1
EXPERIMENTAL STUDY OF MEMBRANE FATIGUE RESPONSE FOR ASPHALT 2
MULTISURFACING SYSTEMS ON ORTHOTROPIC STEEL DECK BRIDGES 3
X. Liu1, G. Tzimiris2, A. Scarpas3, R. Hofman4, J. Voskuilen5 4
5
(1) Corresponding author 6
Section of Road and Railway Engineering, Delft University of Technology 7
Stevinweg 1, 2628 CN Delft, the Netherlands 8 Phone: + 31 (0)15 27 87918 9 Email: x.liu@tudelft.nl 10 11
(2) Section of Road and Railway Engineering, Delft University of Technology 12
Stevinweg 1, 2628 CN Delft, the Netherlands 13 Phone: + 31 (0)15 27 89388 14 Email: g.tzimiris@tudelft.nl 15 16
(3) Section of Road and Railway Engineering, Delft University of Technology 17
Stevinweg 1, 2628 CN Delft, the Netherlands 18 Phone: + 31 (0)15 27 84017 19 Email: a.scarpas@tudelft.nl 20 21
(4) Rijkswaterstaat, Centre for Traffic and Navigation 22
Schoenmakerstraat, 2628 VK Delft, the Netherlands 23 Phone: + 31 (0)887982284 24 Email: rob.hofman@rws.nl 25 26
(5) Rijkswaterstaat, Centre for Traffic and Navigation 27
Schoenmakerstraat, 2628 VK Delft, the Netherlands 28 Phone: + 31 (0)88 7982304 29 Email: jan.voskuilen@rws.nl 30 31 Submission Date: 29/06/2013 32 Word Count: 33 Body Text = 3615 34 Abstract = 135 35 Figures 12×250 = 3000 36 Tables 3×250 = 750 37 Total = 7500 38 39
ABSTRACT 40
In order to adequately characterize the fatigue response of the various membranes with 41
surrounding multilayer surfacing layers on orthotropic steel decks and collect the necessary 42
parameters for FE modeling, the details of the cyclic Membrane Adhesion Tester (MAT) are 43
introduced. The fatigue damage in membrane interface is related to the amount of dissipated work 44
computed by using the measurement of actuator load and piston deformation during the loading cycle. 45
The dissipated work, which is equivalent to the lost part of the total potential energy of the membrane, 46
has been utilized to explain the incremental damage during the testing. Furthermore, using the 47
experimental data obtained from MAT, ranking of the bonding characteristics of various membrane 48
products is demonstrated as well as the role of other influencing factors, such as the types of substrate 49
and test temperatures. 50
Keywords: membrane; orthotropic steel deck bridge; fatigue; dissipated energy; adhesive 51 bonding strength 52 53 INTRODUCTION 54
Orthotropic steel deck bridges (OSDB) are widely used in most of the major long span 55
bridges around the world. The lightweight and flexibility make OSDB a cost-effective solution for 56
cases where a high degree of pre-fabrication or rapid erection is required (Gurney, 1992), in seismic 57
zones, for movable bridges, long-span bridges and for rehabilitation to reduce bridge weight (Mangus 58
and Sun, 1999). 59
An OSDB consists of a deck plate supported in two mutually perpendicular directions by a 60
system of longitudinal stiffeners and transverse crossbeams. Usually, the deck plate is surfaced by 61
bituminous wearing courses. It is known that surfacings except of their function of skid resistance or 62
waterproofing, reduce the stresses in the steel structure. In the Netherlands, an asphaltic surfacing 63
structure for OSDB mostly consists of two structural layers. The upper layer consists of porous 64
asphalt (PA) because of reasons related to noise hindrance. For the lower layer, a choice between 65
mastic asphalt (MA), or guss asphalt (GA), can be made (Medani, 2006). Two layers of membrane 66
are required to bond the two aforementioned structural layers. Earlier investigations have shown that 67
the bonding strength of membrane layers to the surrounding materials has a strong influence on the 68
structural response of OSDB. The most important requirement for the application of membrane 69
materials is that the membrane adhesive layer shall be able to provide sufficient bonding to the 70
surrounding materials. 71
In order to characterize adequately the adhesive bonding strength of membranes with 72
surrounding materials on orthotropic steel bridge decks, a Membrane Adhesion Tester (MAT) device 73
has been developed by Delft University of Technology and presented in the 92nd TRB annual meeting 74
(Liu et al. 2013) and in the technical report of this project (Liu & Scarpas, 2012). A total of eight 75
types of membrane products, representing the most commonly used for waterproofing in OSDB 76
constructions, have been tested under monotonic static loading conditions on different substrates. The 77
effects of temperature on the bonding characteristics of the membrane are investigated. By comparing 78
the critical strain energy release rate (Gc values) of different membranes at the same test condition, it 79
was concluded that the test product labeled C1 performs quite well as bottom membranes while the 80
test products labeled A2 and C2 represent the best choices for top membranes in Dutch OSDB, Table 81
1 & 2. 82
The monotonic MAT tests provided a fundamentally sound, mechanistic methodology for the 83
expedient ranking of the bonding characteristics of membrane products. In the second phase of this 84
project, the MAT device was modified to enable the investigation of the fatigue response of the three 85
top ranked membrane products under cyclic loading conditions at different temperatures. 86
In this paper, the cyclic MAT tests are presented. The characteristics of the tested membrane 87
are briefly introduced. The concept of “dissipated energy” and its utilization for quantification of the 88
damage induced in the membrane due to cyclic MAT loading are discussed. 89
In the last part of this paper, experimental results of the selected membrane products on 90
various substrates tested at two different temperature conditions (100C and 300C) and three different 91
cyclic loading levels (150N, 250N and 350N ) are presented. The values of dissipated work for each 92
membrane interface are compared, as well as the relationship between the membrane debonding 93
length and the number of load cycle. 94
EXPERIMENTAL SETUP AND TEST CONDITIONS 95
The membrane fatigue tests were performed at two temperatures (100C and 300C). The fatigue tests at 96
100C, were performed with a sinusoidal loading F ranging between min
F
50N
andF
max=150N at a 97frequency of 5 Hz for 432 000 cycles. Fmax was increasing every 432 000 cycles, starting from 150N, 98
then 250N and finally, 350N, see Figure 1. 99
FIGURE 1 Schematic show of fatigue loading scheme 100
For the fatigue tests at 300C, the sinusoidal loading F varied from min
F
50N
to 101max
F
100N
at a frequency of 5 Hz. The number of applied load cycles was 864 000. 102In order to run the tests under temperature controlled conditions the set up needed to be 103
properly insulated. For this project, a climate chamber was used to enable testing under different 104
temperatures. 105
The MAT setup is capable of operating in both a monotonic or a cyclic mode. For cyclic tests, 106
the maximum allowed load is 500N. The frequency range is 1 – 5 Hz and the maximum allowed 107
displacement from the bottom position is 150mm. The schematic diagram of MAT setup is shown in 108
Figure 2. 109
111 112
FIGURE 2 MAT set-up 113
LIST OF MEMBRANE PRODUCTS AND THEIR MECHANICAL PROPERTIES 114
Product A1 and A2 are waterproof membranes manufactured with SBS(styrene butadiene styrene) 115
elastomeric bitumen and internally reinforced with a non-woven polyester textile. These two products 116
are implemented on concrete decks, steel decks, sand asphalt or asphalt concrete. Product A1 is 117
applied on the steel plate, while product A2 is applied on the Guss aspahlt. 118
Product A1 and A2 can be bonded to the prepared substrate by melting the film on the 119
membrane surface and softening of the bitumen. Details of the product specifications can be seen in 120
Table.1. 121
Table 1 Specifications of product A1 and A2 from company A 122
Test and specification Units Standard
A1 A2 Nominal values Critical values Nominal values Critical values
Main surface thickness mm EN 1849-1 4 3.8 4.8 4.6
Tensile strength at break
(20oC,100mm/min) N/5cm EN 12311-1 950 820 950 820
Elongation at break
(20oC, 100mm/min) % EN 12311-1 40 35 40 35
There are two types of membranes from company C. Product C1 is used only as a bottom 123
membrane, whist product C2 can be used both as top and bottom membranes in asphalt surfacing 124
systems on steel bridge decks. 125
Product C1 is a 2.4 mm thick single-ply membrane, with non-woven polyester fleece. This 126
product is used for the single-ply sealing under stone mastic asphalt, mastic asphalt or bituminous 127
concrete. 128
Product C2 is a 4.7 mm thick single-Ply membrane, with 1.5 mm strong fleece. This 129
membrane is provided with a modified bituminous mass of 1.6 mm thickness on both sides. Product 130
C2 is a waterproof membrane for bridges, and provides high resistance to traffic loading. The details 131
of specifications for products C1 and C2 are shown in Table 2. 132
Table 2. Specifications of products C1 and C2 from Company C 133
Test and specification Units Standard C1 C2
Thickness mm EN 1859-2 2.4 4.7
Tensile strength MD/TD N/50 mm ISO 527 1350/1150 1350/1150 Elongation at tensile strength
MD/TD % ISO 527 50/70 50/70
134
SPECIMEN TYPES 135
In the Netherlands an asphaltic surfacing structure for orthotropic steel bridge decks mostly consists 136
of multilayers, see Figure 3. The upper layer consists of Porous Asphalt (PA) for noise reduction. For 137
the lower layer a choice between Mastic Asphalt (MA) or Guss Asphalt (GA), can be made. One 138
membrane layer is utilized to bond the surfacing layers together and an other one to bond the 139
surfacing layer to the steel deck . In order to characterize the interface adhesive bonding strength 140
between various membrane products and the surrounding asphalt and steel material layer, four types 141
of specimen, i.e. steel-membrane specimen (SM1), Guss Asphalt concrete-membrane specimen (GM1 142
and GM2) and Porous Asphalt-membrane specimen (PM2) were tested. The GM system consists of 143
two interfaces, with membrane-1 at the bottom of the Guss Asphalt (GM1) and membrane-2 at the top 144
of the Guss Asphalt (GM2), see Figure 3. Therefore two types of GM specimens have been 145 investigated. 146 147 148 149
FIGURE 3 Schematic of a typical Dutch asphalt surfacing system on a steel bridge deck 150
Table 3 illustrates the combinations of membrane materials and surfacing layers tested in this research. 152
Products A1 and C1 have been tested only at SM1 and GM01 interfaces, product A2 only at GM02 153
and PM02 interfaces while C2 product in all interfaces. 154
155
Table 3 Product/interface combination 156 Membrane type A C A1 A2 C1 C2 SM1 √ √ √ GM01 √ √ √ GM02 √ √ PM02 √ √
DISSIPATED ENERGY APPROACH FOR FATIGUE ANALYSIS OF MAT 157
Dissipated energy was used as an indicator of damage in asphalt materials (Van Dijk 1975; Van Dijk 158
and Visser 1977). These researchers postulated that the fatigue life depends on the accumulation of 159
dissipated energy from each load cycle. In later studies, damage was related to the rate of change in 160
dissipated energy from one cycle to the next (Carpenter and Jansen 1997). 161
In this study, the dissipated energy concept has been utilized for MAT cyclic loading tests to 162
characterize the fatigue life of membrane products bonded on the different substrates at different 163
temperatures and loading levels. The fatigue damage in the interface between membrane and substrate 164
is related to the amount of dissipated work computed by using the measurement of the actuator load 165
and the membrane deformation during each loading cycle. The dissipated work per loading cycle, 166
which is equivalent to the lost part of the total potential energy supplied to the membrane by the 167
actuator per cycle, was used in this study as a measure of the incremental damage in the interface 168
between the membrane and substrate during the testing. 169
Dissipated Energy Concept
170 171
Applying a load to a material, the area under the stress-strain curve represents the energy being input 172
into the material. During the loading-unloading process, if the unloading curves do not coincide with 173
the loading but trace different paths, an energy loss is happened within the material. Part of the energy 174
is dissipated out of the material system due to the external work, in the form of mechanical work, heat 175
generation, or damage. 176
The dissipated energy from cycle loading can be determined by calculating the energy losses 177
associated with the phase angle, see Figure 4 (a). The area of the hysteresis loop in Figure 4(b) 178
represents the dissipated energy and the following equations can be used to calculate its value in a 179
linear viscoelastic material. 180
181
FIGURE 4. Oscillating stress, strain and phase angle. (b). Hysteresis loop (one load cycle) 182
i i i i DE sin (1) 184 where 185 i
DE
dissipated energy in cycle i; 186i
=
stress level in cycle i; 187i
=
strain level in cycle i; 188i
=
phase angle betweenσ
andε
in cycle i; 189When stress
σ
is not a directly measurable quantity, the above equation can be expressed in 190terms of dissipated work as: 191 i i i i DW F sin (2) 192 where 193 i
DW
dissipated work in cycle i; 194i
F =
force level in cycle i; 195i
=
displacement level in cycle i; 196i
=
phase angle between F andδ
in cycle i; 197The relative change in the amount of the energy dissipated is directly related to damage 198
accumulation. A low amount of relative change in energy dissipation can be found either in high 199
fatigue resistance materials, low external loading amplitudes, or both. Such relative change in 200
dissipated energy represents the total effect of fatigue damage without the necessity of considering 201
material type, loading modes and severity separately, Shen and Carpenter (2007). 202
This concept was first initiated by Carpenter and Jansen (1997) who suggested using the 203
change in dissipated energy to relate damage accumulation and fatigue life. The work was refined and 204
expanded by Ghuzlan and Carpenter (2000), and then well applied and verified by Carpenter et al. 205
(2003) who used the ratio of dissipated energy change (RDEC) as an energy parameter to describe 206
HMA( Hot Mix Asphalt) fatigue damage. This ratio can be represented as: 207
a b a aDE
DE
RDEC
DE b a
(3) 208 where 209 aRDEC
the average ratio of dissipated work change at load cycle a, comparing to next cycle b; 210a, b = load cycle a and b. the typical cycle count between cycle a and b is 100, i.e., b-a=100; 211
a b
DE , DE =
the dissipated work produced in load cycle a and b respectively; 212Similar as Eq. (3), Eq. (4) can be expressed by using the ratio of dissipated work change 213 (RDWC) as: 214
a b a aDW
DW
RDWC
DW b a
(4) 215 where 216 aRDWC
the average ratio of dissipated work change at load cycle a, comparing to next cycle b; 217a, b = load cycle a and b. the typical cycle count between cycle a and b is 100, i.e., b-a=100; 218
a b
DW , DW =
the dissipated work produced in load cycle a and b respectively; 219RDEC or RDWC eliminates the energy that is dissipate in other forms without producing 220
damage. This provides a true indication of the damage being done to the mixture from one cycle to 221
another by comparing the previous cycle’s energy level and determining how much of it caused 222
damage. 223
As introduced by Ghuzlan (2001) and Carpenter et al. (2003, 2006), the damage curve 224
represented by RDEC vs. loading cycles shows three stages: a rapid decrease, followed by a plateau 225
stage (stage II) for the majority of the fatigue cycles. The plateau stage (stage II), an indication of a 226
period where there is a relatively constant percentage of input energy being turned into damage, will 227
extend throughout the main service life until a dramatic increase in RDEC, which gives a sign of true 228
fatigue failure (stage III). The schematic chart is given in Figure 5. 229
230
FIGURE 5. Typical RDEC plot with three behaviour zones (Carpenter et al. 2003) 231
EXPERIMENTAL RESULTS 232
The first group of tests was conducted at three load levels (150N, 250N, 350N) at 100C. The results 233
are presented below for all products (A1, A2, C1 and C2) and all interfaces (SM1, GM1, GM2, PM2). 234
Figure 6 shows the ratio of dissipated work change (RDWC) curves for three different 235
membrane products at three different load levels (150N, 250N and 350N) at 100C. Figure 7 shows the 236
dissipated work and the membrane debonding length for three different membrane products versus the 237
number of load cycles. The comparisons have been conducted for all products at the steel/membrane 238
(SM1) interface. 239
240
241
FIGURE 6. Ratio of dissipated work change of steel/membrane interface 1 (SM1) 242
In Figure 6 it can be observed that, after the initial loading period, the plateau stage was 243
reached. At the SM1 interface, the plateau values of product C1 were always lower than the other two 244
products indicating that less input energy turned into SM1 interface damage. Particularly, at 150N and 245
250N load levels, almost no energy was turned into fatigue damage for product C1. 246
247
FIGURE 7. Dissipated work and debonding length of steel/membrane interface 1 (SM1) 248
Product C2 was found to have better fatigue resistance than product A1 at 150N and 250N 249
load levels. However, at 350N load level product A1 was found to have lower RDWC values than 250
product C2. That can explain why at 350N load level, membrane C2 was debonded earlier than 251
product A1, see Figure 7. 252
253 254
255
FIGURE 8. Ratio of dissipated work change of Guss/membrane interface 1 (GM1) 256
In Figure 8, the results for product A1, C1 and C2 at GM1 interface are presented. Similarly 257
as in SM1 interface, product C1 was found to have the lower RDWC values followed by product C2 258
and A1 respectively. At 250N load level, the RDWC curves for product C2 and C1 overlapped 259
showing a similar response at GM1 interface. Product A1 could not be tested at 350N load level 260
because it was fully debonded during the first 100,000 cycles of the 250N load level. 261
262
263
FIGURE 9. Ratio of dissipated work change of Guss/membrane interface 2 (GM2) 264
At the GM2 interface, only products A2 and C2 were tested. These products were fully 266
debonded at 250N load level, hence only the RDWC curves of 150N and 250N are presented in 267
Figure 9. Although slightly lower RDWC values occur for product C2 at 150N load level, it can be 268
observed that both products demonstrated a similar response . At the 250N load level, RDWC values 269
are higher for product A2 than for product C2 proving that product C2 had better fatigue resistance at 270
GM2 interface than product A2. 271
272
273
FIGURE 10. Ratio of dissipated work change of Porous AC/membrane interface 2 (PM2) 274
At the PM2 interface, product C2 was found to have slightly lower values of RDWC at 150N 275
load level, Figure 10. However, a significant difference of RDWC between those two products can be 276
observed at 250N load level. It can also be seen that the RDWC curve of product A2 stopped 277
developing after 200,000 load cycles because it was fully debonded. 278
The results from the tests conducted at 300C are presented below. These tests were performed 279
at one load level, 100N for 864 000 cycles for all products and interfaces. Nevertheless, at some 280
interfaces the tests were completed before the maximum of 864000 load cycles due to large 281
deformation of the samples. 282
284
FIGURE 11. Dissipated work and debonding length at 100N and 30oC 285
Figure 11 shows the results for products A1, A2, C1 and C2 at the SM1, GM1, GM2 and 286
PM2 interfaces. For product A1 at the SM1 and the GM1 interfaces, the SM1 interface was found to 287
be the one with the lower values in terms of dissipated work producing a steady response throughout 288
the fatigue test. At the GM1 interface, the debonding process started since the beginning of the test 289
resulting to full debonding after the first 100,000 loading cycles while the calculated dissipated work 290
was also found to increase since the beginning of the test. 291
For product A2, the GM2 interface is better than PM2 in terms of dissipated work. High rate 292
of debonding was observed at PM2 interface at 100,000 cycles. 293
The results of product C1 are also presented in Figure 11. This product is used only as a 294
bottom membrane which means that is only applicable on steel plate and Guss asphalt. SM1 interface 295
was found to be the one with the lower values in terms of dissipated energy therefore producing a 296
steady response throughout the fatigue test. It can be noticed that the SM1 interface has zero 297
debonding. On the other hand, GM1 was found to have 45mm of debonding length at the end of the 298
test. 299
For product C2, it can observed that SM1 interface is the worst one in relation to the 300
membrane debonding speed. GM1 and GM2 interfaces demonstrate almost the same response. 301
302
303
FIGURE 12. Ratio of dissipated work change at 100N and 30oC 304
In Figure 12 the RDWC of three different products, tested at a temperature of 300 C and a load 305
level of 100N, is compared. For both SM1 and GM1 interfaces at 300 C, product C1 was found to 306
have lower RDWC values than product A1 and C2. At GM1 interface, products C2 and A1 were 307
found to have similar response with the RDWC curves overlapping. However, a significant difference 308
can be observed at the SM1 interface between products C2 and A1; higher RDWC values occurred for 309
product A1 at the SM1 interface. 310
At the GM2 interface, product A2 was found to have lower RDWC values than product C2 311
while at the PM2 interface, product C2 was found to have lower RDWC values. Finally, it can be seen 312
that product C2 at PM2 interface reached its plateau stage earlier than product A2. 313
CONCLUSIONS 314
Based on the presented results, the following conclusions can be made: 315
The MAT set up is capable of characterizing the fatigue response of the various membranes 316
bonded on the different substrates. The test results allow a better understanding of the 317
membrane performance on the bridge structure allowing thus optimization of maintenance 318
activities; 319
The fatigue response of a membrane product is influenced not only by the surrounding 320
substrate but also by the environmental temperature and loading level applied on the 321
membrane; 322
The concept of dissipated energy/work provides a fundamental and expedient means to 323
evaluate the fatigue life of membrane products on different substrates; 324
Product C1 performs quite well as the bottom membrane, both at 100C and 300C, in term of 325
values of dissipated work and debonding length. 326
Product C2 and A2 are considered as the best choices for the top membranes. 327
The observations from the MAT cyclic loading tests are coincident to the observations from 328
MAT monotonic static loading tests in the previous paper. It means that the findings of this 329
study is a further proof that the methodology utilized in this research project is adequate for 330
ranking the bonding characteristics of various membrane products on different substrates for 331 OSDB construction. 332 ACKNOWLEDGMENTS 333
This research project is part of the research program of InfraQuest which is a collaboration 334
between TNO, TU Delft and the Dutch road authorities. The research was partially funded by the Dutch 335
Transport Research Centre (DVS) of the Ministry of Transport, Public Works and Water Management 336
(RWS). Their financial support is highly appreciated. 337
REFERENCES 338
1. Gurney, T. Fatigue of steel bridge decks. HMSO Publication Centre: London. 1992, pp. 165. 339
2. Mangus, A.R. and S. Sun. Orthotropic Bridge Decks. Bridge Engineering Handbook, ed. W. 340
Chen and L. Duan 1999, Boca Raton: C.R.C. Press. 341
3. Medani, T.O. Design principles of surfacings on orthotropic steel bridge decks. Delft 342
University of Technology, Delft, 2006. 343
4. Liu, X., Scarpas, A., Li, J., Tzimiris, G., Hofman, R. and Voskuilen, J. Development of Test 344
Method for Assessing the Bonding Characteristics of Membrane Layers in Wearing Course 345
Laid on Orthotropic Steel Bridge Decks. In proceeding of Transportation Research Record, 346
Washington D.C., 2013. 347
5. Liu, X., and A. Scarpas. Experimental and Numerical Characterization of Membrane 348
Adhesive Bonding Strength on Orthotropic Steel Deck Bridges, Part 1. Project report 349
CITG2012-1, Delft University of Technology, The Netherlands, 2012. 350
6. Van Dijk, W. Practical Fatigue Characterization of Bituminous Mixes. Proceedings of the 351
Association of Asphalt Paving Technologists, vol. 44, pp. 38-74. 1975. 352
7. Van Dijk, W. and W. Visser. The Energy Approach to Fatigue for Pavement Design. 353
Proceedings of the Association of Asphalt Paving Technologists, vol. 46, pp. 1-40. 1977. 354
8. Carpenter, S. H., and M. Jansen.. Fatigue Behavior Under New Aircraft Loading Conditions.. 355
In Aircraft/Pavement Technology: In the Midst of Change, seattle, Washington, 17-21 August 356
1997. Edited by F.V. Hermann. American Socity of Civil Engineers, New York. pp. 259-271. 357
1997. 358
9. Shen, S., Carpenter, S.: Dissipate Energy concept for HMA Performance: Fatigue and Healing, 359
Department of Civil and Environmental Engineering. University of Illinois at Urbana-360
Champaign: Urbana, Illinois. 2007. 361
10. Ghuzlan, K, and S. H. Carpenter. Energy-Derived/Damage-Based Failure Criteria for Fatigue 362
Testing, In Transportation Research Record: Journal of the Transportation Research Board, 363
No.1723, TRB, National Research Council, Washington D.C. pp. 141-149. 2000. 364
11. Carpenter, S. H., Ghuzlan, K, and Shen, S. Fatigue Endurance Limit for Highway and Airport 365
Pavements, In Transportation Research Record: Journal of the Transportation Research Board, 366
No. 1832, TRB, National Research Council, Washington D.C., 2003, pp. 131-138. 2003. 367
12. Ghuzlan, K. Fatigue Damage Analysis in Asphalt Concrete Mixtures Based upon Dissipated 368
Energy Concept. PHD thesis. University of Illinois at Urbana-Champaign, August, 2001. 369
13. Carpenter, S. H. and Shen, S. A Dissipated Energy Approach to Study Hot-Mix Asphalt 370
Healing in Fatigue. Transportation Research Record (TRR): Journal of the Transportation 371
Research Board, No. 1970, pp.178-185. 2006. 372
373 374 375 376