAT Caterpillar
CRCMining Cooperative Research Center for Mining
DAS Dynamic Automation Systems
EBX embedded board expandable
EMR electro-magnetic radiation
FIR finite impulse response
GUI graphical user interface
HUD heads up display
IRI International Road Roughness Index
IRRE international road roughness experiment
IRSQ instantaneous road surface quality
1
KE kinetic energy
LAN local area network
LHD load-haul-dump
OBC on-board controller
ODM Olympic Dam Mining Operations
OS operating system
PDF probability density function
PSD power spectral density
RMS root mean square
RPM revolutions-per-minute
RR road roughness
RSQ road surface quality
SCS supervisory control system
UDP user datagram protocol
WMC Western Mining Corporation
2
Terminology
• Articulate The action of a load-haul-dump (LHD) vehicle steering about it’s
articulation point.
• Damage Fatigue or direct stress felt by a LHD.
• Differential RSQ The RSQ minus the RSQ at the start of the shift. This is
a convenient way of looking at the increase in RSQ.
• Draw-point The area at the bottom of the Stope from where the ore is
extracted.
• Empty-tram The trip from ore-pass to draw-point. Named empty-tram be-
cause of the lack of payload.
• Full-tram The trip from draw-point to ore-pass. Named full-tram because of
the payload.
• Grizzly The steel grid that is placed over the ore-pass to prevent oversize
passing.
• Haul A LHD tramming with a full bucket is said to be hauling.
• Haul route The road connecting the draw-point and ore-pass traversed by
an LHD.
3
• Ore-pass Where ore is dumped to be taken to the surface via the underground
rail system through.
• Oversize The name given to large, insufficiently broken ore, too large to fit
through the grizzly and proceed to the next stage in the stope mining process.
• Road Roughness Variation in elevation of road profile.
• Run The trip from the draw-point to ore-pass and back again (a run).
• Stope An area in an underground mine where ore is blasted for extraction.
• Tram The action of a LHD vehicle moving from one place to another (tram-
ming).
• Tramming route See Haul route.
4
Chapter 1
The rationale for monitoring
underground road surface quality
1.1 Introduction
Load-haul-dump (LHD) vehicles, see Fig. 1.1, are used in underground mining to
transport blasted ore from a draw-point to an ore-pass. They are critical production
units at most underground mines, and the ongoing imperative to increase production
has lead to the development, in the last five years, of fully autonomous LHD systems.
This thesis makes contributions to the problem of monitoring the condition of the
roads on which these vehicles operate with the specific aim of developing a road
quality monitor suitable for autonomous LHDs. The role of this monitor is to
provide the automation system with feedback on road condition in a similar fashion
to the way the ‘roughness of the ride’ provides information of road condition to a
human operator physically present on the machine. In the absence of such feedback,
it has been found that these automated vehicles are often run for extended periods
on poor quality roads to the detriment of the mechanical integrity of the machine.
The specific objective of this thesis is to establish whether it is possible to monitor
road condition with sufficient fidelity to determine degrading road surface quality
5
1.1. INTRODUCTION
Figure 1.1: RH2900 autonomous LHD at Olympic Dam Operations.
using measurements of vertical acceleration made from the chassis of LHD vehicle.
An underlying assumption for the work is that there exist critical levels of road
quality at which the road must either be graded or the travelling speed reduced to
maintain control of the vehicle and preserve mechanical integrity of the machine.
Operators do this implicitly by gauging the roughness of ride, but an automation
system requires explicit guidance. The system developed in this thesis is intended to
provide such guidance. When passing over a rough section of road, for example, the
monitor would advise the automation to change down a gear (or two!) to preserve
good control and avoid damage.
The remainder of this chapter gives background information for the problem and
provides a more detailed motivation for the work. Chapter 2 reviews the various
methods that have been proposed for the description of road roughness against the
6
1.2. BACKGROUND
specific objective or developing a road roughness measure for underground load-
haul-dump vehicles.
Chapter 3 describes the hardware and software developed to monitor road surface
quality. This system integrates with the Dynamic Automation Systems (DAS) Au-
totram control system, broadcasting road surface quality information over the local
area network connecting the vehicle to the surface. This road condition information
is displayed in realtime on the surface control console, allowing degrading roads to
be visualised and monitored.
Chapter 4 develops a measure of road condition and reports on the performance
of this measure on data obtained from experimental trials at Olympic Dam. It is
shown that the approach can provide information on road surface quality.
Chapter 5 explores the possibility of using the monitor to detect collisions between
the LHD and the operating environment with a view towards alerting the Autotram
system or remote operators to potentially significant damage events.
Chapter 6 gives recommendations from the study including the proposal for a lon-
gitudinal trial of the technology.
1.2 Background
1.2.1 Olympic Dam Mine (ODM)
The experimental component of this thesis was conducted at Olympic Dam Mining
Operations (ODM), located 560km northwest of Adelaide near Roxby Downs in
South Australia. The ODM orebody was discovered in 1975 and production began
in 1988. The mine has the fourth largest known remaining copper resource, the
fourth largest gold resource, and the largest uranium resource in the world. ODM is
7
1.2. BACKGROUND
also the largest underground mine in Australia. In 2004, ODM produced 224, 000t of
copper, 4, 400t of uranium, 87, 600ounces of gold and 868, 000ounces of silver.
ODM comprises a fully-integrated underground mine and above-ground metallurgi-
cal complex. All processing is done on-site to remove and contain the high-levels of
radioactivity present in the ore.
ODM utilises the stope mining method, which involves progressively blasting rock
in layers from the ore-body to form a cavity, see Fig. 1.2. The rock falls through an
opening, at the base of the cavity, to a draw-point where it is extracted by an LHD.
The LHD hauls ore from the draw-point to an ore-pass, where the ore is dumped.
The ore is then transported using an underground rail system to a high-speed lift,
where it is taken to the surface and along a series of conveyors to the stockpile,
ready for processing.
1.2.2 The RH2900 Load Haul Dump (LHD) vehicle
The LHD used in this study is an RH2900 manufactured by Caterpillar-Elphinstone.
The RH2900 is an articulated vehicle: steering is achieved by the actuation of hy-
draulic rams that pivot the vehicle about a central pivot point, see Fig. 1.3. The
axles and chassis are rigidly joined; suspension is provided entirely by the air-filled
tyres. The RH2900 weighs 50t when the bucket is empty and up to 67t when fully
loaded.
1.2.3 The Dynamic Automation Systems (DAS) Autotram
system
The automated LHD is a retrofit to a standard RH2900 LHD that allows it to
operate autonomously. This retrofit is a product of Dynamic Automation Systems
8
1.2. BACKGROUND
(a) Before Blasting (b) After Blasting
Figure 1.2: Stope mining process. (Figures taken from http://www.uraniumsa.org)
(DAS), a subsidiary of Caterpillar. The retrofit involves installing multiple sensors,
an on board computer-based control system, and communications equipment.
The retrofitted vehicle has three modes of operation: (i) manual mode; (ii) co-pilot
mode; and (iii) Autotram mode. Manual mode is a pure teleoperation mode that
allows the operator full control of the LHD using the joystick. The operator obtains
information from an on-screen display that includes:
• Real-time video footage from cameras mounted on the front and rear of the
LHD.
• Real-time updates of walls and other obstacle locations relative to the LHD.
9
1.2. BACKGROUND
Figure 1.3: The RH2900 LHD. Top view shows articulation motion. Side view showsvarious bucket positions. Figure taken from the CAT RH2900G data sheet withoutpermission.
• Maps of the tramming route.
• LHD dashboard information such as speed, gear, battery voltage, oil temper-
ature, headlight status, engine RPM, angle of articulation, and bucket status
(up or down).
10
1.3. MOTIVATION FOR MONITORING ROAD SURFACE CONDITION
In Autotram mode, the LHD follows a pre-set path and uses its onboard sensors to
correct for errors and avoid collisions. No operator input is required when in this
mode. Co-pilot mode is a middle ground between manual and Autotram modes. It
is similar to manual mode in that the operator controls speed, direction of travel and
steering mechanism, however the computer system intervenes to prevent collisions
with the operating environment.
Each mode of operation is intended for a specific purpose. The operator, working
from a surface console, digs (or bogs) ore from the stope draw-point in manual mode.
The system is then switched into Autotram mode and the LHD autonomously travels
(or trams) to the ore-pass, dumps the ore, and then returns to the stope. On arrival
it switches back to manual mode and the cycle repeats. Co-pilot mode is not used
during normal operation, but is commonly used when moving the LHD around to
deal with exceptional circumstances outside the normal bog-haul-dump-tram cycle.
Autotram systems offer numerous benefits over manual operation, including: (i)
improved safety by removing the operator from the machine and thereby the un-
derground environment; (ii) productivity increases due to higher tramming speed,
improved dumping precision and continued operation while blasting is in progress;
(iii) production cost reductions through the ability to have one operator, control
several LHDs simultaneously; and (iv) lower development costs that come by virtue
of less stringent ventilation requirements.
1.3 Motivation for monitoring road surface con-
dition
The primary motivation for this thesis is to facilitate productivity improvement in
underground stope mines through better utilisation of LHD vehicles. The second
level of motivation is to provide LHD operators with information on potentially
11
1.3. MOTIVATION FOR MONITORING ROAD SURFACE CONDITION
damaging LHD collisions. A third motivation is to provide a means for better time
management of grader vehicles using logged road roughness information.
1.3.1 LHD productivity
The most important indicator of LHD productivity is the number of bucket loads
of ore an LHD transports from the stope to the ore-pass per hour. The total time
taken to transport one bucket can be broken into four components:
• time to bog the ore, Tb;
• time to haul the ore to the ore-pass, Tf (the full-tram));
• time to dump the ore, Td; and
• time to tram back to the ore-pass, Te (the empty-tram).
Therefore, total time to move one bucket of ore is given by:
T = Tb + Td + Tf + Te (1.1)
Since the quality of the road has little effect on bog or dump times, the improvement
in productivity due to improved road quality will be seen in the tramming and
hauling parts of the cycle. Put simply, only the haul and tramming times are
functions of road roughness (RR).
T (RR) = Tb + Td + Tf (RR) + Te(RR) (1.2)
The RH2900 has three gears for travelling in each direction, termed first, second,
and third. The machine is typically operated continuously with full-throttle to
maintain hydraulic pressure and, ideally, the vehicle runs in third gear (highest
speed) to minimise hauling and tramming times. A manual operator makes gear
12
1.3. MOTIVATION FOR MONITORING ROAD SURFACE CONDITION
changes according to his judgement and skill. Under autonomous operation, the
gear changes are programmed to occur at various points along the haul route. The
full-tram time (Tf) and the empty-tram time (Te) can be represented by the time
spent in each gear, i.e.
Tf(RR) = Tf1(RR) + Tf2(RR) + Tf3(RR), (1.3a)
Te(RR) = Te1(RR) + Te2(RR) + Te3(RR). (1.3b)
The time spent in each gear depends on the distance travelled in that gear, so
Tf(RR) =Df1(RR)
Vf1
+Df2(RR)
Vf2
+Df3(RR)
Vf3
, (1.4a)
Te(RR) =De1(RR)
Ve1
+De2(RR)
Ve2
+De3(RR)
Ve3
. (1.4b)
Where the total tramming route length is given by
D = Df1(RR) + Df2(RR) + Df3(RR) (1.5)
= De1(RR) + De2(RR) + De3(RR). (1.6)
Therefore total time for one cycle is
T = Tb +Td +Df1(RR)
Vf1
+Df2(RR)
Vf2
+Df3(RR)
Vf3
+De1(RR)
Ve1
+De2(RR)
Ve2
+De3(RR)
Ve3
.
(1.7)
The minimum cycle time is (with the entire run done in third gear):
Tmin = Tb + Td +D
Vf3
+D
Ve3
(1.8)
An example will illustrate the benefits that can be derived from being able to travel
in higher gears. The data used was from the 58 Orange 57 haul route at ODM. The
production rate for this haul route was approximately fourteen buckets per hour
with the gear change timing configuration in place at the time of testing. The LHD
spent approximately 40% of the tram in 1st gear and 60% in 2nd gear. No time was
13
1.3. MOTIVATION FOR MONITORING ROAD SURFACE CONDITION
programmed in 3rd gear (where the machine runs fastest) for two reasons: (i) the
conditions of the roads made steering difficult; (ii) the mine site lacks confidence in
Autotram and chose not to risk the system. The latter is related to the acceptance of
new technology. The former is an area for improvement, and is where this project’s
efforts are primarily focused.
If the machine were to operate in third gear along the entire haul route, productivity
would increase to approximately twenty buckets per hour. This is equivalent to a
42% increase in productivity. The productivity associated with the time spent in
each gear is shown in Fig. 1.4 along with a marker showing the current configuration.
For stopes with longer tramming routes, the theoretical increase in productivity is
higher, approaching 115% for a very long tramming route.
Figure 1.4: Time spent in each gear vs. number of buckets/hr. Buckets/hr is shownfor current gear configuration.
14
1.3. MOTIVATION FOR MONITORING ROAD SURFACE CONDITION
To improve road quality we must first understand how haul roads degrade with time.
Only by measuring the road condition, can the degradation trends be analysed and
the contributing factors established. Road degradation information could then be
linked to the associated costs, leading to a feasibility study of better road surfacing
techniques.
1.3.2 Abnormal event detection such as collisions
In addition to general road surface quality, there is the need to be able to identify
what might be called exceptional events whose occurrence can adversely affect the
integrity of the machine. These events might include the development of a significant
pothole or a large rock falling across one of the wheel-tracks. Similarly, there is the
possibility of the Autotram LHD colliding with the side walls or other damage
causing events such as the bucket impacting with the steel grid or grizzly that is
placed over the ore-pass to prevent oversize passing.
In principle, an on-board monitor could be used to detect these events and report
and log them. In the case of a pothole or spilt rock across the wheel-tracks, this
information could be used to identify the presence of these irregularities and to
action corrective maintenance.
In the case of glancing wall collisions, provided they were not too severe, the LHD
would otherwise continue operating. The occurrence of a collision probably indicates
that control of the vehicle was lost prior to the collision. Providing notification of the
collision to a supervisory operator would allow him to address this by, for example,
changing gear to maintain vehicle control.
The early detection of other damage causing events, possibly related to operator
behaviours in manual mode, could be used as feedback to alert the operator of an
issue and allow for prompt corrective action.
15
1.4. SUMMARY
1.3.3 Grader scheduling
Graders are large machines used to reduce the unevenness of rough unsealed roads.
This is done by removing the top layer and filling depressions, smoothing the road
surface. Currently, LHD operators decide when the haul road has reached a critical
level of roughness and inform a grader operator of the problem. The grader operator
then travels from their current location and grades the road. Travelling from job-
to-job can take up a significant portion of a grader’s time and the process could
be improved with better forward planning. A system that provides a continuously
updating map of the mine, displaying road roughness on all traversed tunnels would
allow this forward planning to take place.
1.4 Summary
This chapter has introduced the thesis topic. The next chapter surveys previous
work in the area of road condition monitoring and analysis with a view towards
identifying the most effective method for monitoring road surface quality.
16
Chapter 2
Methods for measuring and interpreting
road roughness
Road roughness is an encompassing term used to describe the variation in elevation
of road profiles and is a major contributing factor to vehicle operating efficiency,
quality of ride, and the rate of accrued damage. Interest in road roughness is often
initiated by a desire to understand vehicle ride, so it is often studied alongside specific
vehicle dynamics. Various methods have been proposed for the description of road
roughness. It is the purpose of this chapter to review these against the specific
objective or developing a road roughness measure for underground load-haul-dump
vehicles.
2.1 Road roughness measurement
There are two basic approaches to measuring road condition: (i) indirect methods
based on the measurement of the response of a vehicle as it passes over a road, and
(ii) direct methods based on road profile measurement. The first category are known
as response-type methods; the second as profilometric methods. Table 2.1 summarises
the advantages and disadvantages of each, based on the discussion in [1].
17
2.1. ROAD ROUGHNESS MEASUREMENT
Dat
aAcq
uisition
Spee
d*
Equ
ipm
entCos
t*Acc
urac
y*D
esig
nCom
plex
ity*
Dus
tSe
nsitiv
ity
Req
uire
sReg
ular
Rec
alib
ration
?*
Response-type Fast Low Med Low Low Yes/NoProfilometric Fast High High High Med No
Table 2.1: A comparison of response-type and profilometric methods for measuringroad roughness *[1].
Response-type methods measure the response of a vehicle as it passes over a road.
They are more commonly used than profilometric methods because of the ease and
speed at which data can be gathered and because the required equipment is relatively
inexpensive and simple to design. Regular re-calibration of a response-type system
is required if a standard measure is being calculated from the data, such as the
International Road Roughness Index (IRI) (see Section 2.2).
Vehicle response is commonly established by either measuring the suspension dis-
placement, that is the change in position between the axle and body of the vehicle, or
by measuring the vertical acceleration of the vehicle body. Measurement of suspen-
sion displacement is generally preferred because the results show better correlation
with the IRI. Chassis acceleration is, however, often used in practice because it is
easier to measure.
Profilometric methods have the advantage of providing a description of the geometric
profile, and so can be used for calibration of response-type measurements. Profilo-
metric methods do not require constant re-calibration. Two kinds are in common
use:
1. Manual profilometers (rod-and-level surveys) for slow, low cost, accurate rough-
18
2.2. ROAD ROUGHNESS MEASUREMENT STANDARDISATION AND THEINTERNATIONAL ROUGHNESS INDEX (IRI)
ness measurement.
2. High speed laser profilometers for high speed, but more expensive, accurate
roughness measurement.
2.2 Road roughness measurement standardisation
and the International Roughness Index (IRI)
In the early 1980s, the World Bank sponsored an experiment named the interna-
tional road roughness experiment (IRRE) [7]. The World Bank’s interest in road
condition was motivated by their agenda to improve the efficiency of world trade.
Transport of goods and services is a significant cost to all economies. Road transport
is the dominant transportation method, and well maintained roads allow goods to
be transported cheaper and faster than is possible over poorly maintained roads.
The IRRE led directly to the development of the so-called International Rough-
ness Index (IRI) and involved all commonly used response-type and profilometric
road roughness measurement systems to ensure compatibility with most roughness
measurement systems. Sayers, Gillespie and Queiroz et al. developed the IRI to
reconcile these different road roughness measures [8, 7, 1, 9, 3].
2.2.1 Definition of the IRI
Sayers, Gillespie and Queiroz considered many models for calculating the IRI, using
the following four criteria [7]:
1. Time stability: The IRI must not change with time.
2. Transportability: It must work in all parts of the world and with all mea-
surement methods (Response-type and profilometric methods).
19
2.2. ROAD ROUGHNESS MEASUREMENT STANDARDISATION AND THEINTERNATIONAL ROUGHNESS INDEX (IRI)
3. Relevance: It must reflect the road condition in terms of ride quality, vehicle
operating costs and safety.
4. Validity: Procedures put in place for IRI measurement must be such that
IRI measurements are repeatable.
A quarter-car analysis, see Fig. 2.1, was selected because it satisfied these criteria as
well as possessing the following desirable attributes: (i) it can be calculated directly
from road profile; (ii) it shows good correlation with response-type systems (since it
is a dynamic model of a response-type system); and (iii) it acts as a high and low
pass filter, removing unwanted frequencies from the data.
The IRI results were found to be sensitive to the dynamic parameters of the quarter
car model so standard values were selected for the mass, stiffness and damping
according to which frequencies should be filtered. The resulting quarter car model
is known as the Golden Car, see Fig.2.1.
Figure 2.1: The Golden Car model [3].
To obtain good correlation between results from different response-type instruments,
the traversing speed must be the same. For the IRI, a reference speed of 80km/h is
used, chosen to reflect the typical speed of a light vehicle travelling on a highway.
The influence of the dynamic properties of the vehicle from which the response-type
20
2.2. ROAD ROUGHNESS MEASUREMENT STANDARDISATION AND THEINTERNATIONAL ROUGHNESS INDEX (IRI)
measurements are made have been explored by [8]. It was found that while the use
of different vehicles resulted in different IRI values, there was a strong correlation in
the results. Therefore, the IRI can be used to compare results found using different
vehicles as long as calibration is performed. The method for this is discussed in the
following pages.
2.2.2 Calculation of the IRI
For response-type methods, the IRI is calculated as follows:
1. The vehicle is driven over the road in question at 80km/h (standard traversing
speed).
2. Displacement of the vehicle suspension is accumulated over the length of the
road.
3. The accumulated suspension displacement is then divided by the distance trav-
elled to get the average rectified slope (ARS).
4. The ARS is correlated with the IRI calculated from the profile of the same
road, measured using a profilometric method.
5. This correlation is used to find a calibration equation and the IRI is calculated
from the ARS.
Steps 4 and 5 must be repeated on a semi-regular basis to recalibrate as vehicle
dynamics change with time, causing drift in the calculated IRI. The IRI is calculated
from profilometer data as follows:
1. Data is filtered using a moving average (of 250mm) to simulate the enveloping
effect of the tyres and to reduce the sensitivity of the IRI algorithm to the
sample interval [3].
21
2.2. ROAD ROUGHNESS MEASUREMENT STANDARDISATION AND THEINTERNATIONAL ROUGHNESS INDEX (IRI)
2. This data is then used as an input to the Golden Car quarter car model.
3. Displacement of the suspension is accumulated over the length of profile data.
4. The IRI is calculated by accumulating the suspension displacement and divid-
ing by the distance travelled (the ARS for the golden car).
Simply put, this is a response-type simulation of the IRI golden quarter car model
passing over a measured road profile. The IRI is the ARS calculated using this
method [8, 7].
2.2.3 Summary of the IRI
Calculation of the ARS, and thus the IRI requires measurement of the suspension
displacement, therefore it is required that the vehicle used to obtain the data has
some form of suspension between the chassis and axle. It is also necessary for
calculation of the IRI that roughness measurements are taken while travelling at
80km/h.
In the case of LHDs on haul roads these conditions cannot be satisfied because:
(a) LHD vehicles rely on only rubber air-filled tyres for suspension so the ARS
cannot be calculated; and (b) the LHD used in this study has a maximum speed
of approximately 20km/h. For these reasons, the IRI is not directly applicable to
LHDs. However many of the ideas resulting from investigations into the IRI are
applicable to LHDs, specifically, the strong dependence of the measure on vehicle
speed. This has important implications for this project as the LHD operates at one
of three speeds depending on gear selection.
22
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
2.3 Power Spectral Density (PSD) methods
The power spectral density (PSD) provides a means for extracting frequency infor-
mation from road roughness data. When calculated from time domain data, the
resulting power spectral density (PSD) is in the frequency domain, whereas a PSD
calculated using spatial domain data is in the spatial frequency (or wave number)
domain.
2.3.1 Hypotheses for simplified road roughness description
In 1973, Dodds and Robson carried out an extended study of road surfaces [4]. They
suggested road surfaces could be adequately described using a Gaussian, homoge-
nous and isotropic random process, provided data spikes (transients) were removed
from the data and analysed separately. The elements of this hypothesis are:
• Homogeneity: The Spectral characteristics are the same independent of the
measurement starting location.
• Isotropy: Spectral characteristics are independent of the direction of travel.
• Gaussian process: Data is random and distributed such that its distribution
is Gaussian, allowing the distribution to be described using only a mean and
standard deviation (first and second order moments).
For an isotropic and homogenous road surface, spectral characteristics taken starting
at any location and in any direction, could be used to develop a description of
the entire surface. Dodds and Robson [4] aimed at justifying the hypotheses of
homogeneity and isotropy with a view to significantly reduce the complexity of road
descriptions. It is assumed that the road profile follows a Gaussian distribution.
23
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
2.3.2 Homogeneity
Dodds and Robson [4] undertook dual-track geodesic surveys on three types of road:
motorway, minor road, and paving. After removing low frequency components (road
slope) and high frequency transient components (potholes etc.), the spectral char-
acteristics of the road surface were analysed. The spectral information obtained
from the two wheel-paths showed good agreement for all road types, supporting the
hypothesis of homogeneity.
2.3.3 Isotropy
To test the hypothesis of isotropy, Dodds and Robson [4] formulated a coherency
function (the cross-spectral density normalised by the one sided spectral density),
assuming isotropy. This coherency function was then compared to experimental re-
sults and the two were found to agree for low wave numbers (see Fig. 2.2). Dodds
and Robson [4] argued that this confirmed the accuracy of the homogeneity hypoth-
esis whilst backing the isotropic assumption. As seen in the figure, however, the
validity of this assumption is questionable for higher wave numbers.
Although limited experimental evidence was available, Dodds and Robson concluded
that the hypothesis of isotropy was supported in the case of all three types of roads.
They stated that further experiments would show that the proposed classification
method completely defines road-surface roughness.
Kamash and Robson [10] proposed that not all road profiles are isotropic. They
provided conditions of admissibility that a surface must satisfy to be classified as
isotropic. If these conditions are satisfied, the road can be described by a one
dimensional PSD. If a profile does not satisfy the conditions of admissibility, however,
it is not independent of travelling direction and must be described using a two-
dimensional PSD.
24
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
Figure 2.2: Comparison between measured coherency functions and those calculatedassuming isotropy. (A: Motorway, B: Minor road, C: Paving) [4]
2.3.4 Transient identification
Roulliard, Sek and Perry [6] suggested that when analysing road surfaces with large
peaks or transients, it may be possible to use crest factor analysis (a description of
the quantity and amplitude of sharp peaks in the data) to identify transients. It
was suggested that crest factor should be studied using a moving window on the
data, calculated by finding the highest peak in the window and dividing it by the
root mean square (RMS) for that window. This was called ‘crest factor variation’
and was shown to be useful for identifying smaller as well as larger crests in the
data. This method of transient identification is also used successfully by Bruscella,
Roulliard and Sek [5] and Roulliard, Bruscella and Sek [2] in later works.
The spatial acceleration domain (double derivative of profile with respect to direc-
tion of travel) has been used (Bruscella, Roulliard and Sek [5]) to identify transient
events (outliers) with greater accuracy and reliability than when using road elevation
as an analysis domain. As seen in Fig. 2.3, the transient is within the 95% confidence
interval in the road elevation domain, whereas it is an outlier in the spatial accelera-
25
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
tion domain. Crest factor analysis has been used in the spatial acceleration domain
as an effective means of identifying transients (Bruscella, Roulliard and Sek [5]). An
alternative method proposed by Roulliard, Bruscella and Sek [2] for removing and
recording the amplitude and location of transients is discussed below.
Figure 2.3: Road elevation domain vs. road spatial acceleration domain (raw dataand statistical distribution) [5].
2.3.5 The Gaussian assumption and problems associated
with non-stationary PSDs
Roulliard, Sek and Perry [6] question the validity of assuming a Gaussian distribu-
tion. It was shown that for rough roads, the data was approximately Gaussian, but
less rough roads showed values of kurtosis as high as double the normal Gaussian
26
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
value (here, we refer to the ‘kurtosis proper’, which is three for Gaussian distribu-
tions). It was concluded that the Gaussian assumption is applicable for rough roads
but not for roads with smaller amplitude surface irregularities.
For the PSD and probability density function (PDF) of a random process to be
meaningful, the data is required to be stationary. If the data is not stationary (which
it often isn’t), the PSD and PDF will produce results of little value. Bruscella,
Roulliard and Sek [5] show that road elevation is ”a highly nonstationary, non-
Gaussian process containing transients”.
Transient events can be identified and removed using crest factor analysis (as dis-
cussed), but since skewness and kurtosis are sensitive to non-stationarity as well
as transients, the data is still non-Gaussian. Roulliard, Sek and Perry [6] suggest
the need for a method to analyse data with nonstationary components, stating the
majority of road surface data is nonstationary.
Roulliard, Bruscella and Sek [2] proposed a universal classification methodology for
bitumen and other roads with nonstationary RMS. They suggest that the proposed
method is an “accurate means to characterise road roughness levels for roads with
either stationary or nonstationary RMS”. They propose that any road surface can
be described by two fundamental components:
1. Underlying road surface irregularities
2. Transients
Transients are removed using crest factor analysis and the remaining data is split into
sections of constant RMS. This effectively divides the road up into many discrete
segments according to roughness. This was implemented on the entire 415km of
road samples, resulting in two components: a series of constant RMS (stationary)
road segments with the transient components removed, classified according to RMS
level; and data containing the number of removed transients and their amplitudes.
27
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
Figure 2.4: Statistical distributions of spatial acceleration and transient amplitudes[2].
It was found that the amplitudes of the transients could be described using a sta-
tistical distribution as could the discrete RMS sections (Fig. 2.4). A set of nine pa-
rameters were defined to describe these statistical distributions (Table. 2.2). These
nine parameters were used to fully describe the distribution of stationary segments
according to RMS and the distribution of transients according to amplitude.
Parameter Definition Typical Limits(1) (2) (3)
ραT Transient density (transients/unit length) 0 - 80µαT Transient amplitude mean 100 - 600σαT Transient amplitude standard deviation 50 - 800MαT Transient amplitude median 50 - 400RαT Transient amplitude range 20 - 1500
µαRMS RMS distribution mean 30 - 110σαRMS RMS distribution standard deviation 10 - 60MαRMS RMS distribution median 15 - 95RαRMS RMS distribution range 5 - 250
Table 2.2: The nine parameters used to completely describe a road surface [2].
2.3.6 PSD shape
Roulliard, Sek and Perry [6] also found that for roads of different roughness, the
shape of the PSD curve remained almost unchanged although the RMS varied as
28
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
the PSD shifted vertically according to the road roughness (Fig. 2.5). This is an
observation also confirmed by Bruscella, Roulliard and Sek [5].
Figure 2.5: Four PSDs for roads of different roughnesses [6].
This is an important attribute of road surfaces, as it shows that the area under the
PSD (the RMS) increases with road roughness.
2.3.7 Conclusions
There is much useful information to be gained from this large body of work. The
following are the most useful points (in no particular order):
• Roads can be classed as locally homogenous.
• PSD shape remains unchanged with changes in road roughness.
• Data must be stationary for PSD to be meaningful.
• Spatial acceleration domain is useful for identifying transients with greater
accuracy.
• Methods for the following are suggested:
– Testing whether a road profile is isotropic.
– Transient identification and analysis.
– Analysis of nonstationary data.
29
2.3. POWER SPECTRAL DENSITY (PSD) METHODS
This information is useful in quantifying haul road roughness. Specifically, methods
for identifying transients caused by potholes and spilt rock and analysing nonsta-
tionary data may prove to be extremely useful in analysing the road surface.
2.3.8 Response-type versus profilometric methods for un-
derground haul roads
When deciding on which of these methods to use, it is important to ask which
information is more valuable, the profile of a road, or a vehicle’s response to that
road. Is it degradation of road profile or vehicle wear that is of more value? The most
relevant measure is decided on a case-by-case basis depending on the application.
For underground LHDs on haul roads, we are primarily interested in LHD productiv-
ity and control of the LHD at high speeds. Measuring the profile of the road would
provide a direct measure of road unevenness, but since we are more interested in the
effect of the road on the vehicle, it is more appropriate to measure the LHD response
directly. LHD vertical acceleration holds more significance than actual road profile.
It is not possible to measure suspension displacement, as LHDs have no suspension
per se. Accelerometers were selected for measuring vehicle response for this reason
and because they are relatively simple to install and provide accurate measurements
of vehicle vibration.
30
Chapter 3
Design and development of the road
surface quality monitor
This chapter describes the hardware and software developed in this project to monitor
road surface quality. The monitor integrates with the DAS Autotram control system,
broadcasting road surface quality information over the local area network connecting
the vehicle to the surface. The road quality information is displayed in real time on
the console used by the remote operator, allowing degrading roads to be visualised
and monitored.
3.1 An overview of the Autotram control system
The Autotram control system comprises an on-board controller (OBC), under-
ground, on the LHD and a supervisory control system (SCS) at the surface. The
OBC is responsible for autonomous navigation of the vehicle and communicates
with the SCS via a local area network (LAN) (Fig. 3.1). Sensor information as
well as streaming audio and video is sent from the OBC to the SCS. In remote
operation mode, the operator garners this sensor information with experience to
31
3.1. AN OVERVIEW OF THE AUTOTRAM CONTROL SYSTEM
make control commands, e.g. steer, accelerate, lift bucket, raise engine revolutions-
per-minute (RPM), switch operation mode and so on, that are sent to the LHD
over the LAN. Under automatic operation, this information is used for automated
supervisory control of the vehicle.
Figure 3.1: A schematic representation of the Autotram network with the RSQmonitor.
The OBC consists of a control computer connected to several sensors, notably:
• scanning ladar devices used to map the environment and, in particular, localise
the LHD with respect to the haul route walls;
• radio receivers used to establish LHD position from beacons located along the
haul route;
32
3.2. INTEGRATION OF THE RSQ MONITOR WITH THE AUTOTRAMSYSTEM
Figure 3.2: The Autotram control room supervisory control system (SCS) worksta-tion.
• wheel speed sensors; and
• an articulation angle sensor.
The OBC also makes used of information from various other sensor measurements
including engine speed and temperature, fuel and oil levels, and the binary state
(on or off) conditions of several quantities such as mode of operation and bucket
position.
3.2 Integration of the RSQ monitor with the Au-
totram system
The road surface quality (RSQ) monitor integrates with the Autotram system as
shown in Fig. 3.3. Its primary purpose is to provide the SCS with an indication of
33
3.2. INTEGRATION OF THE RSQ MONITOR WITH THE AUTOTRAMSYSTEM
road roughness. To achieve this, road surface quality information is inferred from
the measured output of accelerometers mounted on the vehicle chassis, sampled at
moderate rates (200Hz). A basic requirement of the RSQ monitor is that it reduce
the significant quantity of measured sensor data acquired during this process to an
indicative measure of road surface quality called the RSQ index. The theory used
for this data reduction is detailed in Chapter 4.
Figure 3.3: The relationship between the Autotram OBC and SCS and the RSQmonitor.
The existing LAN infrastructure at ODM has the capacity to transmit data to the
SCS, so the basic system design described below is predicated on this capability.
Road surface quality information is sent in user datagram protocol (UDP) pack-
ets [11], over the LAN using a format styled on the pre-existing control system
packet structure used to communicate information between the OBC and SCS. The
RSQ monitor connects to the Autotram network as shown in Fig. 3.1.
Information on road surface quality received by the SCS is displayed on the haul
34
3.3. RSQ MONITOR HARDWARE
Figure 3.4: The RSQ index as displayed on the Autotram operator control console.
route map that appears on the operator console. A screen dump showing an example
haul route with an LHD in operation and road surface quality information is shown
in Fig. 3.4. The magnitude of the RSQ measure is used to colour triangles of the
Delaunay triangulation generated and used by the LHD navigation system. These
colors are graded so that green indicates good quality roads and red poor roads.
3.3 RSQ monitor hardware
The RSQ monitor consists of an Advantech PCM-9572 PC-104 embedded board ex-
pandable (EBX) computer fitted with an Advantech PCM-3718 analogue-to-digital
(AD) converter. The PCM-9572 computer reads and processes acceleration data
from two accelerometers via the PCM-3718 to produce an index representing road
35
3.3. RSQ MONITOR HARDWARE
surface quality. Measured data is archived to a compact flash solid state memory
disk and the road quality index measure is broadcast synchronously over the LAN
at a reduced rate. The RSQ monitor (Fig. 3.5) is mounted centrally on the LHD in
the location shown in Fig. 3.6.
Figure 3.5: The RSQ monitor hardware.
The RSQ monitor and accelerometers have been designed to be robust to:
1. High vibration/shock: The LHD experiences high energy vibration and
shock whilst in normal operation due, inter alia, to road surface unevenness.
2. High temperatures: The engine on the LHD generates significant heat. Am-
bient air commonly exceeds 60◦C, with the LHD body reaching much higher
temperatures.
36
3.3. RSQ MONITOR HARDWARE
3. Corrosion: Mine dust at ODM contains high quantities of copper which
promote corrosion. Additionally, underground vehicles are cleaned using high-
pressure salt water, a powerful corrosive.
Figure 3.6: The LHD with accelerometer and RSQ monitor mounting locationsshown.
These environmental factors place significant demands on the hardware system,
requiring careful component selection. The RSQ monitor is physically protected
by an IP66 Rittal AE1057.600 steel case. This is to protect against falling rock
and other potential damage. It also has good heat conduction properties which is
important for this application, as airflow cannot be used to cool the RSQ monitor
due to the corrosive environment in which it operates. A fan mounted inside the
case ensures that the ambient temperature is constant throughout for maximum
heat conduction.
The RSQ monitor is attached to the LHD using vibration isolation mounts. These
mounts are effective at isolating the monitor from high frequency vibrations, however
low frequency motions are still transmitted to the box with the potential to fatigue
fasteners. Military specification connectors are used for all connections and are
37
3.3. RSQ MONITOR HARDWARE
treated to prevent corrosion.
3.3.1 Accelerometers
The primary source of information used by the RSQ monitor is obtained from two
accelerometers. One mounted at the front of the LHD; the other at the rear. The
accelerometers are mounted as shown in Fig. 3.7 in the locations shown in Fig. 3.6.
The accelerometer at the front of the vehicle measures accelerations in three orthog-
onal directions: the lateral (X), longitudinal (Y), and vertical (Z) directions. The
rear accelerometer measures only vertical (Z) acceleration.
Figure 3.7: The method of mounting the accelerometers to the LHD. A lid is thenused to seal the protective case and prevent internal corrosion.
The key consideration for accelerometer selection is dynamic range, namely the
range between the smallest and largest accelerations that can be measured. How-
ever, the expected range of accelerations for motion of the Autotram LHD was not
known at the design stage of the RSQ monitor. The Autotram LHD has a mass
of 55, 000kg suspended on air-filled tyres and the suspension is dynamically com-
plex and therefore difficult to predict. For this reason, ± 25g accelerometers, as
38
3.3. RSQ MONITOR HARDWARE
commonly specified in automotive applications, were chosen. The minimum accel-
eration that these accelerometers can measure is ± 0.010g, sufficient sensitivity for
this purpose. This was verified during trials.
The accelerometers are contained within individual cases for protection against phys-
ical damage and high pressure water. These cases are stiff and securely attached to
the LHD body for maximum accuracy in the acceleration measurements.
3.3.2 Anti-aliasing filter
The accelerometer signals pass through an anti-aliasing filter (Fig. 3.8) to remove
any frequencies above 20Hz. This value is used because the sampling frequency of
the AD card is 200Hz, and the guideline adopted to ensure aliasing does not affect
the data is to use a sampling frequency of ten times the highest frequency measured.
This filter ensures that high frequency vibration such as engine noise is removed, so
the road roughness data remains unaffected by aliasing upon digitisation.
3.3.3 Analog-to-digital converter
The PCM-3718 AD converter accepts analogue inputs from the accelerometers as
well as an engine-on signal and converts these signals to digital format for use within
software.
The analogue-to-digital converter used in the RSQ monitor is twelve-bit, so it has
212 (4096) discretisation levels spread over the range of the input. The selected
accelerometers have a 0g voltage of 2.5V and a range of ±2V, so the AD converter
is set to unipolar mode, measuring inputs over the range 0 ∼ 5V. This gives a
resolution of:
δ =5
4096= 1.22mV
39
3.3. RSQ MONITOR HARDWARE
Figure 3.8: Signal processing in the RSQ monitor.
The selected accelerometers have a sensitivity of 80mV/g, so this AD converter
resolution is equivalent to a resolution of 0.015g, a slightly lower resolution than the
accelerometers are capable of (0.010g).
Too low a sampling frequency can result in the loss of information relating to higher
frequency road surface irregularities. On the other hand, too high a sampling fre-
quency can lead to large quantities of data, requiring more storage space. A sam-
pling frequency of 200Hz is used for this application. This is sufficient for capturing
LHD vibrations in the range 0 ∼ 20Hz. This range is sufficient, because due to
the LHD’s high mass and low tyre stiffness (and therefore low natural frequency)
and low traversing speed, the bulk of the vibrational energy is contained in low
frequencies (below 20Hz).
40
3.3. RSQ MONITOR HARDWARE
3.3.4 The delay-boxes
The Autotram OBC uses an engine-on signal internally. This signal is also fed into
the RSQ monitor via the AD converter (see Fig. 3.8) to provide a means to shut
down the RSQMon software and operating system (OS) before power is cut. When
the engine-on signal is lost, a delay-box waits twenty seconds before disabling power.
This ensures that the system has time to shut down cleanly. A second delay-box is
used to delay the RSQ monitor from starting once the LHD is activated. When the
LHD is used for on-board operation, the DAS system is deactivated. As a result, the
engine-on signal does not activate when the engine is started, so the RSQ monitor
remains disabled.
A switch is used to control the source of the engine-on signal. This switch has
three positions: Engine-on; simulated engine-on (5V DC); and disconnected. The
first of these is the engine-on signal from the DAS system as discussed (the default
position). The second mode is a simulated engine-on signal which allows the unit
to be activated for testing and debugging when the LHD engine isn’t on. The third
mode allows the RSQ monitor to be deactivated.
3.3.5 Compact flash data storage
The RSQ monitor uses ‘compact flash’ solid state memory. Two high-speed 1GB com-
pact flash cards are used: one for the OS and one for data storage. Solid state
memory is used because of its reliability under high vibration, high temperature
conditions and although flash memory does not have the fast write speeds or large
storage capacity of hard disk drives, it is sufficient for this application.
To maximise storage space, data is compressed upon archival. Sixty hours of data
can be stored on a 1GB card at the 200Hz sampling rate used.
41
3.3. RSQ MONITOR HARDWARE
3.3.6 Power
A circuit diagram for the RSQ monitor is shown in Fig. 3.9. It uses approximately
40W (1.6A at 24V) when in operation using power sourced from the 24V DC battery
on the LHD. This power source is passed through a diode bridge rectifier whose
purpose is to supply an output voltage with a polarity independent of the input
polarity. The bridge rectifier prevents current back-flow if the supply potential
drops below that of the internal batteries.
Two 12V, 1.2Ah lead-acid batteries in series act as a uninterruptable power supply
for the monitor. These batteries smooth the input voltage supplied to the monitor
and are charged from the LHD’s alternator. The power line is then passed through a
wide-range voltage regular, that takes an input of 10-40V DC and outputs 24V DC.
This unit is also used to protect the internal components from variations in supply
voltage in case the internal batteries are unable to smooth the voltage sufficiently.
It is also used to ensure a constant 24V DC output independent of the input, as the
LHD battery voltage can vary between 16V DC when low on charge and 28V DC
when charging. With the current drawn by the RSQ monitor, the wide range voltage
regulator can supply 24V DC output from a power source as low as 16V.
The two time-relays discussed in Section 3.3.4 are used to ensure that the RSQ
monitor is only operational when Autotram is running to avoid draining the LHD
battery.
Voltage regulators (24-12V DC and 24-5V DC) are used to obtain the necessary
voltage lines for the computer components and accelerometers. The 5V DC line
powers the motherboard, the AD converter, the flash memory, and the RAM. The
12V DC line is used for the tri-axial accelerometer and is passed through a 12-
5V DC regulator for the single axis accelerometer. This separate voltage regulator
is necessary to protect the single axis accelerometer from noise on the 5V line during
periods of high power drain because this accelerometer does not contain its own
42
3.3. RSQ MONITOR HARDWARE
Figure 3.9: Power flow wiring diagram for the RSQ monitor.
43
3.4. RSQ MONITOR SOFTWARE
voltage regulator. A cable with both 12V and 5V lines is available for any necessary
additional peripherals required during OS installation or system maintenance (e.g.
a CD-ROM).
3.4 RSQ monitor software
The RSQ monitor computer systems runs the QNX realtime operating system. The
monitor software, known as RSQmon, is written in the C programming language.
The functional requirements of this software are to:
1. Read accelerometer data synchronously at 200Hz.
2. Listen on the LAN for relevant information communicated between the OBC
and SCS, e.g. LHD speed and position.
3. Buffer the accelerometer data, and the OBC information gathered from the
network, in memory.
4. Reduce the measured accelerometer data to an RSQ index.
5. Broadcast this index over the LAN.
6. Compress and archive buffered data for analysis.
7. Listen on the network connection for instructions and status and data infor-
mation requests, e.g. start logging, stop logging, extraction of data.
8. Monitor appropriate machine signals to identify times when the operating
mode must change, e.g. if the engine is turned off, it is necessary to shutdown
the logger to prevent battery drain.
44
3.4. RSQ MONITOR SOFTWARE
3.4.1 RSQ monitor software architecture
The RSQ monitor software has been designed as a finite state machine to realise
the functional specification. Specifically, the software has several identified states.
These include (Fig. 3.10):
• a READY state which serves as a platform for entering other system states;
• a MONITORING state in which the system calculates the road surface quality;
• a LOGGING state in which accelerometer and OBC data is written to disk
without calculation; and
• a FAULT state to handle exceptions.
Figure 3.10: A diagrammatical representation of the finite state machine RSQ mon-itor software.
45
3.5. SUMMARY
The state machine is event-driven with events falling into two categories, termed
internal and external events.
Internal events are events triggered by a change within the system. These include
synchronous events such as ‘clock ticks’, whose occurrence initiates the process of
collecting and processing accelerometer data; events that trigger on criteria, for
example writing data to disk when a buffer becomes full; and error events, generated,
for example, on failure of a unit operation such as a service request made to the
operating system to write data.
The sources for external events can be exogenous happenings that trigger according
to predefined criteria, or can be explicit instructions to the RSQ monitor. An
example of an exogenous happening is the LHD engine being turned off. This event
is detected by monitoring the appropriate digital signal line and when detected,
raises an event that instructs the RSQ monitor to shut down. An example of an
explicit instruction is the command to start monitoring. The RSQmon software
listens on a dedicated port for UDP messages encoding such commands, and on
receipt of a command to start logging, an event is raised that brings the RSQmon
finite state machine into the MONITORING state.
3.5 Summary
The hardware and software solution described in this chapter has been deployed suc-
cessfully at ODM for periods of several months at a time and has survived the service
conditions of the mine with only minor repair required between service inspections.
The next chapter details the calculation of the road surface quality measure gener-
ated by the motor and displayed at the operator console as in Fig. 3.4.
46
Chapter 4
Monitoring road surface quality
The purpose of this chapter is to develop a measure of road quality and to use this
measure to analyse road roughness data obtained from experimental trials on site
at Olympic Dam. A measure of road surface quality is proposed and justified using
data obtained from these trials and its limitations are discussed.
4.1 Introduction
Figure 4.1 shows underground haul roads in different conditions. An Autotram
operator has no clear view of the road condition, yet it is desirable that he or she
be able to establish when the road is in poor condition, such as in Fig. 4.1(b). This
chapter develops and tests a simple measure that we call the RSQ index, derived
from accelerometer data obtained using the system described in Chapter 3. The RSQ
index is intended to provide road condition information to the Autotram operator
that could be used to make decisions such as to reduce speed on some section of
road or to call in a grader to recondition the entire haul road surface.
The data used in this chapter was collected from an installation of the monitor at
ODM on stope 57 Orange 58 in October 2004. Unless otherwise stated, the data
47
4.2. DEFINITION OF THE RSQ
(a) Smooth road (b) Rough road
Figure 4.1: An example of smooth and rough underground haul roads. The lightsource is displaced vertically downwards to allow road surface irregularities topresent more clearly.
used is that collected during the night shift of the 16th/17th of October 2004 on a
DAS Autotram LHD. The data was pre-processed using the methodology described
in Appendix A to extract useful segments of data.
4.2 Definition of the RSQ
The key requirement of the RSQ index is that its magnitude should increase as
the road becomes rougher. The decision to base the RSQ index on accelerometer
signals is founded on the observation that the vertical response of a vehicle becomes
more energetic as the road on which it runs degrades. The vertical acceleration
directly measures the excitation of the vehicle in response to road unevenness. A
48
4.2. DEFINITION OF THE RSQ
simple measure of the “energy” of this excitation is provided by computing a moving
average of the square of the measured acceleration. That is, we define the road
surface index by the mean-squared vertical acceleration over a moving window.
Let z̈k be the measured acceleration at sample k. Then the RSQ measure at sample
j is thus defined to be
RSQj =α
N
j∑
k=j−N
(z̈k)2 (4.1)
where N defines the length of the window and α is a unit-less scaling factor used to
bring computed RSQ values to a reasonable magnitude for display purposes.
The hardware described in Chapter 3 samples acceleration measurements at 200Hz.
Trials for different window sizes suggested a reasonable window size was 2.5 seconds,
resulting in N = 500. The scaling factor was set to α = 1000.
There are various interpretations we can make of Eqn. 4.1. One such interpretation
is that it defines a low-pass filter on the squared acceleration signal. We call this
the filter interpretation and a Bode plot of the filter frequency response is shown in
Fig. 4.2. We note that the 3dB bandwidth of the filter is approximately 1 rad/s.
A second interpretation of Eqn. 4.1 is as an estimate of the expected value of the
squared acceleration, E[z̈2], over the moving average window. This interpretation
emphasises that the squared acceleration signal is stochastic. It is well established,
see for example Chapter 5 of [12], that the mean square of a stationary random
process is equal to the area under the graph of spectral density versus frequency.
Assuming the acceleration signal is stationary over the moving average window this
links the RSQ index to more classical spectral analysis methods.
In early testing of the RSQ index this spectral interpretation was explored, and,
in particular, more complicated versions of the RSQ index were tried involving
weighted spectral densities, wherein a weighting function was multiplied with the
spectral density plot in an effort to improve the sensitivity of the measure. The
49
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
−250
−200
−150
−100
−50
0
50
Mag
nitu
de (
dB)
10−2
10−1
100
101
102
103
−90
0
90
180
270
Pha
se (
deg)
Bode Diagram
Frequency (rad/sec)
Figure 4.2: Bode plot showing the frequency response of the RSQ filter applied onsquared acceleration measurements.
weighting functions explored amplified the resonant frequencies and attempted to
attenuate other frequencies on the basis that the resonant frequencies are those most
likely to cause damage to the vehicle.
In the final analysis, however, such enhancements did not appear to increase the
ability to discriminate road condition and for this reason are not pursued further
in this thesis. The spectral content of the acceleration signals are, nevertheless,
investigated below to assist in the selection of the accelerometer information used
for computing the RSQ.
4.3 Choice of data for computing the RSQ index
Large quantities of data are captured by the RSQ monitor during Autotram oper-
ation. It is necessary to identify which data is most useful for representing road
quality. This will ensure that the resulting RSQ index is as accurate as possible.
50
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
Figure 4.3 shows a plan view of the haul route with the draw-point and ore-pass
marked.
Figure 4.3: A plan view of the tunnels that make up the 58 Orange 57 haul routewith draw-point and ore-pass marked. The path taken by the Autotram is shownby the blue dashed line. The portion of the haul route between the green trianglemarkers is referred to as the ‘main’ section of the tramming route.
The section of road between the triangle markers in Fig. 4.3 is referred to as the
‘main’ section of the haul route throughout this chapter. Only RSQ information
collected while travelling along this portion of the tramming route is used for analysis
because, while turning sharp corners and while digging and dumping at the draw-
point and ore-pass, the LHD is subject to significant vibrations from sources other
51
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
than the road surface. Additionally, the LHD speed outside of the main section is
variable, so results are not comparable.
The RSQ monitor described in Chapter 3 receives inputs from two accelerometers,
one mounted on the front section of the LHD and one on the rear. Section 4.3.1
argues that the accelerometer mounted on the rear section of the LHD should be
used in calculating the RSQ index as it is more representative of the road surface
than the front accelerometer.
LHD vehicles transport ore from stope to ore-pass. Each run can be looked at as
comprising two individual stages: (i) the tram from draw-point to ore-pass with a
full bucket, which we call the full-tram; and (ii) the return trip, which we call the
empty-tram. Section 4.3.2 argues that the empty-tram data is more representative
of the road surface.
4.3.1 Which accelerometer?
The rationale for deploying two accelerometers (one on the front and one on the rear
of the LHD) is that the different bodies to which the accelerometers are attached have
different associated dynamics, and so will experience different vibration patterns. It
was not known a priori which of these accelerometers would produce the most useful
signal for calculating a road surface quality measure. The objective of this section is
to justify the selection of the rear mounted accelerometer as the signal from which
the RSQ is calculated.
Figure 4.4 shows PSD plots for the front and rear accelerometers. This data is
based on the vehicle travelling in straight line motion and the accelerometer signals
both reasonably satisfy the requirement of stationarity [13]. The plots show two
modes, at 1.5Hz and 2.4Hz. There are various plausible vibration modes associated
with the LHD and it is likely that these peaks in the PSD are associated with the
dynamic modes resulting in the most significant motion in the vertical plane. The
52
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
(a) Front accelerometer
(b) Rear accelerometer
Figure 4.4: Front vs rear accelerometer frequency information. Empty-tram data.
53
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
likely associated mode shapes include:
• Bounce of front and rear bodies
• Pitching
• In-phase and anti-phase motions of the front and rear bodies.
To resolve the mode shapes, more detailed analysis including a larger number of
measurements would be required. For the purpose of this section, however, it will
be assumed that the peak at 1.5Hz is associated with the vibration of the rear section
of the LHD and the peak at 2.4Hz is associated with vibration of the front section.
It is observed that the PSD of the rear accelerometer shows higher magnitude val-
ues than that of the front accelerometer. That is, the rear body undergoes larger
accelerations.
4.3.2 Which tramming direction?
Each run comprises a full- and an empty-tram. Acceleration measurements taken
during full-tram are not comparable to those from an empty-tram. The purpose of
this section is to argue the use of data measured during the empty-tram for RSQ
calculation.
There are several factors that influence the magnitude of the measured accelerations.
The most significant are: (i) vehicle dynamics, (ii) vehicle traversing speed, and
(iii) road quality. The road surface quality measure is intended to describe (iii) alone,
so it is desired to be independent of the (possibly varying) vehicle dynamics and
speed. The Autotram system (conveniently) provides us with accurate, piecewise
control on speed: vehicle gear changes are triggered by software at the same location
on the road each time the LHD passes. Additionally, the vehicle’s dynamic properties
54
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
do not change as it passes along the road. The LHD speed and dynamics differ,
however, between full and empty-trams. The major differences are:
1. Bucket payload. The bucket is approximately 15t heavier when full.
2. Bucket position. When the bucket is full, the LHD traverses with the bucket
raised. When empty, the LHD traverses with the bucket lowered.
3. Vehicle speed. At full engine RPM1, the LHD reverse gears are faster than
forward gears. This is a consequence of the standard RH2900 gear-box con-
figuration which allows for faster travel during reverse tramming.
4. Wheel-path, which is consistent for each pass in the same direction but usually
differs for opposite directions of motion.
5. Vehicle orientation. The LHD travels forward when empty and backwards
when full.
Figure 4.5 shows PSDs of the accelerometer data for a sample run with the bucket
full and the vehicle travelling in straight line motion. When compared with PSDs
from the empty-tram, see Fig. 4.4, it is apparent that the full-tram presents a new
observable mode in the z-acceleration measurements at approximately 10.5Hz. This
additional mode is conjectured to be associated with the bucket position during the
full-tram.
It is evident that the dynamics observable in the accelerometer measurements differ
between full and empty-tramming. To provide an appropriate control to marginalise
this variation, it is clear that one direction of tramming should be adopted, and the
other ignored.
Empty-tramming is chosen because the vehicle is lighter by approximately 15t and
the dynamics are simpler, in particular the 10.5Hz mode is not observable, viz.
1The engine always operates at full RPM (full throttle) to maintain hydraulic pressure forsteering and bucket control.
55
4.3. CHOICE OF DATA FOR COMPUTING THE RSQ INDEX
(a) Front accelerometer
(b) Rear accelerometer
Figure 4.5: Front vs rear accelerometer frequency information. Full-tram data.
56
4.4. THE TRAMMING ROUTE
Figs. 4.4 and 4.5. It is also to be noted that because the vehicle travels slower in
forward, the RSQ provides better resolution of the road surface during the empty-
tram.
Because the vehicle travels back and forth on a reasonably regular basis (once every
couple of minutes) the restriction of the calculation to one direction of motion does
not make this approach impracticable. It is, in fact, unnecessary to receive updates
on the road quality more often than this, as the tram in the opposite direction has
minimal effect on the quality of the road.
The analysis techniques used in this chapter are designed specifically as a means of
calibration and apply only to the data attained during the trials that took place on
the 58 Orange 57 route at ODM in October 2004. This clarification is necessary
because there are important differences between stopes. Most importantly: the
full-tram and empty-tram are not necessarily carried out (respectively) in reverse
and forward; and all haul routes are designed differently, consequently leading to
large variation in the length of the route and the number of corners. Although this
calibration is specific to this stope, the techniques used can be transferred. It is
reasonable to expect that the RSQ measure would differ in magnitude from stope-
to-stope. This is seen as a practical issue that would need to be addressed in a
broader implementation, but is beyond the scope of this thesis.
4.4 The tramming route
Figure 4.6 shows a plan view of the LHD path along the tramming route at stope 58
Orange 57 at ODM used in September and October 2004. The draw-point is at the
bottom and the ore-pass at the top. Curves in the road are designated by the areas
in-between the dotted lines. The reference co-ordinate frame is arbitrarily chosen
by DAS when setting Autotram up for a new stope, and since the RSQ receives
position information from Autotram in this reference frame, it is used in the data
57
4.4. THE TRAMMING ROUTE
analysis. This LHD tramming route will be used as reference in the exposition of
RSQ data.
Figure 4.6: A plan view of the path the LHD takes along the 58 Orange 57 haulroute (see Fig. 4.3 for a map of the route itself). O designates the coordinate originand dotted horizontal lines designate sections of curved road.
Figure 4.7(a) shows the y-coordinate and speed data over the length of the haul
route for all 94 runs completed on the night shift of the 16th/17th of October 2004.
Load-haul-dump vehicles operate at full throttle to maintain hydraulic pressure for
bucket control and steering. Speed changes (as seen in the figure) correspond to gear
changes which are activated by the Autotram system software. The timing of the
gear changes is configured when Autotram is first set up on a stope. As mentioned
previously, and confirmed by Fig. 4.7(a), the result of this configuration is that the
speed profile of the LHD as it passes over the road is the same for each run. As seen
from this figure, there is minimal variation in speed between runs.
For the empty-tram, the LHD travels from ore-pass to draw-point. This can be
verified by looking at the change in y-coordinates in the Fig. 4.7(a). This plot can
58
4.4. THE TRAMMING ROUTE
(a) The LHD y-coordinate and speed vs sample number. All 94 runs are plottedon top of one another to show the lack of variation in speed from run-to-run.
(b) The LHD RSQ and speed vs sample number. Only one example run is plottedfor clarity.
Figure 4.7: Data from the LHD during the empty-tram runs on the night-shift of the16th/17th October 2004. Two separate y-axes are shown for each figure. Verticallines represent important features. The solid lines represent the boundaries of thearea of road that is used to obtain results, the dashed lines represent the zones wheregear changes occur and the dotted lines represent the segments of curved road.
59
4.4. THE TRAMMING ROUTE
be used to relate LHD position to sample number and sample number can be used
as an index for looking at vehicle speed, RSQ and various other data. Read the y
coordinate from the left y-axis, trace horizontally to the plotted y-coordinate data
and down to the x-axis, noting that for most y values there is only one associated
sample and for those y-values with two, it is straight-forward to establish which is
of interest.
As the LHD changes speed, for example at sample numbers 2000, 4350 and 14950,
so does the resulting RSQ. This dependency on speed is the reason that constant
traversing speed is a primary assumption underpinning the RSQ measure. Although
the speed changes, the periods of acceleration and deceleration are transient and as
discussed, gear changes occur at the same location in each run. The run can therefore
be split piecewise into sections of constant speed, with the sections falling into two
groups. These groups can be thought of as two different RSQ measures: first gear
RSQ (RSQ1) and second gear RSQ (RSQ2). These sections are not comparable to
each other, but for any given section of road, the RSQ trend with increasing passes
can be observed.
Figure 4.7(b) shows the RSQ index, and the LHD speed (for convenience), as a
function of the number of samples for an example run from the night shift of the
16th/17th of October 2004. This figure correlates the RSQ measure with the road
layout and gear changes. This is important in ensuring that the RSQ index is
representing the road surface rather than another phenomenon.
From Fig. 4.7(b), the following conclusions are drawn:
1. The RSQ does not increase significantly on gear (speed) changes, although
study of more data than is presented here suggests that on down gear changes,
a small, transient increase in the RSQ can be seen (see Fig. 4.7(b)). This is
not definitive and it is not known if the third gear change (at sample number
14950) coincides with a segment of road of increased roughness. The collected
60
4.5. ROAD SURFACE QUALITY ALONG THE TRAMMING ROUTE
data does not allow this issue to be resolved, but the phenomena has the
potential to present more visibly on tramming routes requiring numerous gear
changes.
2. The RSQ does not appear to be sensitive to gentle cornering, supported by
the fact that the RSQ does not show increased magnitude around corners.
The RSQ index has its highest value at a sample number of approximately
13300, corresponding to a y-coordinate value of −87m. This peak falls on a
section of curved road (marked by the dotted lines), but it can be concluded
that the road surface is the reason for the increase in RSQ rather than the
tunnel geometry. There are two facts that form the basis for this conclusion:
this peak occurs at the start of the curved road section, the gentlest part of
the curve (refer to Fig. 4.6); and the RSQ begins increasing before the LHD
enters the curved section.
3. The main section of the road (the ‘straight’ - from approximately sample 5000
to 14500) provides the highest RSQ values. The LHD maintains constant speed
(second gear) during this time, so the RSQ values are comparable within this
region. This is the part of the road that degrades the fastest, as the LHD is
travelling at top speed.
4.5 Road surface quality along the tramming route
Figure 4.8(a) shows the RSQ measure at the start of the night shift of the 16th/17th
October 2004, averaged over the first five runs of the shift. Figure 4.8(b) shows the
RSQ measure at the completion of this shift, averaged over the final five runs. Ninety
runs separate the two plots.
There is a clear increase in the RSQ measure with the number of runs. This can be
attributed to road quality, as the LHD speed and vehicle dynamics do not change
61
4.5. ROAD SURFACE QUALITY ALONG THE TRAMMING ROUTE
(a) The RSQ at the start of the shift
(b) The RSQ at the end of the shift
Figure 4.8: A 3D realisation of the RSQ measure (from the night-shift of the16th/17th October 2004) mapped against the haul route. The x and y-axes arethe x and y haul route map coordinates respectively, and the z-axis is the RSQcorresponding to the x-y coordinate.
62
4.6. ROAD DEGRADATION WITH NUMBER OF RUNS
over the course of the shift. Specifically, the ‘ride’ as shown by the RSQ becomes
increasingly rough with the number of runs.
4.6 Road degradation with number of runs
The two waterfall plots in Fig. 4.9 show the variation in RSQ over the duration of
the night shift of the 16th/17th October 2004 in two forms. Each plot in both the
waterfall figures is an average of five runs of RSQ information. These two forms are
used to highlight different aspects of the data and present different results. The two
forms are as follows:
• Figure 4.9(a) shows the change in RSQ with number of runs in 2D form.
The nature of this plot means that the magnitude of the RSQ is lost, but it
allows the viewer to clearly see the reference sample number for any sections
of interest so the location of the LHD can be determined.
• Figure 4.9(b) shows the same data as (a) but in 3D waterfall form. The data
in this plot is limited to the main section of road (between the black vertical
markers in (a)) because the RSQ outside this area prevents easy viewing of
the data of interest. This plot allows the magnitude of the RSQ to be seen
clearly.
Both figures clearly show that the RSQ increases with increasing number of passes. It
is seen from Fig. 4.9(a) that the areas of consistently high RSQ are at approximately
sample number 5800, 7900, 11800 and 13500, where the latter shows the highest
magnitude values. These correspond (respectively) to y-coordinates of −5m, −25m,
−75m and −90m. The fact that the areas of high RSQare consistently high and
that the RSQ increases steadily suggests that these high RSQ values are due to road
unevenness rather than spilt rock, potholes or other external sources.
63
4.6. ROAD DEGRADATION WITH NUMBER OF RUNS
(a) A 2D waterfall plot of the RSQ over the entire road
(b) A 3D waterfall plot for the main section of road (bounded by the black linesin (a))
Figure 4.9: Waterfall plots of the RSQ versus sample number with increasing runsfor the entire night-shift on the 16th/17th of October 2004. Each plot is an averageof five runs; the first plot represents runs 1 to 5 etc. Both plots are presented becausethey clearly show different aspects of the RSQ trend
64
4.6. ROAD DEGRADATION WITH NUMBER OF RUNS
Figure 4.10 shows a 3D waterfall plot of the differential RSQ. This allows the degra-
dation of road surface quality with increasing runs to be seen more clearly. The
differential RSQ is found by subtracting the first plot (the mean of runs 1 to 5) from
the remaining plots.
Figure 4.10: A waterfall plot of the differential RSQ (minus the initial RSQ) versussample number for the main section of road (between the vertical solid black markersin Fig. 4.7) with increasing runs for the entire night-shift of the 16th/17th of October2004. Each plot is an average of five runs; the first plot represents runs 1 to 5 etc.
It is seen from Fig. 4.10 that the RSQ increases not just at the peaks, but over
the whole road. However, it is apparent that the section of road that shows the
most significant growth is at approximately sample 13500, the part of the road
producing the highest RSQ values, possibly due to the formation and propagation
of corrugations in the road surface. It is also seen that the RSQ shows more rapid
growth over the entire road in the first 20 runs. After this, the RSQ continues to
increase, but at a reduced rate.
65
4.7. THE RMS-RSQ: A BROAD MEASURE OF ROAD SURFACE QUALITY
4.7 The RMS-RSQ: A broad measure of road sur-
face quality
The plots in the previous section are used to describe the localised RSQ. This is good
for establishing the difference in RSQ along the haul route, but it can also be useful
to describe the haul road in its entirety. Figure 4.11 shows a broad-based measure
of road surface quality, the RMS-RSQ, as a function of run number. The RMS-RSQ
is calculated by finding the root-mean-square of the differential RSQ (Figure 4.10)
for each run.
Figure 4.11: The RMS of the RSQ vs run number for every run of the night-shifton the 16th/17th of October 2004. A moving average curve is plotted in red to helpidentify the trend in the data.
The increasing trend in RSQ is clearly visible from this figure, as it is with the
localised RSQ discussed in the previous section. The format of this plot makes it
more convenient for identifying the form of the upward trend in the RSQ. It is clear
66
4.7. THE RMS-RSQ: A BROAD MEASURE OF ROAD SURFACE QUALITY
that the RSQ rate of growth is not constant: it appears to have the approximate
form A(1 − e−at).
There are two interpretations that can be used to explain this diminishing increase
in RSQ:
1. The road continues to degrade but the rate at which it does so reduces with
time. This is possibly due to a theoretical ‘maximum roughness’ that the road
can reach.
2. The road continues to degrade at the same rate, but the road is unable to im-
part the continually increasing vibrational energy on the LHD (due to the mass
of the LHD). As the road continues to degrade, the LHD does not experience
any increased vibration.
This point is made in passing. To identify the cause, further experiments would need
to be conducted, including correlation of results with regular road surveys which is
beyond the scope of this thesis.
The RMS-RSQ could be used alongside the RSQ measure to aid road maintenance
procedures. When the RMS-RSQ exceeds some threshold (established by more sys-
tematic correlation of the RSQ with actual road quality), the road would be flagged
as ‘rough’, and road maintenance personnel alerted. This is not a replacement of,
rather a compliment to, the RSQ measure, as the RSQ measure would be used
to identify localised events, consistent with say spilt rock or potholes. The RSQ
measure would provide road maintenance personnel with additional, more detailed
information.
67
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
4.8 Case studies in the use of the RSQ index
This section presents three case studies providing specific analysis based on the RSQ
measure.
4.8.1 Case study 1: Day-shift of October 4, 2004
Water is used to suppress dust at the draw point. On the day shift of October 4,
2004, water from draw point made its way to the 58 Orange 57 haul road causing
it to become muddied. As a result, the road degraded rapidly, resulting in the road
surface shown in Fig. 4.12.
Figure 4.12: The haul route road at the end of the day shift on the 4th of Octo-ber 2004. The light source is displaced vertically downwards to allow road surfaceirregularities to present more clearly.
Figure 4.13 shows the RSQ over the day shift of October 4, 2004. The magnitude of
the RSQ is significantly larger over this entire shift than that on any other shift seen
during testing (e.g. Fig. 4.10). The portion of road worst affected by this accelerated
68
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
degradation is between samples 10000 and 14000. This corresponds to the section
of road between the y-coordinates −100 and −50. This is the segment of the road
travelled in second gear, closest to the draw-point.
The photographs in Fig 4.12 and 4.1(b) depict the road surface of this badly affected
area at the end of the shift and it is clear from these that corrugations were present
on the road surface, as a repeating pattern is observed. It is well established that
road corrugations form with a wavelength corresponding to the dynamic properties
and speed of the vehicle repeatedly passing over it. They are the most likely cause
of the increased RSQ measure.
Figure 4.13: A waterfall plot of the differential RSQ over the entire day shift on the4th of October 2004.
Fig. 4.14 suggests the road surface was already badly degraded at the start of the
shift and continued to degrade with increasing runs. In other shifts, the RMS-RSQ
approached a maximum value of between 15 and 20. In this instance, however,
69
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
although the trend follows the same shape, the RMS-RSQ appears to approach
approximately double the value seen in other shifts. Water on the roads is deduced
to be the cause of these high RSQ values.
Figure 4.14: The RMS-RSQ for every run over the shift on October 4, 2004.
4.8.2 Case study 2: Day-shift of October 18, 2004
The case study deals with the putative appearance of what seems most likely to be
spilt rock. Figure 4.15 shows a 2D waterfall plot of the RSQ with increasing passes.
When compared to the same plot from the night-shift of the 16th/17th (Fig. 4.9(a)),
taking into account the difference in the number of runs, it can be seen that the
data measured on the 18th shows some spikes of high RSQ that are non-existent in
the data from the 16th/17th. An aspect of these spikes worth noting is that they
are not consistent throughout the shift.
Two spikes are present in the data. The first occurs at sample number 7700 in the
70
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
Figure 4.15: A 2D waterfall plot of the RSQ over the entire day shift on the 18th ofOctober 2004.
data obtained from run 6 to 40. This corresponds to a y-coordinate of −25m(refer
to Fig. 4.7(a)). A second spike appears at sample 7000 (a y-coordinate of −15m)
from sample 46 onwards.
There are several plausible reasons for the spikes in the RSQ data, but the most likely
is spilt rock on the tramming route. It was recorded that this shift had particularly
bad quality ore. This would explain the appearance, disappearance, and change in
location of the high RSQ spikes. The vertical acceleration of the rear part of the
LHD during the time corresponding to the spike at sample 7000 during run 60 is
shown in Fig. 4.16, suggesting an impulse input consistent with spilt rock.
The LHD body clearly shows a sudden high magnitude downwards acceleration,
followed by an even larger upwards acceleration. This is conjectured to be associated
71
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
Figure 4.16: A plot of the rear accelerometer data (converted to g) from run 60 onthe day shift of October 18, 2004 at the time of the large RSQ spike.
with the LHD’s front wheel passing over the rock on the road. The front section of
the LHD moves vertically upwards, causing the rear section to pitch and the rear
end of the LHD to accelerate vertically downwards. As the rear wheel passes over
the rock, the rear accelerometer experiences even larger acceleration, this time in
the positive z direction (vertically upwards).
By looking at the RSQ data for individual runs (not presented here), it can be seen
after which runs the RSQ rock(s) on the road appeared and disappeared. It can
then be deduced that the rock causing the RSQ spike at sample 7700 was dropped
from the bucket onto the road during the full-tram during run 4, remaining there
until it was removed when the road was graded in-between runs 41 and 42 (it is clear
that the road was graded during this time because the RSQ over the entire road
shows a definite decrease in magnitude, see Fig. 4.15). Another rock was dropped,
72
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
or the same rock was knocked back, onto the LHD wheel-path on run 45 or 46 and
continued to affect the RSQ until the end of the shift.
The appearance of two rocks on the road during a single shift seems improbable,
so it is more likely that the offending rock was displaced by the grader but then
knocked back onto the road, possibly due to run-to-run wheel track variation by the
Autotram LHD.
4.8.3 Case study 3: Day-shift of October 19, 2004
On the day-shift of October 19 2004, a sudden reduction in RSQ occurred as shown
in the 2D waterfall plot in Fig. 4.17(a). The system was unattended for this shift
and so the cause is not directly known. Each plot in this figure is an average of three
runs, so it is not clear whether this reduction occurred instantly or over several runs.
However, by looking at the RSQ for each of the first 25 runs in Fig. 4.17(b), it is clear
that the reduction in RSQ occurred between runs 16 and 17. This coincides with a
break in the shift of one-and-a-half hours, so this reduction in RSQ is assumed to
be associated with the road being graded.
As expected, following the grading, the RSQ began to increase again, see Fig. 4.18.
The RMS-RSQ seems to approach a final value between 15 and 20, which matches
the final RMS-RSQ from the 16th, shown in Fig. 4.11. Additionally, as in the data
from the 16th, it takes approximately 20 runs for the RMS-RSQ to increase to a
value of 15. Grading of the haul road clearly improves road quality, but in this case
it would appear that grading was carried out sooner than necessary because the
RSQ was not yet near its highest value.
After grading, the RSQ shows significantly lower values over the entire road surface,
with some minor exceptions. In particular, run 17 shows areas of high RSQ (in
particular, at sample 10000) that are not visible in subsequent runs. This is most
likely due to loose road surface after grading. The road quickly settles, however.
73
4.8. CASE STUDIES IN THE USE OF THE RSQ INDEX
(a) RSQ data plotted for the entire shift
(b) RSQ data plotted for the first 25 runs of the shift
Figure 4.17: 2D waterfall plots of the RSQ data obtained on the 19th of October2004.
74
4.9. SUMMARY AND CONCLUSIONS
Figure 4.18: The RMS-RSQ of the RSQ data obtained during the day shift of the19th of October, 2004.
4.9 Summary and conclusions
From the data presented in this chapter, it is clear that an LHD retrofitted with a
single accelerometer on the rear of the vehicle can be used to identify the difference
between a smooth road and a rough road. Specifically, an index is calculated to
provide this information to Autotram operators. It has also been shown that the
RSQ can be used to identify the road degradation trend and predict significant road
deterioration before the LHD accrues damage.
The major caveat on the results presented is that actual road quality has not been
systematically compared with the computed RSQ. Field trials planned to explore
this issue did not yield the required data because of site equipment failures and
75
4.9. SUMMARY AND CONCLUSIONS
unacceptable digging conditions. Calibration of RSQ data with road quality obser-
vations is something that would help prove the usefulness of the RSQ index and is
recommended for future work.
The next chapter reveals some interesting outcomes associated with a specific equip-
ment failure that occurred during trials. Chapter 6 then presents an outline of a
research plan to address the unanswered questions arising from this chapter and the
next.
76
Chapter 5
The RSQ monitor as a collision detection
tool
This chapter explores the possibility of using the RSQ monitor to detect collisions
between the Autotram LHD and its operating environment and to alert the Autotram
system or remote operators to potentially significant damage events. Such events
were observed in field trials during one of the data collection periods and resulted in
LHD breakdown with significant production loss.
5.1 Introduction
Consider the following scenario. A remote operator is using an Autotram LHD to
clear large rock from the ore-pass with a scooping motion. The LHD bucket makes
continued and regular contact with the exposed edge of the grizzly through which
the ore must pass. These impacts are sufficient to eventually cause a catastrophic
failure of the drive transmission.
Arguably, if the machine were being operated by an on-board operator, he/she
would have sensed these impacts and taken action to avoid them (indeed it is not
77
5.2. OPERATOR FEEDBACK
uncommon for operators to suffer injury as a result of such events). The remote
operator, however, remained blissfully unaware of the damage being inflicted. This
chapter examines the potential and limitations of the RSQ monitor for detecting
such damage events for the purpose of raising an appropriate alarm to the operator.
5.2 Operator feedback
There are three major forms of sensory feedback relevant to collision detection in-
herent in on-board operation that are not provided to remote operators:
• Vehicle vibration, impact (collision) and other motion.
• Full 360 degree view from the cabin.
• Directional sound. e.g. knowing from which direction the damage sound came.
Audio and video is fed to the Autotram control room as a partial solution to items
two and three but there is no feedback to account for the lack of the most important
of these three forms, the physical motion felt by the operator when sitting in the
cabin. This has three important implications:
1. Reduced spatial awareness: Remote operators don’t get the same ‘feel’
for their surroundings as they do when they are operating on-board. With
this reduced awareness, remote operators expose the Autotram to collisions
of increased frequency and intensity. Anecdotal evidence is that this is more
significant for inexperienced remote operators, such as operators in training,
but also remains an issue for more experienced operators. In the case discussed
in this chapter, the collision occurs between the tip of the bucket and the grizzly
lip.
78
5.3. LHD UL038 - 15/16TH OF JUNE, 2005
2. Difficulty determining whether a collision has occurred: An LHD’s
brakes are often applied at the last minute to avoid a collision. Following such
an incident it is difficult for remote operators to know whether the brakes
were applied in time, or whether the LHD made contact with the obstacle.
As a result, operators are unable to make the necessary corrections to their
technique or know whether possible LHD damage has occurred. It is common
for operators to repeatedly and unknowingly damage Autotram because they
no longer feel the damage occurring.
3. Difficulty determining whether a collision has resulted in LHD dam-
age: An on-board operator can readily assess collision intensity from the
physical feedback of the machine - it is not uncommon for them to suffer mi-
nor injury as a result of a collision. As discussed, one of the advantages of
Autotram is that remote operators avoid injury, but this has an inherent dis-
advantage: remote operators no longer receive feedback of the intensity of a
collision. Consequently, Autotram operators have no way of knowing whether
damage has occurred to the LHD.
Together, these effects lead to an increased number of LHD collisions and reduced
knowledge of the severity of collision. The damage caused can result in signifi-
cant downtime and high maintenance costs associated with repairs and replacement
parts. This chapter investigates whether the RSQ monitor can be used to provide
information on LHD collisions with a view to avoiding LHD damage.
5.3 LHD UL038 - 15/16th of June, 2005
At 10:00 am on the 16th of June 2005, the RH2900 LHD UL038 was found to have
sustained significant damage. During the LHD’s routine twice-daily service, the
maintenance technician found the transmission cover plate had failed catastrophi-
cally (Fig. 5.1) along with the transmission mounts, allowing the transmission to
79
5.3. LHD UL038 - 15/16TH OF JUNE, 2005
Figure 5.1: Damage to transmission cover plate.
Figure 5.2: Damage to DAS remote control logic box and main steering hose.
80
5.3. LHD UL038 - 15/16TH OF JUNE, 2005
move approximately 20cm. The DAS remote control logic box (attached to the
transmission) had collided with the main steering hose, damaging both (Fig. 5.2).
This was determined to be a result of the LHD colliding with an obstacle whilst
travelling forwards.
Neither of the remote operators, nor the auto-mechanic who serviced the LHD at
1:30 am, nor the on-board operator who trammed the LHD to the workshop at
10:00 am, noticed anything wrong. Either the damage occurred between 1:30 and
10:00 am, or it occurred before this time and went unnoticed at the 1:30 am service.
In either event, the damage could have occurred over an extended period of time
(multiple collisions) or as a result of a single high-energy collision.
5.3.1 Background information
Ore that is too large to fit through the grizzly is called ‘oversize’. Oversize needs
to be broken to prevent blockage and so it can proceed through the system for
processing. This is known as ‘clearing (or cleaning) the ore-pass’ and is carried out
using one of two methods:
1. A rockbreaker splits the oversize and feeds it through the grizzly while the LHD
is bogging the next bucket. This way the LHD can continue uninterrupted,
resulting in maximum productivity.
2. The LHD bucket is used to scrape along the ground in a scooping motion
up to, and over the grizzly, picking up the oversize before raising the bucket
and dropping it back onto the grizzly. This is repeated until the ore has been
sufficiently fragmented to pass through for processing. It is a time consuming
activity for LHD operators and productivity is adversely affected.
On board operators are assigned a rockbreaker if there is sufficient oversize in the ore
to justify it. However, personnel (including rockbreaker operators) are not permitted
81
5.3. LHD UL038 - 15/16TH OF JUNE, 2005
in the Autotram area for health and safety reasons so remote-operators are forced
to use the second method and clean the ore-pass themselves. A remotely operated
rockbreaker has been used to overcome this problem at ODM in the past, but is no
longer in operation.
Figure 5.3: Grizzly with rails to prevent LHD damage.
Not only does the second method of clearing the pass lead to reduced productivity,
but it can also lead to other problems. One such problem is the focus of this
chapter. When a grizzly is initially installed, rails are also installed leading up to
and terminating at its edge (see Fig. 5.3). When clearing the ore-pass, an Autotram
operator lowers the bucket into a scooping position and moves forward, scraping the
bucket along these rails and over the grizzly. The force between bucket and rail is
often high so the heat generated due to friction can be intense.
A flaw in the old rail design has, in some cases, caused the rails to curl upwards
under this intense heat, blocking access to the ore-pass. The warped section of rail
is removed to allow production to continue and the rails are not replaced because
82
5.3.LH
DU
L038
-15/16T
HO
FJU
NE
,2005
Figure 5.4: LHD and grizzly. Detail shows how bucket tip collides with edge of Grizzly.
83
5.4. RESULTS
over a week of downtime is required at substantial cost. Subsequently, the ground
in front of the grizzly is no longer protected from wear and with time, the grizzly
lip becomes exposed (see Fig. 5.4). This exposed edge is a hazard because the LHD
bucket can make contact with it whilst clearing the ore-pass, experiencing excessive
jarring forces, resulting in possible LHD damage.
On the 15th and 16th of June, 2005 the Autotram operator noticed that the bucket
tip was connecting the grizzly lip. He took steps to avoid it, but due to the limited
capability to accurately position the bucket, the operator was unable to prevent the
collisions occurring, as an on-board operator would. Limited feedback also meant
that the operator did not realise the severity of the impacts and was possibly less
careful than he would have been had he known of the damage caused. As a result of
the damage, the LHD was out of operation for a week with production loss running
to many hundreds of thousands of dollars.
5.4 Results
During the events that occurred on the 15th and 16th of June 2005 the RSQ mon-
itor was able to measure the effect of the collisions on the LHD as they occurred.
This was not a planned experiment, but the resulting data shows the RSQ mon-
itor’s ability to provide feedback in addition to that relating to road roughness.
When selecting the accelerometers for measuring road roughness, one tri-axial ac-
celerometer was used in case the x and y (lateral and longitudinal) acceleration data
contained useful information. During the incidents of the 15th and 16th of June,
the y-direction (longitudinal) acceleration was logged and has since been found to
contain information that is useful for determining the nature of the collisions.
LHD collisions are seen as ”spikes” in the y-direction accelerometer data and can
therefore be identified simply by looking for large acceleration values. Any instan-
taneous acceleration of over 10g is unlikely to be caused by anything other than the
84
5.4. RESULTS
sudden reduction in speed associated with a collision. Therefore the magnitude will
be used as the sole criteria for detecting collisions.
For information on the occurrence of said collisions to be of immediate use, it is im-
portant that this data is fed to the Autotram control room in realtime to inform the
remote operator of the event. This can be done by attaching additional information
to the RSQ data packets.
Data logged on the 15th and 16th of June 2005 will be analysed to establish the
RSQ monitor’s ability to accurately determine, for each collision:
1. the cause (Section 5.4.1); and
2. the intensity (Section 5.4.2).
With this additional information, the operator is not only alerted to potentially
damaging collisions, but is supplied with information on the possible reason for
these collisions and an estimate of the resulting damage. This allows the operator to
make more informed decisions when taking evasive action. A log of this information
including the total number of incidents and the time of occurrence for each is kept
for analysis.
5.4.1 Identifying cause of collisions
It is useful to know the cause of each spike in the acceleration data, so this infor-
mation can be passed onto the operator. In this specific case, reports given by the
Autotram operators on shift identify the LHD damage as resulting from the bucket
coming in contact with the grizzly lip. The question is, however, whether it is pos-
sible to see this from the data alone and whether it is possible to establish other
causes for acceleration spikes.
85
5.4. RESULTS
There are many aspects of the data that can be looked at when attempting to
identify the reason for a collision, the most useful being:
1. LHD location; and
2. direction of travel.
Once the LHD location and direction of travel are known, the LHD’s activity (bog-
ging, tramming, dumping, cleaning pass) at the time of the collision can be hypoth-
esised. These aspects of the data will therefore be looked at in more detail in the
following pages.
LHD Location
The first step towards determining the cause for an acceleration spike in the data
is to look at the location of the LHD at the time of the incident. The location of
the LHD when the large acceleration spikes occurred for the 15th and 16th of June
is plotted against a map of the tramming route in Fig. 5.5, with the dump point
shown in greater detail.
All acceleration spikes occur at the ore-pass and the majority of these are due to
the LHD colliding with the grizzly. The circle boundary in the figure corresponds to
the estimated position of the LHD when this collision occurs (7.5m from the grizzly
centre - based on the distance from the centre of the grizzly to the coordinate centre
of the LHD as shown in Fig. 5.4). Any spikes occurring outside this central group
are due to the LHD colliding with something other than the grizzly lip, possibly the
wall at the back of the ore-pass or the side walls. These are marked with a cross.
It can also be seen that although most of the points occur near this circular line,
many do not occur on the line. There are several possible reasons for this:
1. Error in DAS positioning co-ordinates. (relatively insignificant - approximately
10cm maximum).
86
5.4. RESULTS
Figure 5.5: Longitudinal LHD acceleration vs time (hrs) for the 15th and 16th ofJune. Points of high acceleration are shown: green = > 10g, yellow = > 15g, red= > 20g, where 1g is approximately 9.81m/s2. The X is the location of the grizzlycenter and the traced circular line has a radius equivalent to the distance from theLHD’s coordinate centre to the grizzly when the LHD bucket is touching the grizzlyedge. The acceleration spikes marked with a cross are spikes deemed to be a resultof something other than grizzly collision.
87
5.4. RESULTS
2. When the LHD approaches the grizzly at an angle, the corner (rather than the
centre) of the bucket can make contact with the grizzly edge, or the bucket can
make contact with the corner of the grizzly. As a result, the LHD coordinate
centre is further from the grizzly centre at the time of impact.
3. The bucket may have not been lowered completely, and angled towards the
ground so the bucket teeth are at ground level. Consequently, the LHD appears
closer to the grizzly centre at the time of collision.
With this taken into account, the location of the LHD in the data remains sufficiently
accurate to be a valid indicator of LHD location. It can therefore be determined
whether the collision took place near the ore-pass, the stope or whilst tramming
between. It can also be hypothesised whether an LHD collision at the ore-pass is
with the grizzly or the surrounding walls. This information can be displayed on the
Autotram graphical user interface (GUI) as feedback to the operator. In this case,
it can be seen that the majority of data spikes are due to collisions with the grizzly
lip.
Direction of Travel
The LHD’s direction of travel at the time of collision can be used to confirm these
observations. It is more reliable to look at the sign of the acceleration spikes than the
sign of the velocity data because, according to DAS, the raw speed data is inherently
inaccurate and are more likely to be subject to delay, as they are received over the
network. High acceleration represents a sudden change in LHD velocity and in this
case, the accelerations are negative, so the sudden change in speed is due to the LHD
stopping suddenly (decelerating). The sign convention used is as follows: a positive
spike comes from the LHD stopping suddenly whilst travelling in reverse; and a
negative spike comes from the LHD stopping suddenly whilst travelling forwards.
It can be seen from Fig. 5.6 that all bar one of the acceleration spikes are negative,
88
5.4.R
ESU
LT
S
Figure 5.6: Longitudinal acceleration vs Time of UL038 over two days. Vertical dividers represent shift change and ’S’
represents a LHD service. Time is in 24 hour time in the format HH:MM.
89
5.4. RESULTS
corresponding to the LHD stopping suddenly whilst travelling forwards. This sup-
ports the previous evidence indicating that the spikes are due to the LHD colliding
with the grizzly lip. The positive acceleration spike is discussed in Sec. 5.4.2.
5.4.2 Estimating Collision Intensity
Excessive damage to the Autotram LHD caused by collisions was the motivation for
this work. It would therefore be desirable to estimate the resulting LHD damage.
Unfortunately, since a physical model of the #RH2900 LHD is unavailable, this is
not possible. It is possible, however, to estimate the energy dissipated in, and the
forces inflicted on the LHD for each collision. This information can then be used as
an indicator of the damage caused to the LHD.
The mass of the LHD and the magnitude of the acceleration spike can be used to
estimate the force experienced by the LHD at the time of collision. The bucket is
empty when performing the scooping motion which leads to bucket/grizzly collision,
so the LHD mass is assumed to be approximately 55000kg. For a worst-case scenario,
we will look at the collision resulting in the largest acceleration spike, an acceleration
of approximately 23.5g (or 230.5m/s2). With the force on the LHD is given by
F = ma, the maximum force on the LHD is 12.677MN.
Since the LHD is stopping almost instantaneously, and the sampling frequency is
limited, aliasing becomes a problem. This can be seen by looking at one of the
acceleration spikes in more detail. Fig. 5.7 depicts a time-scaled plot of y-acceleration
vs time, showing one of the collisions in detail. It can be seen from this plot that
the sampling rate is insufficient to describe these high speed data spikes accurately.
The initial acceleration is negative, and the acceleration oscillates towards zero, but
the magnitude of the initial acceleration and the number of oscillations is uncertain.
Fig. 5.8 shows a time-scaled view of the positive data spike mentioned previously. It
would be easy to assume that this is due to a collision of the LHD whilst travelling
90
5.4. RESULTS
in reverse, but on closer inspection, it is apparent that the initial collision occurred
whilst travelling forwards and this spike is due to post-collision LHD body vibration.
This is a prime example of the detrimental effect of aliasing on the measurement of
collision data spikes.
Of particular relevance is the resulting uncertainty associated with the magnitude of
acceleration spikes. Without an accurate knowledge of magnitude, there is no way
of determining LHD damage. As a result, it is not possible to estimate LHD damage
with any accuracy unless the sampling frequency is increased to 500Hz or 1kHz. It
is, however, still useful to know of the occurrence of said collisions. This should be
carried out with caution, with limited faith placed in the accuracy of the magnitude
of these acceleration spikes. The measured magnitudes of the acceleration spikes can
therefore be considered conservative estimates of the acceleration felt by the LHD.
91
5.4. RESULTS
Figure 5.7: A time-scaled plot of y-acceleration Vs. time for one of the collisions.Sample points, represented by hollow circles, are 1/200sec apart. Time units areHH:MM:SS:SSSS in 24 hour time.
Figure 5.8: A time-scaled plot of y-acceleration Vs. time for the positive accelerationspike. Sample points, represented by hollow circles, are 1/200sec apart. Time unitsare HH:MM:SS:SSSS in 24 hour time.
92
5.5. CONCLUSION AND FUTURE WORK
5.5 Conclusion and Future Work
It is proposed that a system be put in place to alert remote operators to longitudinal
and lateral impacts of over 10g using a visually obvious graphical indicator on the
Autotram main display. The RSQ monitor has the ability to provide this informa-
tion, requiring only minor software alterations. Once operators have been alerted,
they can take the necessary precautions to ensure that these collisions do not recur.
The RSQ monitor’s sampling frequency could be increased to improve the ability to
measure the amplitude of high frequency acceleration spikes, however this modifica-
tion is unnecessary for two reasons: Firstly, the current sampling frequency of 200Hz
is sufficient for the RSQ monitor’s primary purpose, measuring vertical vibration;
and secondly, it is unnecessary to accurately know the magnitude of the acceleration
spikes so long as the number of spikes is known. The current system provides this.
93
Chapter 6
Conclusions and recommendations for
future work
This chapter summarises the work completed and makes recommendations on a pos-
sible structure for field trials that would allow for more complete correlation of the
RSQ measure with actual road condition.
6.1 Summary
The main outcome from this thesis has been the demonstration that it is possible
to monitor road condition from measurements of vertical acceleration on automated
LHD vehicles. The thesis has exploited the controls that these automated vehicles
provide, most importantly the control of speed to discrete levels corresponding to
gear changes and the consistent timing of these gear changes.
The next logical phase of the work is to undertake a systematic longitudinal study
to establish the performance of the monitor and the associated RSQ measure. This
will establish more precisely how the RSQ is affected by vehicle speed and payload,
the ability of the RSQ to detect spilt rock, potholes and corrugations, and to better
correlate changes in the RSQ with changing road condition.
94
6.1. SUMMARY
Seven proposed aims of the field trials are given and five experiments are suggested
to achieve these aims. The methodology for this is sketched out below.
6.1.1 Aims of field trials
It is suggested that the systematic study be based on a series of well defined exper-
iments with the following aims.
Aim 1: Establish how the road degrades with increasing number of passes
The first aim addresses a question raised by the data in this thesis, namely, why does
the RSQ change rapidly at first, but at a reducing rate as the road degrades, see
Fig. 4.11. The suggested reasons are:(i) the road reaches a critical level of roughness
and due to the mechanics of the LHD and the contact between the tyres and road it
does not degrade any further; or (ii) the road continues to degrade, but the vertical
acceleration of the LHD does not change dramatically because there is a limit to
the vibrational energy the LHD can receive. The question is important because
it directly asks the nature of the non-linear relationship between RSQ and road
conditions.
To resolve this question, surveys of the road throughout the shift need to be under-
taken to determine whether the road degradation rate diminishes or continues to
degrade at the same rate. This needs to be compared to RSQ data.
Aim 2: Establish how traversing speed affects LHD vertical accelera-
tion/RSQ
The literature, especially that relating to the development of the IRI (Sayers, Gille-
spie and Queiroz et. al. [8, 7, 1, 9, 3]), argues that the speed of a vehicle as it
passes over a road is a major determining factor of the resulting road roughness
95
6.1. SUMMARY
measurements as measured by a response-type instrument. However, these papers
are interested only in light vehicles travelling mainly on highways. It is not clear to
what extent the RSQ measure is affected by vehicle speed.
This study differs from the bulk of those done in the literature in many ways, the
most significant differences are: (i) lack of spring suspension; (ii) significantly higher
mass (approximately 50-75t); (iii) discrete operating speeds; (iv) dual chassis design;
and (v) unsealed roads. It is possible that these differences mean that traversing
speed is less of influence on the RSQ than it is on the road roughness measures in
the literature.
Specifically, this aim seeks to develop a relationship between operating gear (as a
proxy for vehicle speed) and the resulting RSQ to allow direct comparison while
travelling in different gears.
Aim 3: Establish how traversing speed affects road degradation rate
This aim is related to Aim 2, but looks more specifically at the effect of traversing
speed on road degradation rate, rather than vehicle response. This aim also stems
from the differences between this experimental situation and those studied in the
literature. As discussed in Aim 2, there are many differences, but the one of interest
here is the difference in the road surfaces studied. LHD vehicles operate on unsealed
roads, whereas most studies looked at in the literature concentrate solely on sealed
roads. Those that do look at unsealed roads, usually do so only in passing; the
degradation rate of these roads is not studied at all.
It is plausible that for higher traversing speeds, the road degrades more quickly.
To test this hypothesis, it is necessary to operate the LHD at different speeds with
similar initial road conditions and see how the road degrades with time.
96
6.1. SUMMARY
Aim 4: Establish how payload affects LHD vertical acceleration/RSQ
Payload is a contributing factor to the RSQ measure, as the dynamics of the LHD
are affected by increased bucket mass. As part of its normal cycle, an LHD travels
with the bucket full from draw-point to ore-pass and empty on the return trip. The
LHD is 15t heavier when the bucket is full. Although there is a clear difference in
LHD mass between the full- and empty-tram, there are other variables that prevent
these data sets being compared directly.
To establish the effect payload has on the RSQ, the LHD must be driven over the
same road with the bucket full and then empty, while travelling in the same direction
at the same speed.
Aim 5: Investigate the possibility of detecting spilt rock or potholes using
LHD vertical acceleration/RSQ
Rock often spills from the bucket onto the haul route during the full-tram. Some-
times these rocks are over a tonne in mass and can cause the LHD to lose control
or drop more rock onto the road. If the RSQ monitor could detect spilt rock on the
road and alert the operator, he/she could remove this rock before continuing.
To achieve this aim, it is necessary to intentionally place or spill rock onto the road,
take note of its location and continue normal operation.
Aim 6: Establish the value of RSQ that requires a reduction in operating
gear or road regrading
This is perhaps the most important aim that can be achieved. As it stands, the
RSQ has not been correlated with actual road profiles. For the RSQ to be useful, it
needs to be calibrated and a value of RSQ associated with predefined critical levels of
road roughness. When the road degrades to a certain point, manual operators decide
97
6.1. SUMMARY
that the road is too rough and needs grading. The RSQ needs to be calibrated with
operator knowledge, so Autotram operators can be supplied with this information
and prevent unnecessary damage to the LHD.
6.1.2 Experiments addressing longitudinal study aims
Five experiments addressing the above aims are discussed below.
Experiment 1: Road degradation trend under ’normal’ operation condi-
tions
This proposed experiment addresses Aims 1, 5, and 6, based on the following
methodology:
1. Grade roads.
2. Do twenty ‘normal’ runs.
3. Stop Autotram operation.
4. Survey roads and take note of estimated road roughness and corrugation am-
plitudes.
5. Mark the following information on a map of the tramming route:
a) an estimate of the quality of the road;
b) location and estimated size of spilt rock; and
c) location, amplitude and wavelength of corrugations.
6. Repeat steps 2-7 until road is ‘too rough to pass’.
7. When the Autotram operator believes that the roads have degraded to a level
where it is unsafe to continue Autotram operation, he/she logs: (i) the current
98
6.1. SUMMARY
time; (ii) the reason for this decision; (iii) the location on the track that is
most dangerous; and (iv) the course of action taken.
This procedure would result in data showing roads going from smooth to rough,
with manual observations of road surface throughout the shift for comparison. The
road survey data would come in the form of a map of the tramming route, colour
coded according to estimated road roughness, with the location and estimated size
of spilt rock and corrugations marked. Photos of key sections of road would refer to
markers on the map indicating the location and direction the photo was taken.
Experiment 2: The effect of limiting LHD speed
This proposed experiment addresses Aims 1, 2, 3 (when combined with experiment
3), and 6, based on the following methodology:
1. Grade roads.
2. Do nine ‘normal’ runs (to give an overall cycle of ten).
3. Limit maximum speed (this can be done easily using the Autotram GUI in-
terface).
4. Do 1 run traversing both directions at the slower traversing speed.
5. Remove speed limit.
6. Repeat steps 2-5 until road is ‘too rough to pass’.
7. When the Autotram operator believes that the roads have degraded to a level
where it is unsafe to continue Autotram operation, he/she logs: (i) the current
time; (ii) the reason for this decision; (iii) the location on the track that is
most dangerous; and (iv) the course of action taken.
99
6.1. SUMMARY
Experiment 3: The effect of increasing LHD speed
This proposed experiment addresses Aims 1, 2, 3 (when combined with experiment
2), and 6, based on the following methodology (a complement of experiment 2):
1. Grade roads.
2. Limit maximum speed (this can be done easily using the Autotram GUI in-
terface).
3. Do nine runs at reduced speed (so that each cycle consists of ten runs).
4. Remove speed limiting.
5. Do one ‘normal’ run.
6. Repeat steps 2-5 uninterrupted until road is ‘too rough to pass’
7. When the Autotram operator believes that the roads have degraded to a level
where it is unsafe to continue Autotram operation, he/she logs: (i) the current
time; (ii) the reason for this decision; (iii) the location on the track that is
most dangerous; and (iv) the course of action taken.
Experiment 4: The effect of varying LHD payload
This proposed experiment addresses Aims 1 (in part), 4, and 6, based on the fol-
lowing methodology:
1. Grade roads.
2. Do nine ‘normal’ runs.
3. Do one run without emptying payload at ore-pass. Travel to and from ore-pass
with full bucket.
100
6.1. SUMMARY
4. Do one ’normal’ run (take full payload to ore-pass, empty it and return to
stope).
5. Do one run without filling bucket at stope. Travel to ore-pass with empty
bucket and return with empty bucket.
6. Do seven ’normal’ runs.
7. Repeat steps 3-6 uninterrupted until road is ’too rough to pass’.
8. When the Autotram operator believes that the roads have degraded to a level
where it is unsafe to continue Autotram operation, he/she logs: (i) the current
time; (ii) the reason for this decision; (iii) the location on the track that is
most dangerous; and (iv) the course of action taken.
Experiment 5: Spilt rock detection
This proposed experiment addresses Aims 1(in part), 5, and 6, based on the following
methodology:
1. Grade roads.
2. Do five ‘normal’ runs for a base data-set.
3. Stop Autotram.
4. Scatter oversize onto the road using LHD in manual mode. Dimensions of
oversize should be determined according to operator experience.
5. Survey roads and take note of estimated road roughness and corrugation am-
plitudes. Take photos of key sections (spilt rock, corrugations, muddy areas,
etc.) with a length reference in frame for size comparison. Mark the following
information on a map of the tramming route (especially b)):
a) Estimated quality of the road.
101
6.2. CONCLUDING REMARKS
b) The location and estimated size of spilt rock.
c) The location, amplitude and wavelength of corrugations.
d) The location and direction of each photo taken.
6. Restart Autotram.
7. Do ten ’normal’ runs.
8. Repeat steps 3-7 until road is ’too rough to pass’ (only repeat 4 if the spilt
rock is no longer affecting the LHD).
9. When the Autotram operator believes that the roads have degraded to a level
where it is unsafe to continue Autotram operation, he/she logs: (i) the current
time; (ii) the reason for this decision; (iii) the location on the track that is
most dangerous; and (iv) the course of action taken.
6.2 Concluding remarks
The work planned for this thesis sought to address, inter alia, the aims for the
suggested experimental program. Three trials were attempted to collect the required
data. All were unsuccessful in meeting the requirements. In the first trials, a sensor
failure curtailed work. In the second, the transmission dropped out from under
the Autotram LHD, see Chapter 5, and only limited data was collected. The third
attempt yielded no data because the stope conditions were too extreme for effective
Autotram use.
Nevertheless, the basic material of this thesis provides a sound foundation on which
these trials could proceed and from which a genuinely effective road condition mon-
itor could be developed. This is an essential next step for this work.
102
Bibliography
[1] M W Sayers, T D Gillespie, and W D O Paterson. Guidelines for conducting
and calibrating road roughness measurements. Technical Paper WTP-46, World
Bank, 1986.
[2] V Rouillard, B Bruscella, and M Sek. Classification of road surface profiles.
Journal of Transportation Engineering, 126(1):41–45, 2000.
[3] M W Sayers. On the calcluation of international roughness index from longi-
tudinal road profile. Transportation Research Record, 1501:1–12, 1995.
[4] C J Dodds and J D Robson. The description of road surface roughness. Journal
of Sound and Vibration, 31(2):175–183, 1973.
[5] B Bruscella, V Rouillard, and M Sek. Analysis of road surface profiles. Journal
of Transportation Engineering, 125(1):55–59, 1999.
[6] V Rouillard, M A Sek, and T Perry. Analysis and simulation of road profiles.
Journal of Transportation Engineering, 122(3):241–245, 1996.
[7] M W Sayers, T D Gillespie, and C A V Queiroz. The international road rough-
ness experiment. Technical Paper WTP-45, World Bank, 1986.
[8] M W Sayers, T D Gillespie, and C A V Queiroz. The international road rough-
ness experiment: A basis for establishing a standard scale for road roughness
measurements. Transportation Research Record, 1084:76–85, 1986.
103
BIBLIOGRAPHY
[9] M W Sayers. Two quarter-car models for defining road roughness: Iri and hri.
Transportation Research Record, 1215:165–172, 1989.
[10] K M A Kamash and J D Robson. Implications of isotropy in random surfaces.
Journal of Sound and Vibration, 54(1):131–145, 1977.
[11] W R Stevens, B Fenner, and A M Rudoff. Unix Network Programming, vol-
ume 1. Addison-Wesley, Boston, MA, 3rd edition, 2004.
[12] D E Newland. Random Vibrations, Spectral and Wavelet Analysis. Longman
Scientific and Technical, Essex, England, 3rd edition, 1993.
[13] J S Benadat and AG Piersol. Random Data: Analysis and Measurement Pro-
cedures. Wiley Interscience, New York, 2nd edition, 1985.
[14] A N Heath. Application of the isotropic road roughness assumption. Journal
of Sound and Vibration, 115(1):131–144, 1987.
[15] A N Heath. Modelling and simulation of road roughness. Vehicle System
Dynamics, 18-Supp:275–284, 1989.
[16] W Jeong, K Yoshida, H Kobayashi, and K Oda. State estimation of road surface
and vehicle system using a kalman filter. JSME International Journal Series
III, 33(4):528–534, 1990.
[17] H Prem. A laser-based highway-speed road profile measuring system. Vehicle
System Dynamics, 17-Supp:300–304, 1988.
[18] V Rouillard, M Sek, and B Bruscella. Simulation of road surface profiles.
Journal of Transportation Engineering, 127(3):247–253, 2001.
[19] M Waechter, F Riess, H Kantz, and J Peinke. Stochastic analysis of surface
roughness. Europhysics Letters, 64(5):579–585, 2003.
104
BIBLIOGRAPHY
[20] D M Xu, A M O Mohamed, R N Yong, and F Caporuscio. Development of a
criterion for road surface roughness based on power spectral density function.
Journal of Terramechanics, 29(4/5):477–486, 1992.
105
Appendix A
Data analysis methodology
Data gathered during trials was collected for post-processing to assist with the de-
velopment of the RSQ monitor. The purpose of this post processing is to extract
and summarise the relevant information in the data, including information relating
to road degradation trends and the difference between full/empty bucket data. To
achieve this:
1. relevant data is extracted from the compressed text files;
2. this data is processed to isolate the useful information; and
3. the results are plotted.
This process has been thoroughly tested to ensure that the resulting plots are accu-
rate and provide insight into the data gathered by the RSQ monitor.
When loading the text files, only the variables of interest are extracted. These
are: time of clock-tick; z direction raw accelerometer data from front and rear
accelerometers; x and y coordinates, representing the location of the LHD; the LHD
bearing angle (direction the LHD is facing); the speed of the LHD; and the RSQ.
Bad data due to corruption (as discussed in Chapter 3) is replaced with data from
106
the measured accelerations immediately prior to the corruption to prevent the data
glitches from affecting the RSQ significantly. The RSQ is then recalculated using
this patched accelerometer data.
The processing step is used to isolate the desired information from the vast quantity
of data stored by the RSQ monitor. In the previous step, data is loaded according
to the day it was attained. This data is then sorted into ‘shifts’, as this is a more
relevant time-frame to work in than a 24 hour day. A single shift is then selected
for analysis (e.g. day shift on October 17th, 2004) and the shift is divided into a
number of ‘runs’ (the trip required to get one bucket of ore - from the ore-pass to
draw-point and back again). Data is then removed if it meets any of the following
conditions:
1. Shift is less than 30 minutes long: Shifts are only stopped short when there
is a problem with operation. Road roughness trends cannot be established
from such limited data.
2. Data measured while the LHD is stationary: This data is of little use
when measuring road roughness.
3. Short runs: This means data collection has been interrupted, usually due to
the Autotram system shutting down because of a wireless network fault. This
means the LHD comes to a halt while the network connection is re-established
and the system is restarted. As a result, the standard speed profile along the
haul route is disturbed and the data is unreliable and therefore cannot be
compared to other data.
4. Runs with unscheduled speed variations: Runs must all have the same
velocity gradient for the results to be comparable. If this velocity gradient
differs in certain runs, they must be removed so they do not affect the result.
107
Runs are then individually divided into ore-pass to draw-point and draw-point to
ore-pass datasets and collated. The relevant data is then used to create the plots
seen in this thesis.
108