INFORMATION TO USERS

This manuscript has been reproduced from the microfilm master. UMI fiims
the text directly from the original or copy submitted. Thus, som e thesis and
dissertation copies are in typewriter face, while others may be from any type of
com puter printer.

The quality of th is reproduction is d e p e n d e n t upon th e quality of th e
copy su b m itted . Broken or indistinct print, colored or poor quality illustrations
and photographs, print bleedthrough, substandard margins, and improper
alignment can adversely affect reproduction.

In the unlikely event that the author did not send UMI a complete manuscript
and there are missing pages, these will be noted.

Also, if unauthorized

copyright material had to be removed, a note will indicate the deletion.

Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand com er and continuing
from left to right in equal sections with small overlaps.

Photographs included in the original manuscript have been reproduced
xerographically in this copy.

Higher quality 6" x 9” black and white

photographic prints are available for any photographs or illustrations appearing
in this copy for an additional charge. Contact UMI directly to order.

Bell & Howell Information and Learning
300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA
800-521-0600

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

NOTE TO USERS

This reproduction is the best copy available

UMI

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Development of Rutting Progression Models by
Combining Data from Multiple Sources

by

Adrian Ricardo A rchilla

Civil Engineer (National University o f San Juan, Argentina) 1989
M aster o f Science (University o f Calgary, Canada) 1993

A dissertation submitted in partial satisfaction o f the
requirements for the degree of
D octor of Philosophy

in

Engineering -Civil and Environmental Engineering
in the
GRADUATE DIVISION
of the
UNIVERSITY O F CALIFORNIA, BERKELEY

Committee in charge:
Professor Samer M adanat, Chair
Professor John H arvey
Professor Juan Pestana-Nascim ento
Professor Paul Ruud

Spring 2000

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

U M I N um ber: 9 9 7 9 5 4 8

UMI
U M I M ic ro fo rm 9 9 7 9 5 4 8
Copyright 2 0 0 0 by Bell & H ow ell Inform ation and Learning C om pany.
All rights reserved . T his m icroform edition is protected against
unauthorized copying u nder Title 17, United S ta te s Code.

Bell & H ow ell Inform ation and Learning C o m p a n y
3 0 0 North Z e e b Road
P .O . B ox 1346
Ann Arbor, M l 4 8 1 0 6 -1 3 4 6

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Development o f Rutting Progression M odels by
Combining Data from Multiple Sources

©2000

by

Adrian Ricardo Archilla

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

To my lovely wife, Marcela,
to my daughters, Maria Fernanda and M aria Florencia,
and to my parents, Irene and Jose.

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

. who made the AASHO Road Test data set available to me. Professor Samer M. for his excellent advice and friendship. Finally. Further reproduction prohibited without permission. particularly during coffee breaks. They have contributed to this dissertation by enhancing my understanding o f asphalt concrete pavements. I particularly w ant to thank my dissertation advisor. Marcela. M any thanks also to Professor John Harvey for his time and advice. This work has been enhanced by his constructive feedback and his involvement in the WesTrack project. I would also like to thank my parents for always being there with their love and support and my wife and daughters for following me everywhere. My gratitude is also extended to professor Marshall Thom pson. During these three and a half years I have been lucky to make some very good friends: Jorge Prozzi. and other related areas. His enthusiasm in my work and interest in my career are also appreciated. I had the privilege o f benefiting from his insight and suggestions. thanks for w ithstanding a grumpy dad during the m ost difficult semesters. thank you for offering me help before I asked you and for being a great ping-pong partner. econometrics. R einaldo Garcia. and to professor Juan Pestana-Nascimento for his detailed review o f this dissertation. thanks for the love and care you put into everything that you do for me. Madanat. To Fem i and Flori.Acknowledgm ents This dissertation has benefited from the contributions o f my dissertation and candidacy exam com m ittee members. They have made my years at Berkeley very enjoyable. I owe to Samer my involvement in the infrastructure m anagement area. To Jorge. R eproduced with permission of the copyright owner. and Da-Jie Lin.

.....................................................................44 4..........2 Factors that affect ru ttin g .......2 W esTrack Road T e s t...........................3 Sum m ary....................5 Evidence from the empirical literature............................ 6 1........................................... 9 C h a p te r 2: R u ttin g o f A sp h a lt C o n crete P a v e m e n ts ............4 Outline o f the dissertation ...................................................................................................... 64 R eproduced with permission of the copyright owner.......................................................... ................................................................................1 AASHO Road T est....................................... 60 4... 34 3................5 1...................................................................55 4..........4 Evidence from the M echanistic-Empirical L iterature...........2 Specification o f a model for rutting originating in the asphalt concrete lay er............................................. 59 4......................................................T able o f C ontents C h a p te r 1: In tro d u ctio n a n d S u m m a ry ............................ 32 C h a p te r 3: D ata S o u rc e s .........................................27 2................................................................................ 44 4.......................................................2 Research objectives................................... 12 2......................................................................................16 2.........................................................................................................................34 3................................... Further reproduction prohibited without permission..................................................3 Contributions....................................................1 Specification o f a model for rutting originating in the underlying layers................................................................. I 1................................42 C h a p te r 4: R u ttin g M odel fo r the A A SH O R o ad T est D ata S e t .......3 Complete M odel Specification.................................................................. 1 1...4 Incorporation o f overlays in the m odel specification.........1 Causes o f ru ttin g ....................................................5 M odel estimation resu lts................ 30 2.............................................................. 40 3.............................3 M odeling A pproaches................................................................................................1 Introduction...................................................... 12 2..............................................................................

.......................6 S u m m a r y ................................................ 115 6............................ 154 A p p en d ix B: E stim atio n R esu lts..................................................................................................................93 5.....2 Com patibility o f the AASHO and W esTrack data sources..........................................................................1 Contribution and C onclusions.....................................................................A A S H O ........5 S um m ary................... 140 A ppen d ix A: E stim atio n A p p ro a c h e s ..........2 F urther research...3 M odel specification..........3 Random -effects ap p ro ac h ...............................................................................103 5............................................................... 131 C h a p te r 7: C o n c lu sio n s........... P red icted R u t D ep th s ............... Further reproduction prohibited without permission.......................................................................................................... 132 7............................................. 119 6....................................................173 A p p en d ix D: O b serv ed vs.....................................................................................................................................................................................2 Fixed effects ap p ro ach .....1 Joint estim ation.......................................................................................135 R efere n ces............................................... 1 Ordinary Least S quares....................................................1 M ixture Performance at W esT rack...............................................80 5.................................207 R eproduced with permission of the copyright owner.................. ............................................150 A...........................................................87 5........................112 6.............2 G radation Index..................................................................................................... 80 5.......... 108 C h a p te r 6: J o in t E stim ation o f th e A A SH O a n d W esT rack M o d els............................ 149 A.................. 151 A.......77 C h a p te r 5: R u ttin g M odel fo r th e W e sT ra c k R oad T est D ata S e t............................................. 159 A p p en d ix C : O b se rv ed vs..............................................................W esT ra ck ..............4.....................4 S u m m ary.......................................................................................................4 M odel estimation resu lts................................................................................................... P red icted R u t D ep th s ..112 6......3 M odel specification and estim ation resu lts................................. 132 7...................

.4: Estimate o f the correlation m atrix for the param eter estimates obtained for the AASHO model using the random-effects approach without including the overlay inform ation............................... Further reproduction prohibited without permission......................................... 105 Table 6............................. 162 R eproduced with permission of the copyright owner....................................................122 Table B ........... 69 Table 5................1: Assignment o f axle loads to the various lanes o f the AASHO Road Test .............................. 161 T able B.................159 Table B..........................................................................36 Table 3...94 Table 5..............................................1: Parameter estim ates o f the AASHO model obtained using the random -effects approach............................................................................................. 160 Table B................................l: Parameter estimates obtained for the AASHO model using the fixed-effects approach without including the overlay inform ation......................2: Pavement characteristics included in the AASHO Road Test and W esTrack data s e ts .................................................. ................2: Parameter estim ation results for the WesTrack m o d e l.......................................43 Table 4.....vii List o f Tables Table 3....................3: Parameter estimates obtained for the AASHO m odel using the random -effects approach w ithout including the overlay inform ation........................2: Estimate o f the correlation matrix for the param eter estimates obtained for the AASHO model using the fixed-effects approach without including the overlay inform ation..................................1: Joint estimation re s u lts .........................................................................................1: Characteristics o f the original W esTrack m ixes....................................................

..........................viii Table B......................7: Parameter estimates obtained for the AASHO model using the random -effects approach including the overlay inform ation.........................................................ll: Parameter estim ates obtained for the W esTrack model using the random -effects approach...............10: Estimate o f the correlation m atrix for the parameter estimates obtained for the WesTrack model using the fixed-effects ap p ro ac h .............164 Table B.. 171 R eproduced with permission of the copyright owner..............169 Table B..... 167 Table B............................................................. 170 Table B ............... Further reproduction prohibited without permission...................................5: Parameter estimates obtained for the AASHO model using the fixed-effects approach including the overlay inform ation......................................................................8: Estimate o f the correlation m atrix for the parameter estimates obtained for the AASHO model using the random-effects approach including the overlay inform ation...........l3: Estimate o f the correlation m atrix for the param eter estimates obtained using joint estim ation from the AASH O and W esTrack data sources w ith the random-effects approach...................... 163 Table B........................165 Table B..................................................................................12: Estim ate o f the correlation m atrix for the param eter estimates obtained for the W esTrack model using the random -effects ap p ro ach ........................................................................6: Estimate o f the correlation m atrix for the parameter estimates obtained for the AASHO model using the fixed-effects approach including the overlay inform ation...168 Table B ................................................................................... ..................................9: Param eter estimates obtained for the W esTrack model using the fixed-effects approach.......166 Table B.......

......4: Observed and predicted rut depth vs.. ......................75 Figure 4....................................................... 42 Figure 4.....1: Anticipated relation between the a'.....2: Thawing index com puted at the AASHO Road T e s t....................3: Effect o f wheel load and tire inflation pressure on the vertical stress distribution assuming a semi-infinite elastic m aterial.........5: Comparison o f observed and predicted valu es.............................................................81 Reproduced with permission of the copyright owner........................ 50 Figure 4....................... and coarse W esTrack m ixes.................................................................7: Predicted rut depth vs..................................39 Figure 3...... time for tw o sections used to estim ate the model p aram eters.. 72 Figure 4...........................................6: Observed and predicted rut depth vs..............13 Figure 3...1: Target gradation for fine.................... time using the param eter estimates obtained with and without the information for overlaid sectio n s........... rut depth m easurem ents with AASHO device ( R D i) .......................................1: Rut depth measurements with a straightedge (R D i) vs..................... Further reproduction prohibited without permission............. 53 Figure 4................ix List of F igures Figure 2........1: Pavement surface deform ations induced by the traffic lo a d s................ 76 Figure 5................. 74 Figure 4.........................3: Truck configuration used in W esTrack...........................58 Figure 4............................... coefficient and the strength o f a p av e m e n t....... time for tw o sections not used to estim ate the model param eters......................................... 39 Figure 3.................................................... fine-plus...2: Negative rut depth m easurem ents with the A ASHO device.....

......... 106 Figure 5.... 84 Figure 5............ 83 Figure 5............3: Observed rutting vs.................................................92 Figure 5............................................................................6: Illustration o f the new m axim um aggregate size definition for plotting m axim um density lines for a hypothetical gradation curve 3 ............ asphalt content at WesTrack after 130...............................1: Predicted total rut depth and rut depth originating in the asphalt concrete layer vs................... voids filled with asphalt after 130.................................... 125 R eproduced with permission of the copyright owner........300 vehicle passages..............................9 )................ 90 Figure 5........................................................10: Observed and predicted rut depth vs................5: Problems with the Superpave definition o f M aximum Aggregate S iz e .2: Observed rutting vs........ time with the param eter estimates from the joint estim ation for two AASHO sectio n s......................................6 C ............................................... time for eight W esTrack sectio n s............ 86 Figure 5........................................................ in-place air voids at W esTrack after 130................................. Further reproduction prohibited without permission.......... 109 Figure 6................91 Figure 5.............................................................................................................8: Functional form o f f(-) in equation (5 ...................9: First two term s o f m„ for different values o f G l as a function o f VFA when the mean m axim um temperature is above 28..........300 vehicle passages.................300 vehicle passages..........7: Illustration o f the new m axim um aggregate size definition for plotting m axim um density lines for a hypothetical gradation curve 4 .........................................................4: Observed rutting vs................. 101 Figure 5.............................................................................. ................X Figure 5.......

.... and 1 6 2 ... 146... 120....... 111............. 165.......... 159............. 175 Figure C............5: Observed and predicted rut depth values for AASHO sections 145.4: Observed and predicted rut depth values for AASHO sections 137....... 262..l: Observed and predicted rut depth values for AASHO sections 107............. 270............. 129... 158.. 179 Figure C........... 180 Figure C.. and 1 3 6 ........2: Predicted total rut depth and rut depth originating in the asphalt concrete layer vs..... 139...... 130....... 268.... and 1 5 4 ............................ 254.... and 1 1 4 ...... 128.. 258. 177 Figure C....... 259...... 263. 255........... 110.123.127 Figure C ........ 109......... 141....113..... and 298 ............ 112................ 147...7: O bserved and predicted rut depth values for AASHO sections 163.... 108........ ... 176 Figure C.................................... 166......... 267...................8: O bserved and predicted rut depth values for AASHO sections 257....................... 253...... 261. Further reproduction prohibited without permission. time with the param eter estimates from the jo in t estim ation for two AASHO sections trafficked w ith -27 kN lo ad s.... 156........ 260. 174 Figure C.... and 1 4 4 ... and 264 ... 142......135...... 122................................... 266.......9: O bserved and predicted rut depth values for AASHO sections 265............... 153..... 148........... 119... 138...6: Observed and predicted rut depth values for AASHO sections 155....... 161.. 269. 152....181 Figure C....... 297........................ and 1 2 4 ................. 157..... 143... 121........ and 256 .2: Observed and predicted rut depth values for AASHO sections 117....... 131... 178 Figure C.............................................................................3: Observed and predicted rut depth values for AASHO sections 127..... 118.................. 140. 182 Reproduced with permission of the copyright owner............. 132........ 151......160.... 164.....xi Figure 6...................

.l4: Observed and predicted rut depth values for AASHO sections 332...................... 306.............. 469. 4 3 7 . 4 2 3 .......... 420...................................4 1 8 ........187 Figure C .. 426................. 186 Figure C .........4 4 5 .....4 1 9 ................ 183 Figure C ...........188 Figure C .............. 442.....4 1 7 ..l5: Observed and predicted rut depth values for AASHO sections 414................4 3 9 ............................ 311....................... and 4 2 9 ...............4 4 7 .4 3 8 ...................... 327. 319....... 330. 314......l8: Observed and predicted rut depth values for AASHO sections 4 4 4 ... ........... 322........... 190 Figure C .l6: Observed and predicted rut depth values for AASHO sections 422..................l9: Observed and predicted rut depth values for AASHO sections 4 5 2 .. 441...l3: Observed and predicted rut depth values for AASHO sections 324.............. 328..... 448. and 4 7 1 .. 456........191 Figure C .......... and 3 1 5 ....... 333...... 320.... 313..................4 5 5 ..... 309..... and 307 ..... and 3 3 1 ....... 427. 4 1 5 ....................l7: Observed and predicted rut depth values for AASHO sections 430... 300......................... 41 1........ 325.........................4 2 4 ..........192 R eproduced with permission of the copyright owner.......4 4 6 ..l 1: Observed and predicted rut depth values for AASHO sections 308...4 5 4 ............ 302............. 428.................... 440.................. 303..4 1 6 . 449........... and 323 ..........xii Figure C................... and 443 . 312................ 326... 304..... 321..... 184 Figure C....... and 4 1 3 ... 334.......... 336................ and 4 5 1 ............................ 470..... 450........ 318................ Further reproduction prohibited without permission.... 310............4 5 3 ...4 2 5 ............10: Observed and predicted rut depth values for AASHO sections 299.. 301.. 189 Figure C ...... 185 Figure C .. 317...........12: Observed and predicted rut depth values for AASHO sections 316.. and 4 2 1 ........ 329. 412. 335...

... 616...................629.............. 627.. 603.. 597.....................23: Observed and predicted rut depth values for AASHO sections 577.................. 620 .... 587...... 200 Figure C............................... and 7 2 7 ......................................... 626........ 625..24: Observed and predicted rut depth values for AASHO sections 585................... and 584 .. 628...... 579.... 632............................ ................195 Figure C............. 598.. 602........4 7 7 ............. 595.. 202 R eproduced with permission of the copyright owner.........4 7 6 ....................... 594... 717....... 6 1 7 . 473.... 582.. 718.............. 196 Figure C.... 481.. 4 7 5 ........... and 600 ..... 590... 198 Figure C...... and 6 2 2 ...... 605...... 193 Figure C................... 201 Figure C.... 591............. 589..................4 8 4 ....25: Observed and predicted rut depth values for AASHO sections 593........ and 608 .197 Figure C..... 575.... 581.. 720. 588..27: Observed and predicted rut depth values for AASHO sections 615.... 624................ 487.29: Observed and predicted rut depth values for AASHO sections 631....28: Observed and predicted rut depth values for AASHO sections 623...................... 4 8 2 ................ 199 Figure C...........6 2 1 ..21: Observed and predicted rut depth values for AASHO sections 480........... 572. 474.4 7 8 .... 583..... 607..... and 576 ......4 8 3 ................................22: Observed and predicted rut depth values for AASHO sections 569......... and 592 . and 4 7 9 ...26: Observed and predicted rut depth values for AASHO sections 601..... and 6 3 0 .......... 580... 570..................................... 578...............4 8 5 ...................194 Figure C.....20: Observed and predicted rut depth values for AASHO sections 472............. 599................. 596. Further reproduction prohibited without permission.... 719....6 1 9 ........... 574......... and 488 . 573...xiii Figure C.... 606... 571................ 586....................... 721......6 1 8 . 604.......

....................................................................... 770..............................742 ...................... and 776 ................. 741................... 760..................................4: O bserved and predicted rut depth values for WesTrack sections 25 and 2 6 .............. 774........................................................................................ 209 Figure D............. 764.. 737................................... 771......... and 756 .................xiv Figure C. 208 Figure D...........3 1: O bserved and predicted rut depth values for AASHO sections 743...........3: O bserved and predicted rut depth values for W esTrack sections 17 to 2 4 .....210 Figure D........ 204 Figure C.................... 203 Figure C ................ 739..... 206 Figure D ..................................................................................... 769... 205 Figure C....... 755..... 775................................... 750..... 740. ............l: O bserved and predicted rut depth values for W esTrack sections 1 to 8 .................... 730....................... 745...........30: O bserved and predicted rut depth values for AASHO sections 729................ 749............................................................................ 763.................. 211 Reproduced with permission of the copyright owner.................................. 744..................................2: O bserved and predicted rut depth values for WesTrack sections 9 to 16.................... 746.... and 772 ................. 738. and...........32: O bserved and predicted rut depth values for AASHO sections 759........... Further reproduction prohibited without permission.......33: O bserved and predicted rut depth values for AASHO sections 773................................

is a serious safety issue for road users because there is a potential for 1 The explanation o f how rutting is measured is deferred until the sources o f data for this research are described in Chapter 3. with concom itant effects on user benefits. and their interactions with the pavem ent’s microclimate. R utting1. and large (“bathtub”) depressions. small (“birdbath”) depressions. dynamic loads.C hapter 1: Introduction and Sum m ary 1. Rutting has historically been a primary criterion o f structural performance in many pavement design methods. riding quality and safety (Paterson 1987). the main types o f structural distress are cracking and permanent deformation. O ther types o f perm anent deform ation are generally much less tractable for direct m odeling since they depend to a larger degree on construction quality. O f these. loosely defined as longitudinal depressions along the wheel paths in asphalt concrete pavements. Perm anent deformations in pavements include rutting.1 Introduction The deterioration o f asphalt concrete pavem ents manifests itself through several distress types. These constitute distress because they directly increase road roughness. These deform ations have been typically controlled through construction and material specifications and their evolution is thus m odeled indirectly through roughness progression (Paterson. 1987). material properties. heave. raveling. shoving (material instability). This dissertation is concerned with the developm ent o f rutting progression m odels using m ultiple experimental data sources. Reproduced with permission of the copyright owner. . and roughness. Further reproduction prohibited without permission. potholing. their local variations. such as cracking. edge depressions. perm anent deformations.

as rutting increases steering becom es more difficult. 1975). With increasing magnitudes and repetitions o f loads and increased tire pressures. al„ 1994). The accurate prediction o f rutting developm ent is an essential element for the efficient m anagem ent o f pavem ent systems. can be used to study the effect of different loading levels. Even in dry conditions. it is very im portant to be able to specify and estim ate accurate deterioration models.4 in. but w ith limited success. The occurrence o f hydroplaning depends on the driving speed and the depth o f the pool o f water. and in particular rutting progression models. Since m aintenance planning in turn depends upon the ability to accurately predict future pavem ent condition. the rutting problem has become severe in many highway pavements (Haas et. . Further reproduction prohibited without permission. As an exam ple. The hydroplaning phenomenon consists o f the buildup o f a thin layer o f w ater between the pavem ent and tire and results in the tire losing contact with the surface. properly specified pavement deterioration models. In addition to their im portance in M & R decision-making. and thus in allocating cost responsibilities to various vehicle classes for their use o f the highway system. causing a loss o f steering control (Yoder and Witczak.hydroplaning when water accumulates in the ruts. Reproduced with permission of the copyright owner. M aintenance planning for highway agencies typically involves the allocation o f limited resources for maintenance and rehabilitation (M &R) to a num ber o f pavements that make up the highway network. hydroplaning can occur at speeds faster than 80 km/h (50 mph) with pools o f water deeper than 10 mm (0.). Considerable research has been conducted over the years towards developm ent o f models to predict the progression o f rutting.

The latter causes an econom etric problem known as endogeneity. the control o f constructed layer thicknesses is o f lower quality. In field data. they can be used to evaluate different strategies for design. Furthermore. As stated before. This is difficult using field data alone. in field data. Further reproduction prohibited without permission. asphalt and air voids in asphalt concrete mixes to m inim ize rutting potential.3 Furtherm ore. In full-scale experimental road tests. Discrimination of the effect o f each load level from a distribution o f loads is a difficult problem . Finally. and the design thicknesses are usually a function o f traffic. maintenance and rehabilitation. Estim ation o f parameters by Ordinary Least Squares (OLS) regression results in biased and inconsistent results in the R eproduced with permission of the copyright owner. Field data generally involve a distribution o f loads whose measurement is not very accurate. usually the pavem ent characteristics and the loading for the different sections are the result o f an experimental design. which is not present w hen each pavem ent is subjected to a known load level. In particular. has the following characteristics: Advantages: The main factors affecting rutting such as axle loads and layer thicknesses are carefully controlled. layer thicknesses are endogenous since they are usually a function o f predicted traffic. the focus o f this dissertation is the developm ent o f rutting progression models from multiple full-scale experimental data sources. Endogenous variables are determ ined through the jo in t interaction w ith other variables w ithin the model. therefore the researcher can capture their effects on rutting progression. such models can be used in the design o f new pavem ent structures. as opposed to field data collected from condition surveys o f in-service pavements. . The use o f experim ental data from full-scale road experim ents for model development. they can also provide inform ation for proportioning o f aggregate.

This problem is avoided or reduced in experim ental data. The second reason for focusing on experimental data is that the developm ent o f an adequate data set from in-service pavements to develop rutting models w ould have required time and resources beyond the scope o f this research. Further reproduction prohibited without permission. The developm ent o f a good experimental model serves as the basis for the specification o f a model for in-service pavements. al. it is im portant to clearly identify the individual effects o f the different factors that affect rutting. . This database is part o f the LTPP program that was designed as a 20year study and is managed by the Federal Highway A dm inistration (Ostrom et. it will be shown in Chapters 4 and 6. T he work in this study is restricted to experimental data sources for two reasons. traffic w ander (the lateral distribution o f the loads over the lane width) may not be adequately replicated in accelerated pavem ent loading tests. as part o f this research a preliminary analysis was made o f the Long-Term Pavement Perform ance (LTPP) database. For exam ple. 1997). whose param eters (perhaps after some sim plifications) can be obtained using the in-service data alone or in com bination with experim ental data. First. As m entioned before.4 presence o f endogenous variables.. For exam ple. R eproduced with permission of the copyright owner. D isadvantage: The main disadvantage is that experimental data may not represent the true deterioration mechanism o f in-service pavements. For example. this is hardly possible using field data alone. that the commonly used fourth power law for load equivalencies can be extremely misleading for rutting originating in the asphalt concrete layer.

which again is not a trivial task. one is left with only a few pavement sections w hen all the possible factors affecting rutting (such as load distributions. at present. monitoring. research project.2 Research objectives The objectives o f this research are as follows: • to develop accurate models o f rutting on asphalt concrete pavem ents. maintenance. Because o f the above. a search over more than 190 tables is necessary to determine the fields o f interest. Further reproduction prohibited without permission. the inform ation for the different pavement sections needs be put together. extensive data such as inventory. and • to explore the feasibility o f developing pavem ent rutting m odels by efficiently com bining different data sources. The am ount o f information in the LTPP database is overwhelming. traffic. This may not be a problem in the near future since the LTPP database is being continuously updated. etc. rehabilitation. and seasonal testing have been collected from m ore than 2000 sites across the U. 1. material testing. environment. it also makes the data extraction quite laborious.) are considered.5 As part o f the program. Finally. Though this is beneficial for research. In addition to the above. the development o f rutting progression models from in-service pavem ents is left for a future.S and Canada since 1989. climatic. larger scale. several simplifications are necessary to make the data usable for estimation. Then. asphalt mix characteristics. . R eproduced with permission of the copyright owner. In the first place.

6 1. asphalt concrete mix characteristics. For example. is adequate if the overlay thickness is enough to reduce any additional rutting originating in the underlying layers and the additional rutting originating in the overlay is relatively small. thawing and high air temperatures in a single equation. consider the case where it is known that rutting originates in the underlying layers. The model would indicate whether the reduction in thickness o f the asphalt concrete layer could produce significant R eproduced with permission of the copyright owner. single and tandem axle loads. consider a pavement in an urban street where it is known m ost o f the rutting originates in the asphalt concrete layer o f the pavem ent and where the designer is constrained by the m axim um pavement surface elevation. and the allocation of cost responsibilities to different vehicles classes are: • The model considers two different mechanisms o f rutting: rutting in the underlying layers and in the asphalt concrete layer. which reduces the exiting rutting close to zero. The m ost important features o f this model and their im plications for both pavem ent design and rehabilitation. The identification o f the origin o f rutting can be used to plan appropriate rehabilitation strategies. The model incorporates the effects o f layer thicknesses.3 Contributions The main contribution o f this dissertation to pavement engineering is the specification and estim ation o f a new empirical model for predicting rutting in asphalt concrete pavements. For another exam ple. Further reproduction prohibited without permission. In this situation. The application o f the model may suggest that an overlay strategy. asphalt concrete mix design. milling may be a viable option. .

. For exam ple. • The model captures the effect o f overlays. The estimated load equivalence coefficient for rutting originating in the asphalt concrete layer. ~ 4). and it accounts for the hardening o f the underlying materials with traffic before the overlay construction. it can help the designer to select between alternatives with high initial construction costs but free o f rehabilitation or with low er initial construction costs but requiring several rehabilitation jobs over the design life o f the pavement. This result has im portant implications for the allocation o f cost responsibilities to different vehicles classes. m ust be included. including user costs. it can be used to determine an econom ically optimal strategy for original construction and rehabilitation with rutting as a measure o f performance. is significantly different from the value usually assum ed in pavement engineering for the calculation o f standard axle loads (i. For example. For such analysis. Thus. the equivalence coefficients for axle loads are not restricted to be equal for rutting originating in the asphalt concrete layer and for rutting originating in the underlying layers. all the relevant costs. • During estimation o f the model. In particular.57. Further reproduction prohibited without permission. no restrictions are imposed on the values o f the param eters representing equivalencies between axle loads or the parameters representing pavem ent layer resistance..7 additional rutting originating in the underlying layers and how much additional rutting is expected to originate in the asphalt concrete layer. 0. it shows that in pavem ent sections where most o f the rutting originates in the asphalt concrete layer the use o f a fourth power R eproduced with permission of the copyright owner.e.

• The model incorporates the effects o f extrem e environments.8 equivalence coefficient assigns an excessive am ount o f the damage to the heavier loads and an insufficient amount to the lighter loads. Thus. Reproduced with permission of the copyright owner. The three mix properties are the gradation index. • Three properties are sufficient to m odel the rutting performance o f the asphalt concrete m ixes accurately. In particular. which is obtained from the aggregate gradation curve. the m odel can be used to guide the selection o f appropriate mix characteristics for adequate rutting performance. . In particular. • The m odel predicts rut depths by adding predicted values o f the increment o f rut depth for each time period. the voids filled with asphalt obtained for the construction mix in the Superpave gyratory compactor. the model estim ation results indicate that there is a value for air voids that optimizes rutting perform ance. and the initial in-place air voids. it accounts for the effect of high air tem peratures in asphalt concrete m ixes and the effect o f freeze-thaw cycles in unbound materials (when a source o f w ater is present). This is particularly advantageous in a pavement m anagem ent context where the current rut depth value is known and the interest is in predicting the rut depth value for the next time period (and hence only the change in rut depth). Further reproduction prohibited without permission. a result that is consistent with traditional asphalt concrete mix design.

• Chapter 3 briefly describes the two experimental data sources used for model estimation: the A ASHO and W esTrack Road Tests. this research also illustrates. 1.9 A more fundamental contribution o f this dissertation is that it dem onstrates how the application o f structured econometric techniques coupled w ith basic engineering knowledge can be used to estim ate accurate pavement deterioration models. Further reproduction prohibited without permission. In particular.4 O utline of the dissertation The rem aining o f this dissertation is organized as follows: • Chapter 2 presents a literature review and gives an introduction about the problem o f rutting in asphalt concrete pavem ents. Reproduced with permission of the copyright owner. It describes the sources o f pavement rutting. Two features o f jo in t estimation are exploited in this research: the identification o f the param eters that are not identifiable from one data source alone. This chapter also highlights the complementary characteristics o f these two data sources. it illustrates the use o f the fixed-effects and random-effects approaches to account for unobserved pavement heterogeneity. some o f the factors that affect rutting. and the gain in statistical efficiency because o f the use o f all available information. and the different approaches that have been used to date. the advantages o f estimating a model jointly from multiple data sources. w ithin the context o f pavement deterioration m odels. . Finally.

Further reproduction prohibited without permission. The model presented in that chapter accounts for rutting originating in the asphalt concrete layer as well as for rutting originating in the underlying layers and accounts for the effect of asphalt concrete overlays. . R eproduced with permission of the copyright owner. which plays a fundamental role in the model specification for WesTrack. After establishing their compatibility.10 • Chapter 4 presents the specification and estimation o f a pavement rutting progression model using the AASHO Road Test data set. The concepts behind the fixed-effects and randomeffects estimation approaches are first introduced in this chapter. • C hapter 7 concludes the dissertation and suggests future research directions to com plem ent the work presented here. The model in this chapter accounts for rutting originating only in the asphalt concrete layer. • C hapter 6 presents the concept and purpose o f joint estimation. It also describes how the models o f the previous two chapters can be estimated jointly and analyzes whether the two data sources are compatible for the application o f this estim ation approach. This chapter also introduces the concept o f a gradation index. the chapter presents the jo in t estimation results. • C hapter 5 presents the specification and estimation o f a pavement rutting progression model using the W esTrack Road Test data set.

• Finally. .• A ppendix A describes in some detail the equations used for the param eter and covariance matrix estimation o f non-linear models under three commonly used frameworks: ordinary least squares. R eproduced with permission of the copyright owner. fixed-effects and random-effects. respectively. predicted rut depth values for all sections at the A A SH O Road Test and WesTrack road test. • A ppendix B presents the tables with the parameter estim ation and covariance matrix estimation results obtained w ith the fixed-effects and random-effects approaches. Appendixes C and D show graphically the observed vs. Further reproduction prohibited without permission.

Section 2. The description in that section o f where pavem ent rutting originates has shape some aspects o f the specification o f the models presented in later chapters.4 and 2. or when bituminous courses are subjected to high temperatures (OECD 1988). The description o f these factors has two purposes. especially when the pavem ent subgrade undergoes seasonal variations in bearing capacity.12 C hapter 2: Rutting o f A sphalt C oncrete Pavem ents This chapter gives an overview o f the problem o f rutting in asphalt concrete pavements.1 describes the causes o f pavem ent rutting. 2. Section 2. Sections 2. On the one hand.5 briefly review the evidence from the mechanistic-empirical and empirical literature respectively. rutting can result from a single or a com paratively few excessive loads or tire pressures causing stresses that approach or exceed the shear R eproduced with permission of the copyright owner. The ruts develop w ithin pavem ent layers when the traffic loading causes layer densification and/or when the stresses induced in the pavem ent materials are sufficient to cause shear displacements within the m aterials. Section 2. On the other hand. .2 describes som e o f the factors identified in the literature as affecting rutting. the factors not included point out limitations o f the models. many o f these factors will be included in the model specifications presented in this dissertation.1 C auses o f rutting Although rutting is strongly influenced by traffic loading. As pointed out by M onism ith (1976) and later by Paterson (1987). clim ate can also have a large influence. Further reproduction prohibited without permission.3 describes the different m odeling approaches that have been used to model pavement rutting to date and explain why an empirical approach has been selected for this research.

The densities required are typically lowest in subgrade 1 A more precise term for the mechanism is permanent plastic and/or viscous flow. Repeated loadings at lesser load and tire pressure level cause sm aller deformations.13 strength o f the m aterials. or time-dependent deform ation in viscous m aterials (e. The pavem ent surface deformations are illustrated schematically in Figure 2. in some cases. Typically. resulting from tighter packing o f the material particles and. static or long-term loadings can cause creep. bituminous materials) and in soils. there are two m echanism s of traffic-associated deform ation that are im portant to the modeling efforts: densification and plastic flow 1.1: Pavement surface deformations induced by the traffic loads. w hich have been selected in accordance with the expected loadings and pavement type. resulting in depressions under the load and often heave alongside the loaded area. T ire Heave Pavement depression Pavement surface 7777 7777777 Figure 2. R eproduced with permission of the copyright owner.1.g. . w hich accumulate over time and becom e a significant rut if the loadings are channelized into wheel paths. Finally. from the degradation o f the particles into sm aller sizes.. As explained by Patterson (1987). densification is controlled through com paction specifications at the tim e o f construction. Further reproduction prohibited without permission. D ensification involves volum e change in the material.

where the stresses induced under traffic loadings are least. most design methods require higher strength materials in the upper layers o f the pavement. CBR. as pointed out by Paterson (1987). al. Further reproduction prohibited without permission... They usually relate one type o f structural response such as com pressive strain or shear strain to the rut depth. which result in both depression and heave as material m oves to the edges o f the loaded areas. the vertical compressive strain at the subgrade surface is usually lim ited to some tolerable amount associated with a specific number o f load repetitions. It occurs when the induced shear stresses exceed the shear strength o f the material or are sufficient to induce creep. Plastic flow is the second mechanism. This is reflected in the many m echanistic rutting models developed to date. R utting is the result o f a complex com bination o f densification and flow.14 layers. and M arshall or Hveem stability for bitum inous materials). This mechanism is controlled in pavem ent design by the selection o f materials according to a surrogate m easure o f shear strength (e. As a result.g. Other researchers have related the rut depth in the asphalt concrete layer to the permanent shear strain at a R eproduced with permission of the copyright owner. . T hese models are useful mostly for thin pavem ent structures where the deform ation of the subgrade is a m ajor contributor to rutting. these design models are not useful for performance modeling because o f the need to predict not the limit. the California Bearing Ratio. More elaborate models that predict the trend o f rutting over time consider the vertical com pressive strain in the middle o f each layer (e. but the trend o f rutting during the life o f the pavem ent. 1998).g. In pavem ent design.. Flow involves essentially no volume change and consists o f shear displacements. and highest in the upper layers o f the base and surfacing where the induced stresses are highest. However. for soils. Ali et.

Further reproduction prohibited without permission. it is im portant to stress that the different responses observed w ith variations in the above factors may allow the approxim ate identification of how m uch rutting is originating from the different sources without an explicit modeling o f w hether the rutting is due to densification or plastic flow. For exam ple. high temperature is m ost relevant for rutting originating in the asphalt concrete layer. All these factors are analyzed in m ore detail in the follow ing chapters. for pavem ents with thick asphalt concrete layers. which is usually attributed prim arily to densification o f the subgrade material. al. in em pirical models such as the ones developed in this dissertation. For now. and loading have different effects depending on where the rutting originates. Thus. subgrade rutting. In contrast. plays the most im portant role for thin pavem ent structures. high temperatures.. it is difficult to separate them. shear displacem ents in the asphalt concrete layer are the most im portant element controlling rutting. Fortunately. Finally. The major modeling difficulties arise for the m ost com m on intermediate pavem ent structures. the effect o f the load magnitudes and tire inflation pressures are also expected to be different depending o f where rutting is originating.15 depth o f 50 mm below the pavem ent surface (UCB et. These models are applicable mostly to pavements with thick asphalt concrete layers. In contrast. factors such as freeze-thaw. where both m echanism s are present in varying degrees. As illustrated in the examples above. 1994). . for rutting originating in the subgrade. R eproduced with permission of the copyright owner. freeze-thaw may play a fundamental role in the rut depth developm ent if the subgrade material is frost susceptible.

m any issues are sim ilar to those presented for aggregates in asphalt concrete mixes and the information that pavem ent managers have about these m aterials is more limited. water and air for granular materials and soils. Inter-particle friction is essential for rutting resistance since it provides the necessary shear resistance. The following paragraphs first describe the factors for asphalt concrete mixes and then very succinctly the factors for granular materials.2 Factors affecting rutting This section briefly describes some o f the most im portant factors affecting the rutting performance o f asphalt concrete pavements. porosity. and polished. and aggregate mineralogy. rough. The factors can be classified into three categories: material characteristics. and loading. but not because they are less important. However. asphalt cement and air for asphalt concrete mixes and mineral particles. Material characteristics: The asphalt concrete mixes and granular materials or soils used in road construction are com posed o f three phases: mineral aggregates. The first two (very rough and rough) are considered desirable for good rutting R eproduced with permission of the copyright owner. . shape. Further reproduction prohibited without permission. The three phases have im portant im plications for the rutting performance o f pavem ents. climate. Aggregate p h a se: The most im portant characteristics o f the aggregates (not their gradation) recognized as affecting the mix resistance to perm anent deform ation are surface texture.16 2. The surface texture o f the aggregates is usually classified as very rough. Granular materials are not discussed at length. smooth. Surface texture and aggregate shape are im portant contributors to inter-particle friction.

for any duration o f applied load. and Brown (1993). The aggregate shapes can be angular. At one end o f the spectrum . Thus. “you can’t get two technicians to agree on fractured face count” . Further reproduction prohibited without permission. In addition to being qualitative in nature. The use o f naturally angular or crushed aggregates is usually recommended. subrounded.17 performance. it is very unlikely that pavem ent managers will have accurate know ledge o f these two characteristics other than some indicators such as percent natural sand or whether the aggregate is crushed or not. As pointed out by Mr. total deformation increases as the percent o f natural sand in the mixture increases. W ade Betenson in a discussion in Hughes and M auping (1987). aggregate gradation is considered a very im portant factor for good inter-particle friction. Along w ith surface texture and shape. described special provisions used in Pennsylvania for heavy-duty pavements. Perdomo and Lytton (1990) show results suggesting that. Cross. their inclusion as explanatory variables in a rutting progression model seems very difficult. and (b) at least 75 percent manufactured sand in the fine aggregate (research in the past has shown that m anufactured sand is generally angular and its incorporation in the mix increases rutting resistance). Two o f these provisions are to use (a) coarse aggregate with at least 85 percent having two or more fractured faces. Kandahl recom m ends at least 4 percent m inus No. 200 fraction in the mixture to stiffen the asphalt binder and R eproduced with permission of the copyright owner. Natural sands are usually rounded and polished. Button. For exam ple. despite their possible importance. The literature contains many references where these two aggregate characteristics are em phasized. Kandhal. . or rounded.

respectively (Hughes and M aupin. Larger maximum aggregate sizes are believed to be better for rutting resistance since they usually give lower voids in mineral aggregate (VMA). At one end o f the spectrum . Further reproduction prohibited without permission.18 maximum permissible percentages o f minus No. These are the Marshall and Hveem stability tests. aggregate gradation is considered a very important factor for good inter-particle friction. Kandahl recommends at least 4 percent m inus No. R eproduced with permission of the copyright owner. . 200 fraction in the mixture to stiffen the asphalt binder and m axim um perm issible percentages o f minus No. Brown and Bassett (1990) test results indicate that mixes with larger aggregate size designed with an air voids content o f 4 percent were generally stronger than m ixes prepared with smaller m aximum aggregates. (1986) presented results indicating that both stability2 and tensile strength decrease as V M A increase. Because o f the 2 The stability value indicates the ability o f the pavement to resist permanent deformation under the action o f loads. 1987). respectively (Hughes and M aupin. 200 of 6 and 5 percent for wearing and binder m ixtures. heavy concentration o f aggregate interlock in large-stone mixes allows for efficient dissipation o f com pressive and shear stresses that are otherwise known to be responsible for rutting in flexible pavem ents. Brown at. The mixes with larger aggregates also require significantly less asphalt due to lower surface area to volume ratios. Along with surface texture and shape. 1987). The m axim um aggregate size is at the other end o f the spectrum . According to M ahboub and Allen (1990). Two different laboratory tests are commonly used for its determination (though their values are nor directly comparable). 200 o f 6 and 5 percent for wearing and binder mixtures. al.

It is generally accepted that a dense gradation gives more points o f contact and consequently more aggregate interlock and more shear resistance. large stone mixes have recently been used to reduce rutting on major highways in several states (Button et. except for friction courses or ATPB (asphalt treated permeable bases). a reason why these mixtures are restricted to use in thin and very thin layers. generally contributes to instability.19 benefits o f the use o f larger aggregate sizes.” H owever. a dense graded m ix is preferable. where their great workability promotes optimal placement. this view is not general. even a small one. They believe that. Brosseaud et al (1993) support this view stating that: “Introducing a gap in the grading curve.. . We don’t get much rutting now that the m ix is designed and we do get long lives o f 15 years and more. In the U nited Kingdom we use a total gap grading with harder asphalt and even some rounded sands in our Hot Rolled Asphalt. A nother issue with gradation is whether the aggregates are dense graded or gap graded. you can do more with a mix if properly designed and it doesn’t rut or crack. Brown and Cooper (1984) report that a gap graded mix showed markedly less resistance to permanent deform ation than the continuously graded material. 1990).” R eproduced with permission of the copyright owner. Further reproduction prohibited without permission. M aurice Akeroyd in the discussion o f Huber and H eim an (1987) states ’’You will find that by opening up the grading to a semi-gap or even full gap. leading to good cohesion properties. Mr. al.

Stiff binders assist in resisting permanent deform ation. Asphalt cements with 100 pen are more viscous than 200 pen asphalt cements. 1993). 3 Penetration is used as a measure o f consistency o f the asphalt and is determined with a standardized test. For a given stability range. For good perform ance not only the quality but also the amount o f asphalt in the mix is important. 200 and 100 pen'3. In spite o f the above. (1993) concluded from laboratory experim ents that there exists a critical level o f filling o f the voids by the binder beyond which the material becom es unstable. Huges and M aupin (1987). and the rate o f permanent deform ation accumulation is directly related to the magnitude o f the shear strain in the constant height simple shear test (Sousa et. probably because the gradations are chosen to provide sufficient aggregate interlock to m inim ize the effect o f binder viscosity. For example. Brown and C ooper (1984) have found that binder grade has no effect on resistance to permanent deform ation at a test temperature o f 30°C for two dense bitumen m acadam road base m ixes which had sim ilar grading and volumetric proportions but contained different grades o f binder. the 100 and 200 pen asphalt cements could have the same viscosity at 30°C. al. . since the magnitude o f the shear strain is less for each load application com pared with mixes containing low stiffness binders. it is usually accepted that the rate o f rutting will be higher for high asphalt contents than for low ones.. However. concluded from mix design data that the viscosity o f the asphalt-additive binder combination apparently does not appear as im portant as the gradation. Brosseaud et. R eproduced with permission of the copyright owner. Further reproduction prohibited without permission. al. since the test is carried out at 25°C .20 Asphalt phase: The most im portant characteristic o f the asphalt cem ent for rutting performance is its stiffness at high temperatures.

it controls the air voids in the mix. That is why higher asphalt contents than the ones that are optimal for rutting are generally used. .Although from a rutting perspective the asphalt content necessary to obtain the highest stability should be used. In summary. However.g. it is im portant to rem em ber here that mix design is a com prom ise between rutting resistance and durability. the air voids content is a function o f construction compaction. Voids filled w ith asphalt include the effect o f both air voids content and voids in the mineral aggregate (VMA). Brown (1989) concluded that the m ajor causes o f rutting were excessive asphalt content and low air voids in the asphalt mixtures. There are also numerous interactions between factors. Huber and Heiman (1987) studied 11 pavement sections that carried sim ilar traffic volum es but exhibited different rutting performances. Here. Finally. results from experiments also indicate that for air voids below 3% the stability will usually decrease substantially. Air phase: Given the aggregate gradation and asphalt content. more rut resistance pavements). This makes it quite difficult to selectively control the factors affecting rutting since m ost o f them are highly interrelated.. More com paction will usually lead to higher stability (i. They concluded that asphalt content and voids filled w ith asphalt were the basic param eters affecting rutting. Further reproduction prohibited without permission.. mix behavior is very complex. the effects o f some variables have R eproduced with permission of the copyright owner.e. excessive asphalt content) can invalidate other good properties. aggregate gradation controls the voids in the mineral aggregate and. com bined with the asphalt content and the com paction energy (and the porosity o f the aggregate). All o f the factors are important because one bad property (e.

However. This phenom enon. in mix design these are balanced w ith the factors for the other performance criteria as well (such as cracking).. Another im portant factor that affects rutting o f the underlying pavem ent layers in cold climates is freeze-thaw cycles. they may be classified as either cohesive or cohesionless. Granular materials a n d soils . is described in more detail below. Geotechnical engineers have also made great advances in constitutive modeling o f soils and the use o f the finite element method to predict the perm anent deform ation on geotechnical structures (e. R eproduced with permission of the copyright owner. M any o f the issues with soils are sim ilar to the ones described for aggregates in asphalt concrete mixtures above.22 been discussed with respect to the rutting performance only. water content. such a detailed analysis is not justified for the m anagement o f a netw ork at this point. Depending on the amount and type o f fine material in the aggregates. 1990). 1981).unbound layers: The properties o f these materials depend on gradation. perhaps the most im portant difference with the aggregates in asphalt concrete m ixes is the presence o f w ater in granular layers. and the aggregate shape and texture (Haas et. Thus. density. In addition to the lower rutting resistance o f unbound materials. 1994).g. Further reproduction prohibited without permission.. al. stress state. the data requirements are also high and the scales o f the problems are quite different from the problem s that are confronted in modeling pavem ent rutting. Wood. . related to the water content o f soils. Unfortunately. The resistance variations with water content and compaction energy have been extensively studied in the geotechnical engineering literature and will not be discussed here (Holtz and Kovacs.

Silts are generally the most frost susceptible soils whereas ice lenses will simply not form in coarse-graded soils. At higher tem peratures they becomes softer.23 C lim ate: Climate can also severely affect the rutting performance o f a given pavement structure. . In general. Low tem peratures also have a great effect on the rutting perform ance o f the underlying layers. This was found for a temperature range from 42°C to 60°C. the upper portions of the asphalt concrete pavem ent are the most severely affected by high temperatures. Ice lenses also have little chance to form in non-cracked clay soils because their permeability is very low. these soils may also experience frost problems. Since in the field there exists a gradient o f temperatures with depth. Reproduced with permission of the copyright owner. Seasonal weather variations introduce variations o f material properties and therefore periodic changes o f pavement characteristics (OECD 1988).000 cycles as at 60°C and 10. which allow s som e water movement to the frost line. Climate affects to a large extent the performance and mechanical properties of pavem ent components and hence the pavem ent structure’s ability to withstand traffic loads. clay soils near the surface may be cracked and fissured. for a soil to be frost susceptible. Consequently. However. the same percentage o f rutting was found at 50°C and 100. asphalt cem ents are very temperature susceptible. In particular. Further reproduction prohibited without permission. the bonds between particles then becom e weaker and consequently the relative m ovem ent between particles becomes easier. In frost-susceptible soils4. The translation represented a factor o f 10 in num ber o f cycles. Brosseaud et al (1993) studied bitumen-treated granular materials in the laboratory and found that an increase in the test temperature o f 10°C (measured in the core o f the material) causes a lateral translation o f the rutting curve (percentage o f rutting versus num ber o f cycles in log-log coordinates). it must have fine enough connected pores for relatively high capillary pressures to develop and yet not so fine that the flow o f water is restricted. the action o f frost may cause frost heave and a loss o f 4 Yoder and Witczak (1975) and H oltz and Kovacks (1981) present criteria for determining whether a soil is frost susceptible.000 cycles.

24 bearing capacity during thawing which jeopardizes the pavement structure if the subgrade is not protected from frost (OECD 1988). a material with a very high w ater content) immediately under the pavement. it is a well-known fact that for a given am ount o f traffic. the pavem entsupporting capacity may be greatly reduced as a result o f a layer o f unfrozen soft material (i. For frost heave to occur. W hen a pavem ent cracks. and (3) a supply o f water. it allows the ingress o f water from the pavem ent surface to the underlying layers w ith the consequent detrimental effects on rutting performance.e. ice crystals will continue to grow until ice lenses begin to form. If the soil is highly susceptible to capillary action. W ith respect to precipitation. This is know n as the spring break up (Y oder and Witzak 1975) and it can be accentuated by periods o f high rainfall during the fall and winter and particularly during the frost-melting period. When water freezes. severe heaving cannot be accounted for only by the expansion o f soil w ater during freezing. . R eproduced with permission of the copyright owner. the supercooled water (w ater that remains liquid w ell below the freezing temperature) and ice will have a strong affinity with the result o f w ater being draw n to the ice crystals that are initially formed. As explained by Yoder and W itczak (1975). the following factors m ust be present (Yoder and W itczak 1975): (1) a frost-susceptible soil. surface distress will occur more rapidly if the surface course is frequently wet (OECD 1988). Thus. The effect o f precipitation is more im portant in cracked pavements.. it expands about 9 percent o f its original volume. During thawing periods (even without frost-heaving taking place). Further reproduction prohibited without permission. the lenses in turn grow until frost heaving results. (2) slowly depressed air temperatures.

Nj = number o f standard loads o f magnitude Pj and where y is a constant with a typically assumed value o f about 4.g.. The advantage o f having an equivalence criterion is that the effect o f traffic can be sum m arized by a single variable.e. Unfortunately. in the discussion o f Huber and Heiman. Further reproduction prohibited without permission. Rutting accum ulates faster as the load duration increases. real traffic data (i. the literature contains only qualitative statements about this (e. The use o f ESALs to develop pavement distress progression models has been pervasive in the R eproduced with permission of the copyright owner.25 Loading: Time o f loading affects rutting because o f the Theological properties o f the asphalt cement. = number o f loads o f m agnitude P. the notion o f equivalence between loads in term s o f the dam age they cause gained widespread acceptance at the AASHO Road Test (HRB 1962). That is. . 1987 or Phang. This variable is the number o f repetitions o f Equivalent 80 kN (18.000 lbs) Single Axle Loads (ESAL). W ith respect to the axle load. 1976). loads and num ber o f applications) are converted to a single num ber o f 80 kN (18. to cause failure. The equivalence is usually expressed in the form (OECD 1988): (NJNj) = ( P J P y where N. 1988) or results o f laboratory experiments that have not been quantitatively corroborated with field data (Lai and Hufferd. This is particularly important for sections where traffic moves slow ly such as sections with recurring traffic congestion or clim bing lanes. Tw o other important factors related to loading are the axle load and axle configuration.000 lbs) single axle load applications that cause the sam e damage on the pavem ent structure.

1993). Kandhal et. it is generally assumed that a 143 kN (32. al. Phang (1988) states that tw in depression ruts appearing in C anada’s highways are a consequence o f a major change to the use o f radial ply truck tires with inflation pressures o f about 750 kPa (110 psi) from bias ply tires normally inflated to 500 kPa (75 psi). but is laterally distributed over the traffic lane. As is well known.000 lbs) tandem axle load has alm ost the same dam aging power as a 80kN (18.26 pavem ent research community. 1988. the equivalencies given in the AASHTO G uide for Design o f Pavement Structures (AASHTO 1993) and previous versions o f it are based on the serviceability concept. T he wheelpath distribution o f the loads is known as w ander and also R eproduced with permission of the copyright owner. O ther factors that influence the rutting performance o f pavements are tire type. the use o f ESALs may not entirely appropriate for the development o f rutting progression models. Tire inflation pressure affects pavem ent damage by changing tire contact patch size and tire vertical stiffness. Since the equivalency factors for rutting can be different from the ones for serviceability.000 lbs) single axle load (the equivalency factor between these two loads is 1 in the AASHTO G uide (AASHTO. traffic does not move in a single wheel path. Further reproduction prohibited without permission. 1993)). tire inflation pressure and traffic wander. The present serviceability index is a measure o f the user perception o f the pavement condition developed at the AASHO Road Test. Many authors attribute the increased rutting observed on pavements to a combination o f increased traffic loads and increased tire truck pressures (Phang. Unfortunately.. The axle load configuration also affects the damaging power o f the axle. . For example.

2. Although. In order to perform these R eproduced with permission of the copyright owner. . The description given below is an oversimplification o f m ost models. The mechanistic com ponent is an analysis in which a pavem ent structural response such as the vertical com pressive strain or the shear strain is com puted by using either a layer-elastic analysis or a finite element analysis.3 M odeling Approaches Rut depth progression models can usually be categorized according to one o f the following approaches: • M echanistic-Empirical. One would expect that the w ider the lane.27 affects the rutting performance. the mechanistic-empirical m odeling framework for rutting has two m ain com ponents. Further reproduction prohibited without permission. “M echanistic design procedures are" [currently] "based on the assum ption that a pavem ent can be modeled as a multi-layered elastic or visco-elastic structure on an elastic or visco-elastic foundation. the higher will be the w ander and consequently the lower the rut depth (with probable wider ruts). H ow ever as its nam e implies. and • Empirical According to the AASHTO Guide for Design o f Pavem ent Structures (AASHTO. 1993). this variable is never measured (other than in some research studies). it can potentially be taken into account by considering the lane width as an indicator.” This approach will not be explained in detail since it can take several forms.

28 analyses. el. . and (3) the ability to extrapolate from lim ited field and laboratory results. 1993). Further reproduction prohibited without permission. (2) ability to predict specific types o f distress. mechanistic empirical design procedures are still lim ited for use with flexible pavements (AASHTO. Improved reliability for design will only be achieved if the material characterization and the effects o f clim ate and environm ent are better modeled. it is not clear that the benefits cited above can be realized in the short run. According to the AASHTO Guide for the Design o f Pavement Structures (AASHTO. The approach has not been widely used because o f the difficulty in obtaining elasto-plastic or visco-plastic characterization for the various paving materials in-situ (Ali et. the ability to predict specific types o f distresses is certainly not limited to the mechanistic em pirical-approach. A mechanistic approach is certainly appealing to civil engineers since it is based on principles that have been applied successfully to solve other civil engineering problems. Furthermore. In spite o f the above. environm ental inform ation. Finally. Other inputs necessary for this analysis are traffic. 1993) the primary benefits which will accrue from the successful application o f m echanistic procedures are: (1) improved reliability for design.. R eproduced with permission of the copyright owner. In the empirical component the structural response is related em pirically to the rutting performance. 1997). material properties such as moduli and Poisson ratios are needed and are usually obtained from either laboratory experim ents or back-calculated from pavement deflection inform ation. and material responses to changes in environmental conditions and assumed loading.

However. The shift factor ends up being a function o f alm ost all variables affecting rutting. not practical in a pavem ent m anagem ent context except when using very sim ple m odels. Work in this area is currently being performed at UC Berkeley and elsewhere. at least with current technology. however. The critical question here is at what points are the stresses com puted? Selecting a given point (or a few points) in the pavem ent structure represents an aggregation o f information. First. Thus. The current practice is to aggregate the effect o f the environm ent by simulating the pavement structure at representative conditions (or at just a few seasonal conditions). R eproduced with permission of the copyright owner. improvements in the constitutive m odeling o f asphalt concrete mixes are needed to be able to model plastic deformations. some conditions should be considered for its successful application. plastic strains are related em pirically to either strains or stresses com puted using elastic or visco-elastic models. . Thus. Further reproduction prohibited without permission. the change o f material response with environm ent must be taken into account. which m akes this approach.29 In the present context. All these aggregations. Second. Third. to really achieve the advantages o f this approach. the determ ination o f the shift factor is a problem alm ost as com plex as the original problem. the effects o f the loads are dram atically different at high temperatures or during the spring break-up in cold regions than at other times. one way or another. The wander o f traffic m akes these issues even more complicated since two consecutive identical loads will most likely cause different stresses and strains at a given point in the pavement structure simply because the loads were laterally shifted. more com prehensive structural analyses may be required. go into shift fa c to rs that are normally used to make the predictions agree with the em pirical observations.

relevant results from the literature on the m echanistic-em pirical approach to modeling pavement rutting are review ed so as to identify suitable model forms. Despite this focus. the most com m on model found in the literature is o f the form ep = a N b Reproduced with permission of the copyright owner. the empirical approach is adopted for this research. By far. the focus o f this research is in the estim ation o f em pirical models of pavem ent rutting performance.” Thus. (2. Further reproduction prohibited without permission.. in light o f the above.30 In summary. HDM III. Rutting is the result o f the integration o f the plastic strains over the pavem ent structure.1) .4 E v idence from th e M echanistic-E m pirical L ite ra tu re As m entioned above. The following citation from the HDM-4 study (Kannemeyer. although the prediction capabilities. is probably the m ost widely used pavement managem ent system in the world): ” . and thus the general acceptability. 1999) serves as a further illustration o f the problem s with their application (its predecessor.. while the direction o f pavem ent engineering is moving tow ards mechanistic approaches. of the HDM -4 rut depth models might be improved by incorporating such detailed m odels. their developm ent is still limited. Numerous m odels have been used to relate plastic strain accum ulation to the number of load or stress repetitions.” [mechanistic] “ the detail required to utilize these m odels would generally make them impractical for network level applications. 2.

N = num ber o f stress applications. 1991) as well as for asphalt concrete m ixes (Khedr. (1997): ep = a ' se N b ( 2 . 1996.1).3) . (2. Given a level o f elastic strain. a. The following equation form is used by Kenis (1977) and by Ali et. and material characteristics. In the Texas Flexible Pavement System (TFPS). functions of applied stress. Behzadi and Yandell.31 where £p = permanent or plastic strain. this equation is equivalent to equation (2 . the permanent strain on asphalt concrete m ixes is assumed to behave in essentially the sam e manner as above: =a" e e N~b" dN R eproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1986). al.2 ) where ec is the elastic strain and a 'a n d £ 'a r e permanent deformation parameters. 1976. The above model form has been proposed for subgrades and unbound materials (M onismith. Vuong and A m strong.b = estimated coefficients. Diylajee and Raymond. This equation is based on the proportionality between the plastic and elastic strains in a pavem ent structure under traffic loading. 1982.

a is influenced by these factors to a greater extent than b. 2. Their estimates vary widely among researchers depending on the materials involved and test procedures.5 Evidence from the em pirical literature The m ost common relevant finding in the empirical literature is the concave shapes o f rut depth w ith cumulative num ber o f load repetitions. Thom pson and Nauman. .32 where a "and b " are estim ated param eters and ee is the elastic strain (Button et.. al. the elastic strain is assum ed to remain constant throughout the life o f the pavement. b \ and b ') are usually considered functions of applied stresses and material properties. 1981 and Paterson. al. it produced a mediocre fit. In general. M ost rutting models developed to date have been limited to linear specifications and do not account for the effects o f the environm ent (e. a". Paterson (1987) developed a non-linear model w ith data from in-service pavem ents that included the effect o f the environment. Furthermore. in spite o f its com plexity. 1987). most developed models specify such a concave shape (e. R eproduced with permission of the copyright owner. The coefficients a and b (and also a'. 1982. the form in Equation (2.g.. Such trends have been observed with Heavy Vehicle Simulators by Maree et. (1982) and by Harvey et. 1990). (1997) and in other experim ents such as the AASHO Road Test (H RB. Unfortunately.2) (for a constant ee) is obtained if this equation is integrated. In the TFPS analysis.g.. 1962). 1993). Lister. Further reproduction prohibited without permission. O bviously. Saraf. al.

al. between air voids. Reproduced with permission of the copyright owner. which may explain in part the lack o f success in the development o f these models. These load equivalence and structural coefficients were derived using the concept o f serviceability. Finally. In addition. a structural num ber was also com puted using the structural coefficients developed at the AASHO Road Test. the model does not account for the effect o f high temperatures.33 In all o f these m odels the traffic was converted to ESALs using the load equivalencies developed at the AASH O Road Test and in Paterson’s model in particular. and stability. As mentioned before. other researchers such as Kandhal et. Finally. this model also ignores the relationships that exist. M odels like this only predict rutting developed in the asphalt concrete layers so they are useful only for thick pavements. . Further reproduction prohibited without permission. voids in the mineral aggregate. (1993) have developed model forms containing inform ation only for the wearing course mix. the corresponding parameters for rutting are not necessarily the same. for exam ple.

34

C hapter 3: Data Sources

This chapter briefly describes the two data sources that will be used for model
estimation. The chapter starts w ith a description o f the AASHO Road Test in Section 3.1,
followed by a description o f the WesTrack Road Test in Section 3.2. The chapter
concludes

with

a

summary

in

Section

3.3 that highlights

the

complementary

characteristics o f these two data sources.

3.1

AASHO Road Test

The AASHO Road Test (HRB, 1962) is the most com prehensive closed-loop, traffic
controlled road test experiment performed targeted at evaluating performance o f
pavem ent sections. The test was carried out between 1959 and 1962 near Ottawa, Illinois
about 80 mi southw est o f Chicago.

The test facilities consisted o f six loops. Each loop was a segm ent o f a four-lane divided
highway. The centerlines divided the pavement into inner and outer lanes, called lane I
and lane 2 respectively.

All variables for pavem ent studies were concerned w ith pavem ent designs and loads
w ithin each o f the sections. The length o f the sections used in this study was 30 m
(100 ft). Short transitions separated the pavement sections.

R eproduced with permission o f th e copyright ow ner. Further reproduction prohibited without permission

35

The sections were subjected to traffic for slightly more than 2 years. Every vehicle in
each one o f the traffic lanes had the same axle load and axle configuration. The
assignment o f axle loads and vehicle types to the various lanes is shown in Table 3.1.

The climate o f the Road Test area had an average annual precipitation o f about 864 mm
(34 in). The average mean sum m er temperature was 24°C (76°F) and the average mean
w inter temperature -3°C (27°F). The soil usually remained frozen during the winter with
alternate thawing and freezing o f the immediate surface.

The north tangent o f each o f the six loops in the AASHO Road Test was constructed o f
flexible pavement (i.e., asphalt concrete pavement). The six tangents included a total o f
234 structural sections or 468 test sections. A majority o f the test sections in each
pavem ent section tangent comprised a complete factorial experiment, the design factors
o f which were surfacing thickness, base thickness and subbase thickness. These
experim ents were referred to as the main factorial designs. The data used in this research
consists o f data for this main factorial design.

The inform ation available for model estimation consists essentially o f initial thicknesses
o f the asphalt concrete, base and subbase layers, overlay thickness (if applicable) and
biw eekly inform ation on axle load and number o f repetitions, mean m axim um and
m inim um temperatures, and mean rut depth. The design asphalt concrete, base and
subbase material characteristics were the same for all sections.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

36

T ab le 3.1: A ssignm ent o f axle loads to the various lanes o f the AASHO Road test.

Loop

L an e

2

1

2

2

3

W eig h t in kN

T ru c k
C o n fig u ratio n

F ro n t axle

L oad axle

Gross
w eight

r

9.1

9.1

18.2

V'T

9.1

27.2

36.3

1

18.2

54.5

127.2

3

2

27.2

109.0

245.2

4

1

27.2

81.7

190.6

4

2

40.9

145.3

331.5

5

1

i1

27.2

101.7

231.6

5

2

tl 1

40.9

181.6

404.1

6

1

40.9

136.2

313.3

6

2

54.5

217.9

490.3

--------r

T T i

---------IT

For the test sections used in this research, subbase thickness ranged from 0 to 406 m m
(16-in.) in 102 mm (4-in.) increments; base thickness ranged from 0 to 229 mm (9-in.) in
76 mm (3-in.) increments; and surfacing thickness ranged from 25.4 m m (1-in.) to 152
m m (6-in.) in 25.4 m m (1-in.) increments.

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

37

The upper 900 mm (3-ft) o f embankment within the test loops was constructed with a
selected A-6 soil w ith laboratory and field (spring) C BR s between 2 and 4. The top
surface o f the em bankm ent was graded to the required lines and grades.

The subbase material was a uniformly graded sand-gravel mixture with 100 % passing
the 25.4 m m (1-in.) sieve. The base material was crushed dolom itic limestone with 100%
passing the 38 mm (1-1/2-in.) sieve. For the base and subbase materials, the fraction
passing the No. 40 sieve had no plasticity.

The asphalt concrete surface course, used for the upper part o f all asphalt concrete layers,
was a m ixture o f dense-graded crushed dolomitic lim estone aggregate, 19 mm (4/3-in.)
m axim um size, and natural sand with about 5.4 percent o f 85-100 penetration grade
paving asphalt. With the exception o f the top 38 mm (1-1/2-in.) that were constructed
with the surface course mentioned above, an asphalt concrete binder course was used for
asphalt concrete layer thicknesses 76 mm (3-in.) and above. The binder course was a
m ixture o f dense-graded crushed dolomitic lim estone aggregate 25.4 mm (1-in.)
m axim um size and natural sand with about 4.5 percent o f 85-100 penetration grade
paving asphalt.

The discussion so far has been purposely vague about w hat is m eant by rut depth. The
following paragraph relates the rut depth measurements at the AASHO Road test with
more recent A ASH TO definitions.

A ccording to A A SH TO (1993), to determine the m ean rut depth, a 4-foot (1.2 m)
straightedge should be laid across the rut and the m axim um depth measured. Although

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

R eproduced with permission of the copyright owner.2 m (4-ft).38 the same definition was given in the glossary o f terms in the AASHO Road Test report (H RB.1. Further reproduction prohibited without permission.1 illustrates. . in the text o f that report it is mentioned that the routine biweekly rut m easurem ents were carried out with the device show n in Figure 3. 1962). 1962). however. that the m easurem ents obtained w ith this device and a straightedge are not necessarily the sam e as Figure 3. experimental data to corroborate this assumption are not available. N evertheless. as shown in Figure 3.2 m straightedge. Notice. This means that it could be safely assumed that the measurements from the A A SH O Road Test are equivalent to those that would be obtained with a 1. The distance between the legs o f this device is also 1. Further. M anual measurements with a straightedge would normally be registered as zero in cases like this. According to cross sections shown in the AASHO Road Test report (H RB. the difference should be negligible. Unfortunately. for small rut depths or for larger rut depths where all the layers are contributing to rutting (the high and low points are far apart transversally and the low point in located near the middle o f the rut).2. in a recently constructed pavement (norm al crow n section) one can m easure negative rut depths w ith the device used at the AASH O Road test. this seems to have been the case for m ost sections at the AASHO Road Test.

Rut depth measurements with a straightedge (R D i) vs.39 AASHO Device 1.2: Negative rut depth m easurem ents with the AASHO device. Reproduced with permission of the copyright owner. AASHO Device Pavem ent surface N e g a tiv e m easurem ent Figure 3.2 m straight-edge RD F igu re3. Further reproduction prohibited without permission. rut depth m easurem ents with AASHO device (RDi).1: RD. .

Thus.) o f engineered fill which was obtained from the natural subgrade m aterials. Further reproduction prohibited w ithout p erm ission.) o f dense-graded crushed aggregate base.40 3. 300 mm (12-in. The 26 test sections o f hot-mix asphalt (HM A) concrete were constructed on the straight tangents between the curves. The oval track consists o f two tangent sections connected by two spiral curves. T he 26 sections have the same structural design. A single primary aggregate source was com bined to develop the three gradations selected for the original experim ent. and 150 m m ( 6 -in. 1996). and three air-void content levels.2 W esT ra ck R oad T est W esTrack consists o f a specially built track in the state o f Nevada. 300 mm (12-in. al. asphalt contents. three aggregate gradation levels.) o f scarified and mixed subgrade soil. 1998).. . The aggregate was a partially crushed fluvial deposited aggregate. The experimental design consists o f three asphalt content levels. consisting prim arily o f andesite (Hand. w hich consisted o f 150 mm ( 6 -in) o f hot m ixed asphalt (HM A) concrete. The original test sections included both fine and coarse graded Superpave mixtures developed from a locally available crushed gravel and a non-modified PG64-22 binder as graded in accordance w ith the Superpave binder classification system (AASHTO 1996). air void contents and aggregate gradations were systematically varied among the sections to represent typical construction variations (Epps et. The em phasis o f this test was on studying the effects o f deviations about target values in materials and construction factors (W esTrack. R ep ro d u ced with p erm ission o f th e copyright ow ner. 1998).

and a project developed laser device (Hand. in the form o f both rain and snow. al. From the cross section profiles.000 lbs) with the steering axle carrying 54 kN (12. Each load axle carries 89 kN (20. an A rizona Department o f Transportation (ADOT) transverse profile device.000 lbs). Traffic loading is provided primarily by an autonomous vehicle technology. These profiles were measured over time with three different devices: the “D ipstick'’.41 The natural soil at the test site and consequently the engineered fill are mostly uniform fine-grained sandy materials. is approximately 100 mm (4” . The climate at the test site is typical o f high desert climates in the intermountain western portion o f the United States. . Frost penetration into the fill and subgrade m aterial is unlikely at this site (Epps et. Two o f the load axles form a tandem axle with load o f 168 kN.3. Four tractor/triple-trailer com binations are utilized for pavem ent loading. These materials range from clayey sands to silty sands. The target vehicle speed for the test is 64. 1998)..). 1998).4 kph (40 mph). A few sum m er air temperatures approach 40°C’ (K M T) and some w inter air tem peratures may fall to -18°C . Radial tires inflated with pressures o f 655 to 689 kPa (95 to 100 psi) are used. routine cross-section profiles were obtained approximately every two weeks. Annual precipitation. The tractor/trailer com bination has 7 load axles and a steering axle. 1998). as shown in Figure 3. Further reproduction prohibited w ithout p erm ission. R ep ro d u ced with p erm ission o f the copyright ow ner. rut depth measurements were obtained in accordance with the LTPP protocol (FHW A. In the W esTrack project.

AASHO provides controlled variations in thicknesses and traffic loading between different sections but the materials used in all sections. On the other hand. thus reducing their depths.42 The lateral distribution o f the loading (wander) in WesTrack is too complex to describe because it w as partly affected by temperature. they were set at specific positions for several days.3 Summary The two data sources have complementary characteristics that are sum m arized in Table 3.3: Truck configuration used in WesTrack. were the same. Further reproduction prohibited w ithout p erm ission. clear reductions in rut depths (as high as 7 mm) between m onitoring sessions were noted. W esTrack provides controlled differences R ep ro d u ced with p erm ission o f the copyright ow ner. However. . 53.4 kN ( 1 2 K) 89 k N 89 kN 89 kN 89 kN 89 k N 89 k N 89 kN (20 (20 K) (20 K) (20 K) (20 K) (20 K) (20 K) Fr o n t a x l e K ) T a n d t t m A x le Figure 3. Such effects seem to be transient because the rut depths usually return quickly to the rutting levels before the changes in antenna location. Since the antennas were not continually moved. These reductions can be connected to changes in the vehicle wander that flatten the peaks o f the previously created ruts. As pointed out by Hand (1998). 3. For some sections. except for construction variability.2. by the end o f the test. the distribution was achieved by m anually moving guidance antennas transversally on the front bumpers of the trucks. it approached that o f normal traffic.

the environm ental effects are different in both tests.43 in asphalt concrete properties but all sections have the same design and loading. In the AASHO Road Test. . Factor AASHO Road Test Layer Thicknesses • Load magnitude • Load repetition • Thawing • W esTrack • High air temperatures • Asphalt concrete mix characteristics • R ep ro d u ced with p erm ission o f the copyright ow ner. high air temperature was the most important environmental variable affecting the performance o f the asphalt concrete mixes. in the WesTrack Road Test. In contrast.2: Pavem ent characteristics included in the AASHO Road Test and WesTrack data sets. Table 3. the freeze-thaw effects on the subgrade material played the most important role on pavement performance. Further reproduction prohibited w ithout p erm ission. In addition.

(4.5 describes the param eter estim ation results and the chapter concludes w ith a summary in Section 4.1 Specification o f a model for rutting originating in the underlying layers Based on the evidence from the material laboratory testing literature reviewed in chapter 2 .6. The model accounts for rutting originating in the asphalt concrete layer as well as for rutting originating in the underlying layers. the following two equations can be used as starting points for the model specification: RD„ = 5 . The data set used for model estim ation also contains information on the progression o f rutting in some overlaid sections. Section 4. Section 4. These two models are then put together in Section 4. Section 4. and from the concave trend o f deform ation with respect to the num ber o f load applications appearing in m ost empirical and accelerated test studies.Af„4' R ep ro d u ced with p erm ission of th e copyright ow ner. Further reproduction prohibited w ithout p erm ission .44 Chapter 4: Rutting Model for the AASHO Road Test Data Set This chapter presents the specification and estimation o f a pavem ent rutting progression model using the AASHO Road Test data set. The sum o f the rut depths originating from these two sources yields the total rut depth at the surface o f the pavem ent structure. Thus. + 0.3. The chapter is organized as follows.2 describes the specification o f the model that accounts for rutting originating in the asphalt concrete layer. 4.4 describes the model modifications necessary to handle asphalt concrete overlays.1 describes the specification o f the model that accounts for rutting originating in the underlying layers and Section 4.1) .

1) over (4. in the derivations that follow. = rut depth construction for pavem ent section This is an unobserved value that contributes to the total initial rut depth.1) have received most o f the attention in the material laboratory testing literature partly because. Therefore. ( \ . That model. the apparent advantage o f equation (4. the m odel becomes non-linear in the variables and the parameters regardless o f which equation is used. Thus. Equation (4. +a. However.2) is the one selected for the m odel specification in this dissertation. they are easy to transform into a linear form.e b' N“ ) (4 .45 or RD„ = 5.1) have been used by Archilla and Madanat (2000) to develop a model o f rutting for the AASHO Road test data set. .2) disappears. N„ = a variable representing the cumulative number o f load repetitions applied to pavement section / up to tim e period t (a more complete definition is given later).2 ) where RD„ = rut depth for section / at tim e t (mm). their param eters are relatively sim ple to estimate by linear regression. and 5. if the intercept term is ignored or can be estimated. Further reproduction prohibited w ithout p erm ission. An equation o f the form o f (4. a and = functions o f the characteristics o f pavem ent / such as layer originating in the underlying layersim mediately after thicknesses. which is the basis for R ep ro d u ced with p erm ission o f the copyright ow ner. Equations o f the form o f (4.

.1). This is unlikely. the definition o f N is straightforward. the model developed from equation (4. but this is not so for the AASHO Road Test where different pavem ent sections were subjected to various load levels and different load configurations (single or tandem axles). This is the main reason why equation (4. In W esTrack alm ost all rutting originated in the asphalt concrete layer. This is even more com plex for actual pavement sections since each section is subjected to a distribution o f loads and configurations. R ep ro d u ced with p erm ission o f the copyright ow ner.46 the specification in this section. H owever. This is actually what m any researchers have done in the past. For laboratory experiments that are usually carried out at a given stress level. One possible solution to this problem is to use the cumulative num ber o f equivalent single axle loads (E SA L s)1.1). H aving selected equation (4. did not account for rutting in the asphalt concrete layer as the complete model presented in this chapter does. The problem w ith this approach is that it is assum ed that the load equivalency factors that were based on the serviceability index are appropriate in the case o f rutting. 1 A definition o f ESALs was given in Chapter 2. attention is now given to its components.2) as the basis for the model specification. This w as particularly evident with the W esTrack model presented in the next chapter.2) was chosen over equation (4.2) is able to model better the hardening o f the asphalt concrete materials than a m odel developed starting from equation (4. and consequently biases may be introduced in the estimation if this path is followed. when this source o f rutting is considered. Further reproduction prohibited w ithout p erm ission.

AL2. FL. = load in front axle o f truck used in section / (kN or lbs). = load in tandem load axle(s) (rear axle(s)) o f truck used in section / (kN or lbs). and thus this concept can be used to define N. In 2000. Further reproduction prohibited w ithout p erm ission.= 1 or 2).47 N evertheless.000 lbs) single axle load. A ll. which is the standard practice in pavem ent engineering. the concept (if not the specific values) o f axle load equivalencies is well accepted in pavem ent engineering. . Tandem axles have been standardized by ft? ■80 kN. R ep ro d u ced with p erm ission o f the copyright ow ner. SAL = 80 kN if loads are expressed in kN or 18. These param eters determine the equivalencies between axle loads.S = number o f vehicle passes on section / during period i-. In the 1960’s the tire pressures on heavy trucks usually were between 500 and 600kPa.000 lbs if loads are expressed in lbs. = number o f load axles in truck used in section / (R. and pj = parameters to be estim ated (j=5. = load in single load axle(s) (rear axle(s)) o f truck used insection i (kN or lbs). w hich is the standard tandem axle 2 N o dependency on tire pressure was introduced. SAL (4. AL1. In equation (4.3) where AV. R.3) all the single loads have been standardized to an equivalent 80 kN (18.7). A (3.6. the tire pressures are about 700 kPa. A SAL AL2.t as follow s2: A = i> P „ Ik \S A L j A + R.

and b. Thus. it seem s that a plausible assum ption for b.48 producing the sam e rutting as a single 80 kN axle. Further reproduction prohibited w ithout p erm ission . the identification o f a different exponent for tandem axles is considered important. On the other hand. it is reasonable to assume that the equivalence coefficient for single axles is the same regardless o f w hether the axle has single or double wheels. Specifications for a'. seems to be dependent on pavem ent strength. Notice that /ifc captures the equivalency between different load m agnitudes for tandem axles. R ep ro d u ced with p erm ission of th e copyright ow ner. b.2 m) than the separation between tires o f dual wheels (o f the order o f 0. in the low er portions o f the pavem ent the distributions o f stresses are similar. The reason is that the axle separation in tandem axles is larger (1. the differences in stresses between single and tandem axles are substantial at greater depths. = b is assum ed to be constant for all sections. On the other hand. It is assumed that the exponent for the front axle load is the same as the exponent for single axle loads (/%).0 to 1. In this dissertation. This definition o f A'„ makes it independent o f the units being used. However. Since the model presented in this section is intended to m odel the rutting originating in the lower portions o f the pavement only. The only difference between these axles is that the front axle had a single wheel whereas the rear single axles had double wheels.3 m). These two different wheel configurations produce different stress distributions in the upper portions o f the pavem ent structure. a'.: From the literature review in chapter 2. . is that is relatively constant or at m ost that it varies linearly w ith pavement strength.

Thom pson and N aum an’s results indicate that as the structural response decreases. Specifically. To model pavem ent strength. this conclusion is indicative o f the inability to estimate the true relation. the strength o f the pavem ent is modeled as: i W .'s for structural responses less than a certain value. (although this is alm ost identical to the structural number. a concept sim ilar to the structural num ber defined in A ASH TO (1993) is used. vary with strength in the m anner illustrated in Figure 4.1. R ep ro d u ced with p erm ission o f the copyright ow ner. observed low a'. their a 'te rm decreases. The erratic trends for weak pavements may be a consequence o f their estim ation approach. they concluded that their a ' term followed threshold type relations. Further reproduction prohibited w ithout p erm ission.e. However. the less the accum ulated rut depth for a given traffic.4) where RN t = resistance num ber for p a v e m e n t. . This m eans that the a'. It is more likely that the relation is rapidly varying near their threshold. after performing 192 regressions with a specification sim ilar to equation (4. i.3 (4..49 Thom pson and N aum an (1993). Based on these results. or it may be because the variance o f the intercept increases with its mean value. 2 + p 3 7’.1 + P 2 7. This is also what intuition would suggest. The exponential function provides a way to obtain such a shape. a different name is used to m ake explicit that this num ber is specific to rutting) Tn = thickness o f the asphalt concrete layer for pavem ent / (m).1). but high m agnitudes and erratic trends above that value. Ti2 = thickness o f the granular base layer for pavement / (m). which simply says that the stronger the pavement.= p l 7’. as the pavem ent strength increases.

fa = contribution o f the j lh layer to the pavement resistance. N ow if a thickness T.1: Anticipated relation between the a'.3 = 0.\ = Ta = T. and subbase respectively.2.1) (4 5) a ' . Assume that T. The above equation admits the following interpretation.? o f subbase material is added to the pavem ent structure. . (1 .: a \ =P'4 e ~AV* = p ' 4 e . coefficient and the strength o f a pavement. Further reproduction prohibited w ithout p erm ission. The following expression is used to relate a to RN. then the rut depth caused by the first standard axle load passage is reduced in a proportion R ep ro d u ced with p erm ission o f the copyright ow ner. traffic loads move over the subgrade material (or over a thin w earing course that does not add structural resistance). In such a situation a'. base. where j = 1.(P>7^ +P:r' : +lh/.50 Ti3 = thickness o f the subbase layer for pavement / (m).eA) = represents the rut depth caused by the first standard axle load passage (N lt = £ ’4 1 (1 . Strength Figure 4.e6) in equation 4.2). that is.3 for asphalt concrete.

(1 . an environm ental variable is defined with the inform ation available.. so it is not necessary to worry R ep ro d u ced with p erm ission o f the copyright ow ner. A sim ilar reasoning for the base and asphalt concrete layers leads to equation (4. These P ’s are functions o f the subgrade..3 ). AccumFze t_\ . Thus an accum ulated freeze index for period t is computed as follows: AccumFze. In what follows. Further reproduction prohibited w ithout perm ission.. base. .eh) represents the rut depth originating in the underlying layers caused in the pavement structure by the first standard axle load. N evertheless.e6) = P ’ 4 exp(-P 3 T.5) (times (1 . the rut depth caused by the first standard axle load is now a'. and asphalt concrete materials. Environmental effects: M ost sections in the A ASHO Road Test showed an evident increment in the rate o f rut depth progression during the spring months. (1 .6) where MeanM inTt is the mean minimum tem perature (°C) in the two week period t (in the A ASHO road test there was no freezing in period 1. The environmental inform ation available in the database for the A ASHO Road Test was very limited. subbase.M eanM inT t) / = 2. 7) (4.3 ) (1 .eA)). T hat is. the rut depth originating in the underlying layers approaches zero asymptotically.. a thawing index is computed with the following reasoning: Freeze will only accum ulate when tem peratures are below 0°C. This is a convenient interpretation because as the pavement becomes more resistant. = m ax(0 -M eanM inTt ) t=1 Accum Fzet = m ax(0.51 given by exp(-P 3 T. a '. from the information about the m axim um and minimum temperatures.e6). In summary.

= Accum F ze. This thawing index w ill be zero when the mean maximum tem perature in the period is below zero or when there is no accumulated freeze. The effect o f thawing will be the greatest when there is considerable accum ulated freeze from previous periods and the tem peratures in the current period are substantially above zero. is the mean m axim um temperature (°C) in the two week period preceding t. this variable starts increasing. Thus. Thus. as illustrated in Figure 4. At some point in time the minimum tem perature again exceeds 0°C. R ep ro d u ced with p erm ission o f the copyright ow ner. When there are enough periods with tem peratures above zero the accumulated freeze is exhausted and therefore the variable AccumFze. then reaches a maximum and then returns to zero at the end o f the thawing period.52 about what happened before that) and where T. •m ax(M eanM axT’ . w hen thawing starts. Once the m inim um temperature fails below 0°C. freezing starts to accum ulate. becomes zero again.0) (with units o f °C2) (4. Further reproduction prohibited w ithout p erm ission. there will be large am ounts o f water in the pavem ent structure with the consequent detrimental effects. thus reducing the accumulated freeze. In such cases.2. . is the number o f observations for section i. a thawing index representing this interaction of cum ulative freeze with tem peratures above zero is defined as follows 77.7) where MecmMcixT.

The rise o f the pavem ent surface was attributed to the presence o f frost in the structure and embankment (subgrade R ep ro d u ced with p erm ission o f the copyright ow ner. . Follow ing the AASHTO soil classification system.53 2000 1800 1600 1400 1200 2 1000 00 1 800 ^ 600 400 200 A N ov Jan . Due to the low perm eability o f clays. In addition. they are not susceptible to frost action unless they are cracked or fissured. 1962). Further reproduction prohibited w ithout p erm ission. For freezing to occur in the pavem ent structure.I _ j — Mar i M ay i i \ i i Jul S ep I i N ov Jan M ar i i M ay i \ ■ Jul i S ep N ov D ate Figure 4. which can be identified as a claylike material. the subgrade material in A ASH O was an A .2: Thawing index com puted at the AASHO Road Test. this seem s to have been the case. N evertheless. studies o f the seasonal changes in elevation o f the pavements indicated that the pavem ent surface rose in the early spring (H RB. G iven the precipitation and w ater table inform ation at the AASH O road test site. there should be enough water available.6 soil. the subgrade soil has to be frost susceptible.

. a'. a first order Taylor series approxim ation was used to approxim ate equation (4.9) and a . AN. Having defined the thawing index. — a '. Further reproduction prohibited w ithout p erm ission.. it is now explained how it is incorporated in the model. This indicates that all the unbound materials in the A A SH O Road Test w ere som ew hat susceptible to frost action. Obviously. Thus. b. e p"u" R ep ro d u ced with p erm ission o f the copyright ow ner. + a . thaw ing alters the m aterials' properties so one could try to incorporate its effect in a'. ySA Lj P 7 SAL \flb (4. (4. e""* = /?£> . e p' Tl' AN.54 material).2): RD„ * R D . the introduction o f a correction factor for a.. A I2. With this new form ulation..10) . increases. x b N ote that equation (4. T he problem is that equation (4.2) is not suitable for this kind o f adjustm ent. e hN“ (4. then it is possible that the function decreases afterw ards when there is no more thawing. The reason is that one would like to obtain a m onotonic increasing function with traffic.8) where: EL ySA L j + R.+ a. If during thawing. w hen the environm ental conditions change is done as follows: RD„ a R D .. AN.^ + a..5) still defines a{ if is replaced by J34 = /3'4 * b. AL\.

+ } S= a .|.55 where b has been replaced by /%. . Or. 2.)) is 1 and whenever there is thawing the factor is greater than one im plying that the pavem ent will rut faster during the corresponding period. In this section a sim plification o f that model is presented for the AASHO Road Test. etc.11) I W henever the thawing index is zero. the intent is to capture m ost o f the freeze-thaw effects at the AASHO Road T est so as not to cause bias in the estim ation o f the loading and resistance parameters. The W esTrack model specification has the following form: (4.e J 77' ] ^1000^ AN IS e p. the new multiplicative factor (exp(Pg 77.2 Specification o f a model for rutting originating in the asphalt concrete layer The next chapter presents in detail the specification and estim ation o f a model o f rutting for the W esTrack Road Test that captures rutting originating in the asphalt concrete layer. Further reproduction prohibited w ithout p erm ission . .12) R ep ro d u ced with p erm ission of th e copyright ow ner. RD tl * 6 . RDU. It is im portant to stress that the use o f the thawing index is not intended to be a precise description o f the freeze-thaw problem in more general cases.A'“ (4. However.11) is the model specification used for rutting originating in the underlying layers.. Equation (4. substituting successively the values o f RD. 4.

Since the asphalt cements used for both R ep ro d u ced with p erm ission o f the copyright ow ner. As will be shown in the next chapter. All the sections at the AASHO Road Test were designed to have the sam e asphalt concrete mix properties. 13 ) where TempDnmt is equal to one if the M eanMaxT. is greater than 28. ( 4. Since m.s = a function o f three asphalt mix characteristics. and loading. it is logical to assum e that the response to loading was sim ilar for all sections.6°C threshold is transferred from the W esTrack model presented in the next chapter. The variability in the observed behavior is caused mainly by the inevitable variability in construction. . Therefore. Further reproduction prohibited w ithout p erm ission. some sim plifications are required for the model specification for the AASHO Road Test because the asphalt concrete mix properties for the as-constructed sections was not available. and loading.s is a function o f some asphalt mix characteristics.s = a variable representing the cum ulative num ber o f load repetitions applied to pavem ent section i up to time period s (a m ore complete definition is given in page 60). temperature. and P u = model parameters. The 28.6°C and 0 otherwise. Note that it is implicitly assumed that the effects o f high air temperatures are the sam e as for the W esTrack Road Test. m. N '. it can be approxim ated by: m . a. tem perature.s depends on som e characteristics o f the asphalt concrete mix and for the AASHO Road Test data set.56 where m.s =Pio + P 1 1 TempDum.

. Further reproduction prohibited w ithout p erm ission. for a constant tire inflation pressure. the stresses caused by different load m agnitudes immediately below the pavem ent surface are identical. since this variable is intended to capture the effects o f loading on the upper portion o f the pavem ent.57 tests are classified with different specifications that are not directly com parable. H owever. equation (4. after the WesTrack m odel has been introduced.3).3 illustrates the effect o f these two factors on the variation o f the vertical stress with depth using the theory o f elasticity on a semi-infinite m edium and the assumption that the tire contact pressure is uniformly distributed. However. N'„ can be defined with an equation sim ilar to (4. different tire inflation pressures produce different stress distributions near the pavem ent surface but the stresses deeper in the pavement are virtually identical. Figure 4.3) different equivalence coefficients are assumed for single and tandem axles. For example.13) should provide an adequate approxim ation to the mis described in the next chapter. this assum ption may not be realistic. As noted before. On the other hand. This issue is revisited in C hapter 6 . For a given load.3) are questionable here. a higher total load increases the vertical stresses below the surface. This is reasonable for the underlying layers because for tandem axles the pressure bulbs o f each axle in the configuration overlap at the depths corresponding to these layers. Since the design air voids for the mixes in the AASHO Road Test was relatively low. in equation (4. some aspects o f (4. The state o f stresses at a point on a pavement structure caused by a load at the pavement surface depends upon the applied pressure and the total load. for rutting originating in the asphalt concrete R ep ro d u ced with p erm ission o f the copyright ow ner. However.

2 0. Most o f the rutting occurs near the surface.500 KPa 20 kN = 700 KPa -A— 40 kN .1 0.3 0. r AL\. where the pressure bulbs o f the axles in a tandem axle configuration do not overlap. Therefore. (4.700 kPa Figure 4. Tire Vertical Stress (kPa) at the centerline 100 200 300 400 500 600 700 800 0. the loading in the asphalt concrete layer is specified as follows: N'„ = 1 > K „ f L ) SAL) 2 + R. Further reproduction prohibited w ithout p erm ission. Thus.14) .58 layer. this assumption is unrealistic.4 o. < D 0.0 0.3: Effect o f wheel load and tire inflation pressure on the vertical stress distribution assuming a sem i-infinite elastic material. a tandem axle acts in effect as two single axles each carrying h alf the load o f the tandem axle.500 kPa -40 kN .5 . l 2 ' fia 2 v SAL j R ep ro d u ced with p erm ission o f the copyright ow ner.• — 20 kN . V '2 \ SAL j + _ ( A L 2 .

For convenience the equation is summarized below: RD„„ = 5 .3 Com plete Model Specification The specification o f the com plete model that accounts for rutting originating in the underlying layers as well as in the asphalt concrete layer is given by the sum o f the models presented in the previous two subsections. This parameter. a.12). the result is multiplied by 2 since there are two single axles with half the load for each tandem axle.A/"AN is + a .15) . Further reproduction prohibited w ithout p erm ission .14) AL2 is divided by 2 to obtain the load in each o f the single axles in the tandem configuration.( 1 0 0 ° ) e P ^ J =1 ( ^ ( P i o + P u T em pD um .i4 + ^ ^ ° 00' e p. which determines the equivalencies between axle loads.59 where all the variables are the sam e as for equation (4. is expected to be significantly sm aller than either (3j or /?<$ o f equation (4.s AN]s J=I p 4 e~ (P|7.11) and (4.3) and where (3/? is a param eter to be estimated.)e^n N“ A N is J =l where R ep ro d u ced with p erm ission of the copyright ow ner. (4. the sum o f equations (4.=+p3W e P. m ls <?Pl' N. 4. Then the load is standardized by the standard single axle load o f 80 kN and the resulting ratio is raised to the load equivalence coefficient for single axles. e S= 1 = P. Finally.' +Pj7.3). That is. Notice that in equation (4.

The rest o f the sections provided information about the pavem ent behavior during the first spring-thaw b u t o f these.and all the other variables are as defined in the previous two sections. However. the inform ation available for the original sections o f the AASHO Road Test consists o f a heavily unbalanced panel data set. only about 50% (25% o f the total number o f sections) also provided inform ation about the pavem ent behavior during the second spring-thaw. 4. This w ould not be much o f a problem if the num ber o f observations for a given section were determ ined exogenously. M ost o f the pavement failures occurred during the spring m onths due to thawing. Further reproduction prohibited w ithout p erm ission. However. About 50% o f the sections provided no information or very little information about the pavem ent behavior during the first spring-thaw. the num ber o f observations per section is likely to have been determined by with p erm ission o f th e copyright ow ner.4 In c o rp o ratio n of overlays in the model specification Trafficking o f the sections at the AASHO Road Test lasted for about two years. Because o f the above. . many o f the sections did not survive that long.

09 m (3.) to 0. it is very difficult to discern w hether the second jum p is sm aller because there is hardening o f the materials in the underlying layers or because the thawing index during the second thawing period was sm aller than the index during the first thawing period (Figure 4. The reason for this exaggeration may be that there is an overrepresentation o f early life behavior in the data set for the original sections. this result was considered suspect. In addition. Further reproduction prohibited w ithout p erm ission. using the param eters obtained from the sample for the original pavem ent sections and with the m odifications shown below. M any o f the sections that failed early were overlaid with asphalt concrete layers between 0. roughness. The application o f equation (4. This causes the problem o f self­ selection that may result in biased param eter estimates. the number of observations per section is determined endogenously.15) to the overlaid sections. most o f the evidence from laboratory experim ents indicates that all materials harden to a certain extent with loading. R ep ro d u ced with p erm ission o f the copyright ow ner. .5 in. etc. Despite the high t-statistics obtained and the fact that the observed and predicted rut depth values were generally close. Unfortunately. many o f the sections that survived the two spring-thaw periods showed evidence o f a lower rut depth jum p during the second spring-thaw than the jum p during the first spring-thaw.15). Estim ation o f the param eters in (4.61 the levels o f cracking. indicated no hardening o f the underlying layers (flu = 0 )1. using the data set for the original sections.05 m (2 in.2). at least in part. Thus.). rutting. First. showed that the rut depth ju m p during the second spring-thaw was somewhat exaggerated by the model.

after the overlay construction. Nowadays. the incorporation o f the effects o f overlays in the model also has considerable practical importance. Only a few m odifications are necessary to incorporate overlays in the model. was constrained to be R ep ro d u ced with p erm ission o f the copyright ow ner. the summation in the equation that defines N it must include all periods since the original construction o f the pavem ent section.62 By incorporating the overlay information. the underlying layers have already been hardened by the traffic applied up to that point. the asphalt concrete thickness in the underlying layers sub-model is increased by the thickness o f the overlay. First. N ote that when an overlay is applied. In order to account for this. During estimation. the parameter greater than or equal to zero. Further reproduction prohibited w ithout p erm ission. the summations o f rut depth increments (in both sub-models) m ust include only the periods after the overlay construction since most o f the previous 1 These results are shown in Appendix B. . the rutting originating in the asphalt concrete layer will occur mostly within the asphalt concrete overlay. In contrast. N 'it m ust include only the periods after the overlay construction. the consideration o f overlays is im portant even for new designs since it allow s the determ ination o f econom ically optim al design and rehabilitation strategies at least for rutting as a performance criterion. Similarly. the incorporation o f overlays in the model greatly increases its scope o f application. Thus. a larger and larger proportion o f the total expenditure on highway im provem ents is being spent on maintenance and rehabilitation rather than on new construction. This is reasonable since an overlay will have the effect o f reducing the stresses in the underlying layers and consequently the rutting originating deeper in the pavem ent structure. Beside the potential importance for the estimation o f the model parameters. In addition. the unbalance in the panel data set is reduced. Thus.

(4.: + p .18) .3 ) e B f Tl' ] 10 00 J g (5.V„ ^ s=k (4.63 rutting is eliminated by the overlay. +O T „ ) + p . Further reproduction prohibited w ithout perm ission.(P. OT„ = sum o f the thicknesses o f overlays applied to section i up to time period r. the final model specification. (7]. \ Pi: v SAL j All. T. g .x A N ls s. which includes the effects o f overlays.17) (4.u■ A fter incorporating the above m odifications.16) t (PlO + Pl Tem pDum .„ + ^ P . t ( F L ? 12 > AV<s ■Z‘ kSALj + R.k w here index o f the period in which the last overlay was applied. Actually. it is assumed that im m ediately after the overlay construction there is a rut depth equal to (5.) e Pl3 s. is: t = P . +2 f A L 2 >P|J v 2 -SALj s =k R ep ro d u ced with p erm ission o f th e copyright ow ner. or the index corresponding to the original construction if no overlay was ever applied. 'ALL'*' SA L ' Pe v^P7 SA L j AL\. £ AK ( F L ^ ySAL j + R. T. .

Further reproduction prohibited w ithout p erm ission.. P) + e„ i=l..19) where T.5 M odel estim ation results Equation (4. . the vector X„ contains the whole history o f loading through the AH’s.i. .16) this model is nonlinear in the variables and the parameters...64 The above specification is general enough to incorporate pavement sections that are overlaid more than once... the sections were overlaid at the m ost once.\Xj. several m ethods o f pooling the data can be used.. . O ne could estim ate separate cross-section regressions (each using observations for R ep ro d u ced with p erm ission o f the copyright ow ner. T .. /7/j P m ..16) is the expression o f the conditional expectation function o f rut depth for section i at time /. .y corresponding to the tlh observation for section / and on the vector o f parameters /? = (/?/. (4. . however. 4.S. P u s )' where S is the total num ber o f sections. As can be seen from equation (4. E(RDi... t = l .. . W hen a data set consists o f observations for different pavem ent units through time. Such data sets are know n as panel data sets. The model can be expressed as the following set o f regression equations: RD„ = E (R D jt \X it... is the num ber o f observations for section i and e„ is the error term which is assum ed to have m ean 0 and constant variance crE2.fi). . AV. Moreover.. At the AASHO Road Test. All these factors make the estim ation o f the model complex.FLhR hA L l „AL2„TI. This function gives expected rut depth conditional on the set o f regressors A". = { { J u J ^ T ^ A V .-.

..19) with E(RD„\Xlh^ ) given by equation (4. Layer com paction can also influence the intercept term in a m ore subtle way. Exam ples o f unobserved heterogeneity are the initial cross-section profile and layer com paction. afterward.65 different pavem ent sections at the sam e point in time) or separate tim e-series regressions (each with observations for a single pavem ent section over time). This is the case if all observations are the result o f a single underlying deterioration process. if the model param eters are constant over time and over cross-sectional units. estimates with lower variance) can be obtained if all the data are com bined and a single regression is run. som e unobserved heterogeneity (unobserved and persistent pavem ent-specific factors) is still expected am ong different pavem ent sections. The form er directly influences the intercept term in the m odel. These are examples o f the kind o f unobserved heterogeneity that will be accounted for in the model estimation.u = p u is the sam e for all The problem with this procedure is that despite the reasonableness o f the assum ption that all the observations are the result o f a single underlying process. Further reproduction prohibited w ithout perm ission. the conditions are similar to the ones that w ould have been obtained with good com paction. In the present context.e. a layer that has not been adequately com pacted will density rapidly w ith the first traffic loads on the w heel paths. this w ould mean to perform a regression using equation (4. then m ore efficient param eter estim ates (i. H owever. R ep ro d u ced with p erm ission o f th e copyright ow ner. This effect will show up mostly in the intercept term. For example.16) and assuming that p . The sim plest technique is to com bine all cross-section data and time series data and perform ordinary least-squares regression on the entire data set.

Since the inclusion o f different constant terms (P . the num ber o f sections. 1997).i-t) is taken to be constant over time and specific to the individual pavem ent section /. This approach produces consistent results (i. + e (. Specifically.e. Both approaches assume that the unobserved heterogeneity can be captured through the constant term. 3 ) e Ps 1 1 0 0 0 J e P .( P i (Tn + O T „ ) + p j Tt l + p 3 r . where w.u ) represents a lack o f knowledge about the model. The two m ost widely used frameworks for m odeling unobserved heterogeneity are called fixed. An alternative approach is the random effects specification.)ePu N“ AjV^+w.. In the fix e d effects approach. approaches infinity) but it is costly in terms o f the num ber o f degrees o f freedom lost. s=k R ep ro d u ced with p erm ission o f the copyright ow ner. (4. consistent as S . it is natural to view the section specific constant terms as randomly distributed across pavem ent sections. N u ^ s=k +^ ( P 10 + P i 1Tem pD um .66 The advantage o f panel data set over a cross sectional data set is that it allows the researcher greater flexibility in modeling differences in behavior across individual units (Greene. because a different intercept term is required for each pavement section.20) . is a random disturbance characterizing the section and is constant through time w ith mean E(w/) = 0 and constant variance equal to ctu . With these assumptions the random-effects specification is: R D it = Pl4+ ‘ / P 4 e . it is assumed that p /1 4 = P u + «<. the individual effect (P.and random-effects respectively. Further reproduction prohibited w ithout p erm ission.

2000). This gain in efficiency is the reason R ep ro d u ced with p erm ission o f the copyright ow ner.67 Without a priori ways to distinguish between individual sections. The model was estimated for rutting in the outer wheelpath using both the fixed effects and random effects approaches. Therefore.g. it was unnecessary to perform a Haussman specification test to verify if the regressors are uncorrelated w ith the individual effects ut.1 shows the estimation results using the random-effects approach. The estim ation of the m odel param eters here is more com plicated since the model is nonlinear in the variables and the parameters.. special routines had to be program m ed for estimation o f the model. treating the intercepts as random variables is a fam iliar expression o f the researcher’s ignorance (Ruud. However. the approach yields consistent param eter estimates only if the regressors are uncorrelated w ith the individual effects u. . It can be noted that all the t-statistics obtained from the random -effects approach are higher than the corresponding t-statistics obtained with the fixed-effects approach. Table 4.and random-effects approaches.035 for the original sections and 2. Greene. there are different numbers o f observations for different pavem ent sections). This can be tested using a Hausman specification test (Greene.. The parameter estimates obtained using the fixed-effects approach are presented in Appendix B. Thus. The estimation approach for linear models can be found in the literature (e.478 observations (7.443 for overlaid sections) corresponding to 260 pavem ent sections were used. Further reproduction prohibited w ithout p erm ission. and the panel is unbalanced (that is. The parameter estim ates were alm ost identical from both the fixed. 1997). A total o f 9. Details o f the estimation approaches mentioned above are given in Appendix A. 1997).

the asphalt concrete layer is only 1. a 4-power law for load equivalencies.79 indicates that a tandem axle load o f 143. that the first spring-thaw had a greater influence than the second one.220 lbs) has the same effect on rutting as an 80kN (18. the coefficient o f the thawing index.47 (p//p>) times more effective in reducing rutting in the underlying layers than the base layer.000 lbs) single axle load.000 lbs single axle was equivalent to a 32.2 kN (32. Further reproduction prohibited w ithout p erm ission. Further. The coefficient P 7 = 1. for exam ple. . The contribution o f the base is 1. as it was previously observed. assum e that the thaw ing indexes are the sam e for corresponding R ep ro d u ced with p erm ission o f the copyright ow ner.87 and Pg = 3. This illustrates the advantage o f not having presupposed. All but one coefficient are statistically significant at a 5% significance level and they all have the expected signs. The coefficients p^ = 2. According to these results. P? has the expected negative sign and indicates that the underlying materials harden over time. To understand this. This also implies.?) times the contribution o f the subbase. /?j (single axle load) and p& (tandem axle load) are significantly different from each other at a 5% significance level.13 (P.55 are significantly different from 4. which indicates the appropriateness o f not having presupposed that Ps = p^.0 at a 5 % significance level. shows the im portance o f the consideration o f the environm ental effects. The significance o f Ps. This is in agreem ent w ith the assumption made at the AASHO Road Test (HRB.?/p. 1962) that an 18.000 lbs tandem axle.68 why the random-effects approach is preferred over the fixed-effects approach w hen its application is appropriate.

04e-6 -10.51 .0 0 2 3. Thus.69 periods o f two spring-thaws.55 19.s/1000) is the same for both periods.79 83.05 U nderlying layers Tandem axles P6 3. the factor exp(77.51 28.06 ctu 2 = 8.1 .00 LEC* Single axles Ps 2. P a ra m e te r description P a ra m e te r A sy m p to tic E stim ate t-sta tistic P a ra m e te r AC layer resistance coefficient P i 7.2 0 e .87 22. However.33 Intercept Pn cte 2 = 4.89e-l 11.6 -24.46e-7 3.80 Ratio o f tandem /single standard axle loads P i 1.15 A C deformability coefficient Pio 1.32 H ardening o f the asphalt concrete layer Pn . 6.Asphalt concrete layer P i 0 .18e-5 27.93 Subgrade deformability coefficient Pa 8.1: Parameter estimates o f the A ASH O model obtained using the random -effects approach. the factor exp ( / ? 9 N IS) is sm aller for the second thaw period because N ls is greater and fig is negative.23 1.59 H ardening o f the underlying layer materials P9 -1.24 Subbase resistance coefficient Pi 4. T ab le 4. Further reproduction prohibited w ithout p erm ission .49 15.61 Tem pDum coefficient P n 0 .0 0 LEC* .42 N um ber o f observations = 9.82 Thaw ing index coefficient Ps 4.33 Base resistance coefficient P2 5.478 * Load Equivalence Coefficient R ep ro d u ced with p erm ission o f th e copyright ow ner.78e-3 24.10 28.

indicating that the asphalt concrete mixes at the AASHO Road Test were not severely affected by high tem peratures. w hich captures the hardening occurring in the asphalt concrete layer.70 During estim ation. The estim ated value of /?/? (0. particularly for the pavements w ith thick asphalt concrete layers that are now com m only used. the rut depth cannot decrease with loading. Given the design characteristics o f these mixes. It m ust be noted that this coefficient m ay capture in part the effects o f tire inflation pressures. near the pavem ent surface the effect o f the load is not as im portant as it is at greater depths. this result seems reasonable.2. Further reproduction prohibited w ithout p erm ission. (3/j. also has the expected negative sign and it is o f the same order o f m agnitude as fig. N ote that a value o f /?/? equal to zero would indicate that each axle passage would induce the same increase in the rut depth originating in the asphalt concrete layer regardless o f the weight o f the axle. The constraint on /?// is binding. . Finally. the coefficients fiio and fin were constrained to be greater or equal to zero since under normal conditions. Since /?/> is greater than zero. A discussion o f why this is so is delayed until Chapter 6 after the W esTrack model specification is introduced. This is surprising since the literature on material testing indicates that asphalt concrete mixes harden faster than unbound layers. That is. the load plays a factor but it is o f m uch less significance than usually assumed. R ep ro d u ced with p erm ission of the copyright ow ner. which are positively correlated w ith the axle loads.39) agrees w ith the analysis presented in Section 4. This result could have im portant implications for the allocation o f cost responsibilities to different vehicle classes.

is close to the accuracy with which rut depth was measured at the AASHO Road Test.23 indicate that the individual effects produce more than 50 % o f the variance. This is shown in the figures in Appendix C. The result is even better in a pavement m anagem ent context where the random effects are less im portant since previous observations o f rut depth are used to predict the future observations. The axle loads are expressed in kN (KiloNewtons) and S o r T indicate that the pavem ent was loaded w ith a single or tandem axle respectively. Further reproduction prohibited w ithout p erm ission. The fits with the parameters obtained in this chapter are slightly better. The section in the top part o f Figure 4. w hich show the observed and predicted rut depth over time for all the sections in the estimation sam ple2. This shows that the size o f the unobserved heterogeneity is significant. The values after the word "D esign” in the section titles indicate the thicknesses in meters o f the asphalt concrete.1 mm. The estim ated standard error o f the regression ( a/c t 2 +<t 2 ). this was generally the case. 4. As can be observed in the figure the pavement behavior o f these sections is replicated quite well even for the overlaid sections. In this case.4 shows a comparison o f the predicted rut depths to the observed rut depths for two o f the sections in the estimation sample. the estim ate o f ctc is more relevant which is only 2. base. and subbase layers. 8. Figure 4.6 mm. As 2 The figures in Appendix C were drawn using the parameter estimates obtained from the joint estimation described in Chapter 6. 3. respectively. . With some exceptions. the thickness o f the asphalt concrete overlay plus a l+” sign appears before the original asphalt concrete thickness.4 survived the w hole experim ent whereas the section in bottom part was overlaid after the first springthaw period.71 The estimates o f cru2.42 and o f cte2. For overlaid sections. R ep ro d u ced with p erm ission o f the copyright ow ner.

.1 5 -0 ......... Figure 4..................S • • .2 0 A x le loads: 8 0 ............6 shows two exam ples for these sections....................3 0 A x le loads: 80 . N ov Jan M ar M ay Jul Sep N ov Jan M ar M ay Jul Sep Nov D a te 4 0 -------- - Section: 587 D esign: 0 ....1 3 -0 ....0 0 -0 ..... more than 28% o f the predicted values are within 1 mm o f the observed values.... and 89% w ithin 5 mm............5........72 illustrated in Figure 4.. Further reproduction prohibited w ithout p erm ission.... R ep ro d u ced with p erm ission o f the copyright ow ner....... time for tw o sections used to estimate the model parameters. The model replication o f the pavement behavior was further confirm ed by a prediction test with a set o f pavem ents not used for estimation...................... 54% are within 2 mm .. . 72% within 3 mm.0 8 + 0 ........S ............................. 83% within 4 m m ... ■ • • ■ : O b s e rv e d : P re d ic te d N ov Jan M ar M ay Jul Sep N ov Jan M ar M ay Jul Sep Nov D a te F ig u re 4.....4: Observed and predicted rut depth vs......1 3 -0 ........................ 3 5 ..... 4 0 Section: 581 D esig n: 0 ..

in addition to the random variations o f the intercepts. the extension o f the model to include the effects o f overlays greatly expands its scope o f application. the increase in variance with loading only creates a problem o f inefficiency in the param eter estimates and because it is relatively small. /?:. this approach is also m ore complicated. As already mentioned. Unfortunately. there may be differences between the compaction energy used for the different sections leading to variations in fit. Further reproduction prohibited w ithout p erm ission . In this m odeling framework. It should be noted. and fis. Precise estimates for these param eters are considered im portant because o f the short period o f time during which the experim ent R ep ro d u ced with p erm ission o f the copyright ow ner. a mixed-effects (random coefficients) approach is more appropriate. how ever. the increase in variance is to be expected because the unobserved pavem ent sections’ heterogeneity may m ost likely manifest not only through the intercepts but through the other model param eters as well. A nother major consequence o f including the inform ation for overlays is the different estim ates o f the param eters representing the hardening o f the m aterials {fig and fin). 1995). som e o f the other param eters can also be assumed to vary randomly between pavem ent sections (D avidian and Giltinan. it must be noted that the value o f cyu 2 seems to be overestim ated because it is capturing m ost o f the increase in variance that occurs with increasing loading. this issue is left for future research. leading to variations in fiio. that the residuals from some sections indicate that the variance increased slightly w ith loading and/or thaw ing periods. Since. Actually. .73 The above results indicate that the model assumptions seem to be generally valid. For exam ple. In order to model these unobserved differences between pavement sections. which leads to som e estimation inefficiency. or in mix characteristics such as aggregate gradation. However.

it is essential to model material hardening correctly. for the set o f parameters obtained w ith the overlay information. . These two exam ples are representative o f w hat happened for m ost sections. The estim ated rut depths are reasonable with both sets o f param eters. for the set o f param eters obtained without the overlay information. the rut depth jum p during the second thaw period seems to be underestim ated for the section that was not overlaid. 100 90 80 70 60 50 40 20 10 0 0 2 3 4 5 6 [Predicted Rut Depth .7 show s the observed rut depths over time for a section that survived the whole experim ent and for an overlaid section. R ep ro d u ced with p erm ission o f the copyright ow ner. However. Figure 4.5: Com parison o f observed and predicted values. The predicted rut depths are shown with the param eters obtained with and without the overlay information. the rut depth jum p during the second thaw period seems to be overestim ated for the overlaid section. Further reproduction prohibited w ithout p erm ission.74 took place. In contrast.Observed Rut Depth| F igure 4. If one wants to predict the pavem ent behavior over longer periods.

The reason why the overlay inform ation strongly influences the estimated values o f the hardening parameters may be that the information in the original sample may contain an overrepresentation o f early life.2 3 -0 . R ep ro d u ced with p erm ission o f the copyright ow ner. Remember that only 25 % o f the sections provided inform ation about the second spring-thaw period. time for two sections not used to estim ate the model parameters.2 0 A x le loads: 2 1 4 . .2 3 -0 . the addition o f the overlay inform ation may have biased the param eter estimates toward the stronger pavements.6: Observed and predicted rut depth vs.75 S ection 271 Design: 0 .2 0 A x le loads: 133 .§ 2 5 f 20 • <D ■O 1 5 4 K 10 -*-r- t ■ t.1 5 -0 .1 —^ 0 — Nov Jan M ar M ay Jul S ep Nov Jan M ar M ay Jul S ep Nov D a te S ection: 2 7 2 Design: 0 .1 5 -0 . Further reproduction prohibited w ithout p erm ission.T Observed P redicted t 25 Q.S 40 35- E 30’ . 20 | ~ TJ 15 0£ 10 Jan M ar M ay Jul Sep N ov Jan M ar M ay Jul Sep Nov D a te F ig u re 4. In contrast.

.....7: Predicted rut depth vs..2 0 A x le loads: 133 ..........w ith overlay information 30 E E 25 ■ P redicted ... ' ....... These observations may indicate that the “true” values o f the hardening param eters may be som ew here in between the estimated values with and w ithout the overlay inform ation..S 40 35 ■ • Observed P redicted .. . ~= 20 ■o 15 cc 9 ' «• 10 N ov Jan M ar M ay Jul Sep Nov Jan M ar M ay Jul D a te F ig u re 4..1 5 -0 ....S 40 35 N ov Jan M ar M ay Jul S ep Nov Jan M ar M ay Jul Sep N ov Sep N ov D a te Section: 3 0 3 Design: 0 ... Further reproduction prohibited w ithout p erm ission. R ep ro d u ced with p erm ission o f the copyright ow ner....1 5 -0 ....w ithout overlay inform ation ..76 S e c tio n 257 Design: 0 ....... time using the param eter estimates obtained with and w ithout the information for overlaid sections.....0 8 + 0 ..1 0 -0 ....3 0 A x le loads: 133 .. This issue is explored further in Chapter 6 ....1 5 -0 ...

particularly. However. • No restrictions have been imposed on the values o f the param eters representing equivalencies between axle loads or the param eters representing pavem ent layer resistance. This may allow the determ ination o f strategies for the original construction and rehabilitation for optim al rutting perform ance. In particular. . The model specification uses concepts that are familiar to pavement engineers such as load equivalencies and structural coefficients. 0.39. The estimated load equivalence coefficient for rutting in the asphalt concrete layer.6. Further reproduction prohibited w ithout p erm ission . the model presented in this chapter has som e unique features sum m arized below: • Tw o different mechanisms o f rutting are considered: rutting in the underlying layers and in the asphalt concrete layer.77 4. In such pavem ent R ep ro d u ced with p erm ission o f the copyright ow ner. for pavements with thick asphalt concrete layers. A non-linear model in the variables and the param eters was specified and estimated. the equivalence coefficients for axle loads are not constrained to be equal for rutting originating in the asphalt concrete layer and for rutting originating in the underlying layers. This may have im portant im plications for the allocation o f cost responsibilities to different vehicles classes. is significantly different from w hat is usually assumed in pavem ent engineering for the calculation o f standard axle loads (usually a value close to 4). Summary This chapter has presented the specification and estimation o f a pavem ent rutting progression model using the AASHO Road Test data set.

especially considering the num ber o f sections and observations that were used for their estim ation. • The model captures the effect o f overlays. but for the mixes at the AASHO Road Test this was not an issue. • The model predicts rut depth increm ents for each period. it can be used to determine an econom ically optimal construction and rehabilitation strategy using a certain level o f rutting as a constraint. • The model includes a thawing index variable that captures the effects o f the environmental factors at the A A SH O Road Test. The results showed that the size o f the unobserved heterogeneity was significant. Both fixed-effects and random-effects specifications were used to account for unobserved heterogeneity. the use o f a 4th pow er for the load equivalence coefficient may overestim ate the effect o f heavy loads on rutting originating in the asphalt concrete layer and underestim ate the effect o f lighter loads. R ep ro d u ced with p erm ission o f th e copyright ow ner. and accounts for the hardening o f the underlying materials with traffic before the overlay construction. T he model fits were good.78 structures. . a prediction test with a set o f pavem ents not used for estim ation confirm ed that the model replicates well the pavem ent behavior at the AASHO Road Test. Finally. An allow ance is also made for the effects o f high temperatures on the asphalt concrete m ixes. w hich is particularly advantageous in a pavement m anagem ent context. Further reproduction prohibited w ithout perm ission. Therefore.

the model has the following limitations. First. Several factors affecting the rutting performance o f asphalt concrete mixes. such as asphalt content.79 Despite the significant estimation results in this chapter. For sections with slow moving traffic such as climbing lanes and urban highways with recurring congestion. the model is limited to the m aterials and tire pressures used at the AASHO Road Test. . Still. there are changes in the materials characteristics o f in-service pavement sections due to aging that are not captured by the model because o f the relatively short period o f time in which the experim ent was performed. the model may underestimate rutting because it does not consider the creep that may occur in such situations. In addition. the model is limited to the base and sub-base materials used at the AASHO Road Test. in Chapter 6 . However. the experim ental data may not represent the true deterioration mechanism o f in-service pavem ents because o f differences in factors such as traffic wander. the model is adapted to overcom e this limitation for asphalt concrete mixes. and m aterial aging. are not accounted for. R ep ro d u ced with p erm ission o f the copyright ow ner. Finally. traffic speed. Further reproduction prohibited w ithout p erm ission.

1 illustrates the target gradations on a 0. Figure 5.1 describes the observed behavior o f the different pavem ent sections (different asphalt concrete m ixes) at W esTrack and relates it to some factors affecting the mix rutting perform ance.2).45 pow er chart (an explanation o f this chart is given in Section 5. Further reproduction prohibited w ithout perm ission. The fine and fine-plus gradations are very sim ilar and so was their behavior.2 introduces a gradation index. 5. Section 5. Section 5.1 M ixture Performance at W esTrack This section describes the observed behavior o f the different pavement sections (different asphalt concrete mixes) at WesTrack. Section 5. and coarse.4 describes the parameter estim ation results and the chapter concludes with a sum m ary in Section 5. The model accounts for rutting originating only in the asphalt concrete layer.3. As confirm ed by trenching studies. fine-plus. The chapter is organized as follows. the fine graded mixtures provided the best perform ance followed by the fine-plus graded R ep ro d u ced with p erm ission o f th e copyright ow ner.80 Chapter 5: Rutting Model for the WesTrack Road Test Data Set This chapter presents the specification and estim ation o f a pavement rutting progression model using the W esTrack Road Test data set. which plays a fundamental role in the model specification presented in Section 5. The results o f the analysis in this section provide a basis for the model specification o f Section 5. The mixes for these three gradation levels were termed fine. C ontrary to w hat the designers had anticipated. . A ggregate G radation: The experimental design in W esTrack included three aggregate gradation levels.3. this was practically the only source o f rutting in WesTrack.5.

R ep ro d u ced with p erm ission o f the copyright ow ner. all 8 o f the coarse mixes had failed from excessive shear deform ations.81 m ixtures. A nderson and Bahia (1997) found sim ilar results with laboratory experiments. and coarse W esTrack mixes. 3 o f the 9 fine-plus mixes had failed.000 vehicle passages. The highest perm anent shear strains were found for the coarse mixes at elevated tem peratures indicating that they are less resistant to rutting. The coarse graded mixtures had the worst performance. After approximately 267. 100 00 c ’c<7/a5 a CL © 00 SO Os rn ri n Sieve Size (mm) i Fine Fine P l u s Coarse Figure 5. They determ ined the final shear strain in repeated shear at constant height tests. . The fine m ixes continued to perform satisfactorily until the end o f the test. fine-plus. and none o f the fine mixes had failed.1: Target gradation for fine. Further reproduction prohibited w ithout perm ission. They reported that these trends w ere inconsistent w ith their prior expectations.

Brosseaud et. With only two exceptions. For a given stability range. . For R eproduced with permission of the copyright owner. Air voids: G iven the aggregate gradation and asphalt content. (1993) concluded from laboratory experiments that there exists a critical level o f filling o f the voids by the binder beyond which the material becom es unstable. Results from laboratory experim ents indicate that for air voids below 3% to 2% the stability (rutting resistance) o f m any m ixes will decrease substantially. Further reproduction prohibited without permission. This mix category does not seem to be affected by asphalt content w ithin the range o f asphalt contents used in the experiment. Figure 5.2 illustrates the rut depths for the 26 mixes vs. These observations are consistent with the pavem ent behavior at WesTrack.82 Asphalt content: For good rutting performance the quality and the amount o f asphalt in the mix is important. 1997).5%. The role o f the asphalt binder is to bind the aggregate skeleton and provide sufficient flexibility for durability (Anderson and Bahia. al. Since only one type o f asphalt was used in W esTrack. there is a clear increase in rut depths when the asphalt content increases above 5. For example. it is usually accepted that the rate o f rutting will be higher for high asphalt contents than for low ones.300 vehicle passages. the initial percent air voids is a function o f construction compaction. The two exceptions are fine mixes. we concentrate on the latter variable. expressed as percent o f the bulk volum e o f the com pacted paving m ixture (AASHTO 1997). Asphalt content is the percent by mass o f asphalt binder in the total m ixture that includes asphalt binder and aggregate (AASHTO 1997). A ir voids is the total volum e o f the sm all pockets o f air between the coated aggregate particles throughout a compacted paving mixture. asphalt content after 130.

2: a F in e -p lu s • C o a rs e Observed rutting vs. As can be observed in that figure. Further reproduction prohibited without permission. whereas for fine mixtures the trend is slightly upward.7 %). Figure 5. asphalt content at WesTrack after 130. 4J Q A 13 20 A 3 Oi ■ ■ 14 IS ■ 4 ■ 13 4.4 %). yet that mix perform ance was satisfactory.3 shows the rut depth after 130.300 vehicle passages. section 18. A dow nw ard trend o f rutting with in-place air voids for coarse and fine-plus mixtures can be observed. w hich has a high asphalt content (6.5 A sphalt C ontent (%) ■ F in e F ig u re 5.22 %) and very low air voids (2.83 exam ple.5 5 6 5. E c ->s • 24 JS. was the m ix that presented the lowest rutting at that point in time. Section 12 also had what would generally be considered as unacceptably low in place air voids (2. The scatter observed R eproduced with permission of the copyright owner. construction air voids. .5 6. Q. Brown (1989) concluded that the m ajor causes o f rutting are excessive asphalt content and low air voids in the asphalt m ixtures.300 vehicle passages vs.

R. He reports that at least h a lf o f these mixes have void contents in excess o f 8 percent. Further reproduction prohibited without permission. 1998). In addition.3: O bserved rutting vs.300 vehicle passages. in-place air voids at W esTrack after 130. R eproduced with permission of the copyright owner. which is the case for many o f the WesTrack sections. 12 16 In Place A ir V oids (% ) F in e a F in e -p lu s • C o a rs e F ig u re 5. . Benson (1996) analyzed more than 120 sam ples from mixes with 19-mm m aximum aggregate size used in five 1988 projects that were acceptable under the California Departm ent o f Transportation (Caltrans) specifications. For exam ple. some researchers do not consider pavements w ith air voids above 8 %. Davis in Epps et al.84 for coarse and fine-plus mixtures indicates that in-place air voids by itself cannot explain asphalt concrete mix behavior. to be properly com pacted (Discussion by Mr. Nevertheless. such high values do occur in practice.

Since A V and VMA are in part. . because. For these samples.4 show s the observed rut depths after 130. neither the VMA nor the VFA are known for the as constructed mixes and therefore their effect cannot be evaluated. as the effective asphalt content increases the air voids content decreases. Further reproduction prohibited without permission. expressed as a percentage o f the total volume o f the specimen (AASHTO 1997). the design number o f gyrations for the design traffic in W esTrack). In fact. a function o f the com paction energy so is the VFA. Two very clear trends (shown with two 1 VMA is the volum e o f intergranular void space between the aggregate panicles o f a compacted paving mixture that includes the air voids and the effective asphalt content. For W esTrack. Here. and the voids in the mineral aggregate ( VMA)1. This variable is interrelated to the air voids content (A V). Figure 5. Reproduced with permission of the copyright owner. VFA is also interrelated to the effective asphalt content (AC ) (w hich accounts for the volum e o f asphalt binder absorbed into the aggregate).85 Voids Filled with A sphalt: Voids filled with asphalt (VFA) is the percentage o f the voids in the mineral aggregate filled w ith asphalt binder (A A SH TO 1997). one can obtain the VFA as ( VMA-A V)/VMA * 100. However. for a given com paction energy.300 vehicle passages vs. w ith know ledge o f the A V and VMA. VFA obtained in the SGC (after the samples w ere subjected to 96 gyrations. Superpave Gyratory Compactor (SGC) tests (A ASHTO 1997) were performed for quality assurance (QA). all four variables were com puted for the top and bottom lift o f the hot m ix asphalt (HMA) concrete. the data for the top lift are used because that is where m ost o f the rutting occurred. These tests were perform ed on materials sampled from transport trucks at the hot plants (Hand 1998).

rut depth increases significantly with VFA. Further reproduction prohibited without permission.4: O bserved rutting vs. Further the rutting values for the fine and fine-plus mixes at any VFA level are lower than the ones observed for the coarse mixes. The R eproduced with permission of the copyright owner. They pointed out that VFA o f less than 75 percent to 80 percent appeared to be necessary to avoid having a mix whose stability is sensitive to asphalt content or VFA (i.86 hand draw n lines) can be observed from this figure. A sim ilar trend is noted for the coarse mixes but the VFA value at which the rutting increases is somewhere between 60 and 70 %. .300 vehicle passages. voids filled w ith asphalt after 130. rut depth remains approxim ately constant up to VFA values o f about 85 %. 40 35 30 • 25 • 25 24 Q. For the fine and fine-plus mixes. a stability that varies by a substantial am ount with a sm all variation in VFA). A schenbrener and M acKean (1994) found sim ilar trends for Hveem stability (an indicator o f the resistance to rutting o f a m ix) with VFA for fine. m edium . 20 • • 15 26 A 19 10 A 2 2 * 14 ■ 41 IS A 12 ^ 5 0 40 50 60 70 80 90 100 V FA (%) ■ Fine a Fine-plus • C oarse F ig u re 5. A fter that. and coarse gradations.e.

maximum density lines. to other unobserved factors such as aggregate surface texture. R eproduced with permission of the copyright owner. Engineers have long been aware that gradation o f the aggregate is one o f the factors that m ust be carefully considered in mix design. their fine mixes performed worse than their coarse mixes for high VFAs. or simply to different responses between a laboratory experiment and an actual pavement. In this research. but there is still disagreement as to what gradations are the most satisfactory. D espite considerable research during several decades. and some associated issues is in order. there is no acceptable aggregate gradation index that indicates the rutting potential o f asphalt concrete mixes.4. Before introducing this index. These differences in response o f coarse and fine mixes may be attributed to differences in the definitions o f what is m eant by “coarse” and “fine” mixes. . Further reproduction prohibited without permission.1. As show n in Figure 5.45 pow er chart. an explanation o f the 0.45 pow er chart provided a good ordering o f the gradations according to the observed rutting performance. it was found that the sum o f square deviations from the m axim um density line in the 0. the aggregate gradation had a substantial effect on the rutting performance observed in WesTrack. this was not the case for WesTrack mixes.87 m ajor difference from their study was that the curves for stability as a function o f VFA for the fine and coarse mixes crossed for high VFAs. 5.2 G rad atio n In d ex As mentioned in section 5. however. Thus.

v axis and the percent passing each sieve size plotted on an arithm etic y axis.e. These mixes usually have good stability (rutting resistance) because o f aggregate interlock but their VMA can be insufficient to accom m odate the proper quantity o f asphalt to make a durable pavement (i.45 power is an em pirically derived value. This chart was developed by Goode and Lufsey (1962). zero percent passing a zero theoretical sieve size. The chart was created as a tool for determ ining proper adjustments in gradations to provide greater or lesser voids in the m ineral aggregate (VMA) in com pacted mixtures.. upward and towards the right to any specific maximum size. the design gradations do not usually follow the m axim um density line.45 pow er plotted on the . represent “maximum density g r a d a t i o n s and they are referred to as “maximum density lines".45 pow er plot o f an H M A ’s aggregate gradation consists o f the sieve sizes raised to the 0. the m axim um density lines drawn as described above have only approximate maximum densities. the m ore susceptible it is to rutting (Goode and Lufsey 1962). a pavem ent w ith long fatigue life). It m ust be noted that since the 0. Further reproduction prohibited without permission. M ixes prepared with aggregates o f maximum density gradation have a minimum volume o f space betw een the aggregate particles. there is evidence to suggest that the farther away is a m ix from the m axim um density line.88 A 0. Reproduced with permission of the copyright owner. Since m ix design is a com prom ise between rutting perform ance and fatigue performance. All straight lines plotted in the chart from the lower left comer. However. .

were excluded from H uber and Shuler’s regressions with this data set. is o f param ount importance since it determines the slope o f the line and the deviations o f the given gradation from it. two gradations may be very sim ilar but the m aximum density o lines associated w ith them (and consequently the deviations from the m aximum density lines) may be significantly different. This may explain some existing problem s with the definition o f maxim um aggregate size in Superpave. One o f the data sets used by Huber and Shuler was the same one used by Goode and Lufsey (1962). the nominal maximum aggregate size is one size larger than the first sieve that retains more than 10 percent o f aggregate. This definition is based on the results o f Huber and Shuler (1992). As illustrated in Figure 5. who found that VMA correlated best with the sum o f absolute distances from the maximum density line when using the above definitions. Unfortunately. the first six gradations. however. Surprisingly. . The maximum density line for gradation 1 is particularly troublesom e. Further reproduction prohibited without permission. Three o f the points corresponding to these six gradations plot very far from their regression line. the maximum aggregate size is one size larger than the nominal maximum aggregate size. In the Superpave m ix design system (Harrigan et al 1994).45 chart. there is no consensus as to how to define that m aximum aggregate size. There is no reason why the m axim um aggregate size (25 m m in the figure) should be so much greater than the actual m axim um (som ew here between R eproduced with permission of the copyright owner.89 Notice that the definition o f the maximum density line given above does not provide guidelines for choosing the m aximum aggregate size from which the line should be drawn for a given gradation. which were used by Goode and Lufsey to develop the rationale for the 0.5 for two hypothetical gradation curves 1 and 2 . In turn. This definition.

g.. The definition is illustrated for tw o hypothetical gradations 3 and 4 in R eproduced with permission of the copyright owner. that gradation is representative o f most o f the gradations in W esTrack. A schenbrener and MacKean 1994). Further reproduction prohibited without permission. no definite conclusion has been reached. a better definition o f the maximum aggregate size associated with a specific gradation curve is needed. the definition given below for the maximum density line is used. In fact. In light o f the above.90 12.5: Problem s with the Superpave definition o f M axim um A ggregate Size. w hich is more in accordance with the m ethod used by the developers o f the chart (Goode and Lufsey 1962).5 and 19 mm). Clearly. . Although other researchers have examined different ways to define the m axim um density line (e.associated with gradation 1 o oo <N Sieve Size (mm) Gradation 1 Gradation 2 Figure 5. N one o f the gradations from Goode and Lufsey’ study used by Huber and S huler’s had this problem. 100 Superpave maximum density line associated with gradation 2 c <u a ■Superpave maximum density line .

First.7 respectively. for the cases in which the problem described above is not present. draw a straight line from the origin to point A and extend it until it intersects the horizontal line corresponding to 100% passing (point B). identify point A as the point on the gradation curve corresponding to the sieve one size sm aller than the smallest sieve with 100 % passing.6: Illustration o f the new maximum aggregate size definition for plotting maximum density lines for a hypothetical gradation curve 3. Then.6 and 5. This definition o f the maximum density line identifies m axim um aggregate sizes that are always relatively close to the actual maximum aggregate size. Otherwise. use the straight line from the origin to point B as the m axim um density line. . Further reproduction prohibited without permission. If point B is to the left o f the point corresponding to the smallest sieve with 100 % passing (point C). o 00 ■ to Sieve Size (mm) F ig u re 5. draw the maximum density line from the origin to point C. This is also convenient since in Goode and Lufsey’s study it was found that the m axim um mix stability was found for gradations a little to the left of the gradation with m inim um VMA. this line is always somewhat to the left o f the m axim um density line as defined in Superpave.91 Figures 5. too Maximum density line associated! with gradation 3 Gradation curve 3 a. R eproduced with permission of the copyright owner. In addition.

Further reproduction prohibited without permission. The gradation index is a m easure o f closeness to the m axim um density line (as defined above).X ‘. The summation is over the Superpave R eproduced with permission of the copyright owner. A gradation index was defined based on the idea that gradations that are farther away from their corresponding maximum density lines have less aggregate interlock and thus are less rutting resistant. .1) where X t is the percent passing sieve / and X .MDL is the percent passing for sieve / corresponding to the m axim um density line. That is.7: Illustration o f the new maximum aggregate size definition for plotting maximum density lines for a hypothetical gradation curve 4. 1DL) 2 ioo frT (5. The gradation index (GI) is defined as 1/100 tim es the sum o f the square o f the differences in percent passing between the actual gradation and the maximum density line corresponding to that gradation.92 100 M a x im u m d e n sity lin e a s s o c ia te d w ith grad ation 4 c < ou tu G ra d a tio n cu rv e 4 U CL o in 00 <N <n 01 Sieve Size (mm) F igure 5. GI =— Y ( X ' .

1. These sieves are: 37.15.5. other things being equal. Table 5. 0. 2. 25. The gradation index seems to give a good characterization o f the aggregate susceptibility to rutting.36.93 standard nest o f sieves.60.5.3 M odel specification This section presents the rationale for the model specification.075 mm. w hich is essentially the same one used in chapter 4 to specify the model for rutting originating in the underlying layers in the AASHO Road Test: RD. The starting point for the model specification is the following equation. 4.18. . R eproduced with permission of the copyright owner. 5. 19.2) where R D it = rut depth for section /' at time t (mm). 9.5. 12. Many o f the concepts used are taken from the specification o f the rutting progression model developed in the previous chapter. This corresponded closely. and 0. 0.75. Further reproduction prohibited without permission. voids filled with asphalt and asphalt content.30. (5. w ith the way the sections rutted.1 shows the gradation index computed for the 26 W esTrack original m ixes as well as their air voids contents. 0. The index is generally smaller for the fine and fine-plus gradations and greater for the coarse gradations.

19 1.7 63.24 6.86 14.59 10.2 5. Further reproduction prohibited without permission.7 6.5 6.6 57.24 14 1.44 (% ) 9.28 8 2.5 84.35 13 1.5 95.9 86.1 5.28 10.8 93.08 4 1.1 5.1 5.43 Coarse R eproduced with permission of the copyright owner.59 6.3 61.1 5.3 62.07 71.0 5.2 77.85 12 S ection GI 1 100.49 7.84 17 2.6 5.40 21 2.3 5.1 5.0 6.50 1.69 2 1.6 6 6 .43 2.3 77.25 2 2 1.80 9.7 87.2 73.2 71.7 4.1 6.2 5.41 2 0 1.2 2 15 1.9 57.2 5.49 11.0 87.00 5.80 7.91 25 4.55 16 1.5 5.2 79.02 ■"V j 1. M ix T ype Fine Fine-plus AVi VFA AC (% ) (% ) 1.64 8.6 4.94 T ab ic 5.52 6.78 24 4.0 6.4 61.49 2.85 8.1 11 1.90 18 1.38 5. .9 45.03 2.8 96.1: Characteristics o f the original WesTrack mixes.02 2.78 5.33 26 5.43 11.2 4.54 6.0 5.67 6 5.01 19 1.7 6 .73 7 4.36 11.47 6.8 5.4 80.6 6.76 5 5.3 68.70 12.6 10 1.2 2 9 1.7 87.3 74.55 23 3.5 5.5 5.

and y7. m .m - rut depth immediately after construction for pavement section i. Since. This equation is used for the same reasons equation (4.3) where AVis = num ber o f vehicle passes on section / during period s. = y7 is again assumed constant for all sections w hile m is assumed to vary w ith pavement characteristics. Further reproduction prohibited without permission. The equation representing loading in W esTrack is as follows: (5. the predictions o f the model w ill be lim ited to rutting originating in the asphalt concrete layer. = a variable representing the cumulative number o f load repetitions applied to pavem ent section / up to time period t (a m ore complete definition is given later).~ functions o f the characteristics o f pavement / such as aggregate gradations and y. The param eter y7. m is assumed to be a function o f the mix characteristics only. .95 N ’i. The specification o f the loading in WesTrack is sim ilar to the one used for the asphalt concrete layer in the model for the AASHO data set. almost 100% o f the rutting is due to permanent deformation o f the asphalt concrete layer (Hand 1998). Clearly. R eproduced with permission of the copyright owner.2) was used in the previous chapter. in WesTrack.

704 kPa) is higher than those used in the AASHO Road T est . 528 kPa for loops 3 to 5 and 563 kPa for loop 6.e.39. 5 = param eter determining the equivalencies between axle loads. Further reproduction prohibited without permission. R eproduced with permission of the copyright owner. 2 2 The tire inflation pressures at the AASHO Road Test were 169 kPa for loop 2 lane 1 . The only differences between equations (5. 3 1 6 kPa for loop 2 lane 2.2) and (4. It must be noted that the actual value o f 5 for W esTrack could be different because the tire inflation pressure used in W esTrack (i.4) Assum ing a given value for the constant (and consequently a value for <5) affects the m agnitude o f the coefficients multiplying N'„ but not the predicted values o f rutting. which was 0. A L l = load in single load axle(s) (89 kN).96 FL = load in front axle o f the truck (53. The value assum ed in the model is transferred from the A AS HO model in the previous chapter {fin). This does not present a problem for the model estimation since N'„ can simply be expressed as: / iV'„ = constant ^ A VLS t=i (5. AL2 = load in tandem load axle (2 x 8 9 = 178 kN).. In other w ords if the constant is multiplied by a factor k.4 kN).14) are due to the different truck configurations used in both tests. Since the loading is identical for all the sections in W esTrack. the estimate o f the param eter that m ultiplies N'„ is divided by k but the rutting predictions are unaffected. . the coefficient S cannot be statistically identified.

97 In W esTrack. b (5. . .._l +m. Further reproduction prohibited without permission.8) where the subscript t is added to mi to indicate that it varies over time. m. high air temperatures played a very important role in the developm ent o f rutting.2) is not suitable for this kind o f adjustment. In order to keep that functional form. so equation (5.2 .. i + m „ A N 'l l e ^ N" (5. Since the high air temperatures alter the materials’ properties. the problem is that equation (5.4 ^ V 80 y _ f8 9 ' +7 80.5) where: ^ 53.. a first order Taylor series approxim ation was used..m. etc. (5. This gave: RD„ * R D + m \ Yl AN „ e r' s" = RD.6) and m.9) J =1 R eproduced with permission of the copyright owner. = m '. the following equation is obtained (5.5) can be modified as follows: RD„ = R D u . RD.7) W ith this new formulation. AN. Substituting successively the values o f RD. can be assumed to vary when the environm ental conditions change. one could try to incorporate its effect in m As in the previous chapter. (5.

. . the specification o f m. After that. rut depth increases significantly w ith VFA. these increments occur mostly during periods w ith high temperatures. In addition. Further reproduction prohibited without permission.vG /. R eproduced with permission of the copyright owner. it was hypothesized that m. for a given gradation. the effects o f tem perature are considered along with VFA. As m entioned before. has two additive com ponents mi„ and one that is a function o f gradation only and another which is a function o f gradation. Based on these observations. evidence in the literature suggests that aggregates with gradations closer to the m axim um density line provide greater rutting resistance. W hen there is adequate compaction o f the asphalt concrete mix and a minimum am ount o f asphalt to hold the particles together. there seems to be an intrinsic resistance o f the aggregate structure that is independent o f the am ount o f asphalt in the mix.10) W hen the VFA in the m ix exceeds a certain threshold (that also seems to depend on gradation). (5.1. rutting increases significantly. The gradation seems to influence the rut depth values for low VFAs. As pointed out in section 5. the threshold at which higher VFAs cause increases in rutting and the rate o f increment with higher VFAs. Thus. In order to express this dependency on the gradation index a linear function o f GI was used: =ri+. Since the asphalt is very temperature susceptible. gradations with sm aller GIs (closer to the m axim um density line) will deform less than gradations with greater GIs.98 To com plete the model specification.t m ust be finalized. rut depth remains approxim ately constant up to certain VFA values. VFA and temperature.

8.) — ~ V IUU y ( / i ) ■f {M eanM axT. and air temperature.11) has a displaced logistic form show n in curve a o f Figure 5. the following form was originally specified for \ w2„ = ( ^ 3 + / 4 (V F A V G I.1 1 ) where M eanM axT. •G l. Those two properties are accounted for with the above specification. Further reproduction prohibited w ithout permission.) (5 . . it was noted that there seems to be a VFA threshold above which the ait depth increases dram atically and that the threshold depends on the gradation. is the m ean maximum air tem perature during period t and J{-) is a function o f M e a n M a x T The first part o f the equation. is intended to capture what happens at very low temperatures.1. However. which is equal to one. VFA. In the discussion o f th at figure in section 5.) } \ (5.) captures the sharp increments o f rut depth with VFA observed in Figure 5.) = 1 + 1 + e x p (a 2 + a 3 M eanM axT . This function can be written as: / ( Me an M a x T. As show n in that figure.4.99 In order to m odel this com plex interaction between gradation. The logistic part o f the factor is intended to capture the increment o f rut depth that occurs with higher air temperatures. The multiplicative function /{■) in the second part o f equation (5.12) The displacement.. (y 3 + /. it is not expected that temperature will play a role for low air tem peratures. as the air R eproduced with permission of the copyright owner.

This is sim ply an indication that all the rutting related to VFA and air tem perature occurs only at high air temperatures. This is represented by the constant values for high temperatures. ) + (Yj + Y* -G I. m « = (Vi +Yi GI. The obtained threshold for the mean maxim um air temperature was 28. In light o f the above two problems. (5. two problems arose during the estim ation phase. the model specification presented a problem o f perfect collinearity.11) and (5. 3 A relation like the one depicted by curve a in Figure 5. but it is probable that the data set is not rich enough to identify it.100 tem perature increases. the first 1 o f equation (5.) x T em p D u m . Since. . the second term in equation (5. Although a threshold type relation was not specified.8 is likely to exist.13) only plays a role for high air temperatures. the model indicated that such a relation exists3. is equal to one if the M eanM axT. The second problem was that the com binations o f a 2 and a 2 produced a curve like b in Figure 5. > 28.6°C.12) is dropped from the specification.13) V 100 y where TempDum. Reproduced with permission of the copyright owner.6°C and 0 otherwise. Although equations (5.8. Further reproduction prohibited without permission. the specification for m„ was simplified as follows: VFA.12) yield a valid specification. Thus. there are two parameters to define one threshold. Therefore. the asphalt in the mix will lose viscosity and the mix will rut more easily. The first was that a\ was several orders o f magnitude larger than 1 and a 2 and a 3 were several orders o f magnitude sm aller than 1. The behavior o f the mix is not expected to change m uch either once the asphalt has becom e too soft.

0 10 20 30 40 50 M ean M a x im u m air tem perature (°C ) C u rv e a " ■ . N evertheless.11).C urve b F ig u re 5. . Indirectly.101 S'S* s 3</) o o B Q. The 96 gyrations are intended to produce the sam e level o f com paction observed after the pavem ent has been subjected to the design traffic. In addition. The model as currently specified produced very good results. The R eproduced with permission of the copyright owner. this factor gives an indication o f the effects o f the levels o f com paction on the mix rutting performance. the m odel already considers the effect o f air voids near the end o f the pavem ent life. one m ore additive term w as added to m„ to account for the effects o f initial in-place air voids because many engineers consider this variable as a main determ inant o f rutting. T his is because the specification o f the model already includes a term for VFA as obtained in the SGC for 96 gyrations and VFA can be expressed as a function o f A V (or effective A C ) and VMA.8: Functional form o f f(-) in equation (5. Further reproduction prohibited without permission.

.13) accounts for the higher rut depth values usually observed for very low air voids.) / r/i + 0 'j + ) V G / l ) 1 100 t a \ . (5.) (5. the mix will be less rutting resistant (implying greater m. The compactable voids are initially considered equal to the initial air voids and they are reduced during each time period as a function o f the applied traffic. Perhaps a value near 2 % is more appropriate. To include this effect o f high initial air voids in the model. C'/. .' = (yi+Y2GI. This is consistent w ith the usually observed reductions in air voids over time. It is also expected this to be a transient effect that is more pronounced im m ediately after construction. a com pactable voids variable (CV) is defined.15) oo with a negative is questionable.'s).14) ■TempDuml + yi CVtl where CV„ = A V t exp(y 9 N lt) The lower limit o f 0 % for the compactable voids (obtained as N lt value for y g ). W henever CV is greater than zero. until C V becomes zero. Further reproduction prohibited without permission. the new specification for m„ is m . However. for higher air voids (say above 4 %) it is expected that a pavem ent with higher initial air voids will tend to com pact more under traffic than a pavem ent with low initial air voids.102 term including VFA in equation (5. Thus. R eproduced with permission of the copyright owner.

AV„. AVti.16b) VFA. + yi AVl exp ( 100 5.TempDnnhY and on the vector o f parameters y = (^/. Further reproduction prohibited without permission. R eproduced with permission of the copyright owner.io + (5.„ y ) :. The model can be expressed as the following set o f regression equations: R D " = E (R D lt |y .4 / 7 N lt) M odel estimation results Equation (5..16) is the conditional expectation function o f rut depth for section / at tim e /..s Y where S is the num ber o f sections.16a) J =l w ith m ls given (5. ■■■Y').. •TempDum.103 N evertheless.AVt. This function gives expected rut depth conditional on the set o f regressors Yu = (1 . and therefore even a 10 or 20% change in a„ has a negligible effect on the predicted rut depth values.y ) .e „ i= (5. the model specification is given by / m ls e yi . exp (b N„) -> 0 for a negative b. yio. Y i o . the predicted values for rutting are not very sensitive to low values o f C V and thus this difference may be ignored4.17) 4 Recall that in equation (5. .v" AN a » r. E(RD„\ Yit. In summary.1 .6) as N„ -> oo.GIuVFAt.

. The fixed and random effects approaches were used to estimate the model parameters.85. is the num ber o f observations for section / and e(/ is the error term w hich is assumed to have m ean 0 and constant variance cte2. The fact that the individual t-statistics for these two parameters are relatively low but at the same tim e the X 2 statistic for the jo in t hypothesis that they are jointly equal to zero is high com pared to the 5% critical value is explained by the high negative correlation (-0. During estim ation. R eproduced with permission of the copyright owner. this model is nonlinear in the variables and the parameters.83) between these two parameters. The estim ated covariance-correlation m atrix for the param eter estim ates is presented in A ppendix B. As in the previous chapter. that ys. are jointly equal to zero produces a x 2 statistic o f 33. Further reproduction prohibited without permission. M oreover. Therefore. and particularly when GI equals zero. As can be observed.2 shows the estim ation results. and /<?. the hypothesis that ys and yg are jointly equal to zero is rejected. Table 5. only the results using the random effects approach are presented. Since the 5 percent critical value from the chi-squared distribution with 2 degrees o f freedom is 5. they yielded very similar param eter estimates. yi and y j vvere constrained to be greater than or equal to zero since there is no reason to expect a reduction in rut depth for any value o f GI.99. It can also be observed that only seven o f the other nine parameters are statistically significant at a 5 % significance level. A s in the previous chapter. the vector Yit contains the whole history o f loading through the AF’s. However. the constraint ys > 0 is binding. the jo in t hypothesis.104 where 7.

yu. only yi and yi play a role in m tt.8 0 a c2 = 2 .1 5 Yj -2. The positive signs o f the coefficients yi and yi. W hen VFA is low or when the m ean maximum temperature is below 28.7 4 e -l 0 .2 6 e-6 3 . Further reproduction prohibited without permission.3 2 Yio 4 .36e-4 11. for different values o f G I as a function o f VFA w hen the mean m axim um temperature is above 28.2 = 7.105 Table 5.4 6 e -6 -1 2 .93 Ys 0 .3 9 Yl 5 . R eproduced with permission of the copyright owner..8 8 e+ 0 5 .3 le -6 -1 . The figure show s the value o f the first two terms o f m.0 0 Y4 1. an asphalt concrete pavem ent will always suffer some rutting that will be more severe for higher gradation indexes.97 Y6 1 0 . indicate that independently o f tem perature.43 N u m b er o f o b servation s = 860 The interpretation o f yi. .0 8 Yi -2 .8 2 Ys 1. y:.0 0 e + 0 0 .9.0 5 e+ 0 11. P a r a m e te r A s y m p to tic E s tim a te t-s ta tis tic P a r a m e te r yi 8 ..6 7 e-6 6.2: Parameter estim ation results for the W esTrack m od el. ys and y6 is done with the help o f Figure 5.25 Ys o .6°C.63e-6 2 .9 8 a . This is represented in the figure by the different constant values for different values o f G /a n d low VFA's.6°C.

The com bination o f yv.5 F igure 5.00E+00 40 50 60 70 80 90 100 VFA — GI = 0. Recall that VFA can be linked to air voids.t for different values o f G I as a function o f VFA when the mean m axim um temperature is above 28. Finally. ys is positive.5 x . yi is statistically significant from zero at a 5 % level and it has the correct sign. . ys and ys causes the first two terms o f m„ to increase rapidly with VFA after a certain value o f VFA.5 -Tii—GI = 2.00E -04 <<5 4-1 O CO | o 5.6C. which as described in the previous section. R eproduced with permission of the copyright owner.0 —o —GI = 7.00E-05 X X Lu ■X X' ■X X *' 0. It should be noted that this result does not indicate that to reduce rutting one should try to obtain an air voids value as small as possible. has the expected sign. That y? is negative indicates that the material hardens over time.9: First two terms o f m. The VFA at which the sudden increase in m i( occurs is higher for low G F s than for high G Fs.GI = 5.106 I. the same loading increment will produce a sm aller increment in rut depth for more trafficked pavements. In other words. Further reproduction prohibited without permission.

This show s that the size o f the unobserved heterogeneity is significant.98 indicate that the individual effects produce m ore than 70 % o f the variance. Finally. O nly section 19 shows a significant overestim ation o f the rut depths. 7. Thus. Figure 5.10 shows a comparison o f the predicted rut depths and the observed rut depths for eight o f the 26 sections in the estim ation sample. fineplus. The estim ates o f cru2. Only two R ep ro d u ced with p erm ission o f the copyright ow ner. for this section the model performed the worst. f t and f t indicate that there is an optimum air voids at w hich rutting is m inim ized. on when those high air tem peratures occur. Further reproduction prohibited w ithout p erm ission. It should be noted that the optim um is dependent on w hether the asphalt concrete mix is subjected to high air temperatures and. because o f the hardening o f the mix. It can be observed that in general the pavem ent behavior is replicated well.73 mm. is within the accuracy with which rut depth can be measured. Actually.107 and as shown in Figure 5. the estimate o f y$ together with estim ates o f ft.43 and o f a e2. . and coarse are represented in the figure. The result is even better in a pavem ent management context where the random effects are less im portant since previous observations o f rut depth are used to predict the future observations. the estim ated standard error o f the regression (-/^ u +<Tt X 3-23 mm. 2. In this case the estim ate o f a E is more relevant which is only 1. fine. The three mix categories.9. high VFA values (and consequently low air voids) will lead at som e point to substantial rutting increm ents. This result is consistent w ith traditional asphalt mix design.

the voids filled w ith asphalt obtained for the construction mix in the Superpave gyratory compactor. 5. it should be noticed that the residuals do increase som ew hat with time (particularly for sections 13 and 19). and the initial in-place air voids. The under-prediction for these two sections is derived from the under-prediction o f the intercept. The three mix properties are a gradation index. However. Further reproduction prohibited w ithout p erm ission. The model also captures the effects o f high air temperatures at W esTrack. Three properties o f the mix are sufficient to model the performance o f the asphalt concrete pavem ent accurately. This result is consistent with traditional asphalt concrete mix design. R ep ro d u ced with p erm ission o f the copyright ow ner. The effect o f high temperature depends on how far the aggregate gradation o f the mix is from its corresponding m axim um density line. A non-linear model in the variables and the parameters was specified and estimated. The above results indicate that the model assumptions seem to be generally valid. the model estim ation results indicate that there is a value for air voids that optim izes rutting performance. Sections 20 and 22 are two good exam ples o f the intercepts’ heterogeneity. . The model specification presents some unique features.108 other sections presented similar problem s but o f lesser magnitude.5 S u m m ary The goal o f this chapter was to develop a model o f pavement rutting from the W esTrack Road Test. In addition. which is obtained from the aggregate gradation. which may lead to some estim ation inefficiency.

Gl = 4 . 400 . 10- 0 100 200 300 400 100 S ection 23: C o a rse .1 40 i------------- .0 Section 18: Fine.30E E Rut depth (mm) 300 Q. Further reproduction prohibited w ithout perm ission.2 .1 R u td e p th (m m ) 40 . VFA= 87.109 S ection 17: Fine. Gl = 1.2. VFA = 8 0 .P redicted . AVi = 10. AVi = 2.24. AVI =6. VFA= 96.7 40 • . Gl = 3.8 4 0 ------------------------------------------------------------------- 200 400 Section 24: C o a rse . . ~ 30E E.----------------------. VFA = 79. T<5U 3 ce 100 200 Day 300 400 0 100 200 300 Day F ig u re 5.4.59. AVi = 6.6.-----------------30 h 0 100 200 300 400 Section 20: Fine-plus.5 4 .3.30 1 E ! £ i 20 T<Ju [ 10 10 0 0 100 200 300 400 0 Rut depth (mm) Sectio n 21: Fine-plus. Gl = 1.7 40 — ------------------. VFA = 66. AVi = 5. VFA = 95. f 20 i T<Ju a.2 4 0 -----------------------------------------------------------j • O b seu ed | -----.8. AVi = 6.36. AVi = 11. tim e for eight W esTrack sections. R ep ro d u ced with p erm ission o f th e copyright ow ner.----------- 30- 100 200 300 400 Section 22: Fine-plus. Gl = 1. VFA = 71. VFA= 74.03 .0 R u td e p th (m m ) 40 E E 20 0 100 200 300 400 S ection 19: F ine-plus. Gl = 1. Gl =2. Gl = 2.49.59.0.10: Observed and predicted rut depth vs.5 .78. AVi = 2.

Results in the literature indicate that some gap-graded aggregates that depart from the maximum density line can still produced mixtures w ith good rutting resistance. since only 26 sections were used in the model estimation. the repetitive shear at constant height test) can be significantly affected by the am ount o f asphalt in the mix. even the gradations that depart considerably from the maximum density line are not affected by VFA values this low. The results showed that the size o f the unobserved heterogeneity was significant. the model has lim itations similar to the ones presented for the AASHO m odel in the previous chapter. the model is lim ited to the materials used at the W esTrack Test. This is particularly advantageous in a pavem ent m anagem ent context where the current rut depth value is known and the interest is in predicting the rut depth value for the next time period (and hence only the change in rut depth).9. to the range Reproduced with permission of the copyright owner.. at least with the W esTrack data set. D espite the significant estimation results in this chapter. However. . Both fixed-effects and random -effects specifications w ere used to account for unobserved heterogeneity. if the tests were perform ed at a VFA value o f say 50 %. the m odel predicts rut depths by adding predicted values o f the increm ent o f rut depth for each time period. As shown in Figure 5.g. and for those m aterials. The model fits very were good.110 Finally. it would be better to use a mechanical test to measure the resistance o f the aggregate structure to rutting. these results should be taken with caution. Ideally. This would not be the case. however. The gradation index defined in this chapter seems to be a good indicator o f the resistance to rutting provided by the aggregate structure. However. Further reproduction prohibited without permission. First. the results from current candidate tests (e.

and m aterial aging. the model can not account for pavem ent rutting due to the deformations in the underlying layers. Second. the experim ental data m ay not represent the true deterioration m echanism o f in-service pavem ents because o f differences in factors such as traffic wander. For example. The first two limitations mentioned above are partly overcom e in the next chapter where a single model is estimated jointly using the AASHO and W esTrack data sets. . Reproduced with permission of the copyright owner.Ill o f values o f the independent variables used in the test. Finally.2. Further reproduction prohibited without permission. th e m odel may be overly optimistic for G F s lower than 1. traffic speed.

Further reproduction prohibited without permission. Feng-Yeu Shyr (1993) has used this m ethodology to develop a rail fatigue model by combining data from a model o f rail fatigue and field data. a model developed from the W esTrack Road Test is limited to the layer thicknesses and axle loads used in that R eproduced with permission of the copyright owner. In the area o f infrastructure management.1 presents the concept and purpose o f jo int estimation. the different types o f data have com plem entary characteristics that can be exploited using joint estimation. Section 6. 6. . In all these cases. This chapter describes how these two m odels can be estimated jointly.1 Joint estimation Joint estimation is a methodology that has been used by Ben-Akiva and M orikaw a (1990) for statistically combining revealed preference and stated preference data in estimating travel demand models. M orikawa.2 analyzes w hether the two data sources are compatible for the application o f this estim ation approach. The chapter is organized as follows: Section 6. the estim ation results are described in Section 6. A model derived only from the AASH O Road Test is limited to the materials used in that test. and Yamada (1991) have applied the technique to combine revealed and stated preference data to model travel mode choice.4. The sam e motivation applies in our context. Ben-Akiva. A fter establishing the compatibility.112 Chapter 6: Joint Estimation of the AASHO and WesTrack Models Chapters 4 and 5 have presented the specification and estim ation o f rutting progression models using the AASHO and the W esTrack data sets. Similarly.3 and the chapter ends with a summary in Section 6.

Thus. the important features o f the m ethodology are: • Bias correction • Efficiency • Identification Bias correction is better explained in the context o f trying to estim ate a model with field and experim ental data. Further reproduction prohibited without permission. the estim ation o f a properly specified model from experimental data would most likely yield biased param eter estim ates (the parameters would be biased when predicting field pavem ent behavior). a correction for bias can be introduced when using the model developed from experim ental data to predict field behavior. As pointed out by Morikawa. from the W esTrack data set. . Ben-Akiva. R eproduced with permission of the copyright owner. the effect o f axle loads and layer thicknesses can be identified but not the effect o f the characteristics o f the asphalt concrete mix. From the AASHO Road Test data set. In contrast. and Yamada (1991). Since one is ultim ately interested in predicting the pavem ent behavior in the field. the effect o f the characteristics o f the asphalt concrete mix can be identified but not the effect o f loading or layer thicknesses. Joint estim ation in this context allows the estimation o f a rutting model that includes bias param eters for the experimental data source. the estimation o f a correctly specified model using field data alone w ould yield unbiased param eter estim ates.113 test. In contrast.

Z B) where E(/?Z)| • ) is the rut depth conditional expectation function. = vector o f explanatory variables unique to the A data source. .1 . a = vector o f model parameters for w*. which has a functional form g(-). Further reproduction prohibited without permission. p s = vectors o f model param eters for. E{RD 4 1x"!. zB = vector o f explanatory variables unique to the B data source. R i f = rut depth measurement for data source B. i f ) E(RD b | xB. such that. wr4) = g 1(p . P'1. y. x \ x B = vector o f explanatory variables shared by both data sources. The following is a b rief description o f the joint estim ation method in the context o f a rutting progression model. x \ a . Identification is related to the com plem entary characteristics o f both data sources already mentioned. By using jo in t estim ation. and n Y = vector o f model parameters for z . x B. trade-offs am ong attributes that are not identifiable from one o f the data sources can be identified from the other data source.114 Statistical efficiency (low er variance o f the param eter estimates) is obtained because the param eters are estim ated from all the available data. z B) = g B (p fl. Define R D [ = rut depth measurement for data source A. The two data sources are labeled A and B. R eproduced with permission of the copyright owner.v'* and respectively.

the param eter estim ation is always the result o f the optim ization o f some objective function. The bias correction is achieved through the P*’s and efficiency is gained w ith the use o f m ore information.. = p. . R eproduced with permission of the copyright owner. the objective function is the log-likelihood function and the optim ization is a maximization problem.e. the objective function is a weighed sum o f square residuals and the optim ization is a m inim ization problem.2 Com patibility o f the AASHO and W esTrack data sources Before applying jo in t estimation to the AASHO and WesTrack data sources.115 A ssum e for exam ple that the model param eters obtained from the A data source are unbiased. As discussed in Chapter 3. different m easurem ent technologies were used in both tests. Joint estimation also produces estimates o f the param eters a and y o f the variables not shared by both data sources. 1 < k < K and where K is the number o f elem ents in the vector p. when using maximum likelihood. For exam ple. W hen using G eneralized Least Squares (GLS). in principle. Since. Further reproduction prohibited without permission. it is necessary to exam ine their compatibility. the rut depth was 1 The summation o f objective functions is justified because it is assumed (and reasonable) that the error terms for the two data sources are independent. 6. while some o f the parameters from the B data source are biased. In WesTrack. i. say p* = \ik P* for some k's. The first issue is whether the rut depth m easurements in both tests are compatible. the true param eters p and the scalars (i*’s can be estim ated by forming the joint objective function as the sum o f the objective functions for the individual data sources and optim izing it with respect to all the param eters1.

the parameter indicating the hardening o f the asphalt concrete layer. as pointed out in Chapter 3. In addition.0 m. the values obtained should be relatively sim ilar.0 m.2 m straightedge recom m ended in AASHTO 1997. asphalt grade.116 calculated from cross-section profiles as the peak to valley vertical distance. w hether or not the two data sources are compatible depends on whether the factors affecting the asphalt concrete mix behavior in both road tests are similar.20 m straightedge.) to each side o f the centerline. In addition. the AASHO m easurem ents are also expected to be similar to the ones obtained with a 1. the rutting in this test originated in the asphalt concrete layer and most o f the traffic w ander was confined to ± 12. it can be seen that a necessary condition for the application o f the methodology is that the models for the different data sources have some param eters in common. . Therefore. loads. Some relevant factors are: aggregate gradation. Therefore.5 cm (5 in. which means a cross section coverage o f a dual wheel o f about 1. the separation between the peaks o f the rut is o f the order o f 1. Therefore. despite the different technologies used for the m easurem ent o f rut depths in both tests. As explained in more detail in the next section. R eproduced with permission of the copyright owner. the AASHO and WesTrack models have only one param eter in com m on. traffic w ander and traffic speed. From the explanation o f joint estimation in the previous section. Further reproduction prohibited without permission. This implies that the peak to valley m easurem ent in W esTrack is almost equivalent to the use o f the 1.

This can now be explained by analyzing the average characteristics o f the AASHO m ixes in reference to the W esTrack model.6°C obtained for the W esTrack model was transferred to the AASHO model.117 The aggregates used in both tests consist o f crushed stone aggregates. In that chapter.3. It should be noted. Unfortunately. there is no clear indication that the AASHO mixes were affected by high air tem peratures.3 and VFA between 70 and 80%.. Further reproduction prohibited without permission. Recall that the mean maximum air temperature threshold o f 28. the W esTrack model would indicate no susceptibility to rutting at high air temperatures for the A A SH O mixes. . it was pointed o ut that given the design characteristics o f the A ASH O m ixes.6°C. A lthough this can be a source o f discrepancy. this seemed reasonable. the asphalt binders used in both tests are graded with different systems that are not directly comparable. This was also indicated by the results in Chapter 4. The estimated average gradation index for the AASHO mixes is 2. As can be observed in figure 5. In the models presented in the previous two chapters. the asphalt binder type enters indirectly through the definition o f TempDum.9. however. for this com bination o f gradation index and VFA. that the VFA values for the AASHO R eproduced with permission of the copyright owner. which is within the range for the mixes at W esTrack. the threshold for the AASHO mixes is not expected to be too different from 28. A 100 pen asphalt was used for the AASHO Road Test w hereas a PG64-22 was used in W esTrack. The A ASH O m ixes had a gradation index o f approximately 2. In addition.

none o f them seems to be so influential as to render the deterioration m echanism s o f the asphalt concrete layer in both tests different. there are many other factors such as tire types. Further. the w ander near the end o f the test was about 90% that o f norm al traffic (Hand. 1998). although one test may be biased with respect to the other. From all the factors above. . However. it can be seen that there are some differences between the two tests. they will be captured by the error term assumptions. If all these factors have small influences. However.118 m ixes were not obtained w ith the Superpave Gyratory Compactor but from M arshall tests. as explained in Chapter 5. Finally. 64 kph (40 mph) for W esTrack and 56 kph (35 m ph) for A ASHO. etc. The load magnitudes used in W esTrack (89 kN) fall within the range used in AASH O (9 . although the traffic may have been more channelized than normal. that also influence the deterioration mechanism but cannot be accounted for in the models. the wander pattern was not achieved in a normal fashion and this may have caused some transient effects in the rut depth measurements. In AASHO. Thus. aggregate mineralogy. In addition to the aforem entioned variables. Thus. The target vehicles’ speeds w ere sim ilar in both tests. In WesTrack. traffic w ander in AASHO may be closer to the wander o f actual traffic than the wander in WesTrack. the wander pattern may have been similar to that caused by humans on actual traffic. R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.133 k N f . soldiers drove the trucks and they were asked to follow painted paths. the tire inflation pressures at W esTrack were about 25 % higher than the tire inflation pressures used in AASHO for sim ilar loads.

Further reproduction prohibited without permission.2 ) 2 The load ranges are given for single axles since as explained in the previous tw o chapters for rutting o f the mix tandem axles act as two single axles.3 M odel specification and estim ation results The specification o f the models to be estim ated jointly are the specifications o f the models presented in the previous two chapters. For ease o f reference.119 it is expected that the bias is smaller than the precision with w hich the param eter in com m on between the two models can be estimated. 6. ( 6 . or the index corresponding to the original construction if no overlay was ever applied. From the above observations.k where k = index o f the period in w hich the last overlay was applied. these specifications are repeated below: AASHO model: rd „=p. R eproduced with permission of the copyright owner. it is reasonable to assume that both data sources are compatible for jo in t estimation.14+ y p s-k ( 6 . 1) ^ (Pio + Pn Tem pD um t ) e ^ uSl' AN a s. .

3) s =k W esTrack Model: RD„ (6. In addition. (6. However.6 ) I ALA . \ 100 y (6. ) + 0'j -Cr/. . R eproduced with permission of the copyright owner. 2 (6.5) + r* C K CV„ = A Vt exp(y 9 Af„) ( 6 . Further reproduction prohibited without permission. two param eters are shared by these models. This parameter m easures the hardening o f the asphalt concrete layer. .) •TempDum.S=1 where VF^ m„ = (/i +/ : G/.120 t A " „ = ]> V U Pi2 f F L ^Pi: SAL ' A L 2.8 0 .V „ = ^ A r „ *=! v 80 j v 80 +2 y . The parameter Pi: in the AASHO specification is the same as the param eter S in the W esTrack specification. the parameter fl u in the AASHO model is equivalent to the param eter y? in the W esTrack model.4) ~ Y .2 . only one o f them can be identified from both models.7) and where all the other variables are as defined in the previous two chapters. Given the com patibility o f the two data sources. This param eter can only be identified from the AASHO data source. +2 + R. e T7'V“ . i o + y ^ w .

Joint estimation allows the sim ultaneous identification o f the effects o f layer thicknesses. As pointed out before.121 O f the three m ain features o f jo in t estimation pointed out in Section 6. would equal the ratio o f the previously estim ated values o f /?/j and y7.1 presents the joint estim ation results using the random -effects approach. This m ay result in some param eters having higher standard errors w hen the m odels are estim ated jointly. the estim ation results would be the same as the ones obtained in the previous two chapters. Bias correction is not relevant here because there is only one param eter in com m on that is identifiable in the two models. Thus. Further reproduction prohibited without permission. the m odels used for jo in t estim ation are a constrained version o f the m odels estimated individually in Chapters 4 and 5. only som e points are elaborated here. if a bias param eter /u were introduced for exam ple for P i 3 . asphalt concrete m ix characteristics. . Table 6.1. Thus. the estim ated value o f Pi 3 would equal to the previously estim ated value o f y 7 and jj. This is because the param eters /?/j and y 7 are constrained to have the sam e value and no bias parameters are specified. statistical efficiency is obtained because the param eters are estim ated from all the available data. Many o f the interpretations are similar to the ones given in the previous two chapters. identification is the m ost relevant for this research. high tem peratures. The only difference would be that in the AASHO model. However. R eproduced with permission of the copyright owner. and thawing.

28e-7 -3.26 Yi 4.21 Pi 2 .1: Joint estim ation results. 1 le .6 -22.13 . The consequence was that there seemed to be an overestim ation o f the effects o f thawing R eproduced with permission of the copyright owner.06 0.6 -1.74 16. Further reproduction prohibited without permission.0 0 3.68 61.53 8 0.17e-4 Yi 8.07 2.20 Pi 4.13 Pi Ps Pll Pn Pn Pn c t e2 A ASH O M odel Param eter Asym ptotic = 5.81 0 .56 ctu 2 = 8.26e-3 23.79e-6 6.29 Pi P4 1.99 Yi 0 .2 .48 0.67 Yio 0.122 Table 6. which means that there was no hardening in the underlying layers.76 Yi -2 .77 4.05e-6 2. Parem eter estim ate t-statistic Pi 9.06 8.44 25.45 23.56 12.56 12.00 <ru 2 = 7.18 N um ber o f observations = 9504 cte 2 = 3.28 16. 1 le . As noted in C hapter 4.48e-6 2.8 6 24.01 Ys 2.70 PlO 1.40 23. when the AASHO model was estimated without the overlay inform ation the estim ated value o f fa was zero.71 p9 -2.0 0 0.0 0 0 .6 -22.59 Number o f observations = 860 Total number o f observations = 10364 The first important difference that can be noted with the results from chapters 4 and 5 are the changes in the material hardening parameters fa and f3u (yy).71e-6 Parameter W esTrack M odel Param eter A sym ptotic estim ate t-statistic 77 6.47 Yi Pi 1.68 Y9 -3 .0 0 0 .25 Y4 1.38 4.6 le .70e-5 22.87 10. .06 4.79 3.

47.123 during the second thawing period for the overlaid sections. the estim ated value o f p n is now about ten times larger than the estimated value o f Pg. the hardening o f the underlying layers may have been overestimated and consequently the effects o f thawing during the second thawing period may have been underestimated. It was also noted that when the model was estimated with the overlay information.11. Another im portant benefit is that the information about the rutting o f materials in WesTrack. higher than the previously estimated ratio o f 1. A s expected. the ratio p 2/Pi is only changed from 1. which is also higher. allows a better allocation of how m uch rutting is originating in the underlying layers and how much rutting is originating in the asphalt concrete layer in the AASHO model. w hich is essentially the rutting o f the asphalt concrete layer. The hardening parameters play an im portant role in this allocation.13 to 1. when the AASHO model is estimated jointly with the W esTrack model. . This is again reasonable since the W esTrack data set contains better information about the behavior o f asphalt concrete mixes. but these param eters are linked to p 4.95. The ratio o f the estimated values o f Pi and p 2 is now 1. In this case. A fter joint estimation. Pi are higher than before. with joint estim ation one can expect changes in the estimated effectiveness o f a unit thickness o f asphalt concrete material with respect to the estimated effectiveness o f a unit thickness o f Reproduced with permission of the copyright owner. the estim ated value o f P n = y2 falls within the values estimated in chapters 4 and 5. However. Further reproduction prohibited without permission. The estimated values o f p h p 2. Thus. the estimated value o f fig was alm ost the same as the estimated value o f P n .

6. the m odeled rut depth jum ps seem to be m ore realistic when the parameter estim ates from the jo in t estimation are used instead o f the parameter estimates obtained in Chapter 4. In addition. Since the 5 percent critical value from the chi-squared distribution with 2 degrees o f freedom is 5.6. Figure 6. they are both significantly different from 4. It can be seen that most o f the rutting attributed to the underlying layers occurs during the thawing periods. Further. about 50% o f the rutting is attributed to each source.6. the hypothesis that /3s and f3g are jointly equal to zero is rejected. the joint estim ation results reinforce the im portance o f introducing the initial air voids in the m odel specification. the joint hypothesis that both coefficients are equal to zero produces a x 2 statistic o f 36. is statistically significant from zero at a 5% significance level and yg is close to the critical value o f -1. W ith respect to the W esTrack m odel. However.1 illustrates how much o f the rutting is attributed to the asphalt concrete layer and the total rut depth for the same two AASHO pavement sections shown in Figure 4. The difference between the two lines in the figure represents the rutting attributed to the underlying layers. the relative effectiveness o f the base and subbase materials should be unaffected. .96. Thus.124 the base and subbase materials. by com parison with Figure 4. Again.99. The estim ated values o f and $ 5 are lower than before. Further reproduction prohibited without permission. R eproduced with permission of the copyright owner. For these two sections.

predicted rut depth values for the pavement sections used in the estim ation o f the model. Further reproduction prohibited without permission.S CL ■O 15 £ 10 Nov Jan M ar M ay Jul Sep Nov Jan M ar Jul Nov D ate S e c tio n : 3 0 3 40 35 30 E E. This point deserves particular emphasis because the pavem ent behavior is Reproduced with permission of the copyright owner.S 1 O b s e r ved ------. 25 1 j • ! D esign: 0 . It can be observed in those figures that in m ost cases the model replicates the pavem ent behavior well.1 0 -0 .2 0 A x le loads: 1 3 3 . The figures in Appendix C are for the AASHO sections and the figures in Appendix D are for the W esTrack sections.- rrp | 11 • * i • 1 •• • v. tim e w ith the param eter estim ates from the jo in t estim ation for two A A SH O sections. The figures in Appendixes C and D show the observed vs.3 0 A x le loads: 1 3 3 . N ov Jan M ar M ay Jul Sep Nov Jan M ar M ay Jul Sep N ov D ate F ig u re 6.1 5 -0 .125 S e c tio n : 2 5 7 D esign: 0 .1 5 -0 .1 5 -0 .0 8 + 0 .1: Predicted total rut depth and rut depth originating in the asphalt concrete layer vs.T otal r ut depth | - - Rut de pth on A C lay er ■ f <u 20! ! ■o 15 i £ • 101 . .

Further reproduction prohibited without permission. which for AASHO and in general are positively correlated with the loads. the estimated value o f P n = 0. As can be observed in the figures in A ppendix C. . this load produced a rut depth o f about 5 mm at the end o f the test for most sections. for asphalt concrete mixes o f different characteristics and in two different environments. It m ust be noted that the 0. As m entioned before. As noted earlier. For passenger cars the model may give an upper bound on their effect since these vehicles have a smaller width and smaller tire imprints than trucks. After having estim ated the model jointly from both data sources. The low est loads are usually ignored in the calculations o f ESALs. near the surface the stresses are determined by the tire contact pressures more than by the loads.57 indicates that even small loads produce non-negligible rutting in the asphalt concrete layer.126 replicated well over a w ide range o f loads and pavem ent structures. The lowest end o f the load range deserves particular attention.2 for two AASH O sections with axle loads o f 27 kN. more vehicles passages are likely to be necessary to obtain the sam e coverage produced by trucks. o f the three main features o f jo in t estimation. the following model can be used for the estim ation o f rut depths: R eproduced with permission of the copyright owner. Thus. This can be seen in Figure 6. they produced significant rutting in the asphalt concrete layer (about 70 to 80% o f the rutting is attributed to this source). The lowest load applied at the AASHO Road Test was 9 kN. A lthough the axle loads were too small relative to the pavem ent structures to produce any significant rutting in the underlying layers during the thawing periods. the most relevant in this context is identification. However.57 value most likely accounts for the effects o f tire pressures. a consequence o f the fourth power ordinarily assumed for load equivalencies.

O pened |i . M a y y ~ Jul " j Sep u Mov j Date ■*0 35 i.t RD» = P M+ £ p 4 * -# .: 10- . ...... n:r38Ces^ aoM!M..• • ' ' M ay Jut .S ' ' ' • • .. . 0 *-**▼ * * '**■ No v Jan - Ma r .... 5 fTT * Ja" Ma r Ma y Ju.in i1 £ ..( a . 127 i... or 5f ‘' . ^ '•I°ooJiO09 ^ e S*k e " AN t (6.. £ 25 f- f 20 S 15f .. ' 25 ~ <u ' - - Jul Sap" Nov . .8) ■-r = zXr where all the terms are as previously defined. Total rut depth - .— 35 - 2?s . 0 a . an Mar SeC“° n 746 Design: 0.. .. sap 1 N „ ' Jan ~ ' D ate - - - - ar • .08-0 08-0 . «.. d e p t h o n A C l a y er " i 15- K 10- w u sa m J S 20 2 «... .he seeltens trafficked w ith 27 kN l„ads. m.. : Axle loads: 2 7 .. . I Sep parameter estim ates from ..- R u t . j= n I ' .. . S « .oAxre^ 4 0 . Reproduced with ’ - Nov ~ f tW 0 A AS« 0 . ? 30[ .

128 T he equations presented before for N it and N„' are very specific to the vehicle configurations used in the AASHO and W esTrack road tests.9) -1 R N' iy ll = \S A L j i= l * r =1 + 2 ni 2 SAL ( 6 . Sr = the single axle load representative o f load range r. As pointed out in the previous two chapters. this can be accounted for by redefining N it and Nu as: > K s. the experimental data may not represent the true deterioration mechanism o f in-service pavem ents because o f differences in factors such as traffic w ander. the results are limited to the R eproduced with permission of the copyright owner. Even after using joint estim ation. and material aging. . -J— y { SAL) j= l / r Tr +n P6 p 7 SAL (6. A ssum ing that there are R load ranges. field pavem ent sections are loaded by different vehicle configurations with a distribution o f loads. 10 ) where n nlsr = the number o f single axle loads over pavem ent section i' w ithin load range r during loading period s. In addition. traffic speed. Further reproduction prohibited without permission. the model still has som e limitations. n isr = the num ber o f tandem axle loads over pavem ent section i w ithin load range r during loading period s. and Tr = is the tandem axle load representative o f load range r. However.

w hich is the presence o f a considerable am ount o f w ater in the underlying layers interacting with traffic loads. the asphalt type did not seem to play a role in the rutting o f the AASHO m ixes). Further.129 subgrade. Further. . Since the problem is essentially the same as with thaw ing. it is also limited to asphalt concrete mixes with a nom inal m axim um aggregate size o f 19 m m and to asphalt cements PG64-22. subbase. M ost o f the limitations m entioned above are com mon to all existing em pirical m odels. it could be taken into account in a similar fashion as the thaw ing index. A nother factor not included in the model is the effect o f cracking. Further reproduction prohibited without permission. som e o f these lim itations are not so severe. it allow s the ingress o f w ater from the pavement surface to the underlying layers w ith the consequent detrim ental effects on rutting performance. (Recall that because o f the aggregate characteristics at the AASHO Road Test. the problem o f cracking can be greatly reduced. W hen a pavem ent cracks. In addition. For stronger subgrades. particularly in a pavem ent m anagem ent context where the predictions are limited to one or tw o years. the inform ation after severe cracking occurred at the AASHO road test was lim ited because either the sections were taken out o f the test or they were overlaid. For exam ple. the subgrade material at the A A SH O road test is representative o f weak subgrades that are affected by frost. U nfortunately. one can bound the prediction o f rutting by R eproduced with permission of the copyright owner. It should be noted how ever that the effects o f cracking on rutting performance are most significant only for regions with intense precipitation and poor maintenance. and base materials used at the AASHO and W esTrack road tests. A sim ilar situation occurred in WesTrack. if preventive maintenance such as crack sealing is perform ed timely.

The loading information is also sim ple to obtain for routes w ith weight in m otion (W IM ) scales. 1997) contains the inform ation for thousands o f w eather stations throughout the United States and Canada. Finally. In addition. Nowadays. if no source o f w ater is present and there is no freeze-thaw cycles. the LTPP database (Ostrom et. at least seasonally. an estimate o f rutting in the asphalt concrete mix may be all that is necessary since m ost o f the rutting will occur in that layer. at least in the United States. this is necessary to obtain precise estimates o f rut depth. which is still much sim pler than the com putation o f strains and stresses in a mechanistic-empirical approach. However. To use the model for prediction. to obtain the required environmental information for tw o-w eek periods should pose no problem. an estimate o f the total loading can be prorated accordingly. M ost mechanistic-empirical m odels require at least the sam e level o f detail to have some merit. this model can be program m ed in a spreadsheet. . But even without the detailed information from W IM scales. the asphalt cem ent used in the m ix does not play a role in the rutting perform ance for pavem ent sections in very cold regions. In addition.130 using the com plete specification for an upper bound and the specification for rutting in the asphalt concrete mix for a low er bound. Thus. The interval between observations at the A A SH O Road test was 2 weeks. In WesTrack the interval between observations was variable but on average was also approxim ately 2 weeks. the analyst can usually obtain some information about the traffic distribution. R eproduced with permission of the copyright owner. the user needs to have the loading and environm ental inform ation in two-week periods. ah. For example. The use o f 26 periods for every year may seem a burden for practitioners. Further reproduction prohibited without permission.

Suggestions o f when some o f these lim itations can be ignored are also offered. statistical efficiency. and identification have been presented. One o f the limitations is that the experimental data may not represent the true deterioration m echanism o f in-service pavements. asphalt concrete mix characteristics. . and environm ents for which the model was estimated.4 Summary The joint estimation approach and its three key features: bias correction. thawing and high air temperatures in a single model. along with the model limitations. The other limitations are mainly related to materials used at the AASHO and WesTrack Road Tests and the environm ents in those tests. the jo in t estim ation results reinforce the need to include the in-place air voids in the W esTrack model. It also allows a better discrim ination o f the effect o f the asphalt concrete layer thickness relative to the thicknesses o f the base and subbase. This approach allows the identification o f the effects o f layer thicknesses. Further. The inform ation about the rutting o f materials in WesTrack. w hich is essentially the rutting o f the asphalt concrete layer. The model in most cases replicates the pavement behavior well. Reproduced with permission of the copyright owner. which is particularly significant considering the w ide range o f loads and pavement structures.131 6. asphalt concrete mix characteristics. Some suggestions are presented for the use o f the model in practice. Further reproduction prohibited without permission. single and tandem axle loads. allows a better allocation o f how much rutting is originating in the underlying layers and how much rutting is originating in the asphalt concrete layer in the A A SH O model.

The model specification uses concepts familiar to pavem ent engineers such as load equivalencies and structural coefficients. a new empirical model for predicting rutting in asphalt concrete pavem ents was developed using joint estimation with data from the AASHO and W esTrack Road Tests. the equivalence coefficients Reproduced with permission of the copyright owner. the m odel presented in this dissertation has some unique features that differentiate it from the state o f the art em pirical models. • The model captures the effect o f overlays accounting for the hardening o f the underlying materials with traffic before the overlay construction. .1 Contribution and Conclusions In this dissertation.132 Chapter 7: Conclusions 7. no restrictions are im posed on the values o f the parameters representing equivalencies between axle loads or the param eters representing pavement layer resistance. Thus. and the allocation o f cost responsibilities to different vehicles classes are: • The model considers two different mechanisms o f rutting: rutting in the underlying layers and in the asphalt concrete layer. The identification o f the origin o f rutting can be used to plan appropriate rehabilitation strategies. However. The m ost important o f these features and their im plications for pavement design and rehabilitation. In particular. itcan be used to determ ine an econom ically optimal strategy o f original construction and rehabilitation using a rut depth threshold as a criterion for rehabilitation. • During estim ation o f the model. Further reproduction prohibited without permission. asphalt concrete m ix design.

is significantly different from w hat is usually assumed in pavem ent engineering for the calculation o f standard axle loads (usually a value close to 4). This has important implications for the allocation o f cost responsibilities to different vehicles classes. • Three properties are sufficient to model the perform ance o f the asphalt concrete mixes accurately.133 for axle loads are assumed different for rutting originating in the asphalt concrete layer and for rutting originating in the underlying layers. which is obtained from the aggregate gradation.57. • The model predicts rut depths by adding predicted values of the increm ent o f rut depth for each time period. In this research. Reproduced with permission of the copyright owner. Thus. Further reproduction prohibited without permission. For exam ple. The estimated load equivalence coefficient for the asphalt concrete layer. the voids filled with asphalt obtained for the construction mix in the Superpave gyratory com pactor1. 0. . the model can be use to guide the selection o f appropriate ' Other ways o f estimating VFA may also be appropriate. it shows that in pavem ent sections where m ost o f the rutting originates in the asphalt concrete layer the use o f the a fourth pow er equivalence coefficient assigns an excessive am ount o f the dam age to the heavier loads and an insufficient am ount to the lighter loads. and the initial inplace air voids. This is particularly advantageous in a pavement management context where the current rut depth value is known and the interest is in predicting the rut depth value for the next tim e period (and hence only the change in rut depth). the results from the Superpave gyratory compactor were used because they were readily available. The three mix properties are a gradation index.

especially considering the number o f sections and observations that were used for their estimation. In particular. asphalt concrete mix characteristics. the model estim ation results indicate that there is a value for air voids that optim izes rutting performance. In particular. . a result that is consistent w ith traditional asphalt concrete mix design. This allows the integration o f pavem ent design and mix design for adequate rutting performance. Further. single and tandem axle loads. • The model captures the effects o f extreme environm ents. thaw ing and high air temperatures in a single model. The individual models for the AASHO and W esTrack data sources were estim ated using the fixed-effects and the randomeffects approaches. In addition to the efficiency gains from using all the available inform ation. This research has also illustrated the advantage o f estimating a model jo in tly from m ultiple data sources.134 mix characteristics for adequate rutting performance. which implies efficiency gains with respect to an ordinary least squares estimation. once sim ilar models are developed for Reproduced with permission of the copyright owner. two tasks that have traditionally been performed separately. it captures the effects o f high air tem peratures in asphalt concrete mixes and the effects o f freeze-thaw cycles in fine frost susceptible subgrade materials (w hen a source o f w ater is present). The model fits were very good. the joint estimation approach has permitted the identification o f the effects o f layer thicknesses. Further reproduction prohibited without permission. Both approaches showed that the size o f the unobserved heterogeneity was significant.

Despite this. it may have small consequences for the prediction o f the mean rut depth. However. sim ulations using the parameters and variance com ponents obtained using the m ixed-effects approach would yield variance estimates that are more appropriate. This more com plicated approach yields more efficient param eter estimates. In such situations.2 Further research Despite the significant estim ation results in this dissertation. the model has some limitations that need to be addressed in future research. 7. jo in t estim ation has played a fundamental role in the assignm ent o f how m uch rutting originates in the asphalt concrete layer and how m uch rutting originates in the underlying layers for the AASHO Roar Test sections. in addition to the random variations o f the intercepts. . since m ost o f the rutting in W esTrack originated in the asphalt concrete layer. in the context o f this research. First. it was stated in Chapter 4 that it may be more appropriate to estimate the model parameters using a mixed-effects approach. In this modeling framework. w hich already are quite accurate.135 other distress types. the model presented in this research could be used in conjunction with those m odels to obtain economically efficient designs. som e o f the other param eters can also be assumed to vary randomly between pavem ent sections. Further reproduction prohibited without permission. Finally. a major m otivation for using the mixed-effects approach is that in many instances the interest is not only the prediction of the mean rut depth but also o f its variance. R eproduced with permission of the copyright owner. from an econom etric perspective.

Again. Further reproduction prohibited without permission. That is. Thus. . there may be som e potential use o f this information for obtaining the effect o f bitum inous-treated and R eproduced with permission of the copyright owner. It m ust be noted that there is a considerable amount o f information in the special base experim ents at the AASHO Road Test that can potentially be used to identify the coefficient for bituminous-treated and cement-treated bases. Because o f the uncertainty in the effective base thickness. sub-base. the coefficient for crushed stone bases can be used to obtain an approximation o f the rut depth. In contrast. The estim ation o f for bituminous-treated and cem ent-treated bases is considered o f particular importance since these are commonly used materials. and subgrade materials used at the AASHO Road Test for the main factorial experiment.136 The sub-model for rutting in the underlying layers is limited to the base. until a coefficient is estim ated for bituminous-treated bases. In this regard. However. /% and fli) corresponding to other m aterials need to be estimated. evidence from special experiments at the AASHO Road Test indicated that. the bases used for these special experiments were built as wedges. To increase the scope o f the model. Unfortunately. joint estimation with other data sources could play a fundam ental role in the identification o f these coefficients. the level o f rutting obtained in those special experiments with cem ent-treated bases was significantly lower than with crushed stone bases (about 30% lower). the level o f rutting achieved on pavements with bituminous-treated bases was sim ilar to the level obtained with crushed stone bases. this information was not used in this research. except for a m inor tim e lag. with the model obtained in this research in hand. the parameters {$■> . the thickness o f the base w as gradually changed from the beginning o f the section to the end o f the section.

Further reproduction prohibited w ithout permission. further research is needed to find an indicator that is applicable to all kinds o f gradations. base thickness. Other data sources are needed to incorporate the effect o f different subgrade m aterials. A lthough the gradation index defined in this research seems to be a good indicator o f the resistance to rutting provided by the aggregate structure for W esTrack and AASHO mixes.. As mentioned in Chapter 5. further research is needed to characterize the resistance o f different aggregate gradations and the effect o f different asphalt concrete grades. particularly for regions with considerable freeze and thawing or with considerable precipitation and poor drainage. In particular. This may be possible because there are many sections influenced by the same factors (e. The developm ent o f such a test would also help to incorporate R eproduced with permission of the copyright owner. the best option may be to device a m echanical test experiment that yields an indicator o f the resistance to rutting o f the aggregate structure.) but with different base types. The inform ation contained in other data sources could also be used to estim ate the param eter for bitum inous-treated and cem ent-treated bases.g. With respect to rutting originating in the asphalt concrete layer. results in the literature indicate that some gap-graded aggregates that depart from the m axim um density line can still produce mixtures with good rutting resistance.137 cem ent-treated bases relative to the effect o f crushed stone bases. asphalt concrete thickness and material. sub-base thickness and m aterial. . there are many HVS (Heavy Vehicle Sim ulator) experim ents prim arily focused in the effects o f base types that can be used for that purpose. subgrade m aterial. Given the difficulties o f sum m arizing the whole gradation curve with a single index. loading. etc.

In addition. Although this is one o f the most com mon maximum aggregate sizes for m ixes.138 the effect o f the maximum aggregate size used in the mix. For example. For stronger subgrades. particularly in a pavement management context where the predictions are limited to one or two years. some o f these lim itations are not so severe. the subgrade material at the AASHO road test is representative o f weak subgrades. as pointed out in Chapter 6 . The model should be validated by comparing its predictions w ith the performance o f the m ixes in other data sources. Here again. In addition. . The difference in this research is that those param eters are specifically obtained for rutting. the experimental data may not represent the true deterioration mechanism o f inservice pavem ents because o f differences in factors such as traffic wander. joint estim ation with data from in service pavem ents can play an im portant role in the development o f bias param eters for the prediction o f field rutting performance. an estimate o f rutting in the asphalt concrete mix may be R eproduced with permission of the copyright owner. Finally. It is im portant to stress that most o f the above limitations are com m on to all existing empirical m odels. traffic speed. and material aging. The results in this research are limited to asphalt concrete mixes with a maximum aggregate size o f 19 mm. if no source o f w ater is present and there are no freeze-thaw cycles. it is still desirable to incorporate the effect o f this variable. Further reproduction prohibited without permission. one can bound rutting by using the complete specification for an upper bound and the specification for rutting in the asphalt concrete mix for a lower bound. which usually assume values for the layer coefficients and load equivalence coefficients. Bias correction is the only key feature o f jo int estim ation that has not been taken advantage o f in this research.

R eproduced with permission of the copyright owner.139 all that is necessary since m ost o f the rutting w ill occur in that layer. Further reproduction prohibited w ithout permission. . unless extrem ely soft asphalt is used. the asphalt cement used in the m ix does not play a role in the rutting performance for pavem ent sections in very cold regions. Finally.

Calibration o f a M echanistic-Empirical Rutting Model fo r In-Service Pavements. A m erican Association o f State Highway and Transportation Officials. D. A. Transportation Research Board. W ashington. AASHTO Guide fo r Design o f Pavem ent Structures. and S. A merican Association o f State Highway and Transportation O fficials. AA SH T O (1996).M. M adanat (2000).C. A ASH T O (1997). W ashington.U.C. Transportation Research Record 1583. Reproduced with permission of the copyright owner.140 References AASH T O (1993). Development o f a Pavement Rutting M odel fr o m Experimental Data. D. R. D. H. Tayabji. Washignton.R.C. and F. Washington.. June 1996 AA SH TO Provisional Standards. Ali. National Research Council. Standard Specification fo r Superpave Volumetric M ix Design. AASH TO Provisional Standard. Further reproduction prohibited without permission. Bahia (1997). A rchilla. Designation MP2-95. W ashington. A merican Association o f State Highway and Transportation Officials. D. Evaluation a n d Selection o f Aggregate Gradations fo r Asphalt M ixtures Using Superpave.C.C. and H. D. La Torre (1998). A nderson. pp 91-97.. Presented for publication at the 77lh Annual M eeting o f the Transportation Research Board. . Provisional Standard MP 1. American Society o f Civil Engineers. Subm itted for Publication at the Journal o f Transportation Engineering. S.

W ashington. E. D. Delorme. pp 99-108. 6 .C. Transportation Research. and C. Bassett (1990). E. Factors That A ffect the Voids in the M ineral Aggregate o f Hot-M ix A sphalt.. Behzadi G. P.. Further reproduction prohibited without permission. Transportation Research Record 1544. Hiernaux (1993). National Research Council.. 58. and S. Use o f LPC IVheel-Tracking Rutting Tester to Select Asphalt Pavements Resistant to Rutting. Cross (1989). W ashington. Vol. pp 1-8. Association o f A sphalt Paving Technologists. Vol 24A. and T. M orikawa (1990). Y. in Transportation Research R ecord 1469. D. A. Estimation o f switching models fr o m revealed preferences and stated intentions. Transportation Research Record 1384. Transportation Research Record 1540 TRB. R. National Research Council. M. R. D. pp 59-68. . pp 97-104. and R. J. Transportation Research Board. Brown. National Research C ouncil. and C. Benson. Yandell (1996).C. Transportation Research R eproduced with permission of the copyright owner.L. Effects o f Maximum Aggregate Size on Rutting Potential and Other Properties o f Asphalt-Aggregate Mixtures. (1996). Asphalt Paving Technology. E. and W . 485-495. Study o f in Place Rutting o f Asphalt Pavements.C. pp.C. W ashington.E. Brown. No. D. Transportation Research Board.A. National Research Council. M acKean (1994).... T.O. Determ ination o f Elastic and Plastic Subgrade Soil Param eters fo r Asphalt Cracking a n d Rutting Prediction. W ashington. 1-39. Transportation Research Board. Brosseaud. Ben-Akiva. Specification o f Asphalt Concrete Using Total Quality M anagem ent Principles.141 Aschenbrener.

M. S. Giltinan (1995). Transportation Research Record 1259. D.E. pp.W . 107-119. User Highway . S.C. Leahy. D. (1984). Raymond (1982).L. Volume 53. D. R. 738-782. pp. Journal o f Geotechnical Engineering Division. W ashington. June 1998. R eproduced with permission of the copyright owner. A sphalt Paving Technology. FHWA (1998).. Perdomo. S. 12151229. Repetitive Load deform ation of Cohesionless Soil.. ASCE 108 (10). Influence o f Aggregate on R utting in A sphalt Concrete Pavements.M .. N onlinear M odels fo r Repeated M easurem ent Data. Davidian. Brown.L. W ashington.F..C. M onographs on Statistics and Applied Probability 62. V.08 Federal Adm inistration.H. W ashington.C. M. Ashmore. and R.C.B. Documentation. C. WesTrack Perform ance-Interim Findings. Proc. pp 415-439. and T. Diylajee. Association o f Asphalt Paving Technologists. Button. Long Term Publication Pavement Number: Performance PROQUAL FHW A-TS-98-00-01. Seeds. and K. M itchell (1998).A. Further reproduction prohibited w ithout permission. N ational Academy o f Sciences.142 Record 1259. V olum e 67. M onism ith. pp. National Academy o f Sciences. Transportation Research Board. S. The M echanical Properties o f Bitum inous M aterials fo r Road Bases a n d Basecourses. Transportation Research Board. London. Journal o f the A ssociation o f Asphalt Paving Technologists..P. 141-152. Lytton (1990). Epps J. Chapman & Hall. and D. Cooper. and G. D. J. Alavi. V2.

Institute o f Transportation Studies. Berkeley. The Superpave M ix Design System. and A.B. M odern Pavement M anagem ent. University o f Nevada. (1997). Strategic Highway Research Program. Upper Saddle River. Reno. D. Florida. R. C Scheffy (1997). J. A New G raphical Chart fo r Evaluating Aggregate G radations. Test Methods. Association o f Asphalt Paving Technologists. M anual o f Specifications. Greene.C. du Plessis. Ph. Long. Greene. Report prepared for California D epartm ent o f Transportation. N ew Jersey. SHRP-A-379.. F. E. pp 176207. Dissertation.. J.T. Guada.: Econom etric Software. Vol. Hand. Relationships Between Laboratory M easured HblA M aterial and M ixture Properties a n d Pavement Performance at WesTrack. 1962.Section 500RF. Bellport. Third Edition. Krieger Publishing Company. Adam J. Zaniewski (1994).D. a n d Practices. University o f California.F. Lufsey (1962).. Prentice H all. and J. R. W ashington. 31.G. Haas.S. Econometric Analysis. Further reproduction prohibited without permission. Pavement Research Center.143 Goode. Harvey. Deacon. H udson. National Research Council. Leahy.. Version 7. I. J. C A L-A P T Program: Test Results fro m Accelerated Pavement Test on Pavem ent Structure Containing A sphalt Treated Permeable Base (A TPB).0: User's Manual. W illiam H. . L.. J A. W illiam H. Hung. (1995). (1998). W.T. D. Reproduced with permission of the copyright owner. Y oucheff (1994). L1MDEP.Y. R. Harrigan. N. Proc.

Report 5 — Pavement Research. HRB (1962). R . and G. and T. G . Further reproduction prohibited without permission. Washington. Proc.S. H ughes. Effects o f A ggregates and M ineral Fillers on Asphalt M ixture Performance. Brown (1993). pp. Philadelphia. An Introduction to Geotechnical Engineering. Prentice Hall.D.S. pp. Proc. Pa. and W . M aupin Jr. Effect o f A sphalt Concrete Parameters on Rutting Perform ance: A Field Investigation. . Publication No. A ssociation o f A sphalt Paving Tecnologists. Asphalt Paving Technology. Transportation Research Record 1384. K ovacs (1981). and G.S. N ew Jersey. Special Technical Publication 1147. C. P. National Academy o f Sciences . Heiman (1987). Asphalt Paving Technology. Highway IE. ASTM .C.144 Holtz. Association o f Asphalt Paving Technologists. 4958. DC. 1-32.. Huber. Volum e 56. H. 225-251. D. Shuler (1992). Englewood Cliffs. 954.D. The AASH O Road Test Research Board Special Report 6 . Experimental Bituminous M ixes to M inimize Pavement Rutting. S. R eproduced with permission of the copyright owner. Washington. Cross. 33 -6 1 .National Research Council. H eavy-Duty Asphalt P avem ents in Pennsylvania: Evaluation fo r Rutting. H uber A. Kandhal.R. Transportation Research Board. and E..A. Volume 56.W . National Academy o f Sciences. (1987).A . pp. Providing Sufficient Voids Space f o r Asphalt Cement: Relationship o f M ineral Aggregate Voids a n d Aggregate Gradation.

101-130. Transportation Research Board. K enis. S.R. (1981). Characterization o f Rutting Potential o f LargeStone Asphalt Mixes in Kentucky. D. part 2. Transportation Research Record 616. 133-140. Paris. (1977). W ashington. (1986). TRB. pp. R eproduced with permission of the copyright owner. Draft Report..W . H ighw ay D evelopm ent and M aintenance System 4. Heavy Wheel Loads and R oad Pavements . D. M ahboub. Deform ation M echanism in Asphaltic Concrete. J. Lai. N ational Academ y o f Sciences. J. Predictive Design Procedures: A D esign M ethod fo r Flexible Pavem ents Using the VESYS Structural Subsystem .J. Allen (1990). M odelling Rutting in Flexible Pavements in HDM-4. . Journal o f Transportation Engineering. M ichigan. Transportation Research Record 1259.J. 4th International Conference on the Structural Design o f A sphalt Pavements. M aree.145 K annem eyer L..C. A nn Arbor. The P erm anent Deform ation o j Pavements with Untreated Crushed-stone Bases as M easured in H eavy Vehicle Simulators Tests. 29-45. Hufferd (1976). van Zyl and P. Sym posium on Heavy Freight Vehicles and their Effects. and W. Proc.L. N. Freeme. W .C. K hedr. 41-43. A SC E 112(1). C.S.F. L. N. Proceedings 11th Conference. A ustralian Road R esearch Board. Washington. (1999). Savage (1982). Lister. pp.H . K. National A cadem y o f Sciences. and D. Melbourne.A.W . Further reproduction prohibited without permission.Damage Relationships. O rganization for Econom ic Cooperation and Development. Predicting Permanent Deformation o f A sphalt Concrete From Creep Tests.

Published for The World Bank. Further reproduction prohibited without permission. (1988). pp. A. FHW ARD-97-001.K. R eproduced with permission of the copyright owner. The H ighw ay Design and M aintenance Standard Series. O strom . and Kikuko Yamada. B. O ffice o f Engineering R&D. Paris. (1976). D.C. O. W ashington.. M oshe B en-A kiva. W alker. S. D.146 M onism ith. Baltimore. VA. OECD Publications and Information Centre [distributor]. Transportation Research Board. Heavy trucks.C. Report No. D. Forecasting Intercity rail ridership T ransportation Research using revealed preference a n d sta ted preference Record 1328. C. M orikaw a. W . W ashington. Rutting: The Contribution o f High Tire Pressure and Rem edial Measure.C. N ational Academ y of data. 30-35. Federal Highway A dm inistration. Paterson W. L. . and S Rowshan (1997). France . Presented at 3rd 1RF M iddle East Regional Meeting. Phang. Sciences. National Academ y o f Sciences. The John Hopkins University Press. O rganisation for Economic C o-operation and Developm ent. T ransportation Research Record 616. Long-Term Pavem ent Perform ance Information M anagement System Data Users Reference Manual. (1987). R utting Prediction in A sphalt Concrete Pavements. (1991). TRB. McLean. O EC D (1988). Riyadth. Takayuki. Road deterioration and maintenance effects: m odels fo r plannin g and management. Harris. W ashington D. prepared by an OECD scientific experts group. D. 2-8. climate a n d pavem ent damage.

Verification and C alibration o f the PD M P and COLD Com puter Programs. (1993). Oxford University Press. Cambridge. and D. Sousa. Com bining laboratory and fie ld data in rail fatigue analysis. TRB.C. Ann Arbor.D. J. Transportation Research Record 1384.147 Ruud. pp. R utting Rate Analyses o f the AA SH O Road Test Flexible Pavements. W ashington.C. M . Thom pson. National Research Council. Tayebaly. Proceedings Fifth International Conference on Structural Design o f Asphalt Pavements. Naum an (1993).R. National Academ y o f Sciences. D. (1982). . C. and C. University o f California at Berkeley. Incorporated. Further reproduction prohibited without permission. MIT.Y. F. Strategic Highway Research Program.L. 36-48. D. 321-332. pp. MA. and SW K Pavem ent Engineering (1994). D. Washington. A n Introduction to C lassical Econometric Theory. A. TRB. P. Dissertation. Michigan. Austin Research Engineers. Ph.C. A.. Transportation Research Record 1384. University o f M ichigan and Delft University o f Technology. Oregon State University.. Report SHRP-A-398. (2000). National Academy o f Sciences. 69-79. Inc.B. Reproduced with permission of the copyright owner.. P. Shyr. Stage I Validation o f the Relationship Between Asphalt Properties and Asphalt Aggregate M ix Performance.. W ashington. Sensitivity o f Strategic Highway Research Program A-003A Testing Equipment to Mix Design Param eters fo r Permanent Deformation and Fatigue. H endricks. J. Harvey. M onismith (1993). Saraf.

D. Australian Road Research Board. (1975). R eproduced with permission of the copyright owner. N ew York.J. Accelerated F ield Test o f Perform ance-Related Specifications fo r H ot-M ix Asphalt Construction... W esT rack Team (1996). W esTrack Team. E. Final Draft. B.W. Interim Report. Soil Behaviour and Critical State Soil M echanics. (1990). Repeated L oad Triaxial Testing on the Subgrades fro m M idgrace A L F Site. Principles o f Pavement D esign. Task G. .M. John Wiley & Sons. M. and W itczak.148 V uong. Amstrong (1991). Further reproduction prohibited without permission. and P. Yoder. W ood. M arch 1996. Cambridge U niversity Press.

In this approach.149 Appendix A: Estimation Approaches The m odels developed in Chapters 4.3 then describe the two m ost com mon frameworks used to extend the estimation to a panel data setting.l) where S is the number o f sections. This assumption is quite unrealistic with panel data since the error terms for a given pavem ent section are likely to be correlated because of com m on unobserved factors. D ifferent assumptions about the error term s lead to different estim ation approaches.. Using as an exam ple the AASH O Road Test model. the rut depths can be expressed as the following set o f regression equations: RD l t = E(RD „|Ar//.P) + 6 „ / = 1 . These two frameworks are the fixed-effects and random -effects approaches respectively. Further reproduction prohibited without permission.eyj) = 0 for all i * j or t * s. and 6 give the conditional expectation function o f rut depth for section / at time t../3) for the AASHO model and E(RD„ | Yiuy) for the W esTrack model. T. is the number o f observations for section / and e„ is the error term... all three . (the vector o f explanatory variables corresponding to the /th observation for section /) and on the vectors o f param eters or y for the AASHO model and W esTrack m odels respectively. 5.2 and A.S. In R eproduced with permission of the copyright owner. Sections A. this function gives the expected rut depth conditional on the set o f regressors X it or Y. This appendix describes three commonly used estim ation frameworks. r= l T..l describes the nonlinear least squares estimation approach. (A . E(RD it | X it. Section A . As explained in those chapters. it is assum ed that Cov(e„.

2) Under fairly general conditions (given for exam ple in Greene 1997 or A m em iya 1986). A . A consistent estimator for a E2 can be com puted using the residuals (G reene 1997): s t. (A. the nonlinear ordinary least squares (O LS) estimate o f (3. variance a e2. Further reproduction prohibited without permission. the error terms are assum ed to have zero mean and constant variance.l Ordinary L east Squares Under the assum ptions that the 8 ./s are distributed with mean 0.150 fram eworks. Most o f the equations have been modified from G reene (1995. can be obtained by minimizing: s t. T hese sections are intended to give an overview o f the approaches w ith an emphasis on how the estimates o f the parameters and covariance matrices are obtained. and Cov(Ei„Eys) = 0 for all / # j or t * s. 1997) to account for the non-linearity o f the models and the unbalanced panel data sets. . P 0 i5 .3) i=i R eproduced with permission of the copyright owner. The three approaches are explained in the context o f the AASHO model. this estim ator is consistent and asymptotically normal. (A.

4) where X ° is an n x K matrix given by X ° = [ x ? \x Q 2\ .7) . . |* / „ P ) = P r + / ( * / . Further reproduction prohibited without permission. equation (A .5) x K) row vectors given by = „ he r e y = g n +. E(RD„ | X n. *= 1 In order to estimate the variance covariance matrix (A.A sy.5 > .4). For example. A. (A . o f the regressors and the rest o f the parameters. 6 ) is evaluated at the least squares estimate.V a r$ ]= a l ( X 0' X ° ) (A. 1=1 The sam ple estimate o f the asym ptotic covariance matrix is (G reene 1997) E st. x Q n ']' and where the x f are (1 (A . t= R eproduced with permission of the copyright owner.p) can be w ritten as: E ( * A . .6. (A . for the AASH O Road Test model.2 Fixed effects ap p ro ach The conditional expectation functions presented in Chapters 4 and 5 can be written as the sum o f a constant term plus a function.5 . . / • ) . >P2) / = 1 .151 s where „ .--.

Given the above assumptions. (A. A way to proceed with the estimation is to elim inate the individual effects. there is no theoretical problem with the estim ation o f the parameters o f equation (A. p. The fixed-effects estim ator o f p i can then be obtained by minimizing the following objective function: 2 Reproduced with permission of the copyright owner. S is the number o f pavem ent sections. P k-\).9) where e„ is the error term which is assumed to have mean 0 and constant variance a E2. 8 ) Therefore.k + / ( * . K -\ is equal to the number o f param eters excluding the constant term s. even for a relatively small number o f pavem ent sections such as for W esTrack (S = 26) the nonlinear optimization becomes unfeasible due to the large number o f parameters to be estim ated (S+K-l). p i. Further reproduction prohibited without permission.. the regression model can be expressed as: (A. and 7) is the number o f observations for section /.9) by the OLS approach explained in the previous section. . ■■■. 10) . That is: P) = (3..152 where /?? = (J3/... However. .S . The fixed-effects approach assumes that differences across pavem ent sections can be captured in differences in the constant term (G reene 1997).K. p 2) / = 1 . t= (A . by taking deviations from the corresponding mean o f the observed rut depths and o f the conditional expectation function.

13) The estimator o f the covariance matrix for f32 is Esl.(•) instead o f £(•) and £. in practice the optimization o f equation (A. Further reproduction prohibited without permission. P2) It R eproduced with permission of the copyright owner.Var[(. 10) is done using _/(•) and /.]=s- V V ( X ° .l 1) and (A. .(•) W ith the estimator o f /?> in hand. 12) and where f t is the mean ofy(-)fc>r section It should be noticed that J3.153 w here T.* . (A. 14) / w ith the (1 o x /T . the implicit OLS estimator o f each (5xK can be com puted with: (A.1 ) vector ^ .. 15) .(•)• Thus..0 given by d f ( X „ . (A .k is /= 1 elim inated from the estim ation because it appears in both £(•) and £. 0)’( * 2-*?■ (A.

S . is a random disturbance characterizing the /lh section and is constant through tim e w ith mean E(u.„ p )): t=\ (A. A.E ( X " . it is natural to view the section specific constant term s as random ly distributed across pavem ent sections. approaches infinity) but it is costly in term s o f the number o f degrees o f freedom lost. 17) As before eit is assumed to be distributed w ith zero mean and constant variance crE‘ . Further reproduction prohibited without permission.*. Reproduced with permission of the copyright owner.1) V Ti . it is assumed that P. Specifically. y i2 (rd„ . $ ) + ut +eu (A.) = 0 and constant variance equal to ctu2.( K . because a different intercept term is required for each pavem ent section. . represents a lack o f know ledge about the model.-. p. Since the inclusion o f different constant terms.3 R andom -effects a p p ro a c h An alternative to the fixed-effects approach is the random -effects approach.a: = Pk + u„ w here u./ ( J f „ . p 2))2 V t.1) This approach yields consistent estimates for f r (consistent as S. 16) S 77 . [rd„ 2 <=I .E ( A r .154 X f is the m ean o f X f t over the 7) observations for section i and w here s2 is given by 5 s t.( K . These estimates are also asym ptotically normal.S ..With these assum ptions the random -effects specification is: —Pat + / ( ^ w / ’P 2 ) + w / + e « .p * . the num ber o f sections.

Then. = E[w. disturbance terms for section i is given by Q. 0 if i * j .} .2] = cju1 . 19) w. *’.22) Thus the covariance matrix for the T. E|£.i E[Sit My] = 0 for all /.i. observations corresponding to pavem ent section i. = s„ + u. E[£. U ' .f) = a} . £[w„2] = a } + a. My] = (A. .’].] = 0 .155 The following is the com plete set o f assum ptions for the disturbance com ponents in the random-effects approach (G reene 1997): E[£„] = £[w. (A. or R eproduced with permission of the copyright owner. t * s. Now let Wi. and j . 18) if t * s or / * j . wa . n ] ’ (A.t. = [w.21) E[w„ w„] = aru2. (A .20) and for the T. = 0 E[W. (A. £[n. Further reproduction prohibited without permission.

( P ) } > . identity matrix and i is a T.25) If the variance com ponents a j and a.26) ."> ii' = — W.{RD.E . Further reproduction prohibited without permission. I \ 2 ^. Since the observations for a given section are independent from the observations for the other sections. .E. = 1 a. (A. x 1 vector o f Is.t o U a: J i = 1. 5 (A . (P )J / =1 R eproduced with permission of the copyright owner. . is needed to obtain the generalized least-squares estim ator. the random-effects estim ator o f /?can be obtained by minimizing the following objective function: (P)^= ^ {RD.where I is a T. . a tE + T. .. . 0 0 Q= (A. x T. This inverse matrix is given by Q.24) 0 0 ••• The inverse o f Q.2 are known. . the error covariance matrix for the full set o f observations is 0 Q.

(X tt. I e // e .. (A . ..30) and 6 l= 6i-6lQ s where R eproduced with permission of the copyright owner. Further reproduction prohibited without permission. .27) /= 1 .. However. p) RDj. . the estim ation o f a j and a j can be done using the following equations (Greene 1997): A2 / =l ^=1 a e = .(P) is the vector o f predicted rut depth values for section /'.. . Then. (A. the following residuals and related quantities can be computed: eu = R D lt. A standard way to proceed w ith the parameter estim ation is first to estim ate p using an alternative approach such as OLS or the fixed-effects approach that yields a consistent estimate p .157 where RDj is the vector o f rut depth observations for section /' and E. S\ r = l . a I and cr„‘ (and consequently W.) are unknown. r . ..29) T.E l { X n . P) are the m ean o f the observed and predicted rut depths where respectively for section /.. „ = R D l . Then. .. and £ .E { X it$ ) i = 1 .^ ----------------------- (A .3 1) .5 (A.

. T. P ^ -. the covariance matrix o f ( 3 ^ . For this to be possible.° are T. equation (A. Although the estim ation described above is standard. t_= 1. . is equal to the vector o f explanatory variables corresponding the t[h observation for section /. R eproduced with permission of the copyright owner.30) and (A.31) at each iteration.35) / =! 'N P / where X. Further reproduction prohibited without permission.. Starting with an initial estimate o f p.34) Finally.. the Feasible Generalized Least squares estimates for the random -effects approach are obtained by m inim izing equation (A. is estim ated by a:. the values o f the error com ponents are recalculated with equations (A.26). and where X.33) s (A . for this research a different approach has been followed. with the estim ates o f a f and au2 in hand.158 (A.32) S -(K -l) (A . The use o f this non-standard procedure may partly explain the surprisingly similar results obtained for all the models estim ated in this dissertation with the fixedand random-effects approaches. x K matrices with flh row equal to ----- 1 / f) \ "’ (evaluated at P ^ -). Finally..26) is minimized directly w ith respect to the vector o f parameters. 1x? (A . .0' q .

81 P u a 2 = 5 .83 Pe 2 . Further reproduction prohibited w ithout p erm ission .159 Appendix B: Estimation Results The Tables in this appendix show the estimation results obtained using the fixed-effects and random-effects approaches for the models specified in Chapters 4.15 P s 2.4 5x1 O’2 15. and 6 .73xlO+0 2. 16 P* 2 . 78 P u 0 . 12xl0'7 -19. 01xl0+o 2.18xl0_1 24.20 -2. 39xl0_1 23. 99xl0'7 4 .38xl0+o 6. .76X10'1 4 .71 P 2 5. 88xl0"2 59. 93 P i 5. 30xl0_1 11.l: Parameter estim ates obtained for the AASHO model using the fixed-effects approach without including the overlay information.99xl0'5 9.OOxlO+0 P>2 6. The elem ents in the diagonal in the tables showing correlation matrices are the variance o f the param eter estimates. T a b le B .08 P7 1.46 P9 0. 93xl0’3 1. OOxlO+0 P to 1. 74xl0'2 24.25xl0'4 31.21xl0'6 1.90xl0'6 6. 72xlO+0 2. 42xlO+0 9.83xlO+0 1.70 mm Asymptotic t-value Sample size = 7035 R ep ro d u ced with p erm ission o f the copyright ow ner. ameter Parameter estimate Standard error P i 7. 58xl0*7 20.77 P s 3. 07xl0_1 24 . 5.

188 0.431 0. 1 8 x 1 0 ' •13 P l2 0.454 0. 345 9.30x10 P* 0 .080 -0. 2 75 5.588 0 .013 0.544 4.176 0.071 0.163 -0.776 -2 -13 -3 1 . Further reproduction prohibited w ithout p erm ission.97x10 P3 A P s P a P i P s -l 0 .071 0.411 P l2 -0.103 -0.244 -0.25x10' 14 R ep ro d u ced with p erm ission o f the copyright ow ner.179 -0.524 0.2 P io 9 .49x10 P s 0.114 -0.033 0. P2 P i P i 3.099 0.894 -0.89x10 Ps 0 . .56x: 1 .147 -0.628 P io P i 2 P l3 P .328 0.337 1.2: Estim ate o f the correlation m atrix for the param eter estim ates obtained for the A A S H O m odel u sin g the fixed -effects approach without including the overlay inform ation.355 0.148 -0.73x10 0.120 0.155 P l3 0.399 4.185 0.001 0.052 P io 0. 452 0. 390 P i 0.258 0.161 0.090 0.086 0.28xl0'4 1.016 0.380 0. 504 0.436 -0.319 0.495 P i 3 -0.38xl0'2 8.118 0.116 0.160 Table B.550 Ps -0. 356 0. 9 8 x 1 0 ' •3 -0.135 -0.041 0.098 -0.426 0.

03xl0_1 27 .20x10*° 1. 72xlO+0 2. 54xl0"2 28. . 65 P9 0. 01x10*° 1. Further reproduction prohibited w ithout perm ission.83x10*° 1. 42x10*° 8.42 P2 5. 91xl0~6 6.40 Parameter P n Pn Asymptotd t-value CTg =4. 35xl0'7 23. 50xl0_1 13.82X10'1 27 . 16xl0'7 4 . 34 Ps 2. lOxlO"1 27. 78xlO'0 -22.82 mm Sample size = 7035 R ep ro d u ced with p erm ission o f th e copyright ow ner.3: Param eter estim ates obtained for the A A S H O m odel u sin g the random e ffe c ts approach w ithout including the o v erlay inform ation.21xl0’6 9. 99xl0'5 8.38xlO+0 5.57 P4 2.26 P s 5.00x10*° P 12 6. 52xl0"2 68. 51 -2.76X10'1 3. Parameter estimate Standard error P i 7 .72 P s 2.84 P n 0. 86xl0'2 17. 72x10*° 2.60 1.161 Table B.00x10*° P io 1. 88X10'1 6.47 P 7 1.48 mm or^ =7. lOxlO"4 35. 93xl0'3 1.29 P s 3.

P io 0 .0 . 1 1 1 0 017 0 . 027 .4: Estim ate o f the correlation matrix for the parameter estim ates obtained for the A A S H O m od el using the random -effects approach w ithout in clu d in g the overlay inform ation.06x10 -2 6 . . 051 1 . 177 -0. . 263 P l2 0.415 . P i A . 0 2 x 1 0 -i P2 0 .0 . 0 .052 0.. 078 . 100 0 ..039 0 . 771 . 3 . 008 .41x10 P s 0 .0 . 391 P7 0 .0 .. 97x10' •13 P i 2 0. 067 0.896 .0 .. 5 3 x 1 0 ' -2 R eproduced with permission of the copyright owner.0 . 021 0 .352 0 .083 .0 .499 0 . 0.355 0 . 0 1 2 0.093 0. 327 0. 147 -0 . 065 . 633 0 .79x10 P s 0 .450 0 . P i -3 49x10' 3 1 80 3 . . 107 . P2 Pi P a P3 P s Pi P e Ps P i 3 . .0 . 139 ..113 -2 . 547 3 . 189 049 -0. 341 7 . Further reproduction prohibited without permission.0 . 025 . 124 P i 3 0 .395 3 .428 0 . . 378 0 .432 0 .240 . 526 0 .28 x10 P e 0 .0 ..0 .456 0 .175 0 .550 .038 0 -4 . 272 4 . P io P l2 075 P l3 1. 057 0. 180 0 .162 Table B.0 . 320 9 . . 330 0... 168 0. 481 1. 153 .34x10 0 .0 . 035 P i 4 P io 6.0 .2ix: 0. 238 . .13 -0 .0 ..0 . 0 80 0. 145 -0 . 161 P s . 588 0 . 096 -0 . 126 0.447 -0 . 3 0 x 1 0 -2 Pa 0 . 5 6 x 1 0 ' •15 P i A .

ObxiQ"1 24 . 62xl0"8 -21.36 P ? 1.89xl0*7 24 .73 03 4 . 48X10'1 19.5: Param eter estim ates obtained for the A A S H O m odel u sin g the fix ed -effects approach in clu d in g the overlay inform ation.20xl0'6 5. 58 0 9 . 34 P e 3. 94xl0‘2 9.10x10"° 2.39 P s 3. 78X10'1 25.79x10"° 2.11 P n 0..47x10"° 5. 05X10'1 17.51x10"° 1. 44xl0~2 73.41xl0'7 3. 63 P s 2.163 T a b le B . 32 P4 8 . Further reproduction prohibited without permission.1 . 8 9 x l 0 _1 3.00x10"° 012 3 . .21x10"’ 21. 66xl0_1 13.20 P2 5.78xl0"3 2.32 P n or =5.87x10"° 1.55x10"° 2. 17xl0"7 -8 . 18xl0’5 4 . 04xl0"6 1.36 mm Sample size = 9478 R eproduced with permission of the copyright owner. Parameter Parameter estimate Standard error Asymptotic t-value P i 7 .89 010 1.20xl0'7 2.85 -1.

2 5 x 1 0 -2 P3 0 . .458 5.115 0 . . 5 79 4. . 027 -0 . 181 0. P2 Pi Pa Ps Ps Ps Pi Pi 3 . 277 -0 . Ps P io P l2 -13 2. . 132 0. .0 . .152 0 . 089 0. 047 0 .0 . . .. 3 9 x 1 0 ' •13 P i 2 0. 98 x 1 0' •4 0 . 16 3 -0. . 170 -0 .213 . 123 . .. 175 0. 042 0 . 0 35 0 . 647 -0 ..390 0 . 337 0 P e 0 . . .0 .093 -0 . 103 -0 . 548 1... 3 . . 160 -0 .260 0.27 6 4 . 149 .0 . 0 2 x 1 0 Ps 0 . 044 -0 .0 .. 122 0 . 37x10 P io 0.075 . 391 . 1.212 0.203 P i 0 . 0...114 P s 0 .031 0 . 965 . 108 Ps 0. 1 9 x 1 0 -2 0 . 114 Pi 3 56x10' -3 425 103 0 . . 106 0. 197 P i 3 .033 0.224 0 . .6: E stim ate o f the correlation matrix for the parameter estim ates ob tain ed for the A A S H O m od el u sin g the fixed-effects approach including the overlay inform ation. 08 0 .091 0 . 8 8 x 1 0 0. . 1 8 x 1 0 -2 Pa 0 .023 1.. 219 .0 .062 0 .164 Table B. .. 189 Ps 3. .0 ..488 3 . 144 0 . 0 31 . 833 ..239 0 .0 . 204 2 . Further reproduction prohibited without permission. . . 2 0 x 1 0 -2 0..063 0 . 120 0. . 187 0. 2 1 X 1 0 ’1 p2 0 . 228 0 . 044 P l2 0. 091 0. . 137 4. 4 5 8 -0 .0 .14 P s P i . 324 0 .0 . . 1 6 x 1 0 ' ■i s R eproduced with permission of the copyright owner. 309 P io 0 .

2 7 x l 0 ‘7 27. 80x l 0 ’7 3. 30xl0_1 22.14xl0'2 83. 6 3 x l 0 _1 6. 02xl0'7 -10.33 1.00 Ps 2.00x10*° P 12 3.78xl0‘3 1.55x10*° 1. 93 P4 8.24 Ps 4 . Parameter estimate Standard error Asymptotic t-value Pi 7 .93x 10"8 -24 .82 P s 4 . 94xl0"4 24. Further reproduction prohibited without permission.87x10*° 1.79X10"1 19.05 Ps 3.06x10*° 1.47xlO+0 Ji.80 P7 1.7: Param eter estim ates ob tained for the A A S H O m od el using the random e ffe c ts approach includ ing the o verlay inform ation.51 Pn a 2 = 4.4 lxlO-7 2 . 56X10'1 28. .20xl0'6 4 .32 P is -1 . 04xl0’6 1.165 Table B.79x10*° 2 . 4 2 mm Sample size = 9478 R eproduced with permission of the copyright owner.23 mm ct2 = 8 . 18xl0'5 4 . CD OO X h-* o1 V— Parameter 15.61 P n 0.59 p* -1.51x10*° 1.15 P to 1 . 4 4xl0"2 11. 10x10"° 1 o f—1 X 1—1 OO 1—t 28.89X10"1 3. 33 P2 5.

081 0.004 Pio 012 013 PlA Pi 4 Ps -2 -2 -W Ps 1 . 326 0 .094 0.064 0.151 0.0 . 0 2 5 -0.8: E stim ate o f the correlation matrix for the parameter estim ates ob tained for the A A S H O m od el u sin g the ran d om -effects approach including the o verlay inform ation.214 0 .059 0 .43x10 Pa 0. 0 4 2 -0. 173 0 .116 0. 105 0. 130 0 . 490 2.045 0. 5 8 x 1 0 ' •4 1 . 157 -0.225 0 . 117 -0.0 . 113 0. 100 -0.143 0.22x10 -2 4 .091 0.0 .179 Pi.106 0.649 -0.214 Pl2 0. 193 0. 115 0 . 7 6x10"' .003 0.221 -0.025 -0. 173 3. 6 9x 10 .837 -0. 581 Pi 0. 338 0.0 .84x10 Ps 0.966 -0.424 2 . Further reproduction prohibited without permission.029 0 .38x10 P2 0.046 Ps 0.091 0 .032 -0.041 0.277 3.114 0. 0 4 x 1 0 ' ■14 Pio 0.195 1 .276 -0.199 -0.0 .197 0 .030 .040 0. 002 0.259 0. 6 6 x 1 0 ' •2 R eproduced with permission of the copyright owner.188 0.022 7.459 Ps 0. 028 0 . 4 2 x 1 0 ' •15 PlA .166 T a b le B . 121 0 .063 0.308 0.460 -0.2 Pe 0. 396 0. 537 Pu -0. 3 .004 0. 1 8 x 1 0 ' ■3 2 .386 -0.242 0.147 -0. 167 .228 0.097 Pio 0.010 0.034 -0. 164 -0.076 . 132 0.3 . 8 2 x 1 0 ' ■13 Pl2 0 .031 0. Pi 02 03 Pa Pi 06 05 Ps -l fix 2.26x10 P3 0 .089 -0.184 1 .

06xl0+° 0. .36xl0"4 1.167 T a b le B. .4 6xl0"6 2.7 0xl0+° 11. 58 Y9 -2. ...33xlO+0 4 .66 Y7 -2...32 Ys 1 .53 Ye 10.17x 10’7 -11.18 Yj 0..19xl0'7 6.98 Ys 8..74xl0'6 3..68xl0'6 9. Further reproduction prohibited without permission.9: Parameter estim ates obtained for the W esTrack m odel using the fixed-effects approach.63xl0"6 6.01 Y2 5 ...87xl0+o 2. Parameter Parameter estimate Standard error Yi 8.36xl0"5 9... ......00xl0T° Y4 1 . 30xl0'6 1 .72xl0'6 -1.36xl0’7 2. . 34 Asymptotic t-value Sample size = 860 R eproduced with permission of the copyright owner.25xl0‘6 2 ..

4 78 0. 7 1 x 1 0 ' ■14 77 . 012 0. 2 . 8 5 x 1 0 0 .435 . .0 . .. 9 24 -0 .067 Ya 0 . 4 3 x 1 0 ' -0 4 . 643 . 72 Yi Yi .0 .13 1. .0 .0 . .0 .064 0. .168 T ab le B. 694 0 . .010 7a .10: Estimate o f the correlation m atrix for the param eter estimates obtained for the W esTrack model using the fixed-effects approach. . 046 -0 . .545 0 .0 . . . 4 7 x 1 0 +0 10 8 -0 .0 .0 .009 .764 R eproduced with permission of the copyright owner. 616 -0 . Further reproduction prohibited without permission.447 8..0 ...0 . . . .468 . 62 7 0 .788 5.5 1 x 1 0 -12 72 75 74 .008 .448 . 4 97 .10 0. 307 ..074 .326 76 75 76 7a 7i 7 . 0 4 x l 0 " 13 -0 .068 0 . .0 ..0 . 695 4 . 683 79 0. 4 5 x 1 0 0.

08 77 -2..80 d....46xl0'6 1 ..4 3 mm Sample size = 860 R eproduced with permission of the copyright owner.36xl0~4 .. l l : Param eter estimates obtained for the WesTrack m odel using the random effects approach..59xlOt0 0 - 1 .98 mm d^ = 7 .b .39 72 5..93 73 0...21xl0"5 11....14xl0+0 5. = 2 ..58xl0'7 2.169 T able B .97 76 10.25 7s 8.32 0 ..43xl0‘6 3.92xl0'7 -12. . 47xl0_1 0 ... xl0~e 8..88xlO+0 2.7 5xl0'6 ....05xl0+° 0.15 7s 79 7/o 1 .25xl0~6 2. 31xl0"6 1 . ...63xl0'6 1 -2.. Parameter Parameter estimate Standard error Asymptotic t-value n 8..82 7 .OOxlO+Q 74 1 ..18xl0'7 6.67xl0+° 11. Further reproduction prohibited without permission.

46x10 Ys 0.121 0 . 184 4 . 9 0 x 1 0 ' -12 Y2 0 .308 • 13 -10 -0 3 . Further reproduction prohibited without permission. 028 0 .242 0. 3 5 x l O +0 Reproduced with permission of the copyright owner. 3. 7 5 x l 0 ' 13 Yio Yio Ys 0 . 164 -0.114 0.097 . 5 7 x l 0 +° Yi 0. Yi Ya Y2 Ys Ye Yi Ys Yi 5 .063 0.45x10 Ye 0.225 0 .581 Ys 0 .064 0.022 0.12: Estim ate o f the correlation matrix for the param eter estim ates obtained for the W esTrack model using the random -effects approach.649 -0.459 Ys 0.031 0.214 0 .396 0. 041 0.966 -0.69x10 Ya 0. 338 0 .326 0 .046 Yio 0. 030 -0.228 0.06x10 0.460 -0.091 0. 167 -0. 6 9 x l 0 " 14 5 .277 6.113 0.106 0.089 -0.117 -0.143 0. 197 0.170 T able B.490 1.

208 . . 0 5 7 1 . 048 0. 075 0 . 001 0 ..3 P e 0 . 036 0. 060 0 . . 00 0 . . 000 0. 001 0. 007 . . 054 . 02 7 .0 . 065 0 . 002 .267 -0 .0 .24x10 P s 0 . 162 . 394 P i 0 . 574 P s 0 .0 . 0 42 0.0 . 001 ..20x10 . .0 .0 . 029 .019 Y2 .0 . 13 0 0 .026 -0 . 172 0.086 0 .. .0 .435 0 . . P i P 2 @3 P 4 P s P e P i P s Pi 3 .0 . 000 0 . 084 0 .. .58x10 . 005 R eproduced with permission of the copyright owner..065 . 184 001 .0 . 009 0.0 . 342 -0 .. 024 0. 073 . Y4 . 000 . 28X10'1 P2 0 .0 ..0 .0 .0 . 070 0. 001 7 .0 . 139 0 . 039 . 015 . 004 .208 0 .2 0 .37x10 .273 0 . 001 .282 0 . 0 1 1 3 . 0 07 .0 . 002 0.313 0 . 014 . . 167 0 .. 568 -0 .0 . 005 .346 0 . 002 .277 .. .0 . 107 -0 . 021 0.0 .065 P i 2 0.4 0. 149 .062 0 . 124 0 . .77x10 . 035 Ys 0.0 .249 0 .016 -0. 0 3 9 .008 0 .0 .0 . 177 -0 .0 .079 . 351 4 .230 .0 . .009 .0 . 029 .0 .150 .. .0 ..0 . .. .0 . 00 1 . 337 -0 . 022 0.093 2 P s 0 . 003 057 0. 001 . 000 . .491 3 .212 P i 3 0 .276 0 . 051 .040 . . .1 3 . 006 . . 012 Y io 0. 947 -0 . . . 057 0 . 004 0. 001 0.0 .21x10 -2 P3 0 .37x10 .0 .0 .0 .0 P i 4 .. 002 Ys 0. . 020 -0 .0 . 048 .099 0.0 .0 .448 0 .0 . 073 ..447 0 . 372 9 . 091 -0 . 001 .0 . 001 0 . 006 .039 . .0 .0 .0 .0 . 017 . 650 P io 0. . 001 0. 2 2 9 0.069 0 . 178 0 . 017 0. Further reproduction prohibited without permission.002 -0 . 342 0 . Ye 0. Yi . .0 .2 P< 0 . 001 . 0 4 2 0 .0 .T a b le B . 000 0. 120 0 .13: E stim ate o f the correlation matrix for the parameter estim ates obtained u sin g jo in t estim ation from the A A S H O and W esTrack data sources w ith the randome ffe c ts approach. Ys 0.053 . 108 .097 0 .0 . 102 . 171 0 .347 0 . 003 .0 .019 -0 .013 .

86x10 Pio 0 .004 -0.080 0.041 -0.001 79 0.143 .007 -0.185 -0.383 8 .70xl0'1 Reproduced with permission of the copyright owner.061 0. Further reproduction prohibited without permission.265 -0.46x10" 710 0. 796 0 .69x10 Pl2 0.208 78 -0.30x: 75 0.172 T a b le B.107 0.156 0.258 0 . 083 -0. 78xl0"2 6 .049 0.020 -0.0 . 103 -0. 221 7i 0.060 0.646 -0.13: Continued P9 Pio Pl2 Pl3 Pi 4 n 74 72 3.002 0. 335 5.0 .172 -0.004 -0.017 -0.300 -0.0 .008 0 .053 72 0.007 0.014 -0.15x10 -0.769 78 -0. 385 5. 106 -0.090 0.60x10 -13 7io 8 .459 0. .525 0 .313 0.239 -0.012 .077 -0.791 .204 -0.0 .432 0.819 4.13x10 76 -3 73 79 -15 2 . 1 0 x 1 0 -12 O PlA -13 5.014 0.325 -0.116 -0.500 Pl3 . 004 -0.072 1 o X* . 004 -0.191 0 .030 0 .167 -0.008 -0. 0 9 x 1 0 ’ 12 4.143 -0. 3 2 x l 0 tO 1 .003 0.272 0. 322 9.483 -0. 165 0. 2 4 4 -0.073 -0.011 -0. 002 710 75 75 2.064 -0. 138 76 -0.554 0 .017 0 .229 79 0.15 Po 74 0.510 0.220 -0.071 -0.149 0.005 0. 2 6 x 1 0 ' -1 76 -0.

The axle loads are expressed in kN (KiloNewtons) and S or T indicate that the pavem ent was loaded w ith a single or tandem axle respectively. the thickness o f the asphalt concrete overlay plus a ‘+” sign appears before the original asphalt concrete thickness.173 A ppendix C: O bserved vs. Further reproduction prohibited without permission. base.AASH O The figures in the following pages o f this appendix show the observed and predicted rut depths versus time for all the AASH O Road Test sections used to estimate the model. . and subbase layers. Predicted Rut D epths . For overlaid sections. The values after the word “ Design” in the section titles indicate the thicknesses in meters o f the asphalt concrete. R eproduced with permission of the copyright owner.

.. 35 35 ......T 40 .....30 E ...S Section 112 Design 0 0508-0 1524-0 2032 Axle loads 1 0 7 .... 0.... 30 ► 20 i- Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov 40 £ — i— :— Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date Date Nov Jan Mar May Jul S ep Nov J a n Mar May Jul S ep Nov Date Section: 113 Design: 0 0508-0 0762-0 Axle loads 53 • S Section: 114 Design: 0 0508-0 0762-0 Axle loads 107 ..T 40 -.. 111... ........l: Observed and predicted rut depth values for AASHO sections 107................174 Section: 107 O asign 0 0762-0....0762-0 1016 Axle loads S3 . 108 Design...30 E £ c 25 ■ t 20 J 10 5 25 I 20 1 I 151 x £ o» i (S r 15 10 5 o' Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Nov J a n Mar May Jul S ep Nov J a n Mar May Jul S ep Nov D ale Date Section: 109 Design 0 0762-0-0 2032 Axle loads: 53 • S Section: 110 Design 0 0762-0-0 2032 Axle loads 107 • T 4 0 ....... and 114...S 40 Section. 112.......... 110.. R eproduced with permission of the copyright owner.....35 J- l l> ~ 30 (E ! i 20 J 9 <E 10 5f0 E E I Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate S e c tio n -111 Design 0 0505-0 1524-0 2032 Axle loads 53 . 108... 109..... Further reproduction prohibited without permission... 113.....T i 35 L 35 30 r 30 j- 25f" § 2 5 ' 20 i 15 10 VV Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date 5 0 Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Dale F ig u re C ...... .0762-0 0762-0 1016 Axle loads: 107 ....

. .... i • i ..T 15 i: : 10 H r T 5 0 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov □ale Section 121 D esign 0 1016-0 0762-0 2032 Axle loads 53 .... 118..... ! \ 1 : ! 1 : : : ' ! 1 .. "....... 7 . .S 40 .i S> x Section: 118 Design: 0 0762-0 1524-0 Axle loads 107 . : i . ' .. 119....S 40 r ------.....T 35 * 30 r 25 -l 20 1 20 I i- 15 >1 0 J- *..2: O bserved and predicted rut depth values for AASHO sections 117. ... 1 ■ ' Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sap Nov Date Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section 119 Design 0 1016-0-0 Axle loads 53 .V Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov □ ale D ate Section: 123 Design: 0 1016-0 1524-0 1016 Axle loads: S3 . . ...........- - 15 u 1 ^ Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date . 120.175 Section: 117 Design: 0 0762-0 1524-0 Axle loads: 53 .... 1 ■ .... Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date F ig u re C.. .. r i S 20 2 .. ■ 1 1 i i . ------ Section: 124 Design: 0 1016-0 1524-0 1016 Axle loads: 1 0 7 ..: ...' E . 25 - f 20 L ......S .- Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Section... ....■ i. . 122 Design 0 1016-0 0762-0 2032 Axle loads 107 .. *1 ...— 3 35 .... i i —4 i __ i . Further reproduction prohibited without permission..i 35 - ~ 30 .. ' ' 30 f 25 r .. ...T I I .. 123.S Section: 120 Design 0 1016-0-0 Axle loads 107 ...---------.. .. 121.........: : . • ..... R eproduced with permission of the copyright owner. i ... . . .......... i I ' : i : ............ ... 122. .5 E 9 x 10 . ..T 40 ---------------------------------— ---. ... and 124.... ..

...r e . .......e r Nov Jan Mar May Jul S ep Nov J a n Mar May Jul Sep Nov Date Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Date Section 131 Design 0 0762-0 1524-01016 Axle loads 53 • S Section 132 Design 0 0762-0 1524-0 1016 Axle loads 1 0 7 .......' s 35l. I « 20 L 2 E ai ! C. ....................... ..... 35 | f 25 20 j.... 130....-~t- Section 130 Design 0 0762-0 0762-0 2032 Axle loads 107 .. 35 j- 35 - - £ 30 - ..1524-0 Axle loads: 107 . Further reproduction prohibited without permission. • ■ 11 .176 Section: 127 Design: 0 0508-0 1524-0 Axle loads: 53 .T 40 r 40 ! ..... and 136...... .. i .......... Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date Nov J a n Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Section: 135 Design 0 0506-0 0762-0.-------------------.. X 10 t 5 ^ 0 Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Oate Nov J a n Mar May Jul Sep Nov J a n Mar May Jul S ep Nov D ate F ig u re C...— ..3: Observed and predicted rut depth values for AASHO sections 127.30 • E £ 25................l J i :: “ h r . i 1 - E !r : : 25 i ..... ..T 40 ......... 128.............. ......S 40 .1016 Axle loads 53 • S 40..S Section: 126 Design: 0 0508-0...25^ © 20 •c 15 r ca . at X 10 h 5 .....T 30 £ e £ a © 5 r a x i........ 132. 15 •.... a<o A.. ......- ^ ■ | 20 15r L -- 3 E 15 a 10 'O p r a> > ■ 1 i ...... . c I I r 20 - at X 10 r 5* .... 129........ 135..T 40 r * ------... ■ ■ ■ ...... 'r D ata .. 35 * 35 - E 30 L- M E c a...... 131.. R eproduced with permission of the copyright owner.......... 5r 5 Nov Jon Mar May Jul Sop Nov Jan Mar May Jul S ep Mcv Date • ..... 0 r>* Nov J a n Mar May Jul Sep New Jan Mar May Jui S ep Nov Date Section 129 Design 0 0762-0 0762-0 2032 Axle loads 53 ..... Section: 136 Design 0 0508-0 0762-0 1016 Axle loads 107 ......... ...

4: Observed and predicted rut depth values for A ASH O sections 137. 142.T 40 35 ~ 35 i 30 £ f 25 f 20 ! 15 - K 10 a £ g> 30 L a 10 l 5 5 > .S Section 144 Design 0 0508-0-0 2032 Axle loads 107 • T 40 40 35 f- 35- 30 i- ~ £ 30- 25 . 143 Design 0 0508-0-0. Nov Jan Mar May Jul S ee Nov J a n Mar May Jul Sep Nov D ate Nsv J a n Mar May Ju! Sep Nov J a n Mar May Jul Sep Nov D ate Section 139 Design 0 1016-0 1524-0 2032 Axle loads 53 .T 40 35 30 ~ 30 25 20 15 10 5 o — . 141. .. _j Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Section 141 Design 0 1016-0-0 1016 Axle loads 53 . R eproduced with permission of the copyright owner. 138. . . . .S Section 142 Design 0 1016-0-0 1016 Axle loads 107 . i .T Date 40 40 35 I- 35 ■ 30 j- - E 30 - — 25 ■ 25 h r S 20" f 15c O) <r 1 0 - 5• « G? Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov D ate D ate Section.S 40 Section 140 Design.S S ection 138 D esign 0 0762-0-0 Axle toads: 107 . 143.177 Section. 9 15 'h 10 X 10 - 5 0 Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Nov J a n Mar May Jut S ep Nov Jan Mar May Jul Sep Nov Date D ate F ig u re C. 20 ^ f 15c. i . and 144. 139. .2032 Axle loads 53 . 140. . . 137 Oesign: 0 0762-0-0 Axle loads 53 . . . Further reproduction prohibited without permission. 0 1016-0 1524-0 2032 Axle loads 107 .

6 30 • 25 r | 25.. - I 20- 20 £ '5o a 10 - 15 • 10 5 - 5 • Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date 0 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date * Section: 153 D esign 0 1016-0-0 2032 Axle loads 53 . 153..... — Section: 152 Design 0 1016-0 0762-0 1016 Axle loads 107 .j Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov □ ate Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov D ate Section 151 O esign 0 1016-0 0762-0 1016 Axle loads 53 • S 40 ^ ... » ’ : Nov Jan Mbt May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date F ig u re C.....S Section 148 Design. and 154.178 Section: 145 Design 01016-0 0762-0 Axle loads: 53 ..... 35 • 35 - 30 .... ---------. ..S 40 - Section: 146 Design 0......5: Observed and predicted rut depth values for AASHO sections 145.1016-0 0762-0 Axle loads 107 ........ Further reproduction prohibited without permission..0762-0 0762-0 Axle toads 107 • T 40 40 ”b ? 30F 30 f 25 1 25 L ... 25 L 20 I........0762-0 Axle loads 53 ... 151. 0.....i -....T 40 35 - « 20 2 15- f.... Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sop Nov Date 0 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oale Section 147 O esign 0 0762-0.T 40 r 35 I- 35 30tE 75 c © o 20 - 30 j. .. R eproduced with permission of the copyright owner..: ~ -------....t- 1 4 :wi'f- 9 X 15 • 10 L i .... .........T 40 . x 10 r 1 • sl. 20 -- 15 ■ 10 t 20- 5 £ 15 .... 146......i.. X9 - r 10 ........— . 147..j .... 152. 148.! o f f E C ...---------— — -— •——-—... .S Section: 154 Design 0 1016-0-0 2032 Axle loads 107 ....* ' s ......-i— ....

..T 40 - 30 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Section: 157 Design: 0 0508-0 1524-0 1016 Axle loads 53 • S 40 T Section.T 4 0 . 161.............179 Section: 155 Design: 0 0762-0 1524-0 2032 Axle loads: 53 . ......2032 Axle loads: 107 ..._ _ _ _ _ _ _ _ _ _ _ _ _ Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Section: 162 Design 0 1016-0 1524-0 Axle loads 107 ...L L ^ j u I I J -I 30 25 i ' 20 j. Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section: 161 D esign 0 1016-0 1524-0 Axle loads 53 • S 5 j. 157.. 6 : O bserved and predicted rut depth values for AASHO sections 155..T 40 r S 25 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ata F igure C... 158 Design: 0 0506-0..... 158....S Section 160 Design 0 0508-0 0762-0 2032 Axle loads 1 0 7 ....... . 156. 35 i 30 r 25 j- « 20 15 r 10 i... --r-T-r -.....................T — -------------*...S 40 — 30 (- £ i f 25- I 20 - Section: 156 Design: 0 0762-0 1524-0..1524-0 1016 Axle loads 1 0 7 .... Reproduced with permission of the copyright owner...2032 Axle loads 53 .....r Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Section 159 D esign 0 0506-0 0762*0...........: i ... ... 159....... 40----35 -I.160..... Further reproduction prohibited without permission..... and 162.......i o *rr .......15 10 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date r j...

..- | 35 i ? 3o r £... ... : ■ ~ £ 25 L r - 15 1 i ....... oO c Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 255 Design: 0 1524-0 0762-0. 164......T 4 0 . 10 hf Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section 253 Design 0 1016-0 1524-0 4064 Axle loads 133 • S 40 ... 0 0508-0-0 Axle loads S3 .S Section: 164 Design: 0........... Section 254 Design 0 1016-0 1524-0 4064 Axle loads 214 ........ 1 6 6 ... ..• • ■ f 20 - 5 15 S D> ..4064 Axle loads 133 • S 40 Section 256 Design 0 1524-0 0762-0 4064 Axle loads 214 ...: ■I 20 !.2 5 4 ..........■ 0 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date 10 i 10 Section 165 Design.T -E 30 u E E r a © ■ o S 20 3 £ X9 Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov D ate Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Date F igure C.. -....... 30- .2 5 1 ........S 40 35 ir ..... si-. . 166 Design: 0 0508-0-0 Axle loads 107 • T 40 . £ tOj.* _ Section...... and 256.. ....... .......... .. ... Further reproduction prohibited without permission...... .....2 5 5 ......... 165...180 Section 163 Design: 0 0762-0-0 1016 Axle loads: 53 ........._ 35 • E E 0.2 5 3 . .. s £ g> ....7: Observed and predicted rut depth values for AASHO sections 163.. R eproduced with permission of the copyright owner.... ...........■ ....i 5 1. ..0762-00 1016 Axle loads: 1 0 7 -T 40 35 ~ 30 E ~ 30 E f £ 25 o 20 25 t 20 3 1£5 | }-- 1a 4 i £ i 5 - *..

2286-0 2032 Axle loads 214 .262. •o 2 0 i- s s 10 s> ce >- Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov D ate Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Section 263 D esign 0 1524-0 22B6-0 2032 Axle loads 133 .T r 25 a a> 20 r 15 l 10 im V - tr 10 r £ 1 5 'f MM Nov Jan Mar May Jul S ep Nov Jan Mar May Jul S ep Nov D ate ' Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Section 261 D esign 0 127-0 0762-0 3048 Axle loads 133 . and 264.1524-0 1524-0. 260. 261.3048 Axle toads: 214 .T 40 E E a.T 40 35 • ~ 30 • E E E E c Q.S Section 262 Design 0 127-0 0762-0 3040 Axle loads 214 . Q. a a © T3 u £a a 2 £ f | ^ a £ K 30 20L t- 10 Nov Jan Mar May Jul S ep Nov Jan Mar May Ju l Sep Nov D ate Nov Jan Mar May Jul S ep Nov J a n Mar May Jul Sep Nov D ate Section.1524-0.-------. R eproduced with permission of the copyright owner.3048 Axle toads: 133 • $ Section: 258 Oesign: 0. 258. Further reproduction prohibited without permission. . 263.T D ate 35 .1524-0. .S 40.— ■------.30E £ 25. 8 : O bserved and predicted rut depth values for AASHO sections 257.1524-0 2032 Axle loads 133 . V •o 2 £5» CC 5 Nov Jan Mar May Jul S ep Nov Jan Mar May Jul S ep Nov D ate *»*•' Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate F ig u re C. 259.181 Section: 257 Design: 0. 260 Design 0 127-0 1524-0 2032 Axle loads 214 .1524-0.S Section. 259 D esign 0 127-0.-------- Section 264 Dosign: 0.

35 • 35 £ 30 #25.T 6 g £ £ ■o E a x Nov J a n Mar May Jul S ep Nov J a n Mar May Jut Sep Nov Date Nov J e n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Soction: 267 D esign 0 101643 2266-0 3046 Axle loads 133 .. ... £ 20 .2 7 0 .T 35 i r £ £ £ f s E 25 r •• Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Date F ig u re C.. . Further reproduction prohibited without permission.. R eproduced with permission of the copyright owner..S Section 298 Design 0 1524-0 0762-0 2032 Axle loads 214 . ■ IT 10 r . ..... 297.. Section 270 Design: 0 1016-0 0762-0 2032 Axle toads 2 1 4 -T 40 -----------------.2 6 6 ... .. .S 40 — . and 298...■ f ? 15- 5 K 10 • 20 •.9: Observed and predicted rut depth values for AASHO sections 2 6 5 ..• • £2* 15....2 6 7 .. 268.S 40 — Section 268 Design: 0 1016-0 2286-0 3048 Axle loads 214 . - £ 30 r ^ 25.T 35 r £ E c 3 i X Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Date Section 269 D esign 0 1016-0 0762-0 2032 Axle loads 133 . 2 6 9 ...182 Section 265 Oesign: 0 127-0 2206-0 4064 Axle loads 133 • S 40 Section 266 Design 0 127-0 2286-0 4064 Axle loads 214 . .-----------------. r 5 r '* Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Dale Section 297 D esign 0 1524-0 0762-0 2032 Axle loads 1 3 3 ...

.......... © •u 3 2 S> (T 5 »■ Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov O ate Nov J a n Mar May Jul S ep Nov J a n Mar May Jul Sep Nov Section 301 Design: 0 1524-0 1524-0 4064 Axle loads 133 * S Section 302 Design: 0.......... _ Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Date D ate Figure C..10: O bserved and predicted rut depth values for AASHO sections 299.....*I i s l ............. 301.. -o 1 E S> x Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate D ate Section 306 Design 0 127-0 1524-0 3048 Axle loads 214 ..... R eproduced with permission of the copyright owner............183 Section 299 Design 0 1016-0 0762-0 3048 Axle loads 1 3 3 .......S 40 35 r 30 - 30 £ far--..... and 307.........____ _______ ..... ..------ Section: 300 Design: 0 1016-0 0762-0 3046 Axle loads 214 • T £E E £ £ £ TZ. 304............. 306. 35 > ~ 30 E £ 25.......... 300.........i ♦ i........ Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov D ate Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate Section 303 Design 0 1016-0 152443 2032 Axle loads 133 .S ... !■ J r■ : £oE> 10 L 25 8 2 15 ■ 10 5 0 L-i..1524-0 1524-0 4064 Axle loads 214 • T 40 r 35 30 E 2 20 r 0 - ... .. Further reproduction prohibited without permission.T 35 ~ Section 307 Design 0 127-0 1524-0 3048 Axle loads 133 ............----Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov 0 ec Z j i—i . 302.... 40 — ------ Section 304 Design 0 1016-0 1524-0 2032 Axle loads 214 * T 40 — .....-.....S 40 ■ — -----... 5... 303.....____ i _ f i _ _ ........ a « 2Qr *...

S ■ .S 35 I- £ f 25 I 20 L ' I «- a '0 <« - Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Oote Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section: 312 Design 0 1524-0 2286-0 3048 Axle loads 214 .S — 30 ► £e H- c.■ Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 31G Design 0 1016-0 2266-0 4064 Axle loads 214 • r Section.3 1 4 .- 0 .S E E £S. 3 1 0 .3 1 1 . ^ ------------------------------------.3 1 2 .184 Section 308 Design 0 127-01524-0 3048 Axle loads 2 1 4 -T 35 h ' 20 Section: 309 Design 0 1016-0 2286-0 4064 Axle loads: 1 3 3 .. | f 5 201 ? 151- I t a: 10 I- Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Dale Nov J a n Mar May Jut Sep Nov Jan Mar May Jul Sep Nov Date Section: 314 Design: 0 127-0. i.. ..3048 Axle loads 1 3 3 .2286-0.■-* -• t.7 ^ .2032 Axle loads: 214 -T Section: 315 Design: 0 127-0. Reproduced with permission of the copyright owner.0762-0 4064 Axle loads: 1 3 3 . and315. I f . Further reproduction prohibited without permission.i— .. « 2 x Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Nov J a n Mar May Jul Sep Nov Jan Mar May Jut Sep Nov Oate F ig u re C .. 311 Design.3 1 3 .T Section 313 Design 0 127-0 228643 2032 Axle loads 1 3 3 . 0 1524-0 2286-0. 309. l l : Observed and predicted rut depth values for AASHO sections 308.

...... R eproduced with permission of the copyright owner.... 3 2 0 ...35 I- 35 30 t 25 l 20 . '•M.. ..... 15 r 10r Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate Section..... 0 1016-0 0/62-0 4064 Axle toads 214 . Further reproduction prohibited without permission.. 317...T Nov Jan Mar May Jul Sop Nov J a n Mar May Jul Sep Nov D ate Section 319 Design 0 127-0 0762-0 2032 Axle loads 133 • S 4 0 .3 2 1 .S 40 ~ 30 6 - 25 i 1 20 r 20 f- u H x9 10 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section 316 Design.........-i—■— -----------... Section: 323 Design: 01016-0.i i M l i Nov J a n Mar May Jul Sep Nov Jan Mar May Ju l S ep Nov Date Nov J a n Mar May Jul S ep Nov J a n Mar May Jul Sep Nov O ate Figure C..... 35 ......12: O bserved and predicted rut depth values for AASHO sections 316..--------------—i—.......................2286-0. 320 D esign 0 127-0 0762-0 2032 Axle loads 214 .......S Ml*] # l«* A t Tt. ................ and 323...T Section 321 Design 0 1016-0 2286-0 2032 Axle loads 133 • S 40 . 3 1 8 .....3 2 2 ....185 Section 316 Design 0 127-0 0762-0 4064 Axle loads: 214 • T 40 • Section 317 Design 0 101643 0762-0 4064 Axle toads 1 3 3 ..it. Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sop Nov Date D ate Section: 322 Design: 0 1016-0.1524-0 3048 Axle loads: 1 3 3 ............ 30 - Q.......2032 Axle toads: 214 * T 40 ....3 1 9 . ------------------........

326.3046 Axle loads 214 -T 40 . -t--i— -------------------------------------.. .4064 Axle loads: 214 -T Section: 331 Design: 0.. •’ Nov J a n Mar May Jul Sep Nov J a n Mar May Jul S ep Nov O ate Figure C. f s s IX cc Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov D ate *.S .328.---------35 £ £ 30 • - E 30- £ c 25 - f 20 L I c ca 15 r xc> 10-! (£ Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate Nov Jan Mar May Jut Sep Nov J a n Mar May Jul Sep Nov O ate Section 328 D esign 0 127-0 1524-0 4064 Axle loads 214 . . .1016-0 0762-0. -----. 327. .--------------------. 35 • 30 > Nov Jan Mar May Jut S ep Nov Jan Mar May Jul Sep Nov D ate Nov J a n Mar May Jul Sep Nov J a n Mar Moy Jul Sep Nov O ate Section: 330 Oesign: 0....-------- S ection 325 Design: 0 1524-0 1524-0 2032 Axle toads: 1 3 3 . . ? E c.13: O bserved and predicted rut depth values for AASHO sections 324.r Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate 0 Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate 5 f • . 325. and 331.. Reproduced with permission of the copyright owner.1016-0 0762-0 4064 Axle loads 1 3 3 .S 40 — • .3 3 0 . 40 .T 40 Section 329 003*00:0. Further reproduction prohibited without permission.186 Section 324 O esign 0 1016*0 1524*0. « |* » I * *' Section 326 O esign 0 1524-0 1524-0 2032 Axle loads 214 • T Section 327 Oesign: 0 127-0 1524-0 4064 Axle toads 1 3 3 ..127-0 2286-0 3048 Axle loads: 1 3 3 .S 40 r — .. 3 2 9 .S 35 i £ 30 L E c.

14: O bserved and predicted rut depth values for AASHO sections 332.. Nov Jan Mar May Jul Sep Nov J a n Mar May Jut Sep Nov Dale Date Section 336 Design: 0 1524-0 0762-0 3046 Axle loads 214 .. .. -----. ......3 3 6 .. . .4 1 2 .S 40 ... . 411 Oesign 0 1016-0 0762-0 1016 Axle loads 100 .......T Section.1524-0 0762-0 3048 Axle loads: 133 ..j — . ..S 35 • - E 30 .. 333.._ 178 • T Section: 413 Design: 0. ~ 25 i- c...4 1 1 . .. and 413.....T Section 333 Design 0 1524-0 2286-0 4064 Axle loads 1 3 3 ..’ ' ■ Nov Jen Mar May Jul Sep Nov Jan Mar May Jul Sep Nov o 19L-L-----.. R eproduced with permission of the copyright owner. . .. . Further reproduction prohibited without permission. % 20 1 ►♦ I T i ...S 40 35 30 25 20 15 a. .... 3 3 4 ...J ' .. .1016 Axle load 40 .V . Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Nov Jan Mar May Jul S ep Nov J a n Mar May Jut Sep Nov Date Dale Section 412 Design: 0 1016-0 0762-0. .. ... ----. 10 10 5 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Date Date Section 334 D esign 0 1524-0 2286-0 4064 Axle loads 214 . ..T 4 0 ..0762-0 2286-0 1016 Axle loads: 100 • S 35 — 30 £ f 25 | 20 I£ 1150 ? £ r & 5 ■o i tr 5 0 Nov Jen Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Oate Nov J a n Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Data Figure C. .3 3 5 .*• * • 5 1.r ..187 Section 332 D esign 0 127-0 2286-0 3048 Axle loads 214 .. . .. - Section: 335 Design 0.— .. - 35 ■- 35 - 20f15k ' 5^-» ► .......

.. ..4 1 8 .... Further reproduction prohibited without permission........ I 20 L n I ^ § 1015 Nov Jan Mar May Jul Sap Nov Jan Mar May Jul Sep Nov Dale Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov D ate Section 420 Design: 0 0762-0.. j _ l l ...-■ ....... Nov J a n Mar May Jul Sep Nov Jon Mar May Jul Sep Nov Dale 0 » ... .... .............t L I ■ I —L I : .. M iZ — i i i ...T 40 ..1524-0 2032 Axle loads: 1 7 6 -T 40 Section 421 D esign 0 127-0 0762-0 2032 Axle loads 100 ..L ^ .18S Section 414 Oesign 0 0762-0 2286-0 1016 Axle loads 178 .. . 4 1 6 ........ i...S 40 35 35 30 30 25 20 20 15 £9 15 X 10 10 5 5 1 1 -1 0 gTfTl—:-----i... and 421...S 40 .. Section 415 Design 0 0762-0 0762-0 3048 Axle loads 100 • S 35 - 25 r ® 20 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov D ate Section 416 Design 0 0762-0 0762-0 3046 Axle loads 1 /8 * r Section 417 D esign 0 10164) 2286-0 2032 Axle toads 1 0 0 .......-l..j Nov Jan Mar May Jul Sep Nov J a n Mar May Ju l S ep Nov D ate F ig u re C. Reproduced with permission of the copyright owner... i . .L . — 30 i E E 6 c 25 ..2286*0 2032 Axle loads 178 -T Section 419 D esign 0 0762-0 1524-0 2032 Axle toads 100 . _ --------...........4 2 0 . 415....S 30 25 ' 8 20 tr 10 i Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date Section 418 Oesign 0 1016-0.4 1 7 ...... i t ...4 1 9 ..j ..15: Observed and predicted rut depth values for A ASHO sections 414. . ....

.. t. 6 £ Q.. .....S --------..4 2 5 ......---........ -----------... 40 .. .r--» 35 - 35 - m Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov « Nov J a n Mar May Jut Sep Nov J a n Mar May Jut S ep Nov D ate Oate Section 426 D esign 0 1016-0 1524-0 3048 Axle loads 1 7 8 -T 40 ~r . ....T 40 ---------------------------......r 40 •S Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate 70 -1 Section 425 Oesign 0 1016-0 1524-0 3048 Axle loads 100 • S ...........-----.-..... g £ 25 r S 2 cSc i <£ •••« Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov O ate Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Section: 428 Design: 0 127-0 2286-0 3048 Axle loads 178 . .. ....... 4 2 7 .— — ------..• . . and 429. 423. ...16: Observed and predicted rut depth values for AASH O sections 422......... E 11 1 *r i ' f “ i -i - Nov Jan Mar May Jul Sep Nov J a n Mar May Ju l Sep Nov Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Oate □ale F ig u re C.. 178 . 42 4 ....S □ale e 6 E £ £ a £ a 10 - i S> a.. . .. R eproduced with permission of the copyright owner... Further reproduction prohibited without permission... 426.... ...... Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov O ate Section 424 D esign 0 127-0 1524 0 1C16 Axle ! ----... 35 -• - £ - ' - - ..• Section 423 Design 0 127-0 1524-0 1016 Axle loads 100 ......4 2 8 .....189 Section 422 O esign 0 127-0 0762-0 2032 Axle loads.... ...T Section: 429 Design 0 0762-0 0762-0 3048 Axle loads 1 0 0 ... . ------------- Section 427 Oesign 0 127-0 2266-0 3048 Axle loads 1 0 0 .... S s £ o E CT> ••••< * cr a.... .. 35 30 .- 35 • 30 • E E £ Q.S 40.

....... and 443......... Further reproduction prohibited without permission......... ...... Section 437 O esign 0 1016-0 2206-0 1016 Axle loads 100 ... 0 127-0 0762-0.4 4 1 ..S 35 r ~ 30 - •f’ Nov J a n Mar May Jul S ep Nov Jan Mar May Jul S ep Nov Date Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Date F ig u re C ......4 4 2 .... 35.... 439 Design 0 127-0 0762-0 1016 Axle loads 100 ...17: Observed and predicted rut depth values for AASHO sections 430....r T i |0 " ■ m 5 Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep N qv Date ____ - V -rr M • H »•< «• >..4 4 0 .. 30 ► - E 30- £ a 20 25.190 Section 430 Design 0 0762-0 0762-0 3048 Axle loads 176 .. 5 2 cr9 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov 15...... I £9 X 15 10 5 0 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov O ute Date Section 442 D esign 0 0762-0 2286-0 3040 Axle loads 178 • T Section 443 O esign 0 1016-0 0762-0 3048 Axle loads 100 . Reproduced with permission of the copyright owner................ 35 10 30 > 3 . .. ...... 4 3 9 ..... „ .. .........S ....... ......S 40... 35 ~ E 30 - a 20....... 438.. 437. .. Nov Jan Mar May Jut Sep Nov Jan Mar May Jul Sep Nov Date Section 433 Design 0 1016-0 2266-0 1016 Axle loads 178 .. e §■ 25 >r I 20 — £ 20 X - - i.f 40 Section....1016 Axle loads 178 ..........5 ....S 40 — ...........---------------. #*• ........T 40 .... 10 > D ate Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section: 440 Design......T Section: 441 Design 0 0762-0 2286-0 3048 Axle loads 100 ..

.................................1■ .... Further reproduction prohibited without permission....4 4 8 ..... and 451.. sz 2 tr > y~*** i Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 446 O esign 0 127-0 1524-0 3048 Axle loads 178 • T 4 0 --------------------------------------35 - Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date ........................ ----............ 35 • 35 - .... 445................S 4 0 .... R eproduced with permission of the copyright owner...... .......4 5 0 .. ....... Section 449 Design 0 0762-0 1S24-0 1016 Axle loads 100 .......T Section: 451 O esign 0 0762-0 0762-0 2032 Axle loads: 100 • S 40 40 n — r ^ n r T T T - - i 35 • 35 i- 30*- 30 >25 l-iI 20 ^ 20 I 2 15 I. —.. ..30 • E .... ... .. 446...T 40............ .......- 15 l I :: ....... 448 D esign 0 127-0 2286-0 2032 Axle loads 178 ............... 4 4 9 .. .. 35 • E E n Q. i i..• Section: 447 O esign Q 127-0 2286-0 2032 Axle loads 100 • S 40 35 30 25 £ 20 20 15 X 10 10H 5 « Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section.... ...........30 r E * - 0 Nov Jan Mar May Jul S ep Nov Jan Mar May Jul S ep Nov Dote 25 20 t15 I10 . .191 Section 444 O estgn 0 1016-0 0762-0 3048 Axle loads 170 -T Section 445 O esign 0 127-0 1524-0 3048 Axle loads 100 ... I0 l 10 5 '■ r - Nov J a n Mar May Jul S ep Nov J a n Mar May Jul Sep Nov O ate 5 .......... 4 4 7 ......S 40 .i 0 Nov J a n Mar May Jul S ep Nov Jan Mar May Ju l Sep Nov O ate F ig u re C.....18: Observed and predicted rut depth values for AASHO sections 444.... . ..- Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Oato Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section: 450 O esign 0 0762-0 1524-0 1016 Axle loads: 178 ...........

.i Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Date Section <154 Dc 40 ■ • 101G-G 1524-0 2032 Axle loads l~Q .19: O bserved and predicted rut depth values for AASHO sections 452.• * 5 i...*.. R eproduced with permission of the copyright owner. 453. _______ »«>’ Section 469 Design 0 127-0 1524-0 2032 Axle loads 100 • S 40 -----. a: 10 • 10 **••* . .S 40 -----.T Section 455 Design 0 10160 1524-0 2032 Axle loads: 100 ..■ r~r—. and 471.r _ .T Section 453 Design 0 1016-0 1524-0 2032 Axle loads 100 • S 40 35 F 35 ? 30.._ .2032 Axle loads 178 * T Section 471 O esign 0 0762-0 2286-0 2032 Axle loads. 455. 5r 0 .• ■ 1 15 v - ^ 15.• ■ - 5 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov O ate o i -------------------a . f 25« 20 | • ■• ! 15. 100 • S 40 40 Oate 35 30 £ £ 25 z. .4 6 9 . 3 0 - ■ ..2032 Axle loads 178 • T 40 ----------.• 30 ■ 25 35 . ■ - 30 25 20 15 c.------------35 30- Nov J a n Mar May Jul S ep Nov Jan Mar May Jul S ep Nov Oate Nov Jan Mar May Jut Sep Nov J a n Mar May Jut S ep Nov Section: 470 Design: 0.. 454.- }25)-f -' 20 V. 2 o> 10r r - 20 - £ 10f - 5 U^ . . . * 20 S 15 a:a» 10 f aS 5 Nov J a n Mar May Jut S ep Nov J a n Mar May Jul Sep Nov D ate Nov J a n Mar May Jul Sop Nov J a n Mar May Jul S ep Nov Oate F ig u re C. . 4 5 6 .127-0 1524-0..4 7 0 .. -------- 35 ■ •■ . Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date 0 rn fZ T Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Oate Section: 456 Design: 0 1016-0 1524-0. Further reproduction prohibited without permission..192 Section 452 O esign 0 0762-0 0762-0 2032 Axle loads 1 7 8 .

25 . Nov J a n Mar May Jul S ep Nov J a n Mar May Ju l S ep Nov Date Figure C. .S 40 -------. 0 0762-0 2286-0 2032 Axle loads 1 7 8 -T Section 473 O esign 0 1016-0 1524-0 1016 Axle loads: 100.S E £ Date E a f 25 1- I | 20 - i a: I5 « rr x 10 - Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov D ate V J..r .cn 475 D esign 0 127-0 2286-0 1016 Axle loads 100 • S Section 40 .S 35 - £ 30 - - ~B »C* ...T Section 477 Design 0 1016-0 2286-0 3048 Axle loads 100 ... 5a» iOh< _. 473.. .----------------- 35 i q. 475. 0 127-0 2286-0 1016 Axle loads 178 . and 479..4 7 8 ..T Section: 479 Design 0 127-0 0762-0 3048 Axle loads 1 0 0 . r ..20: Observed and predicted rut depth values for AASH O sections 472....i 30 E | 25 f 20£I 15. 476.. 5 - Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Dale Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov Oate 474 Design 0 1016-0 1524-Q1Q15 Axle lead 3act.. _— . .. 35 .. . . Further reproduction prohibited without permission. 4 7 7 . R eproduced with permission of the copyright owner. 1o r r• - Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov D ate Ncv J a n Mar May Jul Sep Nov Jan Mar May Ju l S ap Nov Section: 476 Design: 0 1016-0 2266-0 3048 Axle loads: 178 . 4 7 4 . fi 20 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov O ate Nov J a n Mar May Jul S ep Nov Jan Mar May Ju t S ep Nov Oate Section 476 Design.193 Section 472 Design. . -----------------..

............ 51 ..4 8 7 ...4 8 3 .... 481. 4 8 5 ...........V S 10 K r T 5 #*♦ 0f Nov Jan Mar May Jul Sep Nov Jan Mar May Ju l Sep Nov Date Nov J a n Mar May Jul Sop Nov Jan Mar May Jul Sep Nov Section: 487 Design: 0 0762-0....S 35 i- 35 j— 30 • 6 £ 25... .S 40 .......S Section 488 Design: 0 0762-0 1524-0 3048 Axle loads 178 ....T 40 .....4 8 4 ...194 Section 460 O esign 0 127-0 0762-0 3046 Axle loads 178 • T 40 - Section 461 Design 0 1016-0 0762-0 2032 Axle loads 100 ...... R eproduced with permission of the copyright owner.........• • I 15. c cc 10 j...1524-0 3048 Axle loads: 100 ....--------------35 - - 30 - £ 30 > - 25 ~ 25 ‘ € 20 '■| 15...T Oate c r 30 £ 25 k ~ 25 ^ ..r ..... 35 • ~ 20 I £ f 20 h 5 E s» 15 r i f 10- I S' E alM * 4 Nov Jan Mar May Jut S ep Nov Jan Mar May Jut S ep Nov Oate Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov O ate Section 484 O esign 0 127-0 2206-0 1016 Axle loads 178 -T Section 465 Design 0 0762-0 0762-0 1016 Axle loads 100 ... and 488....-..... ® 20 -..— — . Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Nov J a n Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date Section: 462 Oesign 0 1016-0 0762*0 2032 Axle loads 178 .....* . ..— * — . Further reproduction prohibited without permission......21: Observed and predicted rut depth values for AASHO sections 480.. *a 1 5 - Nov Jan Mar May Jul S ep Nov Jan Mar May Ju l S ep Nov O ate Nov J a n Mar May Jul Sep Nov Jan Mar May Jut S ep Nov Date F ig u re C......... 4 8 2 ...... Section 483 Design 0 127-0 2266-0 1016 Axle loads 100 • S 40 ........

. r ..1 .. .. 5 7 3 ...5 7 5 ...5 7 4 ... .. 35 35 .1016-0 0762-0.. ... R eproduced with permission of the copyright owner.......T ---• -----------------------.. ................... .............. ...................... sn r^ .... Further reproduction prohibited without permission..... Date . and 576... 30 25 o& 20 a 15 a* 10 K 5 Nov J a n Mar May Jul S ep Nov J a n Mar May Jul Sep Nov Date 0 Nov J a n Mar May Jul Sep Nov J a n Mar May Ju l Sep Nov Date F igure C .. ........ 2 5....... ......... — r --------r—1 > • • ' > ■ . 11 1 1a»»".... 60 • S 40 r ....... ........• ....... .. ... ■1 .. 1 5 ' 1 1 1 1 i 1......... 570............ £ £ c...S Section 572 Design 0 0762-0-0 3048 Axle loads 142 .• ■ ....e|] r1.... 5 71.... 20 ■ .. ....... 40 - 35 r 35 - 30 r 30 • E E 25: ............ . ' ' ' . Section 574 Design 0 0762-0 0762-0 2032 Axle loads 142 • T 40 .. ■ ■ 1■ 'r i • 1•• .....3046 Axle loads: 80 . 0 101643...... . ... . [.......... ■ ■ ' ■ ... .T 40 r 40 35 - S.22: Observed and predicted rut depth values for AASHO sections 569. ... .30 E 20 1 1 . ..... ‘ ...\ .. ..S Section 570 Design 0 0762-0-0 3048 Axle toads' 142 .. : i ■ l ......... S 20 u X 10r Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Section.. 573 D esign 0 0762-0 0762-0 2032 Axlo loads. 30 r 25 i 20 > I IS i 15 i i• 1 ■ 10 10.57 2 ..195 Section 569 Design 0 0762-0-0 3048 Axle loads 80 ....0762-0 3048 Axle loads 142 .. : Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Date 0 Nov Jan Mar May Jul Sep Nov J a n Mar May Ju l Sep Nov Date Section: 575 D esign 0.......S 40 r 40 35 35 Section 576 O esign... 15- K 101 10 i 5: • 0 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Dale Section 571 Design 0 0762-0-0 3048 Axle loads 80 ....T .. .T .......... . j_ i.. ..

'■ IT 30 i 25 f 20 l '5 <r 10 5 Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Data 35 - 30 f 25 E | 20 I3 15 f.S Section 584 Design 0 1016-0-0 1016 Axle loads 142 ■T 40 r r I 40 35 I. 5 8 1 . R eproduced with permission of the copyright owner.I 10 f.. 578.. .i — Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Data Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section: S83 Design: 0 1016-0-0 1016 Axle loads 80 . Section: 582 Oesign: 0 127-0...5 8 2 .1524-0 3043 Axle loads 142 • T 40 ( 40 - .i -1 rI -1 5 Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate F igure C. 580.—.. i. 579.196 Section 577 O esign 0 1016-0 1524-0 2032 Axle toads 80 . Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate 35 i 30 - 30 ► ~ 25 a 20 h X 20 0 i— >— .23: Observed and predicted rut depth values for AASHO sections 577.--— •— . Further reproduction prohibited without permission..S Section 578 Design 0 1016-0 1524-0 2032 Axle loads 142 • T 35 • X 10r- Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate Section 579 D esign 0 1 2 /0 0 7 6 2 0 1016 Axle toads 80 * S Section 560 Design 0 127-0 0762-0 1016 Axle loads 142 -T 40 40 35 • 30 v 25 20 15 10 5 4 0 Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section: 581 D esign 0 127-0 1524-0 3048 Axle loads: 80 . and 584. 583..i .i ..S 35 E •- • ..

197
Section: 565 Oesign 0 0762-0 1524-0 1016 Axle loads 60 - S
40
t . _ ----------

Section 506 O esign 0 0762-0 1524-0 1016 Axte loads 142 - T
40 , - ............................
- ...............................

35 .

35 f

-

30 :

30)

25 . . .

.

25 -

20 • - • •

■ 20|
I «l|
C

15

10- , • -

a

,

a

i

10

r

5

51-

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
Oate

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov
Date

Section 567 Design 0 127*0*0 2032 Axle loads 60 • S
40 --------------- - , .
-..... . ....

Section 580 Design 0 127-0-0 2032 Axle loads 142 • T

35 • • •
30 25 ■ <- 20 - - 15 -

10 - •
r

5 . - .

f

*


0'
Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
Oate

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
Date

Section. 569 Design: 0 1016-0 0762-0 2032 Axle loads: 00 - S

Section: 590 D esign 0 1016-0 0762-0 2032 Axle toads. 142 - T
40
_ f . . . _____ ___________

35 •

35 f

~

E 30 •
I 28 •
I 20.

~

I

15 f

s

1 0 f- r 7

5?

.I

6
I
S
f
a

30 L

2520
'5
10
5

0

Nov Jan Mar May Jut Sep Nov J a n Mar May Jul Sep Nov
D ate

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov
□ ale

Section 591 Design 0127-0 1524-0 2032 Axle loads 00 - S
40 ■

~

25

Section 592 D esign 0 127-0 1524-0 2032 Axle loads 142 - T

-

25

0 rfrt- j- ±.
Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov
Date

—i—

_j—1—__ ,—- _—,

—i_U—.—j.—.

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov
D ate

F ig u re C.24: Observed and predicted rut depth values for AASH O sections 585, 586,
587, 588, 589, 590, 591, and 592.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

198
Section 593 Oesign 0 127-0 0762-0 3046 Axle loads 60 - S
40

Section 594 D esign 0 127-0 0762-0 3048 Axle loads 1 4 2 -T
40

35

35

30 ,

1-

r- - - ■ -

.............' i ' —

.

S>
a? 10 r

4 .'
1

^
- - - ijtl

*
.. . . . .
.
Nov Jan Mar May Jut Sep Nov Jan Mar May Jut Sep Nov
D ate

0 ^ * * .♦
Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
Oate

Section 595 Design 0 1016-0 1524-0 1016 Axle loads 00 - S
40
.

Section 596 O esign 0 1Q16-Q 1524-0 1Q16 Axte loads 142 • T
4 0 .............................................................

£
E

c.

* 20

~

I
5
£9
cc

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov

Nov Jan Mar May Jul S ep Nov Jan Mai May Jul Sep Nov

Oate

Date

Section: 597 Oesign 0 1016-0 0762-0 2032 Axle loads 80 • S
40 -

Section. 598 O esign 0 1016-0 0762-0 2032 Axle loads 142 • T

30 E
E

30 .

c

Q.

5 15.

E

g>

X

10

r

Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
Date

Nov Jan Mar May Jul S ap Nov Jan Mar May Jul Sap Nov

Section: 599 Oesign: 0 0762*0 0762-0 1016 Axle loads 80 • S
40 r - r —
^
r
..................... ..

Section: 600 D esign 0 0762-0 0762-0 1016 Axle loads 142 • T
40, ^
'

D ate

35 ~

.

30f

H r1 «-. . .i . . .• . -i. . .•' f

L ;

i

20 1

rj r i : i

5
£
a tofi- r t
j_ ^

.
i

~ ■

i-

' ! ! ! ! ! !

1

> i

i i

|

;

!

1
r

r

: ! :

t

i ~i ' r

t

r

~

i

, ■ , ,

1
;

i. l : . i
O'
Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov

Oate

;

*

;

'

1

-

r

r -i

-

r

r

■ .............................................

.........................
:

f

i
r
i

. •
t 1
i .

.
i

i

i

.

i

i

i

'!

1

■ !

l


'

:

,

:

: :

1

,

Nov Jan Mar May Ju l Sep Nov Jan Mar May Jul Sep Nov
Oata

F ig u re C.25: Observed and predicted rut depth values for A A SH O sections 593, 594,
595, 596, 597, 598, 599, and 600.

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

199
Section 601 Design 0 0762-0 1524*0 3048 Axle loads: 80 - S
40

Section 602 Design 0 0762-0 1524-0 3048 Axle loads 142 • T
40

35

35 -

30

E
E
r
H.

25

20

i 20,

15

10
5

0"
Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov

---- ..... ... _ ^ . •: . .

...t. .

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov
Oate

Section 603 Design 0 IQISG-Q 2043 Axle loads. 80 • S

S eato n 604 Q estgn 0 1016-0-0 3048 Axle loads 142 - T

40

40

35

35 •

30
25
a
91
■o

20

20•

15

,••••• »•

10
5

Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov

0
^ • - - .................... . — 4 - . i- ~ . . .i
Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
Date

Section 605 Design 0 127-0*0 1016 Axle loads 60 - S

Section 606 Design 0 127-0-0 1016 Axle loads 142 -T

40 ,
35 L
30 f-

E
E
£

25 1

30;

20(15 L

10i-

Nov J a n Mar May Jul Sep Nov J a n Mar May Jul Sep Nov
Date

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
O ate

Section: 607 Design: 0 0762-0-0 2032 Axle loads 60 • S
40 r r

Section: 606 Design: 0.0762-0-0.2032 Axle loads: 142 • T
40 r
T
“T T 7 T
"7"

1T l ’T*

35 L r
?
F

c

35r
£• 30k

30)- -

E

2 5 --

o
■Q 20 L

I

I

f

20 U

15-

£»
ce 10 L 5 ••

i -i -•
4

i

-i

Nov J a n Mar May Jul Sep Nov J a n Mar May Jut Sep Nov
D ate

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov
O ate

F ig u re C .26: Observed and predicted rut depth values for AASHO sections 601, 602,
603, 604, 605, 606, 607, and 608.

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

200
Section 615 Design 0 127-0 1524-0 1016 Axle loads 60 - S

Section: 616 Design 0 127-0 1524-0 1016 Axle loads 142 - T

E 30 ,

~

30 -

E

c

5.
4)
13

1

Nov J a n Mar May Ju( Sep Nov Jan Mar May Jut Sep Nov
Date

Nov Jan Mar May Jut Sop Nov Jan Mar May Jul Sep Nov
Date

Section 617 Des*gn C G762-G 0752-0 304S Axle loads 60 * S
40 -

Section

Nov J a n Mar May Jul S ep Nov Jan Mar May Jut Sep Nov
Date

618

Design 0 0762-0 0762-0 3048 Axle loads 142 - T

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
O ate

Section: 619 Design 0 1016-0-0 2032 Axle loads 80 - S
40 -

Section 620 Oesign 0 1016-0-0 2032 Axle loads 142 • T

35 ■
30 -

E

30 •

6

25 r

c

B.

20 f-

*
I

I 151
S 10[
I

C

S>

(X

0 flL-n
Nov J a n Mar May Jut Sep Nov Jan Mar May Jul Sep Nov

Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov

O ate

D ate

Section: 621 Design: 0 127-0-0 3048 Axle loads 60

Section: 622 Design: 0.127-0-0 3048 Axle loads 142 - T

40
35

S

30

-

30

25

-

25

20
15

10
5

0
Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov
D ata

Nov Jan Mar May Jul Sop Nov J a n Mar May Jui Sep Nov
D ate

F ig u re C .27: Observed and predicted rut depth values for AASHO sections 615, 616,
6 1 7 ,6 1 8 ,6 1 9 , 620, 621, and 622.

R eproduced with permission of the copyright owner. Further reproduction prohibited without permission.

.... 625....1524-0 3048 Axle loads 80 • S Section 626 O esign 0 1016-0 1524-0 3048 Axle loads 142 ......... 2 201 15 10 ~ r 5 L.... ........T Nov J a n Mar May Jul S ep Nov Jen Mar May Jul Sep Nov Date Nov Jan Mar May Ju l S ep Nov Jan Mar May Ju l Sep Nov Data 5 1<r F ig u re C ..0 Nov Jan Mar May Jul Sop Nov Jan M3r May Jul Sep Nov O ate Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Dale Section: 629 Design: 0 127-0 1524-0 1016 Axle loads 80 .. and 630.. 627..... .... 35 35 • 30 ..... Further reproduction prohibited without permission.S 40 Section 626 O esign 0 1016-0 0762-0 1016 Axle loads 142 ......30 E 25 - 20 E 25. Reproduced with permission of the copyright owner.......... 6 2 8 . Q...... 35 .. 626.... aa *> 20 cr 10 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov O ate Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 625 Oesign 0 1016-0........ 2 20 2 S 2> cr Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov O ate Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 627 Oesign 0 1016-0 0762-0 1016 Axle loads 60 . 30 - E E c..S Section 630 D esign 0 127-0 1524-0 1016 Ajtle loads: 142 .......... 624..... „ .30 E E E c.......S Section 624 O esign 0 0762-0 1524-0 2032 Axle loads 142 -T 40 .28: O bserved and predicted rut depth values for A ASH O sections 623......................... -T ...... ......T 40 r .......201 Section 623 Oesign 0 0762-0 1524-0 2032 Axle loads 80 ........T 4 0 ----------------...6 2 9 ..

.29: O bserved and predicted rut depth values for AASH O sections 631..S 40 r 35 [ ii 40 : :I i ... S 20 r ^ I 15f^ £ 10 r T Nov Jan Mar May Ju l S ep Nov J a n Mar May Jul Sep Nov Date F ig u re C.. ■ L ...--------.S ^ ... 1 IX 10 ^ ' r i e -i _i‘ r r r i n ...S 40 ( Section 720 D esign 0 0254-0 1524-0 1016 Axle loads 27 ... .... . R eproduced with permission of the copyright owner... ..“ r r t i ■*i~ ' r r t.........T 35 • E E 20 rl e» x Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section 717 Design 0 0264-0 0/62-0 1016 Axle loads 9 . ....7 1 9 ........ and 727...... 35 i - 30 I s S> X 10 U T i 10 • 5 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jon Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Date Section: 721 Design: 0. 721. <r 10 • * Nov J a n Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Oate • Nov Jan Mar May Jul S ep Nov Jan Mar May Jul Sep Nov Oate Section: 719 Design 0 0254-0 1524-0 1016 Axle loads 9 ......• • ........... .30 r E % 20 .. .202 Section 631 Design Q 127-0 0762-0 2032 Axle loads 80 . i : 1 1 ■1 [ ' 1 1 1 : 1 1 1 i 1 * . ---------- Section 710 Design 0 0254-0 0762-0 1016 Axle loads 27 ... 40 .■ ! «5........S t * ' ’ 1• ' ** 1> Section: 727 Oesign.S Section 632 D esign 0 127-0 0762-0 2032 Axle loads 142 .......7 1 8 ...0254-0-0 Axle loads: 9 .... 1ll. l i ..... < i ^ 0 Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Date 35 #25........... 0 0254-0-0 1016 Axle loads 9 .S 40 .. 7 1 7 ... Further reproduction prohibited without permission... ■£ 30 ■ E •■ ....S 40 ..... 632... 720......~j 1• 1 5 t i ■i i .. ..

• ■ £ S O 20 i.......... and 742.. .... 730..0762-0-0.. 35 35 • 30 - 30- 25 | 25.S - Section: 742 Design: 0 0508-0 0762-0 1016 Axle loads 27 . 737...-• ♦ - 25 4) 20 15 ...r s'\-l 5 0 w Nov Jan Mar May Jul S ep Nov Jan Mar May Jul S ep Nov Nov Jan Mar May Jui S ep Nov Jan Mar May Jul Sep Nov D ate D ate Section: 741 Design: 0 0506-0 0762-0....... Further reproduction prohibited without permission. ..S Section 730 D esign 0 0506-0-0 1016 Axle loads 27 • S ....■: 10I..1016 Axle loads: 9 • S 6 e £ I 5 Section 740 Design: 0 0762-0-0 1016 Axle loads 27 ..-.. 07 Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov Nov Jan Mar May Jut S ep Nov Jan Mar May Jul Sep Nov Dale Date Section: 739 Design: 0..5 - £ 10 • a 10 5Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov D ate Nov Jan Mar May Jul S ep Nov Jan Mar May Jul S ep Nov Date Section 737 Design 0 0 5 0 8 0 1524-0 1016 Axle loads 9 ..30: O bserved and predicted rut depth values for A A SH O sections 729.. 739. 40 40 .. 10 5 5 _ . 741. 10 ..... R eproduced with permission of the copyright owner.. 740..203 Section: 729 Oosign 0 0508-0-0 1016 Axle loads 9 ... 738.....1016 Axle loads 9 .S 40 35 - 35 30 . .S f 2 i tx Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov D ate Nov J a n Mar May Jul S ep Nov Jan Mar May Ju l S ep Nov D ate F igure C...• S 15f-: .. 30 e6 25 .• 20 f 20 r • 15 3 £ .S 40 Section 738 Design 0 0508-0 1524-0 1016 Axle loads 27 • S 35 30 E £ £ 25 2«L 20 •o 5 15 c9c> 10 a a £ 5....

749... S-i «f«t Nov Jan Mar May Jul Sep Nov J a n Mar May Jul Sep Nov O ate N ov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov D ate Section: 755 Design: 0 0254-0 1524-0 Axle loads: 9 ...L ^ -1 ------_ 1_ - ..... -i«‘ r Nov J a n Mar May Jul S ep Nov J a n Mar May Jul Sep Nov O ate Nov Jan Mar May Jul Sep Nov J a n Mar May Jul S ep Nov D ate F ig u re C... i ~r ........ 745............• • • Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 746 Design 0 0762-0 0762-0. 1 .•..S 40 35 - 35 r - £ 30 • £ 25 • - E « 20- « 20 .S Section.... 0 0254-0 1524-0 Axle loads 27 • S • : . ...31: O bserved and predicted rut depth values for AASHO sections 743....... _ Section 744 Oesign 0 0254-0 0762-0 Axle loads 27 .... « • „ ' “ ■f......... ...... . ....S Section 750 Oesign: 0 0762-0 1524-0 Axle loads 27 • S O ate X l Okf r ......./ P ---- .. . 744........... i .. i... 746.1524-0 Axle loads: 9 . ..... • I '5....C ' o> £ 10 ........... 756 Design........ . Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Oate Section 745 Design 0 0762-0 0762-0 1016 Axle loads 9 . ■ ■ I : ....♦ -• •* . : ...204 Section 743 Oesign 0 0254-0 0762-0 Axle loads 9 • S 40 — . h ► i > . . .: ......... r r : » ...........M t V : i . 1 .......... ■ >. Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Section: 749 Oesign 0 0762-0....S 40 35 30 f 20 20 % 15 E £P 15 10 10 5 o •' v* * 0 ........ ....... E X 10at 5 ~ 0 30 - wr......-i ..S 5 r ... . ?% isrr?~r........... and 756... 1 (.......1016 Axle loads 27 . Further reproduction prohibited without permission.... ......... 755........ 750........I .... Reproduced with permission of the copyright owner..... .. i.. ...

. 764. 760. . . ... 10- 35 - 6 ! 25 r I 20'.. R eproduced with permission of the copyright owner.. . ..S Section: 772 Design.. 771. 5 1 Nov Jan Mar May Jul S ep Nov Jan Mar May Jui Sop Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov O ate Section 771 D esign 0 0508-0-0 Axle loads: 9 . .... ... 769.’— 35 I...— 0 — Nov Jan Mar May Jul Sop Nov Jan Mar May Jui Sep Nov Date ‘ ‘ Sep ~ Nov Jan Mar May Jui Nov Jan Mar May Jul Sep Nov Date Section 769 Design 0 0762-0-0 Axle loads 9 .. Nov Jan Mar May Jul S ep Nov Jan Mar May J i i Sep Nov D ate Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Date Section 703 O esign 0 0762-0 1524-0 1Q16 Axle loads 9 ..rrr...S .S .....32: Observed and predicted rut depth values for AASHO sections 759.. 770..... 763..205 40 Section 759 Design 0 0506-0 0762-0 Axle loads 9 .30 E ^25 | 20 I 10 15 e 10 oc 5 I- l ofc Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul S ep Nov Oate F ig u re C. 10 5 Q#11***.S 40 35 - 30 i- S 15 I I a.... and 77... . Section 760 O esign Q 0508-0 0762-0 Axle toads 27 . Further reproduction prohibited without permission....— i.S 40 r ...... . --— . 0 0500-0-0 Axle loads: 27 ..«■ -■ —▼ o ^ *— « • • . 40 Section 764 Oesign 0 0762-0 1524-0 1016 Axle loads 27 • S 40 35 • 35 • 20 t- « 20 5 0* — ....S Section 770 Design 0 0762-0-0 Axle loads 27 ♦ S 40 35 v - 30 r | 25 i I 20' 5 | £ 10 r .. .

206 Section 773 Design: 0 0762-0 0762-0 Axle loads... S> X 5r 10 5 Ai ■» 7 * ohrr— r 0 Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Section 775 Oesign 0 0508-0 1524-0 Axle loads 9 . 774.. . 9 . 10 ■ 5r XS> —-"A —■ Nov J a n Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Date F ig u re C..S 40 35 i ....■ ... £ 25- 20 .• - .. 35 • 30 ■ . ..• • 15 r 10 ^ .. 15 . . R eproduced with permission of the copyright owner.S Section 774 Design 0 0762-0 0762*0 Axle loads 27 . . .. 775.. 20 ■ .33: Observed and predicted rut depth values for AASHO sections 773... . ......S Section 776 Design 0 0508-0 1524-0 Axle loads 27 • S 40 35 ■ 30 25 S..... 20 15 V c. .... Further reproduction prohibited without permission.... and 776... . ......... ....30 ■ E 25 . ..

W esT rack The figures in the following pages o f this appendix show the observed and predicted rut depths versus time for all the W esTrack sections. R eproduced with permission o f the copyright owner. Further reproduction prohibited without permission .207 A ppendix D: O bserved vs. Predicted R ut Depths .

52. VFA = 68.49.3. AVi = 11 2. AVi = 5 3. AVi = 14 9.l: O bserved and predicted rut depth values for W esTrack sections 1 to 8.5 40 o 100 200 Day 300 400 Section 8: Coarse. Gl = 5 49. AVi = 6. VFA = 86. VFA = 73 5 40 40 30- 30. Gl = 1.9. Gl = 5 00.1 Section 2: Fine. Gl = 1. AVi = 7 9. Fine. VFA = 45. 400 . Reproduced with permission of the copyright owner. Gl = 2.86.0 40. AVi = 6.3. CL > ■0D I o 100 200 Day 300 400 Section 7: Coarse. VFA = 66.47. Gl= 1 44. • — Otoseved Predicted 30 • CL 101 0 100 200 Day 300 400 F ig u re D .2 200 300 400 Day Section 4: Fine. VFA = 61 1 40 40 30 30 ■ 20 200 300 400 Day Section 5: Coarse.6. VFA = 77 6 300 400 Section 6 Coarse. Gl= 1. VFA = 57 2 40 40 30 30 £ a 10 20- 10 ■ 100 100 200 300 400 Day Section 3. Gl = 4. AVi = 11.20 8 Section 1: Fine.36. AVi = 9. Further reproduction prohibited without permission.

VFA = 100 0 40 • 0 100 200 300 400 Day Section 11: Fine-plus. Fine-plus. 30 20 20 10 10 0 0 400 100 200 Day 300 400 0 0 100 200 Obseved Redicted 300 Day F igure D. 100 200 Day 300 Section 16: Fine.6. 10' • • • • 200 300 400 Day Section 13: Fine-plus. Rut depth (mm) 30: 20. AVi = 9 2. Gl = 1 43.7. Gl = 1. Gl = 1 43. Gl = 1. R eproduced with permission of the copyright owner. Gl = 1.85.2: O bserved and predicted rut depth values for W esTrack sections 9 to 16.80. AVi = 7 5. AVi = 1 6. Further reproduction prohibited without permission. VFA = 57. VFA = 84 0 Rut depth (mm) 30' 0 100 200 300 400 D ay Section 12. VFA = 87 5 30 • 20. AVi = 11 1. AVi = 5 8.85.38. AVi = 8. VFA = 7 1 2 40 ' Rut depth (run) Section 9 Fine-plus. 400 .5 40. AVi = 12. AVi = 2 7. VFA = 93 1 4 0 ------------------. VFA = 63.------- 200 300 400 Day Section 14: Fine.209 Section 10: Fine-plus.6 40 Rut depth (rrm) • — 30. Gl = 1 19. Gl = 1. Gl = 1. 0 Section 15: Fine.-------------------------------------------. VFA = 77 7 40.70.

2 .8 Section 24: C o a rse .4. VFA = 74 7 40 30 30- 10- 0 100 200 300 400 100 200 300 400 Day Section 23: Coarse. AVi = 6 3. AVi =6 5 .0 40 40 30- — 30 : E E f 20 20- i- cu ■o tr 1 0 : 200 300 400 Day Section 19: Fine-plus. Gl=2. VFA= 80. AVi = 5 2 . 30' 30 10- 4) "O 2 10 0 - 0 100 100 200 300 400 Day Section 21: Fine-plus. Gl = 2.210 Section 17 Fine. VFA = 66 0 Section 18: Fine.8. AVI = 2. AVi = 10 0 VFA = 87 1 40 .. VFA = 9 5 1 40 200 300 400 Day Section 20 Fine-plus. 400 . Gl = 1 59. Gl = 1 24. Reproduced with permission of the copyright owner. Gl = 3 78. Further reproduction prohibited without permission. Gl = 1.2 40 ---------------------------------------------------------------------. VFA = 79.3: Observed and predicted rut depth values for W esTrack sections 17 to 24. VFA = 96. • i Obseved • Predicted 30 I f •o K 1010 20 10 100 200 Day 300 400 0 0 100 200 300 Day F ig u re D. VFA = 71. AVi = 2. AVi =11 6. Gl = 1 59.49.7 40 200 300 400 Day Section 22: Fine-plus. Gl ^ 5 4 .36. AVi = 6. .03 .

7. Gl = 4. R eproduced with permission of the copyright owner. .211 Section 25: Coarse.6 Section 26: C o a rse . AVi = 2. Gl = 5.02.4: Observed and predicted rut depth values for W esTrack sections 25 and 26.28. Further reproduction prohibited without permission.3 40 40 • Rut depth (mm) 30 20 Observed FYedicted . AVi = 10. VFA= 62.— 30 a a - CC 10 0 100 200 Day 300 400 300 400 F igure D. VFA = 87.3.