Vol.4, No.

1 EARTHQUAKE ENGINEERING AND ENGINEERING VIBRATION June, 2005

Article ID: 1671-3664(2005)01-0069-05

Centrifuge modeling of PGD response of buried pipe

Michael O'Rourke 1., Vikram Gadicherla 2. and Tarek Abdoun l~

1. Department of Civil and Environmental Engineering, Rensselaer Polytechnic Institute, Troy, New York, USA
2. Hance Engineering Solutions, Hyderabad India

Abstract: A new centrifuge based method for determining the response of continuous buried pipe to PGD is presented.
The physical characteristics of the RPI's 100 g-ton geotechnical centrifuge and the current lifeline experiment split-box are
described: The split-box contains the model pipeline and surrounding soil and is manufactured such that half can be offset, in
flight, simulating PGD. In addition, governing similitude relations which allow one to determine the physical characteristics,
(diameter, wall thickness and material modulus of elasticity) of the model pipeline are presented. Finally, recorded strains
induced in two buried pipes with prototype diameters of 0.63 m and 0.95 m (24 and 36 inch) subject to 0.6 and 2.0 meters (2
and 6 feet) of full scale fault offsets and presented and compared to corresponding FE results.

Keywords: earthquakes; buried pipe; permanent ground deformation; centrifuge models; fault crossings; lifeline
earthquake engineering

1 Introduction pursued, specifically centrifuge modeling:

In the recent past, Bransby et al. (2002a, 2002b),
Buried pipelines are commonly used to transport have used centrifuge models to investigate the upheaval
oil, water, sewage and natural gas. These pipelines are capacity and resistance o f buried offshore pipe. The
sometimes referred to as lifelines as they are essential work reported herein, is to the author's knowledge,
for the support o f life and maintenance o f property. the first reported application o f centrifuge modeling
Pipelines in seismic zones are prone to permanent procedures to the problem of PGD effects on buried
ground deformation (PGD) and wave propagation pipelines.
hazards. Through the PGD hazards are limited to small
areas within the pipeline network; the potential for 2 Centrifuge m o d e l i n g
damage is high since PGD imposes large deformation
on the pipelines. For problems in which soil-structure interaction
As such, fault crossing, lateral spreads and other forces are dominate, small-scale models by themselves
types o f PGD are arguably the most severe seismic cannot replicate similar stresses at comparable points on
hazards for continuous buried pipelines. Current the model and the prototype. This is because soil loading
analysis and design procedures, to a great extent, are and stiffness at a particular depth are related to the self-
based upon Finite Element (FE) modeling. There are, weight o f the soil above. For example, the vertical
unfortunately, relatively few full-scale case histories stress at a depth o f 10 m for a soil having a density of
which could be used to benchmark or validate the 10 kN/m 3' would be 100 kPa. However, in a tenth scale
applicability o f FE assumptions. That is, full-scale model, the stress at the corresponding location (i.e. 1 m
verification o f the predicted behavior o f buried pipeline depth in the model) in the same soil is only 10 kPa. As a
subject to PGD is, at best, sparse. result, the similitude o f soil-structure interaction forces
Herein, an alternate approach to validation is is not maintained. This difficultly can be overcome by
effectively increasing the weight o f soil by a factor o f
Correspondence to: Michael O'Rourke, Rensselaer Polytechnic 10. In a centrifuge, this is accomplished by spinning at
Institute, Department of Civil and Environmental Engineering, a speed corresponding to a centrifugal acceleration o f 10
Jonsson Engineering Center, Room 4046, 110 8th Street, Troy, times the earth's gravity (10 g). This is the underlying
New York, 12180, USA. concept behind centrifuge modeling o f soil and soil-
Tel: (518) 276-6933; Fax: (5[8) 276-4833; structure systems. A partial list of scaling law in terms
E-mail: [email protected] o f a centrifugal acceleration Ng is presented in Table 1.
*Professor; ~Engineer; w Professor When modeling the response o f a buried pipe to
Supported by: National Science Foundation Under Grant No. PGD, care must be taken to maintain similitude. In
CMS-0085256 terms o f soil, similitude is maintained by using the same
Received 2004-11-17; Accepted 2005-05-13
70 EARTHQUAKE ENGINEERING AND ENGINEERING VIBRATION Vol.4

Table 1 Selected similitude relations for centrifuge modeling Eptp = N (3)

Parameter Model units Prototype units Emtm
Length 1/N 1
which is the scaling relation for EA.
Strain 1 1
In order to maintain the similitude of bending
Stress 1 1
moment effects in the buried pipe the flexural rigidity,
Axial rigidity 1/N2 1
El, needs to be scaled as
Flexural rigidity 1/N4 1
Eplp -- N 4 (4)
type of soil in the model as exists for the prototype. This Eml
ensures that both the soil density and friction angle in the where I is the moment of inertia for the pipe. When
model and prototype match. In the simple example cited the t/d ratio is small, the moment of inertia may be
above, the soil density was assumed constant. Clearly, approximated as I = zcaPt/8. Hence, E1 similitude
the effects of g-level upon soil density if any, would need requires
to be included if similitude is to be maintained. Note,
however, that in general, soil density doesn't change as a Epdp3tp - X 4 (5)
result of an increase in g-level (static loading) but rather Emd~3tm
due to vibrations or shaking (dynamic loading). In a
recent study by Byrne et al. (2004) changes in relative However as noted before, d p/ d m scales
.
as N, hence the
density were monitored over a large range of g-levels (1- similitude relation for E1 is identical to that for EA as
120g). It was concluded that the change in the relative given in Eq. (3).
density for a deep soil deposit subject to 120g was on the
order of about 4%. For the soil profile used herein, the 3 Experimental equipment
change in soil density would be about 0.5%.
For buried pipe, the longitudinal forces at the soil- The Rensselaer Geotechnical Centrifuge facility is
pipe interface are influenced by the roughness of the located at the basement level of the Jonsson Engineering
pipe surface. Hence, this quality should be similar in Center. The centrifuge itself is located in a below
both model and prototype. Finally, since the soil forces ground closed circular room with a diameter of roughly
acting on the pipe are linearly proportional to the pipe 6 m (18 feet). Operation of the centrifuge and data
diameter; the pipe diameter needs to scale as N. acquisition is performed in the adjacent control room.
In relation to the buried pipe, the similitude of axial The centrifuge used in the series of tests reported herein
force effects is maintained if the axial rigidity, EA, (the was an acutronic model 665-1.
product of modulus of elasticity and cross section area) The in-flight radius of the centrifuge is 3 m. The
scales by a factor of 1/Nz, when the model is subjected to centrifuge is capable of carrying a maximum payload of
a centrifugal acceleration of Ng. Note, that stress has the 1 metric ton at 100 g (i.e. 100 ton g ton). The maximum
same dimensions or units as modulus of elasticity (Pa speed of the centrifuge is 200 g. The soil model sits
etc.) and hence from Table 1 both have a scaling factor on the testing platform which is located at the end of
of 1. However, again from Table 1, area (having units the centrifuge arm. The maximum model or payload
of length squared) scales by a factor o f N 2. Hence, axial dimensions are 100 cm 80 cm 80 cm. The transfer
rigidity, which is the product of modulus of elasticity of data is enabled by wireless DAQ system and the
and area, scales as control signals is enabled by 64 electric slip rings; 50 for
analog signals, 12 for power and two for video signals.
EpAp _ N2 (1) The hydraulic rotary joints have a total of six passages,
two of which are hydraulic oil passages rated at 3000 psi
(lpsi~0.7 and the remaining four are air/water
where E is the modulus of elasticity, A is the cross- passages rated at 300 psi.
sectional area and m, p refer to model and prototype
respectively. When the wall thickness to diameter ratio, 3.1 Split box
t/d, is small, the area may be approximated as A = ~dt.
Hence, the scaling relation to maintain similitude for EA As noted above, the split box allows simulation of
becomes fault offsets and the relative horizontal displacement at
the margins of a lateral spread. The split box had inside
gptpdp - N2 (2) dimensions of 1 m 0.354 m x 0.203 m. It consists
Emt~dm of two halves; one is fixed while the other can move
horizontally on rollers to simulate an offset. The moving
As noted above, the ratio of diameters d p/ d m scales portions of the container are supported and guided using
with N to ensure similitude of soil forces on the pipe roller bearings to provide precise movement with
interface, hence Eq. (2) becomes minimal friction. The sliding interface between fixed
No. 1 Michael O'Rourke et al.: Centrifugemodelingof PGD response of buried pipe 71

and movable portions of the container utilizes low (50) or 15 in order that Eq. (3) and (5) are satisfied. The
friction Teflon seals protected by steel shields. resulting diameter and wall thickness for the models are
A hydraulic cylinder is used to displace the movable also presented in Table 2. The same soil, was used for
half of the split box. The driving shear force is provided both pipeline tests; dry sand with a unit weight of 18.9
by a 3000 psi hydraulic actuator system, which includes kN/m 3 and a friction angle of 35 ~
a flow-metering valve, a solenoid valve for remote Figures 1(a) through 1(d) show the axial and bending
operation, and hoses for connecting to the centrifuge's strains along the pipe as measured by strain gauges on
quick connects. The load cell, located between the the Pipeline #1 model for an offset of 0.4, 0.8, 1.2 and
actuator and the movable portion of the split box, 1.6 cm respectively. The largest of these correspond to a
measures the force applied by the actuators. The prototype offset of 0.8 m.
maximum relative displacement of the movable section As one would expect, the bending and axial strains
is 8 cm, simulating 4m offset at 50 g. The flow-metering are increasing functions of the offsets. For these tests
valve controls the rate of movement of the split box. The with a 90 ~ interaction angle between the pipe axis and
motion of the actuator is controlled by a servo valve and the fault trace, the bending strains dominate. Also note
a feedback control system, while an LVDT measured the that the fault location is a point of counter flexure (i.e.
offset. zero bending moment) due to the asymmetric nature of
the offset.
3.2 Anchor points Figures 2(a) through 2(d) present similar information
for the Pipeline #2 model, specifically axial and bending
The connection between the pipe model and the strains for model offsets of 1, 2, 3 and 4 cm respectively.
split box end walls consist of a steel rod that sits in For these tests at 50 g's, the largest of these offsets
one of the metal plates attached to the split box wall. correspond to 2 m in prototype scale.
This assembly allows the pipe to rotate freely about the
vertical axis. The assembly also has a space for two 5 Finite element model
LVDT's, which can measure rotation of the pipe model
at the support. This assembly is covered by a plastic A finite element idealization of the centrifuge model
enclosure and rubber sheeting to prevent the entry of was constructed. The pipe was modeled with beam
sand near the LVDT's. elements while the soil was modeled as elasto-plastic
springs. The burial depth for the pipeline models was
3.3 Instrumentation and data acquisition system 2.4 cm (little under 1 inch) which corresponds to 1.2
m (-4 feet) in prototype scale. Since the burial depth
Strain gauges were installed on the pipe model to was constant across the model, and the offset had no
measure the axial and bending strains at various points. vertical component, there was no need for vertical soil
The strain gauges were model CEA-032UW-120 from springs. That is, the FE model was two dimensional in
the Measurement Group Inc. Ten strain gauges were set nature. Hence, the soil was modeled by longitudinal
in a quarter bridge configuration and the remaining four and transverse - horizontal elasto plastic springs. The
strain gauges in a half bridge configuration. stiffness of the soil springs was based upon relations
The date acquisition system is capable of recording in the 1984 ASCE Gas and Liquid Fuels Guidelines.
128 channels of data at a 10 kHz sampling rate per Specifically the elasto-plastic soil springs were
channel. A Pentium 4 PC generates the digital input characterized by a maximum resistance (force per unit
signal, which causes the split box to move by the desired length) and the corresponding "yield" displacement. For
offset. The servo controller on the centrifuge ann example, the maximum resistance for the longitudinal
receives the signal through the slip rings. A hydraulic soil spring, f, is
pump is used to pressurize the actuator in the split box
before sending the signal. f = m / 7 / H ( ~ -~) tan k~b (6)

4 Centrifuge tests where y is the unit weight of soil, H is the burial depth
to the pipe centerline, ko is the coefficient earth pressure,
Two pipelines were tested in the Rensselaer q~ is the friction angle at the soil and k is the coefficient
centrifuge at 50 g's. The diameter and wall thickness related to the condition of the soil pipe interface.
of the steel prototype lines are listed in Table 2. The The pipe model was taken to be pinned at the split
diameter and d/t ratios are common for larger gas and box wall (i.e. end of FE model located 40 cm (20 m in
liquid fuel pipe in the U.S. prototype scale) each side of the fault location). The
Commercially available small diameter pipe typically offset was simulated by displacing the base of all the soil
do not have large d/t ratios. For that reason, aluminum springs located on one side of the fault as well as the pin
was chosen as the model pipe material. That is, since the end on that side of the fault.
E rn /E p ratio for the aluminum model and steel prototype The results of the FE simulation are also shown in
is about 0.30, the scale factor for wall thickness was 0.3 Figs. 1 and 2. For Pipeline #1 (see Fig. 1) the axial and
72 EARTHQUAKE ENGINEERING AND ENGINEERING VIBRATION Vol.4

Table 2. Pipeline properties

Property Pipeline #1 Pipeline #2
Prototype Model Prototype Model
Diameter 0.64 m (25 in) 1.27 c m (0.5 in) 0.95 m (37 in) 1.9 c m (0.75 in)
Wall thickness 1.1 c m (0.43 in) 0.71 m m (0.0279 in) 1.85 cm (0.73 in) 1.27 m m (0.05 in)

2000 3000
....... Expected axial strain ~y~ - ...... Expected axial strain
150~ .......................Expected bending strain i\ ..........7-- Expected bending strain
9 Observed axial strain 2000 / ~ ~ Observed axial strain
1000 // \ x Observed bending strain
.-. 1000
500 !
........ ~ ........ ~ ..... . . . . . . . , ..... ~__~. . . . . ~, ...... ~,--
0 ~-~ 0
.9

r~
-500
-1000
....... 8o
-1000 5f 8o
-2000
-1500
-2000 -3000
Position (cm) Position (cm)
(a) 0.4cm offset (b) 0.8cm offset
4000 l 5000
3000 / ....... Expected axial strain 4000 ....... Expected axial strain
........................Expected bending Strain ................... Expected bending strain

2000/] ~ Observed axial strain 3000 A Observed axial strain

x Observed bending strain .--. 2000 Observed bending strain
1000 / g lOOO ......... i;; ................ ?'A . . . . . . . A.......... ~---
........... ~........ ~ .......... ~...
.9 0 0
.9
_lOOOl 20 ................. E -1000 20 4~ 9 ///~,0' ....... 80
c~ -2000
-2000[ ~ /
-3000
-3000~ '\',,~}'' -4000
4000 ~............................................................................ -5000
Position (cm) Position (cm)
(c) l . 2 c m o f f s e t (d) 1.6cm offset

Fig. 1 Distribution of measured and simulated strains for Pipeline #1 model

4000
2000 - ...... Expected axial strain ~..
....... Expected axial strain
1500 ./ ~ .....~ Expected bending strain 3000 ............................... Ex!oected bending strai~
/ ", Observed axial strain 2000 5, A Observed axial strain
1000 ............, ~ ~ Observed bending strain Observed bending strait
500 ~..:,,,2 .................... ~ ~.,,~ ..... ~ .......... ~ . . 1000
................... ~ ~ 7 , .... x .......... ~- -
.9 yr ;g ~<
.9 0
e~
-1000 20 40~ 60 ....... 80
~t3 \ ///
~i i~"'~9
iii 20 40,'~ 60 ................80 ~I
-2000
-3000
t
-2000 L................................................................................................................................................................................ -4000
Position (em) Position (era)
(a) l c m offset (b) 2 e m offset
5000 ............................................................................................................................................................. 5000
4000 ....... Expected axial strain 4000
* ~ Expected bending strain
3000 Observed axial strain 3000
2000 / ' ,. ~ Observed bending strain 2000
i
1000 1000
0 .......* ,~ . .... I 0
20 40" 60 [ .9
-1000( -1000(
._=
-2000 ',, ,J .... ! -2000
-3000 -3000
-4000 " ti -4000
-5000 ...................................................................................................................................................... -5000
Position (cm) Position (cm)
(c) 3 c m offset (d) 4 c m offset

Fig, 2 Measured and simulated strains for pipeline #1 model

No. 1 Michael O'Rourke et al.: Centrifugemodelingof PGD response of buried pipe 73

bending strains from the FE simulation match well the knowledge of the actual stress-strain behavior of the
measured strains. The match is remarkably good for centrifuge pipe material. Nonetheless, the results
offsets of 1.2 and 1.6 cm while somewhat less remarkable presented herein strongly suggest that centrifuge
at lower offsets. For Pipeline #2 (see Fig. 2) the match techniques are a valid method to determine PGD
between measured strains in the centrifuge experiments behavior of buried pipe beyond the elastic range.
and corresponding values from the FE simulation is quite
good for offsets of 2 cm or less. As the offset increases Acknowledgements
beyond 2 cm, the correspondence between measured and
simulated strains deteriorates somewhat. In this range, The research work described herein was sponsored
where the pipe material is beyond yield, the bending by the National Science Foundation through Award No.
strains compare reasonably well, but the measured axial CMS-0085256. The original NSF program manager
strains are somewhat lower than those from the FE was Vijaya Gopu, who was succeeded by Peter Chang.
simulation. It is thought that this mismatch may be due The construction of the split box was sponsored by
to compliance or inward movement at the anchor point. the National Science Foundation through the George
That is, in the FE model the anchor points do not move E. Brown, Jr. Network for Earthquake Engineering
while in the centrifuge model, they may move slightly, simulation program (NEES). This support is gratefully
resulting in less measured axial stress. acknowledged. However, all statements, results and
conclusions are the authors and do not necessarily reflect
6 Conclusions the views of NSF.

The paper describes the first known attempt to use References

centrifuge modeling to determine pipe strains induced
by PGD. The tests were successful in the sense that the American Society of Civil Engineers (ASCE) (1984),
experimental equipment apparently functioned properly Guidelines for the Seismic Design of Oil and Gas
and strains were measured as the split box was being Pipeline Systems, Committee on Gas and Liquid Fuel
offset in-flight. Furthermore, the measured pipe strains Lifelines, ASCE, pp. 150-176.
matched analytical results from a FE simulation for
Bransby M, Newson T and Brunning P (2002),
offsets wherein the pipe remained in the linear elastic
"Centrifuge Modeling of the Upheaval Capacity of
range.
Pipelines in Liquefied Clay," Proc. ISOPE, Kitakyushu,
The agreement between pipe strain values measured
Japan, pp.899-904.
in the centrifuge and those predicted by FE modeling
was quite good for small offsets where the pipe remained Bransby M, Newson T, Davies M and Brunning P
elastic. At larger offsets, the agreement between the (2002b), "Physical Modeling of Upheaval Resistence
measured and simulated bending values remained good. of Buried Offshore Pipelines," 4th Int. Conf. Physical
However, the measured axial strains were less than those Modeling in Geomechanics, St.Johns, Newfoundland,
predicted by the FE method. It is thought that the inward pp.899-904.
movement of the split box wall at the pipe anchor points Byrne P, Park S, Beaty M, Sharp M, Gonzalez L,
may account for these differences. Abdoun T (2004), "Numerical Modeling of Liquification
Finally, it must be noted that FE simulation of and Comparison with Centrifuge Tests," Canadian
large offsets (i.e. pipe in the inelastic range) requires Geotechnical Journal, 41(2): 193-211.