See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228408364

A Large-Scale Triaxial Apparatus for Prototype Railroad Ballast Testing

Article  in  Geotechnical Testing Journal · July 2009

DOI: 10.1520/GTJ102033

CITATIONS READS
29 548

3 authors, including:

Adam Sevi Louis Ge

Norwich University National Taiwan University
9 PUBLICATIONS   90 CITATIONS    93 PUBLICATIONS   1,127 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Low-carbon Biogeotechnologies and Sustainable Geomaterials View project

New Geotechnical Testing Technologies and Fundamentals of Soil Behaviors View project

All content following this page was uploaded by Adam Sevi on 02 April 2018.

The user has requested enhancement of the downloaded file.

 Geotechnical Testing Journal, Vol. 32, No. 4
Paper ID GTJ102033
Available online at: www.astm.org

1 Adam F. Sevi,1 Louis Ge,2 and W. Andy Take3

2 A Large-Scale Triaxial Apparatus for Prototype

3 Railroad Ballast Testing

5 ABSTRACT: Shear strength and compressibility of large grain-size materials are critical parameters for the geotechnical design of road bases,
6 rock fill embankments, and railroad sub-base. However, due to the correspondingly large scale of triaxial specimens necessary for rock-fill and
7 railroad ballast material testing, the numbers of facilities that are capable of testing these materials are few. In this paper, a cost effective design is
8 documented for a triaxial apparatus capable of testing prototype railroad ballast material containing particle sizes up to 63.5 mm (2.5 in.). Unique to
9 this testing apparatus is the use of vacuum as confinement to allow an unobstructed digital image measurement of specimen volume change during
10 testing. The specimen preparation methodology, manufacture of latex membrane, and instrumentation are also discussed. Finally, the results of a
11 cyclical triaxial test are presented to demonstrate the quality of the testing data from this triaxial apparatus.
12 KEYWORDS: triaxial testing, railroad ballast, PIV
large scale of triaxial specimens necessary for rock-fill and ballast 40
13 Introduction
material testing, the numbers of facilities that are capable of testing 41
14 Two of the most important characteristics in the design of roadbeds, these large grain-size materials are few (Marachi et al. 1969; Ray- 42
15 rock-fill embankments, and railroad structures are shear strength mond and Davis 1978; Indraratna et al. 1998; Varadarajan et al. 43
16 and compressibility of the granular material. Testing of these prop- 2003). Shear strength and compressibility of large grain-size mate- 44
17 erties is typically performed using the triaxial testing apparatus. rials are critical parameters regardless of grain size; therefore a cost 45
18 The triaxial test configuration allows a confining pressure to be ap- effective technique for estimating these quantities for large grain- 46
19 plied through a flexible membrane surrounding the sample during size materials is necessary. 47
20 axial loading. It has been demonstrated that the largest grain size
21 that can be accurately examined in the triaxial apparatus must be
22 one-sixth the diameter of the testing specimen. Additionally, a Triaxial Testing Facility 48
23 sample height to diameter ratio of two is necessary to alleviate end
24 platen confinement of the specimen during testing (ASTM D5311-
A prototype large grain-size triaxial testing apparatus has been de- 49
25 92).
veloped at the Missouri University of Science and Technology 50
26 It is not uncommon for railway ballast materials to contain indi- (Missouri S&T) which is capable of testing specimen diameters of 51
up to 419.1 mm 共16.5 in.兲. In a traditional triaxial test apparatus, 52
27 vidual particles of sizes up to 63.5 mm 共2.5 in.兲. One such railway
the confining pressure is provided by a fluid contained within the 53
28 ballast, a dark fine-grained igneous rock mined and cone crushed
triaxial cell. Due to the sample size requirements for railway bal- 54
29 by Iron Mountain Trap Rock (a subsidiary of Fred Webber Inc.,
last, the adoption of a similar approach would require a triaxial cell 55
30 MO), was chosen to evaluate the performance of the prototype
that would be sufficiently heavy to require mechanical assistance 56
31 large grain-size triaxial apparatus. This ballast material is consid-
32 ered a very strong railroad sub-base material with a Los Angeles
100
33 abrasion value of 15. The gradation of the ballast material contains
34 a maximum particle size of 63.5 mm 共2.5 in.兲 and is marketed as #3 90
35 modified railroad ballast (63.5 mm to 9.5 mm
80
36 共2.5 in. to 3 / 8 in.兲) as shown in Fig. 1.
37 If the 1 + 6 maximum particle size to sample diameter ratio is 70

38 met, the minimum diameter of a triaxial specimen of this material 60

39 should be approximately 381 mm 共15 in.兲. Therefore, due to the
% Passing

50

Manuscript received August 18, 2008; accepted for publication February 4, 40

2009; published online December XXXX.
1
Assistant Professor, Department of Civil Engineering, Norwich University, 30
158 Harmon Drive, Northfield, VT 05663, E-mail: asevi@norwich.edu. For-
20
merly, Ph.D. Student, Missouri University of Science and Technology, Rolla,
MO 65409. 10
2
Assistant Professor, Department of Civil, Architectural, and Environmental
Engineering, Missouri University of Science and Technology, 1401 North Pine 0
1 2
10 10
Street, Rolla, MO 65409, E-mail: geyun@mst.edu Grainsize (mm)
3
Assistant Professor, Department of Civil Engineering, Queen’s University,
Kingston, ON K7L 3N6, E-mail: andy.take@civil.queensu.ca FIG. 1—Gradation curve of #3 modified railroad ballast.

Copyright © 2009 by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959. 1
 2 GEOTECHNICAL TESTING JOURNAL

thick was bolted to the bottom of the bottom platen allowing the 99
sample to be lifted using a fork lift. The two ports drilled and tapped 100
into the bottom platen were not used in this testing program as they 101
tended to accumulate fine debris caused by specimen attrition. They 102
would be useful in the case that water was to be drained out of the 103
specimen if saturated surface dry conditions were of interest. How- 104
ever, for the current testing, these ports were plugged and not used. 105
Although vacuum confinement greatly simplifies the triaxial 106
testing system, it does lead to measurement difficulties; in particu- 107
lar, the measurement of sample volume change. In a traditional tri- 108
axial test apparatus, volume change is recorded by controlling or 109
measuring the volume of flow in or out of the triaxial sample. In the 110
case of vacuum confinement, no such measurement is possible. To 111
overcome this limitation, two methods of calculating volume 112
change have been adopted: on-sample local radial strain transduc- 113
ers and digital image analysis. 114

FIG. 2—Top and bottom platen details.

Building a 419 mm „16.5-in.… Diameter Specimen 115
57 for assembly. This fact, combined with the acceptability of the rela-
58 tively low confining stresses, in the range of 20– 80 kPa
Once the material was delivered to the Missouri S&T civil engi- 116
59 共2.9– 12 psi兲 for testing railway ballast, led to the adoption of neering laboratory the barrels were emptied into drying trays, oven 117
60 vacuum backpressure applied inside the specimen as an alternative dried, and then sieved. All testing gradations were mixed from this 118
61 means of apply confining pressure for the large grain-size triaxial sieved fresh material. A 419 mm 共16.5-in.兲 diameter and 864 mm 119
62 testing apparatus. This method of confinement is only sufficient for 共34-in.兲 tall triaxial specimen yields a volume of 0.12 cubic meter 120
63 materials relatively close to the surface, and therefore under rela- 共4.21 cubic ft兲. At a density of 1571 kg/ m3 共98 pcf兲 this is a speci- 121
64 tively low confinement. The maximum confinement available using men of almost 190 kg 共420 lb兲. The gradation of the specimens was 122
65 vacuum is 101 kPa 共1 atmosphere兲. important, therefore sieving and remixing of each sample before 123
66 Both the top and bottom caps, as shown in Fig. 2, were ma- and after triaxial testing was performed. Sieving was performed 124
67 chined from 1-in. thick aluminum plate. Each cap has two port using Gilson shakers with 609.6 mm by 762 mm (24 in. by 30 in.) 125
68 holes, drilled and tapped for standard pipe thread. These ports can screens. The sieve system used was capable of taking approxi- 126
69 be used for access to the inside of the specimen for regulating mately 23 kg 共50 lb兲 of material at a time. In general, the speci- 127
70 vacuum and vacuum measurement. Around the perimeter of both mens were broken into ten equal parts of approximately 18 kg 128
71 caps a 9.5 mm 共3 / 8-in.兲 groove was machined centered at 8.5 mm 共40 lb兲 each. In this fashion, two buckets comprised a fifth of the 129
72 共1 / 3 in.兲 from the exterior surface. This groove was placed to hold sample. This was useful as the sample was built using five equal 130
73 an o-ring in place around the entire diameter sealing the latex mem- lifts of 37 kg 共82.5 lb兲 each. 131
74 brane to the cap. While this groove was instrumental in keeping the After a specimen was triaxially tested, the sample was dumped 132
75 o-rings from rolling off the top and bottom of the caps, additional into a slurry tray. This steel tray was accompanied by a wooden 133
76 o-rings placed outside the groove helped maintain a tight seal be- pallet for ease of movement throughout the lab. The tray was raised 134
77 tween the latex and the cap circumference. to the bottom platen after triaxial testing using a fork lift. The 135
78 Figure 2(a) shows the design of the top platen, which was fitted vacuum in the sample was then allowed to dissipate and the sample 136
79 with six bolts in a circular pattern 317.5 mm 共12.5 in.兲 in diameter. was pushed into the tray. The tray was then placed on a stack of 137
80 These were placed to surround the load platen of the MTS load pallets and tilted to approximately 45 degrees. Five gallon buckets 138
81 frame, maintaining alignment of the load platen with the top cap were then placed under the low end of the tray and the material 139
82 during testing. Hollow plastic rods were placed over the bolts to shoveled/funneled into the buckets. While the buckets were not 140
83 protect the top cap. In addition, the MTS 880 load frame uses a weighed during this step, approximately 18 kg 共40 lb兲 of material 141
84 spherical seat at the top platen to allow free tilting, thereby avoiding was placed in each bucket. In this fashion ten buckets were used to 142
85 stress concentrations along the top cap. The two port holes of the contain the entire sample. This allowed a consistent number of 143
86 top platen were fitted with a 6.4-mm 共1 / 4-in.兲 and a 9.5-mm buckets to be used for transporting the sample throughout the 144
87 共3 / 8-in.兲 standard pipe thread tapped into the aluminum plate. At- sample preparation process. 145
88 tached to these ports are ball valves that allow vacuum and vacuum The buckets were then poured into the sieve machine with the 146
89 measurement at both ports. The two ports were useful in times proper sieve stack for the specific gradation. The material captured 147
90 when the membrane leaked. A large volume of vacuum could be on the different sieves was then placed in delineated buckets. After 148
91 applied to the specimen using the two ports, until the sample was sieving the entire sample, the material captured on the respective 149
92 stabilized through patching of the membrane. sieves was weighed for attrition measurements. To bring the grada- 150
93 The design of the bottom cap is given in Fig. 2(b). The bottom tion back to the specified testing gradation, material typically 151
94 cap was fitted with four steel angles bolted to the bottom and di- needed to be either added or removed from some or all of the sieves. 152
95 rected straight down from the bottom of the cap. On the vertical In the case of adding material to a sieve, fresh material was always 153
96 face of these angles a hole was centered at 63.5 mm 共2.5 in.兲 down added. In the case that more material was on a sieve than needed 154
97 from the bottom cap to allow the bottom cap to be bolted to the material was removed from the sample material and wasted. In this 155
98 MTS 880 bottom load platen. An aluminum table 13 mm 共 2 -in.兲
1
process some fresh material was added to the sample in each test. 156
 SEVI, GE, AND TAKE ON PROTOTYPE RAILROAD BALLAST TESTING 3

The hoop is then removed from the molten creating a cylinder of 195
latex. For thicker membranes a slower rate of extraction from the 196
pan of molten latex is used. There was no latex company willing to 197
custom build membranes 419 mm 共16.5 in.兲 in diameter and 198
889 mm 共35 in.兲 in length. Therefore, latex membranes for this 199
testing were constructed in house. Rolls of latex of dimensions 200
0.635 mm by 1067 mm by 21 m (0.025 in. by 42 in. by 69 ft) were 201
purchased. These rolls were then cut into 1358.9 mm 共53.3 in.兲 sec- 202
tions (pi by 16.5 in.+ 1.5 in.). The extra 38.1 mm 共1.5 in.兲 was used 203
to form an overlapping glue seam. The 38.1 mm 共1.5-in.兲 wide 204
seam was found to adequately seal the sample at the low confine- 205
ments used in this testing. Sandpaper was used to prepare the two 206
38.1 mm 共1.5-in.兲 overlapping seam portions of the membrane be- 207
fore gluing the seam together. 208
In order to glue the seam together, the latex was thoroughly 209
cleaned of all powders that are used during shipping. These pow- 210
ders keep the latex from adhering to itself while in a roll. Cleaning 211
FIG. 3—Specimen mold including PVC rigid mold outside, rubber liner inside,
and latex membrane laying at bottom of sample mold.
of this dust was performed using a clean wet rag. It was found that 212
latex tended to curl once glue was applied and started to cure. In 213
157 Typically, no more than 1 % of the original materiel was added or
order to allow the glue to be applied before the latex curled onto 214
158 removed from any sieve during this process. itself a method of temporarily adhering the latex to metal channels 215
159 With the proper weights of material at the different sieve sizes was developed. The channels served to both hold the membrane flat 216
160 obtained, the material was then thoroughly mixed. Mixing was per-
during glue application and serve as flat surfaces to sandwich the 217
161 formed by pouring the different sizes of material into a concrete
membrane seam between while the glue was allowed to cure. The 218
162 drum mixer. Mixing was performed at 12 r / min while adjusting the
channels were first cleaned of all debris including any fine powder 219
163 tilt of the mixing bucket several times to ensure proper mixing.
or leftover glue from previous seam building. Both the latex and 220
164 Leaving the tilt at one setting did not mix the material. The mixed
165 gradation was again poured into the steel tray and again tilted on
metal were then wetted and the latex was laid flat on the channel. 221
166 top of several pallets to be poured into buckets. The ten buckets All air bubbles were squeezed out from between the two surfaces. 222
167 were filled to exactly 18.7 kg 共41.3 lb兲 each. In this manner two The system is then allowed to dry overnight allowing the latex to 223
168 buckets comprised one 37.4 kg 共82.5 lb兲 sample lift, once com- lightly adhere to the metal channel. This additional step was found 224
169 pacted to 1570 kg/ m3 共98 pcf兲. sufficient to properly apply the glue and join the two edges of the 225
membrane forming an air tight seam. 226
At this point Scotch-Grip contact adhesive 1357 from 3M was 227
170 Specimen Mold applied to both 38.1-mm 共1.5-in.兲 wide scratched latex surfaces. 228
171 Due to concern of puncturing the latex membrane surrounding the The best method of applying the glue was found to be placing a 229
172 ballast sample during sample building and densification a unique heavy bead on one surface and brushing this bead uniformly across 230
173 sample mold was designed. The mold consisted of a PVC tube the 38.1 mm 共1.5-in.兲 seam. Two light beads were run the length of 231
174 457 mm 共18 in.兲 outside diameter and cut to 889 mm 共35 in.兲 tall the adjoining seam. Several seconds were allowed to let the glue 232
175 with a sheet of rubber placed inside. The latex membrane was then cure initially before the seams are joined by placing one channel on 233
176 placed inside the rubber sheet and ballast placed inside of this top of the other. Several weights were placed on the top channel at 234
177 membrane. The PVC tube was cut in half lengthwise creating two this point to maintain contact. If the glue was not allowed to dry 235
178 halves each standing 889 mm 共35 in.兲 tall and semicircular. Inside before contact or too much weight was applied to hold the seam 236
179 the rigid PVC tube a sheet of 9.5 mm 共3 / 8-in.兲 rubber BUNNEAU together the result was that the glue was pushed out of the seam 237
180 60A was cut to fully cover the inside walls of the rigid PVC mold as
resulting in poor seam bonding. The glue was then allowed to cure 238
181 shown in Fig. 3. Both the PVC half tubes and the rubber lining
for at least an hour. At this point the latex was peeled off the chan- 239
182 sheet were fitted with matching columns of port holes drilled into
nels and the seam inspected. If there were areas that did not adhere 240
183 each component. These holes were used as ports for applying
for the full width of 38.1 mm 共1.5 in.兲, more glue was applied to the 241
184 vacuum to pull the membrane tight against the mold during sample
185 building. The lining sheet of rubber was intended to avoid pinching/
voids and the latex seam was again pressed together between the 242
186 punching of the latex membrane between particle corners and the flat channel surfaces. At this point the membrane was structurally 243
187 rigid PVC mold during sample construction. complete. The assembling of the seam of the latex membrane is 244
188 The mold was designed to be one inch taller than the sample. displayed Fig. 4. 245
189 One inch was included at the base of the mold allowing it to be Visual texturing was applied in order to provide texture for the 246
190 hose-clamped around the 25.4 mm 共1 in.兲 tall bottom cap. In this digital image PIV strain analysis. A flat black oil-based paint was 247
191 fashion a sample 864 mm 共34 in.兲 tall was constructed. used for applying texture to the latex. This paint was flicked onto 248
the membrane using a paint brush. The best method for applying 249
many small nonuniform dots of paint was running a finger through 250
192 Constructing Latex Membrane the bristles and allowing the bristles to then flick the paint onto the 251
193 Typical latex membranes used in triaxial testing are made by dip- membrane. The paint was then allowed to dry and the other side of 252
194 ping a wire hoop oriented horizontally into a pan of molten latex. the membrane was painted after being allowed to dry overnight. 253
 4 GEOTECHNICAL TESTING JOURNAL

tested in preparation for the triaxial testing program. The most ef- 265
fective method found will be described here. A uniform initial test- 266
ing density of 1571 kg/ m3 共98 pcf兲 was established as the target 267
density based on initial densities of materials placed in the field. 268
Relative density is not typically determined for materials con- 269
taining particles of the size included in railroad ballast. This is pri- 270
marily due to large particles edge effects on void ratio along the 271
walls of a containing mold. All samples for triaxial testing were 272
prepared to an initial density of 1571 kg/ m3 共98 pcf兲. This density 273
was achievable for all the gradations and considered stable after the 274
sample had been constructed. To assess the density of the sample 275
during construction, magnets were placed in the sample at the in- 276
terface of respective lifts for monitoring. These magnets were ring- 277
shaped lightweight magnets (bonded neodymium-iron-boron). By 278
covering the hole of the ring, using thin sheet metal, the unit weight 279
of the magnet was made close to the unit weight of an intact ballast 280
particle. These magnets could then be monitored during densifica- 281
tion of subsequent lifts. This monitoring was achieved by using a 282
high power magnet outside the sample mold. By running the high 283
power magnet along the mold the location of the magnets at the 284
respective lifts could be monitored. It was found that five lifts were 285
effective for controlling sample density throughout. Magnets were 286
FIG. 4—Assembling the seam of the latex membrane.
not used during the building of samples that were used for triaxial 287
testing; however, painted ballast particles were used to monitor lift 288
254 Obtaining Uniform Initial Density interfaces. 289
During preliminary sample compaction it was found that once a 290
255 Compaction of granular material to a uniform initial density has lift was densified to 1522 to 1602 kg/ m3 共95 to 100 pcf兲 the lift 291
256 traditionally been performed using some form of vibratory ham-
would not appreciably further compact during the placement and 292
257 mer. Marachi et al. (1969) used black washers in the sample with
densification of subsequent lifts. All densification was performed 293
258 clear latex to investigate density during specimen building. Vibra-
by attaching the sample onto a horizontal shake table. The shake 294
259 tory compaction methods typically create considerable membrane
260 rupture as well as concerns of uniform density (Knutson 1976; table, as shown in Fig. 5, used was a Kimball K-3396 table assem- 295
261 Frenkel 2000). bly with an MTS 204 actuator and operated by a MTS 407 control- 296
262 In order to construct a quality sample for triaxial testing, the ler. This piece of equipment is more customarily used for earth- 297
263 density must be controlled throughout the sample. Several strate- quake loading simulation; however, it was found to be quite 298
264 gies for obtaining uniform density throughout the sample were satisfactory for the densification of railroad ballast. 299

FIG. 5—Sample mold on the shake table.

 SEVI, GE, AND TAKE ON PROTOTYPE RAILROAD BALLAST TESTING 5

Monotonic loading was performed on a MTS 880 load frame. 309

The load frame is controlled on an analogue MTS 448.85 Test Con- 310
troller and the load rate is dialed into a MTS 410.80 Function Gen- 311
erator. After the sample was placed on the MTS 880 loading platen, 312
a confining vacuum was maintained at all times to prevent sample 313
collapse. Vacuum was maintained during sample preparation using 314
the laboratory vacuum system. If there was a membrane leak sus- 315
pected before removing the sample mold, a second vacuum system 316
was connected to the sample. This second portable vacuum pump 317
had a large volume capacity. This large volume capacity allowed 318
the patching of the membrane in all but one occasion. Once a stable 319
vacuum was attained, the sample mold was removed and wire ex- 320
tensionometers placed. As shown in Fig. 6, wire extensionometers 321
were placed surrounding the samples circumference at 1/4H, 1/2H, 322
and 3/4H. This positioning was used to allow calibration between 323
the wire extensionometers and digital image radial strain measure- 324
ments throughout the sample. The wire extensionometers used in 325
this study were loaned by Raymond Frenkel of the University of 326
Massachusetts-Amherst and are the same extensionometers used in 327
the Suiker et al. (2005) study. Two LVDTs were placed at opposite 328
sides of the specimen as a redundant measurement of total vertical 329
deflection. 330
For photographic strain analysis three digital cameras were set 331
FIG. 6—Triaxial specimen fitted for testing. up surrounding the sample 120 degrees from each other. The cam- 332
era shutter was controlled by sending a voltage to the remote shut- 333
300 Instrumentation
ter control on the individual camera. All three cameras were oper- 334
ated from the same voltage output allowing all three cameras to 335
shoot at the exact same time. Each camera stored images from each 336
301 Monotonic Testing
data gathering event on local memory. These memory cards were 337
302 Monotonic strain controlled triaxial compression testing was per- then downloaded to an independent computer after testing was 338
303 formed on the prototype railroad ballast. Axial load, axial strain, complete. The elimination of the confining cell of the prototype 339
304 volumetric strain, and confining pressure were continually moni- large grain-size triaxial testing apparatus, allowed unobstructed 340
305 tored during testing. The monotonic testing was carried out at a and direct visual (and photographic) viewing of the specimen dur- 341
306 displacement rate of 0.17 mm/ s (1 in. per 150 s) or ing testing. In the case of digital image strain analysis this simpli- 342
307 10.2 mm per minute 共0.4 in. per min兲. Confining pressure was fication avoids geometric corrections necessary for light traveling 343
308 held at 20.7 kPa 共3 psi兲 throughout the monotonic testing. through a cylindrical chamber of fluid or air and the glass of the 344

FIG. 7—Prototype large grain-size triaxial testing apparatus.

 6 GEOTECHNICAL TESTING JOURNAL

TABLE 1—Initial unit weights of prepared prototype ballast specimens.

Unit Weight (pcf)

Lift Test #5 Test #7 Test #12 Test #16 Test #19 Test #21 Test #24
1 97.0 100.7 105.2 97.6 97.0 99.2 102.1
2 102.6 95.7 93.9 101.1 97.9 96.1 100.9
3 97.3 100.6 103.7 97.2 97.6 100.4 92.6
4 95.7 98.8 90.6 95.9 99.7 96.6 97.2
5 97.9 94.8 98.5 98.5 96.3 98.1 98.1
Average 98.1 98.1 98.4 98.1 98.1 98.1 98.2

345 chamber itself (Macari et al. 1997; Gachet et al. 2006). A sample during the cyclical load testing included axial load, axial deforma- 393
346 being tested in a light controlled environment can be seen in Fig. 7. tion, circumferential deformation, and confining pressure. An addi- 394
tional data channel was dedicated to attach a time stamp to digital 395

347 Cyclic Testing

photographs captured during testing. All of these parameters were 396
monitored at a rate of 100 data points per second. The slower load- 397
348 Stress ratios that the material was cyclically loaded to were calcu- ing during data readings allowed digital cameras placed at three 398
349 lated based on the monotonic testing previously discussed. Once different angles around the sample to operate taking a picture every 399
350 the sample was prepared, cyclical loading was performed using the
two seconds. This allowed ten pictures per cycle. 400
351 MTS function generator in conjunction with the MTS 880 loading
The ballast material tended to stiffen and plastic deformation 401
352 system. The loading system was again run at the 10 % capacity
353 setting allowing 49900 kg 共11 000 lb兲 of capacity from a system
continued to accumulate during testing. Due to these sample char- 402
354 with a 49 900 kg 共110 000 lb兲 capacity. This allowed more accurate acteristics monitoring of the median load and span was required as 403
355 control of the system for the loads required for this testing. Loads load applications continued. Data readings were taken while load- 404
356 required for this testing ranged from 113 kg 共250 lb兲 to 1815 kg ings continued at 1 Hz and the midpoint and span adjusted accord- 405
357 共4000 lb兲. ingly. A separate span setting was needed to achieve proper stresses 406
358 The following procedure was used to perform the cyclical load- at a loading rate of 0.05 Hz. Once this slower loading span was set 407
359 ing of railroad ballast samples using the analogue system available. it was fine tuned at the beginning of each data recording. In this 408
360 At the beginning of the test the control of the load was performed fashion the fifth loading of the ten loadings recorded would be clos- 409
361 by manually dialing the load using the MTS test controller in load est to a perfect loading. Therefore, the fifth loading provided the 410
362 control. In this manner the seating load of 113 kg 共250 lb兲 for all
data assigned to the specific load cycle. 411
363 cyclical tests was dialed in. At this point the data acquisition was
364 started. The load was then increased from the seating load to the
365 median load. The median load is the load that defines the midpoint
366 for the cyclical loading using this MTS system. The first cycle was
367 continued manually up to the maximum load and then decreased
368 back to the seating load. During this initial load, the load span was
369 set for subsequent cycles. In some cases it was necessary to manu-
370 ally control another cycle in this fashion in order to get the proper
371 midpoint and span controls correct. The load was then returned to
372 the median load in preparation for the function generator controlled
373 cycles. All cycles controlled by the function generator were sinu-
374 soidal load controlled by function midpoint and span. Both of these
375 functions were controlled using analogue dials adjusting the load
376 exerted at the platen of the load frame.
377 Preliminary testing concluded that there was no difference in the
378 ballast reaction to loadings performed between 0.05 to 3 Hz.
379 Therefore, a testing program was established to both allow rela-
380 tively slow loadings while data were being gathered, and acceler-
381 ated between readings to expedite the testing. Therefore, all mea-
382 surement readings were taken at 0.05 Hz 共20 s per cycle兲, while
383 loadings between readings were performed at 1 Hz for expedience.
384 Data were gathered during the initial twelve cycles of a test and
385 then for ten cycles at 100, 200, 500 cycles and then every
386 1000 cycles thereafter. The initial two cycles were recorded during
387 the setting of the load span and midpoint load 共1 to 2 cycles兲. Then
388 a ten cycle data stream was recorded. All further data points con-
389 sisted of ten cycle data streams. This ten cycle recording allowed
390 fine tuning of the loading to be performed as these cycles contin-
391 ued. The fifth loading was then taken as the data readings for the
392 specific loading number (100, 200, 500, 1000…). Data recorded FIG. 8—PIV patches on triaxial specimen and corresponding strains.
 SEVI, GE, AND TAKE ON PROTOTYPE RAILROAD BALLAST TESTING 7

412 Test Results

413 Initial unit weight measurements of each lift of prepared prototype
414 ballast specimens is displayed in Table 1. Lift 1 represents the very
415 bottom layer of compacted ballast after horizontal shaking. It
416 shows that unit weights of specimens for Test #5, 7, 16, 19, and 21
417 were controlled within 5 % variation of average initial unit weight.
418 Among them, unit weights of specimens for Test #19 and 21 were
419 even down to 2.5 % variation. On the other hand, the specimen for
420 Test #12 yields the most scattered unit weight distribution of all
421 specimens prepared. It was up to 8 % variation of the average value.
422 Based on the data displayed in Table 1, horizontal shaking is found
423 to be an effective method to achieve the uniform distribution of
424 initial unit weight for railroad ballast samples.
425 As indicated earlier, digital image technique was used to capture
426 volumetric strain behavior during triaxial testing. Figure 8 shows a
427 set of particle image velocity (PIV) patches on a triaxial specimen
428 and corresponding strain measurements at different elevations on
429 the sample as found using this technique. The PIV method uses the
430 movement of “PIV patches” to calculate local strains at the surface
431 of the sample. These strains are calculated by tracking these PIV
432 patches from one image to the next, as testing continues. The pixel FIG. 9—PIV (line) and wire extensionometers (dots) measurements.
433 pattern within a PIV patch is established on the first image in time.
434 As testing continues a second image is taken. The PIV algorithm
435 searches the second image for a pattern similar to a patch in the sample. The measurements from these two methods can be seen to 449
436 first. Once this patch match is made between the two images, the be well matched. 450
437 distance this patch has moved in the image can be calculated. It is Figure 10 presents the cyclical stress strain results for a proto- 451
438 imperative that each patch have a unique pixel pattern and consis- type ballast gradation sample tested using a stress ratio 0.94. The 452
439 tent lighting. A random pattern (in this case splatter paint) is needed stress ratio is defined as the ratio of the applied cyclical deviatoric 453
440 on the sample membrane in order for the PIV algorithm to function. load to the peak deviatoric load from the monotonic tests. The cy- 454
441 A regular pattern, or no pattern at all, on the sample membrane clical test was performed for 10 000 cycles under a constant con- 455
442 would create many patches that were similar. This would increase fining pressure 20.7 kPa 共3 psi兲. Data were recorded periodically 456
443 the likelihood of one patch being mistaken as another on a different throughout the 10 000 load cycles. Data are presented in Fig. 10 457
444 region of the sample. This mismatching would lead to grossly erro- corresponding to the 100, 500, and 1000th load cycle, and then 458
445 neous strain calculations. A more technical and thorough explana- every 1000 load cycles until completion of the test. The final data 459
446 tion of the PIV analysis method can be found in White et al. (2003). stream corresponds to the 10 000th load cycle. Figure 11 shows the 460
447 Figure 9 displays the radial strain measurements by both wire ex- permanent or accumulated axial and volumetric strains at the se- 461
448 tensionometers and the PIV technique at different elevations on the lected number of cycles (100, 500, 1000, …, 10 000). 462

150
Deviatoric Stress (kPa)

2000th − 10000th
100th 200th 500th 1000th
100

50

0
−0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Axial Strain (%)

0.6
Volumetric Strain (%)

0.4

0.2

−0.2
−0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Axial Strain (%)

FIG. 10—Cyclic stress-strain curves at various number of cycles.

 8 GEOTECHNICAL TESTING JOURNAL

0.8 University of Science and Technology (formerly University of 498

Missouri-Rolla) for their help with testing and sample preparation. 499
Permanent εa (%)

0.6
Additionally, Raymond Frenkel, of the University of 500
0.4 Massachusetts-Amherst, is recognized for developing and donating 501
0.2
the wire extensionometers used in this study. All ballast material 502
used in testing was donated by Gary Nickelson of Fred Webber Inc. 503
0
2 3 4
10 10 10

0.4 References 504

Permanent εv (%)

0.3
[1] ASTM Standard D5311–92, 1992, “Standard Test Method for 505
0.2 Load Controlled Cyclic Triaxial Strength of Soil,” Annual 506
0.1
Book of ASTM Standards, ASTM International, West Consho- 507
hocken, PA. 508
0
10
2
10
3
10
4 [2] Frenkel, R., “Radial Strain Measurement Device for Triaxial 509
Number of Cycles Testing,” Masters of Science Degree Project Report, Geotech- 510
nical Report No. TUD00–464P, University of 511
FIG. 11—Permanent axial and volumetric strains versus number of cycles. Massachusetts—Amherst, 2000. 512
[3] Gachet, P., Geiser, F., Laloui, L., and Vulliet, L., 2006, “Auto- 513
463 The following observations can be made regarding the proto- mated Digital Image Processing for Volume Change Measure- 514
464 type cyclical test results. The specimen exhibited contraction upon ment in Triaxial Cells,” Geotech. Test. J., Vol. 30, No. 2, pp. 515
465 axial cyclical loading. This indicates the material is below the criti- 1–6. 516
466 cal void envelope at a confinement of 20.7 kPa 共3 psi兲. All devia- [4] Indraratna, B., Ionescu, D., and Christie, H. D., 1998, “Shear 517
467 toric stress strain curves exhibit a hysteretic loop for each loading Behavior of Railway Ballast Based on Large-Scale Triaxial 518
468 and unloading cycle. This loop indicates work adsorbed by the bal- Tests,” J. Geotech. Geoenviron. Eng., Vol. 124, No. 5, pp. 519
469 last. Work generally decreased with further loading. Additionally, 439–449. 520
470 these stress strain loops became closer together with further load- [5] Knutson, R. M., 1976, “Factors Influencing the Repeated 521
471 ing but did not reach an overlapping situation. In this fashion Load Behavior of Railway Ballast,” Ph.D. Thesis, University 522
472 “shake down” was not achieved, where the material ceases to accu- of Illinois at Urbana-Champaign. 523
473 mulate plastic strain. [6] Macari, E. J., Parker, J. K., and Costes, N. C., 1997, “Measure- 524
ment of Volume Changes in Triaxial Tests Using Digital 525
Image Techniques,” Geotech. Test. J., Vol. 20, No. 1, pp. 103– 526
474 Summary and Conclusions 109. 527
[7] Marachi, N. D., Chan, C. K., Seed, H. B., and Duncan, J. M., 528
475 The development of a cost effective large-scale triaxial testing ap- 1969, “Strength and Deformation Characteristics of Rockfill 529
476 paratus is presented. The apparatus can test prototype railroad bal- Materials,” Report No. TE-69–5, State of California Depart- 530
477 last material with grain size up to 63.5 mm 共2.5 in.兲. This configu- ment of Water Resources. 531
478 ration uses a vacuum routed inside the sample to provide [8] Raymond, G. P. and Davies, J. R., 1978, “Triaxial Testing on 532
479 confinement. This method of providing confinement allows direct Dolomite Railroad Ballast,” J. Geotech. Engrg. Div., Vol. 104, 533
480 visual (and photographic) contact with the triaxial sample, simpli- No. GT6, pp. 737–751. 534
481 fying PIV strain assessment. This method of providing confinement [9] Suiker, A. S. J., Ernest, T. S., and Frenkel, R., 2005, “Static 535
482 is useful for testing of materials that are to be tested at relatively and Cyclic Triaxial Testing of Ballast and Subballast,” J. Geo- 536
483 low confinements, below 101 kPa 共1 atmosphere兲. The specimen tech. Geoenviron. Eng., Vol. 131, No. 6, pp. 771–782. 537
484 preparation, instrumentation, and testing procedure are also dis- [10] Take, W. A. and Kemp, J., 2006, “Application of Digital 538
485 cussed. The cyclical test data shown here demonstrate the quality Image-Based Deformations Measurements in Geotechnical 539
486 results that are possible when using this apparatus. An attempt of Testing,” 57th Canadian Geotechnical Conference, Vancou- 540
487 incorporating a digital image technique to capture the volumetric ver, BC. 541
488 strain during testing is also discussed. This possibly represents a [11] Varadarajan, A., Sharma, K. G., Venkatachalam, K., and 542 AQ:
489 great advantage of this triaxial configuration by removing the sur- Gupta, A. K., 2003, “Testing and Modeling of Two Rockfill 543 #1
490 rounding chamber and fluid, allowing unobstructed imaging during Materials,” J. Geotech. Geoenviron. Eng., pp. 206–218. 544
491 testing. [12] White, D. J., Take, W. A., and Bolton, M. D., 2003, “Soil De- 545
formation Measurement Using Particle Image Velocimetry 546
(PIV) and Photogrammetry,” Geotechnique Vol. 53, No. 7, 547
492 Acknowledgments pp. 619–631. 548
[13] White, D. J., Take, W. A., Bolton, M. D., and Manachen, S. E., 549
493 This work could not be accomplished without the assistance of the 2001, “A Deformation Measurement System for Geotechni- 550
494 following people. Gary Abbott, Jeff Bradshaw, and Steve Gabel are cal Testing Based on Digital Imaging, Close-Range Photogra- 551
495 recognized for their extensive input during both design and con- phy, and PIV Image Analysis,” 15th International Conference 552
496 struction of this testing equipment. Lucas Carr, Chris Cook, and of Soil Mechanics and Geotechnical Engineering, Istanbul, 553
497 David Weidinger, former undergraduate students at the Missouri Turkey, pp. 539–542. 554
NOT FOR PRINT! FOR REVIEW BY AUTHOR NOT FOR PRINT!

AUTHOR QUERIES — 004904GTJ

#1 au: Please supply volume number

View publication stats