 

. U ., ..‘ .. ... .V n... -‘v'v~‘- -......‘_......., ”.5... ....,....u...-.‘.... a..\...,,'._‘u.

 

u

 

 

 

         

     

:—_

       

 

, _ , . ‘

 

This is to certify that the
dissertation entitled

EFFECT OF PAVEMENT CONDITION ON VEHICLE
OPERATING COSTS INCLUDING FUEL CONSUMPTION,
VEHICLE DURABILITY AND DAMAGE TO TRANSPORTED
GOODS

presented by

IMEN ZAABAR

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Civil Engineering

 

 

A/Ajﬁ D?

Major Professor’s Signatilreil
57/31/20 I 0
I

Date

 

 

MSU is an Afﬁrmative Action/Equal Opportunity Employer

LIBRARY
Michigan State
Ul liversity

 

 

 

 

.-.—.—.—-_.—.-.-.—.—.-g—.-.-.—a~o—u— ‘4

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 KzlProleccaiPresIClRCIDateDue.indd

 

 

 

 

EFFECT OF PAVEMENT CONDITIONS ON VEHICLE OPERATING COSTS
INCLUDING FUEL CONSUMPTION, VEHICLE DURABILITY AND DAMAGE TO
TRANSPORTED GOODS

By

Imen Zaabar

 

 

 

 

 

 

 

 

 

 

 

 

 

 

EFFECT OF PAVEMENT CONDITIONS ON VEHICLE OPERATING COSTS
INCLUDING FUEL CONSUMPTION, VEHICLE DURABILITY AND DAMAGE To
TRANSPORTED GOODS
By

i Imen Zaabar

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Civil Engineering

2010

 

 

 

ABSTRACT
EFFECT OF PAVEMENT CONDITIONS ON VEHICLE OPERATING COSTS
INCLUDING FUEL CONSUMPTION, VEHICLE DURABILITY AND DAMAGE TO
TRANSPORTED GOODS
By

Imen Zaabar

Vehicle Operating Costs (VOC) including fuel consumption, repair and maintenance
and damage to goods are an essential part of life cycle cost analysis. They are inﬂuenced
by vehicle technology, pavement condition, roadway geometrics, environment and speed.
Many of these models were developed on the bases of data generated years ago for
vehicles that vary substantially from those used currently in the US. Therefore, there is a
need to collect new information that could help in reﬁning these models or developing
models that would better apply to US conditions.

Regarding fuel consumption, the thesis presents the calibration exercise of the Highway
Development and Management model (HDM 4) for US conditions using ﬁeld data
collected as part of this thesis. The results showed that the calibrated HDM 4 model was
able to predict very adequately the fuel consumption of ﬁve different vehicle classes
under different operating, weather and pavement conditions. The better accuracy
achieved aﬁer calibration has improved the prediction of the effect of roughness. The
comparison of sensitivity analyses before and after calibration showed that the effect of
roughness on fuel consumption increased by 1.75 for the van, 1.70 for the articulated
truck, 1.60 for the medium car, 1.35 for the SUV and 1.15 for the light truck. Also,
analysis of covariance was successfully used to extract the effect texture and pavement

type from the collected data. The analysis showed that a 67 % decrease in mean texture

depth will result in a 1.3 % and 0.9 % decrease in fuel consumption for heavy truck at 56
and 89 km/h, respectively. The analysis also showed that, the mean difference of fuel
consumption between asphalt and concrete pavements is statistically signiﬁcant only at
low speed for both heavy and light trucks and for summer conditions.

Regarding repair and maintenance (R&M) costs, two approaches were reported in the
thesis. The ﬁrst approach is to update Zaniewski’s repair and maintenance costs (i.e., the
latest comprehensive research conducted in the US) using the inﬂation rate of R&M
costs. The second approach is to use the mechanistic-empirical model developed as part
of this thesis to conduct fatigue damage analysis using numerical modeling of vehicle
response. The comparison between the results from both approaches showed that the
results agree up to an IRI of 5 m/km (95 percent of the roads in the US have IRI lower
than 5 m/km). These ﬁndings Show that the mechanistic—empirical approach is promising
because it is more ﬂexible than the empirical approach. Also, because of the mechanistic
nature of the model, it could be used by state highway agencies (SHA) to correct for the
effect of roughness features on vehicle durability at the project level.

Regarding damage to goods, a mechanistic-empirical approach was proposed to
conduct product fragility assessment using numerical modeling of vehicle and product
vibration response. The model could also be used at the project level.

In summary, this research provided models applicable to the United States. Such
models will provide SHA with the tools necessary for considering VOC in evaluating
pavement-investment strategies and identifying options that yield economic and other
beneﬁts. In addition, this thesis proposes a newly developed tool to detect, localize and

identify roughness features.

Copyright by
IMEN ZAABAR
2010

DEDICATION

To my father’s soul
To my mother
Without whom this work could not have been completed

A l'ame de mon pere

A celui qui m'a indiqué la bonne voie en
me rappelant que la volonté fait toujours
les grands homes

A ma mere
A celle qui a attendu avec patience
les fruits de sa bonne education

L'amour que je vous porte n'a d‘égal
que l'infmi

ACKNOWLEDGEMENTS

ke to express my gratitude to Professor Karim Chatti for his friendship,
18 and guidance throughout my stay at Michigan State University. His
sion and expertise guided me in the development of this research. He always
ck my conﬁdence whenever I lost it. I also would like to extend my
to Professors. Ronald Harichandran, Clark Radcliffe and Richard Lyles and

ed Emin Kutay for reading this dissertation and offering constructive

my parents, my husband and my family for their continuous emotional

ted to my fellow graduate students at the Civil Engineering Department for

the fuel consumption ﬁeld tests.

ch in this dissertation has been supported by the Michigan Department of

m and the National Cooperative Highway Research Program of the National

vi

 

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................... xi
LIST OF FIGURES .................................................................................................... xiv
CHAPTER 1
INTRODUCTION ......................................................................................................... 1
1.1 MOTIVATION ............................................................................................... l
1.2 RESEARCH HYPOTHESIS AND OBJECTIVES ........................................ 5
1.3 RESEARCH SCOPE ...................................................................................... 7
1.4 DOCUMENT ORGANIZATION ................................................................ 10
CHAPTER 2
BACKGROUNG AND LITERATURE REVIEW ..................................................... 12
2.1 INTRODUCTION ........................................................................................ 12
2.2 BACKGROUND .......................................................................................... 12
2.3 OVERVIEW OF EXISTING VOC MODELS ............................................ 15
2.3. 1 Introduction ......................................................................................... 1 5
2.3.2 Empirical Versus Mechanistic VOC Models ...................................... 18
2.3.2.1 Empirical models ............................................................................. 18
2.3.2.2 Mechanistic-Empirical models ........................................................ 19
2.4 SUMMARY ................................................................................................. 19
CHAPTER 3
FUEL CONSUMPTION MODELS ............................................................................ 21
3.1 INTRODUCTION ........................................................................................ 21
3.2 IDENTIFICATION OF EXISTING FUEL CONSUMPTION MODELS .. 21
3.2.1 Empirical Models ................................................................................ 22
3.2.2 Mechanistic Models ............................................................................. 23
3.3 HDM 4 FUEL CONSUMPTION MODEL .................................................. 26
3.4 SUMMARY ................................................................................................. 27
CHAPTER 4
CALIBRATION OF THE HDM4 FUEL CONSUMPTION MODEL ....................... 34
4.1 INTRODUCTION ........................................................................................ 34
4.2 TESTING OF THE ACCURACY AND PRECISION OF TEST
EQUIPMENT ............................................................................................... 34
4.2.1 Principles of Engine Control Units ...................................................... 35
4.2.2 Calculating Fuel Efﬁciency ................................................................. 36
4.2.3 Repeatability and Accuracy Testing .................................................... 37
4.2.3.1 Repeatability/Precision .................................................................... 37
4.2.3.2 Data acquisition system accuracy/calibration ................................. 43
4.3 FIELD TRIALS ............................................................................................ 45
4.4 CALIBRATION OF THE HDM 4 MODEL ................................................ 54

vii

4.4.1 Calibration of the HDM 4 Engine Speed Model ................................. 54

4.4.2 Calibration of HDM 4 Fuel Consumption Model ............................... 59
4.5 EFFECT OF ROUGHNESS AND TEXTURE ON FUEL CONSUMPTION
64
4.6 EFFECT OF PAVEMENT TYPE ON FUEL CONSUMPTION ................ 70
4.7 SUMMARY AND CONCLUSION ............................................................. 77
CHAPTER 5
CALIBRATION OF REPAIR AND MAINTENANCE COSTS MODEL ............... 8O
5. 1 INTRODUCTION ........................................................................................ 80
5.2 OVERVIEW OF EXISTING REPAIR AND MAINTENANCE MODELSSO
5 .2. 1 Introduction ......................................................................................... 80
5.2.2 Empirical Models ................................................................................ 81
5.2.3 Mechanistic Models ............................................................................. 85
5.2.4 HDM 4 Repair and Maintenance Costs Models .................................. 86
5.2.5 Summary and Conclusion .................................................................... 87
5.3 CALIBRATION OF THE REPAIR AND MAINTENANCE COSTS
MODEL ........................................................................................................ 88
5.2.1 Data Collection .................................................................................... 88
5.3.1.1 Michigan Department of Transportation (MDOT) data .................. 88
5.3.1.2 Texas Department of Transportation (MDOT) data ........................ 92
5.3.1.3 NCHRP 1-33 data .......................................................................... 100
5.2.2 Assessment of Data Applicability ..................................................... 102
5.2.3 Updating Zaniewski et a1. Tables ...................................................... 105
5.4 SUMMARY AND CONCLUSIONS ......................................................... 109
CHAPTER 6
IDENTIFICATION OF PAVEMENT ROUGHNESS EVENTS ............................ 110
6. 1 INTRODUCTION ...................................................................................... 1 10
6.2 IDENTIFICATION AND CHARACTERIZATION OF PAVEMENT
ROUGHNESS FEATURES ....................................................................... 110
6.2.1 Pavement Faulting ............................................................................. l 10
6.2.2 Pavement Breaks ............................................................................... 113
6.2.3 Curling ............................................................................................... 114
6.2.4 Potholes ............................................................................................. l 14
6.3 EVALUATION OF EXISTING METHODS OF PROFILE ANALYSIS 114
6.3.1 Time domain analysis ........................................................................ 115
6.3.2 Power Spectral Density Method ........................................................ 117
6.3.2.1 PSD Analysis ................................................................................. 118
6.3.2.2 Disadvantages of PSD method ...................................................... 121
6.3.3 Joint time frequency analysis method ............................................... 121
6.3.3.1 Different types of Joint Time Frequency Analysis (JTFA) ........... 122
6.3.3.2 Wavelet transform method ............................................................ 124
6.4 FINALIZATION OF ROUGHNESS IDENTIFICATION METHODS... 125
6.4.1 Filtering out the topography .............................................................. 125
6.4.2 Method for Identifying Joint/Crack Faulting .................................... 129

viii

6.4.2.1 Introduction ................................................................................... 129

6.4.2.2 Discrete Elevation Difference Method .......................................... 143
6.4.3 Method for Identifying Pavement Breaks and Potholes .................... 147
6.4.4 Method for Identifying Slab Curling ................................................. 147

6.4.4. 1 Introduction ................................................................................... 147

6.4.4.2 Discrete Slope Method .................................................................. 155
6.4.5 Summary ............................................................................................ 157
6.4.6 Development of a Window-Based Software System ........................ 157

6.5 FIELD TRIALS .......................................................................................... 157
6.5.1 Criteria for Selecting Pavement Sections .......................................... 157
6.5.2 Field Tests ......................................................................................... 159
6.5.3 Results ............................................................................................... 161

6.6 SUMMARY ............................................................................................... 162

CHAPTER 7 , .

IMPACT OF ROUGHNESS ON VEHICLE DURABILITY ................................... 167
7.1 INTRODUCTION ...................................................................................... 167
7.2 RESEARCH APPROACH ......................................................................... 168

7.2.1 Artiﬁcial Road Proﬁle Generation .................................................... 168
7.2.1.1 Introduction ................................................................................... 168
Category ................................................................................................ 171
Flexible .................................................................................................. 173

7.2.1.2 ISO 8608-1995 .............................................................................. 174
7.2.2 Artiﬁcial Generation of Roughness Features .................................... 176

7.2.2.1 Artiﬁcial Generation of Faults ....................................................... 177

7.2.2.2 Artiﬁcial Generation of Breaks/Bumps ......................................... 179

7.2.2.3 Artiﬁcial Generation of Curling .................................................... 180

7.2.2.4 Artiﬁcial Generation of Potholes ................................................... 182
7.2.3 Dynamic Vehicle Simulation ............................................................ 183
7.2.4 Vehicle Fatigue Damage Analysis .................................................... 187

7.3 RESULTS AND CONCLUSIONS ............................................................ 189
7.3.1 Suspension Failure Threshold ........................................................... 190

7.3.1.1 Suspension Damage in Cars .......................................................... 191

7.3.1.2 Suspension Damage in Trucks ...................................................... 195
7.3.2 Mechanistic versus Empirical Approach ........................................... 197
7.3.3 Case Study 1: Effect of Faulting on Suspension Damage ................. 199

7.3.2.1 Damage to Car Suspensions .......................................................... 200

7.3.2.2 Damage to Truck Suspensions ...................................................... 201
7.3.4 Case Study 2: Effect Of Breaks on Suspension Damage ................... 204

7.3.3.1 Damage to Car Suspensions .......................................................... 204

7.3.3.2 Damage to Truck Suspensions ...................................................... 205
7.3.5 Case Study 3: Effect of Curling on Suspension Damage .................. 205

7.3.4.1 Jointed Reinforced Concrete Pavements ....................................... 206

7.3.4.2 Jointed Plain Concrete Pavements ................................................. 208

7.4 SUMMARY AND CONCLUSIONS ......................................................... 209

 

CHAPTER 8

IMPACT OF ROUGHNESS FEATURES ON DAMAGE TO GOODS ................. 211
8. 1 INTRODUCTION ...................................................................................... 2 1 1
8.2 RESEARCH APPROACH ......................................................................... 212

8.2.1 Artiﬁcial Road Proﬁle and Roughness features Generation ............. 212
8.2.2 Dynamic Vehicle Simulation ............................................................ 212
8.2.2.1 Vehicle Model ............................................................................... 212
8.2.2.2 Discussion ...................................................................................... 215
8.2.3 Product Vibration .............................................................................. 216
8.2.4 Product Damage Analysis ................................................................. 217
8.2.4.1 Electronic and appliances products ............................................... 217
8.2.4.2 Horticultural produce ..................................................................... 220
8.3 RESULTS AND CONCLUSIONS ............................................................ 225
8.3.1 Case Study 1: Effect of Faulting on Damage to goods ..................... 225
8.3.2 Case Study 2: Effect of Breaks on Damage to Goods ....................... 227
8.3.3 Case Study 3: Effect of Curling on Damage to Goods ...................... 229
8.3.4 Case Study 4: Effect of Interaction between Roughness Features
Magnitude, Frequencies and Trip Length on Damage to Goods ....... 230
8.4 SUMMARY AND CONCLUSIONS ......................................................... 231

CHAPTER 9

CONCLUSIONS AND RECOMMENDATION ...................................................... 234
9. 1 INTRODUCTION ...................................................................................... 234
9.2 SUMMARY OF FINDINGS ...................................................................... 234

9.2.1 Fuel Consumption ............................................................................. 235

9.2.2 Effect of Roughness on Repair and Maintenance Costs ................... 237
9.2.3 Effect of Roughness and Roughness Features on Vehicle Durability and

Damage to Goods .............................................................................. 237

9.3 CONCLUSIONS ........................................................................................ 240

9.4 RECOMMENDATIONS ........................................................................... 242

APPENDIX '

TYPICAL CONDITIONS IN THE US ..................................................................... 244

UPDATED REPAIR AND MAINTENANCE COSTS ............................................ 258

ROUGHNESS FEATURES-FIELD TRIALS DATA .............................................. 267

DIAGNOSTIC TOOL FOR ROUGHNESS FEATURES IDENTIFICATION AND

LOCOLIZATION- USER MANUAL ...................................................................... 285

BIBLIOGRAPHY ..................................................................................................... 299

LIST OF TABLES

Table 2.1 Categories of VOC Models (Empirical versus Mechanistic) ...................... 18
Table 2.2 Summary of VOC Models .......................................................................... 20
Table 3.1 HDM 4 Fuel Consumption Model ............................................................... 29
Table 3.2 HDM 4 Traction Forces Model .................................................................. 30
Table 3.3 Final Parameters for the Engine Speed Model [17] .................................... 32
Table 3.4 Final Parameters for Cs Model [17] ............................................................ 32
Table 3.5 Final Parameters for CR2 Model [17] ......................................................... 32
Table 3.6 Final Parameters for Effective Mass Ratio Model [17] .............................. 33
Table 4.1 Flint Loop Repeatability Test for Run 1, Run 2 and Run 3 ........................ 40
Table 4.2 Flint loop repeatability test for run4 and run5 ............................................. 41
Table 4.3 I-69 repeatability test between run 1 and run 2 ........................................... 42
Table 4.4 Summary of ﬁlCl consumption data using AutoTap and Graphtec ............. 45
Table 4.5 Data Collection for Engine Parameters, Environmental Factors and
Pavement Surface Characteristics during ﬁeld trials ................................. 46
Table 4.6 Field test matrix ........................................................................................... 47
Table 4.7 Recorded weather conditions ...................................................................... 47
Table 4.8 Characteristics of the Vehicles Used in the Field Trials ............................. 51
Table 4.9 Vehicle Classiﬁcation Used in the Engine Speed Model Calibration ......... 55

Table 4.10 New Coefﬁcients for the Engine Speed Model By Vehicle Class for Wet
and Dry Conditions .................................................................................... 58

Table 4.11 Recommended Coefﬁcients for the Engine Speed Model by Vehicle Class
................................................................................................................... 59

xi

 

Table 4.12 Summary of the Model Performance ........................................................ 63

Table 4.13 Lack of Fit Tests ........................................................................................ 65
Table 4.14 Tests of Between-Subjects Effects at 89 km/h .......................................... 66
Table 4.15 Tests of Between-Subjects Effects at 56 km/h .......................................... 66
Table 4.16 Estimated Marginal Means — Articulated Truck ...................................... 72
Table 4.17 Estimated Marginal Means - Light Truck ............................................... 73
Table 4.18 Pairwise Comparisons — Articulated Truck at 56 km/h .......................... 73
Table 4.19 Pairwise Comparisons — Articulated Truck at 72 km/h ........................... 74
Table 4.20 Pairwise Comparisons — Articulated Truck at 88 km/h ........................... 74
Table 4.21 Pairwise Comparisons - Light Truck at 56 km/h ..................................... 74
Table 4.22 Pairwise Comparisons — Light truck at 72 km/h ...................................... 75
Table 4.23 Pairwise Comparisons — Light Truck at 88 km/h ..................................... 75
Table 4.24 Summary of Pairwise Comparison for all vehicles .................................. 76
Table 5.1 HDM 3 Maintenance Model Parameters [17] ........................................... :. 83
Table 5.2 HDM 4 Maintenance Model Parameters [l7] ............................................. 88
Table 5.3 Repair and maintenance costs and inﬂation rates ..................................... 106
Table 5.4 Change in repair and maintenance costs as a function of [RI ................... 108

Table 6.1 Number of Pavement Sections' for Veriﬁcation of Roughness Diagnosis

Tool .......................................................................................................... 158
Table 6.2 Number of Deﬁnition of Severity Level .................................................. 159
Table 6.3 Summary of measured faults in section 1 ................................................ 161
Table 6.4 Summary of measured faults in section 2 through 4 ................................ 161
Table 6.5 Summary of observed breaks from ﬁeld trials and predicted breaks using

the algorithm ............................................................................................ 165
Table 7. 1 Roughness Parameters for the White-Noise PSD Model [46] ............... 170

xii

Table 7. 2 Ranges of Variable Values for PSD Equations [48] ............................... 171

Table 7. 3 PSD Coefﬁcients in the Roughness Model [46] ..................................... 173
Table 7.4 Roughness Content of Real Proﬁles ......................................................... 194
Table 8.1 Parameter values for a “standard vehicle” [77] ......................................... 214
Table 8.2 Summary of material properties [80] ........................................................ 219

Table 8.3 Summary of truck and packaging parameter values for horticultural produce
................................................................................................................. 224

Table B.l Updated Repair and Maintenance Costs ($/ 1000 km) — Small Car .......... 259
Table B.2 Updated Repair and Maintenance Costs ($/1000 km) — Medium Car ...... 260
Table B.3 Updated Repair and Maintenance Costs ($/1000 km) — Large Car .......... 261
Table B.4 Updated Repair and Maintenance Costs ($/ 1000 km) — Pick up .............. 261
Table B.5 Updated Repair and Maintenance Costs ($/1000 km) -— Light Truck ....... 263
Table B.6 Updated Repair and Maintenance Costs ($/1000 km) — Medium Truck.. 264
Table B.7 Updated Repair and Maintenance Costs ($/1000 km) — Heavy Truck ..... 265

Table 3.8 Updated Repair and Maintenance Costs ($/1000 km) — Articulated Truck

................................................................................................................. 266
Table C.1 Distress Data Collection for Site 1 ........................................................... 267
Table C.2 Distress Data Collection for Site 2 .......................................................... 270
Table C.3 Distress Data Collection for Site 3 .......................................................... 273
Table C.4 Distress Data Collection for Site 4 .......................................................... 277
Table C.5 Repeated Measurements for Site 2 .......................................................... 282
Table C.6 First Repeated Measurements for Site 3 .................................................. 273
Table C.7 Second Repeated Measurements for Site 3 ............................................. 284
Table D.1 SUmmary of Records in an ERD File Header .......................................... 289

xiii

LIST OF FIGURES

Figure 2.1 Components of Road User Costs [17] ....................................................... 13
Figure 2.2 Ranges in Terms of Texture Wavelength ................................................. 13
Figure 2.3 Summary of Relative VOC Costs for Trucks (after [2]) ........................... 14
Figure 2.4 World Bank VOC Models Development [13] ......................................... 16
Figure 2.5 United States VOC Models Development [13] ........................................ 17
Figure 4.1 Fuel consumption data acquisition system (a) Different parts (b) OBD II
connection ..................................................................................................... 35
F igtire 4.2 Map View of Flint Loop ........................................................................... 38
Figure 4.3 Roughness and Texture Depth versus Distance (Flint loop) ..................... 39
Figure 4.4 Grade versus Distance (Flint loop) ............................................................ 39
Figure 4.5 Fuel Consumption versus Distance for runl through run3 ....................... 40
Figure 4.6 Fuel Consumption versus Distance for run 4 and run5 ............................. 41
Figure 4.7 Grade and roughness versus Distance (I69E near Lansing) ...................... 42
Figure 4.8 Mass Air Flow Rate versus Distance (1—69 B near Lansing) ..................... 43

Figure 4.9 Instrumentation used by University of Texas at Arlington research group
(a) Fuel Meter (b) RTD Temperature Probe (c) Data Acquisition System .. 44

Figure 4.10 Experiment setting used by University of Texas at Arlington ............... 45

Figure 4.11 MDOT Test Vehicles (a) Rapid Travel Proﬁlometer (b) Pavement
Friction Tester ........................................................................................... 48

Figure 4.12 Different Vehicle Used During Field Trials (a) Medium Car (b) SUV (c)

Van ((1) Light Truck (e) Articulated Truck ................................................ 50
Figure 4.13 (a) Loading of Light Truck (b) Loaded Light Truck (c) Loading of Heavy
Truck ((1) Loaded Heavy Truck .................................................................... 52

Figure 4.14 Examples of Collected Data — All Sections ........................................... 53

xiv

Figure 4.15 Comparison Between the Original HDM 4 Model And HDM 4 With the

New Engine Speed Model ......................................................................... 54

Figure 4.16 Calibration of the HDM 4 engine speed model —— Van and Medium car
...................................................................................................................... 56
Figure 4.17 HDM 4 engine speed model Calibration - SUV and Light Truck .......... 57
Figure 4.18 HDM 4 engine speed model Calibration - Heavy Truck ........................ 58
Figure 4.19 Predicted and Measured Fuel Consumption versus Distance ................. 61

Figure 4.20 Observed Versus Estimated Fuel Consumption (a) With Cruise Control

and (b) Without Cruise Control .................................................................... 62
Figure 4.21 Observed Fuel Consumption versus Estimated Using HDM 4 Model — All
Vehicles ........................................................................................................ 63
Figure 4.22 Energy Distribution in a Passenger Car versus Speed as a Percentage of
the Available Power Output 'at the Engine [after 116] ................................. 67
Figure 4.23 Effect of Surface Texture on Fuel Consumption ..................................... 68
Figure 4.24 Effect of Roughness on Fuel Consumption ............................................. 69
Figure 4.25 Mean and Standard Deviation of Fuel Consumption for Different
Pavement Type and Speed— Articulated Truck ............................................ 71
Figure 4.26 Mean and Standard Deviation of Fuel Consumption for Different
Pavement Type and Speed — Light Truck .................................................... 72
Figure 4.27 Effect of Speed on Asphalt Pavements [After 72] .................................. 74
Figure 5.1 Michigan Regions ...................................................................................... 89
Figure 5.2 Distribution of roughness for Michigan regions (Michigan DOT) ............ 90
Figure 5.3 Percent of sections with IR] > 2.4 mfkrn ................................................... 91

Figure 5.4 Distribution of repairs by district and vehicle class (Michigan DOT) ...... 92

Figure 5.5 Summary of the repair and maintenance data (Michigan DOT) ................ 93
Figure 5.6 Texas districts ............................................................................................ 94
Figure 5.7 Distribution of IRI by district for Texas (Texas DOT) .............................. 95

XV

 

 

 

n.—-
--

‘&._.1

4“

“king: ti {Eta} ’

Figure 5.8 Distribution of repairs by district for Passenger cars (Texas DOT) .......... 95

Figure 5.9 Distribution of repairs by district for (a) Light and (b) Medium trucks

(Texas DOT) ................................................................................................. 96
Figure 5.10 Distribution of repairs by district for (a) Heavy and (b) Articulated trucks
(Texas DOT) ................................................................................................. 97
Figure 5.11 Summary of the repair and maintenance data for light and medium
vehicles (Texas DOT) ................................................................................... 98
Figure 5.12 Summary of the repair and maintenance data for heavy vehicles (Texas
DOT) ............................................................................................................. 99
Figure 5.13 NCHRP‘l-33 Raw data [23] .................................................................. 100
Figure 5.14 NCHRP 1-33 data corrected for age and mileage [23] .......................... 101
Figure 5.15 Categories of roughness considered in the statistical analysis (Texas
districts) ...................................................................................................... 102
Figure 5.16 Comparison between HDM-4 predictions and data from truck ﬂeets
(NCHRP 1—33) ............................................................................................ 104
Figure 5.17 Approach for updating Zaniewski et a1. tables ...................................... 108
Figure 6.1 Theoretical samples from faulting of slabs .............................................. 111
Figure 6.2 Faulting from actual proﬁle .................................................................... 111
Figure 6.3 Components of the created dummy proﬁle ............................................. 112
Figure 6.4 Dummy proﬁle ......... , .............................................................................. 112
Figure 6.5 Faulting of dummy proﬁle ...................................................................... 113
Figure 6.6 Faulting detected by Pathway’s procedure ............................................. 117
Figure 6.7 Relationship between Wigner (Wx) and Cohen’s class distribution (A)
[37] .............................................................................................................. 124
Figure 6.8 (a) Butterworth ﬁlter bode plot (b) Transfer functions of the Butterworth
ﬁlter for various orders ............................................................................... 127
Figure 6.9 Cascade ﬁltering diagram ....................................................................... 128
Figure 6.10 Daubechies 3 wavelet analysis .............................................................. 131

xvi

Figure 6.11 Proﬁle and slope .................................................................................... 133

Figure 6.12 Proﬁle and curvature ............................................................................. 133
Figure 6.13 Proﬁle and discrete elevation difference ............................................... 134
Figure 6.14 Discontinuities detected by slope and adaptive ﬁltering ...................... 134
Figure 6.15 Discontinuities detected by curvature and adaptive ﬁltering ............... 135

Figure 6.16 Discontinuities detected by discrete elevation difference and adaptive

ﬁltering ....................................................................................................... 135
Figure 6.17 Detected faulting from the discrete elevation difference method ........ 137
Figure 6.18 Detected faulting from the slope method .............................................. 137
Figure 6.19 Detected faulting from the Haar wavelet method ................................. 138
Figure 6.20 Detected faulting from the db2 wavelet method ................................... 138
Figure 6.21 Detected faulting from the coifl wavelet method ................................. 139
Figure 6.22 Detected faulting from the sym2 wavelet method ................................ 139
Figure 6.23 Detected faulting from the discrete elevation difference method &

adaptive ﬁltering ......................................................................................... 140
Figure 6.24 Detected faulting from the slope method & adaptive ﬁltering ............. 140

Figure 6.25 Detected faulting from the Haar wavelet method & adaptive ﬁltering. 141
Figure 6.26 Detected faulting from the db2 wavelet method & adaptive ﬁltering .. 141
Figure 6.27 Detected faulting from the coifl wavelet method & adaptive ﬁltering 142

Figure 6.28 Detected faulting from the sym2 wavelet method & adaptive ﬁltering 142

Figure 6.29 Faulting in the raw proﬁle ..................................................................... 144
Figure 6.30 Effect of the moving average ﬁlter on surface discontinuity ................ 144
Figure 6.31 Different scenarios of recording a fault after ﬁltering ......................... 145
Figure 6.32 Fault detection algorithm .................................................................... 146
Figure 6.33 Illustration of a pavement break ............................................................ 147

xvii

Figure 6.34 Break detection algorithm .................................................................... 148

Figure 6.35 Original and ﬁltered proﬁles ................................................................ 149
Figure 6.36 Curling using GBPF method ................................................................. 150
Figure 6.37 Created curling and extracted curling using wavelets .......................... 151

Figure 6.38 Wigner-Ville joint time frequency distribution of the dummy proﬁle . 152

Figure 6.39 Pseudo Wigner-Ville joint time frequency distribution of the dummy

proﬁle .......................................................................................................... 152
Figure 6.40 Smoothed Pseudo Wigner-Ville joint time frequency distribution of the

dummy proﬁle ............................................................................................ 153
Figure 6.41 Curling extracted with DS method ....................................................... 154
Figure 6.42 Detection of local maxima of the slope function ................................ 155
Figure 6.43 Curling detection algorithm ................................................................. 156
Figure 6.44 Correlation analyses for Site 1 (a) magnitude (b) location .................. 163
Figure 6.45 Correlation analyses for Site 2 (a) magnitude (b) location ................. 164
Figure 6.46 Correlation analyses for Site 3 (a) magnitude (b) location .................. 165
Figure 6.47 Raw proﬁle for Site 1 .......................................................................... 166
Figure 6.48 Filtered proﬁle and predicted curling magnitude ................................ 166
Figure 7.1 Schematic description of roughness features .......................................... 177
Figure 7.2 Proﬁle of a faulted rigid pavement ......................................................... 178
Figure 7.3 Proﬁle of a concrete slab with one break ................................................ 180
Figure 7.4 Proﬁle of a concrete slab with one break ................................................ 181
Figure 7.5 Proﬁle of asphalt concrete pavement with one pothole .......................... 183
Figure 7.6 Schematic of two degrees of freedom quarter-car vehicle model ........... 184

Figure 7.7 Schematic of Two Degrees of Freedom Quarter-Car Vehicle Model 186

Figure 7.8 Schematic Deﬁnition of the Rainﬂow Cycle as given by [63] .............. 188

xviii

 

Figure 7.9 Road surface roughness distribution in the United States ...................... 191

Figure 7.10 Effect of road surface roughness on car suspension ............................ 192
Figure 7.11 Accumulated Car Suspension Damage Using Real and Artiﬁcial Proﬁles
.................................................................................................................... 195
Figure 7.12 Effect of road surface roughness on truck suspension ......................... 196
Figure 7.13 Accumulated Truck Suspension Damage Using Real and Artiﬁcial
Proﬁles ........................................................................................................ 197
Figure 7.14 Comparison between Car Repair and Maintenance Costs Calculated
Using M-E Approach and the Updated Zaniewski Costs ........................... 198
Figure 7.15 Comparison between Truck Repair and Maintenance Costs Calculated
Using M-E Approach and the Updated Zaniewski Costs ........................... 199
Figure 7.16 Car R&M costs induced by different levels of faulting ....................... 201
Figure 7.17 Truck repair and maintenance costs induced by different levels of
faulting — JRCP ........................................................................................... 202
Figure 7.18 Truck repair and maintenance costs induced by different levels of
faulting — JPCP ........................................................................................... 203
Figure 7.19 Car R&M costs induced by different levels of breaks ......................... 204

Figure 7.20 Truck repair and maintenance costs induced by different levels of breaks
.................................................................................................................... 205

Figure 7.21 Car repair and maintenance costs induced by different levels of curling —-

JRCP ............................................................................................. 206
Figure 7.22 Truck repair and maintenance costs induced by different levels of curling
—— JRCP ........................................................................................................ 207
Figure 7.23 Car repair and maintenance costs induced by different levels of curling —
JPCP ............................................................................................................ 208
Figure 7.24 Truck repair and maintenance costs induced by different levels of curling
—- JPCP ......................................................................................................... 209
Figure 8.1 Typical mathematical model of a half truck ........................................... 213
Figure 8.2 High acceleration level Location in the truck [48] .................................. 215

xix

Figure 8.3 Electronic product with Fragile Component, Modeled as Spring/Mass

system ......................................................................................................... 216
Figure 8.4 Collision of horticultural produce treated as inelastic Shocks ................ 216
Figure 8.5 Damage Boundary Curve [79] ................................................................ 217

Figure 8.6 Road-vehicle-load interaction for multi-layered energy absorbing package5222

Figure 8.7 Effect of vibration frequency on the bruise depth of apples at constant peak

acceleration of 1.4 g .................................................................................... 223
Figure 8.8 Damage Induced by Different Levels and Counts of Faulting ............... 226
Figure 8.9 Interaction Effect between Speed and Fault counts on Damage to Apples

.................................................................................................................... 228
Figure 8.10 Damage Induced by Different Levels and Counts of Breaks ................ 229
Figure 8.11 Damage Induced by Different Levels and Counts of Curling .............. 230

Figure 8.12 Damage Induced by Different Fault Magnitude and Trip Length ........ 232

Figure 8.13 Damage Induced by Different Break Magnitude and Trip Length ....... 233

Figure A.1 Latest Pavement Type Frequency Distribution (source: HPMS database)
.................................................................................................................... 246

Figure A.2 Latest Pavement Roughness Frequency Distribution (source: HPMS
database) ..................................................................................................... 247

Figure A.3 Environment Conditions in the US for 2007 (source: National Climatic
Data Center) ................................................................................................ 248

Figure A.4 Latest Aerodynamic Parameters in the US (sources: EPA and CarTest
software) ..................................................................................................... 249

Figure A.5 Vehicle Weight Statistics in the US Grouped By EPA Vehicle
Classiﬁcation (source: FHWA) ....................................................... 251

Figure A.6 Latest Fuel Economy And Efﬁciency Distribution In The US For
Passenger Cars And Trucks (source: EPA report, 2007) ........................... 252

Figure A.7 Relationship between engine efﬁciency and rated power (source: EPA
report, 2007) ............................................................................................... 254

 

CHAPTER 1
INTRODUCTION

1.1 MOTIVATION

Understanding the costs of highway construction, highway maintenance and vehicle
operation is essential to sound planning and management of highway investments,
especially under increasing infrastructure demands and limited budget resources.
While the infrastructure costs paid by road agencies are substantial, the cost borne by
road users are even greater. In 2009, the American Automobile Association (AAA) [1]
reported an average vehicle operating cost of $4.993 per vehicle mile based on 2008
prices. For conventional vehicles, these costs are related to fuel and oil consumption,
tire wear, repair and maintenance. These costs depend on the vehicle class and are
inﬂuenced by vehicle technology, pavement-surface type, pavement condition,
roadway geometrics, environment, speed of operation, and other factors. Therefore,
vehicle operating costs are part of the costs that highway agencies must consider when

evaluating pavement-investment strategies.

A large body of research is available on the effects of pavement condition on vehicle
operating costs and on models used to estimate these effects. Much of this information
and many of the models were developed on the basis of data generated some 30 years
ago in other countries for vehicle ﬂeets that vary substantially from those used
currently in the United States and for roadways that differ from those built in the

United States. Therefore, there is a need to collect new information that could help in

 

reﬁning these models or developing new models that would better apply to US
conditions. These models would be used to estimate the effects of pavement condition

on vehicle operating costs, in order to conduct a rational economic analysis.

Reduction in vehicle fuel consumption is one of the main beneﬁts considered in
technical and economic evaluations of road improvements considering its signiﬁcance.
In fact, the 255 million vehicles in the United States consume about 200 billion
gallons of motor fuel annually. With today’s gas prices, this will translate to about 600
billion dollars. Finding ways to reduce this energy consumption is a national goal for
reasons ranging from ensuring economic and national security to improving local air

quality and reducing greenhouse gas emissions.

Fuel consumption is the most signiﬁcant component of the total vehicle operating
costs (VOC) followed by transport cost including repair and maintenance costs and
damage to goods [2]. The American Association of State Highway Ofﬁcials
(AASHTO) reported that poor road conditions added an estimated $76.8 billion to
transport costs annually [3]. These costs were mainly caused by the interaction
between vehicles and pavements. The excitation of road vehicle can be described by
many factors, such as road proﬁle, road curvature, topology, and driver behavior,
including speed changes and maneuvering. Of these, the most important parameter for
the durability of most components is the road proﬁle. The surface proﬁle of the road
transmits the vibrations through the tires and suspension system to the body of the

vehicle and then to the driver, passengers and cargo. Vehicle manufacturers place a

 

 

major focus on constantly improving the design of these different vehicle components
to respond better to changes in road surface proﬁles. Despite this, changes in the road
surface proﬁle still directly affect the user costs including repair and maintenance
costs and damage to goods.

Road roughness has long been used as one of the primary indicators of pavement
condition. It is today a common practice in pavement engineering to measure
longitudinal pavement roughness and compute a suitable roughness index as an
estimate of pavement serviceability. With the development of high-speed
proﬁlometers, large database of road surface proﬁle has been obtained. Road
roughness is normally characterized by a summary index that applies over a length of
road. Summary index measures are obtained directly by measuring the longitudinal
proﬁle and then applying a mathematical analysis to reduce the proﬁle to the
roughness statistic. All of these summary indices can only give an average condition
for a relatively long section of pavement. However, they do not retain the actual
contents of pavement surface roughness. Such detailed roughness content information
may be useful for maintenance operations, diagnosis of surface roughness as a defect,

and detailed analysis of the trend of pavement performance deterioration.

Sate Highway Agencies extract roughness features (type, extent, and severity) from
video images of the pavement surface periodically. Surface distresses and proﬁle data
of all pavement types are used to automatically compute some form of condition index
and roughness index. The bi-annual change of the pavement’s condition is included in

a performance model to estimate the pavement’s Remaining Service Life (RSL).

 

However, it cannot provide useful information about some distress features such as the
magnitude of faulting, breaks and curling in concrete (PCC) pavements and potholes
in asphalt (AC) pavements. Failure to include speciﬁc roughness features in Pavement
Management Systems (PMS) has the following negative impacts on system
performance:

0 The analysis may overestimate the Remaining Service Life (RSL) of the
pavement.

0 The system process may not ultimately select the most cost-effective ﬁx to
extend pavement life.

0 Missing information on speciﬁc roughness features reduces the reliability of
recommending the most cost-effective strategy for preserving the pavement
network.

0 The analysis may underestimate the effect of localized roughness events on
vehicle durability and damage to goods.

A signiﬁcant improvement to the ﬁeld would be the implementation of tools to

extract speciﬁc roughness features, through the use of the raw proﬁle data.

The developed tools should detect, locate and identify the level of surface
irregularities; however, they do not in themselves provide guidance on acceptable
roughness levels to limit user costs. Therefore, there is a need to develop a
methodology to determine such roughness threshold. If this threshold value exists, it
could be a useful preventive maintenance (PM) tool, whereby a PM action, such as

smoothing the pavement surface, is taken to reduce the user cost.

 

1.2 RESEARCH HYPOTHESIS AND OBJECTIVES

The goal of this research is to recommend models for estimating the effects of
pavement condition on vehicle operating costs including fuel consumption, repair and
maintenance costs and damage to goods. The models shall reﬂect current vehicle
technologies in the United States. Accomplishment of these objectives was possible

through conducting the tasks discussed under research scope.

The fuel consumption of a vehicle is proportional to the forces acting on the
vehicle. These forces are rolling resistance, gradient, inertial, curvature, and
aerodynamic forces. The rolling resistance forces are functions of pavement
conditions, tire parameters and vehicle characteristics. The rolling resistance has a
signiﬁcant effect on ﬁrel consumption. Therefore, one way to decrease rolling
resistance is improve pavement conditions. In fact, it was reported that pavement
roughness has signiﬁcant effect on rolling resistance forces. A decrease in pavement
roughness by 2 m/km will result in a 10 percent decrease in rolling resistance [4]. A
ten percent reduction in average rolling resistance promises a l to 2 percent decrease
in ﬁrel consumption. This would save about 3 to 6 billion gallons of ﬁre] per year. In
this context, a l to 2 percent reduction in the fuel consumed would be a meaningful
accomplishment.

Regarding repair and maintenance costs and damage to goods, road roughness
induces dynamic loads in traveling vehicles and these dynamic loads cause fatigue
damage to the vehicle components. Also potential damage to the products could be

caused by severe Vibrations (and shocks) during transportation. Therefore, road

 

roughness in general, and roughness “events caused by certain pavement distresses play
a signiﬁcant role in increasing operating costs and further damaging the pavement,
traveling vehicles and transported goods.

This interaction between the vehicle and road surface can explain the change in fuel
consumption due to roughness, damage to vehicle components and to transported
goods due to localized roughness events associated with certain pavement distresses.
Modeling this interaction can provide a tool to warn a given highway agency about
potential increase in VOC due to roughness under the following hypothesis:

0 Pavement roughness has an effect on vehicle operating costs including fuel
consumption, repair and maintenance costs and damage to goods.

0 The ampliﬁcation in the load magnitude due to the transient events of the road
proﬁles can lead to a tangible damage to vehicle suspensions and transported
goods.

Also, it is well understood that a half or quarter car model cannot be expected to
predict loads on a physical vehicle exactly, but it will highlight the most important

road characteristics as far as fatigue damage accumulation is concerned [5, 6, 7].

The objectives of this research are to:

1. Determine the effect of roughness on VOC including fuel consumption, repair
and maintenance costs and damage to goods.

2. Develop a proﬁle-based diagnosis tool for identifying localized roughness
features that may have an impact on vehicle durability and damage to goods:

3. Provide guidance on acceptable roughness levels to limit vehicle operating
costs and damage to goods.

1.3 RESEARCH SCOPE
The work to be performed in this research was accomplished in two phases through

the execution of the six general tasks outlined below:

Part I
Task 1 .' Determine the effect of pavement roughness on fuel consumption

Subtask 1.]: Identiﬁcation and evaluation of current fuel consumption models.
Identify the models currently available for estimating the effects of pavement

condition on fuel consumption.
Subtask 1.2: Field Trials

Subtask 1.3: Calibration and validation of the candidate ﬁrel consumption

model
Task 2: Determine the effect of pavement roughness on repair and maintenance costs

Subtask 2.]: Identiﬁcation and evaluation of current repair and maintenance
models. Identify the models currently available for estimating the effects of

pavement condition on repair and maintenance costs.

Subtask 2.2: Calibration and validation of the candidate repair and

maintenance costs

Part H

Task 3: Development and implementation of localized roughness features detection

methods

Subtask 3.1: Identiﬁcation and characterization of localized roughness
features. Surface features that can be detected from sensor (proﬁle) data will
be identiﬁed and characterized, based on the proﬁle characteristics. Surface
features to be included are: (l) Faulting at Crack/Joint; (2) breaks or tilting of
ﬁill-depth patch or pavement slab; (3) curling of FCC pavement slabs; and
(4) localized bumps and depressions (e.g., potholes and punch-downs) in

asphalt pavements.

Subtask 3.2: Evaluation of existing methods of proﬁle analysis: Available
methods for identifying localized roughness features from a surface proﬁle
will be reviewed. For each available roughness identiﬁcation method, the

following information will be provided whenever applicable:
(1) Theoretical basis for the process
(2) Past experiences - both successes and problems

Known available roughness identiﬁcation methods to be reviewed are as

follows:

(1) A “Height Variance Filter” method used by Pathway to identify
a faulting crack and measure its faulting severity ,

(2) Power spectral density method

(3) Joint time-frequency method

Subtask 3.3: Finalization of roughness identiﬁcation methods. A
comprehensive analysis will be performed to decide on the best identiﬁcation
method for each proposed surface roughness features, which will be clearly
justiﬁed. This sub-task will compile the relevant information to accomplish

the ﬁrst objective.

Subtask 3.4: Development of a window-based software system. For objective
2, a user-friendly, window-based computer software system for detecting and

quantifying roughness features (faulting, etc) will be developed.

Subtask 3.5: Field trials. First the criteria for selecting pavement sections will
be established. These criteria will then be used to select pavement sections
for ﬁeld veriﬁcation. The raw proﬁle data of the selected pavement sections
will then be extracted from PMS ﬁles. Field surveys will be performed to
collect distress data. The above information will be used to validate the

roughness identiﬁcation method.
Task 4 : Numerical modeling of vehicle response and products Vibration

A mechanistic model to simulate vehicle and products vibration responses

will be developed.
Task 5: Vehicle damage analysis and product fragility assessment

A mechanistic-empirical approach to conduct fatigue damage analysis and
product fragility assessment will be followed. A relationship between
dynamic loading/product vibrations and damage to vehicle and goods will be

developed.

 

 

The key element of this analysis is the critical amplitude and frequency to

induce damage.

Task 6: Quantiﬁcation of the effect of roughness features on vehicle and product

damage.

A sensitivity analysis will be performed to quantify the relationship between

height and width of roughness features, and vehicle and product damage.

1.4 DOCUMENT ORGANIZATION

The thesis is divided into nine chapters including the introduction.

In chapter 2, the literature review on current practices, and other information
relevant to estimating the effects of pavement condition on vehicle operating costs is
presented.

In chapter 3, the models currently available for estimating the effects of pavement
condition on ﬁrel consumption are identiﬁed and evaluated.

In chapter 4, the calibration and validation of the candidate models are discussed.

In chapter 5, the models for estimating the effects of pavement condition on repair
and maintenance costs are evaluated and discussed.

In chapter 6, the methods of identifying speciﬁc pavement roughness features

through the use of proﬁle data are investigated.

10

In chapters 7 and 8, a detailed approach to estimate the effect of roughness features
on vehicle durability and damage to goods, respectively, and to provide guidance on
acceptable roughness levels are discussed. Case studies are also provided.

The ﬁndings and the impact of this research on the state-of-art are Summarized in

chapter 9.

ll

CHAPTER 2
BACKGROUND AND LITERATURE REVIEW

2.1 INTRODUCTION

In this chapter the ﬁndings from the review of the published literature that deal with
VOC costs are summarized. The review included FHWA reports, NCHRP reports as
well as other research reports and publications.

The review was focused on research that has identiﬁed factors affecting VOC costs
including pavement conditions. The most relevant reports were found to be those that

include relationship between pavement conditions and VOC costs.

2.2 BACKGROUND

The cost of transportation is often referred to as road user costs. Figure 2.1 shows
the components of road user costs which essentially comprised of vehicle operating
costs, travel time delay, safety, comfort and convenience, and environmental impacts.

Vehicle operating costs are the costs associated with owning, operating, and
maintaining a vehicle including: fuel consumption, oil and lubrication; tire wear,
repair and maintenance, depreciation, and license and insurance. Vehicle operating
cost components modeled include fuel and oil consumption, repair and maintenance
costs, tire wear as well as vehicle depreciation. Each of these cost components are
typically modeled separately and summed to obtain overall vehicle operating costs.
Empirical or mechanistic-empirical relationships are used to address inﬂuence factors.

Common to many of these relationships is a road roughness factor used to describe the

12

condition Of the road. One such roughness measure is international roughness index

(IRI) developed as part of the World Bank HDM standards studies [68].

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Road User Costs I
i ,
r ------------- " ‘ _
l Vehicle Operating I Travel Time Safety & Comfort and Envrronment
I Costs ' Delay Accidents Convenience Impacts
l H I
Fuel epair I oa . . Ir
: Consumption Maintenance | Construction Fatalities Road Dust Pollution
: I L g I 1 RI d N i.
- - Road . . oa orse
I Trre Wear Capitol Cost | Maintenance Injuries Roughness Pollution
I I
. L _ I . I I .
| Oil License & | Road Property Trafﬁc 8011
I Consumption Insurance I Roughness Damage Noise Pollution

 

 

 

 

 

[— ____________ I

 

 

 

 

 

 

 

 

 

Figure 2.1 Components of Road User Costs [17]

 

 

 

Road roughness is a broad term describing a range of irregularities from surface

texture through road unevenness. To better characterize the inﬂuence of road

roughness on vehicle operating costs, the total texture Spectrum was subdivided into

four categories. Figure 2.2 deﬁnes these categories.

mm m

 

 

 

 

 

 

 

 

 

 

 

_/ ¥ A
r 0.001 0.01 0.1 1 IO\/ 0.1 I 10 D
I l I I I l I I I I I I I I I
I I T I I I I I I I I I I I I I
0.005 0.05 0.5 5 50 0.5 5 50
. Mega- ’ 7
ISO 1347-1:1997 I Micro-texture I Macro-texture I I Roughness I
texture
7 . Mega-
PIARC (1987) Micro-texture Macro-texture Roughness
texture
Fine Coarse Me a-77
Laser RST macro- macro- g Roughness
texture
texture texture

 

 

 

 

Figure 2.2 Ranges in Terms of Texture Wavelength

l3

 

This characterization of roughness allowed a better evaluation of the surface factors
inﬂuencing fuel consumption. As with fuel consumption, road roughness impacts
maintenance and repair costs, tire replacement and the market value of vehicles. In [2],
Barnes et a1. summarized vehicle operating costs from various data sources including
technical reports and trucking literature. Some of their ﬁndings are shown in Figure
2.3, which indicates the relative cost of fuel, tire replacement and maintenance/repair
expressed as a percentage of total operating expenses, including driver and other non-

marginal costs.

 

     
     
   
      
    

 

 

 

 

 

 

18
’2: e. -
é 16 I ﬁgs...-
Jli, h
o ,, M '
2 12 9‘4?”
. ‘,
43 10 5.5.
I" 8 a,” .
9 3515' t
O) 6 ' i" *9."
.2 .1
E 4 E‘s,
d.) j # . .
a: 2 214-: .
0 V O A

fuel Maint./Repair Tires

Cost Component

Figure 2.3 Summary of Relative VOC Costs for Trucks (after [2])

Clearly fuel consumption is the primary cost component followed by maintenance
and repair costs, and tire wear. Therefore, the research will be focused on fuel

consumption and repair and maintenance costs.

2.3 OVERVIEW OF EXISTING VOC MODELS
2.3.] Introduction
Based on the literature review, a number of major models that have been developed
in various Countries were identiﬁed. The most relevant models include:
o The World Bank’s HDM-III and IV VOC module;
0 Texas Research and Development Foundation (TRDF) VOC model;
0 MicroBENCOST VOC module;
0 Saskatchewan VOC models [69];
0 British COBA VOC module;
0 Swedish VETO model;
0 Australian NIMPAC VOC module;
0 ARF COM model of fuel consumption;
0 New Zealand NZVOC; and

0 South African VOC models

Most of the present VOC models have beneﬁted from the World Bank’s HDM
research to some extent. Figures 2.4 and 2.5 outline the chronological development of
these models. As shown in Figure 2.4, the basis of HDM research dates back to the
seminal study by Weille [8] for the World Bank, which lead to the development of the

Highway Cost Model and subsequently to the most recent HDM IV module.

15

 

 

Kenya, India &
III Caribbean I
1971-1986

HighwayCostModel TRRL&CRRI 3 ................... E
--+ 1971 + HDM-III w VOC =
MIT,TRRL&LCPC I VOC1987 3
BraziIStudy I : ................... :

1+ 1975-1984 ~
TRDF&TRRL

 

 

 

 

De Weille
1966

 

 

 

 

 

 

 

 

 

 

     

 

TRDF VOC

E Background Work 1980-82

ccccccccc

Major VOC Model

VETO
NITRR

NZVOC
PMlS
C B-Roads

D Other VOC Model

 

 

 

 

Figure 2.4 World Bank VOC Models Development [13]

Figure 2.5 highlights the VOC research conducted in the United States, which was
primarily initiated by Winfrey [9] followed by Claffey [10]. These initial efforts laid
to the foundation for an assembly of VOC data and estimation models in the American
Association of State Highway and Transportation Ofﬁcial (AASHTO) Red Book by
1978 [70]. In 1982 new VOC models were developed by the Texas Research and
Development Foundation (TRDF) [11].

The TRDF model was representative of current vehicle technology at that time. The
model also considered the Brazil HDM study results, particularly in the effect of
pavement roughness on VOC. The TRDF models were also incorporated into the
MicroBENCOST model which was intended to be a modern replacement for the
AASHTO Red Book. It should be noted that IRI was not an accepted roughness index

at that time.

16

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

   

 

Wrnﬁ'ey, Intermedrate US Data on France Price
Claffey 3’02" SW)’ 1970' v hicle I d ' 1976
1968-1971 1975-1980 S e “ ”mg
I I I I I
E Red Book 3 MicroBENCOST
AASHTO , e: TRDF i195 MOdel § e: VOC E
1978 1991-1992
I
Canada:HUB United States: FHW A
Alberta HIAP State DOT
:1; 1:: Counties
Munici alities
[:I Background Work State DOT P
Major VOC Model
C Other VOC Model

 

 

Figure 2.5 United States VOC Models Development [13]

More recently, a user-friendly model for personal computers, “Vehicle/Highway
Performance Predictor” (HPP), was developed for highway designers, planners, and
strategists to estimate fuel consumption and exhaust emissions related to modes of
vehicle operations on highways of various conﬁgurations and traffic controls [12].
This model simulates operations of vehicles by evaluating the vehicle external loads or
propulsive demands determined by longitudinal and lateral accelerations, positive and
negative road grades, rolling resistance, and aerodynamic drag for various

transmission gears.

17

2.3.2 Empirical Versus Mechanistic VOC Models

The literature indicates that the existing models range from mechanistic models,
which require a wide range of input parameters, to simple empirical models that need
very few input data.

Mechanistic models take into account not only vehicle-related parameters, such as
the cost of purchase of the vehicle, its depreciation costs, maintenance costs, ﬁrel
consumption costs, etc, but also those other user costs such as value of time, road
roughness, road texture, tire life and replacement costs, etc. On the other hand, simple
models will exclude many of these more detailed (and difﬁcult to establish)
parameters [2]. The formulation of VOC models can be categorized into two basic
approaches: empirical and mechanistic models.

Table 2.1 shows approaches for the most common VOC models developed in
different countries. A more detailed description of the models is provided in Chapter
3.

Table 2.1 Categories of VOC Models (Empirical versus Mechanistic)

 

 

 

 

VOC Models
Feature HDM COBA TRDF HDM
-III 9 VETO NIMPAC “RI COM MicroBENCOST -IV
‘ Empirical \/ J - J - J -
Mechanistic / - f - \/ - /

 

 

 

 

 

 

 

 

 

 

2.3.2.1 Empirical models

Empirical models are based on a traditional approach in which data on vehicle
operating costs (derived from previous records) are subjected to a regression analysis,
from which a model is derived. The development of such models is data-intensive, and
they require frequent updating and re-calibration as a result of changes and

ﬂuctuations in prices, and in vehicle and road parameters.

18

However, they have the advantage of requiring less input data, and may be more
suitable for those applications where the availability of such data is limited. Their
disadvantage is that they cannot be applied to radically new situations or scenarios

because of their empirical origins.

2.3.2.2 Mechanistic-Empirical models

This type of model is based on mathematical representations of the mechanical
relationship between vehicle and road conditions. A number of these relationships are
broadly established and mathematically derived. Calibration of these models is
generally less data-intensive than for empirical models. Within this class of models,
deterministic models provide a single result from the input data, having a discrete
value. Probabilistic models use distributions of data as input, and provide, as output, a
distribution of values for the result, with an indication of the likely reliability of that
result. This type of models is capable of predicting the outcome for a wide variety of
scenarios, providing the appropriate input data is well known. However, the input data
necessary for reliable operation of these models is often very extensive, and

sometimes difﬁcult to establish reliably.

2.4 SUMMARY

Table 2.2 summarizes the essential features of the existing VOC models. In
summary, most of the recent available VOC models in the literature were developed in
various countries other than the USA. Keeping in view the mechanistic nature of these
newer models, most of these may still be applicable to US conditions with some

modiﬁcations/calibration.

19

Table 2.2 Summary of VOC Models

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

VOC Models
Feature HDM- TRDF
III VETO ARF COM HDM-IV MicroBENCOST
Empirical J - - - J
Mechanistic J J J J -
Level of A ggregation
Simulation - J J -
Project Level J J J J J
Network Level J - J J -
Vehicle Operation
Uniform Speed J J J J J
Curves J J J J J
Speed Change - J J J J
Idling - J J J J
Typical Vehicles
Default J J J J
User Speciﬁed J J J J J
Modern Truck J J J J
Road-related Variables
Gradient J J J J J
Curvature J J J _
Super-elevation J J J -
Roughness J J J _
Pavement Type J J J _
Texture - J J J -
Snow, water .. J J J J
Wind, Temperature - J J J _
Absolute Elevation J J J ..
VOC Components
Fuel, Oil, Tires,
Repair/Maintenance,
Depreciation J J J J J
Interest J J - J _
Cargo Damage - J - _
Overhead J J - _
Fleet Stock - J - -
Exhaust Emissions - J - J -

 

 

 

 

 

 

20

 

CHAPTER 3
FUEL CONSUMPTION MODELS

3.1 INTRODUCTION

This chapter identiﬁes and evaluates the existing models. Based on the detailed
literature review, the most relevant fuel consumption models to date were identiﬁed.
These models were studied in detail in terms of their assumptions, required inputs and

their applicability to US conditions.

3.2 IDENTIFICATION OF EXISTING FUEL CONSUMPTION MODELS
The fuel consumption models can be grouped into empirical- and mechanistic-based
models. The only available US. models are those of the Texas Research and
Development Foundation (TRDF) developed by Zaniewski et al [11]; an updated
version of this model is in the MicroBENCOST VOC module (1993). The most recent
models have been developed outside the US, and are mechanistic-empirical in nature.
The relevant models are:
0 The World Bank’s HDM 3 and 4 VOC models;
0 Australian NIMPAC VOC models (adopted in HDM III with some
modiﬁcations) and ARF COM model of fuel consumption (adopted in HDM

IV with some modiﬁcations);

0 Saskatchewan VOC models;

0 Swedish VETO models.

21

3.2.1 Empirical Models

Early work conducted in the US by Winfrey [9] established charts and tables for
calculating fuel consumption cost based on vehicle class only. Later Zaniewski et a1.
[1 1] updated the fuel consumption tables based on empirical models derived from
experimental ﬁeld trials. Although this is the most comprehensive study conducted in
the US to date, it did not treat all aspects of the problem. While fuel consumption tests
were carried out for idling, acceleration, deceleration, and constant speed driving, the
effect of pavement conditions was only considered in the constant speed case.
Constant speed mode was used for most of the experimental effort in these ﬁeld trials,
which also tested the effect of speed, grade, surface type, and pavement condition. No
tests were carried out for larger truck combinations, and relations were assumed for a
3-S2 unit. Also the fuel consumption values were based on only one test vehicle in
each class, except for the medium size car, where two identical vehicles were used so
that the variance between the two identical cars could be used in the statistical
analysis. However, the tests on the effect of pavement conditions showed no
signiﬁcant difference between the two identical cars, which means it was not
necessary to do these tests after all [I 1]. According to Zaniewski’s tables and charts,
pavement conditions had a minor effect on fuel consumption. They found that grade,
curvature, and speed were the major factors that affect fuel consumption.

The US Department of Transportation (USDOT) recently conducted a study to
investigate highway effects on vehicle performance [12]. The study developed the

following fuel consumption model based on regression analysis:

22

FC=— (3-1)

FE
T c
FE=a[—2—+b] (3.2)
where:
F C = Fuel consumption in L/km
FE = Fuel economy (km/L)
T = Engine torque (N -m)
a, b,c = regression coefﬁcients, depending on gear number

3.2.2 Mechanistic Models

Mechanistic models predict that the fuel consumption of a vehicle is proportional to
the forces acting on the vehicle. Thus, by quantifying the magnitude of the forces
opposing motion one can establish the fuel consumption. Mechanistic models are an
improvement over empirical models since they can allow for changes in the vehicle
characteristics and are inherently more ﬂexible when trying to apply the models to
different conditions. Some of the most recent mechanistic fuel consumption models
are given below. It was noted that most of the models are derived from earlier ones.
The following models are discussed chronologically. '

The South African fuel consumption model considers that the fuel consumption is
proportional to the total energy requirements that are governed by the total engine
power and an engine efﬁciency factor [14]. Equation (3.3) shows the form of this

model.

P
PC = 10003-121- (3.3)
V

where:

23

F C = Fuel consumption in mL/km

ﬂ = Fuel efﬁciency factor in ml/kW/s or mL/KJ
Pm, = Total power requirement in kW
v = Vehicle velocity in m/s

The South African model assumes that the fuel efﬁciency of the vehicle is
independent from the driving mode. However, a number of studies that were
conducted in the early 1980’s in Australia to model fuel consumption found that the
fuel efﬁciency increases in the acceleration case [15]. An improved mechanistic model

was then developed to predict fuel consumption using the following relationship.

 

,BzMazv
IFC = a + P + (3.4)
ﬂ ” 1000
where:
0! = Steady state fuel consumption in mL/S
,6 = Steady state fuel efﬁciency parameter in mL/(KJrn/s)
,62 = Acceleration ﬁrel efﬁciency parameter in mL/(KJm/sz)
M = Vehicle mass in kg
v = Vehicle velocity in m/s

Some studies in the later 1980’s in Australia found that the fuel efﬁciency is not
only a function of tractive power but also a function of the engine power. The
following mechanistic model (ARRB ARFCOM model) was developed to predict the
fuel consumption as a function of the input (engine) and output power. The general

form of the model is described by the following equations [16]:

IFC=max(a,,6x(P0ut—Peng)) (3.5)

24

ﬂzﬂbx(l+ehpxP0ut /Pm,x)

where:
P = The total output power of the engine required to provide
out tractive force and run the accessories (KW)
Peng = The power required to run the engine (KW)
Pmax = The rated power or the maximum power (KW)
ﬂb = Base fuel efﬁciency parameter in mL/(KJrn/s)
ehp = Proportionate decrease in efﬁciency at high output power

(3.6)

The model predicts the engine and accessories power as a function of the engine

speed. These relationships are from a regression analysis and are given below as

Equations (3.7) and (3.8).

 

 

2.5
Pacs = EALCX RPM + ECFLCmeax RPM
TRPM TRPM

2
RPM
2 * __
Peng ceng+beng (1000]
where,

EALC = The accessory load constant (KW)
ECFLC = The cooling fan constant

Pmax = The rated power or the maximum power (KW)
RPM = Engine speed

T RPM = Load governed maximum engine speed

ceng = Speed independent engine drag parameter
beng = Speed dependent engine drag parameter

25

(3.7)

(3.8)

However, Biggs [16] noted that the determination of the parameter values for the
engine drag equation was quite problematic with low coefﬁcients of determination and
high standard errors. Also, Biggs estimates the engine speed as a function of the
vehicle speed in order to compute the engine power. There are two different equations
in the engine speed model: One for a vehicle in top gear; the other for a vehicle in less
than top gear. However, these equations lead to a discontinuous relationship between
vehicle speed and engine speed when the vehicle shifts into top gear. Such
discontinuities lead to inconsistent fuel consumption predictions and should therefore

be avoided [16].

Recently, the World Bank updated the mechanistic fuel consumption model in the
HDM 4 module [17]. The model adopted is based on the ARRB ARFCOM
mechanistic model (Australian model) described above, but with a change to the
prediction of engine speed, accessories power, and engine drag. The following section

presents the details of the HDM 4 model.

3.3 HDM 4 FUEL CONSUMPTION MODEL

The general form of the model is expressed conceptually by Equation (3.9).

IFC = f(Ptr,Paccs + Peng) = max (mix Ptotx(1 + dFuel)) (3.9)
where:

IF C = Instantaneous Fuel consumption in mL/s

Ptr = Power required to overcome traction forces (kW)

26

[Daccs

Peng

a

dFueI

The engine efﬁciency decreases at high levels of output power, resulting in an
increase in the fuel efﬁciency factor I; The total power required is divided into tractive
power, engine drag, and vehicle accessories, respectively. The total requirement can
be calculated by two alternative methods depending on whether the tractive power is
positive or negative as shown in Table 3.1. The tractive power is a function of the
aerodynamic, gradient, curvature, rolling resistance and inertial forces. The
aerodynamic forces are expressed as a function of the air density and the aerodynamic
vehicle characteristics and are given in Table 3.2. The gradient forces are a function of
vehicle mass, gradient, and gravity. The curvature forces are computed using the slip
energy method. The rolling resistance forces are a function of the climate and the

vehicle characteristics. The inertial forces are a function of the vehicle mass, speed,

= Power required for engine accessories (e. g. fan belt, alternator etc.)
(kW)
= Power required to overcome internal engine friction (kW)

== Fuel consumption at Idling (mL/s)
= Engine efﬁciency (mL/KW/s)

(Ptot T Pens)

max

{={b 1+ehp

= Engine efﬁciency depends on the technology type (gasoline versus

diesel)
= Rated engine power

= engine horsepower

= Excess fuel conception due to congestion

and acceleration.

27

3.4 SUMMARY

As mentioned above, a large research body is available on the effects of pavement
condition on VOC and on models used to estimate these effects. Much of this
information and many Of the models were developed on the basis of data generated
years ago in other countries for vehicle ﬂeets that vary substantially from those used
currently in the United States and for roadways that differ from those built in the
United States. However, some relevant information was collected in the United States
in recent years that could help in reﬁning these models and/or developing new models.

In addition, it was also revealed that the most current research on fuel consumption
are highly based on ﬁrndamental principles of mechanics. These types of models offer
the following advantages [2]: (1) fully user deﬁnable; (2) require minimum data
collection; (3) provide a full audit trial (i.e. no “black box” calculations); (4) can
explicitly quantify uncertainty associated with fuel consumption; (5) take into account
region’s economy, vehicle technology, driver behavior, regulations, and ﬂeet

operating decisions (i.e. transferability).

28

Table 3.1 HDM 4 Fuel Consumption Model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Description Unit
P = _tr + Paces + Pen for P 2 0, uphill/level
Total power (PM) tat 90" g tr . kW
Ptot = edtthr + Paccs + Peng for P". < 0, downhill
edt Drive-train efﬁciency factor
Engine and Pengaccs = KPea meax
accessories power x ( Paccs _ a1 + (Paces _ a0 — Paccs _ a1) kW
Pengaccs = Peng >< RPM —RlI/H’Idle
+ Paces RPMlOO—RPMIdle
KPea Calibration factor
Pmax Rated engine power kW
P _—b+\Ib2—4xaxc
accs a1—
— 2Xa
I 2 lOO—PctPeng
= X X X
Paccs_a 1 a 5b ehp kPea Pmaxx 100
ibszkaeameax
C = —a
I
Engine efﬁciency depends on the technology type mL/kW/s
5b (gasoline versus diesel)
ehp Engine horsepower hp
a Fuel consumption at Idling mL/s
P 0 Ratio of engine and accessories drag to rated engine
aces—a power when traveling at 100 km/h
Percentage of the engine and accessories power used by 0
PCtPeng the engine /°
, RPM=a0+a1xSP+a2xSP2+a3xSP3 .
Engrne speed (RPM) Rev/mm
SP = max (20, v)
v Vehicle speed m/s
a0 to 03 Model parameter (Table 3.3)
RPM 1 00 Engine speed at lOOKm/h Rev/min
RPMIdle Idle engine speed Rev/min
Traction ower
(P ) P P =V(Fa+Fg+FC+F,+Fi) kw
tr if 1000
Traction forces (Table 3.2) N

Fa, F , Fc, Fr, F,-

 

 

 

29

 

Table 3.2 HDM 4 Traction Forces Model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Description Unit
Aerodynamic forces (Fa) Fa = 0.5 * p * CDmult * CD * AF * v2 N
CD Drag Coefﬁcient
CDmult CD multiplier
AF Frontal Area m2
,0 Mass density of the air Kg/m3
U Vehicle speed m/s
Gradient forces (F g) Fg = M "‘ GR * g N
M Vehicle weight Kg
GR Gradient radians
g The gravity M/s2
i 2 )
[M 202 —M x gxe]
Curvature forces (F c) F C = max 0, wa Cs ><10—3 N
x J
R curvature radius m
Superelevation (e) e: max j 0,0.45—0.68*Ln( 1a)] m/m
Nw Number of wheels
M M 2
Tire stiffness (Cs) Cs=KCSx a0+alx-——v—v-+a2x[m—)
KCS Calibration factor
a0 to a2 Model parameter (Table 3.4)

 

 

3O

 

Table 3.2 (cont'd)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Name Description Unit
Fr 2 CRZXFCLIM
Rolling resistance (Fr) x(b1 1x Nw+ CR1x(b12><M +b13x v2 )) N
CR1 Rolling resistance tire factor
_ , Ib11 = 37 x Dw
Rollrng resrstance .
parameters < b12 : 0.067 / Dw old tires
0.064 / Dw latest tires
(b1), b12, b13) 2
kb12 = 0.012x Nw/ Dw
Romng “gimme = Kcr2[a0+a1desp +a2><IRl+a3><DEF]
surface factor (CR2)
Kcr2 Calibration factor
a0 to a3 Model coefﬁcient (Table 3.5)
T dsp Texture depth using sand patch method mm
IRI International roughness index m/km
DEF Benkelman Beam rebound deﬂection mm
Climatic factor F CLIM = 1 + 0.003 * PCT DS + 0.002 * PCTDW
(F CLIM)
. a2
Inertial forces (Fi) Fl = M X 610 'I' 01X arCtan ['7] X a
2)
a0 to a2 Model parameter (Table 3.6)

 

 

 

31

 

Table 3.3 Final Parameters for the Engine Speed Model [17]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Engine speed
Vehicle Type a0 a1 a2 a3
Motorgycle -162 298.86 -4.6723 ~0.0026
Small car 1910 -12.311 0.2228 -0.0003
Medium car 1910 -12.311 0.2228 -0.0003
Large car 1910 -12.311 0.2228 -0.0003
Light delivery car 1910 -12.311 0.2228 -0.0003
light goods vehicle 2035 -20.036 0.356 -0.0009
four wheel drive 2035 -20.036 0.356 -0.0009
light truck 2035 -20.036 0.356 -0.0009
medium truck 1926 -32.352 0.7403 -0.0027
heavy truck 1905 -12.988 0.2494 -0.0004
articulated truck 1900 -10.178 0.1521 0.00004
mini bus 1910 -12.311 0.2228 -0.0003
light bus 2035 -20.036 0.356 -0.0009
medium bus 1926 -32.352 0.7403 -0.0027
heavy bus 1926 -32.3 52 0.7403 -0.0027
Coach 1926 -32.352 0.7403 -0.0027
Table 3.4 Final Parameters for Cs Model [17]
. <=2500kg >2500Kg
coefﬁcrent Bias radial bias radial
a0 30 43 8.8 0
a1 0 0 0.088 0.0913
a2 0 0 -0.0000225 -0.00001 14
Kcs 1 1 l 1
Table 3.5 Final Parameters for CR2 Model [17]
Surface surface <=2500kg >2500Kg
class type a0 a1 a2 a3 30 al a2 a3
Bituminous AM or ST 0.5 0.02 0.1 0 0.57 0.04 0.04 1.34
Concrete JC or GR 0.5 0.02 0.1 0 0.57 0.04 0.04 0
unsealed GR 1 0 0.075 0 l 0 0.075 0
unsealed - 0.8 0 0.1 0 0.8 0 0.1 0
CB, BR or
block SS 2 0 0 0 2 0 0 0
unsealed SA 7.5 0 0 0 7.5 0 O 0

 

 

 

 

 

 

 

 

 

 

32

 

Table 3.6 Final Parameters for Effective Mass Ratio Model [17]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vehicle Type Effect Mass ratio Model Coefﬁcients
a0 al a2
Motorcycle 1 .1 0 0
Small car 1.14 1.01 399
Medium car 1.05 0.213 1260.7
Large car 1.05 0.213 1260.7
Light delivery car 1.1 0.891 244.2
light goods vehicle 1.] 0.891 244.2
four wheel drive 1.1 0.891 244.2
light truck 1.04 0.83 12.4
medium truck 1.04 0.83 12.4
heavy truck 1.07 1.91 10.1
articulated truck 1.07 1.91 10.1
mini bus 1.1 0.891 244.2
_light bus 1.1 0.891 244.2
medium bus 1.04 0.83 12.4
heavy bus 1.04 0.83 12.4
coach 1.04 0.83 12.4

 

 

 

 

 

 

33

CHAPTER 4
CALIBRATION OF THE HDM 4 FUEL CONSUMPTION MODEL

4.1 INTRODUCTION

In the previous chapter, the most relevant existing fuel consumption models were
presented. From the literature review, it became apparent that most of the existing
modelsare derived from previous ones; each model improves the problems found in
the previous ones. The most recent VOC model found in the literature is HDM 4.
Therefore, the latest mechanistic-empirical model, i.e. those in the HDM 4 model, is
recommended. In this chapter, the research approach for adopting the appropriate
models to estimate the effects of pavement condition on fuel consumption will be
discussed. The approach proposed herein is: (1) ﬁeld trials, and (2) calibrating and

validating the HDM 4 fuel consumption model to US conditions.

4.2 TESTING OF THE ACCURACY AND PRECISION OF TEST
EQUIPMENT

The goal of this research is to estimate the effect of roughness on fuel consumption.
Therefore, in this case, the repeatability and accuracy of the measurement is a key
criterion for data interpretation. Preliminary tests were conducted to validate the
accuracy, stability, and repeatability of the equipments that will be used during ﬁeld
tests. Our focus was on ﬁnding data acquisition systems that could access and log data
from the vehicle’s Engine Control Unit (ECU) via On Board Diagnostic (OBD)

connector. Several equipments were identiﬁed. The data acquisition system used

34

 

during ﬁeld tests is AutoTap. Figure 4.1 shows the different parts of the data

acquisition system and how it will be connected to the vehicle.

      

-*.’ 7.; . i 9 ,_ I
(a) Different parts (b) OBD II Connection

Figure 4.1 Fuel Consumption Data Acquisition System

4.2.1 Principles of Engine Control Units

The Engine Control Units (ECU) controls various aspects of an internal combustion
engine‘s operation. ECUs determine the quantity of fuel, ignition timing and other
parameters by monitoring the engine through sensors. These sensors include Manifold
Air Pressure (MAP), Mass Air Flow (MAP), throttle position, air temperature, oxygen
sensor and many others. Often this monitoring and control is done using a control loop
(such as Parameters ID controller). Modern ECUs use a microprocessor which can
process the inputs from the engine sensors in real time. For example, for an engine
with fuel injection, an ECU will determine the quantity of ﬁre] to inject based on a
number of parameters. If the throttle pedal is pressed further down, this will open the
throttle body and allow more air to be pulled into the engine. The ECU will inject
more fuel according to how much air is passing into the engine. If the engine has not
warmed up yet, more ﬁrel will be injected (causing the engine to run slightly 'rich'

until the engine warms up).

35

An ECU contains the hardware and software. The hardware consists of electronic
components. The main component is a microcontroller. The software is stored in the
microcontroller so that the hardware can be re-programmed by uploading updated
code or replacing chips. Different protocols such as the Controller Area Network
(CAN) bus automotive network, SAE J 1850, etc. are often used to achieve
communication (messages) between these devices. The data logger will intercept these

messages (via the OBD connector), decode them and convert them to ASCII format.

4.2.2 Calculating Fuel Efﬁciency

Given the vehicle speed and the mass airﬂow rate, we can determine the
instantaneous fuel efﬁciency knowing a couple of other constants.

The ﬁrst constant is the engine’s air/ﬁre] ratio. In modern low-emissions vehicles,
the air/fuel rate is maintained at a constant chemically ideal ratio of 14.7 grams of air
to 1 gram of gasoline. We convert grams of air per second into grams of gasoline per
second by dividing by 14.7.

The second constant needed is the density of gasoline in grams per gallon. The
density of gasoline varies somewhat according to the fuel grade and ambient
temperature. For example, the unit weight of the unleaded gasoline is 6.17 pounds per
gallon. Knowing that there are 454 g in a pound, we can divide the mass airﬂow rate
by 14.7 and by 2,801 (i.e., 6.17 X 454) to determine the fuel ﬂow rate in gallons per
second. We then multiply this number by 3,600 (the number of seconds in one hour) to

determine the gallons per hour. In order to calculate the instantaneous fuel efﬁciency

36

in miles per gallon (MPG), we divide the vehicle speed in miles per hour by this last

number. To summarize, we use the following equation to calculate the fuel efﬁciency:

 

MPG=14.7*6.17*454*VSS (4.1)
3600 * MAF
Where:
MPG = miles per gallon
VSS = vehicle speed
MAF = mass air ﬂow rate

4.2.3 Repeatability and Accuracy Testing

Preliminary ﬁeld trials were conducted to test the accuracy and precision of the
instrument. Five different locations were selected based on the variability level of their
pavement conditions (i.e. roughness, gradient, texture and pavement type). These
locations include Flint, Lansing and Owosso. Three different vehicles were used: 2008

Mitsubishi Gallant, 2007 Chevrolet Impala and 2008 Mercury Sable.

4.2.3.1 Repeatability/Precision

Two different tests were conducted in two different locations using two different
vehicles to measure the repeatability of the instrument. During both tests, the outdoor
conditions for the identiﬁed sections were measured using a portable weather station.
Running tire pressure was maintained at 2.4 bar (35 psi). One test showed a signiﬁcant
change in the speed and therefore had to be repeated.

The ﬁrst test was conducted near Flint. The vehicle (2008 Mercury Sable) was
driven over pre-selected routes (I69E, I496N and 1758). Figure 4.2 shows an aerial

View of these routes (20 miles loop).

37

 

A: exit to I69E
B: Exit to I475 N
C: Exit to 75 S

 

 

 

 

Figure 4.2 Map View of Flint Loop

Pavement conditions of the loop were presented in Figure 4.3 and 4.4. The start and
end points of each run were marked by distinct ﬂags and road markers. The data
acquisition system was connected to the vehicle during the test. Five runs were made
on the pavement: three runs with a speed of 105 km/h or 65 mph (runl through run3)
and two runs with a speed of 96 km/h or 60 mph (run 4 and 5). Cruise control was
engaged to reduce the acceleration and deceleration cycles.

Figure 4.5 and 4.6 show data collected during run 1 through run 3 and mm 4 through
5 respectively. The Repeatability test result matrices are summarized in Table 4.1 and
4.2. The correlation between run 1 through 3 was almost perfect 0320.98 and .02%

error). Also, run 4 and run 5 were highly correlated (pz0.9 and .05% error).

38

 

 

 

5 , ..
I . l
A 4 I i ' . I i 41.
' ' ' '1 1 1. 11'

g 3 I I I i ’11 i ‘ II[ ‘ . .1 LII/i III I I III: i ,'I i all
E III 1 I", I , all] Igh‘. U1“ it“ UI‘liii‘Iiigiljgi I] v ' ' I‘ III
H 2 d I 1 .nl' .‘ .1 ,1," y 1!" IE1”,

1 s
O Iiﬁ-T— TIT "Tm—”I” T' '"T "*- ‘1 “—r—i "—7 ——:- -—-—-———ﬂ,__ —+ 0

Grade (%)

0 2 4 6 810121416182022242628303234
Distance(km)

PINE/kg) -':__Texru_.r<:tnirnil

Figure 4.3 Roughness and Texture Depth versus Distance (Flint Loop)

 

 

 

 

 

 

 

 

 

 

 

 

 

-5 ‘I’TTT'T' ‘_'I ' 'I __‘ __l______ ' '_ ' I ' I U 1'" I ' '

0 2 4 6 810121416182022242628303234

Distance (Km)

Figure 4.4 Grade versus Distance (Flint Loop)

39

 

 

Texture depth (m)

300 -1

N N

O U!

C O
1 1

ﬂ
LII
O

100 *

ﬁlel consumption (mL/Km)

U!
Q
1

 

O

 

 

 

 

 

 

 

 

 

 

02468

f—T

 

 

 

 

 

 

   

 

 

 

 

 

EXit to Exit to IE6X91tEtO R
I496N 1753 \
'1 I J,
m ,
T”’_"‘I" F—‘"'“l“ "Tﬁ"’7‘"‘—Tf_ﬂ_‘_T"’”—i._T—Tl I '1
10 12 14 16 18 20 22 24 26 28 30 32 34
__D_ist:1£e.(19n> _
I—runl —run2 -—run3I

Figure 4.5 Fuel Consumption versus Distance for Run] through Run3 (Speed 104 km/h)

Table 4.1 Flint Loop Repeatability Test for Run 1, Run 2 and Run 3

Correlations

 

 

 

 

I run1 run2 run3 I

run1 Pearson Correlation 1,000 .995" .980“

Sig. (2-tailed) .000 .00(J

N 638.000 633 638

' run2 Pearson Correlation .995" 1,000 972"

Sig. (2-tailed) .000 .000

N 633 633.000 GIJ

run3 Pearson Correlation 930" .972" 1 000
Sig. (2-tailed) .000 .000

N 638 633 638. 000|

 

 

 

 

**. Correlation is signiﬁcant at the 0.01 level (2-tailed).

40

 

 

 

 

 

 

 

 

 

Exit to -
300 — Ex1t to
I496N 1758
E 250 ﬁ
75! 200 I
E“
.2 I
a 150 -
8
5'3
8 100 -
0
E
50 4 b
O V 7 l l 'Tjhﬁ'ﬂ‘ ﬁT'—' F T
0 2 4 6 8
Distance (km)
F‘ ‘T

 

 

 

 

 

 

 

 

l. I'MI

 

Exit to
I69E

 

 

 

I

III

 

 

 

I—run4 -—run5_J

 

r

l l

1 1 1

10 12 14 16 18 20 22 24 26 28 30 32 34 36

Figure 4.6 Fuel Consumption versus Distance for Run 4 and RunS (Speed 104 lm/h)

Table 4.2 Flint Loop Repeatability Test for Run4 And Run5

 

 

 

Correlations
I run4 run5 I
run4 Pearson “
Correlation 1 '000 .904
Sig. (2—tailed) .000
N 747.000 745
run5 Pearson ..
Correlation 904 1 -000
Sig. (2-tailed) .000
N ’ 745 745.000

 

 

 

**. Correlation is signiﬁcant at the 0.01 level (2-

tailed).

41

The second test was conducted near Lansing. The vehicle chosen was a 2008 Chevy
Impala. The test section selected was a 9 mile stretch along I 69 (E and W). The speed
selected for the runs was 105 km/h (65 mph). Pavement conditions of the loop were
presented in Figure 4.7. The start and end points of each run were also marked by distinct

ﬂags and road markers. Two runs were made on the pavement.

 

4 ~ 2
3.5- ~ 1.5
A 3 ’1 I 1 A
25 . 4* ~ 0.5 $1
. 1 1
‘52 Ill 4 g
g 1.5 w y 4 -05 {15‘
1 -1
0.5 I —1.5
0 _— —l_ *'l"- T—"m—l— __-'7l #_' T’ﬁ‘ "—‘F -2
0 1 2 3 4 5 6 7 8 9
Distance (Km)

1+ IRI(m/km) + Grade (%) I

 

 

Figure 4.7 Grade and Roughness versus Distance (I69E Near Lansing)

Table 4.3 I-69 Repeatability Test between Run 1 and Run 2

Correlations

 

 

 

 

 

I run1 run2 I
run1 Pearson Correlation 1,000 .924"
Sig. (2-tailed) .000
N 4733.000I 4733
run2 Pearson Correlation .924" 1,000
Sig. (2-tailed) ' .000
N 4733 4733.000

 

**. Correlation is signiﬁcant at the 0.01 level (2—tailed).

42

Figure 4.8 shows the raw data collected during the two test runs. The repeatability test
result matrix is summarized in Table 4.3. The correlation between run 1 and 3 was also

almost perfect ($0.92 and .02% error).

Based on these results, we can conclude that the instrument reliably provides repeatable
results, and therefore it is sufﬁciently precise to determine the relative effects of surface

conditions on the change in fuel consumption.

100
907
80 ‘
7o - I ‘ , ‘l. ,l
60 - 1 . . ; '. ’ "I I 0“.

 

40*
301
20*
10~

Fuel Consumption (mL/Km)
M
o
<

 

0 2 4 6 8 10 12 14 16
Disance (Km)

I— run 1.7;“sz

1

 

Figure 4.8 Mass Air Flow Rate versus Distance (I-69 E near Lansing)

4.2.3.2 Data acquisition system accuracy/calibration

The attempt to calibrate the data acquisition system was conducted at the Civil and
Environmental Engineering of the University of Texas at Arlington. A fuel meter

(Graphtech) was used to collect the instantaneous fuel consumption (Figure 4.9).

43

Figure 4.10 shows their experiment setting. The instrumented van was driven under the
same environmental, operating and pavement conditions using both Graphtec and
AutoTap for 19.9 seconds at highway speed. Table 4.4 summarizes the data collected
using both data acquisition systems. The collected data were compared to each other. The
percent difference between the two instruments is 0.41%. Based on previous studies and
preliminary results from our own tests, the effect of surface roughness on fuel
consumption is expected to be in the range of 3 to 5% for the IRI range of 1 to 5 m/km.

Therefore, this level of error is acceptable.

 

(a) Fuel Meter (b) RTD Temperature Probe

    

I 32/7
(c) Data Acquisition System

Figure 4.9 Instrumentation Used by University of Texas at Arlington Research
Group

44

   
  
   

 

 

FUEL ENGINE
TANK
_ Fuel
Sensor 2
Transmission
Shaft

 

 

 

  
     

Data
Acquisition
System

Figure 4.10 Experiment Setting Used by University of Texas at Arlington

Table 4.4 Summary of Fuel Consumption Data Using AutoTap and Graphtec

 

 

 

Data acquisition Total fuel consumed
system (mL)
AutoTap 27.1
Graphtec 27.2

 

 

 

 

4.3 FIELD TRIALS

The main objective of these ﬁeld trials is to validate and calibrate HDM 4 fuel
consumption model. During the ﬁeld trials several data variables related to vehicle engine
parameters, pavement surface characteristics, and environmental factors was collected as

shown in Table 4.5.

45

Table 4.5 Data Collection for Engine Parameters, Environmental Factors and Pavement

Surface Characteristics during field trials

 

Response Variables

Environmental Variables

Pavement Surface

 

0 Calculated Power (kW)
0 Calculated Efﬁciency
(%)

 

 

(%)
Wind Speed (km/h)

 

(Engine Parameters) Characteristics
0 Fuel Rate (liters/h) Ambient Temperature (°C) Roughness (IRI)
0 Engine Revolution Maximum Relative Humidity Vertical
(rev/min) (%) Alignment
0 Fuel Temperature (°C) Minimum Relative Humidity Texture depth

 

Five different locations were selected based on the variability level of their pavement

conditions (i.e. roughness, gradient, texture and pavement type). All These locations were

near Lansing. Table 4.6 shows the ﬁeld test matrix. It should be noted that the tests were

conducted during both wet and dry conditions. The actual weather condition (temperature

and wind speed) were recorded using a portable weather station. Any change of more

than 3°C (5°F) in ambient temperature would introduce an error and the test would have

to be repeated. These pavement and weather conditions were typical in the US (Appendix

A). Table 4.7 summarizes these conditions.

46

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2-: 4N4: QN I SN 33 in: may 80% e53
N.mN-m.NN N.wN-NNN NaNiWN 2 N-o Gov engages 228$
m ~80 N 3Q H @Q N END H b5 $3m§>
pogm H353 .
mGoEUr—OU 5:303 vowuooom 5v 2an
N NN on NN 2 36 SS 395E em €2.35 2 82 2 N
x x
N NN % cm 2 333 3x4 3d 2 €255 E 28865 2 a: a N
x x
N NN cm ow to x S: :om 3: Hm€58 2 $2 m N
N NN 3 ow to 3-2 x Nam 22:28 as: can m 32 e N
N NN em N: we 3.: x 3. 285 § €92 as m N
N NN om NN no 0-3m x 5 5% sen em 2 e N
uo>t SowEDOUr—t 3.53;?
N NN em NN no 0-3 . x E 3: 32:3 2 N N
5>t nomﬁaooh 3533
N Nb 8 NN f w: x o om case: in use: 2 3.36 N _
N NN on NN 2 SEN; x bi use: 2 323 2 £30 a H
as: EEC 8a 3. Em teem Rom H35 awa
83232 A _. c A e 5 EMS: amuse 3: 09C.
. Sam is 83m
E2523

 

 

 

 

 

 

 

 

 

 

 

 

Ea: use 20E 3 29¢

47

The pavement conditions data (Raw proﬁle and texture depth) during the test were
collected by Michigan Department of Transportation. The equipments used by MDOT
are:

0 Rapid Travel Proﬁlometer: This vehicle measures the ride quality or
smoothness of pavements. Operating at highway speeds, it uses a laser to
measure the proﬁle of the roadway and an accelerometer to determine the
movement of the truck (Figure 4.11(a))

o Pavement Friction Tester: This vehicle measures pavement friction by
' applying water on the road and locking up the brakes on one wheel of the
trailer. Horizontal and vertical forces are measured by instrumentation on the
trailer axle. Once these forces are known, a ﬁction value can be determined

for the tire to pavement interface (Figure 4.11(b)).

 

Figure 4.11 MDOT Test Vehicles (a) Rapid Travel Proﬁlometer (b) Pavement
Friction Tester

The slope data surveys were collected by a third party using a high precision GPS.
The sampling rate is every 1 second at highway speed (every 30.5 m or 100 it). The
average error of the measurement is 12.7 mm (0.5 inch) per 0.5 km (0.3 miles), which
is translated to 0.003 % (about twice the error of the total station). The slope data

surveys were collected by a third party using a high precision GPS.

48

Six different vehicles that represents typical vehicle in the US (Appendix A) were
used:
0 Medium car (Figure 4.12 (a))
0 SUV (Figure 4.12 (b))
0 Van (Figure 4.12 (c))
0 Gasoline light truck (Figure 4.12 (d))
0 Diesel light truck (Figure 4.12 (d))

0 Articulated heavy truck (Figure 4.12 (e))

Table 4.8 summarizes the characteristics of these vehicles. Tests for trucks were
conducted under two different loading conditions: loaded (Figure 4.13) and unloaded.
The light truck was loaded with two concrete blocks. The total load was 2.82 metric
tones (6210 lb). The concrete blocks were tightly secured to the trailer. The only
possible movement is in the vertical direction. The trailer of the heavy truck was
loaded with steel sheets. The total load of the steel was 21.32 metric tones (47000 lb).
The Gross Vehicle Weight will be around 36.3 metric tones (80000 lb) which is the

maximum load allowed in US. These loading conditions are typical in the US.

49

 

Figure 4.12 Different Vehicle Used During Field Trials (a) Medium Car (b) SUV (c)
Van (d) Light Truck (e) Articulated Truck

50

Table 4.8 Characteristics of the Vehicles Used in the Field Trials

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. . Vehicle class
Characteristics . _
Medium car SUV VAN Light truck Heavy truck
Make Mitsubishi Nissan Ford GMC International
Model Galant Pathﬁnder E350 W4500 9200 6x4
Year 2008 2009 2008 2006 2005
Dragcoefflcient 0.4 0.5 0.5 0.6 0.8
Frontal area (m ) 1.9 2.9 2.9 4.2 9
Tare Weight (t) 1.46 2.5 2.9 3.7 13.6
Maximum allowable ,
Load (t) - - - 2.9 22.7
GVW (t) - - - 6.6 36.3
Weight of the load (t) - - - 2.8 21.3
Gas type Gas Gas Gas Gas/Diesel Diesel
Tire diameter (m) 0.38 0.4 0.4 0.4 0.57
Tire pressure (psi) 35 39 43 57/75 1 10
Tire type radial radial radial radial bias
Cargo length (m) - - - 4.88 15.85
Other - 4WD 15 seats - Flat bed

 

 

 

 

 

 

 

The vehicle was driven over a surface previously selected based on roughness,
gradient, texture and pavement type. The scanner was connected to the vehicle and the
vehicle was driven at different speeds on cruise control to reduce the acceleration and
deceleration cycles. Multiple and repeated runs were performed. In order to
understand the effect of cruise control on the collected data, all the tests were
conducted at constant speed with and without cruise control. The start and end points

of data logging were marked by distinct ﬂags and road markers.

Figure 4.14 show example data collected during runs at 35 mph except section GH,

HG (I69 east and west) where the speed was 88 km/h (55 mph). All the data collected

during ﬁeld test will be included in Appendix A.

51

 

(d) Loaded heavy truck

Figure 4.13 (3) Loading of Light Truck (b) Loaded Light Truck (c) Loading of Heavy
Truck ((1) Loaded Heavy Truck

52

Fuel consumption

Fuel consumption

 

   

   

 

:- g 400;
g ‘5- 300 i
2: E 3 ‘
5‘ a a 200 I
a 8 V
5 E 100 |
E “ ° ‘
PH - _
Distance = 3.2 Km Distance _ 1'6 Km
SUV 8 Van E Cobalt SUV
. 123 Van Z Diesel Truck
7
Diesel Truck 13 Gas Truck El Gas Truck
(a) Section ABC (b) Section DEF
1000 f‘* ___.,, mg *7 5 2000 '
, '5‘
800 l a. 1500 .
3 600 1 E 31000
5 400 § 5
200 , g 500
0 ‘“ 0
Distance = 4 Km Distance = 10 Km
I Cobalt SUV El Cobalt SUV
123 Van Z Diesel Truck E Van Diesel Truck
E21 Gas Truck 121 Gas Truck
(0) Section GH ((1) Section I] .
2000 —' gem ,_.-7 -1
1500
E 1000
500
0 ,
Distance = 10 Km
El Cobalt SUV
Van Diesel Truck
E3 Gas Truck

(e) Section JI

Figure 4.14 Examples of Collected Data — All Sections

53

4.4 CALIBRATION OF THE HDM 4 MODEL

The instantaneous fuel consumption was calculated using Equation 4.1. The
predicted and measured consmnption data were compared to each other. It was noted
that the HDM 4 model over predicts the engine speed-of the vehicle. The engine and
accessories power will be over predicted and the ﬁJeI consumption will be
overestimated (Figure 4.15). Consequently, when calibrating the the] consumption
model, the traction power (i.e., the effect of pavement conditions) will be
underestimated. Therefore, The HDM 4 engine speed model was calibrated ﬁrst

followed by the fuel consumption model.

 

120 T“ *” "‘* ‘ mi »--—~--——* LEE, — __ ’___, -_-.fv ,___, _.1
l ‘ H .
J (r ‘ l
100 l "a w [1‘ ~. ll
A H l “ ‘ H II \ b ~
80 .. 1 I \ H l l ‘ 1 I! r.“
n \ M 'J‘ l “'3’“ n ‘ 1
2 ‘r. ' 1 . l '1‘ ‘3 u
g 60 '5'“ " i n 2'“ H H IL ‘L [1‘ ' l ‘Jcl' l
E l‘ 1 " . r L it '.‘ ‘ . I. é.“ / +9 i
”-4 40 d 15,"! 1” J, ’ ll" Ill .' i V u " Tub \u ‘
1.. 1 “r, i ‘ =5
1
l
O '1 — —— — 1— _ __ ___H—1 —*rm 1”— '—— i
0 2 4 6 8 10
Distance (km)

 

1 F-HDM‘DESWI + HDM 4 with new engine modelj

 

Figure 4.15 Comparison between the Original HDM 4 Model and HDM 4 with the
New Engine Speed Model

4.4.1 Calibration of the HDM 4 Engine Speed Mode]
The engine speed model was calibrated for all vehicle classes using the data

collected during both winter and summer ﬁeld tests. Table 4.9 classiﬁes vehicle

54

classes into categories. Figures 4.16 through 4.18 show the results of the calibration
using both winter and summer ﬁeld test data. It was observed that the previously
calibrated model using winter data still holds for the data collected in the summer,
except for light truck. We should note. that the light truck used in winter tests had a
misﬁring engine. Therefore, we used only summer test data to calibrate the HDM 4
engine speed model for the light truck. Table 4.10 summarizes the engine speed model
coefﬁcients for winter and summer conditions. These coefﬁcients will be used during
the calibration of the HDM 4 fuel consumption model. Table 4.11 presents the

recommended coefﬁcients for the HDM 4 engine speed model.

Table 4.9 Vehicle Classification Used in the Engine Speed Model Calibration

 

Categories Vehicle classes Vehicle used
Passenger car Small car 0 Medium car
Medium car
Large car
Mini bus
Light delivery vehicle 0 Van
Lighhgoods vehicle
FWD 0 SUV

Light truck 0 Light truck
Light bus
Medium truck 0 Articulated truck
Heavy truck
Articulated truck
Medium bus
Heavy bus
Coach

 

 

Light commercial
vehicle

Four wheel drive (FWD)
Light truck

 

 

 

Heavy truck

 

 

 

 

 

55

 

 

 

3000 l y = -O.0007x3 + 0.2006x2 + 0.868x + 720.05
,N2500

32000

p—I
kl!
O
O

1000 ..

Engine speed

 

0 20 40 60 80 1 00 1 20
Speed (Km/h)

 

‘ A engine speed model (HDM 4)
. 9 measured engine speed-Dry condition
1‘____-:-.Calib_1:ated_model _ L.

(a) Calibration procedure — Medium car

1 l- 0 measured engine speed-Wet condition ;
l
l
l

 

2500 l y = 0.0062x3 - 0.301sz2 + 6.7795x + 671.98
3, 1 2==
6 2000 “i R 0.96 ..
g +AAAA MMA‘A‘AAAAM‘
g 1500 1
8* l
E 1000 ’.,«x:%ﬁ“’
'61) 'me '
:1
LL]

 

0 20 40 0
Speed (Km/h) 6

 

 

! 69 measured engine speed-wet condition l
l' A engine speed model (HDM 4)

‘ 0 measured engine speed- -Dry condition l
l _""" Calibrated model i

(c) Calibration procedure — Van
Figure 4.16 Calibration of the HDM 4 Engine Speed Mode] — Van and Medium
Car

 

 

56

 

y = 0.0019x3 - 0.1331112 + 3.6701x +

 

 

3000 " 982.37 1,.

A ﬂ 2 z

a 2500 R 0.99 A»

S

8

91‘

Q)

G

'50

1::

L11

0 “1“-” f'h— ‘ﬁﬁ " _"~ _ T- "‘ 7’ ____ “1
0 20 40 60 80 100 120
Speed (Km/h)

 

1*”--

1 ° measured engine speed- Wet condition 1
1 9 engine speed model (HDM 4) 1
1 0 measured engine speed- -Dry condition 1
a — Calibrated model 1

 

(a) Calibration procedure - SUV

y = -0.0018x3 + 0.3798x2 -
3000 3.0722x + 550.08

Engine speed (rpm)

 

 

0 20 40 60 80 100 120
Speed (Km/h)

1 ° measured en e speed- wet condition 1
1 . engine spee model (HDM 4)

1 6 measured engine speed- dry condition1 1
— Calibrated model

 

 

 

,..m.. _ __J

 

(c) Calibration procedure — Light truck

Figure 4.17 HDM 4 Engine Speed Model Calibration — SUV and Light Truck

57

Figure 4.18 HDM 4 Engine Speed Model Calibration — Heavy Truck

. 3000 —
E 2500 —
:3“ 2000 1
g 15001
:5; 1000 ~
1:: 500 "
LI.)

0 __,

 

F“ _._..__.. _

° measured engine speedé-Dry condition1
4 engine speed model (HDM 4)

1

3 2
y = 6E-05x + 0.2077x - 5.3791x + 799.6

2
R = 0.99

 

1::Ca11brated model

 

 

1

Table 4.10 New Coefﬁcients for the Engine Speed Model by Vehicle Class for Wet
and Dry Conditions

 

Engine speed coefﬁcients

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vehicle class Wet condition Dry condition
a0 a1 a2 a3 30 a1 a2 33

Small car 823.78 -4.6585 0.2702 -0.0012 720.05 0.868 0.2006 -0.0007

Medium car 823.78 -4.6585 0.2702 -0.0012 720.05 0.868 0.2006 -0.0007

Largg car 823.78 -4.6585 0.2702 -0.0012 720.05 0.868 0.2006 -0.0007

Light delivery

(goods)car 589.6 -0.5145 0.0168 0.0019 671.98 6.7795 -0.3018 0.0062

Four wheel

drive 943 .51 -0.0861 -0.0069 0.0007 982.37 3.6701 -0.1331 0.0019
_I;ight truck 797.01 -25.028 0.9112 -0.0049 550.08 -3.0722 0.3798 -0.0018

Mini bus 823.78 -4.6585 0.2702 -0.0012 720.05 0.868 0.2006 -0.0007
_I;ight bus 797.01 -25.028 0.9112 -0.0049 550.08 -3.0722 0.3798 -0.0018

Medium

truck - - - - 799.6 -5.3791 0.2077 0.00006

Heavy truck - - - - 799.6 ~5.3791 0.2077 0.00006

Articulated

truck - - - - 799.6 -5.3791 0.2077 0.00006

Medium bus - - - - 799.6 —5.3791 0.2077 0.00006

Heavy bus - - - - 799.6 -5.3791 0.2077 0.00006

Coach - - — - 799.6 -5.3791 0.2077 0.00006

 

58

 

Table 4.11 Recommended Coefﬁcients for the Engine Speed Model by Vehicle Class

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Engine speed coefﬁcients and
Vehicle class statistics
a0 . al a2 a3
Small car 720.05 0.868 0.2006 -0.0007
Medium car 720.05 0.868 0.2006 -0.0007
Lagge car 720.05 0.868 0.2006 -0.0007
_L_ight delivery car 589.6 -0.5145 0.0168 0.0019
light goods vehicle 589.6 -0.5145 0.0168 0.0019
four wheel drive 982.37 3.6701 -0. 1331 0.0019
_l_i_ght truck 550.08 -3.0722 0.3798 -0.0018
mini bus 720.05 0.868 0.2006 -0.0007
light bus 550.08 -3.0722 0.3798 -0.0018
medium truck 799.6 -5.3791 0.2077 0.00006
heavy truck 799.6 -5.3791 0.2077 0.00006
articulated truck 799.6 -5.3791 0.2077 0.00006
medium bus 799.6 -5.3791 0.2077 0.00006
heavy bus 799.6 -5.3791 0.2077 0.00006
coach 799.6 -5.3791 0.2077 0.00006

 

 

 

 

 

 

4.4.2 Calibration of HDM 4 Fuel Consumption Model

The HDM 4 fuel consumption model provides two coefﬁcients for calibration [71],

which are:
0 Kcr2 which modiﬁes traction power;
0 era which modiﬁes accessories and engine power.

The procedure used in this study calculates the least sum of square differences
between the observed ﬁeld values and those predicted using I-[DM 4 model (SSE).
Then the coefﬁcients that minimize SSE are obtained. The methodology that was used
is summarized as follows:

1. A random value was assigned to Kcr2, and then the value of era yielding
the lower least square value is determined.

2. This process is continued iteratively until the lowest least square is obtained.

59

The data collected during ﬁeld tests were used to calibrate the HDM 4 fuel
consumption model. It was noted that, with cruise control, low consumption was
underestimated; whereas, high consumption was overestimated (Figure 4.19. It was
also noted that, during the test, when the vehicle is driven over a steep positive slope,
the cruise control will be invalidated and the vehicle speed will decrease resulting in a
decrease in fuel consumption. However, when the vehicle is driven over a steep
negative slope, the HDM 4 model yields negative traction power, which is equivalent
to the vehicle mobilizing by itself without any need for traction force provided by the
engine. Thus, the predicted amount of fuel consumed will decrease. This never
occurred during the tests. Instead, we noted that the speed increased, and this is
because the cruise control never sets a negative force to the engine. Figure 4.20
supports these observations.

However, for some vehicles, it was difﬁcult to maintain constant speed without
cruise control especially when the roads are very rough. This is important especially
when looking at the effect of pavement conditions on fuel consumption. The
observations regarding the effect of cruise control were not applicable to the Medium
car, Van and SUV because they are light vehicles. Therefore, for calibration purposes,
data collected during tests with cruise control were used for these vehicles. The data
collected during tests without cruise control were used for light and heavy truck.
Figure 4.21 shows the results after calibration of the HDM 4 fuel consumption model
for all vehicle classes. Table 4.12 summarizes the calibration coefﬁcients. Statistical
analysis showed that there is no difference between the observed and the estimated

fuel consumption at 95 percent conﬁdence level.

60

Fuel consumption (mL/Km)

 

 

 

 

 

1
2.001 .0 1
1.50 '4 '1' “(5'
~ 1:? 4.
1.00 A '51::‘1 To
‘ Q1.- 1'5;n 1
~‘ .1 '
0.50 4 1
c3 1
1 1
0,001-- .-—- — ——-— -. 1

Distance (km)

29+099919xmunmr1:M99993909999;

 

 

 

 

(a) Van
350 "1*”‘7 “7* r 'V‘ ‘1 “ 'ir ‘ "'—’ —“ "‘—'
' .5» j
A c. .
—1 a 0 a,
£300 0 C; ﬂ}: El ‘
o (i «c» '
n I "0 c _ ». " f?’ f
E 250 1 i I? .53 . 10) . V ‘ I I 0 fa I
v a?" 1 9 0" 0 it " ~ .‘ " ""0 a ” '
I: 2 1‘ 13" 1 l O,"“-.." 0"“ h 01 '3‘ “A". 0‘ “a". r,
@200 3'" .1, , .~ ..,1 u - ,1 1 1., v- ,. J; ‘ .,
a "9 '4" ‘1'! ( :3 "Cl .3 ".0 21:0 0 V ‘i" ‘3) 1
§150 1 '3‘, E, 9:: a" . 4;, .,~, u,
E 1 6v.» 9 '5"- “19“)
3100 1
E); 1
11.. 50 1 1
0.- -- - _... -- 1
0 2 4 6 8 10
Distance (Km) 1
1 ~@— Predicted Fuel Consum tron + Measured Fuel Consum t1on 1
1.. _.__- _ . _ _ _______ ._ _ -2. __ _ . ______.-- - .2 . .
(b) Light Diesel Truck

Figure 4.19 Predicted and Measured Fuel Consumption versus Distance

61

200 ‘

100 '

Predicted Fuel rate (mL/Km)

 

0 100 200 300 400
Measured Fuel rate (mIJKm)

(a)
400

300 '“

200

Predicted Fuel rate (mL/Km)

1001

 

0 100 200 300 400
Measured Fuel rate (ml/Km)

(b)

Figure 4.20 Observed Versus Estimated Fuel Consumption (a) With Cruise Control
and (b) Without Cruise Control

62

 

__—___ 1'

_1

 

  
    

 

  

 

 

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

A 100 “—"— y=X 1———— w-*'— A 200 T'* ”—_T1 y=X 1
g ' R2 — 0 90 ' E 1 1 1 2 - ’
_1___# —. g 1 11R—083
g 80 ESE=4.09 g 150 1w”. 4 Lssrs= 9.58 .
a 601—... r“ * _.._ _ a 1 “— 1
3 ~ “-71,, 1 E 100 ' — —~ ’———--——1
‘1’ r ‘1’ T o
if: 40 '1’“. "'—"—’" ‘ij: vi ’ M“ § * 1 L: '1 1
'5 . .' - '° _._ , an, in
g 20 - 1 H - - ‘5' 50 1 1
1 .0
ﬁ 1 "ﬁ 1 1 1
8 0 ' .1 —'—‘ —1——‘— 2__ — 1—- —— 1 L) 0 ‘TT’T—T"” ‘_.— 7”_'——“ “‘ ’
0 20 40 60 80 100 0 50 100 150 200
Measm'ed Fuel rate (ml/Km) Measured Fuel rate (mIJKm)
(b) Medium car (b) SUV
1‘ 1
E 120 -~——-—~- 3' —. .—~-—~ 1 -~ A 300 4 1 - 4 —-
1ss1~3=419 7‘ 1 ‘ g 1
g .1 9 ' 1 . 1. 1 g
a 60 ‘ .t ‘ "”" w —
:3 0 ° , 1 1 3
L14 -. . 1 _.2.2 -_._ 1 _.__1 LL
8 4° .9 T * ‘ 8 “’0
8 20 J 1 1 .. -— a
i3 1 . é’
8 O i F __r_ ' “*7 ——+' '1 6 O 21 _. “,1.
0 20 40 60 80 100 120 0 100 200 300
Measured Fuel rate (ml/Km) Measured Fuel rate (ml/Km)
(c) Van ((1) Light Truck
y=x 1
A 400 --1—.——— 11w—r—— -—
E , 2 o
g R =0.88 ‘9 1
*1; 300 --—----1SSE=5.29 :o— ro_‘1
8 , o 1
E 200 -1——~—- — ° w—J—~W~1
d)
u? 1' :- 1
“O 100 _____-,_ .. .. ___,___.. _____ 1
3
8 ° 1 , 1
o - -1..-»
0 100 200 300 400
Measured Fuel rate (ml/Km)
(e) Articulated Truck

Figure 4.21 Observed Fuel Consumption versus Estimated Using HDM 4 Model — All
Vehicles

63

 

Table 4.12 Summary of the Model Performance

 

 

 

 

 

 

 

 

 

 

 

Kcr2 era SSE Number of data
(mL/Km) COIlSldCI'Cd
Medium car 0.5 0.25 4.09 456
SUV 0.58 0.56 9.58 250
Light truck 0.99 0.61 10.16 356
Van 0.67 0.49 4.19 352
Articulated truck 1.1 0.35 5.29 456

 

From previous studies and our sensitivity analysis of the HDM 4 model, the effect of
roughness on fuel consumption was estimated at about 5%. If the error of the estimate
exceeds 5%, then the model will not be able to estimate the effect of roughness
correctly. In our case, the error ranges from 2.5 percent for articulated trucks to 8
percent for medium cars. Therefore, more detailed analysis was conducted to look for

the sources of error and to validate the HDM 4 prediction for the effect of pavement

condition.

4.5 EFFECT OF ROUGHNESS AND TEXTURE ON FUEL CONSUMPTION

To estimate the effect of roughness and texture on fuel consumption, a more detailed

analysis was required. The analysis consists of the following operations:

1. Range discretization: The grade data were divided into equal ranges. The
width of the discretization interval was selected to be equal to 0.1% (based

on the sensitivity of fuel consumption to grade).

2. Analysis of covariance (ANCO VA): The grade was treated as a ﬁxed factor,
IRI as a covariate variable (IRI_SI) and the fuel consumption as the
dependent variable (FC_mLKm). The groups that have at least 3 points were

selected to be used in this analysis.

3. Linear regression analysis: A linear function was ﬁtted to the data within
each group of grade.

64

 

 

 

The results of the regression and the lack of ﬁt analysis are presented in Tables 4.13
through 4.15. 95 percent conﬁdence interval has been found to be a convenient level
for conducting scientiﬁc research: Intervals of lesser conﬁdence would lead to too
many misstatements and Greater conﬁdence would require more data to generate
intervals of usable lengths.

The lack of ﬁt test (Table 4.12) conﬁrms that the selected model ﬁts very well the
data (P-value is about 80%). Results summarized in Tables 4.14 and 4.15 show that:

1. 98 percent of variance is explained by the variables

2. Grade is statistically signiﬁcant at 95 % conﬁdence level (P-value is less than
5%),

3. IRI is statistically signiﬁcant at 95 % conﬁdence level (P-value is less than
5%),

4. Texture is statistically not signiﬁcant at 95 % conﬁdence level at 72 and 105
km/h (In Table 4.14, the P-value is about 55% which is more than 5%).

5. Texture is statistically signiﬁcant at 95 % conﬁdence level at 56 km/h (In
Table 4.15, the P-value is less than 5%).

Table 4.13 Lack of Fit Tests

 

 

 

 

 

 

lSource lSum of Squares df Mean Square F Sig. I
Lack of Fit 46.867 119 .394 .650 .792
Pure Error 1.818 3 .606

 

65

Table 4.14 Tests of Between-Subjects Effects at 89 km/h

 

 

 

 

 

 

 

 

 

 

 

| Type 111 Sum
Source of Squares df Mean Square F Sig.
1Corrected Model 4300_7693 14 307.198 769.817 .000
Intercept 19697.944 1 19697.944 49361.721 .000
IRI 23.557 1 23.557 59.032 .000
Texture .147 l .147 .368 .545
Grade 3796.846 12 316.404 792.887 .000
Error 48.684 122 .399

otal 351401.815 137
Corrected Total 4349.454 136
a. R Squared = .989 (Adjusted R Squared = .988)

Table 4.15 Tests of Between-Subjects Effects at 56 km/h
Source Type III Sum
of Squares df Mean Square F Sig.

Corrected Model 1381 13.100a 13 10624.085 2077.841 .000
Intercept 335375.546 l 335375.546 65592.193 .000
IRI 500.674 1 500.674 97.921 .000
Texture 23.920 1 23.920 4.678 .032
Grade 123056.405 l l l 1 186.946 2187.924 .000
Error 628.904 123 5.113

Total 5525978735 137

Corrected Total 138742.004 I36

 

 

 

 

 

 

 

 

a. R Squared = .995 (Adjusted R Squared = .995)

According to Sandberg [115], road surface texture has higher effect on rolling

resistance at higher speed. However, the results from the analysis of covariance

showed that the effect of texture on fUCl consumption is statistically not signiﬁcant at

higher speed. An explanation for these observations is that, at higher speeds, air drag

becomes the largely predominant factor in fuel consumption. The increase in rolling

resistance (i.e., fuel consumption) due to texture will be shadowed by the increase in

66

 

 

air drag due to speed. Figure 4.22 taken from a Michelin publication [116] shows how
mechanical power available at the engine is distributed as a function of the vehicle
speed (no climbing, no acceleration) for passenger car. At a constant speed of 100
km/h on a horizontal road, air drag represents 60% of energy loss while rolling
resistance accounts for 25% and internal ﬁiction (drive line loss) for 15%. For heavy
truck, approximately 12% of the fuel consumption is accounted for by the rolling
resistance losses in the tires at a constant speed of 80 km/h. This energy loss
represents approximately 30% of the available mechanical power from the engine.

Therefore, the effect of texture as a percentage is lower at higher speed as shown in

 
 
 

  
 
 
  
 

Figure 4.23.
100 '
"£3
E 80 Intemal
0 \° Friction
:5 a, 60 1
“S 333 1
g 8. 40 1 Rolling
g 20 1 Resistance .
9-4 1 1
O F 1’__l ' ' ' ' 1'" 1 l "l ' ""1

0 20 4O 60 80 100120140 160
Speed (km/h)

Figure 4.22 Energy Distribution in a Passenger Car versus Speed as 3 Percentage of
the Available Power Output at the Engine [after 116]

67

 

 

 

g 1.8 --
r: 1.61
.2 '
g 1.41 .
53 1.2 " .""
g 11 "“. ‘0‘0
.3 0-81 ' v’ .Go'
a) ,- .—'
«E 061 ”o"
3‘3 0.4I _'.e"' . "vi-11¢"
g0 0.21 . . - ',,--«'W:'ﬂ
_: 1““ ’ 1 ﬁv—m—1—— ——~ r —
U
0 0.5 1 1.5 2 2.5 3
Mean Proﬁle Depth (mm)
1+Mediumcar-35 mph --°--Mediumcar-55 mph 1
1+SUV -35mph ”AP‘SUV-SSmph 1
l—i—Van-3Smph --X--Van-55mph 1
1—0—Lighttruck-35 mph --0--Light truck- 55 mph 1
I

1:.— Articulated truck — 35 mph - o - Articulated truck - 55 mph

1
__4

 

 

 

 

Figure 4.23 Effect of Surface Texture on Fuel Consumption

Since the effect of roughness is statistically signiﬁcant, our focus was also to check
the accuracy of the calibrated model. Figure 4.24 (a) shows the change in fuel
consumption as a function of IRI using the calibrated HDM 4 model and the results
from the regression analysis described above. As seen in Figure 4.24, the results match
very well. Therefore, the HDM 4 model was able to predict the effect of roughness on
fuel consumption reasonably well. Figure 4.24 (b) shows the trends predicted by HDM
4 before calibration. The comparison of sensitivity analyses before and after
calibration showed that the effect of roughness on fuel consumption increased by 1.75
for the van, 1.70 for the (articulated truck, 1.60 for the medium car, 1.35 for the SUV

and 1.15 for the light truck.

68

Change in fuel consumption (%)

Change in fuel consumption (%)

2.5 1 .I.

 

 

 

1.5 1 1, -/. ....... A
1 ~— mv/"jj ,,,, . ------ ‘ _____ - ....... .. ------- °
0.5 1 é ﬁg "
o 1 ——~ *4 a 4 —~ —1
l 2 3 4 5
lRI(m/km)
1r—l—Mediumcar- HDM4 ---E}-- Mediumcar- Regression 1
+SUV-HDM4 "'A" SUV-Regression 1
1—)<-—Van-1-lDM4----+ Van-Regression '1
l—o—Lighttruck-HDM4 ---€>-- Lighttruck-Reglession 1

1—o—Articulated mick — HDM 4 ---<>-- Articulated truck - Regessioni

 

(a) Comparison of HDM 4 with regressed data

4.5 7
4 .1
3.5 1
3 1
2.5 1
2 '
0.5 /
0 , W“ i , a
l 2 3 4 5
IRI (m/km)
1 '—-—'_" ’Méﬁiinnﬁr‘ ' 1
1 +SUV
1 —)(—Van
1 +Light truck
1 _ +_ Articulated truck

(b) HDM 4 without calibration

Figure 4.24 Effect of Roughness on Fuel Consumption

69

4.6 EFFECT OF PAVEMENT TYPE ON FUEL CONSUMPTION
A detailed analysis to ﬁnd the effect of pavement type on fuel consumption was
conducted. The methodology is the following:

1. Locate pavement sections that have the same slope but different pavement
type (i.e., concrete and asphalt).

2. Conduct univariate analysis having IRI as a covariate and pavement type as
ﬁxed factor

3. Repeat for all vehicles at 56, 72 and 88 km/k (35, 45 and 55 mph).

Figures 4.25 and 4.26 show the mean and standard deviation of ﬁle] consumption for
articulated and light trucks respectively driven over different pavement type at 56, 72
and 88 mph. Tables 4.16 and 4.17 present summary statistics for articulated and light
trucks respectively. Tables 4.18 through 4.23 show results of the main effect analysis
using SPSS. It was noted that, for both truck type and for summer conditions, the

mean difference of fuel consumption between Asphalt and Concrete pavements is
statistically signiﬁcant at 56 km/h; whereas, it is statistically not signiﬁcant at higher
speeds (i.e. 72 and 88 km/h).

However, for winter conditions, the mean difference of fuel consumption between
Asphalt and Concrete pavements is statistically not signiﬁcant. The analysis showed
that the mean differences of fuel consumption between Asphalt and Concrete

Pavements for passenger car, van and SUV are also statistically not signiﬁcant (Table
4-24)- These observations could be explained by the viscoelastic behavior of asphalt

Pavement (Figure 4.27).

70

 

 

 

 

 

260.00—
§ 240.00-
1::
c
E.
E
i E E
a 220.00“
0
3
Ln Eli
5 IE
0 200.00-
E
180. I T l
56 72 88
Speed (Km/h)

Pavement_Type

I AC
0 PCC

Error Bars:
95% Conﬁdence
Level

I AC
I PCC

Figure 4.25 Mean and Standard Deviation of Fuel Consumption for Different

Pavement Type and Speed— Articulated Truck

71

 

 

 

 

Pavement_Type

260.00- I AC
A 0 PCC
5, 240.00-
S
E.
g Error Bars:
,1” E E 95% Conﬁdence
s 220.00- Level
D
E I AC
5 I I PCC
:1
a I
a 200.00“
2

180. l r I

56 72 88
Speed (Km/h)

Figure 4.26 Mean and Standard Deviation of Fuel Consumption for Different
Pavement Type and Speed — Light Truck

Table 4.16 Estimated Marginal Means — Articulated Truck

 

 

 

 

 

 

 

 

Speed 95% Conﬁdence Interval
x (km/h) Mean Std- Error Lower Bound Upper Bound
PCC 56 2013913 1.077 198.610 204.171
72 2229383 1.053 220.220 225.655
88 2484013 1.190 245.328 251.473
AC 56 2094223 1.008 206.821 212.024
72 225121331 1.034 222.543 227.883
88 247.5883 1.183 244.535 250.641

 

 

 

a. Covariates appearing in the model are evaluated at the
following values: IRI = 1.1908.

72

Table 4.17 Estimated Marginal Means — Light Truck

Dependent Variable:FC_mLKm

 

 

 

 

Speed 95% Conﬁdence Interval
x (km/h) Mean Std Error Lower Bound Upper Bound
PCC 56 151.1303 1.080 148.344 153.916
72 1877783 1.071 185.015 190.541
88 225.1883 1.294 221.850 228.527
AC 56 156.739a 1.061 154.002 159.475
72 18831983 1.052 185.685 191.112
88 2199739 1.225 216.811 223.134

 

 

 

 

 

 

 

a. Covariates appearing in the model are evaluated at the
following values: IRI = 1.2298.

Table 4.18 Pairwise Comparisons — Articulated Truck at 56 km/h

Dependent Variable:FC_mLKm

 

 

 

 

 

95% Conﬁdence Interval for

. a

Mean 3 Difference
(I) (J) Difference (I-J) Std. Error Sig. Lower Bound Upper Bound
PCC AC 1727* 1.045 .000 -10.436 -5.018
AC PCC 7727* 1.045 .000 5.018 10.436]

 

 

 

 

 

 

Based on estimated marginal means
* - The mean difference is signiﬁcant at the .01 level.

a- Adjustment for multiple comparisons: Least Signiﬁcant Difference (equivalent
to no adjustments).

73

Table 4.19 Pairwise Comparisons — Articulated Truck at 72 km/h

Dependent Variable:FC_mLKm

 

 

 

 

95% Conﬁdence Interval for
. . a
Mean a Difference
(I) (J) Difference (I-J) Std. Error Sig. Lower Bound Upper Bound
|Pcc AC -2242 1.712 .191 -6.678 2.194
IAC PCC 2.242 1.712 .191 -2194 6.678

 

 

 

 

 

 

 

Based on estimated marginal means

a. Adjustment for multiple comparisons: Least Signiﬁcant Difference (equivalent to
no adjustments).

Table 4.20 Pairwise Comparisons — Articulated Truck at 88 km/h

Dependent Variable:FC_mLKm

 

 

 

 

 

 

 

 

 

95% Conﬁdence Interval for

. a

Mean 3 Difference
(I) (J) Difference (I‘D Std. Error Sig. Lower Bound Upper Bound
IPCC AC .354 2.965 .905 -7.342 8.050'
IAC PCC -354 2.965 .905 -8.050 7.342|

 

 

Based on estimated marginal means

a. Adjustment for multiple comparisons: Least Signiﬁcant Difference (equivalent
to no adjustments).

Table 4.21 Pairwise Comparisons — Light Truck at 56 km/h

Dependent Variable:FC_mLKm

 

 

 

 

 

95% Conﬁdence Interval for
Mean Differencea
(I) (J) Difference (I-J) Std. Error Sig.a LowerBound Uppeermd
PCC AC 5475* 1.358 .000 -8.990 -1.960|
AC PCC 5475* 1.358 .000 1.960 8.9901

 

 

 

 

 

 

 

Based on estimated marginal means
*. The mean difference is signiﬁcant at the .01 level.

a. Adjustment for multiple comparisons: Least Signiﬁcant Difference (equivalent to no
adjustments).

74

Table 4.22 Pairwise Comparisons — Light truck at 72 km/h

Dependent Variable:FC_mLKm

 

 

 

 

 

 

 

 

 

 

 

95% Conﬁdence Interval for
. a
Mean Difference
. . a
(I) (J) Difference (I-J) Std. Error Slg. Lower Bound Upper Bound
[PCC AC -.845 1.424 .553 —4.529 2.840]
IAC PCC .845 1.424 .553 -2.840 4.529]
Based on estimated marginal means

a. Adjustment for multiple comparisons: Least Signiﬁcant Difference (equivalent to no

adjustments).

Table 4.23 Pairwise Comparisons — Light Truck at 88 km/h

Dependent Variable:FC_mLKm

 

 

 

 

 

 

 

95% Conﬁdence Interval for
. a
Mean 3 Difference
(I) (J) Difference (I—J) Std. Error Sig. Lower Bound Upper Bound
IPCC AC 5.477 2.826 .054 -l .848 12.803
IAC PCC -5.477 2.826 .054 -12.803 1.848

 

 

 

 

Based on estimated marginal means

a. Adjustment for multiple comparisons: Least Signiﬁcant Difference (equivalent

to no adjustments).

75

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

<2 <2 <2 <2 8 we- 38 33. mm 8::
<2 <2 <2 <2 3 2 28 32 Ne 858:2
<2 <2 <2 <2 0 w :8 4.8.8. 8 .
3 3- 28 Sam 3 2- $9 6.6: mm

3 ed 32 $2 ed 8 we: is: as 8.5 am:
3 S no: 962 o 88 :2 82 em

no 2 32 $2 3 mo- 22 ed: 2

we no 82 was 4.0 8 2.6: 3: Ne >3.
8 no 32 82 3 2 32 22 em

No 3. 32 E: 63 3. mm: 5.: mm

3 2- em: 32 mac 3. 98 6.8 Q 5S
8 No 93 Ex moo 3 2.; w; 3

no 2 4% 8% 3 S- Re 3: mm 86
3 3- Se Se N3 3. 38 6.8 E :58:
no 8 <3 New :5 3. SW 6.2 8 .

DOCOHD — OOGOHO —
8562mm eeeemb com o< eeeeeﬁeem 8&2 com o<
H8.55 868% 323: mmﬂo
. Beam 263$

@855
2238338 33 :82

 

 

 

 

ans—25> =< 8m 838800 0232mm mo meEzm 3% 2an

76

__ 1 ____ #"h— ‘——'—“1
350 ' 1+ sensor 111

 
    
  

300 ,1 1+ sensor 121

1+ sensor 141
250 1 1+sensor 311
200 1 1+ sensor 321

 

 

 

Strain (microns)

 

0 10 20 30 40 50 60 7O 80 90
Speed (Km/h)

Figure 4.27 Effect of Speed on Asphalt Pavements [After 72]

4.7 SUMMARY AND CONCLUSION

In this chapter, we showed that the HDM 4 fuel consumption model was able of
predicting very adequately the fuel consumption of ﬁve different vehicle classes under

different operating, weather and pavement conditions.

The better accuracy achieved after calibration of the HDM 4 fuel consumption model to
US conditions has improved the prediction of the effect of roughness on ﬁiel
consumption. The comparison of sensitivity analyses before and after calibration showed
that the effect of roughness on fuel consumption increased by 1.75 for the van, 1.70 for
the articulated truck, 1.60 for the medium car, 1.35 for the SUV and 1.15 for the light
truck.

Also, the key characteristics of representative vehicles used in the HDM 4 model vary

substantially from those of US vehicles. Therefore, the predicted fuel consumption using

77

the model without calibration was lower than the actual consumption. For example, the
amount of fuel consumed by a light truck in the US is equivalent to the predicted
consumption by a medium truck using the HDM 4 default characteristics of
representative vehicles. Therefore, we recommend the use of the HDM 4 fuel
consumption model after calibration and adjustment to US conditions using the

calibration factors mentioned in Table 4.11.

The Analysis of Covariance of the data collected during the ﬁeld test showed that the
effect of surface texture is statistically signiﬁcant at 95 percent conﬁdence interval for
heavier trucks and at low speed. An explanation for these observations is that, at higher
speeds, air drag becomes the largely predominant factor in fuel consumption. The
increase in rolling resistance (i.e., fuel consumption) due to texture will be shadowed by
the increase in air drag due to speed. For example, at a constant speed of 100 km/h on a
horizontal road, air drag represents 60% of energy loss while rolling resistance accounts
for 25% and internal friction (drive line loss) for 15%. For heavy truck, approximately
12% of the fuel consumption is accounted for by the rolling resistance losses in the tires
at a constant speed of 80 km/h. This energy loss represents approximately 30% of the
available mechanical power from the engine. Therefore, the effect of texture as a

percentage is lower at higher speed.

In this chapter, the effect of pavement type on fuel consumption was also investigated.
The detailed analysis of the data collected without cruise control showed that, for both
light and articulated trucks and for summer conditions, the mean difference of fuel
consumption between Asphalt and Concrete pavements is statistically signiﬁcant at 35

mph (trucks driven over AC consume more than trucks driven over PCC); whereas, it is

78

statistically not signiﬁcant at higher speeds (i.e. 45 and 55 mph). For winter conditions,
the mean difference of fuel consumption between Asphalt and Concrete pavements is
statistically not signiﬁcant. The analysis also showed that the mean differences of fuel
consumption between Asphalt and Concrete pavements for passenger car, van and SUV
are also statistically not signiﬁcant. These observations could be explained by the

viscoelastic behavior of asphalt pavement.

79

CHAPTER 5

CALIBRATION OF REPAIR AND MAINTENANCE COSTS
MODEL '

5.1 INTRODUCTION

This chapter identiﬁes and evaluates the existing repair and maintenance costs
models. Based on the detailed literature review, the most relevant repair and
maintenance costs models to date were identiﬁed. These models were studied in detail

in terms of their assumptions, required inputs and their applicability to US conditions

5.2 OVERVIEW OF EXISTING REPAIR AND MAINTENANCE MODELS
5.2.1 Introduction

Vehicle repair and maintenance costs are mainly comprised of two components:
Parts consumption and labor hours. Maintenance costs are often a signiﬁcant
contributor to the beneﬁts from road improvements.

The current models can be grouped into empirical- and mechanistic-based models.
The only available US. models are also those of the Texas Research and Development
Foundation (TRDF) developed by Zaniewski et al [11]. The most recent models have
been developed outside the US, and are mechanistic-empirical in nature. The relevant
models are:

0 The World Bank’s HDM 3 and 4 VOC models;
0 Saskatchewan VOC models (Canada);
0 South African model

0 Swedish VETO models.

80

The details for the most commonly used repair and maintenance cost models are

presented in this section.

5.2.2 Empirical Models

Winfrey [9] presented maintenance costs based on the results of surveys. However,
Winfrey’s tables and charts did not consider the effect of roughness. These were later
updated by Zaniewski et a1. [11]. The modiﬁcation factors related to roughness were
calculated using data collected from Brazil [18].

Saskatchewan Highways and Transportation (SHT) adopted a vehicle maintenance

cost model that relates maintenance costs to roughness:

Mc:MCmer (5.1)
where:
MC = Maintenance cost ($/km)
= Avera e maintenance cost $/km
Mcf g ( )

K = Road roughness coefﬁcient
mr

HDM 3 allowed users to predict repair and maintenance using relationships derived
from road user cost studies in Brazil, India, Kenya, and the Caribbean. The Brazil
relationships were the ‘standard’ relationships in HDM 3 [18]. Equations 5.2 through

5.6 show the Brazilian model.

(5.2)

PARTS COSPx CKMkp x exp(CSPIRI>< 1R1) for [R] < IRIOSP

5.3
PARTS ( )

CKMkp (a0 + alxIRI) for IRI>IRIOSP

81

C OSP exp(CSPIRI X IRIOSP)(1 - CSPIRI X IRIOSP) (5-4)

00 =

a1 = COSPXCSPIRI exp(CSPIRIxIRIOSP) (5.5)
LH =COLH PARTSCLHPC exp(CLHIRI IRI) (5.6)
where:

PARTS = Standardized parts consumption as a fraction of the replacement vehicle
price per 1000 km

CKM = Vehicle cumulative kilo-meter

IRI = Roughness in IRI m/km

IRIOSP = Transitional roughness beyond which the relationship between parts

consumption and roughness is linear

COSP = Parts model constant
CSPIRI = Parts model roughness coefﬁcient
LH = Number of labor hours per 1000 km

COLH = CSPIRI = Parts model roughness coefﬁcient
CLHIRI = Labor model roughness coefﬁcient

The Brazilian model actually incorporates several vehicle classes including
passenger cars, utility vehicles, buses and trucks. The HDM 3 repair and maintenance
model suggests parameters for parts and labor for all vehicle classes, as shown in
Table 5.1.

The structure of the parts model as shown above is quite complicated because trucks
were found to have a linear response to roughness while passenger cars, utility

vehicles, and buses had an exponential response.

82

Table 5.1 HDM 3 Maintenance Model Parameters [l7]

 

 

 

 

 

 

 

 

 

 

Parts Model Parameters Labour Model Parameters
Vehicle Class COSP CSPIRIP
kp (x10'6) (1110'3 ) IRIOSP COLH CLHPC CLHQI
Passenger Car 0.308 32.49 178.1 9.2 77.14 0.547 0
Utility 0.308 32.49 178.1 92 77.14 0.547 0
Large Bus 0.483 1.77 46.28 14.6 293.44 0.517 0.0715
Light and
Medium Truck 0.371 1.49 3273.27 0 242.03 0.519 0
Heavy Truck 0.371 8.61 459.03 0 301.46 0.519 0
Articulated
Truck 0.371 13.94 203.45 0 652.51 0.519 0

 

 

 

 

 

 

 

As part of considerable studies into VOC, the Council for Scientiﬁc and Industrial

Research (CSIR) in South Africa developed models for parts consumption. The

research can be grouped into two areas: Speed and roughness effects on parts

consumption as well as labor costs. Speed effect on parts consumption is the

following:

“3

PCST = a1 + asz +—+a4xS2
S

Where,

PCST
S

a1 to a4

The roughness effects on parts consumption is described as follows:

Maintenance cost in cents/km
= Speed (km/h)

= Model constants

PARTS = exp(-3.0951 + 0.4514 In CKM + 1.2935 ln(13IRI)) x 103

83

(5.7)

(5.8)

 

The key problem with this model is that it is sensitive to the assumed average speed.

In fact, du Plessis [19] proved that there is a huge difference in the parts cost when

assuming urban versus rural speed.

The South African model includes two separate equations (buses and trucks) for

modeling labor costs:

 

Buses
PARTS 0'5 ‘7
LH = 0.763 exp (0.0715 IRI) —— (5.9)
N VPLT
Trucks
PAR TS
LH =max(3, -0.375+0.0715 ( )+0.1821RI) (5 10)
N VPLT '
Where,
LH = Number of labor hours per 1000 km
PARTS = Standardized parts consumption as a fraction of the replacement vehicle
price per 1000 km
[R] = Roughness in IRI m/km
NVPLT = Cost of new vehicle less the cost of a set of new tires

For HDM 4, the parts model was simpliﬁed over that used in HDM 3 [17]. Details

about the HDM 4 repair and maintenance costs were included in the following section.

84

5.2.3 Mechanistic Models

The only purely mechanistic model is the VETO model which was developed by the
Swedish Road and Trafﬁc Research Institute (VTI) [20]. It contains two approaches to
the calculation of parts and maintenance labor, one empirical and one mechanistic.
The former relies on the HDM Brazil relationships (Equations 5.2 to 5.6) while the
latter employs a "wear index" for vehicle components. The mechanistic model is a
detailed simulation of an idealized two-dimensional vehicle traveling over a surface
with a speciﬁed proﬁle. The model works on the basis that the wear and tear of
components depends upon the product of the number of stress cycles they have been
subjected to and the stress amplitude raised to the sixth power. The number of cycles
is assumed to be constant per unit length of road (independent of roughness) while the
stress amplitude for each component is proportional to the RMS value of the dynamic
component of the wheel load. The model does not take into account the static load.
The model was calibrated by looking at the life expectancy of different components.
Only four components were studied and so the model does not yet provide a total cost
calculation. Nevertheless, it is interesting to note that the change in vehicle wear with
increasing roughness that it calculates is far higher than the change in parts cost
predicted by the empirical model. In spite of developing a complex model,

Hammarstrom and Karlsson [20] concluded that:

"... it would probably be virtually impossible to develop a model which could be
used for calculating the relationship between total repair costs and road

unevenness, component by component. "

85

Hammarstrom and Henrikson [21] produced coefﬁcients for calibrating HDM 3 to
Swedish conditions. The study produced scaling constants (COSP) and was also able
to examine how parts consumption changes with vehicle age (kp). However, it did not
provide information on the effect of roughness on parts consumption. Now, the
Swedish Road and Trafﬁc Research Institute are trying to update the HDM 4 model to

Sweden condition [22].

5.2.4 HDM 4 Repair and Maintenance Costs Models

Equations 5.11 through 5.14 present the HDM 4 repair and maintenance costs
model. The model suggests eliminating the effects of roughness on parts consumption
at low IRI (less than 3 m/km). This is achieved by using the smoothing relationship

given in Equation 5.14.

k
PARTS = 1K0pc [CKM P (a0 + a1 RD] + Klpc)(l + CPCONxdFUEL) (5.11)

 

 

LH .—.— K01h1a2 x PARTS“ ) + K1 ,h (5.12)
R1 = max (1R1,min (IRIO,a4 + a5 * 1R1"6 )) (5.13)
a4 = [RIO—a7
a
_ 7
“5 ‘ IRIO
[RIO a7
[RIO (5.14)
as =
a7
a7 = [RIO—3
11210 = 3
86

Where,

PARTS = Standardized parts consumption as a fraction of the replacement vehicle
price per 1000 km

CPCON = Congestion elasticity factor (default = 0.1)

dFuel = Additional fuel consumption due to congestion as a decimal
Kopc = Rotational callbration factor (default = 1.0)

Klpc = Translational calibration factor (default = 0.0)

30 and 31 = Model parameters (Table 5.2)

LH = Number of labor hours per 1000 km

KOIh = Rotation calibration factor (default = l)

Kllh = Translation calibration factor (default = 0)

32 and a3 = Model parameters (table 25)

IR] = International Roughness Index in m/km

IRIO = Limiting roughness for parts consumption in IRI m/km
a4 to a7 = Model parameters

5.2.5 Summary and Conclusion
Based on the detailed information documented above, we selected the latest
mechanistic-empirical models, i.e. those in the HDM 4 model, for further analysis and
evaluation. Also, the latest comprehensive research conducted in the US, i.e.
Zaniewski’s tables/charts, will be updated to current conditions and used as a basis for

Comparison to US conditions.

87

Table 5.2 HDM 4 Maintenance Model Parameters [17]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Parts consumption model Labor Model
Vehicle Type
CKM kp a0* lE-6 a1*lE-6 a2 a3

Motorcycle 50000 0.308 9.23 6.2 1 161.42 0.584
Small car 150000 0.308 36.94 6.2 1161.42 0.584
Medium car 150000 0.308 36.94 6.2 1 161.42 0.584
Large car 150000 0.308 36.94 6.2 1161.42 0.584
Light delivery car 200000 0.308 36.94 6.2 61 1.75 0.445
hgh.‘ gOOdS 200000 0.308 36.94 6.2 611.75 0.445
vehicle

four wheel drive 200000 0.371 7.29 2.96 61 1.75 0.445
light truck 200000 0.371 7.29 2.96 2462.22 0.654
medium truck 240000 0.371 1 1.58 2.96 2462.22 0.654
heathruck 602000 0.371 1 1.58 2.96 2462.22 0.654
articulated truck 602000 0.37] 13.58 2.96 2462.22 0.654
mini bus 120000 0.308 36.76 6.2 611.75 0.445
light bus 136000 0.371 10.14 1.97 637.12 0.473
medium bus 245000 0.483 0.57 0.49 637.12 0.473
heavy bus 420000 0.483 0.65 0.46 637. 12 0.473
coach 42000.0 0.483 0.64 0.46 637. 12 0.473

 

 

5.3 CALIBRATION

MODEL

5.2.1 Data Collection

OF THE REPAIR

[23], and Texas and Michigan DOTS.

88

AND MAINTENANCE COSTS

5.3.1.1 Michigan Department of Transportation (MDOT) data
MDOT Pavement Management System extracts pavement surface distresses (type,

Pavement condition, vehicle characteristics and repair and maintenance costs data

have been collected from different sources which include NCHRP 1-33 (Truck ﬂeets)

extend, and severity) from video images of the pavement surface bi-annually. The raw

proﬁle is collected by a high speed proﬁlometer. Surface distresses and proﬁle data of

 

all pavement types (rigid, ﬂexible, and composite) are used to automatically compute
Distress Index (DI) and Ride Quality Index (RQI), for a minimum segment length of
161 m (0.1 mile). The bi-annual change of the pavement’s D1 is included in a
performance model to estimate the pavement’s RSL. RQI reﬂects the user’s
perception about pavement ride quality. Figure 5.1 shows a map view of the eight
district of Michigan. Figure 5.2 shows the distribution of roughness (RQI) in the eight
regions of Michigan. According to [72], RQI and IRI have a good correlation, with the
RQI increasing asymptotically with increasing IRI to a plateau value of about 100 as
the IRI-values approach the 4 to 5 m/km range. The IRI-value corresponding to an
RQI of 70 (is about 2.4 m/km (threshold for rehabilitation in Michigan). Figure 5.3

shows the percent of sections with IRI > 2.4 m/km.

 

 

University

Figure 5.1 Michigan Regions

89

  

 

   

100 T
g. 8.1
E 60 *-
3 1
g 40
L“ 20 1
0 1——1l11 ~ '+-+-+
10 40 70 100
RQI
(a) Superior region
60 . .
1,, 501
O -
4° 1
g 30 1
.. :31 | . I
0 . 1 .11... 2.
10 40 70 100
RQI
(c) Grand region
80 T
‘3‘ 6O 1.
s 4., --
E
o 41—4le -' 1 1%--+—1
10 40 70 100
RQI
(e) Southwest region
15 T
E 10
“=3 1
E 5 1
0 1 +_ a1
10 40 70 100
RQI
(g) Metro region

5* 150 ‘1
5 1
E. 100
u- 501- I
f . 0 1 , 1‘1—1— Rwi
More 10 40 70 100 More
RQI
(b) North region
40 T
.>,~ 30 -1--
5 1
g. 20 =
3 1
“‘ '° 11111
1T1 o 1 ' 1 - -~; 1411—47—1
More 10 40 70 100 More
RQI
(d) Bay region
50 '1
>. 401
U
5 30 '1
:3
g 20 1
LL.
10 1 11
More 0 --—~+ 1» 1 1 -1*~+-——¢-——1
10 40 70 100 More
ROI
(f) University region
More

Figure 5.2 Distribution of Roughness for Michigan Regions (Michigan DOT)

90

60% .

 

 

50% 1”

40% —;——

 

 

30%

20% _,

10% —

% of sections with IRI >2.4 m/km

 

 

0% —
superior north gand bay southwest university metro
Region
Figure 5.3 Percent of Sections with IRI > 2.4 m/km
Figure 5.4 shows the distribution of repairs by region for different vehicle classes.

average labor and parts costs versus roughness for different vehicle classes for

Michigan. It should be noted that the graphs were corrected for mileage.

Figure 5.5 shows the average labor and parts costs versus roughness for different
vehicle classes for Michigan. It should be noted that the graphs were corrected for
mileage. A data point in the graphs represents the average labor hours or parts costs

for all the vehicles within a district for each vehicle class.

91

4000 _

3500 ‘*

3000“

2500 4

 

2000

1500 1

Number of vehicle

    

9' '

1000“

500“

 
    

O

\

O

/A:. .

 

1111

southwest university metro

 

superior north

Regions

 

 

1. Passenger Cars Light Trucks 51 Medium Truckﬁ 1

 

 

Figure 5.4 Distribution of Repairs by District and Vehicle Class (Michigan DOT)

5.3.1.2 Texas Department of Transportation (MDOT) data

The state of Texas has 25 congressional districts. Figure 5.6 shows a map view of
Texas districts. Figure 5.7 shows the average roughness measured during the ﬁnancial

year 2006 for each district of the state of Texas.

92

 

0 Parts 0 Lab or

 

 

 

 

 

 

g $600 — s— —:—— ~1— $400
a $500 7"” "*7 "— ”“1_ 9' ' “Fr ’1 a
g $400 A---—-+— ~—-~— -—s~—. —— 1~- ~11 $300 *g
g ‘33 $300 T” 'TTT 'TbTo T '53 ””1” $200 L2;

$200 ____ 1.“- __11'_____ _:_______._____1 $100 '3
[$1 $100 q, 0 1 1.1
12.. $0 1— .; -.8.4_q9_4_s+_ ,1 $0
0.00 0.50 1.00 1.50 2.00 2.50
IRI (m/Km)
(a) Passenger Cars
$2,500 1 “ T T " "'T’ " __ 1 TT T TTT $2,500
.‘ 1 o 1 1
$2,000 1T T 7 ___. "To T "TTT ”T" $2,000

 

Parts Consumption (S)

 

0.00 0.50 1.00 1.50 2.00 2.50 3.00

 

lRI(m/Km)
(b) Light Trucks
g $5,000 7" o‘ (.3——*—" T1- $5,000 A
Ea $4,000 1' h 1 0'77" ‘ -, — 1. $4,000 ?
g 13,000+ - 1g _- _,. $3,000 51
W 1 W: .1.
"’ 1,0001—5—M 1 h ”T 1,000 a:
E $ 1111---3984mQ; .
0.00 1.00 2.00 3.00
IRI(m/Km)
(C) Heavy Trucks

Figure 5.5 Summary of the repair and maintenance data (Michigan DOT)

93

Labor Cost ($)

It should be noted that there is 100% roadbed coverage that includes all state
maintained road network, i.e. interstate, national and state road and farrn-to-market
network. Figures 5.8 through 5.10 show the distribution of repairs by district (Texas)
for different vehicle classes. Figures 5.1] and 5.12 show the average labor and parts

costs by district for different vehicle classes. It should be noted that the graphs present
the corrected data for age and mileage. A data point in the graphs represents the

median values of labor or parts costs for all the vehicles within a district for each

vehicle class.

 

Figure 5.6 Texas Districts

94

 

 

 

 

’51
1
1
€1.51
E 1
.1. ~
.. l

 

1 ‘1’ 7'

 

Til

 

 

 

 

 

 

 

 

.VL

 

 

 

11

 

1

12 3 4 5 6 7 8 910111213141516171819202122232425

Figure 5.7 Distribution of IRI by District for Texas (Texas DOT)

 

Number of Passenger cars

 

District code

 

 

12 3 4 5 6 7 8 910111213141516171819202122232425
Districtcode

Figure 5.8 Distribution of Repairs by District for Passenger Cars (Texas DOT)

95

700 J———:~;~w __-__ ._.w.

 

 

§

”s"

‘9’

number of Light Trucks
.h
8
1
1
1
1
1
1
1
1
1

 

 

 

 

 

§

§

 

O
1

 

”Milli M u

12 3 4 5 6 7 8 910111213141516171819202122232425
Districtcode

(a)

 

180

160 a

number of Medium Trucks
8 8 8 8 '8 8

20~

 

 

12 3 4 5 6 7 8 910111213141516171819202122232425
Districtcode

(b)

Figure 5.9 Distribution of Repairs by District for (a) Light and (b) Medium trucks

(Texas DOT)

96

 

200
180 q
160 ~
140 4
1201 ,
100 4

80‘

number of Heavy Trucks

60a

40 «

20~

 

 

12 3 4 5 6 7 8 910111213141516171819202122232425
Districtcode

((1) heavy trucks

 

20

18~

16a

144

12a

10«

number of Articulated Trucks

 

 

12 3 4 5 6 7 8 910111213141516171819202122232425
Districtcode

(e) articulated trucks
Figure 5.10 Distribution of Repairs by District for (a) Heavy and (b) Articulated trucks
(Texas DOT)

97

 

OPaits OLabor I

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$8001i1 1 1 01 155800
6 $600 1 ’ c710611 $600 53
+5 1 1 *3
8 $400 ’3 609 2' 31 ~—1—$400§
:3 . 06°.» ‘1 1 2
94 $200 0086) 601$ cg) —-~‘- $200 3

$0 1——-~~~a—ﬁ 1 41 ‘9": --—:— $0
1.40 1.60 1.80 2.00 2.20 2.40
IRI (In/km)
(a) Passenger cars
$1,000 1% . 1 $1,000
_?._ﬁ.____.__.___*__Q- ________.r_ A
a $8001 1 , 1 $800 2;;
8 $600] 166 O 900 $600 8
.3 $400 T—ﬁmge‘o’gm“ -- $400 3.
‘3‘ $200 1~—61§%- 11647 $200 3
. 1 e 01 0 2
$01-— 1 r . —+ $0
1.4 1.6 1.8 2 2.2 2.4
IRI(m/km)
(b) Lighttrucks

$1,200 “’ ~ ~ | $1,200

$1,000 — . - ‘FUTT‘ : 00—7? $1,000 g
g $8001~wrr0‘ +~o - 1- $800 .3

1 , .
{.3 $600 ew— -- 6°08 08 8 —~1~ $600 §
6. $400 6*00 61°03 ' Obj $400 ,8
‘39 6 o , a
$200 *' 1 .SW“ 966” 1 $200 *4

$0 +- T ° 1 f 1.1—“~30
1.4 1.6 1.8 2 2.2 2.4
IRI (m/km)

(c) Medium cars

Figure 5.1] Summary of the Repair and Maintenance Data for Light and Medium
Vehicles (Texas DOT)

98

 

 

0 Parts 0 Labor

 

 

 

 

 

 

$1,000 " *‘1 ""’—1-—‘ ‘1——"‘* $1,000
a; $800 1__ '—-‘7 “9'“6‘" 1 $800 a
T”; $6001 1 a 01 ()01 $600 ‘9
m ' o __~._ '0
8 1 8 43 01> . 1m 3’: . 8
$400 ~—-~ -- 1 $400 15
g 1 g (ﬂ 0 O 1 1 '8
m $200 ,1___,__-: :60 6%"; 1 $200 ._:
1 B 0 1 1
$0 —~:———.+-—-—-—~'——- 1 ——~-—~~-——1— $0
1.4 1.6 1.8 2 2.2 2.4
IRI (m/km)
(a) Heavy trucks
$2,000 T—MT " "—— ______ $2,000
893 $1,5001 . 4 67 ~ 1 : 1$1,500 ‘13
8 $1,000 ‘T"—_*"‘—"““°"”§“ga‘—1‘ $1,000 8
1 8o .. I :g
a. $500 1 8+eo§g§e¢gm --1 $500 3
I 8 ‘
I O@ O 3
$0 r— 1 O 1 89: ~+: $0
1.4 1.6 1.8 2 2.2 2.4
IRI (In/km)
(b) Articulated trucks
$800 3 ‘ ‘1 1 1 *O' *1 $800
A 1 1 1 1 1 A
8’; $600 1 * ' ‘4 o """"" ‘ $600 ‘23
+3 1 O 1 1 1 1 g
8 $4001 ~ 101 1 a $400 8
O 1 o O ‘5
a, $200 1 1' ‘5‘ o 0““ ‘f i $200 3
1 . I 1
1 .' 1 1 1 1
$0 ——= -— 61— 5 — 1 ——» 8+ $0
1.4 1.6 1.8 2 2.2 2.4
IRI (m/km)
(c) Buses

Figure 5.12 Summary of the Repair and Maintenance Data for Heavy Vehicles (Texas
DOT)

99

5.3.1.3 NCHRP 1-33 data
Data from previous research conducted as part of NCHRP 1-33 project were also

obtained. It should be noted that the data is only for heavy trucks. Figures 5.13 shows

the raw data and Figure 5.14 show the corrected data for mileage and age.

 

 

 

 

 

 

 

 

14000
“
.0 120001-
|— O
E 10000— 9 .
E O O
E 8000 1 ,
a . '
5 6000 - o t
0 ° : 9
U 0 . Q
i 40001 i g 1 ,
3 0 o
g 2000 1— o o
O
0 1
O 1 2 3 4 5 6 7
AGEyrs
(3) Annual parts cost versus age
180
160 «1
g O
O
n: 140 11 o
8 o
5 120 1 o 3
2 3 ° 0 3
6 ° 3 '
80 -~ 0 3
«'3 ’ 0 °
0
8 60 11 3 ° 0
. O
3 4o 1
5:
20 1
0 . - a
0 1 2 3 4 5 6 7
AGE (yrs)

(b) Annual labor cost versus age
Figure 5.13 NCHRP 1-33 Raw Data [23]

100

COST OF PARTS $11000km

LABOR hrs/1000 km

50

 

45 1
4o -1
35 1
30 1
25 —~
20 1
15 1
1o .,

 

O

0
00M.”

00

 

 

1.2

1.2

1.3 1.4 1.5

1.6

1.7

ROUGHNESS IRI mlkm

(c) Cost of parts ($/1000km)

1.8

1.9

 

1.0-

0.8

 

000. .0. .00 O

O

 

 

Figure 5.14 NCHRP 1-33 Data Corrected for Age and Mileage [23]

1.3 1.4 1.5

1.6

1.7

ROUGHNESS IRI mlkm

1.8

((1) Labor hours per 1000 km (hr/1000km)

101

0

1.9

5.2.2 Assessment of Data Applicability

Based on the data obtained (Figures 5.1 through 5.14), the number of samples (for
all vehicle types) is adequate for making statistical inferences. The MDOT ﬂeet does
not include heavy and articulated trucks (for normal operations). However, this gap is
offset by the Texas DOT and the NCHRP 1-33 data. Therefore, the quantity of data is

adequate.

The quality of the data in the context of this analysis is related to the question of
whether repair and maintenance costs could be tied to pavement conditions (i.e.,
roughness). A crucial criterion for establishing such relationship is that there would be
enough variance in roughness conditions among different regions/districts.
Preliminary analysis of pavement roughness data in Michigan and Texas suggests that
there is enough variability in roughness conditions among different regions, districts
and counties (Figures 5.2, 5.3 and 5.7). For example, one could classify the regions of
Michigan into 3 categories (Figure 5 .2):

0 University and Metro regions (~50% of sections with IRI>2.4 m/krn)
0 Bay and Southwest regions (~30% of sections with IRI>2.4 m/km)
0 Superior, North and Grand regions (~20% of sections with IRI>2.4 m/km)

Similarly, the state of Texas could be grouped into ﬁve groups (Figure 5.15):

0 Group A: roughness > 2.1 mlkm
0 Group B: 1.9 mlkm < roughness < 2.1 m/krn
0 Group C: 1.7 m/km < roughness < 1.9 m/km

0 Group D: 1.5 m/km < roughness < 1.7 m/km

102

0 Group E: roughness < 1.5 m/km
This classiﬁcation was obtained using the average roughness of 1.9 m/km and a

standard deviation of 0.2 mlkm.

-111 11111 1

216191412 5 7 4 9 2520213 3 231510 81716241221118
Districtcode

 

 

 

 

 

 

Figure 5.15 Categories of Roughness Considered in the Statistical Analysis (Texas
Districts)
Repair and maintenance data from vehicle ﬂeets as reported in NCHRP 1-33 were
correlated with pavement condition (IRI) and compared with HDM-4 predictions
(Figure 5.16). It can be seen that:

0 The range of IRI in the NCHRP 1-33 is quite restricted.

0 The parts consumption according to NCHRP1-33 is lower than the
predictions by HDM-4.

0 The labor hours according to NCHRP 1-33 are much lower than those
predicted by HDM-4.

103

 

 

0.004 1
330 E 1
g o 0.003 . o
3 8 0002 -1
>3 1 ‘t
O _.-
a o 0.001 1 .
0 T' 1 T . ‘ 1— ”‘1
O l 2 3 4 5
o NCHRP 1-33 —e—HDMIV1
(a) Parts consumption
5 20 1
O
O 1
g 15 W
3.5, 1
£2 10
8 1
e 5 f
o .
3 0 71—77 ~ t“ - 1 -~1 1
O 1 2 3 4 5
lRI(m/Km) __

 

( _.- NCHRP 1-33 -B—HDM 1V1

 

(b) Labor hours
Figure 5.16 Comparison between HDM-4 Predictions and Data from Truck Fleets
(NCHRP 1-33)

The predictions from HDM-4 do not appear to be reasonable for US conditions.
These observations could be explained by the fact that HDM—4 model was calibrated
using data from developing countries (e.g., Brazil, India). It is well known that the
labor hours in those countries are much higher than in the US. Also, the difference

between parts consumption in the US and those predicted from HDM-4 could be

104

explained by the inﬂation in the parts and vehicle prices. These problems will be
overcome by calibrating the model to US conditions.

Eventhough the quantity of data is adequate, no information on whether there is a
relationship between roughness and repair and maintenance could be extracted. The
following issues were encountered:

0 The scatter plots shown in Figures 5.5 and 5.11 through 5.14 were wide
because of the high variability of the data.

0 The range of roughness is narrow (between 1.4 and 2.4 m/km) and does not
cover the full range in the US (~ 1 to 5 m/km, Appendix A).

0 The maximum roughness level observed from the collected data is less than 3
m/km. Previous study showed that there is no effect on repair and
maintenance costs at low level of roughness (less than 3 m/km) [17, 24].

For the most part, it is expected that the bulk of any correction/calibration to match
US conditions could be achieved using macro-economic model corrections based on
overall (average) economic data (e.g., average labor hours for typical vehicles and
average parts cost comparisons). Therefore, in this research, the latest comprehensive
research conducted in the US, i.e. Zaniewski’s tables/charts, was updated to current

conditions.

5.2.3 Updating Zaniewski et al. Tables

The latest comprehensive research conducted in the US [11] for VOC components
did not involve the development of models (i.e., it only involved updating Winfrey’s
tables [9]). Zaniewski et al. [11] costs for maintenance and repair at constant speed at

level terrain in good condition were updated by multiplying Winfrey’s costs by the

105

ratio of current overall maintenance and repair costs to Winfrey’s overall costs.

Similar procedure was be used in this study.

Current overall Repair and Maintenance (R&M) costs were estimated using MDOT

and Texas DOT databases. The DOT data was ﬁrst sorted by car make, model and

year. Then, only repair costs related to damage from vibrations were extracted (e.g,

underbody inspection, axle repair and replacement, shock absorber replacement). The

inﬂation rate between 1982 and 2007 is calculated as the ratio of current overall

average R&M costs to those reported in the tables by Zaniewski et al. Table 5.3 recalls

Zaniewski et al. costs, current R&M costs and the inﬂation rate for different vehicle

classes. Zaniewski et al. tables were updated by multiplying their reported costs by the

inﬂation rate R&M costs. Appendix B presents the updated cost tables.

Table 5.3 Repair and Maintenance Costs and Inﬂation Rates

 

 

 

 

 

 

 

 

 

 

 

Avera e

Vehicle [SEE/711.3%?) if? Inﬂation véillzigie e vehiclge

Class Zaniewski et Estimated rate ( ) g mileage

al. [11] average cost yr (miles)
Small car 34.3 64.36 1.88 9.23 96215
Medium car 41.6 64.36 1.55 9.23 96215
Large car 48.04 64.36 1.34 9.23 96215
LT 52.81 82.85 1.57 7.80 86963
MT 145* 92.13 0.64 7.40 87449
HT 145* 190.83 1.32 12.50 196378
AT 145* 198.85 1.37 14.64 352633
buses - 190.59 - 22.75 323174

 

 

 

 

 

 

* Assumed equal to Heavy Truck cost

To include the effect of pavement condition, information from Brazilian study

(HDM 3 model) were investigated and adjustment factors were computed by

106

 

Zaniewski et al. In the Brazilian study, two different relationships were established to
estimate parts and labor expenses as a function of surface roughness. A similar
approach was used in our analysis. First, Equations 5 .11 through 5.14 of the HDM 4
repair and maintenance costs were compiled for all vehicle classes for ranges of
roughness levels. Then, these values were compared to the baseline condition which is
assumed to be 2 m/km (”z 3.5 PS1) because Zaniewski’s tables were generated for 3
PSI of 3.5. The change from the baseline condition (3.5 PSI z 2 mlkm) to the
respective roughness level was reported as the adjustment factor applied to the
updated costs. These adjustment factors were calculated using equation 5.15.

AF. = C0ST(IRIZ.)
l COST(2) (5,15)

 

COST(IRIi) = PARTS(IRIi)+LH(IRIi)

where:

AF i( %) = Adjustment factor in percent

PARTS (IRII) = Standardized parts consumption as a fraction of the replacement
vehicle price per 1000 km evaluated at IRIi (Equation 5.11)

LH (IRA) = Labor cost per 1000 km evaluated at IRI," (Equation 5.12)

IRI i = International Roughness Index in m/km

The HDM 4 model suggests eliminating the effects of roughness on parts
consumption at low IRI (less than 3 m/km). This is achieved by using the smoothing

relationship given in Equation 5.14. This assumption was reported in many studies.

107

Also, it was observed from the data that was collected as part of this study. Table 5.4

presents these factors. Figure 43 summarizes the methodology described above.

Table 5.4 Change in Repair and Maintenance Costs as a Function of IRI

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Adjustment Factors
Vehicle Class IRI
(In/km)
l 1.5 2 2.5 3 3.5 4 4.5 5
Small car 1 1 1 1 1.01 1.06 1.11 1.17 1.22
Mediumcar 1 1 1 1 1.01 1.06 1.11 1.17‘ 1.22
Large car 1 l 1 1 1.01 1.06 1.11 1.17 1.22
Pick up trucks 1 1 l l 1.01 1.06 1.11 1.17 1.22
light truck 1 1 1 1 1.02 1.09 1.18 1.27 1.37
medium truck 1 1 1 1 1.01 1.07 1.14 1.22 1.29
heavy truck 1 1 l 1 1.01 1.07 1.14 1.22 1.29
articulated truck 1 1 1 1 1.01 1.07 1.13 1.20 1.26
Average
/ Roughness /
J ‘5 11
Tables Develop Time 3 /
and —' Series Trends ' Update “’ 0’8 0
Charts and

 

 

 

 

 

 

 

 

 

‘ “ 19691382 260'7

Time (years)
Data from DOTﬂ eet/ Previous /
Economic analysis Tables/Data

Figure 5.17 Approach for Updating Zaniewski et a1. Tables

 

 

 

 

 

108

5.4 SUMMARY AND CONCLUSIONS

The analysis of the repair and maintenance data of Michigan and Texas DOT vehicle
ﬂeets showed that the predictions from HDM-4 do not appear to be reasonable for US
conditions. These observations could be explained by the fact that the HDM-4 model
was calibrated using data from developing countries (e.g., Brazil, India). It is well
known that the labor hours in those countries are much higher than in the US. Also,
the difference between parts consumption in the US and those predicted from HDM-4
could be explained by the inﬂation in the parts and vehicle prices. Therefore, it was
recommended to use the updated Zaniewski’s repair and maintenance costs (i.e., the
latest comprehensive research conducted in the US). These costs were updated by
multiplying their reported costs by the inﬂation rate of R&M costs between 1982 and
2007.

The outcome from this analysis will be a model that will be used to estimate repair
and maintenance costs at the network level. At the project level, the effect of localized

roughness events should be included. This analysis is presented next.

109

CHAPTER 6
IDENTIFICATION OF PAVEMENT ROUGHNESS EVENTS

6.1 INTRODUCTION

The excitation of road vehicle can be described by environmental parameters, such
as, road proﬁle, road curvature, topology, and driver behavior such as, speed changes
and maneuvering. Of these, the most important parameter for the durability of most
components is the road proﬁle. Transient events in the road proﬁle do have a
signiﬁcant impact on vehicle durability and damage to goods, and it is therefore
necessary to know about their magnitude and extent. This chapter discusses method to

extract roughness features through the use of the raw proﬁle data.

6.2 IDENTIFICATION AND CHARACTERIZATION OF PAVEMENT
ROUGHNESS FEATURES

The focus of this study is the detection of roughness features that may have an impact
on vehicle durability and damage to goods. These features are
0 Faulting, Breaking and Curling for concrete pavement, and

0 Potholes for asphalt pavement.

6.2.1 Pavement Faulting

Faulting of the joints/cracks is deﬁned as the difference in elevation of the two

adjacent slabs before and after the joint/crack, as shown in Figure 6.1. The Pathway

110

laser proﬁler (MDOT’s current contractor for video and laser data collection) collects
the elevation of the proﬁle at every 19 mm or 0.75 inches (the sampling interval).
These measurements are averaged every 76 mm or 3 inches and recorded (the
recording interval). These proﬁle samples are taken for leﬁ and right wheel paths and
the center of the lane.

Theoretically, when the samples of the proﬁle are collected over a faulted
joint/crack, the difference in elevation should appear in two adjacent samples.
However, from the MDOT proﬁle data the faulting does not appear in two adjacent
points. Faulting of the slabs in the actual proﬁle appears at several sample points, as

shown in Figure 6.2.

Sampled Points
Direction of trafﬁc

9 (Liz N

 

Approach Slab l!
Leave Slab

Figure 6.1 Theoretical Samples from Faulting of Slabs

 

 

l
60 l 2
4o 4 / i
20 - l
-20 1
40 4 ‘
-60 2
-80 r . I

0 50 100 l 50 200
Distance (m)

Elevation (mm)

 

 

 

Figure 6.2 Faulting from Actual Proﬁle

111

In order to explain such phenomenon, a comprehensive analysis using a dummy
proﬁle was performed. The dummy proﬁle was created to include the following:
0 Long wave sinusoid that represents the topography of the site.
0 Short wave sinusoid that represents the curling of the slabs.

0 Discontinuous line that represents the faulting of the joints/cracks.

Figure 6.3 shows the components of the dummy proﬁle listed above. The dummy
proﬁle is the sum of these three functions and is shown in Figure 6.4. A sample

interval of 76 mm (3 inches) was used to meet the current sampling interval used by

MDOT.

 

150
100
50

1

l

i

-50
-100 r
-150

 

Elevation (mm)
o

 

 

 

0 50 100 l 50
Distance (rn)
— Curling — Faulting —— Topography

Figure 6.3 Components of the Created Dummy Proﬁle

112

 

 

 

 

200
E 150 r
V 100 r "
.5 50*
a; 0 1
g. -50 .
m -100 r
-150 1 1
0 50 100 150
Distance (m)
—Dummyproﬁle

Figure 6.4 Dummy Proﬁle

In the dummy proﬁle, faulting was created by a signiﬁcant difference in elevation
of the adjacent samples as shown in Figure 6.5. Also shown in the ﬁgure is a proﬁle
with faulting that is similar to the actual proﬁle data (faulting appears within several
sample points). This proﬁle was obtained by passing the dummy proﬁle through a
moving average ﬁlter with a base length of 300 mm (1 ft). The moving average ﬁlter
is a low-pass ﬁlter that smoothes the proﬁle and is a built—in ﬁlter of the proﬁlometer
in most cases. The number of sample points to show a faulting is highly affected by

the type of the low-pass ﬁlter and its base length. A more detailed discussion will be

presented in section 6.4.

 

80!
601
40~

20‘

 

Elevation (mm)

 

 

 

O 1 1 - 1
4.5 5 5.5 6 6.5 7 7 .5
Distance (m)

+ Dummy proﬁle + Averaged proﬁle

Figure 6.5 Faulting of Dummy Proﬁle

113

6.2.2 Pavement Breaks
A pavement break is a broken portion of the pavement section that starts with a
negative fault and ends with a positive fault. Its important characteristics are:
0 Magnitude of the fault at the start point,
0 Magnitude of the fault at the end point, and,
0 Its extent.
6.2.3 Curling
Curling is the distortion of a slab into a curved shape by upward or downward
bending of the edges. This distortion can lift the edges of the slab from the base
leaving an unsupported edge or comer which can crack when heavy loads are applied.
Sometimes, curling is evident at any early age. In other cases, slabs may curl over an

extended period of time [25].

6.2.4 Potholes

A pothole is a type of disruption in the surface of a roadway where a portion of the
road material has broken away, leaving a hole. Most potholes are formed due to
fatigue of the pavement surface. As fatigue cracks develop, they typically interlock in
a pattern. The pavement portion between fatigue cracks becomes loose and may
eventually be picked out of the surface by continued wheel loads, thus forming a
pothole. The formation of potholes is exacerbated by cold temperatures, as water
expands when it freezes and puts more stress on cracked pavement. Potholes can grow
to feet in width, though they usually only become a few inches deep, at most. If they

become large enough, damage to tires and vehicle suspensions can occur.

114

6.3 EVALUATION OF EXISTING METHODS OF PROFILE ANALYSIS

Although considerable information on proﬁle analysis is documented, only few
studies that discuss diagnosing/localizing distresses from the actual proﬁle data
identify the type of distresses. These methods include: Power Spectral Density,
Moving average, Joint time frequency analysis, wavelet-based analysis. The most

important concepts are discussed brieﬂy on the following paragraphs.

6.3.1 Time domain analysis

In [26], Byrum developed an algorithm that picks up the “imperfection” zone from
the proﬁle of rigid pavements. These zones are then separated from the slab region.
However, the method does not identify the type of the individual distresses that are
present. The detection of the imperfection zone is done by calculating the curvature of
the proﬁle and applying a threshold on the curvature. The threshold value for CVT is
between 10 mm'1 and 49 mm". The curvature variation threshold (CVT) is calculated
by the following equation:

1

C VT = 1—2-[——O.03242 (06C VStDev)0'8405

+ 0.00665 (24C VStDev)l ' 1446]

1 (6.1)

—EI:0.06631(48CVStDev)3 7819]

(0.003 < CVT < 0015 ft”; 1 ft = 0.3048 m); R2 = 0.81

115

Where,

06CVStDev = Standard deviation of the 152.4-mm curvatures in 152.4 m proﬁle,

1/ft *1000

24CVStDev = Standard deviation of the 609.6-mm curvatures in 152.4 m proﬁle,
1/ft *1000

48CVStDev = Standard deviation of the 1219.2-mm curvatures in 152.4 m proﬁle,
l/ft *1000

Fernando and Bertrand [27] used moving average ﬁlters to detect localized
roughness. The points where the deviations from the averaged proﬁle are large are the

rough areas. However, this method does not identify roughness features.

Chang et al. [28] identified localized roughness based on the Texas Department of
Transportation Speciﬁcation Tex-lOOI-S. This method was ﬁrst used by [27]. First,
each elevation point from the two longitudinal proﬁles (left and right wheel paths) is
averaged to produce a single averaged wheel path proﬁle. Then, the resulted. proﬁle is
placed on a 7.5 m (25 ft), centered-moving average ﬁlter. The difference between the
average wheel path proﬁle and the 7.5m—moving average ﬁltered proﬁle for every
proﬁle point is computed. Deviations greater than 3.8 mm (0.15in) are considered a
detected area of localized roughness. Positive deviations are considered as "bumps"
and negative ones as "dips". However, this method does not identify roughness

features.

The current method used by MDOT is the Pathway procedure. First, for each of the

longitudinal proﬁles, the differences in height 76 mm (3 inches) apart are taken and

116

the variance of the differences of the heights is calculated for each point. Then, a
moving average ﬁlter is applied to the calculated variances and compared with the
actual elevation differences. The base length for the variance and moving average
calculations is 6 m or 20 ﬁ (10 ft before and after). This procedure is repeated for all
three longitudinal proﬁles (left and right wheel paths and the center of the lane), and if
the elevation difference is greater than the averaged variance for all three longitudinal
proﬁles, then the point is classiﬁed as faulting. Figure 6.6 shows a dummy proﬁle (see
section 6.2.1 for more details about dummy proﬁles) and the faulting detected by the
Pathway’s procedure. Note that this is only for a single longitudinal proﬁle. The

procedure detects the presence of faulting but in a wide range rather than at a certain

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

point.

2 200
Cf r 150
3 _ 100 E
E 50 E“
$1400000<19<wa90’ 00092.0 g
m “:3
a 0)
3g * -50 Ej
g3 - -100

O -150

0 50 100 150

Distance (m)

° Faulting(y/n) — Dummy proﬁle

Figure 6.6 Faulting Detected by Pathway’s Procedure

117

6.3.2 Power Spectral Density Method

The frequency content and amplitude of vehicle-induced vibrations are the most
important factors in determining the ride quality, as experienced by passengers.
Frequency characteristics of vehicle vibrations are directly related to the frequency
spectrum of the pavement proﬁle through a transfer function of the vehicle. A
pavement proﬁle constitutes various features of different wavelengths, ranging from
pulse like disturbances of a short wavelength (high frequency), to long wavelengths of
the road geometrical curve (low frequency). Power spectrum density (PSD) analysis is
commonly employed to obtain the distribution of proﬁle energy between its
constituting wavelengths.

PSD analysis provides more information about the proﬁle roughness characteristics
than any roughness index and is one of the best available diagnostic tools for
interpreting pavement properties. PSD functions reveal the general level of roughness,
whether the proﬁle is dominated by long or short wavelengths. It also can reveal
whether the energy of the proﬁle affecting rideability is concentrated over certain
wavebands [29]. This information is helpful in interpreting pavement proﬁle
characteristics. Roads with dominant long wavelengths may have poor foundations,
while roads with dominant short wavelengths tend to have cracks, patches, slab
faulting, and other forms of short lived surface disturbances contributing to roughness
[30].

PSD analysis was originally developed to characterize random data. For nonrandom
proﬁle data, only an estimation of PSD can be calculated [31]. One approach to

improve the estimation of PSD amplitudes is based on reduction of the edge effects.

118

This can be achieved by ﬁltering the proﬁle to exclude low frequency geometry trend.
Large amplitudes associated with a road geometrical curve can obscure a lower
amplitude high-frequency portion of the spectrum that contributes more to roughness.
A number of techniques have been developed and used for removing the effect of the

geometry trend.

6.3.2.1 PSD Analysis

Any pavement proﬁle can be broken down into a linear combination of sinusoids of
different frequencies (wavelengths), amplitudes, and phases. An arbitrary N-point set
of proﬁle elevation points p(n) obtained at sampling interval A, can be represented by

a Fourier series as [32]:

p(n) = A0 + A1 cos(27ran + (01) + A2 cos(27r(2f)nA + (p2) + . ..

(6.2)
+ Ak cos(27r(lg")nA+¢k) +...
Where,
p(n) = The proﬁle elevation of the n point
f = The fundamental frequency of the proﬁle (reciprocal of the total length
of the proﬁle)
k = An arbitrary integer
A k and (p k = The amplitude and phase angle at frequency If
The series can also be written in the following equivalent form [32]:
A k cos(27r(lgf)nA + (pk) = a k cos(27£(lgf)nA) + bk sin(27r(kf)nA)
(6.3)

ak = Ak cos(¢k), bk 2 —Ak sin((ok), a]: +b/‘3 = A]:

119

Fourier transform is commonly employed to calculate the amplitude and phase angle
of each frequency component. Coefﬁcients ak and bk are, to within constants, real and
imaginary parts of the Fourier transform. Relating the above series to usual frequency
characteristics gives [32]:

Mean square at frequency [if = A]?

(6.4)
PSD estimate at frequency lg” = A]? / 2f

Plotting PSD amplitudes of the proﬁle versus frequency or wavelength provides an
estimate of energy distribution of the proﬁle among various wavebands. PSD
estimation can be computed for elevation, proﬁle slope, or acceleration.

PSD analysis has been used to ﬁnd out what wavelength range contributes most to
the road roughness in several research experiments [32, 33, 34]. Commonly, root
mean square (RMS) values are calculated for proﬁle (or slope) PSD amplitudes over
various third octave frequency bands. These RMS amplitudes are then correlated to
the corresponding mean panel ratings (MPR) obtained from a controlled panel study.
Therefore, a range of frequencies giving the best correlation with MPR can be
obtained. A pavement proﬁle consists of a low frequency trend coming from the
geometry of the road superimposed by high frequency roughness ﬂuctuations. The
geometry trend in the raw proﬁle taken by reference proﬁlers (e.g. rod and level
survey, walking proﬁler, high seed proﬁlometer) is normally of an order of magnitude

higher than the roughness ﬂuctuations. Therefore, the variation in the proﬁle

120

roughness is obscured by the road geometry curve and, therefore, cannot be
recognized.

Furthermore, from the computational standpoint, the geometry curve introduces
artifacts in the estimation of PSD functions, which are calculated using Fourier
transforms. One of the underlying assumptions in the Fourier analysis is the
periodicity of the analyzed proﬁle. The artiﬁcial periodicity for a non-periodic data
set, such as a measured proﬁle, is achieved by repeating the original data points.
Presence of a geometry curve introduces ’large discontinuities in the artiﬁcially
periodic proﬁle that result in artifacts in the computed spectral amplitudes. Therefore,
it is necessary to remove or minimize effects of the road geometrical curve to observe

the roughness variations and improve the PSD estimation of the pavement proﬁle.

6.3.2.2 Disadvantages of PSD method

As a diagnostic tool, PSD estimation may indicate existence of short-lived surface
defects or repeated waves in the proﬁle. However, it can not point to the location of
any of those frequency-related roughness features. PSD analysis can be also quite
misleading under some circumstances. It has been shown that, two completely
different proﬁles can yield identical PSD estimations [32]. Finally, the effort to
develop ride indices using PSD analysis can be reduced, if there is a transformation
that divides a full range spectrum of a proﬁle into a number of desired bandwidths and
provides energy of the signal in each bandwidth in a single step. This is the reason

why the joint time frequency analyses were introduced to the ﬁeld.

121

6.3.3 Joint time frequency analysis method

One common method of visualizing a signal is in the time domain. This
representation often plots the signal value as a function of time. Another useful signal
representation is the frequency-domain view of the signal. This representation is
typically based on a variation of the Fourier Transform and commonly plots the
frequency or phase content of a signal as a function of frequency. By examining the
frequency-domain view of a signal, one can derive information about the signal that
might not be immediately apparent from an examination of the time-domain
representation.

Although frequency-domain representations such as the power spectrum of a signal
often show useful information, the representations don’t show how the frequency
content of a signal evolves over time. For this task, Joint Time-Frequency Analysis

(JT FA) is useful.

6.3.3.1 Different types of Joint Time Frequency Analysis (JT FA)

There are a number of different transforms available for JTFA. Each transform type
shows a different time-frequency representation. One approach to obtain a local time-
frequency analysis (suggested by various scientists, among them Ville), is to cut the
signal ﬁrst into slices, followed by doing a Fourier analysis on these slices. However,
the ﬁmctions obtained by this crude segmentation are not periodic, which will be
reﬂected in large Fourier coefﬁcients at high frequencies (i.e., the Fourier transform
will interpret this jump at the boundaries as a discontinuity or an abrupt variation of

the signal).

122

Another approach is the Short Time Fourier Transform (STFT). The STFT consists

of applying the Fast Fourier Transform (FFT) repeatedly to short segments of a signal.

The STFT technique suffers from an inherent coupling between time resolution and

frequency resolution (increasing the ﬁrst decreases the second, and vice versa), which

is due to the uncertainty principle. This coupling is not practical for localization

purposes. Low ﬁ'equencies can be hardly depicted with short windows, whereas short

pulses can only be poorly localized in time with long windows.

Other JT FA methods and transforms can yield a more precise estimate of the energy

in a given Frequency-Time domain. Some options include:

Gabor spectrogram [35]: it is a special case of the STFT. It is used to
determine the frequency and phase content of local sections of a signal as it
changes over time. The window used is the Gaussian function. The window
function means that the signal near the time being analyzed will have higher
weight.

Wigner distribution [36]: it is the Fourier transform of the autocorrelation
function of the signal with respect to time shift.

Cohen class transforms [37]: it is the Fourier transform of the autocorrelation
function of the signal with respect to time. Figure 35 shows the relationship
between Wigner and Cohen’s class distributions.

Wavelet transforms [37]: it is similar to the Fourier transform except that it
uses generalized function and not periodic functions.

Although, these options are powerﬁrl for electrical and mechanical signals, we found

not to be effective for the analysis of pavement proﬁles (except for wavelet

transforms). Speciﬁcally, the following issues were noted:

123

l. The time support governed by the window function is equal for all the Gabor
functions for all frequencies. In order to achieve good localization of high-
frequency components, narrow windows are required; as a result of that, low-
frequency components are poorly represented.

2. The non-linearity of the Wigner distribution causes many interference
phenomena, which make it less attractive for detection purposes.

3. Although, the Cohen’s class distribution ﬁrnction has higher clarity than STFT
and Wigner distribution, it still suffers from an inherent cross-term
(interference) contamination when analyzing multi-component signals.

Auto-term

  
 

Cross-term

 

-- .................. 5 O Auto-term
O‘ ,

t

 

 

Cross-term

Figure 6.7 Relationship between Wigner (Wx) and Cohen’s Class (Ax) Distributions
[37]

These limitations in the resolution were some of the reasons for the invention of

Wavelet theory.

6.3.3.2 Wavelet transform method

Wavelets make the basis for a fairly recent mathematical transformation, which is
used to present the changes in spectral characteristic of the proﬁle with distance in a
2D wavelength (frequency) - distance plane. The added complexity of describing the
original 1D proﬁle (or PSD distribution of the proﬁle) in a 2D representation results in

a redundancy that can be advantageously utilized to improve interpretation of the

124

proﬁle data. Application of wavelet transforms in providing a multi-resolution
decomposition of proﬁles and a common basis to relate different available roughness
indices has been investigated recently [38, 39]. Like Fourier transform, wavelet
transform decomposes the discrete proﬁle into a linear combination of wavelets (i.e.,

non-periodic functions).

In [40], De Pont et al. introduced wavelets to be used for identifying the local
features of the proﬁle. The suggested method localizes features of the proﬁle in terms
of short, medium and long wavelengths. However, the method does not identify the
individual roughness features that are present.

Application of wavelet transforms as advanced diagnostic tools in pavement

smoothness evaluation will be investigated further in this chapter.

6.4 FINALIZATION OF ROUGHNESS IDENTIFICATION METHODS

6.4.1 Filtering out the topography

The geometry trend in the raw proﬁle taken by reference proﬁlers (e. g. rod and level
survey, walking proﬁler, high seed proﬁlometer) is normally of an order of magnitude
higher than the roughness ﬂuctuations. Therefore, the variation in the proﬁle
roughness is obscured by the road geometry curve and, therefore, cannot be
recognized. Thus, it is necessary to remove or minimize effects of the road
geometrical curve to observe the roughness variations. A number of approaches have
been used for removing the long-wave topography from the measured proﬁle (e.g.

moving average, 2-pole Butterworth, Chebyshev ﬁlters, etc.). The principles of several

125

ﬁltering techniques have been discussed in [41]. The Butterworth ﬁlters are the most

appropriate for our application since they have the following characteristics [41]:

0 Smooth response at all frequencies
0 Monotonic decrease from the speciﬁed cut-off frequencies

0 Maximal ﬂatness, with the ideal response of unity in the passband and zero
otherwise.

These properties will not distort the signal. Figure 6.8 (a) shows the frequency
response of a low-pass Butterworth ﬁlter. In the frequency domain, the z-transform of
a ﬁlter is written as follow:

— N —1
Z l b 1

b0 +blz‘l +"'+bNb“ (6.5)

 

H (z) = 1 —(N 4)
1+a z‘ +...+a z a
1 Na —1
Where,
H (z) = The z-transform of the impulse response of the ﬁlter
2 = A complex number
bj = The set of forward coefﬁcients of the ﬁlter
Nb = The number of forward coefﬁcients of the ﬁlter
a k = The set of reverse coefﬁcients of the ﬁlter
N = The number of reverse coefﬁcients of the ﬁlter
a

In order to reduce ﬁltering error, steeper slope for the transfer function of the ﬁlter is
required. To achieve that, higher order ﬁlters are usually used. Figure 6.8 (b) shows
the magnitude of the transfer function of Butterworth ﬁlters for different orders. For
order higher than 4, the transfer function are not that different. However, a ﬁlter

implemented by directly using the structure deﬁned by Equation 6.5 is a direct-form

126

IIR ﬁlter. A direct-form IIR ﬁlter often is sensitive to errors introduced by coefﬁcient
quantization and by computational precision limits. Also, a ﬁlter with an initially
stable design can become unstable with increasing coefﬁcient length. The ﬁlter order
is proportional to the coefﬁcient length: As the coefﬁcient length increases, the ﬁlter

order increases and the ﬁlter becomes more unstable. Writing Equation 6.5 as a ratio
of z transforms, which divides the direct-form transfer function into lower order

sections, or ﬁlter stages will lessen the sensitivity of a ﬁlter to error. By factoring
Equation 6.5 into second-order sections, the transfer function of the ﬁlter becomes a

product of second-order ﬁlter functions, as shown in Equation 6.6.

—1 —2
b0k+b1kz +b2kz
k=1 l+a1kz"l+a2kz‘2

 

and Na 2 Nb (6-6)

= The number of stages

= The largest integer less than or equal to (Na/2)

127

 

£3»)!- Cutoff frequency

 

 

  
 

 

o
a -10
g -20
a
'E -30
8 4o.
2

'50 PaSsband

-60 — -.

0: . - - 1., -

’31
3.30:- ............................ L_,_._zv. _
j; l Passband Stopband
(D
g -60
o.

-90 ‘4 - ,~ ..

Lwi- . 1.- _ .1. __. -L-.-._-. . . .2 . ,Ln, innnml_.__n_._1_;n_ _ - - a” ..,
0.001 0.01 0.1 1 10 100 100(
Frequency (radians/sec)
(a)

0.4 -

Magnitude square

9
N

 

 

Normalized frequency

(13)

Figure 6.8 (a) Butterworth Filter Bode Plot (b) Transfer Functions of the Butterworth
Filter for Various Orders

128

The ﬁlter structure defmed by Equation 6.6 could be described as a cascade of
second-order ﬁlters. The overall ﬁlter was implemented by cascading all passes in

series. Figure 6.9 illustrates cascade ﬁltering.

 

 

 

 

 

 

 

 

 

 

x(i)oﬁ Stagel Stage2 _..—..— Stage N5 ———. y(i)

 

 

 

Figure 6.9 Cascade Filtering Diagram

If n is the number of samples in the input sequence, the ﬁltering operation proceeds

as shown in the following equations.

ryo [i] “[13
l Sk [l] = yk—l[l]-alksk[i—1]_02ksk[i_2]
.y k [l] = bOkSk [i]+blksk [i_1]+b2ksk ll" 2] (6'7)

 

foreach samplei=0, 1,2, ...,n— l andk= 1,2, ...,Ns.

Butterworth ﬁlters always create phase distortion. This can be modiﬁed to become
zero-lag ﬁlters when the data are processed in both the forward and reverse directions.
In such cases little improvement is realized by applying multiple passes. This solution

was ﬁrst introduced by Murphy and Robertson in 1994 [42].

6.4.2 Method for Identifying Joint/Crack Faulting

6.4.2.1 Introduction
Four methods were studied for identifying the faulting from a proﬁle; (1) Discrete

Elevation Difference (DED) method, (2) Discrete Slope (DS) method, (3) Discrete

129

 

Curvature (DC) method and (4) Wavelet analysis method. The DED method takes the
elevation difference of the two discrete sample points. These two points would be the
two adjacent points if there were no moving average ﬁlters applied to the proﬁle that
is being analyzed. However, because the faulting appears in several points in a ﬁltered
proﬁle, the two points are not adjacent but exist with some distance apart. This
distance would be dependent of the type and the base length of the ﬁlter applied to the

proﬁle.

The DS method is used to detect discontinuities in digital image processing. It is
similar to the DED method in the sense that it takes the difference of the elevation
heights. But the difference is that the slope (difference in elevation) is taken from the

adjacent sample points. The DC method is similar to the DS method other than that it

takes the curvature (2nd derivative) of the proﬁle.

Wavelet analysis is a popular method for compression and multi-resolution analysis
of discrete signals. It decomposes a given signal into several signals of different scale.
The term ‘scale’ here is related to frequency. Thus, it allows the user to see the signal

in different scales along with time or distance.

All the analyses herein are based on the averaged proﬁle and not the raw proﬁle

except for the artiﬁcially generated dummy proﬁle.

130

Figure 6.10 shows example outputs from Matlab® using De Pont et a1 method [40]
to detect roughness features with discrete wavelet transforms. The proﬁle of an LTPP
section (20-0201) was used for this purpose. “D1” and “D2” may be treated as noise
since they are decomposed of high frequency contents. “D3” and “D4” may provide
useful information depending on the type of the mother wavelet used. The highlighted
locations in Figure 6.10 might have potential of faulting or cracking since they show
high variations in the decomposed signal. The wavelets used were Daubechies (db)
wavelets, which are widely used in engineering applications because there are the only
orthogonal wavelets (allowing unique reconstruction of the original signal) that can be
used in both continuous and discrete analysis. Also, it is the wavelet used by [40].
Besides Daubechies wavelets, the analysis performed in this research included other

types of wavelets: Haar, Symlets (sym), and Coiﬂets (coif).

131

DeeomposltionatlevelB:s=a8+d8+d7+d6+d5+d4+d3+d2+d1.

 

 

 

 

r1:
L.

‘

 

Nd
COCO

 

l
\
r—i—e
l
L\
_._4
I
_\q
l
l
_,_
l_ —
I
._(._.
\
__ _

"éto

oo o oo o o

.L—L
OO O

A
61190109

 

1 00 200 300 400 500 600 700 800 900

Figure 6.10 Daubechies 3 Wavelet Analysis

Once the method is decided, then it is important to decide on the threshold. This is

because the accuracy of the methods depends on the threshold value. While methods

using high threshold values will not detect low level faults, methods using low

threshold values will add errors. Two possible methods are suggested. (1) Threshold,

and (2) Adaptive Filtering. Thresholds may simply be applied on the slope, curvature,

132

DED, and on wavelets. Figure 6.11 shows an example of threshold applied on the
slope with a value of 0.03. With this threshold, all the locations that have slopes in
between -0.03 and 0.03 are considered to be continuous. Although it is not shown in
Figures 6.12 and 6.13, similar thresholds may be applied to the curvature and DED.
Unlike these examples, deciding on a threshold value is not a simple problem. Further
analysis will be performed to decide on a reliable threshold.

Adaptive ﬁltering is a method used in digital image processing. It calculates the
global and local standard deviation and compares them to each other. If the global
standard deviation (GSTD) is greater than the local standard deviation (LSTD) around
a point, then that point in the sample is continuous. If the LSTD is greater than the
GSTD, then the point around where the LSTD is calculated is not continuous. Thus,
the GSTD of a given sample acts like a threshold. The advantage of this method is that
given the samples, the threshold (GSTD) is easily calculated. However, the inherent
disadvantage is that the threshold is dependent on the number of sampled points and if
the signal is very noisy it does not clearly point out the locations of discontinuities (see
Figure 6.15 as an example). Figures 6.14 through 6.16 show the discontinuities on the
proﬁle using the slope, curvature and DED, respectively, along with the adaptive

ﬁltering.

133

Elevation (in)

Elevation (in)

853320 was 2:95 2 .0 853m

0595' @3230 l

8.0-

 

3d-

 

 

 

 

. mod-

 

no .0

 

 

3 d

 

 

 

mo .0

 

 

8% Es ”EOE :6 23mm

o__.§n_lll 32m l

32:02:.

mo .o-
8 .o-
3 .o-
no .9.

no .0
3 .o
oo .0
mo .0
F .o

amrmung

odors

134

(u!) normals!

wEBEm o>u§u< wan 32m 3 wouoBuQ moEscucoomE E .o Emmi

2n Hal 092m Inl

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2.
m- .. .
2.- “>1.
H: . I n m f
2-- .\\I1 J/ H . 5\./ Q r
or E _ /:_ 9* _ . / _ N”.
2. /~/..\\ _ J M

n d < w x. a.

2 u x N

a. <

2

 

 

(H!) "011mm

855wa :ocgﬂm 2885 can oEoE 2.0 .3de

mEoi-l Eomol

 

 

 

 

 

 

 

 

 

 

m3-
3.
mod-

mod
2”
2 6
ad
and

eouarayrq none/1313

(I1!) Supine-1

135

(u!) nonmetal

(a!) "camera

$.38an u>ua€< was 8:80th conga—m 8885 NE @88qu 3233885 3.0 oSwE
088ml 280+

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

m4.- 2.
m- -
3- - > S- J
-- . ﬂ : _.-.
22. . s- m
- .. . . . . . u
D '31 > _ ﬂ _ i . I‘ll. . > . a \ f D m
d H m.
..-11 _ / N 1
ﬂ . . c / K Ho
mm . < s
u
Wm u no
MESH—E D>EQMU< Una QHBMEU E @800qu mugs—HEGOQEQ m H .0 PEWE
ea Eml ﬂawtso l
Wm-
N. 1
m.“-
T m a J
8
W.D| F I. . F r .I \l‘lj’ . = H...- D. h I I _ m
c r u a d' c r: a m m m.
i a a
2.. < <\ 1.
2 L
N 1

 

 

 

 

On

136

The methods mentioned above were tested using LTPP proﬁle data. LTPP section 20-

0201 was selected because it has the highest faulting (14 mm).

Figures 6.17 through 6.22 show a proﬁle with faulting that was measured from the ﬁeld
with an accuracy of i1 mm. The ﬁgures also show the faults that were detected from the
methods described above. The threshold value of 1 mm was used for DEDM and slope
method while 0.5 mm was used for level 1 of wavelets (D1). Figures 6.23 through 6.28
show the same methods but with adaptive ﬁltering instead of the threshold. The discrete
elevation difference method detected the faulting better than the wavelet methods in

terms of location and magnitude.

137

(ww) ﬁumnea

(mm) uoneAaB

0000
00
e-Fmo‘P‘T‘TW‘?

00502 32m 2.: Bot chsmm 038qu 3.0 233m

mEoi IT 82m 11 game 09880 IT

 

 

 

 

 

 

€85.05
9: our 08 8 8 9. om o
. t . _ _ . . .H _ o
- N
1 V i:
- o m.
. - m .m-
- S 1w,
w
4 2 (
- E
E
00502 8:80th 5:565 8885 05 Bob was—sum 088:5 N. L .0 ouswmm
mac-ill $89050 .>m_m $935+ Em. ummenOIT
E 8:920
03 our 02 8 8 3. om o
, . . . . 1 . . o
c «a «~04:€«~_.H m
- m
MM .0
1 m m.
., .- m m.
r or m
- : ,wx
., N.
- 2
- 3
3

 

 

138

(ww) Sumned

(ww) Gunned

0N-
ON.
0 _.1
o :1

ml.

or
0:

02:02 02283 N00 05 Soc @533 0800on 8.0 oHDwE
209.". ll Non IT :33 “520000 in

05 8:920
03 :8 8 8 o:

 

 

 

 

00802 029.83 0003 05 800 chBE 080800 a L .0 80wE

2:2: 11 .8: l :39 8288 11

E 8:990
our 08 8 8 0:

 

 

0:. H

«l

I

ijTT

DVMNromCONtOlDVOONx-O

 

 

IDVCONV-OODQNCDIDVCONFO

(LUUJ) uogreAelg

(ww) uogeAala

139

(ww) ﬁumned

(urtu) 5U9|ﬂB_-]

0N-
ON.
0 7

mm.

07

00502 “20:83 N88 05 800 wﬁzsmm 0800qu NN.0 Emmi
2:20 It N58 It 89 8288 1.1

 

 

 

 

E 8:920
02 ON, :2 8 8 0: ON 0

1. . a _ 1 c l . « n 1 m
- - N
- m

- v
1 m
- 0

- T N
. m

n - S
- - :
0

2 E
2

00502 “228.3 C000 05 800 wcgzmm 08.0300 _N.0 803m
2:90 1.. E8 ..T 239 8880 11
E 8:920
03 ON? 02 8 8 0... ON 0

A : h P Q _ G H I w
- - N
- m

- , v
- . m
- o

- - N
- 8

.. - 0
- E
- - S
- N.
- - 2
, - 3
m:

 

 

 

 

(unu) uogreAela

(111w) uogeAe|3

140

mam-52E o>0000< 0% 00502 002m 05 Bob chNE-m woaoopom 5.0 053m

280 l 820 l :39 8880 .1..-

 

 

 

 

 

 

E 8:920

9N1 VF:

- N.-

8. - 2 S-

a 2- - _ m m
m 8- - __ o: _ 8., 2.: F m 8 8 - 8 wN 0 N-
m. ...r. ‘ 0 . - . m
0 o - _ _ 2 ..
o
101 ﬁ 8
m - . 0 2

S - - NF

, -_ E

2 8

wccﬁzm 0300000. 0% 00522 00:80th 0000811.— 8885 2: 800.0 was—30m 0800qu mm.0 0.5310
0.090 Ii 0009050 .>m_m 90:85 ...T «.000 00280011
E 8:920

8. 3-

T NFC

0N- - - S-

a 2- - .. H W
m S- 0 .n_ 8 . ONF . 02 0 8 8 8 ON 0 W.
W. . ,h__ . m 1 _ . . _ 1 o
m m- 0 ‘ é . - N
1w; o - w . m m
101 w

m - . 0 2

2 - - NF

, - E

8

 

 

 

 

0..

(mm) Bumneg

(unu) Sumneg

l4l

(urur) uogeAan

(rum) uogeAaE]

 

888 8:09;. a. 88: 98:3 N00 90 :50 8230 890900 8.8 2:90

2090 ll N00 IT =33 0020000 II

E 8:920

 

 

 

 

 

 

macozﬁ o>0000< 00 00502 8383 H8: 05 800 qusmm 08030Q 0N0 oSwE
2090 l 000: IT =08 0028.00 IT

30 8:920

 

 

 

 

 

 

(rum) Sumned

(mm) Bumned

142

ow- I

m?

(ww) uogewla

(qu1) uogeAala

or-

or
3

was: 2:33 a 8502 E23,? aim 2: 50¢ maﬁﬁ 333E mg ”Ema
mEoi IT Nﬁsm IT gzﬂ umzwmno IT

E 8:920

 

4L#

 

03m.

om? cor

om

‘I‘I—rI

 

 

$55 02353 Q 8%: 632;» E8 2: 8% maﬁa 8338 8.0 2&5

£9; 1: ES IT g5 823.8 I:

E 8:93

 

 

 

VV'I

T I

 

(DVNOCD‘OVNO

VPFF

“.“T

(qu1) Sumnej

 

(ww) Sumned

143

6.4.2.2 Discrete Elevation Difference Method

For the reasons mentioned above, DED method with threshold was selected for
detecting faulting at joint/crack. The discrete elevation difference method consists of
computing the difference in elevation between points at a given interval. Theoretically,
when the samples of the proﬁle are collected over a faulted joint/crack, the difference in
elevation should appear in two adjacent samples. However, as shown before, the faulting
of the slab in the raw proﬁle appears at several sample points (Figure 6.29). As discussed
before, the elevation is collected at every 19 mm and recorded at every 76 mm (Figure
6.30). Consequently, when the difference in elevation is computed, each fault/crack is
represented by a range of differences. Thus, the largest absolute value (local maximum)
for each range of differences was taken as the fault magnitude. However, as it can be seen
from Figure 6.31, if the faulting is calculated based on the elevation difference of
adjacent points, three different values of the fault magnitude can be obtained, depending
on whether the distance to the fault is a multiple of the recording interval:

1 Exact magnitude, [ Figure 6.31 (a)];
2 75% of the exact magnitude, [Figure 6.31 (b) and (d)];

3 50% of the exact magnitude [Figure 6.31 (c)].
In order to resolve this problem, it was decided to take the difference in elevation

between the points that are 152 mm (i76mm of the point of interest). The algorithm for

the fault detection method is described in Figure 6.32.

144

 

Elevation (mm)
L

 

 

 

Actual Faulting 5 mm
reported

   

 

 

 

  

 

  

Actual Faulting 14 mm

 

 

 

 

 

'16 reported
-20 T
-24 I I I -
100 110 120 130 140 150
Distance (m)

Figure 6.29 Faulting in the Raw Proﬁle

Sampling interval=19 mm

 

 

Figure 6.30 Effect of the Moving Average Filter on Surface Discontinuity

145

 

 

 

76mm

 

 

 

 

 

 

76mm

 

 

 

 

 

 

 

(d)

(C)

Figure 6.31 Different Scenarios of Recording a Fault after Filtering

146

 

/lead Original may

 

 

 

 

 

 

 

Calculate local maximum (AyQLmax

 

 

1

Yes

 

 

 

 

V

 

 

 

No Faulting

 

Faulting (AyQLmax

 

 

 

 

 

 

 

 

I I

Figure 6.32 Fault Detection Algorithm

147

 

6.4.3 Method for Identifying Pavement Breaks and Potholes

Before detecting breaks, different steps are followed. First, differences in elevation are
computed. This step is the same as the fault detection method. Second, if the sign of two
successive faults is different, the method checks the distance between them: if the
distance is less than a pre-selected length, it is considered break; otherwise, there is no
break. In Figure 6.33, a width of 0.9 m was selected. This threshold could be updated
according to the user perception. The algorithm for the breaks detection is described in

Figure 6.34.

   

....................................................

<09 m

Figure 6.33 Illustration of a Pavement Break

6.4.4 Method for Identifying Slab Curling

6.4.4.1 Introduction

Several methods were studied to extract the curling of the slabs from the dummy
proﬁle. The methods studied were; ( 1) Gaussian Band-Pass Filter (GBPF) method, (2)
Joint Time Frequency (JT F) analysis, (3) wavelet analysis method, and (4) Discrete Slope
Method. The moving average of the proﬁles is not critical for identifying curling because
the moving average ﬁlters are low-pass ﬁlters and do not eliminate the curling aﬁer

passing the proﬁle through the ﬁlters.

148

 

ﬂead Original Proﬁ/

 

 

Calculate Ayn = yn-yn+2

 

 

I

 

 

 

Calculate local maximum (Ayn)Lmax

 

No Yes

 

 

 

 

AYn)Lmax >T i
r

 

 

 

No Faulting

 

 

 

Faulting (AyQLmax

 

 

 

No

 

    

Sgn((AYn)Lmax) #Sgn((Ayn+1)Lmax)

No

 

 

   

Dist((Ayn+1)Lmax) - DiSt((AYn)Lmax)

 

l Yes

 

Breaks (AYn)Lmax

 

 

 

 

 

 

Figure 6.34 Break Detection Algorithm

149

As discussed earlier, the topography has a signiﬁcant impact on the curling detection.
To ﬁlter out the topography, a high-pass ﬁlter (91 m long cut-off wavelength) was
applied. Then, to eliminate the effect of proﬁle roughness and micro-curling, a low-pass

ﬁlter (1.5 In short cut-off wavelength) was applied. Figure 6.35 shows an example of this

procedure.

—— Raw Ptofile

 

long w avelength filtering —short w avelength filtering

1-20 T‘W““ _‘____-_______T________.,
43010 20 , 4o 60 80

~4OI~‘ " I“ "\ /

Elevation (mm)

 

Distance (m)

Figure 6.35 Original and Filtered Proﬁles

The Gaussian Band-Pass Filter extracts certain frequency contents from the Fast
Fourier Transform (FFT) algorithm. Thus, the curling can be extracted after performing
FFT to the proﬁle and then applying the ﬁlter to the transformed proﬁle. Figure 6.36
shows the artiﬁcial curling used in the dummy proﬁle along with the curling that was

extracted using the GBPF method. The curling was extracted with a phase lag.

150

 

O)
O

 

 

 

 

 

a 40“
15, 20
C
,3 O
‘3
2 '20
“J 40 4

«60

O 5 10 15 20 25 30
distance (m)
curling (mm) ——fft_bandpass

Figure 6.36 Curling using GBPF Method

Figure 6.37 shows the curling proﬁles extracted from the dummy proﬁle using three

different wavelet analyses; dblO, sym8, and coif5 wavelets.

Joint time frequency analysis was performed on the dummy proﬁle and the LTPP
proﬁle (section 19-0217). The joint time frequency analysis calculates the frequency
energy distribution along with the time/distance. Although, the joint time frequency
analysis method is powerful for electrical and mechanical signals, it was not efﬁcient for
the analysis of pavement proﬁles. Figures 6.38 through 6.40 show the Wigner-Ville
Distribution (WVD), Pseudo WVD (PWVD), and Smoothed Pseudo WVD (SPWVD) of
the dummy proﬁle. As it can be seen from the Fourier spectrum —placed at the left hand
side of the time frequency distribution- the major energy is concentrated in relatively low

frequency components.

151

 

Elevation (mm)

 

 

 

 

distance (m)
— cuning (mm) — db10

(a) Extracted curling of the dummy proﬁle using db10 wavelets

 

 

 

 

 

A 40
E
g 20 ~
.5 04
§
2 -20
LL]

4:0

o 10 20 30

distance (m)
—— curling (mm) — sym8

(b) Extracted curling of the dummy proﬁle using sym8 wavelets

 

Elevation (mm)

 

 

 

 

distance (m)

— cuning (mm) —- coif5
(c) Extracted curling of the dummy proﬁle using coif5 wavelets

Figure 6.37 Created Curling and Extracted Curling Using Wavelets

152

Signal in time

 

ﬁr

 

Real part
$éo~no

 

 

 

Log. scale [dB] WV, lin. scale, contour, Threshold=5%

 

 

 

 

 

 

 

0.08
ll ll II-Io'uwwlwva-I.“IIIII I. I I
.3:
U1
g ' a}: 0.06 .
E ..E IIIIIIIIIIIIIIWIIIIIIIIIII
>\
U
a * ' 8 0 04 .
m a” 9 ’. .:.‘ I‘ .. ‘. ﬂ;
:3 a l | | | I In a.) no: I H
g - In 0.02-
c:
Lu
5
0 -5 -10

 

Time [s]
Figure 6.38 Wigner-Ville Joint Time Frequency Distribution of the Dummy Proﬁle

Signal in time

ﬁr

 

 

Realpart
.Ltbowac»

 

 

 

Log- scale [dB] PWV, Lh=512, Nf=4096,
‘ * lin. scale, contour, Threshold=5%

I II ll'll'lllll'll'll'll'll'l| l'l I I ‘l II'II'Il'Il'll'lli'l"

 

 

'3

 

lllllllllllllllllllllllllllIlllllllllllllllll
9

Frequency [Hz] I
O
2

Energy spectral density
9
8

 

 

 

 

0 -5 -10

 

Figure 6.39 Pseudo Wigner-Ville Joint Time Frequency Distribution of the Dummy
Proﬁle

153

Signal in time

 

    
   

 

 

 

 

 

 

 

 

5 SI
a. 2M
73 .2.
9‘ 4 . . .
SPWV, Lg=102, Lh=256, Nf=2048,
Log. scale [dB] lin. scale, contour, Threshold=5%
' ' 0.08 1 - T . II
2:
g ' ,_ 0.06-
:2 ..1 a II.
a
a . . a ,0,
o G
8- a
is e
is . In 0.02»
a I,
0%? 0 100 200 300 400 500

Time [s]

Figure 6.40 Smoothed Pseudo Wigner-Ville Joint Tirne Frequency Distribution of the
Dummy Proﬁle

The discrete slope (DS) method was identiﬁed as a method to detect slab curling. This
method has been used in signal and image processing for its capability to detect
discontinuities. It is similar to the DED method in the sense that it takes the difference of
the elevation heights. But the difference is that the slope (difference in elevation) is taken
from the adjacent sample points. Figure 6.41 shows a simulated proﬁle and the

corresponding slope.

154

Schematic Proﬁle with Daytime Curling

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

c
.9
‘5
>
2
In. “
-4 . 1 . .
0 20 4O 60 80 100 120 140 160
Distance
Schematic Proﬁle with Nighttime Curling
6
4 .,_, __fif i
C
.2
‘5
>
2
In
0 20 40 60 80 100 120 140 160
Distance
Slope of Profile
0.1
0.05*
3
2 o-—-
a)
-0.054
-0.1 . - , . 1 - 1 '
O 20 40 60 80 100 120 140 160
Distance
Slope of Profile
0.1
0.05« - I
§ 0*
5.005. ~ —
-0.1 a T . . .
0 20 40 60 80 100120 140 160

Distance

Figure 6.41 Curling Extracted with DS Method

155

6.4.4.2 Discrete Slope Method
It can be seen from Figures 6.36 through 6.40 that the GBPF method seems to work

better than wavelet and time-frequency methods. However, this method produces a phase
lag or time shiﬁ which could affect the localization. The DS method produces the same
order of error in magnitude but is more accurate in localization. Therefore, the DS

method was selected.

To compute curling, differences in elevation between the point that corresponds to the
local maximum of the slope ﬁmction and the next point where the slope function is zero
is computed (see Figure 6.42). The algorithm for curling detection is described in Figure

6.43.

Location of maximum curling

—3
db

 

slope (mm/m)

V40 60 80

Local maximum lslopel
Distance (m)

 

 

 

Figure 6.42 Detection of Local Maxima of the Slope Function

156

 

/{ead Original Proﬁ/
Long wavelength cut off

Short wavelength cutoff

 

 

 

 

 

 

 

 

 

V

 

Calculate Ayn = (yn+1- yn)/( dn+1- dn)

 

 

 

 

7

 

Calculate local maximum locations
of the slope (Lmax)

 

 

 

 

 

Calculate local zero locations of the
310136 (L0) (Ayn)L0

 

 

 

 

 

Calculate curling elevation
(Ayn)L0-(Ayn)Lmax

I I

Figure 6.43 Curling Detection Algorithm

 

 

 

 

 

157

6.4.5 Summary

The purpose of the work reported in this section was to review and identify the most
appropriate methods for detecting localized surface distresses. These methods include:
wavelet analysis, joint time-frequency analysis and discrete methods. The methods were
evaluated using simulated proﬁles and proﬁles of LTPP sections. The discrete elevation
difference method was selected for fault and break detection. The discrete slope method
was selected for curling detection. For the analyses presented herein, a threshold value of

1mm was used.

6.4.6 Development of a Window-Based Software System

Based on the preliminary ﬁndings, user-friendly software was developed. The
roughness features detection tool is an engineering software application that allows users
to view and analyze longitudinal pavement proﬁles. The software is an excel ﬁle with
macros that can detect the presence of pavement roughness features from the proﬁle and then
tabulate them in an excel ﬁle. The details of the software are discussed in [43, 44] and is

provided as Appendix D.

6.5 FIELD TRIALS
6.5.1 Criteria for Selecting Pavement Sections

In order to evaluate and verify the detection tool, it is necessary to conduct a ﬁeld
survey. The criteria for selecting the pavement sections are summarized in Tables 6.1 and

6.2.

158

Table 6.1 shows the minimum number of pavement sections that needs to be visited
with each distress type and severity level. However, this table assumes that in a pavement
section, there is only a single type of roughness features corresponding to a single
severity level. Thus, it should be noted that the number of pavement sections to be visited
can be reduced signiﬁcantly if a pavement section has a large number of distresses at

different severity levels, which was the case.

Table 6.2 shows the deﬁnition of the severity levels of different localized roughness
events. Since there is a similar trend in the proﬁle data for faults, breaks, bumps, and
depression, the deﬁnition of severity levels for these distresses may be combined. The
severity deﬁnition of the above distresses is the same as those used by Pathway’s fault
detection algorithm which again agrees with MDOT’s deﬁnition of severity level for

faults.

Table 6.1 Number of Pavement Sections1 for Veriﬁcation of Roughness Diagnosis Tool

 

 

 

 

 

 

Pavement Type Rigid
Distress Type ' Faults/Breaks Curling
Severity LOW 3 3
L 12 Medium 3 3
M High 3 3

 

 

 

 

 

l 2
Notes: A pavement section is 0.1 mile long. A given section may have all three severity levels.

159

Table 6.2 Number of Deﬁnition of Severity Level

 

 

 

 

 

 

Severity Level
Distress Type Low Medium High
Faults, Breaks, . >6. 4 mm, >19 mm
Bumps and 9.4 mm (0.25 1n) $19 mm (0.75 in)
egessron
Curling N/A N/A N/A

 

 

 

 

6.5.2 _Field Tests

The ﬁrst visit to the ﬁeld was held on Tuesday, July 26th, 2005. The trip was scheduled

for measuring faults in rigid pavements. An interstate highway was selected in the
University region. Section 1 was located on 169 close to exit 84 (Airport road). The
pavement type is rigid with a joint spacing of 41 12.5 m (41 feet). The selected pavement
section was approximately 0.64 km (0.4 miles) long and has a lot of faulting at the joints
and cracks. The measurements were taken in between stations 386+36 and 408+95 using
the digital faultrneter (also called the Georgia faultrneter) at the left and right wheel paths
and at the center of the lane. The right wheelpath was 0.9 m (3 it) away from the shoulder
and the left wheelpath was 1.75 m (5.75 it) apart from the right wheelpath. These
represent rough locations of where Pathway lasers would fall. Table 6.3 summarizes the

number of measured faults in terms of the severity level deﬁned in Table 6.2.

The second visit was held on Thursday, October 27th, 2005. Section 2 was located on I-
69 just before exit 105 (Perry Exit). Section 3 is about a mile down the road from the end
of the ﬁrst one. Section 2 was approximately 0.83 km (52 mile) long and has a lot of

faulting at cracks. The measurements were taken between stations 693+00 and 720+OO.

160

Section 3 was approximately 0.83 km (0.52 mile) long. The measurements were taken
between stations 773+00 and 800+00. Both sections are rigid pavements with a joint

spacing of 12.5 m.

The third visit was held on Tuesday, November 8th, 2005. Section 4 was located on I-
69 eastbound at mile marker 130. It is also a rigid pavement with a joint spacing of 12.5
m. The selected pavement section was approximately 1.4 km (0.86 miles) long. The

measurements were taken between stations 130+00 and 174+00.

In sections 2 through 4, faults were measured at the left and right wheel paths using the
Georgia Faultrneter. Repeated measurements were also collected in order to evaluate the

accuracy of the measurement device.

Table 6.4 summarizes the number of measured faults in terms of the severity level and

the extreme magnitudes of faults in each of section 2 through 4.

In order to independently verify the methods using the manually collected data, the
proﬁle data for the same pavement sections collected using the high speed proﬁlometer
were requested from MDOT. The summary of the data collected are presented in

Appendix C.

161

Table 6.3 Summary of Measured Faults in Section 1

 

 

 

 

 

 

Counts
Severity level Low Medium High Total
Indrvrdual 199 1 6 1 216
measurements
Average fault along 69 3 0 72
transverse direction

 

 

 

 

 

Table 6.4 Summary of Measured Faults in Section 2 through 4

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Severity level Section 2 Section 3 Section 4
Low Medium | High Low Medium High Low Medium High
Counts 124 4 | 0 57 12 0 150 48 0
highest
magnitude 14.2 9.7 13.7
(turn) -
Lowest
magnitude -2.8 -3.8 -10.9
(mt!)

 

6.5.3 Results

Statistical analyses were performed to study the accuracy of the method. Figures
6.44 through 6.46 shows the results obtained from these analyses. As can be seen from
these ﬁgures, the predicted magnitude and location of faults for each site match well
with the measured magnitude and location. However, each ﬁgure show a bias, all be it
very small; this error could be the result of inattention in fault measurements or
inaccuracy of proﬁle data. Comparing the slopes from successive ﬁgures, it can be
seen that the bias in the measurements was slightly reduced after each ﬁeld test (slope

value closer to unity).

162

 

The new method was able to detect the magnitude of faulting with an average error
of 0.1% and a standard deviation of 1%. The localization was also highly accurate
(average error in distance is within 1 m or 0.5%). Therefore, we conclude that the
DED method is able to detect the location of a fault and compute its magnitude with a

reasonable accuracy.

Three breaks were reported during the ﬁeld test in Site 1. The tool was able to detect
all three breaks with a comparable magnitude and location. Table 6.5 summarizes the

number of observed and predicted breaks, their width and magnitude for all the sites.

Site 1 also had severe early morning (upward) curling (Figure 6.47). The tool was

also able to detect the curling magnitude and frequency (Figure 6.48).

6.6 Summary

The purpose of the work done in this chapter was to show that the newly developed
roughness diagnosis methods are able to detect, identify, and localize roughness
features that are not readily available from video imaging methods. We proved that the
methods discussed above give results with a comparable magnitude and frequency of
bumps and depressions (faults, breaks ad potholes) with an average error of 0.1% and
a standard deviation of 1%. The localization was also highly accurate (average error in
distance is within 1 m or 0.5%). Thus, these new methods could be used as a
complementing module for the existing PMS. It would help the highway agencies in

deciding whether a particular section of the pavement needs a

163

maintenance/rehabilitation action. Appendix D presents the user manual for the newly

developed detection tool.

15

 

 

predicted faults (mm)
'J‘i

      

   

 

0 I——
-5
-5 0 5 10 15
measured faults (mm)
(a)

”g“ 1000 I y=0.99x
8 800 * R2: 0.99
g 600 7:
_U .
B 400 r
O
'6 200*
8
Q 0 m I

0 200 400 600 800 1000

measured distance (m)

(b)
Figure 6.44 Correlation Analyses for Site 1 (a) Magnitude (b) Location

164

 

predicted faults (mm)

 

 

-5 0 5 10 15

measured faults (mm)

(a)

 

 

predicted distance (m)

0 200 400 600 800 1000

measured distance (m)
(b)
Figure 6.45 Correlation Analyses for Site 2 (a) Magnitude (b) Location

165

predicted faults (mm)

y—a
£11
0
O

1000

predicted distance (m)

I—II—IN
OUIO

.'_..'—.
momom

.___1

500

 

 

-lO

 

 

0

10

measured faults (mm)

(a)

500

1000

measured distance (m)

(b)

20

  

1500

Figure 6.46 Correlation Analyses for Site 3 (a) Magnitude (b) Location

Table 6.5 Surrunary of Observed Breaks from Field Trials and Predicted Breaks

 

 

 

 

 

 

Using the Algorithm
Starting point (m) Width (m) Fault magnitude at the Fault magnitude at the
starting point (mm) end (mm)
Observed Predicted Observed Predicted Observed Predicted Observed Predicted
220 221 0.14 0.15 -4.3 —4.8 3 3.3
320 321.7 0.14 0.15 -5.3 -5.8 3 3
620 624.3 0.5 0.53 -13.9 -l4.2 11.9 12.2

 

 

 

 

 

 

 

 

166

 

Elevation (mm)

Curling magnitude (mm)

 

0 200 400 600 800
Distance (m)

Figure 6.47 Raw Proﬁle for Site 1

6x
I— ,

0 200 400 600 800
Distance (m)

1 4 154$; Litrliringiﬁereﬂmﬁlj

 

Figure 6.48 Filtered Proﬁle and Predicted Curling Magnitude

167

CHAPTER 7
IMPACT OF ROUGHNESS ON VEHICLE DURABILITY

7.1 INTRODUCTION

All road surfaces have some level of roughness even when they are new, and they
become increasingly rougher with age depending on pavement type, trafﬁc volume,
environment etc. This process is mainly driven by the interaction between vehicles and
pavements. An increase in pavement roughness leads to higher dynamic loads. This
ampliﬁcation in the load magnitude can lead to a tangible acceleration in pavement
distress. The pavement distresses lead to surface irregularities which translate into a
transient event in the raw proﬁle affecting the ride quality of a road. The surface
proﬁle of the road transmits the vibrations through the tires and suspension system to
the body of the vehicle and then to the driver, passengers and cargo. Vehicle
manufacturers place a major focus on constantly improving the design of these
different vehicle components to respond better to changes in road surface proﬁles.
Despite this, changes in the road surface proﬁle still directly affect the user costs
including repair and maintenance costs. For example, the American Association of
State Highway Ofﬁcials (AASHTO) reported that poor road conditions added an

estimated $76.8 billion to transport costs annually [3].

The developed tools described in chapter 6 detect, locate and identify the level of
surface irregularities; however, they do not in themselves provide guidance on
acceptable roughness levels to limit user costs. The objective of this chapter is to

develop a methodology to estimate the effect of roughness features on vehicle

168

durability and to provide guidance on acceptable roughness levels. Accordingly, we
propose using a mechanistic-empirical approach to conduct fatigue damage analysis
using numerical modeling of vehicle response. A sensitivity analysis was performed to
quantify the relationship between roughness feature height and width to vehicle
suspension damage. The analysis consists of the following steps:

1. Artiﬁcial generation of road surface proﬁle;

2. Artiﬁcial generation of roughness features (bumps, depressions and curling);

3. Estimation of the response of the vehicle to these transient events;

4. Computing of the induced damage to the vehicle suspension;

5. Repeat step 2 through 4 for different heights and frequencies of roughness
features.

7.2 RESEARCH APPROACH
7.2.1 Artiﬁcial Road Profile Generation

7.2.1.1 Introduction

Various types of road models have been in use for years to represent roads for
analyzing vehicle ride behavior [45]. One of the ﬁrst proposed stochastic models is an

equation of the form:

 

A
Gz(v) ‘ (2m)2 (7.1)
Where,
Gz(v) = PSD function of elevation (2)
v =Wavenumber
A = Roughness coefﬁcient obtained by ﬁtting the OSD of a measured road to

the above equation

169

As the elevation is perceived to be changing with time, it also has a velocity
(proportional to slope) and acceleration (proportional to the derivative of slope), which
also have PSDs for the same road section. Since velocity is the derivative of position,
the velocity PSD is related to the elevation PSD by the scale factor (27w). And,
likewise, the acceleration PSD is related to the velocity PSD by the same scale factor.
The concept of the road as an acceleration input to the vehicle is important to
understand because its ultimate effect - vehicle ride vibration- is invariably quantiﬁed
as accelerations [45]. Gillespie et al. [46] proposed the following equation for the PSD
model:

Ga Gs
Gz(v) = + 2 +Ge (7.2)

(2nd“ (22w)

 

The ﬁrst component, with the amplitude Ga, is a white noise acceleration that is
integrated twice. The second, with amplitude Gs, is a white noise slope that is
integrated once. The third, with amplitude Ge, is a white noise elevation. The model
can also be written to deﬁne the PSD of proﬁle slope by looking at the derivative of

the above equation.

I = Ga
G () (2m)

 

+ Gs+ Ge(2ﬂ'v)2

2 (7.3)

Gillespie et al. [46] suggested ranges of roughness parameters based on the road
proﬁles measured in North America, England and Brazil (Table 7.1). When traversed

by a vehicle, the proﬁle is perceived as an elevation that changes with time, where

170

time and longitudinal distance are related by the speed of the vehicle. The time-

varying elevation can also be characterized by a PSD that has units of elevation.

Table 7. 1 Roughness Parameters for the White-Noise PSD Model [46]

 

 

 

 

 

Surface type GS 6 Ga -6 Ge 6
(m/cyclex 10' ) (1/m cycle x10 ) (m3/cycle x10' )
Asphalt (Ann Arbor) l~300 0.0~7 0.0~8.0
Asphalt (Brazil) 4~100 0.4~4 0.0~0.5
PCC (Ann Arbor) 4w 90 0.0~l 0.0~O.4
Surface treatment (Brazil) 8~ 50 0.04 0.2~1.2

 

 

 

 

 

 

Marcondes et al. [47] developed another equation to predict PSD. Elevation proﬁles
of federal and interstate highways near Lansing, Michigan, were measured with a
proﬁlometer and PSD were calculated. They developed the following equations to ﬁt

the above data:

PDpe(v) = Ale(—kvP), v S V1

(7.4)
PDpe(v) = A2 (v-vo), v > v1

Where,

PDpe(v) = Power density value [in3/cycle] for the pavement elevation
v1 = Discontinuity frequency, cycles/in

v0 = Asymptote frequency, cycles/in

A1, A2 = Constants

k, p, q = Constants

171

Marcondes [48] reported the range of values of parameters for the equations as shown

in Table 7.2.

Table 7. 2 Ranges of Variable Values for PSD Equations [48]

 

 

 

 

 

 

Category A1 k p A2 V0 q
* ~ 7000~ 5.9E-7~ 0~ -2.6~

0C 1'3 7'2 67000 1'64” 4213-5 3913.3 -1.5
* ~ 24000~ ~ 6.0E-7~ 2.5E-3~ -2.2~

NC 1'5 3'4 83000 1'8 2'0 6.0E-5 4.95-3 —1.1
* ~ 63000~ ~ 1,413.4~ 4.6E-3~ -1.1~

AC 1'8 5'7 240000 2'0 2'2 7.7E-4 5213—3 .05

 

 

 

 

 

 

 

*Note: OC: old concrete pavement; NC: new concrete pavement; and AC: asphalt
concrete pavement.

They investigated the relationship between RMS (Root Mean Square) elevation and

IRI (International Roughness Index); the measured data showed that the correlation

. 2 . .
between them lS weak (R <O.7). It was found that a good correlation exrsts between

the IRI and the PSD only for spatial frequencies between 0.002 and 0.015 cycle/in.

From the same data, Marcondes et a1. [49] found strong correlations between the [RI
and the RMS vertical acceleration at the truck bed. The following models were

developed for these relationships:

- For vehicle speed = 45 mph:

2 3 7 2 2 (7'5)
RMS=3.794x10' +1.902x10‘ xIRI—8.89x10‘ xIRI R =0.937
- For vehicle speed = 52mph:

2 3 6 2 2 (7'6)
RMS=4.467X10’ +2.144x10‘ xIR1—1.819><10" XIRI R =0.914

172

 

- For vehicle speed = 60 mph:
3 6 2 2 (7'7)
RMS=0.105+1.25X10‘ xIRI—1.63><10' XIRI R 20.866

Most researchers have used generated road proﬁles for dynamic vehicle simulation.
Road proﬁles, like any other random signal, can be generated using a random number
algorithm. To generate random numbers, Gillespie et al. [46] used a Gaussian

distribution with a mean value of zero, and the standard deviation is:

1/ 2
S =[fi] (7.8)
2A
Where,
G = White-noise amplitude for one of the three coefﬁcients; Gs, Ge and Ga

Interval between samples used for wavenumber.

A

A simulated road proﬁle that matches the target PSD is generated using the

following procedures:

1) Create an independent sequence of random numbers for each of the three
white-noise sources, scaled according to the above equation.

2) Integrate each sequence as needed to obtain the desired distribution over
wavenumber.

3) Sum the outputs of the ﬁlters.

Thus, the sequence corresponding to the Ga term is integrated twice, the sequence

corresponding to the Gs term is integrated once, while the sequence corresponding to

173

the Ge term is not integrated. Table 7.3 shows PSD coefﬁcients and IRI values used

by Gillespie et a1. [46].

Table 7. 3 PSD Coefﬁcients in the Roughness Model [46]

 

 

 

 

 

 

 

 

Pavement Surface IRI Gs Ga Ge
Type Type (in/mi) (m/cyclexlO'6) 1/(mxcycleX10'6) (m3/cyclex10'6)
Smooth 75 6 0.00 0.000
Flexible Medium 150 12 0.17 0.000
Rough 225 20 0.20 0.003
Smooth 80 1 0.00 0.000
Rigid Medium 161 20 0.25 0.100
Rough 241 35 0.30 0.100

 

 

 

 

 

 

In the case of a rigid pavement, faulting and curling/warping should be considered.

The slab roughness between joints has similar characteristics to that of a ﬂexible

pavement; and, therefore, the periodic faults caused by the slab joints are

superimposed on to this model. The resulting road proﬁle over a slab length is:

y(x)=yr(x)+y,f(x)

Where,

er x)
yijC)

The proﬁle due to the slab roughness

(%)x for 0<x<L

0

for

x=L

Joint fault magnitude

174

(7.9)

 

 

L = Joint spacing

Thermal and moisture gradient across the slab thickness result in signiﬁcant bending

moments along the edges of the slab. The curling/warping is modeled as a periodic

hemispheric wave added to the above road model:

[.-gf.[yw(.)_g_ejf-Rz. Regime)

Where,

5 = Mid-slab deﬂection due to warping

R = Radius of curvature of slab

yw = Vertical displacement due to warping
L = Joint spacing

The ﬁnal road proﬁle is:

y(X) = yr (10+ny (x)+yw(X)

7.2.1.2 ISO 8608-1995

(7.10)

(7.11)

The road roughness proﬁles represent stochastic processes depending on the spatial

road coordinate x of the travelling vehicle. The evaluation and processing of numerous

measurement data of road roughness proﬁles shows a common tendency allowing

some standardization (see ISO-8608 [50]). The one—sided power spectral density

(PSD) is a simple but often used road roughness model. All the models described

above are included in the ISO standard and they generate road proﬁles as white noises.

White noise processes are stationary random processes having a constant spectral

density function. Also, white noise processes are characterized by inﬁnite variances

175

and, thus, they are not realistic. Most of the models described above were used to
generate road proﬁles for roughness prediction purposes. Their objective was to
implement an algorithm that is computationally efﬁcient. Assuming white noise
vehicle excitations will reduce considerably the computation effort for determining
vehicle random vibrations, i.e. the dynamic load applied to the pavement. However, in
vehicle durability where the focus is on the vehicle random vibrations, this assumption
is not true. At most, the vehicle excitations will be modeled as a white noise velocity
spectrum. In common practices, the road proﬁle is approximated by a colored noise
(i.e., the response of a ﬁrst-order system to a white noise). In 1979, Robson [7]
suggested the following model to generate the pavement surface roughness proﬁles,

which was included in the ISO 8608-1995:

—n

 

Su(k)=clk (7.12)
Where,
Su(k) = Displacement spectral density, m3/cycle
n = 2.5
= L n=0,1,----(N—1)=Wavenumber
k NA
N = Number of samples in the proﬁles
A = The distance interval between successive ordinates of the surface proﬁle

Characterizes the roughness level

('3
H

The constant c in Equation 7.12 was found to be correlated with the IRI [52] and

could be estimated using Equation (7.13):

176

l 3
C =1.69X10'8 (IRI)2 (m/zcycleéj (7.13)

_ To generate a random road surface proﬁle, a set of random phase angles uniformly
distributed between 0 and Zn is applied to the desired spectral density. Then, the
inverse discrete Fourier transform was applied to the spectral coefﬁcients [7]. The

general form of the road proﬁle is given by Equation 7.14:

N—I (6,, +21%?)
u(r)=Z,/S,, Xe , r=0,1,2, ....... (N—l) (7.14)
:0 .
Where,
Sn 27:
= [NAJXSu (7n)
S (7 ) = The target spectral density (Equation 7.12)
u n
y" = ZAf—Z— = Wavenumber in rad/m
A = The distance interval between successive ordinates of the surface
proﬁle
6" = A set of independent random phase angles uniformly distributed

between 0 and 271'

7.2.2 Artificial Generation of Roughness Features

Recall that the focus of this analysis is to provide guidance on acceptable roughness
levels to limit user costs. The identiﬁed roughness features that were proven to affect
vehicle suspensions are: Faulting, breaks and curling in concrete pavements and

potholes in asphalt pavements. To investigate their effect, these roughness features

177

 

 

were artiﬁcially generated and superimposed on to the generated road surface proﬁle.

Figure 7.1 presents schematic description of roughness features.

 

 

 

 

............. C ,' .‘ 1 - . ‘ ‘ ." .‘l I A" 4'- I
. ............................................................ z.-

(b) F u.“ 1‘

 

 

  

        

3'33 1:3 \\\“
_' . . \ .
3:75? :35

. l .1, Q
‘._- ‘ ‘4 __ ‘». . _ .‘ . \‘a‘ .3 V" ‘ ‘7. . \ -.
by? 534%x\‘\‘3‘.~\\\\\\\.\‘.\\¢\\\v-;\s :- :«r-m

(c) Break ((1) Bump

      

 
 

'x'
.‘l
f/

, ................ ' ............... - ......................... fa- .- .r .4 x ................................ . ---------------------------------- ‘ ........

l (e) Pothole

 
        

Figure 7.1 Schematic Description of Roughness Features

7.2.2.1 Artiﬁcial Generation of Faults

Faulting is the difference in elevation across a joint or crack. It is determined by
measuring the difference in elevation between the approach slab and the adjacent slab.
In the current practice of road surface proﬁle measurement in the US, the reporting
interval for elevation is 0.025 to 0.075 m (1 to 3 inches). Based on previous study by
Chatti et a1 [44], for a sampling interval of 0.019 m and a reporting interval of 0.075
m, the correct height of a fault is detectable when it is calculated as the difference in
elevation between points that are 0.15 m apart. Accordingly, the width of a fault was
taken as this value. The general form of a fault is given by Equation 7.15. Figure 7.2

shows the form used to ﬁnd the mathematiCal description of a fault.

178

Where,

uf(x)

l‘INh“?

0.15

 

 

 

__ ——x or

1 0.15 f
h

\l—O.15

= ——Oﬁ5x for
< h

——L

L—O.15(x ) for
kO for

 

Road proﬁle due to fault

OSxSOJS

(x—l) for ' 0.153xSl

OSxSO.15

0.153xSL

xZL

= Number of faults per 1.6 km
Joint fault magnitude in mm

 

Distance between faults
Joint spacing,

4.6 m for Jointed Plain Concrete Pavement (JPCP)
12.5 m for Jointed Reinforced Concrete Pavement (JRCP)

Fault width in m

Figure 7.2 Proﬁle of a Faulted Rigid Pavement

179

(7.15)

  

The resulting road proﬁle over 1.6 km is:

N f—l
“(x)=“r(x)+ Z “f(x—jl) (7.16)
1:0
Where,
u(x) = Road proﬁle
ur(x) = Road proﬁle due to roughness
uﬂx) = Roughness features (Equation 7.15)
Nf = Number of faults per 1.6 km

7.2.2.2 Artiﬁcial Generation of Breaks/Bumps

Break is a broken portion of the pavement section that starts with a negative fault
and ends with a positive fault. The distance between the two opposite faults should not
exceed 0.9 m (3ft), see [25]. The general form of a break is given by Equation 7.17.

Figure 7.3 shows the form used to ﬁnd the mathematical description of a break. The

same equation is used to generate bumps except that the magnitude is h instead of (—

h).
[_1600
Nb—l
r h
ub(x)= ——x for OSxSOJS
0.15 (7.17)
—h for 0.15.<_xSO.75

th(x—O.9) for 0.7SSxSO.9

 

OM

k for 0.9SxSl

180

u b (x) = Road proﬁle due to break

Nb = Number of Break per 1.6 km
h = Break magnitude in mm

1 = Distance between breaks in m

_ 0.15m

 

-------1- -

   

 

N

I
I
:4
V

Figure 7.3 Proﬁle of a Concrete Slab with One Break

The resulting road proﬁle over 1.6 km is:

Nb —1

u(x)=u,.(x)+ Z ub(x—jl) (7.18)
i=0

Where,

u(x) = Road proﬁle

urfx) = Artiﬁcially generated road proﬁle due to roughness

u bfx ) = Roughness features (Equation 7.17)

Nb = Number of breaks per 1.6 km

7.2.2.3 Artiﬁcial Generation of Curling

Curling is the distortion of a slab into a curved shape by upward or downward

bending of the edges. This distortion can lift the edges of the slab from the base

181

leaving an unsupported edge or comer which can crack when heavy loads are applied.
Sometimes, curling is evident at any early age. In other cases, slabs may curl over an
extended period of time. Curling is described as ellipse and its general form is given
by Equation 7.19. Figure 7.4 shows the form used to ﬁnd the mathematical description

of curling.

2
uc(x)=—hx [1—#]
l

 

 

 

      
 
 

   

(7.19)
l: 1600 S L
Nc —
Where
“c (x) = Road proﬁle due to curling in mm
NC = Number of curling per 1.6 km
h = Curling magnitude in mm
1 = Curling width in m
L = Joint spacing in m
L :
l/2 .
H—-——>:

  

////////////////

Figure 7.4 Proﬁle of a Curled Concrete Pavement

182

The resulting road proﬁle over 1.6 km is:

Nc—l

u(x)=ur(x)+ Z uc(x—jl) (7.20)
j=0

Where,

u(x) = Road proﬁle

arm = Road proﬁle due to roughness

u C(x ) = Roughness features (Equation 7.19)

Nc = Number of curling per 1.6 km

7.2.2.4 Artiﬁcial Generation of Potholes

A pothole is when a portion of the road material has broken away, leaving a hole.
Most potholes are formed due to fatigue of the pavement surface. As fatigue cracks
develop they typically interlock in a pattern known as "alligator cracking". Then, the
pavements between fatigue cracks become loose by continued wheel loads forming a
pothole. The width of a pothole was taken as 0.3 m (1ft). The general form of a
pothole is similar to that for curling and it is given by Equation 7.21. The only
difference is the ellipse width. Figure 7.5 shows the form used to ﬁnd the

mathematical description of a pothole.

 

 

_ 1600
Np—l
r (x—0.15)2 (7.21)
u (x): —hx 1———— for OSxSO.3
p < 0.152

033x31

 

O
35
V

183

“p (x) = Road proﬁle due to potholes
' Np = Number of Potholes per 1.6 km
h Pothole magnitude in m

N
|| 11

Distance between potholes in m

 

 

Figure 7.5 Proﬁle of Asphalt Concrete Pavement with One Pothole

The resulting road proﬁle over 1.6 km is:

u(x) =u, (x)+ 2 up (x—jl) (7.22)
i=0
Where,
u(x) = Road proﬁle
= Road roﬁle due to rou ess
"r09 p ghn
up (x) = Roughness features (Equation 7.21)
Np = Number of potholes per 1.6 km

7.2.3 Dynamic Vehicle Simulation
As comprehensively reported by From [53] and Cebon [7], there exists a number of
numerical models, which have been developed to predict the behavior of vehicles

when traveling on irregular pavement surfaces. The most basic models are based on

184

quarter-vehicles represented by a second-order, two-degree-of-freedom, linear
differential equation (Equation 7 .23), whereby the vehicle response is computed for
the vertical orientation with the pavement surface proﬁle as the excitation function

(Figure 7.5).

r

muiu _CS (is "xu)_ks (xs —xu)+kt (xu —u)=0
1 (7.23)

mSJ'c'S +63 (is —5cu)+ks (x5 -—xu)=0

 

Where,

u ( t) = Road proﬁle

xu = Elevation of unsprung mass (axle)
xs = Elevation of sprung mass (body)
kt = Tire spring constant

ks = Suspension spring constant

mu = Unsprung mass

m s = Sprung mass

CS = Shock absorber constant

In order to account for the more complex behavior of road vehicles, a number of
more sophisticated vehicle models have been developed to describe the dynamic
behavior of half-vehicles (complete axle) and even full-vehicle models [54]. These
have been aimed at predicting the roll and pitch response of vehicles, which have been
found to be signiﬁcant for some vehicle types. It has been suggested, however, that in
most cases, the vertical vibration remains the dominant component of vibration

induced by irregular pavement surfaces [54].

185

 

 

 

 

 

 

 

 

 

 

u(t)

 

 

 

Figure 7.6 Schematic of Two Degrees of Freedom Quarter-Car Vehicle Model

Further improvements of numerical models have been achieved by including the
non-linear characteristics of suspension elements, such as shock absorbers, leaf
springs and air springs [55]. A number of such models have been developed to further
improve the accuracy of vehicle response estimates. One useful application of
numerical vehicle models is the estimation of vertical vehicle vibration, which is
widely accepted as being directly related to vehicle damage. A half or quarter car
model cannot be expected to predict loads on a physical vehicle exactly, but it will
highlight the most important road characteristics as far as fatigue damage
accumulation is concerned; it might be viewed as a “fatigue-load ﬁlter” [7, 53, 56, 57,
58,59,60]

There have been a number of such models with various speciﬁc parameter values
proposed to emulate the response of a wide variety of road vehicles ranging from

small sedans to large trucks with air ride and more conventional steel suspension

186

systems [7]. These models have also been used to predict the response of large
vehicles with different axles in recognition of the differing suspension elements used
for the driver’s cabin and the trailer. ‘Quarter car’ parameters for a passenger car [61]

and full truck with typical air and steel suspensions from [7] are given in Figure 7.7.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Model parameters
Constants Truck [4] Car [69]
Name Unit Steel Air - ms FLXS
Kg 4500 4500 250
ms ks % F4 03
Kg 500 500 40
mu
mu ixu
k MN/m 1 0.4 0.028
S
ku
kt MN/m 2 2 0.125
Cs KNs/m 20 20 2 u(t)

 

 

 

 

 

 

 

 

 

 

Figure 7.7 Schematic of Two Degrees of Freedom Quarter-Car Vehicle Model [4,69]

From Equation 7.23, the state space equations can be derived by proper assumption

of sate variables. Let us assume state space variables as y] = 565;)?2 = xs;y3 = 5C“ ;
and y4 =xu. Substituting these space variables into Equation 7.23, the state space

equations can be written as:

187

 

 

 

 

 

 

 

 

_ f 0 \
(yl \ ms ms ms ms (yl \ 0
y2 l 0 0 0 y2

= x + kt Xu (7.24)
3’3 _C_S_ _ki -38. _(kS.+..k.t._) “V3 7,;
(F4) mu mu mu mu (Y4) 0
K J
( 0 0 l 0 j

A simple generic linear numerical quarter-vehicle model was developed to compute
the vertical vibration level of typical vehicle types from different pavement proﬁles at
constant speeds. This numerical model developed with the Matlab/Simulink®
programming environment effectively computes the solution to the two degrees of
freedom system using the ﬁxed-point method. The inputs to the model are the

longitudinal pavement proﬁle and the velocity of the vehicle.

7.2.4 Vehicle Fatigue Damage Analysis
A common laboratory experiment to predict fatigue damage is to subject test

specimens to a sinusoidal load with amplitude U, and count the number of cycles N to
breakdown. Commonly, the simple parametric model N (U) = C—1U_ﬂ (Basquin’s

relation) is ﬁtted to experimental data. Usually, for vehicle components, the fatigue
exponent [3 takes values between 3 and 8 [57, 58]. A typical [3 value for steel
suspensions is 6.3 upon a “half-life” rule [62]. The “half-life” rule states that the half
life of a steel component will be approximated at a 10 to 12% increase in cyclic load

amplitude.

188

Loads caused by road roughness ﬂuctuate randomly. To assess the fatigue damage, it
is necessary to extract cycles from the load sequence. The load sequence on the sprung
mass of the vehicle model is rainﬂow-counted, to extract the load cycles U). The
rainﬂow counting method was introduced by Endo in 1968 [63]. A simpliﬁed
equivalent deﬁnition was given by Rychlik in 1987 [64]. This deﬁnition (stated below)

can be used to extract cycles straight-forwardly, see also Figure 7.8.

Deﬁnition (Rain/low cycle). Each local maximum in the load sequence
is paired with one particular local minimum, determined as follows

. .h . .
(Figure 7.8): From the It local maxrmum (value Mi) one determrnes
the lowest values in forward and backward directions between M,- and

the nearest points at which the load exceeds Mi. The minima mi‘ and

1121* on each side are identiﬁed. The larger (less negative) of those two

values, denoted by mlRFC, is the rainﬂow minimum paired with M,- ,

i.e. ml-RFC is the least drop before reaching the value M,- again on

. . .th . . .
either srde. Thus the l rarnﬂow pair rs (Mi,mlRFC) and the

rainﬂow amplitude is U i =M i —ml-RF C.

 

Figure 7.8 Schematic Deﬁnition of the Rainﬂow Cycle as Given by [63]

189

The rainﬂow-counting algorithm was written in C programming language. The

algorithm was then compiled to executable ﬁles and called in MATLAB ®

environment.
Palrngren-Miner’s linear accumulation hypothesis is used to estimate fatigue

h
damage. Thus, the damage caused by the jt cycle equals 1/N(Uj),

[_ti

J . The total fatigue damage caused by the rainﬂow-counted

whereUj =Mj—m

load sequence is:

D = C2 Ufa (7.25)
l

Computations of fatigue damage from a given load. sequence was performed using

MATLAB ® environment.

7.3 RESULTS AND CONCLUSIONS

To illustrate the various features of the method described above, several conditions
have been examined. All road surface proﬁles were artiﬁcially generated at every 0.07
m. Road surface proﬁles were ﬁltered out using a moving average ﬁlter with a
baselength of 0.25 m for cars and 0.3 m for trucks representing the tire enveloping.
Then, the ‘quarter car’ vehicle models traveling at a constant speed of 110 km/h were
applied to a 1.6 km of road surface proﬁles. The parameters used in the ‘quarter car’

model are given in Figure 7.7 above.

190

7.3.1 Suspension Failure Threshold

Based on personal communications with vehicle repair specialists [65], vehicle
owners tend to replace their suspensions at about 160,000 km for cars and 400,000 km
for trucks. This value was also reported as the lifetime warranty for suspensions given
by vehicle manufacturers in the US. Consequently, in this study, the average life of
cars and trucks suspensions for typical driving conditions was assumed to be about
160,000 km and 400,000 km, respectively. Moreover, truck suspensions are not
replaced when they actually fail but when certain signs of wear become evident and,
consequently, compromise the safety and comfort of drivers. The amount of service
life used at which owners are urged to replace their suspensions was estimated using
the following procedure:

1. Estimate the roughness (IRI) distribution of US roads (Figure 7.9), with l
m/km corresponding to a smooth road and 6 m/km to a very rough road;

2. Generate 30 road surface proﬁles for each of the IRI values;

3. Calculate the accumulated damage (Di-IRI) induced by each of the road proﬁle

generated in step 2 for a length of 1.6 km and assuming a value of 6.3 for ,8 in
Equation 7.25;

4. Take the average value of the accumulated damage calculated for each proﬁle
set having the same IRI level.

5. Estimate the number of kilometers per IRI (MIR!) value using the distribution
obtained in step 1 and assuming that the road network is 160,000 km and
400,000 km for cars and trucks respectively.

6. Compute the total accumulated damage using Equation 7.26:

191

(7.26)

 

 

Using the above procedure, the value for Dreplace is about 87.3% for cars and 62.2 %

for trucks.

 

 

 

 

   

 

4.3o
4’ 3.1% 2.5% 2.3% 1.7%

Probability density function (%)

 

 

1 1.5 2 2.5 3 3.5 4 4.5 5 6

IRI (mlkm)
Figure 7.9 Road Surface Roughness Distribution in the United States

7.3.1.1 Suspension Damage in Cars

Figure 7.10 shows the damage accumulated in car suspensions after 160,000 km of a
road with a given (constant) IRI value. This damage is obtained by multiplying the
damage calculated in step 3 above (for cars) by 160,000. The error bars show the error
in accumulated damage caused by the variations in the proﬁles; i.e., different proﬁles
will generate different suspension vibrations even though they may have the same IRI.

Based on personal communication with SoMat [66], many car manufacturers design

their vehicles for the 90th -95th percentile road roughness. According to Figure 7.9,

192

3.9 m/km is the 93rd percentile of the roughness distribution in the US. From Figure

7.10, when a car is driven for 160,000 km of road with IRI = 3.9 m/km, the

accumulated. suspension damage will be about 84.5%, which is very close to 87.3%.

 

 

 

l[_’ ' I e; ' A’ ‘i ‘"' ’ I—if 1
, , , , I,
i I ‘ 1 , ; I l
a) 0'8[ 1" ; ********** ? ---_.
go | I i 1 i . l ; ‘
§ ' : ; I : = ' 1
“60.61““ t 7 ; l : t I
“U _ . I
B ' ;
3 . ; 1 i
:1 . I i ‘. f ,
o I
2 f .f
0.2 “ ‘ [I] 1
i ; r
0‘. - 2 222ﬂ¢::11_____-___ ..__ , __j
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
IRI (m/km)

Figure 7.10 Effect of Road Surface Roughness on Car Suspension

Actual road proﬁles from in-service pavements were used to check the accuracy of
the approach discussed above. Table 7.4 summarizes the pavement conditions of the
different proﬁles used in this study. Figure 7.11 shows the accumulated suspension
damage for cars induced by both real and artiﬁcially generated proﬁles at each
roughness level. It can be seen that the results from real proﬁles follow the general

trend of the curve generated using artiﬁcial proﬁles, with some scatter around the

CUI'VC.

193

It is believed that the variance is caused by the actual content of the proﬁles. For
example, the two proﬁles highlighted in Figure 7.11 have an IRI of 2.4 m/km (proﬁle
3 and proﬁle 6); however, proﬁle 3 causes more damage than proﬁle 6. According to
Table 7.4, proﬁle 3 has a break while proﬁle 6 has only low severity faulting and

curling.

The additional cost induced by roughness features when the accumulated damage

reaches Dreplace was assumed to be equal to the price of a new suspension and the

labor hours for replacement. This value is equal to $1,000 according to auto repair

specialists [65].

 

 

   

 

 

 

 

 

   

 

 

1--— I r ; _,_ v H
; i —-Artiﬁcial proﬁles _ , i
0. 8 ' ‘7‘ “Realproﬁles * f _p _ _ g _ _____ ‘ ______ .
a) 1 I
g0 I :
"O ' .
9. I
.59.
E
:5 0'4: 3
O
o .
< I
0.2“ , J 5
0_ ::':‘ : :-: H‘: t - 49.912227110239537 3.2—**. I _3._ -__.-J
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

IRI (m/km)

Figure 7.11 Accumulated Car Suspension Damage Using Real and Artiﬁcial Proﬁles

194

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

- 2:. EN- EN. <2 <2 <2 <2 <2 <2 <2 955 8:?me 8ng
o e 2 m <2 <2 <2 <2 <2 <2 <2 258 8658
- - - - - 2. - - - mm 2. Ass 82: ease; E
@520
- - - - - SN - - - a S 2:80
- - - - - - no 3 .o - - ca 523
- - - - - - 8.2- on- - - Aeavueaaammﬂ £85
. i - i i I N _ i .. 3:500
- - - - we- 2. _ 2:- mm- mm- «.m. 28% 82.64
- - - - S. 2 E e: 3 Ni 4.2 €25 836%
- - - - o o o o o o _ 21mm: 825
- - - - 2 o _ we 2 e 2 magi 8.58
- - - - 8N em 2 o: R <2 82 33
2 8-2 3-2 3.23 3-2 3-2 8.. Elm 3-3 3-2 2-2 322328?
2.0 2 3 so we 3. m3 3 8.0 3o 3o 2% emeﬂ
o< o< u< o< 8m 8m 08 RE R: 08 on: 25 28822
_mwﬁm em 2883 3:320 282 mg: as _mwﬁm $2 $2 $2 $2 eeom
: 2 a w e e m e m N 2 8%

 

 

 

 

 

 

 

 

 

 

 

 

 

$5838 0228-5 Eat moEoﬁ cuntsm .«o 52.286285 2.3885 v.5 2an

195

7.3.1.2 Suspension Damage in Trucks
Figure 7.12 shows the damage accumulated in truck suspensions after 400,000 km of

a road with a given (constant) IRI value. This damage is obtained by multiplying the
damage calculated in step 3 above (for trucks) by 400,000. The error bars show the
error in accumulated damage caused by the variations in the proﬁles; i.e., different
proﬁles will generate different suspension vibrations even though they may have the

same IRI.

 

 

 

 

 

 

1 r 1 ' H V 2* #2 T 77* ' "1' _.
a) 0. 8 ——————————————————————————————————————————— I 75] r J
DD . ,1
cc: ; ,1"
g . : 2i:
1: 0.6 —————————————————————————— 1, ,1 — 1
13 ' /
a) /
H I /
"a 1
g 0.4 1" " ‘ """"" ‘ :14 2

1/1
< i
0. 2 f 17;,” _
0 1. ..._,._ .2221 2, 2_. ..'222k-.2-—,:,—o—__:r" :‘fj’ﬁ _ a l __. V, J _,2-,, _ __ _ .
O O. 5 l 1.5 2 2. 5 3 3.5
IRI (m/km)

Figure 7.12 Effect of Road Surface Roughness on Truck Suspension

Based on personal communication with PACCAR [67], many truck manufacturers

design their vehicles for the 80th -95th percentile road roughness. According to Figure

7.9, 3.2 m/km is the 87th percentile of the roughness distribution in the US. From

196

Figure 7.12, when a truck is driven for 400,000 km of road with IRI = 3.2 m/km, the
accumulated suspension damage will be about 66%, which is also very close to 62.2%.
Therefore, we proved using two different methods that a value of 6.3 for ,8 is
reasonable for both cars and trucks. Accordingly, a typical value for ,B is taken as this
value.

Real road proﬁles were also used to check the accuracy of the approach discussed
above. Figure 7.13 shows the accumulated suspension damage for trucks induced by
both real and artiﬁcially generated proﬁles at each roughness level. Similar
observations can be made; i.e., the results from real proﬁles follow the general trend of

the curve generated using artiﬁcial proﬁles, with some scatter around the curve.

 

 

 

 

 

 

 

 

 

1 1 . : ______,_ “ *' nu“? " l
E ——Artiﬁcial proﬁles E .
; ‘13“ Real pro ﬁles E ; E . 5;.
0.8' “f '“ ..222-..2-2..2.22,22.__.. 2- 22 ,2 22 2 .2 2.2 . __ ,.1
o l
2:”
"o 0.6 r * 5
‘U
o
33
:3
E 0.4
:3
o
0
<1
0.2 ~ 2
l
0 G _-
0 . 4
IRI (In/km)
Figure 7.13 Accumulated Truck Suspension Damage Using Real and Artiﬁcial
Proﬁles

197

The additional cost induced by roughness features when the accumulated damage

reaches Dreplace was also assumed to be equal to the price of a new suspension and

the labor hours for replacement (i.e., $3,000 and $1,800 for air and steel suspensions,

respectively, according to auto repair specialists [65]).

7.3.2 Mechanistic versus Empirical Approach

Figures 7.14 and 7.15 show the results from (1) the mechanistic-empirical approach
using actual proﬁles, (2) the mechanistic-empirical approach using artiﬁcially
generated proﬁles, and (3) the empirical approach (i.e., Updated Zaniewski’s tables)
for cars and trucks respectively. While roughness effects below 3 mlkm are minimal,
we still see a constant R&M cost, which corresponds to other routine maintenance
costs that are not related to roughness. We, therefore, shifted the curves from M-E
analysis to match the empirical curves in the IRI range below 3 m/km. The results
from the mechanistic-empirical approach were compared to the empirical results (i.e.,
Updated Zaniewski’s tables), and were found to be very close until 5 m/km. These
results seem promising since the typical IRI range in the US is between 1 to 5 m/km.
Therefore, for pavement management at the network level, considering only IRI as
measurement of pavement conditions is enough. However, at the project level, the

effect of roughness features should be included.

198

80 2222222 2. , ,- - 22.222222 -2.-.. 222.2 2222222_ 222222222222- ._ _ .2222“.
. 1 . 5 .2

._-___ 2____ .‘

 

1 1 Cr M-E with actual proﬁles 3 J
75’” ' .—M-Ewithartiﬁcialproﬁles 'L' ______
70 ******

L—‘mELngiﬁcalzrlgﬁsvxaki/HDM‘EJ -- _ .- ,_ i ......... E

 

Repair and Maintenance cost ($/1000km)

 

 

Figure 7.14 Comparison between Car Repair and Maintenance Costs Calculated
Using M-E Approach and the Updated Zaniewski Costs

DJ
8

i 55—5--~—-— - 2_2 “-2212. 'f—_—r‘ ____ 2,
. 2 222 _— — -— — 1 ,1

—— Emp iric a1: Zaniews ki/HDM4 ;
M-E with artiﬁcial proﬁles 1 :1 E

N
U}
o.

yyyyyy

M-Ewithactualproﬁles E

 

 

 

Repair and Maintenance costs ($/1000 km)
5?
_9 2 _

100 ~ 3
i
501--
E . 1
L ____ _s_ ..... _l
0O l 2 3 4 5 6
lRI(m/km)

Figure 7.15 Comparison between Truck Repair and Maintenance Costs Calculated
Using M—E Approach and the Updated Zaniewski Costs

199

7.3.3 Case Study 1: Effect of Faulting on Suspension Damage

The ﬁrst case study examines the effect of different combinations of faulting levels
and frequencies per 1.6 km on car and truck (both air and steel) suspensions. The
faulting was modeled as a rising step function of variable amplitude and a width of
0.15 m (Figure 7.2). The vehicle speed was the same for all the runs. Two concrete
pavement types were considered: Jointed plain concrete pavement (JPCP) and Jointed
reinforced concrete pavement (JRCP). JPCP is the most common type of rigid
pavement. It controls cracks by dividing the pavement into individual slabs separated
by contraction joints. Slabs are typically between 3.7 m (12 ft.) and 6.1 m (20 ft.)
long. JRCP slabs are much longer (as long as 15 m (50 it.» than JPCP slabs; so JRCP
uses reinforcing steel within each slab to control cracking. This pavement type is no

longer constructed in the US. due to some long-term performance problems.

7.3.2.1 Damage to Car Suspensions

Figure 7.16 shows the additional car suspension repair and maintenance costs
induced by different levels and frequencies of faulting in both JRCP and JPCP. It was
noted that faulting in JRCP pavements affect car suspensions more than in JPCP. To
explain this phenomenon, spectral analyses of surface proﬁles of both JRCP and JPCP
pavements with similar roughness features were performed. It was noted that the
dominant frequency of JRCP pavements is closer to the resonant frequency of the
vehicle than that of JPCP pavements. The higher the number of faulting per slab, the
closer both dominant frequencies will be, which explains the reduction of the gap

between curves.

200

Repair and maintenance cost

E

($/1000

$20 -
$18 1
$16 '-
$14 1
$12 'j
$10 1
$8

 

$6

$4 5
$2
$0

 

P

JPCP

 

 

1111+?

+ 50% of slabs faulted
—A— 100% uncracked

-9— 100% slabs with 1 crack
—8— 100% slabs with 2 cracks
—-+— 100% slabs with 3 cracks

 

 

 

,// a
/ 2: '/2 ﬂ _i _ _
5 10 15
Fault magnitude (mm)

Figure 7.16 Car Repair and Maintenance Costs Induced by Different Levels of

Figures 7.17 and 7.18 show the additional repair and maintenance costs induced by

obviously increases.

Faulting

7.3.2.2 Damage to Truck Suspensions

different levels and frequencies of faulting in JRCP and JPCP, respectively. The costs
were expressed as dollars per 1000 km. As expected, the effect of faulting on air
suspensions is lower than on steel suspensions because air suspensions provide more
vibration isolation. However, the additional repair and maintenance costs for air and
steel suspensions are close because replacing air springs is much more expensive than

replacing a steel suspension. Also, as the number of faults increases, damage

201

$25 3’

 

 

$20 '

Steel Air

—o— + 50% of slabs faulted

+ + 100% uncracked

—o—- —e— 100% slabs with 1 crack
—0— —a— 100% slabs with 2 cracks
+ —+— 100% slabs with 3 cracks

 

 

a

0

d)

E

5’5" 1
EOW“
§§
§§$10:
H

'3

Q,

0)

ad

$51

$0 --+5

it
n

/.

q
/' 5
' ' '1 ’o
’ ./
7"" /T/:

5
Fault magnitude (mm)

 

A
/

10

Figure 7.17 Truck Repair and Maintenance Costs Induced by Different Levels of

Faulting — JRCP

202

$20

$16
$14
$12

$8
$6
$4
$2

Repair and maintenance cost
($/1000 km)

$0

Figure 7.18 Truck Repair and Maintenance Costs Induced by Different Levels of

$18 1

 

$10

____1._ 2221,,

~

 

 

11$

50% of slabs faulted
100% uncracked

100% slabs with 1 crack
100% slabs with 2 cracks
100% slabs with 3 cracks

+1111?-

 

 

 

4 6 8
Fault magnitude (mm)

Faulting — JPCP

203

7.3.4 Case Study 2: Effect of Breaks on Suspension Damage

The second case study examines the effect of different combinations of
breaks/potholes levels and frequencies per 1.6 km on car and truck suspensions (air
and steel). The breaks were modeled as a step function of variable amplitude and a

length of 0.9 m (Figure 7.3). The costs were expressed as dollars per 1000 km.

7.3.3.1 Damage to Car Suspensions

Figures 7.19 show the additional repair and maintenance costs induced by different
levels and counts of breaks in both JRCP and JPCP. The observations are in two-folds:
(l) Breaks naturally cause more damage than faults; and (2) Damage increases with

increasing frequency of breaks.

 

number of slabs between
breaks
$16

. JPCP JRCP
§ 9 +120 40
q, j 43— 60 20
g A512 ' —11— 30 10
5 5 + 15 5
a 91(- 8 3
.1; § 18 l . *
'o S '
5 63 |
.t: _1
(lg; $4 E
94 3

0 5 10 15

Break magnitude (mm)

Figure 7.19 Car Repair and Maintenance Costs Induced by Different Levels of
Breaks

204

7.3.3.2 Damage to Truck Suspensions

Figure 7.20 shows the repair and maintenance costs for different severity levels of
breaks. The effect of breaks on air suspensions is much lower than on steel
suspensions. The difference in damage between air and steel suspension is even

greater than the one reported in section 7.3.3.2 (i.e., air suspension is even less

    
  

damaged than steel suspensiOHS). waer of Slabs
betWeen breaks
Steel Air
$16 E JRCP
a + 20 —a— 20
5 E 1
E E E ‘ JPCP
“’ 8 $8 -r
2 £79 —a— 60 _B_ 6 0
:3 v + 30 + 30
.23 —>e— 15 ___,_ 15
a $4 1 l + 8 _h 8
m E I "/'4
$0 E z . I : h‘. .v m‘ _ if
0 5 10 15
Break magnitude (mm)

Figure 7.20 Truck Repair and Maintenance Costs Induced by Different Levels of
Breaks

7.3.5 Case Study 3: Effect of Curling on Suspension Damage
The third case study examines the effect of different combinations of curling
magnitude and frequencies for 1.6 km of concrete pavements on both car and truck

suspensions. Curling is modeled as an ellipsoid of variable amplitude and width. The

205

curling width is assumed any value between 3 m (minimum) and slab length

(maximum).

7.3.4.1 Jointed Reinforced Concrete Pavements
The slab width for JRCP is 12.5 m (41 ft). These conditions were analyzed for (1)

uncracked slabs, (2) slabs with one mid-panel crack, and (3) slabs with 2 mid-panel
cracks. Figure 7.21 shows the additional repair and maintenance costs for car
suspensions caused by different severity levels of curling. According to the ﬁgure,
damage increases with increasing frequency of cracks on the pavement (i.e, smaller
curling width). This could be explained by the shift of the dominant frequency of
curled proﬁle with multiple cracks closer to the resonant frequency of car suspensions.

Figure 7.22 shows the additional repair and maintenance costs for truck suspensions
caused by different severity levels of curling. The difference in damage between steel
and air suspensions is signiﬁcant. Since the road isolation ability of (properly
maintained) air ride suspensions is higher than leaf spring suspensions, they will
absorb more energy induced by the vertically accelerated wheel, allowing the frame
and body to ride undisturbed while the wheels follow the bumps/depression in the
road. This difference is less signiﬁcant for curled slabs with multiple cracks. Damage
in steel suspension is relatively insensitive to crack frequency. This has to do with the
“resonance effects” of suspension vibrations versus frequency content of a curled

proﬁle.

206

Repair and maintenance cost
($/1000 km)

,0.) $16
8 $14 .
d)

g ,2 $12
.125 $10 1
'23 8

E S

3 <3;

.9:

<6

8

a:

 

 

"°“ uncracked slab

43— slab with 1 crack
202 slab with 2 crack

 

 

$8
$6
$4 '
$2
$0

 

 

 

 

 

2 4 6 8 10 12
Curling magnitude (mm)
Figure 7.21 Car Repair and Maintenance Costs Induced by Different Levels of
Curling 2 JRCP
Steel Air
-<>- —9- uncrac ked slab
—n— —a— slab with 1 crack
.0. .9. slab with 2 crack
, 4' /: 2 f .
2 4 6 8 10 12 14 16
Curling magnitude (mm)

Figure 7.22 Truck Repair and Maintenance Costs Induced by Different Levels of
Curling — JRCP

207

7.3.4.2 Jointed Plain Concrete Pavements
The slab width for JPCP is 4.5 m (15 ft). Since, the curling width is assumed any

value between 3 m and 4.5m, a maximum of one crack per slab is considered in this
case. All slabs were considered curled. In addition, these conditions were analyzed for
the cases where 20, 30, 60, 90 and 100 percent of slabs are cracked.

Figures 7.23 and 7.24 show the additional repair and maintenance costs for car and
truck suspensions caused by different severity levels of curling. It was noted that the
additional repair and maintenance costs caused by curling in JPCP pavements is lower
than those induced in JRCP pavements (Figures 7.2] and 7.22). We believe that the
slab length being smaller for JPCP induced higher dynamic loads. This increase led to

additional damage.

Percent slabs cracked

 

 

Egg $101 5;- 0% +20%;

8 $8 .2! '+30% +60%

g E , E+90% —~>e—100%J /

3 $6 1; ‘ " “ /

.S o i /

E 8 E " /

'0 : $4 1 /

5 a . /

g $2 I ///

3 $0 1_____ ,_ /2 -’ ...__--___ —1 ﬂ 7‘—

Od .

0 2 4 6 8 10

Curling magnitude (mm)

Figure 7.23 Car Repair and Maintenance Costs Induced by Different Levels of
Curling — JPCP

208

Percent slabs cracked
Steel Air

 

 

 

 

 

8 $18 , +20% +20%

8 164 +60% +60%

5 AS 4:199%-*100%

... o $12 -1

E o $10 4. °

E 2 $8 ~ "

g .5 $6 a

.2: $4 “ /

(6 $2 4 /

Q. " :

g; $0~ / .-———ﬁ
0 2 4 6 8 10

Curling magnitude (mm)

Figure 7.24 Truck Repair and Maintenance Costs Induced by Different Levels of
Curling — JPCP

7.4 SUMMARY AND CONCLUSIONS

In this chapter, we proposed a novel approach to estimate repair and maintenance
costs induced by roughness. The approach proposed herein uses a mechanistic-
empirical approach to conduct fatigue damage analysis using numerical modeling of
vehicle vibration response. To check the accuracy of the method, the accumulated
suspension damages induced by (l) artiﬁcially generated proﬁles and (2) real roads
proﬁles from MDOT and LTPP databases were compared. The results showed that the

computed accumulated damage using real proﬁles follows the curve using generated

209

proﬁles with some scatter around the curve. The variance around the curve is caused
by the difference in real proﬁles roughness contents.

The results from the mechanistic-empirical approach were also compared to the
empirical results (i.e., Updated Zaniewski’s tables), and were found to be very close
until 5 mlkm. These results seem promising since the typical IRI range in the US is
between 1 to 5 m/km. It should be noted that all the models currently in use are
empirical models. The only purely mechanistic model is the VETO model which was
developed by the Swedish Road and Trafﬁc Research Institute (VTI). However, they
adopted an empirical model because their predicted change in vehicle wear with
increasing roughness is far higher than the change in parts cost predicted by the
empirical model. Therefore, we can conclude that the mechanistic-empirical approach
proposed herein is an improvement to the state-of-art. Also, we can also conclude that
considering only IRI as measurement of pavement conditions is enough for pavement
management at the network level. However, at the project level, the effect of
roughness features should be included. This could be done using the same approach.

The approach reported in this chapter could help in better estimating vehicle
operating costs at the project and network level. For routine application, 3 DOT would
need to run a given proﬁle (for a given project) through our program. It generates the
accumulated damage for a given suspension type assuming a life of 400,000 km for

trucks and 160,000 km for cars.

210

CHAPTER 8
IMPACT OF ROUGHNESS FEATURES ON DAMAGE TO GOODS

8.1 INTRODUCTION

The riding quality of a road has, for many years, been used as the primary indication
of the quality of a road. This is mainly due to ﬁndings that most of the deterioration in
the road structure ultimately translates into a decrease in ride quality of the road. This
decrease can be attributed to increased vibrations of a moving vehicle.

The vibrations from the road are transmitted to the transported cargo, resulting in
damages to the cargo. Potential solutions to this problem include improvements to the
packaging of the cargo or improvements in the design of the cargo itself. Both these
potential solutions add another cost to the logistics of the operation, since every piece
of cargo requires improved design or packaging with the sole objective of improving
the trip between the supplier and the customer. This does not add any direct value to

the product, but it merely causes an additional cost.

The objective of this chapter is to develop a methodology to estimate the effect of
roughness features on damage to goods and to provide guidance on acceptable
roughness levels. Accordingly, we propose using a mechanistic-empirical approach to
conduct fatigue damage analysis using numerical modeling of vehicle response and
product vibrations. A sensitivity analysis was performed to quantify the relationship
between roughness feature height and width to damage to goods. The analysis consists

of the following steps:

211

 

1. Artiﬁcial generation of road surface proﬁle;

2. Artiﬁcial generation of roughness features (bumps, depressions and curling);

3. Estimation of the response of the vehicle to these transient events;

4. Estimation of the product vibration to these transient events

5. Computing of the induced damage to transported goods;

6. Repeat step 2 through 4 for different heights and frequencies of roughness

features.

8.2 RESEARCH APPROACH
8.2.1 Artificial Road Profile and Roughness features Generation

Road proﬁles were generated using the methodology described in chapter 7.
Equations 7.12 through 7.14 were used to generate road surface proﬁles. Equations

7.15 through 7.22 were used to generate roughness features.

8.2.2 Dynamic Vehicle Simulation

8.2.2.1 Vehicle Model
As comprehensively reported by Schoorl [73], Burgess [74], Marcondes [75], and

Singh [76], there exists a number of numerical models, which have been developed to
predict the behavior of products during transport. These models are overall road-
vehicle-load systems. In general, trucks are modeled as a two-axle vehicle [77]. This is
shown in Figure 8.1. The body of the vehicle is represented by the mass, M1, with
moment of inertia, I, about its center of gravity. The mass of each axle and wheel is
generally referred to as its unsprung mass (M2, M3). The springs are represented by

stiffness terms K1 and K2. Their load-deﬂection characteristics are assumed to be

212

 

linear. The damping of the springs is represented by Coulomb ﬁiction terms (B1, B2)
and by various damping terms (C1, C2). If a shock absorber (hydraulic damper) is
included in the system under consideration, its value would be included in the terms
Cl and C2. As the shock absorbers are designed to have quite different damping
characteristics in the “bump” and the “rebound” directions, Cl and C2 can have
different bump and rebound values. The tires are represented by stiffness terms, K3
and K4, and their damping by viscous damping terms, C3 and C4. Parameter values

for a “standard vehicle” are given in Table 8.1.

 

 

 

he;

k———>n——+

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

M1 9 I i 9
1321K2 +c2 Bl 1 K1§, I”'(31
M3 M2

 

 

 

 

 

 

 

 

 

u(t)

Figure 8.1 Typical mechanical model of a half truck

213

 

Table 8.1 Parameter values for a “standard vehicle” [77]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Notation Values Unit
Moment of inertia about CG of I 31000 Kg m-
body
Mass of the vehicle body M1 5395 Kg
Distance from front axle to CG of R 3.5 m
body
Distance from rear axle to CG of S 1.09 m
body
Mass of front axle M2 336 Kg
Mass of rear axle M3 1000 Kg
Front spring stiffness K1 250000 N m'1
Front spring viscous damping C1 (bump) 1000 N m'1 3.1
Cl (rebound) 4000
Front spring Coulomb damping B1 2000 N
Rear spring stiffness K2 1295000 N m-1
Rear spring viscous damping C2 (bump) 4000 N mIsJ
C2 (rebound) 4000
Rear spring Coulomb damping B2 4000 N
Front tire stiffness K3 1564000 N m‘1
Front tire viscous damping C3 1000 N m'1 8'1
Rear tire stiffness K4 3078000 N m-T
Rear tire viscous damping C4 2000 N m'1 8‘1

 

A simple generic linear numerical half-vehicle model was developed to compute the

vertical vibration level of typical vehicle types from different pavement proﬁles at

constant speeds (Figure 8.1). This

numerical

developed with the

Matlab/Simulink® programming environment effectively computes the solution to the

two degrees of freedom system using the ﬁxed-point method. The inputs to the model

are the longitudinal pavement proﬁle and the velocity of the vehicle.

214

 

8.2.2.2 Discussion

A common packaging practice is to secure the products to the cargo so that the only
possible movement is in the vertical direction. However, when the truck is not fully
loaded, the packages could move in the longitudinal direction. A recent studies
conducted by Singh et al. [78] provided results that measured and analyzed truck
vibration for major freight distribution routes in USA and Canada. In their study, the
American Society of Testing and Materials (ASTM) and International Safe Transit
Association (ISTA) vibration test methods were used. The Comparison of the data
collected for trucks with both air and steel suspensions showed that the measured
vertical vibration levels were more severe than the lateral and longitudinal levels. The
study also showed that higher vibration levels occurred in the rear of the trailer, and
were a function of the suspension and payload (Figure 8.2). Therefore, a two-axle
truck model (bicycle mode) considered to be reasonable in product fragility

assessment studies.

 

 

 

 

 

 

 

Figure 8.2 High Acceleration Level Location in the Truck [48]

215

 

8.2.3 Product Vibration

There are two basic models depending on the type of product: (1) electronics
products or (2) horticultural produce. Models for electronic and appliances treat fragile
interior components as a single-degree-of-freedom spring/mass system deforming in
an elastic perfectly plastic manner under dynamic loading (Figure 8.3). The wok done
in [74] compares the model predictions with experimental results, which include the
generic ﬁndings referenced in an ASTM standard (ASTM D 3332). It was concluded
that these models have an acceptable outcome. Models for horticultural produce are
based on the principle of conservation of momentum. They treat the collision of

products in multi-layered packs with the surface as inelastic shocks (Figure 8.4).

 

   

Force
A
Product
Fragile ﬂ F0 '
Component - ix

A

 

x0 Compression

Figure 8.3 Electronic Product with Fragile Component, Modeled As Spring/Mass
System

I] Dissipated energy
Shock impulse

Figure 8.4 Collision of Horticultural Produce Treated As Inelastic Shocks

216

8.2.4 Product Damage Analysis

8.2.4.1 Electronic and appliances products
The shock fragility of these products is usually described by two parameters: critical

velocity change and critical acceleration. According to [3 7], a single shock pulse to the
product will not damage any component inside if either its velocity change (area under
the shock curve) or its peak deceleration (height of pulse) falls below these critical
values, respectively. Shock pulses much less severe than this with areas and heights
well below these critical values can damage a component by fatigue through multiple

drops. Figure 8.5 shows a typical Damage Boundary Curve [79].

Critical Velocity Change, (Vc)

\\\\\\\\\\\\ \N

A

OD

er

~§ \DAMAGE REGION\

:5

'8 Critical

ft, \ Acceleration, (Ac)
5 :\

E. : :

K I I

g ; ; 1. 57 Vc N O DAMAGE REGION

9
r

 

 

Velocity Change, mm/sec

Figure 8.5 Damage Boundary Curve [79]

For elastic perfectly plastic spring behavior, the critical velocity change and critical
acceleration for a shock repeated N times are calculated using Equations 8.1 through

8.5. A comprehensive summary of the model parameters is reported in [80].

217

2A A + B/ N
G = 27: —
C f" [2A+B/Nj
A = (Zﬂfnxo)
2g
x1
B = 2A — —
x0
n __ _ __
271' Wx0
Where,
A and B = constants speciﬁc to a given product
f,, = the natural frequency of the most fragile component
g = 981 cm/sz= acceleration due to gravity
W = the weight of the component
F 0 = the constant force producing plastic deformation
x0 = the limit up to which the spring deforms elastically
x1 = the failure limit

(8.1)

(8.2)

(8.3)

(8.4)

(8.5)

The vertical accelerations due to roughness events computed in the previous sub-

sections are the shock pulses applied to the electronic products. The velocity change

AV of the shock pulses and its peak deceleration G are computed as the area under

the shock curve and its height respectively. Typically, a truck is loaded with 12 large

wooden boxes ﬁlled with electronic products. The parameters for one block in the

trailer are given in Table 8.2.

218

 

Table 8.2 Summary of Material Properties [80]

 

 

 

 

 

Name Values
Natural frequency (f?) 33
The elasticity limit of the 1.45
spring (x0) in cm
Failure limit (x1) in cm 2.54
Acceleration due to gravity (g) 9.81
in m/s2

 

 

 

 

The following values for critical velocity and acceleration are obtained using the

parameters in Table 8.2 and assuming that the shock pulse was repeated 1 time (N =

l):

 

2
A = (Zﬂfnxo) : (2><3.14><33x1.45)2 ___ 0 46 m
2g 2x981 '
B = 2A[i-1] = 2x0.46x[§ﬁ—1]=0.69 m
x0 1.45

 

 

AVC = J2g£A+£—) = \/ZX9.81[0.46+0—°16—9—] =4.76m/s
GC=27rfn 2:4; A+B/N
g 2A+B/N
*
:2*3.14*33* ’2 0.46,, 0.46+0.69/l =454
9.81 2*O.46+0.69/1

If the shock pulse has a velocity change and peak deceleration higher than the values

 

calculated above, the entire block will be damaged. The same method will be used for
all the remaining blocks. If the percentage of the damaged blocks is more than 2%,

then the height and/or the width of the roughness event does not meet the threshold.

219

 

The 2% threshold is estimated according to the shipping insurance that shipping
companies give, which guarantees that 98% of the shipment will arrive intact [81].

The parameters presented in Table 8.2 are for electronic products including the
package which explain the high value for the critical acceleration above which damage
will occur. According to [78], the highest vertical acceleration amplitude is 0.89g for
leaf spring and 0.5g for air ride. Also, the lateral and longitudinal levels were
extremely low (<0.1g). Therefore, it is believe that roughness features will not cause

damage to electronic products.

8.2.4.2 Horticultural produce

Movement of fruit in containers during transport is a more serious cause of damage
than simple compression. The frequency and amplitude of vibration will depend on the
size and construction of the container and the suspension system of the vehicle.
Vibrational stress will be highest when the frequency coincides with the natural
resonant frequency of the fruit.

Jones et a1 [77] have investigated the dynamic behavior of fruit and vegetables in
multilayered packs under impact conditions. Columns of apples up to ten deep were
dropped on to stationary and moving surfaces, chosen to simulate impacts in transport.
The drops were photographed using high speed photography and measurements of
bruise volumes on all contact interfaces were made. The studies showed that:

1. At the beginning of the drop, individual apples separated and the whole
column fell with small spaces between each apple;

2. Collisions occurred sequentially;

3. Bruising of apple at each interface took an appreciable time;

220

 

4. While bruising was occurring at any interface, subsequent collisions of
apples further up the column added to the bruising of the interface;

5. When bruising was completed at any interface, subsequent collisions
further up the column had no effect at that interface.

These observations led to the formulation of a model for predicting the behavior of
apples in impacted, multi-layered columns. Consider ﬁrst a collision with a stationary
surface and assume that the package has no rebound. The kinetic energy of the falling
column is dissipated by bruising of apples at the various interfaces. Therefore, for each
collision the principle of conservation of momentum holds and can be used to
calculate the velocity of the impact surface aﬁer impact. Assuming no bounce-back of

the colliding apple, the principle of conservation of momentum is written as:

maua + msus : (ma + ms ) vs (8.6)
Where,
m a = mass of one apple
m S = mass of impact surface
u a = velocity of colliding apple
u S = velocity of impact surface
' Vs = ﬁnal velocity of apple and impact surface

Equation 8.7 calculates the change in kinetic energy (KB) of the impact surface. This
energy has been dissipated in bruising, apportioned between various interfaces of the

columns.

AKE : _ms (“5 "Vs )‘é‘mavs (8-7)

221

 

If the mass of the impact surface is large relative to that of the apple, its motion is
little affected by each collision, although substantial bruising may occur. This is the
situation for a lightly loaded truck. However, as the mass of the load increases relative
to that of the impact surface, the motion of the impact surface can be dissipated in

bruising of apples. This could be the case for a truck that is heavily loaded.

The ﬁrst collision will be between the truck bed and the ﬁrst layer of the package.
The second collision will be between the second layer and the layers beneath it (Figure

8.6). This iterative process will be repeated for all the layers.

—§

 

 

 

 

 

 

I , //,. x'u'x/

Figure 8.6 Road-Vehicle-Load Interaction for Multi-Layered Energy Absorbing
Packages
If the dissipated energy is larger than the energy resistance of the produce, damage
will occur. If the percentage of the transported produce exceeds 5%, then the height
and/or the width of the roughness event are not acceptable [81]. The list of
horticultural produce and their physical properties is exhaustive and cumbersome.
Including all products will be a cumbersome analysis. Apples, oranges, and bananas

are the most popular in the United States. According to the United States Department

222

of Agriculture Economic Research Service [82 and 83], they were ranked ﬁrst, second,
and third, respectively for fresh and processed fruit for millions of pounds purchased
in 2008. It was reported that the average consumer in the United States consumed 21.4
kg (47.1 lb) of fresh or processed apples in 2008. No other fruit was consumed in as
large of a quantity. Bananas were imported in much larger quantities than apples or
oranges. The apples used in this study were of the Golden Delicious variety. This
variety was selected because it is quite susceptible to bruising.

Apples are damaged when the frequency is less than 5 Hz (Figure 8.7) and the

-l
dissipated energy into the apple exceeds 6.4 ml J [77, 84 and 85].

Typical characteristics for trucks and packaging used to transport horticultural

produce are given in Table 8.3. These values were given by packaging companies

 

 

 

 

 

 

 

 

 

 

[81].
12 __-—__k,* __. ,A __.__,, _..,_.____ __.-.__r.. __.— __ __.—__— __.__-_.____,
'5
3 L__mﬁ‘wﬁh_hh_
'U
.3.
E —Ti T k TIT “_- United States standards for
E grades of apples' threshold
a _ IT 7 T T ”T “__. for commercial damage h-
3i _2 - - __./.2 _
5
3’0
8 k — .
a)
>
“1 IIIW
l ‘ I

Frequency (Hz)

Figure 8.7 Effect of Vibration Frequency on the Bruise Depth of Apples at Constant
Peak Acceleration of 1.4 g

223

 

Table 8.3 Summary of Truck and Packaging Parameter Values for Horticultural

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Produce

Parameter Values
Trailer length (m) 16
Trailer width (m) 2.6
Trailer height (m) 4
Maximum allowable GVW 45.4
(metric tonnes)
Maximum allowable payload 36.2
(metric tonnes)
Packaging box length (m) 0.38
Packaging box width (m) 0.32
Packaging box height (mm) 0.38
Packaging box weight (kg) 10
Number of apples per box 120
Number of boxes per trailer 3360
Packaging layout Columns

 

 

 

 

Recently, there has been renewed interest in energy consumption as research on
sustainable agriculture has broadened into analyses of food systems. Many existing
research and analyses of energy use in the food system discuss where the most
signiﬁcant and feasible energy savings might be achieved. Recently, Hendrickson [86]
compiled data from the US. Department of Agriculture to ﬁnd out how far apples
traveled. He reported that the estimated transportation distance for apples is about
2400 km. The study also reported that apples are ranked fourth for the average mileage

travelled. Grapes, lettuce and peaches were ranked ﬁrst, second and third respectively.

224

8.3 RESULTS AND CONCLUSIONS

To illustrate the various features of the method described above, the case study of
US conditions has been examined. All road surface proﬁles were artiﬁcially generated
at every 0.07 m. Road surface proﬁles were ﬁltered out using a moving average ﬁlter
with a baselength of 0.3 m for trucks representing the tire enveloping. Then, the ‘half-
car’ truck model traveling at a constant speed of 110 km/h was applied to a 1.6 km of
road surface proﬁles. The parameters used in the ‘half-car’ model are given in Table

8.1 above.

8.3.1 Case Study 1: Effect of Faulting on Damage to goods

The ﬁrst case study examines the effect of different combinations of faulting levels
and frequencies per 1.6 km on damage to goods. The faulting was modeled as shown
Figure 7.2. The vehicle speed was the same for all the runs. Only jointed plain
concrete pavement (JPCP) were considered in this analysis since Jointed Reinforced
Concrete Pavement are no longer used in the US. Also, the effect of suspension type
on damage to goods was also investigated. The analysis showed that faulting has no

effect on electronic products.

Figure 8.8 shows the effect of different levels and frequencies of faulting in JPCP on
the damage to apples. The damage is expressed as a percentage of the total number of
boxes in the cargo. As expected, the higher the fault magnitude the higher the percent
of damaged boxes. It was also noted that shorter spacing between faults in the roads
will cause less damage to the goods transported in trucks with steel suspension. This

observation is not true for trucks with air suspensions.

225

 

 

 

 

 

 

 

 

 

12 Steel Air
+ —0— 50% of slabs are faulted
A 10 ‘ + -E*- 100% uncracked
E: + + 100% slabs with 1 crack
& 8 I Acceptable damage
a threshold
re 6 a
Q
E
o 4 r
«‘3‘
D—t
2 .4
0 ~ mi . ~ . -- - r . v a
O 2 4 6 8 10 12 14

Fault Magnitude (mm)

Figure 8.8 Damage Induced by Different Levels and Counts of Faulting

It is believed that these observations were the result of the interaction between

speed, proﬁle wavelength content, resonant frequencies of trucks and goods. Figure
8.9 shows the effect of speed on damage to goods. It was observed:

For trucks with steel suspensions (Figure 8.9a)

0 High speed will cause less damage to goods than low speed except for the
case where all the joints in the JPCP pavements are faulted.

o The difference between pavement conditions is even greater with lower
speed.

For trucks with air suspensions (Figure 8%)

0 In concrete pavement where 100% of slabs are with 1 crack, high speed will
cause less damage to goods than low speed.

0 When all the joints in the JPCP pavements are faulted, the speed has no
effect on damage to goods.

226

0 When 50% of slabs are faulted, the multiple resonant frequencies of the truck
model cause the curve to oscillate.

8.3.2 Case Study 2: Effect of Breaks on Damage to Goods

The second case study examines the effect of different combinations of
breaks/potholes levels and frequencies per 1.6 km on damage to goods. The effect of
truck suspensions (i.e., air and steel) is also investigated. The breaks were modeled as
a step function of variable amplitude and a length of 0.9 m (Figure 7.3). The damage
was also expressed as a percentage of the total cargo. The analysis showed that breaks
have no effect on electronic products.

Figure 8.10 show the effect of different levels and frequencies of breaks in JPCP on
the damage to goods. The damage is expressed as a percentage of the total number of
boxes in the cargo. As expected, the higher the break magnitude the higher the percent
of damaged boxes. Also, more breaks in the roads will cause more damage to the
cargo. Vehicles with air suspension cause less damage than those with steel spring
suspension for apples. The difference in damage between steel and air suspensions is
signiﬁcant. Since the road isolation ability of air ride suspensions is higher than leaf
spring suspensions, they will absorb more energy induced by the vertically accelerated
wheel, allowing the frame and body to ride undisturbed while the wheels follow the
bumps/depression in the road. This difference becomes less signiﬁcant as the number

of breaks increases.

227

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

14 Fault Magnitude
5 mm 10mm
12 I; + ‘9‘ 50% of slabs are faulted
§ 10 + -E*— 100% uncracked
ED ; + —a— 100% slabs with 1 crack
(6
E
{U
Q
E
8
B
a.
80 100 120 140 160 [
Vehicle Speed (km/h) ‘

(a) Steel suspension

p—a

Fault Magnitude

5 mm 10mm

+ ‘9‘ 50% of slabs are faulted
+ -B- 100% uncracked

+ + 100% slabs with 1 crack

 

__._._._ __1

 

 

I
— _. J..._-._‘_.L__

 

_l

I? -)

o—‘NUJAUIaQOOKDO
__.__
I
(11'
l
((13
l 12>:

 

 

 

Percent Damage (%)

 

80 100 120 140 160
Vehicle Speed (km/h)

(b) Air suspension

Figure 8.9 Interaction Effect between Speed and Fault Counts on Damage to Apples

228

Number of slabs

  
 
 
  

 

 

 

 

 

25
i betweenbreaks
A201 [Steel FA}:
g\: —50 —501
85 1 +25 +25
m 15
g . +10 +10,
:8 1+5 —e—5 I
c, .
g 10 | +1
Q) l
e i
5 i
0 ‘9 ‘
0 2 4 6 8 10

Break magnitude (mm)

Figure 8.10 Damage Induced by Different Levels and Counts of Breaks

8.3.3 Case Study 3: Effect of Curling on Damage to Goods

The third case study examines the effect of different combinations of curling
magnitude and frequencies for 1.6 km of concrete pavements on damage to goods.
Curling is modeled as an ellipsoid of variable amplitude and width. The curling width
is assumed any value between 3 m (minimum) and slab length (maximum). The
analysis showed that curling has no effect on electronic products.

Figure 8.11 shows the effect of different levels and frequencies of curling in JPCP
on the damage to apples. The damage is expressed as a percentage of the total number

of boxes in the cargo. As expected, high severity curling will cause more damage.

229

Also, more curling in the roads will cause more damage to the cargo. Vehicles with air
suspension cause even less damage for apples than those with steel spring suspension
as compared to breaks. The difference in damage between steel and air suspensions is

very signiﬁcant. This difference becomes less signiﬁcant as the number of curling

 
 
    

 

 

 

 

 

increases.
Steel Air
14 — — 100% uncracked
+ —A— 20% slabs with 1 crack
12 - + —e— 40% slabs with 1 crack
’3 + —e— 60% slabs with 1 crack
85 10 ‘ + .9. 80% slabs with 1 crac
86
E 8 ”
<6
1:
a 6 ~
8
a? 4 -
2
O _qr-__
0 2 4 6 8 10

Curling magnitude (mm)

Figure 8.11 Damage Induced by Different Levels and Counts of Curling

8.3.4 Case Study 4: Effect of Interaction between Roughness Features
Magnitude, Frequencies and Trip Length on Damage to Goods

For all previous case studies, the trip length was assumed constant and equal to the

typical value in the US, i.e., 2400 km. However, trip length ranges from 160 km (local

230

trip) to 2400 km (Shipping from California). The forth case study examines the effect
of different combinations of roughness features magnitude and frequency for different
trip lengths of concrete pavements on damage to goods. Figures 8.12 and 8.13 shows

the results for faulting and breaks.

8.4 SUMMARY AND CONCLUSIONS

In this chapter, we proposed a novel approach to estimate the damage induced to
transported goods by roughness features. The approach proposed herein uses a
mechanistic-empirical approach to conduct product fragility assessment using
numerical modeling of vehicle and product vibration response. A half-truck model
was used to simulate vehicle vibrations. Table 8.1 summarizes the parameters for the
half-truck (bicycle mode) model considered in this study. To analyze damage to
goods, there are two basic models for products depending on their type: Models for
electronics and appliances treat fragile interior components as a single-degree-of-
freedom spring/mass system deforming in an elastic perfectly plastic manner under
dynamic loading. Models for horticultural produce are based on the principle of
conservation of momentum. They treat the collision of products with the surface as
inelastic shocks. The approach reported in this chapter could help in better estimating
vehicle operating costs at the project and network level. For routine application, 3
DOT would need to run a given proﬁle (for a given project) through the program

developed in this study.

231

Percent Damage (%)
w

    

'iii'1‘500

0 Trip Length (km)
(a) 50 % of slabs are faulted

Percent Damage (%)

 

0 Trip length Gan)
(b) 100% of slabs are faulted

Figure 8.12 Damage Induced by Different Fault Magnitude and Trip Length

232

;> ;>
N U)

.o
_.

Percent Damage (%)

 

 

'\..

Break Magnitude (mm) \ -265“ 1000
0 Trip Length (km)
(a) One break per 1.6 km

N DJ

p—n

Percent Damage (%)

  
 
  

_ . 2000 2500
. 7 , 1000 1500
Break Magnitude (mm) ' 500

Trip Length (km)
(b) three breaks per 1.6 km

Figure 8.13 Damage Induced by Different Break Magnitudes and Trip Length

233

CHAPTER 9
CONCLUSIONS AND RECOMMENDATIONS

9.1 INTRODUCTION

This chapter summarizes the main ﬁndings from this research, outlines the main

conclusions from the study and their implications to pavement management and

highlights areas in which further research is needed.

9.2 SUMMARY OF FINDINGS

The objective of this study was to investigate the effect of pavement conditions on

Vehicle Operating Costs (VOC) including fuel consumption, repair and maintenance

and damage to goods. These effects are essential to sound planning and management

of highway investments, especially under increasing infrastructure demands and

limited budget resources. This was done by testing the hypothesis that:

Pavement Conditions have an effect on fuel consumption: the analysis of
covariance (ANCOVA) was successfully used to extract the effect of pavement
roughness. Also, the analysis showed that surface texture has a measurable
effect on fuel consumption for heavier trucks and at low speed. The pairwise
comparison between sections that have similar geometric characteristics but
different pavement type showed that the difference between asphalt and
concrete pavements is statistically signiﬁcant for light and heavy trucks at low
speed and summer conditions.

Pavement roughness has an effect on repair and maintenance costs and damage
to goods: The analysis of repair and maintenance cost data collected from
vehicle ﬂeets showed that roughness less than 3 m/km has no effect on repair
and maintenance. Based on the results from the proposed mechanistic-
empirical approach, the effect will increase exponentially.

The transient events in the road proﬁles have an effect on the damage to
vehicle suspensions and transported goods: The surface proﬁle of the road
transmits the vibrations through the tires and suspension system to the body of
the vehicle and then to the driver, passengers and cargo. The proposed

234

 

mechanistic empirical approach was successfully used to estimate the effect of
these transient events on repair and maintenance costs and damage to goods.

9.2.1 Fuel Consumption

The goal of this part is to investigate the effect of pavement conditions on fuel
consumption. The proposed work has entailed the use of ﬁve instrumented vehicles to
measure fuel consumption over different pavement sections selected based on the
variability level of their pavement conditions. We, ﬁrst, calibrated the World Bank
Highway Development and Management (HDM4) fuel consumption model. We
showed that the calibrated model was able to predict very adequately the fuel
consumption of ﬁve different vehicle classes under different operating, weather and

pavement conditions.

The better accuracy achieved after calibration of the HDM 4 fuel consumption
model to US conditions has improved the prediction of the effect of roughness on fuel
consumption. The comparison of sensitivity analyses before and after calibration
showed that the effect of roughness on fuel consumption increased by 1.75 for the van,
1.70 for the articulated truck, 1.60 for the medium car, 1.35 for the SUV and 1.15 for

the light truck.

Also, the key characteristics of representative vehicles used in the HDM 4 model
vary substantially from those of US vehicles. Therefore, the predicted fuel
consumption using the model without calibration was lower than the actual
consumption. For example, the amount of fuel consumed by a light truck in the US is
equivalent to the predicted consumption by a medium truck using the HDM 4 default

characteristics of representative vehicles. Therefore, we recommend the use of the

235

HDM 4 fuel consumption model after calibration and adjustment to US conditions

using the calibration factors mentioned in Table 4.11.

In addition, the analysis of covariance of the ﬁeld test data collected as part of this
study showed that roughness and texture have a measurable effect on fuel
consumption. For example, a 2 m/km reduction in IRI will result in a 2 to 0.5 %
reduction in ﬁJel consumption for passenger cars and light truck, respectively. Also,
the analysis showed that the effect of surface texture is statistically signiﬁcant at 95
percent conﬁdence interval for heavier trucks and at low speed. An explanation for
these observations is that, at higher speeds, air drag becomes the largely predominant
factor in fuel consumption. The increase in rolling resistance (i.e., ﬁiel consumption)
due to texture will be shadowed by the increase in air drag due to speed. Therefore, the
effect of texture as a percentage is lower at higher speed. For example, a 67 %
decrease in mean texture depth will result in a 1.3 % and 0.9 % decrease in fuel

consumption for heavy truck at 56 and 89 km/h, respectively.

In addition, the good quality of the data allowed the investigation of the effect of
pavement type on fuel consumption. It was noted that, for both heavy and light trucks
and for summer conditions, the mean difference of fuel consumption between asphalt
and concrete pavements is statistically signiﬁcant at low speed; whereas, it is
statistically not signiﬁcant at higher speeds. On the other hand, for winter conditions,
the mean difference is statistically not signiﬁcant. The analysis also showed that the
mean differences of fuel consumption between asphalt and concrete pavements for
passenger car, van and SUV are statistically not signiﬁcant. These observations could

be explained by the viscoelastic behavior of asphalt concrete.

236

9.2.2 Effect of Roughness on Repair and Maintenance Costs

The analysis of the repair and maintenance (R&M) cost data of Michigan and Texas
DOT vehicle ﬂeets showed that the predictions from HDM-4 do not appear to be
reasonable for US conditions. These observations could be explained by the fact that
the HDM-4 model was calibrated using data from developing countries (e.g., Brazil,
India). It is well known that the labor hours in those countries are much higher than in
the US. Also, the difference between parts consumption in the US and those predicted
from HDM-4 could be explained by the inﬂation in the parts and vehicle prices.
Therefore, it was recommended to use the updated Zaniewski’s repair and
maintenance costs (i.e., the latest comprehensive research conducted in the US). These
costs were updated by multiplying their reported costs by the inﬂation rate of R&M

costs between 1982 and 2007.

9.2.3 Effect of Roughness and Roughness Features on Vehicle Durability and
Damage to Goods

In this thesis, 3 novel approach was proposed to estimate repair and maintenance

costs induced by roughness features. First, the approach proposed herein detects,

locates and identiﬁes roughness features from road surface proﬁles. Then, it uses a

mechanistic-empirical approach to conduct vehicle component fatigue and product

damage analysis using numerical modeling of vehicle and product vibration response.

The newly developed roughness diagnosis methods were able to detect, identify, and

localize roughness features. It was reported that the methods discussed in chapter 6

237

give results with a comparable magnitude and frequency of bumps and depressions
(faults, breaks and potholes) with an average error of 0. 1% and a standard deviation of
1%. The localization was also highly accurate (average error in distance is within 1 m
or 0.5%). Thus, these new methods could be used as a complementing module for the
existing pavement management system. It would help the highway agencies in
deciding whether a particular section of the pavement needs a

maintenance/rehabilitation action.

Regarding vehicle component fatigue, different vehicle models were developed to
simulate the dynamic behavior of cars and trucks with air and steel suspensions. Then,
the time histories were rainﬂowcounted to compute the number of cycles and their
magnitude. Later, the accumulated damage was calculated using miner’s rule. Finally,
the repair and maintenance costs were computed by multiplying the accumulated
damage per km by the lifetime warranty of truck and car suspensions (i.e., 400,000 km
and 160,000 km, respectively) and suspension replacement costs. To check the
accuracy of the method, the accumulated suspension damages induced by (1)
artiﬁcially generated proﬁles and (2) real roads proﬁles from MDOT and LTPP
databases were compared. The results showed that the computed accumulated damage
using real proﬁles follows the curve using generated proﬁles with some scatter. The
variance around the curve is caused by the difference in real proﬁles roughness
contents.

The results from the mechanistic-empirical approach were also compared to the

empirical results (i.e., Updated Zaniewski’s tables), and were found to be very close

238

until 5 m/km. The standard error is about 2 %. These results seem promising since the

typical IRI range in the US is between 1 to 5 m/km.

Regarding damage to goods, a half-truck model (bicycle mode) was used to simulate
vehicle vibrations. Two basic models for products were used depending on their type:
0 Models for electronics and appliances treat fragile interior components as a

single-degree-of-freedom spring/mass system deforming in an elastic
perfectly plastic manner under dynamic loading.

0 Models for horticultural produce are based on the principle of conservation
of momentum. They treat the collision of products with the surface as
inelastic shocks.

The analysis of three case studies for horticultural produce showed that:

0 Air suspensions cause less damage to the transported goods than steel
suspensions

0 Shorter spacing between faults in the road will cause less damage to the
transported goods in trucks with steel suspension.

0 Low speed will cause more damage to transported goods in trucks with steel
suspensions than higher speed.

0 More breaks on the road will cause more damage to the transported goods.

9.3 CONCLUSIONS
The research in this study was successful at estimating the effect of pavement
conditions on Vehicle Operating Costs (V OC):
0 The analysis of covariance was successfully used to extract, from ﬁeld data
collected as part of this research, the effect of roughness and texture on fuel

consumption. We also were able to clarify the effect of pavement type on
ﬁre] consumption.

0 We successﬁrlly calibrated a mechanistic-empirical model to estimate the
effect of pavement conditions on fuel consumption. We showed that a

239

 

decrease in pavement roughness by 2 mlkm will result in a 1-2 percent
decrease in fuel consumption depending on vehicle class. This would save
about 2 to 4 billion gallons of ﬁrel per year of the 200 billion gallons
consumed by the entire vehicle ﬂeet in the US. In this context, a 1 to 2
percent reduction in the fuel consumed would be a meaningful
accomplishment.

0 We updated an empirical model to estimate the effect of pavement roughness
on repair and maintenance costs using the inﬂation rate from 1982 to 2007.

0 We developed a novel mechanistic-empirical approach to estimate the effect
of pavement roughness on repair and maintenance costs. It should be noted
that all the models currently in use are empirical models. The only purely
mechanistic model is the VETO model which was developed by the Swedish
Road and Trafﬁc Research Institute (VTI). However, the change in vehicle
wear with increasing roughness is far higher than the change in parts cost
predicted by the empirical model. Therefore, they adopted the empirical
model. In our case, the comparison between the updated (empirical)
Zaniewski model and our mechanistic-empirical model showed that they
agree until 5 m/km. Since the IRI range in the US is from 1 to 5 m/km, we
can conclude that the proposed approach is an improvement to the state-of-
art. We can also conclude that considering only IRI as measurement of
pavement conditions is enough for pavement management at the network
level. However, at the project level, the effect of roughness features should.
be included. This could be done using the mechanistic-empirical approach
proposed in this thesis.

0 We developed a novel mechanistic-empirical approach to estimate the
damage to goods induced by roughness features.

In addition, this thesis developed a new tool to detect, localize and identify
roughness features. This tool can be used at the project leve. At the network level, the
dl.Slfribution of IRI and roughness features per 0.16 km (0.1 miles) will be input to the

VOC models.

9-4 RECOMMENDATIONS

The results of this study lead to the following recommendations.

240

First, it is suggested that the newly developed roughness diagnosis tool should be
incorporated as part of a Pavement Management System program. This would
enable state highway agencies to monitor the pavement condition for the entire
pavement network. Such action does not entail implementing the proposed
detection methods in Chapter 6. The roughness diagnosis computer program was
completed as part of this study; therefore its incorporation as an independent
subroutine into the PMS program should be straightforward.

Second, it is recommended to use the calibrated fuel consumption (chapter 4) and
repair and maintenance (chapter 5) models developed in this study in order to have
a better estimation of the effect of pavement conditions.

Third, it is believed that the effect of roughness features on damage to vehicles and
to transported goods should be taken into account when performing a life cycle
cost analysis. Since IRI is a summary index, it will hide the actual content of the
surface proﬁle. These localized roughness events were proven to contribute the
most to vehicle damage. It is suggested to use the methodology presented in
chapters 7 and 8 and the corresponding computer program developed as part of
this study.

9.5 FUTURE RESEARCH

The results of this study lead to the following recommendations.

1.

Develop a mechanistic formulation for predicting the effect of roughness on fuel
consumption: The vehicle models could be used to predict the dynamic load,
which can then be input in the HDM 4 model to replace the weight.

Develop a method to estimate the cost corresponding to the predicted damage to

goods using the approach presented in chapter 8.

241

 

 

 

 

APPENDIX

 

242

APPENDIX A

TYPICAL CONDITIONS IN THE US

Figures A.l through A.6 show typical input data for pavement condition,
environment (temperature) and vehicle characteristics in the US. Figures A.l(a)
through ((1) show the distributions of pavement types in the US. It can be seen that
65% of the roads in the US are paved. Among the paved roads, 57% are ﬂexible, 6%
are rigid, 11% are composite and 26% are thin surfaced pavements (source: FHWA).
IRI data for all states as of 2006 have been extracted from the F HWA website. Figures
A2 (a) through ((1) show the distributions of IRI of Interstate, US and state highways
in selected states. These data support the distribution of pavement sections by
pavement type and roughness in the proposed experimental matrix for the ﬁeld trials
presented in chapter 4.

Figure A3 (a) shows the average monthly air temperatures (2007) for representative
states above and below the national average, while Figure A.3(b) shows the data for
Michigan conditions. These data will be used to correct for the environmental
conditions during the ﬁeld trials for fuel consumptions.

Vehicle aerodynamic characteristics for all classes in the US have been collected
(source: EPA report, 2007). Figures A.4 (a) and (b) show typical ranges of these
characteristics for passenger cars. Figures A.4 (c) and (d) show these characteristics
for speciﬁc car models of popular US and imported brands. These data was used for
inputting the speciﬁc data for the vehicles used in the ﬁeld trials. Figure A.5 shows

examples for trucks and passenger cars.

243

 

Fuel efﬁciency data for all vehicles in the US market are readily available from
manufacturers and other sources. Figure A.6 shows distributions of these engine
performance parameters for passenger cars and trucks (source: EPA report, 2007).
Speciﬁc fuel efﬁciency values for the vehicles used in the ﬁeld trials will be extracted
from the vehicle manufacturer catalogs. The rated engine power for any vehicle can be
determined using the relationship shown in Figure A.7 (source: EPA report, 2007).

The data from bureau of census database (Census 2004) was also collected. The
database contains detailed information on trucks in US ﬂeets (Truck use, body type,
vehicle size, annual miles of travel, age, vehicle acquisition, truck type, range of
operation, and fuel type, etc.). Figure A.8 shows an example of such data taken from
this source. Figure A.8a shows annual fuel consumption by vehicle class. It can be
seen from the ﬁgure that passenger cars, single unit trucks (SUT), heavy trucks and
buses consume about 45, 33, 21 and 1 percent of the total fuel consumed, respectively.
Figure A.8b presents the percent of vehicle-miles traveled by vehicle class: The data
shows similar trends as fuel consumption. Trucks and buses have the highest fuel
consumption as compared to cars and SUT (see Figure A.8c). Although trucks showed
a smaller percent of the total fuel consumed and mileage traveled, they have the
highest annual traveled mileage by vehicle class (see Figure A.8d). Furthermore, time
series data for average mileage per gallon by vehicle class indicates that there is no
signiﬁcant change from 1995 to 2002. However, with increasing fuel costs and
demands (see an example of forecasted fuel consumption for the State of Oregon in

Figure A.8f), it is anticipated that this trend (in Figure A.8e) will not remain the same.

244

 

This simple example indicates that the

factors needs to be considered.

 

(a) % of mileage by pavement category

, Low Type
4%
lntennediate
18%

Rigid
9% ‘
Composite
15%

Flexible
54%

(c) % of mileage by surface type
(urban roads)

interaction between various VOC-related

. Rigid Low Type
Composne 5% 10%

9/e

   

Intermediate
18%

Flexible
58%
(b) % of mileage by surface type
(rural roads)

Rigid Low Type
Composite 5% 8%
11%
Intermediate
18%
Flexible
57%

(d) % of mileage by surface type
(all roads)

Figure A.l Latest Pavement Type Frequency Distribution (source: HPMS Database)

245

 

100—-—

80 l

°/o OF MILEAGE
8

 

 

 

 

IRI (mlkm)

(a) Interstate Highways from PMS databases

 

 

 

 

 

 

 

 

 

 

1oo—»
a 80
3 60”
E
II. 40
O
a\° 20
0|
0 1 2 3 4 5
lRI(mIkm)
(b) US Highways from PMS databases
100 — ————————— a
,. l
3 8° 2 ‘ / ” ”
a / '
_l 60 . _, -~
s ./ x
I]. ,_ 2 , , JI_L_.,._,.
0 4° /
e\"
20 w -7/ -
0 I. t . T T
o 1 2 3 4 5 6 7
lRI(mlkm)

(c) State Highways from PMS databases

Figul‘e A.2 Latest Pavement Roughness Frequency Distribution (source: HPMS database)

246

90W
80«
70—
60.
50-

404

 

Temperature (F)

30~

204

 

101

0 J _T ' ifi ‘ 7* "H“; T__ _‘_T— _T_‘_‘_—T_' ‘ ”1" .——»“l

Jan Feb Mar Apr May June Jul Aug Sep Oct Nov

 

 

Average —-— Michigan _+ California —e— Ahbama {
._+ Arizona —°— Colorado + Texas |

(3) Temperature distribution in different states

8

Temperature (F)
8 8

N
O
- . _ _.l___

 

 

+ West upper + East upper + Northwest lower l
+ Northeast lower —9— West central lower —8— Central lower ,
—x— East Central lower -£r- Southwest lower -B— South central lower i
—4— Southeast lower .

 

 

(b) Temperature distribution in Michigan 4

Figme A.3 Environment Conditions in the US for 2007 (source: National Climatic Data
Center)

247

0.6

0.5

Drag coefﬁcent

0.1

2.2 .

2.1 }

Frontal area (m2)

1.6 ~

1.5 4

Figure A.4 Latest Aerodynamic Parameters in the US (sources: EPA and CarTest

0.4 .—
0.3

0.2 L

1.9
1.8 .

1.7 ¥

 

 

 

 

 

 

0.49
g __j 0.46
I 1 I l ———i
l l
. l I
“‘ ‘ L__-__.i ____d
0.33
0- 0.3
Small Medium Large
Passenger car categories
(a) Drag coefﬁcient ranges by vehicle class
2.16
i‘ " ‘l
1.95 ]
1.83 [ .__ __.___l
i —' l i
i I l _ j 1.86
J 1 76
1.59
Small Medium Large

Passenger car categories

(b) Frontal area ranges by vehicle class

software)

248

 

 

 

EC?" T Frontal area (m2)
[3 Frontal area 0 0.5 1 1.5 2 2.5 3
4. ~ - : --,, + e4i~ : ~ a ﬁll

 

[ ]
'05 Malibu :—

 

 

IC ]
'05 Blazer LS _—

 

 

'03 Blazer ﬁ—
'03 Monte Carlo ,—
l J

'03 Venture —

[ l

 

 

 

Vehicle model

'03 Impala SS _
'02 Camaro Convertible ,—
[ l

'02 Camaro Calloway 1_

 

 

 

[ l

'02 Corvette —

*7 .LV ,,.___7

0 0.1 0.2 0.3 0.4 0.5

 

Drag coefﬁcient

(c) Frontal area and drag coefficient for Chevrolet

’ I Cd _. Frontal Area (m2)

Efroma'a'ié 0 0.5 1 1.5 2 2.5 3 3.5
t ~ ~+ ~r-—d-— _+— 'A‘fru—“Y ~-— ——-«.' -~—- .
L i

'05 Camry SE Automatic _

L_ j
'05 Tandra 4X4 —

L J
I05 Previa LE _
l_

'04 Land Cruiser _
i— l

'04 Corolla SE —

L
l l

'04 Celica GT Coupe [—
'03 Camry SE Automatic —
'03 Camry _ 7 "1

O 0.1 0.2 0.3 0.4 0.5

 

 

 

 

 

 

 

Vehicle Model

 

 

 

 

 

Drag Coefﬁcient

(d) Frontal area and drag coefﬁcient for Toyota

Figure A.4 Cont’d

249

70% 'l
I

,8

50% d
40% 1

30%)
oo
00 _
o ‘ , ﬂ.-. ‘-..—_.. _ f.EI..-.f_—ﬁj__-.L

6000 10000 14000 16000 19500 26000 33000 More
Truck Weight (Lb)

Percentage of trucks (%)
8
\

o5
o\\

(a) Percentage of truck by truck weight class

4,500 *
4,000 ‘
3,500
3,000 1
2,500
2,000
1,500
1,000
500

Average Car Weight (Lb:

 

 

 

 

R

2005
Year

   

FLLLVL,ﬂﬁrrmr , , L, L
I Srmll Medium CI Large ‘
(b) Average car weight by class for 2004-2006

Figure A.5 Vehicle Weight Statistics in the US Grouped by EPA Vehicle Classiﬁcation
(source: F HWA)

250

 

 

 

 

 

 

 

 

 

 

0.3
0.25 i — —
0.2
’3
E.
m 0.15 4
2
a
m 0.1 ,. —- —
Dill], ll , ,1] Elle.
16.8 18.5 18.8 20.5 21.4 23.2 26.1 27 28.7 30.2 42.9 46.2
Fuel economy (MPG)
(3) Average distribution of fuel economy for passenger car in the US
0.3 f
0.25 a .— _
0.2
’2
a
,,, 0.15
.33 I
8 i
005101 7 —

 

 

 

 

 

 

 

 

 

l I If T T I 17

.020 0.022 00310.033 0.035 0.036 0.0410044 0.046 0.050 0051£0.056
Fuel effe cie ncy (mL/KJ)

(b) Average distribution of fuel efﬁciency for passenger car in the US

Figure A.6 Latest Fuel Economy and Efficiency Distribution in the US for Passenger
Cars and Trucks (source: EPA Report, 2007)

251

 

0.25 -'

sales(%)

0.05

0.3 7

0.25 1 _

.O
m

sales (%)
.o

.0
_L
l
l

0.05

.1800, U ii ill]-

T I 1

14.2 14.7 15 15.7 16 17.7 19.9 20.1 21.8 22.6 26.4 29.7

Fuel economy (MPG)
(c) Average distribution of fuel economy for Trucks in the US

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2.00, [1 ii 0.0...

T l 7 r]—

0.026 0.029 0.034 0.035 0.038 0.038 0.043 0.048 0.049 0.051 0.052 0.054

Fuel efficiency (mL/KJ)
((1) Average distribution of fuel efﬁciency for Trucks in the US

 

 

Figure A.6 Cont’d

252

 

Effeciency (%)
—s N
or o

_s
O

 

 

 

__ T .___ _,._ __—

O 20 40 6O 80 1 00

Percentage of rated engine power

Figure A.7 Relationship between Engine Efﬁciency and Rated Power (source: EPA
Report, 2007)

 

253

Armual ﬁre] consumption (bill. Gal)

Trucks
21 .64%

      

' 44.90%

‘

Single Unit Truck
32.85% Buses
0.60%

(a) Annual ﬁrel consumption by vehicle class

Vehicle—miles of travel (billions)

 

Trucks
7.55%
Single Unit
Truck
33.93%
Cars
58.28%
Buses
0.24%

(b) Average vehicle miles by vehicle class

Figure A.8 Average Fuel Consumption and Miles Traveled by Vehicle Class for the US
(Source: US Census)

254

Avg miles per gallon (Year 2002)

25 .____ __ La____ -_-, -_.. *__.__._____ .________.. __2

 

 

(a) Average miles per gallon by vehicle class

 

Avg. miles per vehicle (x 1000)

30 —— «we ~-—~~—————~

lull

Buses Single Unit
Truck

20 i-

 

(b) Average annual miles by vehicle class

Figure A.9 Average Miles per Gallon and Annual Miles by Vehicle Class for the US
(Source: US Census)

255

 

L»)
O

 

 

 

a

{:57-Chns-4§;lBhani-1rj§RJT"—¥F-T}ucks;

 

 

 

 

 

 

 

N
LI]
0
0
- W _ E>_____‘y_ ________

 

 

 

 

g e :7 ea ¢2_...—re
:: 20 i
m
on A L
H r. A L‘s A am u
3 a
£3 15 i
E
oh
:5 10 I
8 5% M a e e 8
5 . “5 Ff
‘ l
0 _——'L—_“ -—¥ -‘~ --—L - __-;L -__4___ __-_______ _h ___i___[___ -_ :
1995 1996 1997 1998 1999 2000 2001 2002

\ﬂxns

(6) Average miles per gallon by vehicle class, time series

Figure A. 10 Vehicle Mileage Statistics for the US (Source: US Census and Oregon DOT)

256

 

APPENDIX B

UPDATED REPAIR AND MAINTENANCE COSTS

Table B.l Updated Repair And Maintenance Costs ($/1000 km) — Small Car
Table B.2 Updated Repair and Maintenance Costs ($/1000 km) -— Medium Car
Table B.3 Updated Repair and Maintenance Costs ($/1000 km) — Large Car
Table B.4 Updated Repair and Maintenance Costs ($/1000 km) — Pick up
Table 3.5 Updated Repair and Maintenance Costs ($/1000 km) — Light Truck
Table B.6 Updated Repair and Maintenance Costs ($/1000 km) — Medium Truck
Table B.7 Updated Repair and Maintenance Costs ($/1000 km) — Heavy Truck

Table B.8 Updated Repair and. Maintenance Costs ($/1000 km) — Articulated Truck

257

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

a»? 8.8 2.3. 8.8 52m NEW :8 8.8 2.8 8.8 an? 8.: SS 2.? w-
8.2 8.8 8.8 8.3 3.9. $5. aim 53m NEW :8 :8 a? 2.8 8.8 e-
a: 3.: 3.8 8.2 Sam 5a 8:. 8.9. 3.2 we? 8.; 8.8 onmm 8.8 e-
8.8 SE 8.: 38 8.8 3.8 mean 3.2 8.8 1.8 8:. 3.2. 3.9. 8.3. m-
3.3. 8.3 8.8. 3.2 as: an: $8 3.2 $2 8.2 8.8 8.2. 5A 8.2 4.
88 8.5a .88 $8 8.2 9.8 an: 8.: 8.: 2.: 5.3 8.8 8.2 8.8 m-
9.6.. 2.; 8.2 8.8 8.8 3.2 8.: 8.8 a: 3.: 3.: 8.: mm: 8.2 N-
3% $8 8.3 8.8 8.2 3.8 8.8 8.2 2.8 2.: :2 32 a: OS: _-
43m $3 a: 3.8 8.8 2.8 2.2 8.8 8.2 8.: N58 8.: 2.: at: o
8.: 8.2 :8. on: 8.8 3.3 :8 8.2 8.8 2.8 :._N :8 an: a: _
8.3 8.3 3.3 9.2 8.8 6qu a: 8.2 3.8 3.: 8.8 8.8 an: 8.: N
8.8 5.: an? 3.3 8.: a8 :8 $8 8.8 2.8 a: SS 8.: :9 m
3:. 2.3 98 83. 3% Fa 8% 2.2 8.2 3.2 Nos 3.: a: 2.: a
3.9. 3.8 8.3 2.2 $3 2.; 3% mm: :8 w?“ 8.2 a: 2.8 2.2 m
8.3 2:. :8 3.8 Si 88 3.8 8.3 :2 8.4.8. 2.2 8.: 2.8 an: e
3.8 3.? 2.9. 8.2 8.3 3.2 an: 8.2 a: 82 8.2. a: 8.8 :3: a
$5. 2.3. an; 8.3 8.3 an: Era 8.8 a: 8.2 8.8 and 3.8 as: m
N: 2: ea mm ow Q 8 em 8 9. mm X 2 w .8

3:8 98% 886

 

 

 

5 __am - E 82% as sesaaiaaz gas 2 m as

258

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

:8 8.: :8 :8 8.: 8.: 3.: 8.: :8 :.: 8.8 8.: 8.: :8 w-
88 8.: :8 :8 :8 S: 8.: K: :8 9: :8 :3 :8 :8 w
:8 8.: :8 8.: 8.: 8.8 8.: 8.: $8 :8 8.: 8.: 2.: 8.: o-
8.: 88 :1: :8 88 :8 :.: we: 5.: 8.8 9.: E: 3.: :8 m-
:8 :8 :8 88 8.: :8 :8 :8 :8 :8 8.: 8.: 8.: 8.: w-
:.: :8 8.: :8 88 :8 8.: :: :.: :8 6.: R8 :8 :8 m-
2.: :8 88 88 :8 :8 88 :8 8.: 8.: 2.: 8.: :.: 3: N-
E: 8.: :8 88 :8 :8 :8 :8 3.: :8 :8 :.: 8.: 8.: T
:8 a: 8.: 88 :8 :8 E8 :8 8.: :.: :8 :.: 8.: :.: o
8.: 8.: :.: 8.: :8 8.: $8 :8 :8 8.: 8.: :8 3.: 8.: _
8.: :8 S: E: 88 88 :8 88 $8 58 :8 :8 S: :.: N
8.: :8 8.: 8.: 8.: 88 :8 :8 :8 :8 :8 :8 :.: mm: m
3: :8 :.: 8.: :.: 8.: :8 :8 :8 :8 :8 8.: :8 :.: v
8.3. 9.: E: 3.8 :.: 8.: 88 8.: :8 88 8.: on: :8. Q: m
:8 E: 8.8 8.: 3.: 8.: 2.: :8 :8. 88 :8 :.: :8 2.: o
3.: :8 Q: 8.: :8 :8 8.: 88 :8 :8 :8 8.: :8 S: b
8.8 3.: E: :.: 8.: :8 :.: 8.8 :8 88 :8 :8 8.: 8.: w
:_ 2: 8 : 8 :. g cm : 8 : 8 2 w :

2:5: 88m 88m

 

 

 

5 Es: .. 3 so: a: 25:: E :EEE E :5

259

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ovdm meow no.3 nowv omdm $.mm mm.wm vnoo mndo 03% no.3 nn.oo mm.no mono w-
nonm oodm vmom o: .ov 3.? 86¢ wvdv ow. M m oo.mm 56m N: .nm monm mn.wm Qdm n-
3.3 mm.mm _mnm mm. _ m No.3” oowm mo. 3 3.? 3.3 oonv nmwv mi? 3.3 wmom o-
vowm mwom vnw: modm 36m omdm 3 .mm mmdm 8.0m mm.mm wmdm moov 3. :u mwiv m-
mmom mm.wm onom dem 3.: voom 2:3 mw.mm wwnm mmdm Som ngm vﬁmm vodm v-
om. : m :wdm om.mm onom vaN mivm mg: o: .n_ 2.2 owom wo.~m oodm $.mm mmém m-
3.3 Vm. g m nmdm wo.wm _n.om oimm mmém mo.m~ oodm no.2 3.2 3.: wow: 862 N-
2 .mm mw.~m moom mmdm vwnm niom 3 .mm 3.3 oodm on. : m mmom mnd: 8.2 m3: 7
mnom 36m VmNm Eom oodm Snm no.8 oném nmgmm SNN oogm no.om 9:2 no.2 o
mm.wm vwdm :n.mm ow. _ m nﬁom 3.3 3.3 om.mm mo.vm mn.mm ﬁlm mvom no.3 2.3 ﬂ
wodm mmnm 36m 2.?“ om._m Eda monm 3.6m wném omﬁm No.3 onom 5.3 3.2 m
9. _v mmwm wvém mvém Exam moom wwwm :.nm 3.3 mw.m~ Sada SAN ﬂow and: m
¢o.mv mmov ownm wo.mm mo.mm on. ﬂ m nndm Nona 26m Siam mwNm wv. : m mmom 3.2 w
8.3 mm. 3 omdm no.3 oném on.mm moon mn.w~ mwom No.3 omdm ow.§ ooom mod: m
3.? 3.? moov :mwm No.3 mn.mm No. _ m mmdm _mnm mm.mm nn.mm 2 .mm vwom mnd: o
nwnv 3.3 no.3 omdm oo.nm and». dem mmom wmwm o: .80. mmém ovdm wo. _ N two: n
wvdv omdv it? meow mm.wm wwwm 51mm 2.3 mowm mndm ondm mwdm wmAm 3.3 m
N: voﬁ om ww ow an we om wv ow mm vm o: w o\o
EEOC 30mm 2320

 

 

 

5 as- 3 so: .3 85%::EEE 2::

260

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SEN? oodv Stow No.3 owoo ovgn Noon vndn mwdw Stow omnw oodw o: .oo Nooo w-
moom _nom wo.mv oowv nmém moon oodo mmno Nvon Son moon voon no.nn o_.wn n-
ooom omém oo.om omom voév _wov oo._o owéo oo.wm wmoo oodo oméo vmoo on.mo o-
oo.nm nvom _o.mm n_.mm nmdm wmém mo.wm omdv ovov moov noon wogm mmdw mm.mm m-
Nwdm mmnm omom mm.mm oo.Nm SNN mmom modm moon moom on.nm _N.wm omov ooév v-
vn._v vodm Noom oo.mm ovdm oogm ovom no.2 oo.o~ maﬁa nm.mm nwom oo.nm vowm m-
mogﬂq owov mm.wm ooom mném 3.2 mmgm modm wo.wm omnm ow.m_ all om.m_ om.o_ N-
no.9q NoNv wﬁov oo.wm Eom wmém No.Nm noom _mom anm moom wmoN SéN NTNN T
wvnv wmév 33v omom ov.nm moom ﬁndm mogm wwom doom 2 .nm 3.2 nﬁmm moém o
mvdv oﬁov om.mv __.—v ow.wm Nwom owgwm ww.mm no.3. dem on.nm Stow NVﬁN mném _
nmAm monv 26v modv omov oo.wm voom ow.mm ooAm wodm nm.wm Nwom wo.mm 3.3 m
mm.mm ondv on.ov No.3V mogv omdm Nonm owém mn.~m moom on.wm wranm oooN ooom m
Snow mmgm mvwv No.3V modv Vmov :.wm wn.mm om.mm wmgm omom monm mmom nﬁmm v
ooom mm.mm noom N_.nv wmév ongv omdm onom mmdm modm wwdm no.wm 3.8 deN m
modm owém mngm Noow nn.mv modv omov wmnm oﬁom mn.~m ovom wv.wm Nwom wig o
24o ooom mm.mo wﬁom N_.nv nmév mi; mo.wm ooom 0m.mm No.3 on.wm wonm wo.m~ n
oo.No Sow owéo wogm wav ovov 3.? oodm Nwom Sim mmgm Rom anN Nvom w
m: v3 oo mm ow Nn vo om wv ov mm VN o: w 8..
35:3: ”25

 

 

 

g E - 3 :8: .3 85%: E :EEE : m:

261

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mwﬁu mmdm omdm nmém Emom oo.wm mmdm wmdo mvo oo.~o modo mnéo vmoo wmoo w-
oodm oﬁmv N03 Vmov oﬁwv :dv ozm omdm mm.mm Rem Smmm Egon omom 8.3 5.
ﬁlm ow.mm ooom __.wm nwdm mi? 3.? S .3 3.3. gov NW: 3.? $4? 3.3 o-
omdm moon Elm wwdm oo. ﬂ m 3.3 ooém omdm woSm :.wm aqwm mwdm oodv wv. 3 m-
mmdm oi: wo.2 vogm ovdm oo.m~ ovom 3.5m wwwm wwdm ondm mmgm NVNm vmdm v-
ooém ooh». 24m 3.2 m”? mwoﬁ m3: No.2 oodm vogm meN 3.3 wﬁwm 5.3 m-
nwdm S.wm wo.mm ovdm 2.7m coda no.2 3.: 3.2 wm.§ 3.3 2.2 3.2 mw.o_ N-
nmwv 5.9 wﬁdv 36m omém omdm wmdm Sea vw.mm omém 2am Exam mus oo.w T
mwdm m5? v53 mo: wo.wm wndm wmdm ovdm aﬁwm omom moém om.m~ 25m oo.m~ o
No.3 Nwdm vmdv mwdv .GNV 2.0m 006m 36m ovdm _o.wm ﬂow obém Nada 8.3 a
No.3 wvsm mw.mm 36v Sow 3.9» mowm wowm modm mmdm moSN wwwm modm omﬁm N
owoo ooNo Ewm nmém 3.3 8.3 3.? mm.wm 34% v5; Edm woom Nina 3.3 m
M3: mmSo oo.~o Gwm owém omdv coo... oodv 3.5m Kdm modm Ewm ﬂow Com v
2.: aim» mmso oo.mo wt? mwdm mm: oodv Siam boom oﬁmm omdm mmom mmdm m
mmdw 2.: mad. owoo ozo Noon oo._m Vmov woﬁq at? oo.mm dem no.3 owwm o
36w 5.; ﬂow wads vmwo mmdm mmdm 3.: oo.mv mmdm oﬁmm wvgm mvwm omom N.
3.8 oaow wndw wows nwwo ooNo ooom mmgm Sow 8.3 oo.om Nod». omdm boom m
N: X: om ww ow Nb vo om wv ow mm vm oﬁ m X.
3E5: 8on combo

 

 

 

3 £3 - E 22% .3 855%: E E 3%: a is

262

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3% 9.8 2:... 3% 3% No.8 3% 8.3 38 38 RS 3% N30 an? m.
3.8 E: 8.? 2.? 3.x mm? 8.3 38 mm: 2.? ?.? 3% mm? 2.8 :
was. one. 93. 8.2. 8.2. $9. $5. 3:. 2.9. $9. 3.3 3.9.. 9.: mm: o-
3.? 32% E? 8.3 9...: mm? 2.? 8.? 8.9. 3:. 3:. 3:. 3:. on? m-
E? 9.8 a? 5.? we? 3.? ?.cm on: 8.? 3.? 3.? 8.3“ $4.... $4..” v-
3.: 8.: mm: ?.2 3.8 .2.? mm? 8.? ?.?. :1? 8.? S? 8.8 9.8 m-
?.S 8.? 3.? 5.8 2.: 3.: a: 3: mm: 5.2 $2 8.: 8.: 8.: N-
8.? 3.9. 32 S? 3.? ?._m 2.8 3.? a? S? 3? 3w 2.0 8.0 7
S9. 8.3. 8.? 2.9. 8.: mm? 3? 8.? 3% 8.? S? P? S? «a? o
3.? 8.: 2.9. 3.3. 3.2. 8? we: 8.? mm? 5: 3.3 8.? 8.? 3.? :
6.3 5.: 8.? 8.2. :9. 2.? $9. 8.? on? 3.? 3.: a? ?.? 2.? N
RS 2.8 8.? 8.x 9% 8.2. S? 39. no? 3? 3? 3.: 8.? 3? m
2.? 3:. 3.8 ?.? Mn? ?.om 3:. 8.2. 8.2. 8.? 2.? mm? 9.3 8.? v
a? 2.? 3% 3? ?.? a? on? :9. 9.? 2.9. 8.2 8.3. Na: 8.? m
83 3.? 2.? 3% 2.8 8.? 3.? $9. 3.3. 3:. 3? 2.? 2.? 8.? e
was 3% a? 3.? SE 2.8 no: 8.? 2%. ...mi. 9.9. No.3 3.? .2.? ?
E8 3% 3.? 2.? 9.? 88 3.8 E? 8.3 :9. 3:. as... 3.? 3.? m
N: 2: 8 mm om ? E on a. 9. mm a 2 w .x.

3.5: Beam 82o

 

 

 

.38? 5232 I as: 82% 380 85.5.5.2 as has: 8.83 pm 23

263

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9.69 5?: 829 39.9 2?: 9.89 8:: ~39 ?.9._ 9?: 39.: ?.§ :29 :22 9
~29 ?.9_ 8.9: $2: 9.9: 8?: 9?: 3?: ~59 5?: ?._9 829 829 ?.?: 9
?.? «a? 9.8 3.8 ?.8 :39 8.9.9 ?.9: :9: 9.9: :3: 8.9: 8.9: ?.9.: ?.
?.? 9.? ?.? 9.? E? ?.? ?.? 8.? 8.? ?.8 8.? ?.8 8.8 ?.8 ?.
?.? ?.? 1...? ?.? 8.? ?.? ?.? 8.? 8.? ?.? 9.? ?.? a? ?.? 9.-
?.? K? ?.? 9.9. 8.9. on? 8.9. a? 9.? a? 9A.? ?.? ?.? ?.? m-
3? ?.? 8.? 9?? a? ?.?. 9.? S? ?.? 8.? ?.? ?.? ?.? 8.? N-
:..8 :..8 ?.? ?.? ?? 8.? ?.? 9w? ?.? 93.? ?.: 3:9 ?.9 2.? 2-
8.9: 8.9: 9.8 9.? 8.? ?? ?.? ?.? ?.? 9? 9.? a? 89.? 9.? o
9.?9 ?2_ ?.?9 we? ?.? 8.? ?.? 9.? 9.? .?.? 9.? ?.? ?.? 9.? :
9.29 ~39 ?.:_ ?.S_ ?.8 8.8 a? ?.? 8.? ?.? ?.? «n? ?..? 9.? m
3?: $9.: 5?: :9: ?.9: 2.9: ?.8 ?.? ?.? 9.? 9.? ?.? :.? 8.? m
3?: ?.?_ ?.9.: 5?: 9.9: 5.9: 9.9: ?? ?.? 8.? 8.? ?.? 8.? 5.? 9.
9.9: ?.?: :99 9.29 3?: ?.:: 9.9: 9...? 9:8 9:? 5.? ?.? 3? 8.? m
:19 5.: 9.92 ?.o9 3?: 9.?9 5?: 9.9: 9:? ?.? E? 2.? 8.? 8.? ?
?.?? 38: ?.m: 8.2 9?: ?.:.9 9.3.9 8.9: 39: ?.? on? ?.? ?.? ?.? ?
?.?.~ ?.~? ?.?9 mi: 9?: N3: ?:9 89: ?.9: ?.? ?.? 8.? E? ?.? m
N: x: 8 ? 9w ? ? ? 9. 9. mm 9.? 2 w .8

992V 30? 2.90

 

 

 

925. >295 I A89 ooc<mv 980 oocmaouEmE 98 hand BEES 5m 2an

264

 

I'm

a... v

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9.2.9.2 932 ?.22 ?.?2 222.2 ?.?2 ?.?2 ?.?2 ?.?2 ?.?2 ?.?2 ?.?2 ?.?2 92.?2 9
?.?2 ?.?2 28.92 ?22 92.2 9.?2 9.?2 ?.?2 ?.?2 8.?2 8.?2 8.?2 ?.?2 ?.?2 9
9.9: ?.222 ?.92 ?.22 8.9.22 ?.22 2.22 8.22 ?.:2 ?.:2 8.92 9.92 8.92 2992 9
?.? ?.8 ?.8 ?.8 ?.8 ?.8 ?.8 9.? ?.8 8.8 ?.8 ?.? ?.8 :9: ?.
?.? ?.2? 9.? ?.? ?.? 9.? ?.? ?.? ?.? ?.? ?.? :.? ?.? ?.? 2.-
2.? ?.? ?.? 8.2.? ?.? ?.? ?.? ?.? ?.? ?.? ?.? ?.? ?.? ?.? m-
9.? ?.? ?.? ?.? ?.? 2.? ?.? ?.? ?.? 8.? 9.? ?.? ?.9. 9.2.22. 9
:..29 ?.8 9.? ?.? ?.? ?.? ?.? 2o? 8.? ?.9 ?.9 2.? ?.? ?.? 2-
2...?2 ?.22 ?.9: 8.9: ?.8 ?.? 2.? 8.? 9.? ?.? 2.? ?.? ?.? ?.? o
29.2.2 2S2 ?.?2 9.92 9.9: ?.9: 2?.? 29? ?.? 2?.2? 9.? ?.? ?.? ?.? 2
?.?2 ?.?2 :..2.2 9.?2 ?.?2 8.9.22 ?.?2 ?.8 ?.? ?.? ?.? 22.? ?.? ?.? N
28.9: 9.?2 ?.?2 ?.?2. 9.9.2 ?.?2 8:2 239 ?.8 ?.? ?.? ?.? ?.? ?.? m
?.22? ?.?2 8.9: ?.?2 ?.?2 8.9.2 ?.?2 ?.92 9.2.82 ?.8 ?.? ?.? 9.? ?.? :.
?.?? 9..:~ ?.~? ?.?2 ?.?2 ?.?2 8.9.2 ?.?2 ?.22 2.9: ?.8 ?.? ?.? 92.? m
9?? ?.?N ?.2? 8?? ?.?2 ?.2: ?.?2 9.9.2 ?.?2 ?.92 8.? 8.? ?.? 9.? o
2?? ?.?~ ?.9? 2?? 9?? 9.?2 ?.?2 ?.22 ?.22 2992 22.9 ?.8 ?.? 22.? ?
?.8? ?.?~ ?.?~ ?.?~ ?.9N ?.22? 229 ?.22 :92 ?.?2 9.92 2.8 ?.? ?.? w
92 9.9 8 ? ow ? 9.? ? 9. 9. mm 9.? 2 w 8
92:52: 2.8? 02.90

 

 

 

.29.? 2.9.2829... I 2252 9299 980 89.52.2922 2.9 29232 2.39:2: 9m 2.22.

265

APPENDIX C
ROUGHNESS F EATURES-F IELD TRIALS DATA

Table C.1 Distress Data Collection for Site 1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm) Explanation Other
Feet Inches Left Center leghtiAverage
0 0 1.5 l 1.5 1.33 Station 386+36
41 2 2.3 2.4 3.1 2.60 Joint
82 3 0.9 1.9 3.6 2.13 Expansion Joint
123 4 0.6 1.2 1.6 1.13 Joint
164 2 0.5 1.4 1.7 1.20 Joint
191 7 Pothole with D = 6 in.
178 3 Pothole with D = 6 in.
205 6 1.1 2 2 1.70 Joint
246 9 0.2 1.2 1.7 1.03 Joint
271 8 Pothole with D = 4 to 8 in.
287 11 2.3 4 5.5 I 3.93 Joint
301 4 6 in. break (3ft) 2 in. wide Crack
331 0 7 6.7 4.5 6.07 Joint
Negative patch
371 10 2.4 3.9 4.5 3.60 Joint 1ft befoiioint
413 7 5.3 5.3 4.9 5.17 Expansion Joint
419 11 1.6 2.3 2.9 2.27 Crack
435 5 Patch
454 7 6.5 5.9 5.7 6.03 Joint
Shallow pothole
(4 to 5 in. wide)
at l to 2 in.
away from leﬁ
495 8 5 5.7 4.9 5.20 Joint wheelpath
536 8 5.3 5.4 3.9 4.87 Joint
567 6 Station 392+00
577 9 3.8 4.5 2.9 3.73 Joint
607 10 -0.4 0 -2.3 -0.90 Patch start
614 1 -0.7 -0.3 0.1 -0.30 Patch end
618 9 0.7 -1 -3.2 -1.17 Joint
624 9 -0.7 0.5 -0.3 -0. l 7 Patch start
630 8 2.4 2.8 1.6 2.27 Patch end
659 6 6.2 5.2 5.5 5.63 Joint
667 1 Station 393+00
700 6 6.3 5.7 5.3 5.77 Joint
741 7 1.7 2.3 3 2.33 Expansion Joint
782 6 4.4 4.9 4.7 4.67 Joint
823 6 3.8 3.3 3.6 3.57 Joint
841 l Pothole Figure Cl
848 3 -0.8 0.1 0.2 -0.17 Sealed Crack
864 7 3.5 1.7 1.9 2.37 Joint

 

 

 

 

 

 

266

 

Table C.1 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm) Explanation Other
Feet Inches Left 1 Center J RiLht 1 Average
883 11 Pothole Figure C2
905 8 5.4 6.3 6.8 6.17 Joint
946 8 5.4 5.8 4.9 5.37 Joint
987 7 5.3 6.6 6.7 6.20 Joint
1028 8 6 1.8 2.9 3.57 Joint
1069 8 5.1 4.3 5.3 4.90 Expansion Joint
1092 9 -0.4 0.5 0.3 0.13 Sealed Crack
1110 7 0.7 0.8 0.4 0.63 Joint
Y crack

1132 7 2.1 -0.1 -0.7 0.43 Sealed Crack (Figure C3)
1133 11 0 1.8 1.5 1.10 Sealed Crack
1151 9 3.8 3.4 2.9 3.37 Joint
1192 9 3.1 5.3 5.9 4.77 Joint
1233 10 3.3 3.2 2.8 3.10 Joint
1275 0 6.5 5.8 3.7 5.33 Joint
1296 11 0.1 0.2 -0.4 -0.03 Sealed Crack
1316 0 6.4 7.5 7.4 7.10 Joint
1357 0 1.6 2.9 2.5 2.33 Joint
1398 l 4.5 5.2 7.2 5.63 Sealed Crack
1438 0 5.4 5.6 5.7 5.57 Joint

4 in./ 1 ft. of

spalling on
right
1453 O 0.5 2.9 3.1 2.17 Sealed Crack wheelpath
1479 1 1 5.1 5.6 5.5 5.40 Joint
1520 0 3.8 5 5.2 4.67 Joint
1562 0 0.8 2.3 2.8 1.97 Joint
1603 1 7.9 8.2 7.8 7.97 Joint
1644 2 5.3 5.7 6.4 5.80 Joint
1685 1 3.5 3.6 4.4 3.83 Joint
1726 1 3.5 4.2 4.6 4.10 Expansion Joint
1736 l 4.9 4.8 4.9 4.87 Sealed Crack
1769 2 1.7 3.6 4.3 3.20 Joint
1808 1 3.1 5 5.4 4.50 Joint
Right
wheelpath joint]

1849 4 4.8 5.3 5.1 5.07 Joint width = 0.5 in.
1856 6 1.7 2.2 1.5 1.80 Crack
1896 0 4.7 5 5.6 5.10 Joint
1931 7 3 2.6 4.4 3.33 Joint
1972 8 6.3 5.8 5.7 5.93 Joint
2013 11 4.4 4.2 3.9 4.17 Joint
2026 3 -0.8 0 —0.1 —0.30 Sealed Crack
2038 6 -0.6 -0.2 0.1 -0.23 Sealed Crack

 

 

 

 

 

 

 

 

 

267

 

Table C.l Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faultin (mm) Explanation Other
Feet Inches Left Center Right Average
Right
wheelpath
Joint width =
0.5 in.
(Figure CA for
2055 0 23.4 9.5 4.8 12.57 Expansion Joint Breaks)
2095 11 2.2 3.7 3.1 3.00 Joint
2137 l 3 3.1 3.3 3.13 Joint
2158 8 -1 -0.7 -0.2 -0.63 Sealed Crack
2171 4 Station 408+00
2178 5 4.4 5.2 6.3 5.30 Joint
2219 6 1.5 2.6 2.5 2.20 Joint
2267 1 1.6 2.3 2.6 2.17 Joint Station 408+95

 

 

 

 

Direction of Trafﬁc

 

 

    

1 in. 1 in. 2 in.

Figure C.l Pothole at 841 mile

 

 

Direction of Trafﬁc

 

 

3 in. l in. 1

Figure C.2 Pothole at 883 mile

 

 

 

 

 

I Direction of Traffic

/

Figure C.3 Y crack

T Direction of Traffic

 

 

 

 

-85mm -ll.8mm

 

 

 

 

Shoulder

 

 

 

 

 

18 in.

 

 

 

 

Figure C.4 Breaks

 

268

 

 

 

Table C.2 Distress Data Collection for Site 2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm)
Feet Inches Left Center Right Comments
3 2 station 693+00
1 5 6 0.6 0.9 crack
44 1 joint
58 1 1 14.1 1 1 crack
69 1 0 -2.9 3 crack
85 9 joint
101 8 2.1 4.7 crack
126 6 'oint
148 10 0.2 0.7 crack
167 8 joint
1 81 4 1 .8 2.6 crack
209 3 joint
225 2 0.5 1.5 crack
251 0 joint
268 7 4.8 4.4 crack
276 9 -2.4 -1 .8 crack
292 5 joint
307 1 1 2.4 7.2 crack
316 9 0.9 1 .8 crack
334 0 'oint
349 9 7.4 7.9 crack
363 10 -1 .1 -1 crack
375 6 joint
399 10 3.3 4.5 crack
416 1 0 joint
431 11 0.2 2.3 sealed crack
444 2 -1.1 -0.5 sealed crack
458 7 'oint
468 8 0.1 -0.5 sealed crack
477 1 -0.8 -0.6 sealed crack
487 10 1.6 3.2 wide sealed crack
505 0 joint
516 7 0.9 2.7 sealed crack
527 O 0.1 -1.1 sealed crack
541 4 joint
554 4 -O.5 -0.5 crack
571 0 1.2 13 sealed crack
582 10 0.6 -0.7 joint
605 8 station 699+OO
624 4 joint
635 6 -1.7 0 sealed crack
644 9 -0.5 1.5 partial sealed crack
650 8 7 5.9 crack

 

 

269

 

 

 

Table 02 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance FaultianJmm)
Feet Inches Left Center ﬁght Comments
664 11 -0.3 0.2 joint
676 9 -0.2 -0.4 sealed crack
704 6 -1 0.5 sealed crack
770 0 joint
758 1 joint
778 3 0.4 0.5 sealed crack
799 4 joint
812 8 6.5 10.6 sealed crack
820 1 0.2 —0.8 sealed crack
826 9 3.9 2.9 sealed crack
840 6 'oint
858 2 3.5 4.5 sealed crack
867 1 -1.1 -0.3 sealed crack
882 1 joint
900 10 -1.1 1 sealed crack
911 8 0.8 -1.1 sealed crack
923 1 -0.3 -1 joint
942 10 1.8 2.2 sealed crack
953 4 -0.5 -1 sealed crack
963 5 -0.5 -0.4 joint
988 8 2.8 4.8 wide sealed crack
1 004 8 'oint
1024 3 -1.4 -1 sealed crack
1045 4 joint
1063 1 -0.5 -0.3 sealed crack
1075 1 1 -1 0 sealed crack
1086 5 -0.3 0 joint
1 128 1 joint
1167 5 -0.3 -0.3 ' sealed crack
1 169 4 joint
1210 0 0.5 1 .2 pint
1 251 5 0.4 0 joint
1274 1 1 1 ~07 crack
1293 6 joint
1306 8 4.5 6.6 wide sealed crack
1 334 8 1 .8 0.7 joint
1343 5 0.4 0 sealed crack
1374 7 1.9 1.4 joint
141 5 9 joint
1456 8 joint
1497 7 joint
1 538 4 joint
1 579 2 joint
1620 2 joint

 

270

 

 

Table C.2 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faultipg (mm)

Feet Inches Left Center Rjﬂ Comments
1661 3 joint

1688 1 1 -1.2 0.2 sealed crack
1701 6 -0.1 0.3 joint

1732 10 0 0.3 sealed crack
1742 4 joint

1756 5 -0.8 1 sealed crack
1783 3 joint

1824 1 joint

1843 7 0.2 -0.5 sealed crack
1865 4 joint

1906 2 joint

1947 2 joint '

1988 1 joint

2029 1 joint

2069 11 0.9 0.6 joint

2085 4 1.6 2.3 sealed crack
2096 11 5.3 4.6 sealed crack
2101 2 -0.1 2 sealed crack
2110 8 -0.3 0.4 joint

2151 3 0 03 joint

2192 5 0.1 06 joint

2233 0 joint

2274 0 joint

2315 0 joint

2356 0 joint

2397 0 joint

2438 0 joint

2479 0 joint

2506 3 O -0.9 sealed crack
2520 0 joint

2561 0 joint

2572 8 -0.2 partial sealed crack
2602 0 joint

2616 9 -0.2 -0.9 sealed crack
2629 11 4.4 3.8 wide sealed crack
2643 0 'oint

2684 0 joint

2720 1 station 720+00
2725 8 joint

24 I 4 i l [ error

 

271

 

 

 

Table C.3 Distress Data Collection for Site 3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm)
Feet Inches Left Center RLtht Comments
0 0 station 773+00 (core)
22 8 joint
32 9 4.6 8 crack
63 5 joint
87 4 -0.5 0 crack
104 4 joint
145 7 joint
186 4 joint
1 99 4 0.7 4.1 crack
227 2 pint
234 7 1.4 5.3 crack
258 8 1.6 4.3 crack
267 8 joint
300 6 station 776+00
309 4 joint
350 2 joint
362 4 1 3.3 crack
391 2 joint
432 3 joint
456 9 4 9.5 crack
465 0 1.6 1.6 crack
473 1 joint
488 2 0.6 1 crack
500 11 station 778+00
503 3 5.8 6 crack
514 1 joint
526 6 4.4 6.8 crack
554 1 1 joint
596 4 joint
636 1 1 joint
649 3 5.8 7 crack
656 4 3.2 3.3 crack
662 5 -1.6 crack
665 5 8 crack
672 2 24 diagonal crack
678 2 joint
701 4 station 780+00
719 1 joint
747 9 mid-slab (8inch) deep>1 inch
759 10 0.5 0.3 joint
790 1 0 -0.4 2 crack
800 10 joint + station 781 +00
814 1 3.7 7.6 wide crack

 

 

 

 

 

 

 

 

272

Table C.3 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm)
Feet Inches Left Center Right Comments
822 10 0.8 -0.6 crack
841 1 1 joint
854 6 5.6 8.3 crack
882 10 joint
897 9 2.5 6 crack
923 1 0 joint
933 3 2.3 5.9 crack
945 0 8.3 9.6 crack
955 7 0.3 1 .8 crack
964 9 joint
1001 7 station 783+00
1005 10 joint
1046 10 joint
1063 8 4.3 8 crack
1087 10 joint
1100 2 4.1 8.9 crack
1 128 10 joint
1137 10 -0.3 0.7 crack
1143 7 3.5 6.3 crack
1 153 6 ~04 4.4 partial crack
1158 10 -1 —1.8 crack
1 169 2 joint
1 183 5 3.4 6.4 crack
1196 5 3.4 3.5 space wide crack
1202 3 station 785+00
1210 6 0.5 0 'oint
1231 1 1 0 3.1 crack
1240 3 0.1 -1 .4 crack
1251 8 joint
1292 1 1 joint
1302 4 station 786
1304 6 1 .2 4.4 crack
1333 10 joint
1345 1 0.6 6.6 crack
1358 1 5.1 5.7 crack
1365 11 1.7 1.6 crack
1375 2 joint
1384 1 1 2 6.4 crack
1394 4 2.5 1.8 crack
1404 5 2.8 4.2 crack
1416 1 joint
1426 5 4.8 2.8 crack
1438 0 3.9 5.8 crack
1446 7 -0.2 -0.8 crack

 

273

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table C .3 Cont’d
Distance FaultinLme)

Feet Inches Left Center Right Comments
1457 4 joint

1469 4 5.5 1 .1 crack

1476 6 -0.5 crack

1478 9 -0.8 crack

1498 0 expansion joint
1510 2 1 3.2 crack

1540 1 joint

1581 0 joint

1595 2 -0.1 2.5 crack

1612 1 -1.8 -3.8 joint (patch)
1682 3 joint

1663 8 joint

1694 9 0.5 1 crack

1705 0 joint

1706 0 station 790+00
1712 0 -0.4 0.9 fault depth repaired joint
1732 1 -1 .7 4 end of previous joint
1746 1 1 2.5 joint

1787 6 joint

1828 6 joint

1869 6 joint

1910 7 joint

1951 6 joint

1993 5 joint

2034 6 1.6 0.6 joint

2053 8 5.2 6.3 crack

2063 5 -0.4 0.3 spoiled crack
2076 3 joint

2094 6 0 0.8 crack

21 16 8 joint

2158 2 joint

2199 3 -0.3 0.8 joint

2240 5 joint

2255 6 -O.2 0 crack

2281 8 joint

2297 8 3.1 3.7 crack

2323 0 joint

2364 3 'oint

2379 0 1 2.1 crack

2405 4 joint

2408 8 station 797+00
2434 9 0 1 .7 crack

2446 7 joint

2459 4 -0.5 -0.7 crack

 

 

 

 

 

 

 

274

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table C.3 Cont’d
Distance Faultinjjmm)

Feet Inches Left Center Right Comments

2488 1 joint

2504 4 1.5 3.3 crack

2512 3 5.1 2.2 crack

2520 3 -1 .5 -1 .4 crack

2529 3 joint

2541 7 5.6 6.5 crack

2550 1 1 .3 1 .2 crack

2570 11 0.3 1 joint

2612 1 joint

2631 1 4.3 6 crack

2640 1 0.2 -0.3 crack

2653 3 joint

2694 7 joint

2712 1 (core) station (800+00)
r 12 I 0 I 1 error

 

275

 

 

 

Table C.4 Distress Data Collection for Site 4

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faultinﬂnm)
Feet Inches Left Center ﬁght Comments
0 0 station 130+00
0 6 10 9.1 joint
57 1 1 .9 1 repaired joint (patch)
110 4 2 4.3 repaired joint (patch)
166 4 0.7 2.5 spoiled crack (10" wide)
216 4 8.4 7.9 joint
266 1 5.4 4.9 wide crack
287 8 9.4 8.3 joint
323 1 0.5 0.4 spoiled crack (10" wide)
358 9 6.6 15.3 spoiled joint (4" wide)
9
399 4 3.5 7 patch joint
430 4 8.8 6.2 joint
445 7 2.9 5 spoiled crack (3" wide)
501 6 station 135+00
501 11 9.8 7.8 joint
523 8 -0.7 2.3 crack
573 7 6 5.2 joint
644 1 1 3.9 6.4 spoiled joint
665 7 4.5 6.5 repaired joint matchL
716 4 8.1 7.1 joint
738 5 3.7 3.5 repaired joint (patchL
787 6 7.3 6.6 joint
837 2 4.3 4.9 spoiled crack
859 2 6.6 7.2 joint
930 10 4.1 5.7 joint
954 6 4.9 5.3 wide crack
1003 1 3.2 3.3 station 140+00
1045 5 1 .6 1 .2 patch joint
1073 8 7.2 5.4 joint
1106 9 3.7 2.8 crack
1119 8 3.2 4.3 crack
1 145 6 10 8.5 joint
1 166 1 1 .1 5.5 repaired joint (patch)
1 194 5 4.5 4.8 repaired joint (patcip
1217 1 10.8 7.7 joint
1235 1 1 start patch
1242 6 end of patch
1288 4 6.9 5.6 joint
1337 3 start patch
1343 10 -0.4 0.9 end of patch (repaired joint)
1381 9 start patch
1400 0 3.2 5.5 end patch (repaired patch)

 

276

 

 

 

Table C.4 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm)

Feet Inches Left Center Right Comments
1431 8 5.6 5.5 joint

1448 8 6.1 5 crack

1467 5 5.4 8.5 crack

1483 6 start patch

1490 2 end patch (repaired patch)
1520 10 6.7 5.7 wide crack

1546 6 start patch

1553 0 end patch

1575 0 8.2 8.4 'oint

1593 5 0.3 0.8 crack

161 1 10 start patch

1646 10 3.3 2 end patch (repaired joint)
1 666 4 2.8 3.7 crack

1 677 4 8 6.1 crack

1699 3 1 .8 4.3 crack

1706 8 station147+00

1 71 8 6 8 7.8 'oint

1737 1 1 8 9.1 wide crack

1750 9 4.5 3.5 crack

1790 0 8.1 6.6 joint

1861 6 5.5 6.6 joint

1877 0 crack

1894 8 -10.8 -5.9 start patch

1920 4 end patch

1933 0 8.6 6.3 joint

1953 6 start patch

1960 1 end patch

1972 5 start patch

1986 9 end patch + start patch
1995 10 end patch

2004 8 9.7 9 joint

2019 7 start patch

2039 9 end

2053 4 start patch

2059 1 0 end

2075 9 13.8 10.7 joint

2092 1 start patch

2098 6 end

21 18 3 sta rt patch

2124 10 -0.6 0.3 end patch (repaired joint)
2147 4 8.9 6.7 joint

2167 7 start patch

21 73 1 1 end

21 89 0 crack

 

 

277

 

 

 

Table C.4 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faulting (mm)
Feet Inches Left Center Right Comments
21 95 1 0 crack
2208 3 station 152+00
2218 7 13.8 9.1 joint
2234 0 start patch
2240 11 end
2265 11 4.6 1.1 start patch
2272 6 end
2290 4 7.5 6.5 joint
2310 3 6.9 7.6 crack
2324 11 start patch
2344 5 end patch
2361 7 6.3 5 joint
2403 1 -2.6 -4.8 start patch
2420 1 1 6 3.1 end
2461 10 start patch
2468 6 0.9 1.6 end
2504 9 5.1 3.4 joint
2509 8 station 155+00
2532 2 start patch
2538 8 4.3 3.4 end
2557 3 start patch
2563 9 1.8 4.8 end patch
2576 4 12 10.9 joint
2627 6 start patch
2634 4 4.8 5.8 end
2647 8 9.1 7.1 joint
2667 1 start patch
2673 7 6 6.3 end patch
2682 3 start patch
2688 1 1 1 .5 4.1 end
2697 6 start patch
2704 0 end
271 0 7 station 1 57+00
2737 11 -0.8 -2.2 start patch
2752 1 1 end
2771 5 start patch
2778 0 —1.7 1.8 end
2791 0 8.1 10.9 joint
2858 11 start patch
2865 4 3 4.3 end
2892 1 4.6 5.3 crack
2933 8 9 10.5 joint
2967 3 crack
2975 6 start patch

 

 

278

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

279

 

Table 04 Cont’d
Distance Faulting (mm)
Feet Inches Left Center [ﬁght Comments
2982 0 5.4 4.2 end
3005 1 5.7 4.5 joint
301 1 6 station 160+00
3036 7 4.7 3.8 crack
3057 7 start patch
3064 1 end
3076 7 8 8 joint
31 1 1 5 crack
3127 4 5.8 3.9 crack
3148 2 1 1 7.2 joint
3163 9 crack
3219 6 2.2 4.8 joint
3259 1 4.7 4.1 crack
3276 11 1.2 7.8 non measurable crack
5
3281 4 9.4 5.6 joint
3310 3 start patch
3316 1 1 end
3342 8 5.5 7.3 crack
3362 8 9.2 9.2 joint
3406 0 3.1 5.1 crack
3414 4 station 164+00
3434 7 7.3 8.2 joint
3485 1 4.6 diagpnal crack
3487 5 5.1
3506 7 9.5 10.5 joint
351 9 4 crack
3536 5 start patch
3543 0 end
3556 8 start patch
3563 3 end
3599 8 3.3 4.7 crack
3619 4 start patch
3625 1 0 end
3636 6 crack
3649 2 9 7.1 joint
3720 5 10.9 10.5 joint
3791 11 9.8 8.9 joint
3815 7 crack +station 188+00
3827 1 0 crack
3847 4 1 .7 4.5 crack
3863 4 9.9 9.1 joint
3874 1 crack
3884 10 start patch

 

 

 

Table C.4 Cont’d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distance Faultinﬂmm)
Feet Inches Left Center Fight Comments
3891 5 end
3915 9 2.1 2.1 crack +station 169+00
3948 2 start patch
3954 9 end
3966 5 start patch
3972 1 1 end
3986 1 1 6.2 4.2 crack
4021 5 start patch
4040 6 end
4077 7 7.5 8.6 joint
41 16 11 station 171 +00
4120 7 2.3 5.2 crack
4142 0 crack
4149 1 9.7 9.7 joint
4164 1 1 crack
4173 6 0.6 1.3 spoiled crack
4192 4 start patch
4203 6 end
4220 10 3.8 4.7 joint
4241 0 crack
4248 2 2.4 diagonal crack
4249 2 2.4
4266 1 crack
4276 1 crack
4292 2 7 7.4 joint
4308 6 4.5 5.7 crack
4317 1 1 station 173+00
4326 1 1 5.2 6.9 crack
4363 10 10.5 10.3 joint
4375 4 crack
4389 5 8 8.8 crack
4400 0 7.7 8 crack
441 8 7 station 1 74+00
4435 4 8 7.6 joint
4450 4 2.3 -1.7 start patch
4470 7 6 5.6 end
4490 0 7.5 7.8 crack
4506 10 9.2 10.3 joint
4524 11 6.8 25.8 spoiled crack
18.5
F 2| 11 E i lerror

 

280

 

 

 

Table C.5 Repeated Measurements for Site 2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

281

[ Repeatability test I
Distance Faulting (mm I
Feet Inches Left Center Right I Comments
650 8 7 5.9 I crack
Direction
repeated
trafﬁc transversal measure
6 5.1 5.9
6 5.9 5.5
5.8 6.3 5.7
5.9 6.3 5.8
5.1 5.5 5.8
5.5
5.7
5.6
5.7
5.7
5.5
5.7
5.8
5.8
5.6
interval
(inch) 3 6
average 5-75 5-32 5.69
standard 0.11 0.22 0.01
error

 

Table C.6 First Repeated Measurements for Site 3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

282

L Repeatability test (1) I
Distance Faulting(mm) I
Feet Inches Left Center Right I Comments
2541 7 5.6 6.5] crack
Direction
re eated
trafﬁc transversal mgasure
7.7 6.8 7.1
7.1 7 7.2
6.8 6.8 7.1
7.1 7.1 7.1
6.9 6.5 7
7.5 7
7.4 7.1
7
7.2
6.9
6.9
7
7.2
6.8
7.1
interval
jinch) 4 6
average 7.12 7.014286 7.046667
standard
error 0.0976 0.106939 0.013156

 

 

Table C.7 Second Repeated Measurements for Site 3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I Repeatability test (2) I
Distance Faulting(mm)
Feet Inches Left Center Right Comments
1476 6 . -0.5, crack
Direction
re eated
trafﬁc transversal mgasure
-0.4 -0.3 -0.5
-0.4 -0.5 -0.5
-0.2 -0.2 -0.5
-0.7 -0.2 -0.6
-0.6 -0.7 -0.5
-0.7
-0.7
-0.5
-0.5
-0.7
-0.7
-0.6
-0.8
-0.3
-0.5
interval
(inch) 4 6
average -0.46 -0.38 -0.57333
standard
error 0.0304 0.0376 0.015289

 

 

 

 

 

 

283

 

 

APPENDIX D

DIAGNOSTIC TOOL FOR ROUGHNESS FEATURES
IDENTIFICATION AND LOCOLIZATION- USER MANUAL

D.l INTRODUCTION

Chapter 6 was intended to choose and ﬁnalize the most accurate and reasonable
identiﬁcation method for surface distresses. Various proﬁles were studied with several
signal processing techniques. The pathway method was also reviewed. Based on the
preliminary ﬁndings, user-friendly software was developed. The distress detection tool
is an engineering software application that allows users to view and analyze

longitudinal pavement proﬁles.

D.2 USER MANUAL FOR DISTRESS DETECTION TOOL

The software is an excel ﬁle with macros. Macros are written in VBA language. To
make sure that the software is running, the security level for Excel must be set to
“Medium” or “Low”. Figure D.1 shows how to change the security level. When you
run the application, a security alert message will be displayed (Figure D.2). Then, you
should accept and click on “YES”. The main window appears giving the opportunity to

the user to run or close the application (see Figure D3).

The roughness localization window (Figure D.4) contains three tabs each of them

corresponds to one distress type: Faulting, Breaks and Curling. Also, the window

284

 

 

allows the user to import a raw proﬁle (*.ERD ﬁle) and then choose left or right

wheelpaths.

D.2.1 Import/Export Windows

The application allows the user to import ERD ﬁles including the general
information and raw proﬁle, and export results in an excel ﬁle format. An ERD ﬁle
contains two independent sections, the header and data. The header part contains only

text, and the data part contains only numbers. The numbers are written in text form.

D.2.l.1 The Header

The ERD ﬁle header consists of a series of conventional readable text lines. These
lines contain the information used by post-processing tools to read the numerical data.

As a minimum, the header contains three lines of text:

0 The ﬁrst line identiﬁes the ﬁle as following the ERD format.

0 The second line describes the way that the numerical data are stored in the data
section of the ﬁle.

0 The third required line is an END statement that indicates the end of the header
portion. Any number of optional lines can be included between line #2 and the

END line.

Table D.1 summarizes the lines in an ERD ﬁle, and describes the parameters used in

line #2 to describe the numerical data.

285

 

 

 

C Lin 1050“ [M L-1 liooki

SHOW"
1”ch
Compare and Merge ﬂorist-oaks...

 

 

 

um:
WWW “airmen '

 

Figure DJ Set Up of the Security Level

 

. W a x- ‘ strv— ‘rr.’ w v—q wry- -- r‘
ritir.'rosoft Forms ‘ ' “a

“rm“ L “ ”J“ ‘A “ ' L " l ' Eywmstﬂnmwddtﬂe,sebct0Kudﬂnmdsﬂb

r
. . l 1 ..-__

_1_1_

Figure D.2 Security Alert Message

 

 

 

 

286

 

 

Adi-o.- can... .~ 1...... .‘._—-.-—----.-—-... --**-—M-M‘55“-1‘-ﬂ.-ﬁun_‘W-I—“hn-anlv-. , - —.-.- can”-..

IE3,

Roughness Localization

p0.

 

i so rm rso :90 so m
-= I

‘15 < l———

Figure D.3 Main Window

 

 

 

 

 

 

W.

m‘a—fvr- - --—-------§- W" .--—v--v----v- —-. a-‘—- --v"mv ~w‘wmme 1‘7 -- -- r--—-—Q-4'r.—- -T -- - - - - WW ra- -
Locohzed Roughness Ea

Roughness localization

I W 5353 535
I ' sas'
I
I
I
I

 

 

5“: 199* I Cm I
I r
I nun-or . L5"
I ﬂy.“ I J (‘ Racy

 

 

 

 

 

i
I naming litres! (huh) F‘—
I Myth wank) [_—

Crihrh I L]

I 0 U f I C I

I . I I U l
u /'l I I I I r I I A I TO ‘
I
”IL—‘- "’.—-.—_*_ mm _7— k WA: rt"

 

 

 

 

 

 

 

 

 

_ 7 M C-nPro’fiIe 661 Lip-.7; Prone RMS]
;—-----—1_ - .DISIanceﬁn)__ . ..: 1.2

I ..
I l
I 08<
i
I 0.6-
l
I 04

- I

. l

- I

: ' 024

1 — I _

I I

I All -_ -_ _ ' n

:J__J » .55 ALI '

 

 

 

 

Figure D.4 Localized Roughness Window

287

 

Table D.1 Summary of Records in an ERD File Header

 

 

 

Line Description

No.

1 ‘ ERDFILEV2 . 00 -- identiﬁes ﬁle as having ERD format

2 NCHAN, NSAMP, NRECS, NBYTES, KEYNUM, STEP, KEYOPT -- use

commas to separate numbers:

. NCHAN [integer] = Number of data channels.
NSAMP [integer] = Number of samples for each channel.
NRECS [integer] = Number of records of data.
NBYTES [integer] =Number of samples per record.
KEYNUM[integer] Indicates how the data are stored.

o 5, 15 = Formatted ﬂoating-point (text). The format must be

speciﬁed using the FORMAT keyword.

For KEYNUM=5, the data are stored with all channels for the ﬁrst
sample together, then all channels for the second sample, etc.
For KEYNUM=10,11, and 15, the data are stored with all samples for
the ﬁrst channel together, then all samples for the second channel, etc.
. Step [real] = sample interval (e.g., time step)
. KeyOptjinteger1= auxiliary number used by some programs

 

 

 

Optional records. Each record begins with an 8-character keyword, followed
by information associated with that keyword.

 

 

Last END -- indicates the end of the header
Line

 

 

 

Figure D.5 shows an example header which is fairly bn'ef, consisting of the three
required lines and four optional lines. (The optional lines are the ones beginning with
the keywords TITLE, SHORTNAM, XLABEL, and XUNITS.)

Looking at the second line of the ﬁle shown in Figure 3.5, we see that the ﬁle
contains data for 2 channels, with 529 samples per channel, stored as 1 record, that the
data storage format is type 5, that the interval between samples is 1.00, and that the
status of the auxiliary numbers is -1. The header shown in Figure D.5 includes names
of the units for each channel, as identiﬁed with the keyword UNITSNAM. The name
of units for the ﬁrst channel, ft, has only two characters. Thus, it is followed by six
spaces so that the name for the second channel, ft, begins in the correct column

position.

288

 

ERDFILEV2.00

2, 529, -1, -l, 5, 1.00000 , -1,
TITLE 1993 RPUG Study, Dipstick, Section 1, Measurement 1
SHORTNAMLElev. RElev.

UNITSNAMft ft
XLABEL Distance
XUNITS ft

END

0.000000 0.000000

0.416667E-03 -O.141667E-02
0.416667E-O3 0.583333E-O3
0.666667E—O3 0.916667E-O3
0.133333E-02 0.133333E-02
0.750000E-O3 -O.166667E—02
-O.300000E-02 -O.458333E-02
-O.558333E—02 -0.500000E-02
-0.625000E-02 -O.658333E-02
-0.775000E-02 -0.825000E-02

 

 

 

Figure D.5 Short Header for an ERD File with Text Data.

 

Using the format of the ERD ﬁles that MDOT gave to the research team, the input
ﬁles (*.ERD) should include all the lines in the heading presented in Figure D6. The
software will work properly only and only if the input ﬁles follow the mentioned

format.

 

ERDFILEV2.00

2, 9840, -l, 1, 5, 0.2459307, -1,
TITLE I69
SHORTNAMLEleV. RElev.
UNITSNAMft ft
XLABEL Distance
XUNITS ft
SURVDATE 05/07/2005 12:11
DISTRICT 6
COUNTY O
ROADFROM JCT CONN-96
ROADTO JCT US-127
ROADFRMP 4.510
ROADTOMP 4.968

IRILWP O
IRIRWP 0
DATAFR 1
END

 

 

 

Figure D.6 Example of Required Headings for the Input Files

289

 

D.2.l.2 Data

The data part of the ERD ﬁle contains nothing but numbers, organized into columns
and rows. The form in which the numbers are stored depends on the value of the
KEYNUM parameter from line 2 of the header (see Table D.1). The total number of
values that will appear in the data section is NCHAN x NSAMP. All of the numbers in

the data portion are stored in the same format, and there can be no missing values.

D.2.l.3 Import window

The “Import” button in the localized roughness window open a new screen asking
the user to choose an ERD ﬁle name. Then, users should click on the “Open” button
such that the application will extract all the relevant information about the site (Figure

D.7).

l Open EII CI’J'ZISE‘
D:1MDOTIl9043E.ERDI

Figure D.7 Import File Window

 

D.2.1.4 Export window

The “Export” button in the localized roughness window open a new screen asking

the user to choose a ﬁle name and its path. Then, Users could export to excel results

290

 

using the “Save” button. Users should enter a ﬁle name followed by the extension

“.xls” to export to excel ﬁle (Figure D.8).

 

D:1MDOT]19043E.XISI

Br Eli-"15'?

 

Figure D.8 Export Results Window .

D.2.2 Faulting Detection Window

Figure D.9 shows the fault detection window. Users should choose between the right
and left wheelpath from the raw proﬁle. To do that, they should check the radio point
that corresponds to the preferred wheelpath. One checked, the application copy the
raw proﬁle to the column “Proﬁle”.

Before analysis, users should specify the threshold (in inches) as well as the
reporting interval. If the worksheet is not empty, users should click on the “initialize”
button. Then, they should click on “analyze” button.

The output for the analysis will be shown on the “results” tab (Figure D9). The
application allows users to plot the raw proﬁle on the “proﬁle” tab (Figure D. 10).

In order to get summary statistics as well as faulting distribution according to their
severity, the user should click on the “histogram” button (Figure D.11). Users should
enter bin range to get fault distribution; otherwise, default values are used (correspond

to the severity level selected by the research team and deﬁned in Table 6.2), i.e.,

291

o 0.25 low severity level
0 0.5 Medium severity level

0 0.75 high severity level

 

    
 

muquw .(ww w—v. _x

 

 

    

Roughness localization
mm —

 

 

F-nfujamrslcml
I II I
Mn! N . “L5" I
I “m (‘Rﬂev I
“ ﬁgural-unrest.) [———3 I‘ 1
I

 

 

 

  

 

 

 

 

 

 

 

I mun-mgr.) I‘_—.J [ , ', "77”
I Grim-h [WT-__I I I ttttt '
r 1...... I..—..— I . - - -
1 -
| 2 7 7
I 7 3
4 t.
I 5 7 I
6 .,..
| 7 .
I 3 12mm 033 E a.
I 9 12mm on g
.10- 15mm 0.26 g
11 1 no 027 g “‘
I 12 247nm 109 '5
13. 2472900 4150 If
I 71‘ an unu- —- -1-
I 715 g I
I 16 ”‘
I 17 , , . ,
18 7 ' \z “ j, . . u]
19 7 r" ' . ~I A. V _.
2D . ': r . I
21 53 ryﬂ M .Ij' .__L " Distance inch
— - 1 b / . 1]_] p ( )

 

 

 

Figure D.9 Fault Detection Results Window

292

 

,, , _, , ,_7 T
We. ,7 - ..r __.v ___.._,__

Roughness localization

 
 
 
 
  
 
 
 
     

2.“,lﬁe-E'IOIFFL,, , V, n

nun! r———_l..ui, , Gm”
PR5"

twil-

n-r-m-c him-Ion» 3

 

 

 

 

 

 

Distance (inch)

 

 

  

l ‘I .I;

 

L—lr 9” .
Figure D. 10 Faulting Detection Window Displaying the Original Proﬁle

 

 

 

9m 5.: m: w :m W'Amﬂ.’ W 4v -- ﬂ
Sumnmo'Smn'n‘ia _
I! 12
h 0'23 M [159—_—
ﬂhl I 0.21:
W 037 m I 1.09

 

 

 

 

 

 

 

 

 

 

5 c
C
G)
:1
g
t I
z
o
0.2; :5 us Mon
"‘ Bin range (<)(Inch)
_

 

Figure D.ll Fault Results Summary and Distribution Window

293

D.2.3 Breaks Detection Window

Figure D.12 shows Breaks Detection window. Before detecting Breaks, ﬁrst, the
module detects any difference in elevation. Since, this part is the same as the fault
detection module, users should specify the same inputs as for detecting faulting; i.e
reporting interval and threshold. Second, if the sign of two successive faults is
different, the Breaks detection module checks the distance between them: if the
distance is less than 3 feet, it is considered break; otherwise, there is no break. This
module gives as a result:

0 Starting Point of the break;

0 Width of the break;

0 Fault magnitude at the stating point; and,
0 Fault magnitude at the end point

D.2.4 Curling Detection Window

Figure D.l3 shows the Curling Detection Module. First, the module cut off the long
wavelengths (high frequency) and the short wavelengths (low frequency) to ﬁlter out
the topography. A 4th order Butterworth band-pass ﬁlter was used. Second, the
module displays the original and the ﬁltered proﬁle as shown in Figure D.l4. Third,
slope of the ﬁltered proﬁle will be computed and the local maximum and zero points
of the slope function will be detected. At the end, the zero point’s location and the
difference in elevation between zero and maximum point will be resumed. Users

should specify only the reporting interval.

294

 

“‘1. _

 

 

 

Roughness localization
mm- _

 

 

 

 

     
 

a‘eie a‘a‘aiafd

an
L’.‘

 

 

 

 

 

 

 

 

 

 

 

Figure D.lZ Breaks Detection Results Window

295

 

 

 

 

 

Roughness localization

 

 

 

Hanan! I
Fuﬂulbads MIDI
nun": z... a '5"
upnm- (‘ an"

139-WWW” I 3

 

 

 

 

 

 

‘C.II.‘i Fllffred A
- 00153 -0 0300210 ‘
7 3 UCIBEI‘IZ 4303:5372 . e
7 75 09159440031101! _' ~
- viagomaaal 003nm , I _,
12 0.023325 .0 0320129 . ' 7'. , ,
15 0.033493 43032471701 ‘ , , :,,
l8. 0.0357sva.0327952 * ‘
_ 21 00353920033199: I 3522 007 ._ E ‘_
24 41034066 00334955 i 4020 gum g .
270 035964 cream? 1 4422' - , 0.0: g;
g 30 -O.11354%:U.03415611 I my my ' E
33 74103714 410344845 ~ 5532 0;” E .
i 735 41040752 -0 0345155: 1 6027 005' - U
39 0044232 00351505 5439 009 v"
42 43 045133 4103555166 I 6993 ‘ on *I I
45 0050844 {1095mm , 7521 010 "I
48 0.059472 am am A013 *
- 51 4] 033444 0.0337028! I 350': DIR
54 arms“ 0.0371534_ , m , 0h,
57 0.037433 00375351 | 9534 on, F “7' m
50 ‘0 559204 410381597“ __m75___ ‘ n’ 1 ' Distance (Inch)
O D /.

 

 

 

 

 

 

 

 

 

 

 

Figure D. 13 Curling Detection Results Window

 

 

I 681134 'Rngii 0;;

wires. SM]

...,n-.. ---3- - n. --_.4-.-—— .

-.__,-7-.2_;-t—- ...9 E‘e- 9,-3— _ .._

__ rI—E-K 'Wﬂ" ..

 

-.w—v.—._ ...u .-— ~;.— ‘»~—-.-_-'.- on.—.

Roughness localization

F'ilenune I

 

 

__.—— _.,_ _

 

 

 

44405;; '1' “ﬂ

 

 

I Unite! I _.I 6 L5"
I thpnﬁk P RElev
|
I mundane» IT—
-Jo;_ -4 '- __z
-____,_,1_ 0 00015800300210
2 3 0 009312 0 0305872
3. 5 0015944 003110381
4 9 0018984 0 031577
5 12 0 023328 00320129
5 15 0 033405 0 03241701 !
7 18 0 03575 0 03279521
8 21 0 035892 0 0331529:
9 24 0 034055 0 0334955
10 27 0 035954 0 0338282
11 30 0035495 00341550 I
12 33 0 03714 0 03445401
13 35 0 040752 -0 0348158;
14 39 0044232 00351581
15 42 0.045188 0 0355145
15 45 0 050844 -0 0358880.
17 48 0 059472 -0 03528321 I
18 51 0053444 0.0367038! *
19 54 0054544 0 0371534 '
20 57 0 057488 0 0375353“:
21 __50

0059204 0 0381527! .
Ll____l _l'

 

Elevation(inch)

 

 

 

 

4‘r :41-

Distance (inch)

 

I ~Q- Raw prom ‘- Flred porte]

 

 

 

 

Figure D. 14 Curling Detection Window Displaying the Original and the Filtered

Proﬁle

297

 

 

 

 

BIBLIOGRAPHY

298

10.

11.

12.

BIBLIOGRAPHY

. American Automobile Association (AAA), “Your driving costs”, AAA's annual

guide, 2009.

Barnes, G., and Langworthy, P., "Per Mile costs of operating automobiles and
trucks." Transportation Research Record(1864), 71, 2004.

. American Association of State Highway and Transportation Ofﬁcials, “Rough Roads

Ahead -- Fix Them Now or Pay for It Later”, 2009
http://roughroads.transportation.org/RoughRoads_FullReport.Ddf.

Lu, X. P., ”Effects of Road Roughness on Vehicular Rolling Resistance,” Measuring
Road Roughness and Its Effects on User Cost and Comfort, ASTM STP 884, T. D.
Gillespie and Michael Sayers, Eds., American Society for Testing and Materials,
Philadelphia, 1985, pp. 143-161.

Cebon, D., “An Investigation of the Dynamic Interaction Between Wheeled Vehicles
and Road Surfaces”, Ph.D Dissertation, University of Cambridge, 1985.

Cebon, D., “Examination of the Road Damage caused by Three Articulated
Vehicles”, Proceedings 10th IAVSD Symposium on the Dynamics of Vehicles on
Roads and Trucks, Swets nad Zeitlinger, Prague, 1987.

Cebon, D., “Handbook of Vehicle-Road Interaction”, Swets & Zeitlinger, Lisse,
Netherlands, 1999.

Weille, D., "Quantiﬁcation of Road User Savings." World Bank Staff Occasional
Paper, Washington DC, 1966.

Winfrey, R., Economic Analysis for Highways, International Textbook Company,
Scranton, PA, 1969.

Claffey, P. J., "Running Cost of Motor Vehicles as Affected by Road Design and
Trafﬁc." NCHRP Report No. 111, National Cooperative Highway Research Program,
Transportation Research Board, Washington, DC, United States, 1971.

Zaniewski, J. P., Butler, B. C., Cunningham, G., Elkins, G. E., Paggi, M. S., and
Machemehl, R., "Vehicle Operating Costs, Fuel Consumption, and Pavement Type
and Condition Factors." FHWA-PL-82-OOI, Texas Research and Development
Foundation, Austin, 1982.

Klaubert, E. C., "Highway Effects on Vehicle Performance." FHWA-RD—OO-l64,
US. Department of Transportation, Federal Highway Administration, Cambridge,
2001.

299

 

1.1-—

 

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

Bein, P., "VOC model needs of a highways department." Road and Transport
Research, (2), 40, 1993.

Bester, C.J., Fuel Consumption of Highway Trafﬁc. Ph.D. Dissertation, University of
Pretoria, South Africa, 1981.

Biggs, D.C., Estimating Fuel Consumption of Light to Heavy Vehicles. ARRB
Internal Report AIR 454-1, Australian Road Research Board, Nunawading, 1987.

Biggs, D.C., ARFCOM - Models For Estimating Light to Heavy Vehicle Fuel
Consumption. Research Report ARR 152, Australian Road Research Board,
Nunawading, 1988.

Bennett, C. R., and Greenwood, I. D., "Volume 7: Modeling Road User and
Environmental Effects in HDM-4." Version 3.0, International Study of Highway
Development and Management Tools (ISOHDM), World Road Association (PIARC),
2003, ISBN: 2-84060-103-6

Chesher, A. and Harrison, R. Vehicle Operating Cost: Evidence from Developing
Countries. World Bank Publications. Washington, DC. 1987.

du Plessis, H.W. editor. An investigation of Vehicle Operating Cost Relationships for
use in South Africa. In: Road Roughness Effects on Vehicle Operating Costs -
Southern Africa Relations for Use in Economic Analyses and Road Management
Systems. CSIR, Pretoria, 1989.

Hammarstrém, U. and Karlsson, B., VETO A Computer Program for Calculating
Transport Costs as a Function of Road Standard. English Translation of VTI Report
501, Swedish Road and Traffic Institute, Linkoping, 1991.

Hammarstrom and Henrikson. Reparationskostnader for bilar Kalibrering av
Varldsbankens HDM-3-samband for svenska forhallanden. VTI Meddelande Nr. 743,
Swedish Road and Trafﬁc Institute, Linkiiping, 1994.

Hammarstrém, U., Description of VETO, EVA and HDM III With Regard to Vehicle
Effects and Proposals for HDM-4. Internal Report, VTI, Linktiping, 1994.

Papagiannakis, T., "Methodology to Improve Pavement-Investment
Decisions."NCHRP Project 1-33, NCHRP, 2000.

Poelman, MA. and Weir, R.P., "Vehicle damage induced by road surface roughness"
Vehicle, Tire, Pavement interface, ASTM STP 1164, J.J.Henry and 1C. Wambold,
Eds., American sosciety for testing and Materials, Philadelphia, 1992, pp. 97-111

Huang, Y. H., Pavement Analysis and Design, Prentice Hall, 2003.

300

 

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

Byrum, C., “A high speed proﬁler based slab curvature index for jointed concrete
pavement curling and warping analysis”. Ph.D Dissertation. University of Michigan.
2001

Fernando, E. and Bertrand, C., “Application of proﬁle data to detect localized
roughness”, Transportation Research Record, 1813, 02-4050, 55-61, 2002.

ProVAL 2.7, a software for pavement proﬁle viewing and analysis, The Transtec
Group, available for download at www.roadproﬁle.com.

Zaghloul, S. M. “Effect of Poor Workmanship and Lack of smoothness Testing on
Pavement Life-Cycle Costs”. Transportation Research Record 1539.

Sayers, M. W., Karamihas, “S. M. Interpretation of road Roughness Proﬁle Data”.
FHWA DTFH 61-92-C00143 F inal Report, UMTRI, 1996.

Orfanidis, “Introduction to Signal Processing”. Prentice-Hall, Upper Saddle River,
New Jersey, 1996.

Hayhoe, G. F. “Spectral Characteristics of Longitudinal Highway Proﬁles as Related
to Ride Quality”. Vehicle, Tire, Pavement Interface ASTM STP 1164, J. J. Henry and
J. C. Wambold, Eds., American Society for Testing and Materials, 1992, pp. 32-53.

Sayers, M. W. On the Calculation of International Roughness Index from
Longitudinal Road Proﬁle. Transportation Research Record 1501 , 1995.

Sayers, M. W., Karamihas, S. M. Estimation of Rideability by analyzing Longitudinal
Road Proﬁle. Transportation Research Record 1536, 1996.

Cohen, L., T ime-F requency Analysis, Prentice-Hall, 1995

Claasen, T.A.C.M. and Mecklenbrauker, W.F.G, "The Wigner Distribution - A Tool
for Time-Frequency Signal Analysis, part 1: Continuous-Time Signals," Phillips
Journal of Research, vol. 35, no.3, pp. 217-250, 1980.

Mallat, S, A Wavelet Tour of Signal Processing, Academic Press, 1998

Shokouhi, P., Gucunski, N., Maher, A. and Zaghloul, S. M. Wavelet-Based
Multiresolution Analysis of Pavement Proﬁles as A Diagnostic Tool. Proceedings of
the 84th TRB annual Meeting, TRB, 2004, Washington, DC.

Wei, L., Fwa, T. F. Characterizing Road Roughness by Wavelet Transform.
Proceedings of the 83rd TRB annual Meeting, TRB, 2004, Washington, DC.

De Pont and J. J. Scott, A, Beyond road roughness — interpreting road proﬁle data,
Road & Transport Research, VS, 1999, p12-28

301

 

41.

42.

43.

45.

46.

47.

48.

49.

50.

51.

52.

Bogner , R. E., Constantinides, A. G., Introduction to Digital Filtering, Wiley, 1975,
ISBN 0471085901.

Murphy, SD. and Robertson D.G.E. (1994), “Construction of a high-pass digital ﬁlter
from a low-pass digital ﬁlter”, Journal of Applied Biomechanics, 10 (1994) 374—381

Chatti, K., Zaabar, I., and Lee, HS, “Development of a simple diagnosis tool for
detecting localized roughness features”, Michigan Department Of Transportation,
MDOT Project n° 1505, February 2009.

.Chatti, K., Zaabar, 1., and Lee, HS, “Development of a simple diagnosis tool for

detecting localized roughness features”, Proceeding of the 9th International
Conference on Concrete Pavements, August 2008.

Gillespie, TD. and Sayers, M.W., “Role of Road Roughness in Vehicle Ride”,
Transportation Research Record, 836, Transportation Research Board, National
Research Council, Washington DC, 1981, pp. 15-20.

Gillespie, T.D., Karamihas, S.M., Sayers, M.W., Nasim, M.A., Hansen, W, Ehsan, N.
and Cebon, D., “Effects of Heavy-Vehicle Characteristics on Pavement Response and
Performance”, NCHRP Report 353, University of Michigan and Cambridge
University, 1993.

Marcondes, J ., Burgess, G.J., Harichandran, R. and Snyder, M.B., “Spectral Analysis
of Highway Pavement Roughness”, Journal of Transportation Engineering, Vol.117,
No.5, 1991, pp. 540-549.

Marcondes, J ., MODELING VERTICAL ACCELERA TION IN TRUCK SHIPMENTS USING THE
INTERNATIONAL ROUGHNESS INDEX FOR PA VEMENTS, Ph.D. thesis, Department of Civil
and Environmental Engineering, Michigan State University, East Lansing, Michigan,
1990.

Marcondes, J.A., Synder, MB. and Singh, S.P., “Predicting Vertical Acceleration in
Vehicle Through Road Roughness”, Journal of Transportation Engineering, Vol. 118,
No.1, 1992, pp. 33-49.

ISO, Mechanical Vibration - Road Surface Proﬁles - Reporting of Measured Data.
ISOS608:1995(E), 1995.

Cole DJ. and Cebon D., “Effects of ‘road friendly’ suspensions on long-term ﬂexible
pavement performance”, Proceedings of the Institution of Mechanical Engineers, Part
C: Journal of Mechanical Engineering Science, 211(6), 411-424, 1997.

Hardy M.S.A. and Cebon D., “An investigation of anti-lock braking strategies for
heavy goods vehicles”, Journal of Automobile Engineering, Institute of Mechanical
Engineers, 209(D4), 263-271, 1995.

302

 

 

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

Prem H. “A Road Proﬁle-Based Truck Ride Index”. Austroads Publication No. AP-
R177/00. Austroads: Sydney, New South Wales, Australia, 2000.

Gillespie TD. Heavy Truck Ride. SP-607. Society of Automotive Engineers:
Warrendale, PA, 1985.

F ancher PS, Ervin RD, Winkler CB, Gillespie TD. Measurement and Representation
of the Mechanical Properties of Truck Leaf Springs. Paper No. 800905. Society of
Automotive Engineers: Warrendale, PA, 1980.

Bogsjo, K., STOCHASTIC MODELING OF ROAD ROUGHNESS, PhD Dissertation:
Mathematics, Lund Institute of Technology, Centre for Mathematical Sciences,
Sweden, 2005.

Kouta R., METHODS OF CORRELATION BY SOLICITATIONS FOR TRACKS OF VEHICLE TESTS,
PhD dissertation: Mechanical Engineering, Ecole doctorale des sciences pour
l'ingénieur de Lyon, France, 1994.

Bogsjc'i, K., “Development of analysis tools and stochastic models of road proﬁles
regarding their inﬂuence on heavy vehicle fatigue”, Vehicle System Dynamics, Vol.
44, Supplement, 2006, 780-790.

Prem, H., Tu, Z., and Ayton, G., “Ride quality measurement and user ratings: initial
assessment of improved techniques”, 9th International Conference on Concrete
Pavements, San Francisco, California, August 17-21, 2008.

Rouillard, V., “Generating Road Vibration Test Schedules from Pavement Proﬁles
for Packaging Optimization”, Packaging. Technology and Science, Vol. 21, 2008,
501—514.

Sayers, M. W. and Karamihas, S. M., The Little Book of Proﬁling, UMTRI, 1998.

Fuchs, HO. and Stephens, RI. (1980), Metal Fatigue in Engineering, John Wiley &
Sons, 155-156.

Matsuiski, M. and Endo, T., “Fatigue of metals subjected to varying stress”, Japan
Society of Mechanical Engineering, 1969

Rychlik, I., “A new deﬁnition of the rainﬂow cycle counting method”, International
journal of fatigue, Vol. 9, 1987, pp. 119-121.

Consumer Reports and personal communication with auto repair specialists (2008),
www.repairpal.com.

 

Janas, J. (2002), "Relating road roughness to vehicle component fatigue", EE-
Evaluation Engineering, SoMat.

http://www.paccar.com/ (2007)

303

 

68.

69.

70.

71.

72.

73.

74.

75.

76.

77.

78.

Sayers, M. W., Gillespie, T. D. and W. D. O.Paterson, "Guidelines for the Conduct
and Calibration of Road Roughness Measurements." World Bank Technical Paper
No. 46, the World Bank, Washington DC, January 1986.

Berthelot, C. F., Sparks, G. A., Blomme, T., Kajner, L., and Nickeson, M.,
"Mechanistic-Probabilistic Vehicle Operating Cost Model." Journal of Transportation
Engineering, 122(5), 1996, pp 337-341.

American Association of State Highway and Transportation Ofﬁcials, “AASHTO
Red Book -- A Manual on User Beneﬁt Analysis of Highway and Bus Transit
Improvements”, 1977.

Bennett, C. R., and Greenwood, I. D., "Volume 5: HDM-4 Calibration Reference
Manual", International Study of Highway Development and Management Tools
(ISOHDM), World Road Association (PIARC), 2003, ISBN: 2-84060-063-3 .

Lee, D., DEVELOPMENT OF ROUGHNESS THRESHOLDS FOR THE PREVENTIVE
MAINTENANCE OF PA VEMENTS BASED ON DYNAMIC LOADING CONSIDERA TIONS AND
DAMAGE ANALYSIS, Ph.D. thesis, Department of Civil and Environmental Engineering,
Michigan State University, East Lansing, Michigan, 2001.

Schoorl, D., Holt, J.E., 1985, “Veriﬁcation of a model for predicting damage to
horticultural produce during transport”, Agricultural Systems. 16, 67—83.

Burgess, G. 1988. "Product Fragility and Damage Boundary Theory." International
Journal of Packaging Technology and Science. Vol. 1, No. 1, pp 5-10.

Marcondes, J., Singh, P., and Burgess, G. Dynamic Analysis of a Less Than
Truckload Shipment." ASME Transactions, 1988

Singh, J., S. P. Singh, and G. Burgess, “Measurement and Analysis of US Truck
Vibration for Leaf Spring and Air Ride Suspensions and development of Tests to
Simulate these Conditions”, Packaging Technology and Science, Vol. 19, pp 309—
323, 2006.

Jones, C.S., Holt, J.E., Schoorl, D., “A model to predict damage to horticultural
produce during transport”, Journal of Agriculture Engineering Research, Vol 50,
1991, pp. 259—272.

Singh, J., Singh, S. P., and Burgess, G., “Measurement and Analysis of US Truck
Vibration for Leaf Spring and Air Ride Suspensions and development of Tests to
Simulate these Conditions”, V01. 19, 2006.

79. Newton, R. E., “Fragility Assessment Theory and Test Procedure” Technical Report,

80.

Monterey Research Labs, 1968.

Daum, M., “Shock response spectrum and fatigue damage: a new approach to product
fragility testing”, Ph.D. thesis, Michigan State University, School of packaging, 1999.

304

81.

82.

83.

84.

85.

86.

87.

88.

89.

90.

91.

92.

Personal correspondence with insurance companies, Walmart and Sum’s Club Inc.
(Lansing branch)

Rich, P., Van Pelt, T., Enshayan, K., and Cook, E., “Food Duel and Freeways: An
Iowa Perspective on How Far Food Travels, Fuel Usage and Greenhouse Gas
Emissions,” June 200121. Leopold Center for Sustainable Agriculture, 27 February
2008 <http://www.leopold.iastate.edu/pubs/staff/ppp/food_mil.pdf>.

Texas Department of Agriculture. Texas Agriculture Packs a Punch. Texas
Agriculture Facts, Go Texan Program. 12 July 2007
http://www.gotexan.org/gt/channel/render/items/O, 12 l 8, 1670_1 693_0_0,00.html.

Chesson, J.H., O’Brien, M., “Analysis of mechanical vibration of fruit during
transportation”. Trans. ASAE, Paper No. 69-628, 1971.

Tirnm, E.J., Bollen, A.F., Dela Rue, B.T., Woodhead, I.M., “Apple damage and
compressive forces in bulk bins during orchard transport”, Applied Engineering
Agriculture, Vol 14, 1998, pp.165—172.

Hendrickson, John, “Energy Use in the US. Food System: A Summary of Existing
Research and Analysis,” 2004: 8. University of Wisconsin-Madison, College of
Agricultural and Life Sciences, 27 February 2008
http://wwwciaswisc.edu/pdf/energyuse2.pdf.

43. Fancher PS, Ervin RD, Winkler CB and Gillespie TD, “A factbook of the
mechanical properties of components for single-unit and articulated heavy trucks”,
University of Michigan Transportation Research Institute, 1986.

I. Zaabar and K.Chatti, “Identiﬁcation of Localized Roughness Features and Their
Impact on Truck Durability”, 11th International Symposium on Heavy Vehicle
Weights and Dimensions, Melbourne, Australia, 14-17 March 2010.

I. Zaabar and K.Chatti, Calibration of HDM4 Models For Estimating The Effect of
Pavement Roughness on Fuel Consumption For US Conditions”, In Press, Journal of
the Transportation Research Board, 2010.

Biggs, D.C., ARFCOM - Models For Estimating Light to Heavy Vehicle Fuel
Consumption. Research Report AR 152, Australian Road Research Board,
Nunawading, 1 988.

Bonney, R. S. P., and N. F. Stevens, Vehicle Operating Costs on Bituminous, Gravel
and Earth Roads in East and Central Aﬁica. Road Research Technical Paper,
Number 76, London, England, 1967.

Antle, J ., Measurement of Lateral and Longitudinal Vibration in Commercial Truck
Shipments. School of Packaging, Michigan State University, 1989.

305

 

 

93.

94.

95.

96.

97.

98.

99.

100.

101.

102.

103.

104.

Brandenburg, R. K., and J. J. Lee., Fundamentals of Packaging Dynamics.
Skaneateles, New York: L. A. B. 1991.

Brown, T. Julian, and J. J. Troon., Vehicle Operating Costs and Road Roughness
Issues: A Southland Perspective. New Zealand Roading Symposium, 1987.

Clouston, P. B. The Effects of Road Roughness and Seal Extension Works on Vehicle
Operating Costs. Institute of Professional Engineers New Zealand, Transportation
and Trafﬁc Engineering Group, Technical Session, Proceedings, Volume 10, Issue 1,
1984.

Cundhill M., Recommended Relationships for Parts Modeling in HDM-4. Report to
the International Study of Highway Development and management Tools, University
of Birmingham, 1995.

Dieckmnann, D. The effect of mechanical vibration upon man—a review and
summary of research to date. A report of the Max-Planck Institut for
Arbeitsphysiologie, Dortmund, 1955.

Duffner, L. R., L. H. Hamilton, M. A. Schmitz et al. Effect of whole-body vertical
vibration on respiration in human subjects. J. Appl. Physiol. 33: 913-916, 1962.

Emsting, J. Respiratory Effects Of Whole Body Vibration. Air Ministry Flying
Personnel Research Committee Report, 1960.

Federal Highway Administration, Life Cycle Cost Analysis - Summary of
Proceedings - FHWA Life Cycle Cost Symposium, Searching for Solutions: A
Policy Discussion Series, No. 12, Federal Highway Administration, Washington, D.
C., 1994.

Haugodegard, T., Johansen, J., Bertelsen, D., and Gabestad, K. Norwegian Public
Roads Administration: A Complete Pavement Management System in Operation,
Volume 2, Proceedings, Third International Conference on Managing Pavements, San
Antonio, Texas, May 22-26, 1994, Transportation Research Board, Washington, DC.
1994.

Hide, H., et al. The Kenya Road Transport Cost Study: Research on Vehicle
Operating Costs. Transport and Road Research Laboratory Report, Number LR 672,
Great Britain, 1975.

Kher, R., M. A. Karan, and R. Haas. Effects of Pavement Roughness on Vehicle
Speeds. Pavement Design, Evaluation, and Performance 1976.

Lee, N.D. International, Modeling Road User Effects in HDM-4, Prepared for the
Asian Development Bank as part of the International Study of Highway Development
and Management Tools, Final Report, 1995.

306

 

105.

106.

107.

108.

109.

110.

11].

112.

113.

114.

115.

116.

Pierce, C. D., S. P. Singh, and G. Burgess. A Comparison of Leaf-spring with Air-
cushion Trailer Suspensions in the Transport Environment. Packing Technology and
Science 5 (1 1-15): 1992.

Rhodes, N. F. An Investigation of Vehicle Speed and Pavement Roughness
Relationships for Texas Highways. Department of Civil Engineering, Texas A&M
University, College Station, Texas. 1978.

Robinson, R., and P. A. Subramaniam. Road Deterioration Parameters for Estimating
Vehicle Operating Costs in Developing Countries. Overseas Unit, Transport and
Road Research Laboratory (TRRL), Crowthome, Berkshire, England. 1976.

Sawhill, RB. and J .C.F irey, Motor Transport Fuel Consumption rates and Travel
Time, Highway Research Board, National Academy of Sciences, Washington DC,
Bulletin 276, 1960.

Standard Test Method for Measuring Surface Macrotextrure Depth Using Volumetric
Technique, ASTM Standard E 965-87, reapproved in 1995, Volume 04.03, 1996.

Bureau of Transport and Communications Economics (BTCE) of Australia, “Roads,
Vehicle Performance and Greenhouse: Costs and Emission Beneﬁts of Smoother
Highways.” 1997. http://www.dot.gov.au/programs/btce/btcehomehtm. Information
Services, Bureau of Transport and Communications Economics, GPO Box 501,
Canberra, ACT 2601 , Australia.

Census of Transportation, “Vehicle Inventory and Use Survey”, Bureau of Census,
Department of Commerce, http://www.census.gov/svsd/www/vius/2002.htrnl, 2002.

HEM, 1986, Imp://www.hemdata.com/Products/Dawn/index.htm

Stone, R. (1999). Introduction to Internal Combustion Engines, SAE International,
Warrendale, PA.

VTRIS. (2005). "Vehicle Travel Information System."
http://ﬂiwyapm.ﬂrwa.dot.gov/vtris/.

 

Sandberg, Ulf S. 1. Road Macro- and Megatexture Inﬂuence on Fuel Consumption.
ASTM STP 1031 page 460-479, USA 1990

The Tyre Encyclopedia. Vol. 3. Michelin, France 2003.

307