Professional Documents
Culture Documents
Ding Etal 2006 PDF
Ding Etal 2006 PDF
www.elsevier.com/locate/engstruct
Received 2 June 2005; received in revised form 5 December 2005; accepted 5 January 2006
Available online 23 March 2006
Abstract
This paper presents a three-dimensional numerical simulation method for large-scale seismic response calculation based on a newly built
immersed tunnel in Shanghai. The whole analytical model consists of the surrounding soil, the tunnel segments, and the detailed flexible joints.
Both the number of nodes and the number of elements exceed one million. This model takes account of nonlinear material behavior such as soil,
non-reflecting boundary definition, and soil–tunnel interaction. Final calculation has been performed on the Dawning 4000A supercomputer using
the finite-element code LS-DYNA 970 MPP. The result provides a global understanding of this immersed tunnel under seismic loading and reveals
the weak parts. It also provides relevant data and references for the aseismic design of immersed tunnel and flexible joints in the future.
c 2006 Elsevier Ltd. All rights reserved.
Keywords: Immersed tunnel; Soil–tunnel interaction; Seismic response; Supercomputer; FEM; Non-linearity
After the nodal location and the nodal acceleration at time tn Friction in LS-DYNA is based on a Coulomb formulation,
are acquired together with the nodal velocity at time tn+1/2 , the and the maximum frictional force Fv is defined as:
nodal displacement at time tn+1 can be calculated by Eq. (5). In
the time domain, the displacement, velocity, and acceleration Fv = µ| f s | (10)
of each discrete point can be calculated through such integral where µ is the coefficient of friction and f s is the normal force
recursive formulae. at slave node n s .
CDM is suitable for solving complicated problems such The possible frictional force f ∗ of the tn+1 time step
as wave propagation [11] due to the fact that, when some can be obtained through the friction force f n of time step
nodes are first disturbed, neighboring nodes and then other tn . Meanwhile, since the value of f ∗ should not exceed the
nodes will behave gradually with time, which is consistent with maximum frictional force Fv , the real friction force f n+1 of
the features of wave propagation. However, it is necessary to the next time step will be treated differently according to the
guarantee a small time step for modeling wave propagation, contrast between | f ∗ | and Fv :
because the explicit CDM is conditionally stable. In fact, the
time step size depends on the minimum natural period of the f n+1 = f ∗ (| f ∗ | ≤ Fv );
whole mesh to guarantee the calculation stability of the central f n+1 = Fv f ∗ /| f ∗ | (| f ∗ | > Fv ). (11)
difference method. Hence, during the solution, a new time
step size is determined by taking the minimum value over all The time cost of common calculations using explicit
elements: algorithms such as LS-DYNA depends on the element size and
wave speed. However, when it comes to parallel calculation,
t n+1 = α · min{t1 , t2 , t3 , . . . , t N } (6) it is the parallel algorithm involving contact that affects the
calculation efficiency.
where α is a scale factor, N is the number of elements, and t
Hallquist et al. [12] developed a standard algorithm in LS-
is the critical time step size.
DYNA for problems with contact. According to Eq. (1), this
algorithm can be extended and described as follows:
2.2. Parallel algorithm for contact interface
1. Determine the mass matrix M (diagonal matrix).
In the soil–tunnel analytical model, some parts that contact 2. Sub-divide the finite-element mesh into groups, adjust the
each other will possibly have extrusion and sliding behaviors numbers of elements in the groups for load balancing, then
under seismic loading. To handle this problem, it finally comes assign the shared nodes to the groups for contact checking.
down to the search for contact objects and the calculation of For each time step:
contact forces as a nonlinear boundary condition. 3. Perform a calculation to determine (P − F + H − C ẋ),
Such a contact problem is often solved by the symmetrical calculate the contact force vector Rc (including the normal
penalty method [9], which consists of placing normal interface contact force and the tangential frictional force), and add the
springs between all penetrated nodes and the contact surface. contact force vector Rc to the global load vector P to form a
First, the penetration between the slave node n s and the master new one P (within each group).
segment si is judged within each time step. If there is no 4. Compute acceleration (within each group): ẍ = M −1 (P −
penetration, there is no treatment; otherwise, a normal contact F + H − C ẋ).
force f s will be calculated: 5. Adjust the accelerations of shared nodes between the groups
(communicate between processors):
fs = mki n i (7)
ẍ = ẍ g1 + ẍ g2 + ẍ g3 + · · · .
where m is the amount of penetration, n i is the normal to the
6. Compute the velocity and the displacement (within each
master segment at the contact point, and ki is the stiffness factor
group).
for the master segment,
7. Continue with step 3.
f K i A2i
ki = . (8)
Vi 2.3. Material model for soil
Here, f is a scale factor for interface stiffness, and K i , Ai
and Vi are the bulk modulus, the volume, and the surface area Several soil layers close to the tunnel mainly influence
of the element that contains si , respectively. Meanwhile, each of the seismic response of the underground tunnel. Here, the
the four nodes ( j = 1, 2, 3, 4) that comprise si adds a contact soil’s behavior is assumed to be governed by an elastic–plastic
force, constitutive relation based on the Drucker–Prager criterion.
Considering the possible influence by the hydrostatic stress
j
fm = φ j (ξc , ηc ) f s (9) component on the yield strength of soil and rock material,
the Drucker–Prager criterion [13] is described on the basis of
where φ j (ξc , ηc ) is the value of the shape function of si at the modification of the von-Mises yield criterion:
contact point coordinates (ξc , ηc ). Similarly, the treatment will
then be applied to the master nodes and the slave surfaces again. F = α J1 + (J2 )1/2 − k = 0 (12)
1370 J.-H. Ding et al. / Engineering Structures 28 (2006) 1367–1377
where J1 is the first invariant of the stress tensor, and J2 is 2.5. Material model for steel shear key
the second invariant of the stress deviator tensor. The material
constants α and k can be calculated as: The material model called “MAT PLASTIC KENEMATIC”
2 sin φ 6c cos φ in LS-DYNA [14] can be used to model the material behavior
α= √ ; k= √ (13) of the steel shear key among the flexible joint components.
3(3 ± sin φ) 3(3 ± sin φ) This model is suitable for modeling isotropic and kinematic
where ϕ is the angle of internal friction and c is the cohesion hardening plasticity with the operation of including rate
value. Here, the positive and negative signs indicate the tensile effects. Varying a parameter called β between 0 and 1 can
and compressive conditions, respectively. specify isotropic, kinematic, or a combination of isotropic and
The flow rule defining the direction of the plastic flow is kinematic hardening.
given by Meanwhile, the strain rate is accounted for by using the
Cowper–Symonds model [9,16], which scales the yield stress
∂G by a strain rate dependent factor
ε̇ p = λ̇ (14)
∂σ 1/ p
ε̇ p
where G is a plastic potential and λ̇ is a positive scalar quantity σy = 1 + (σ0 + β E p εeff ) (18)
defining the amplitude of plastic flow. For non-associative C
plasticity, the plastic potential is selected so that its derivative where p and C are the Cowper–Symonds strain rate parameters,
with respect to the stress tensor yields and ε̇ is the strain rate. Therefore, the current radius of the yield
∂G 1 surface σ y is the sum of the initial yield strength σ0 plus the
= αψ δi j + √ si j (15) p p
growth β E p εeff . Here, εeff is the effective plastic strain, and E p
∂σi j 2 J2
is the plastic hardening modulus, defined as:
where δi j is the Kronecker delta, si j is the stress deviator,
E tan E
and αψ is defined by a given dilation angle ψ and relations Ep = (19)
analogous to Eq. (13). Moreover, a Drucker–Prager criterion E − E tan
with elastic–perfectly plastic material response is considered where E tan is the tangent modulus and E is the Young’s
with respect to the hardening behavior. modulus.
The surrounding soil layers in the study can be modeled
with the use of the card “MAT DRUCKER PRAGER” in 2.6. Material model for cable
LS-DYNA [14].
The material model called “MAT CABLE DISCRETE
2.4. Material model for rubber BEAM” in LS-DYNA [14] can be selected to modify the
material behavior of prestressed steel cable acting as another
kind of flexible joint components.
As two kinds of flexible joint components, GINA gasket and
This material can be used as a discrete beam element. The
vibration isolation bearing are both made of rubber material.
force, F, generated by the cable is nonzero if and only if the
Rubber shows nonlinear features in both geometric and material
cable is in tension. The force is given by:
behavior. The Mooney–Rivlin model is a model that is widely
used for simulating rubber material in many finite-element F = max(F0 + K L, 0.) (20)
analysis software products [15]. In this model, the strain energy
density function [9] is used to describe incompressible rubber where F0 is the initial tensile force, L is the change in length,
and K is the stiffness. In Eq. (14), L can be calculated
material behavior as follows:
by:
W = C10 (I1 − 3) + C01 (I2 − 3) + C(I3−2 − 1) + D(I3 − 1)2
L = L cur − (L ini − Doff ) (21)
(16)
where L cur is the current length, L ini is the initial length, and
where I1 , I2 , and I3 are invariants of the right Cauchy–Green Doff accounts for the offset used for F0 . Here, the stiffness K is
Tensor; and C10 and C01 are two material constants that can defined as:
be obtained through the calculation based on uniaxial stretch E · Ac
or shear experimental data. Here, C and D are two constants K = (22)
(L ini − Doff )
decided by C10 and C01 , which can be described as:
where E is the Young’s modulus and Ac is the cross-sectional
C10 (5υ − 2) + C01 (11υ − 5)
C = 0.5C10 + C01 ; D= area. The offset Doff can be input as a positive length for an
2(1 − 2υ) initial tensile force in this study.
(17)
2.7. Non-reflecting boundary
where υ is Poisson’s ratio.
In LS-DYNA, this model has been incorporated and can be It is inevitable that the finite boundary of the finite-
used with the card “MAT MOONEY RIVLIN RUBBER” [14]. element model will cause the seismic waves to be reflected
J.-H. Ding et al. / Engineering Structures 28 (2006) 1367–1377 1371
3. Finite-element modeling
Table 3
Material parameters of several upper soil layers surrounding tunnel
Soil layer Dynamic shear modulus (MPa) Poisson’s ratio Density (g/cm3 ) Cohesion (kPa) Angle of internal friction (◦ )
Drab yellow silty clay 15.38 0.40 1.93 19 20.6
Gray sandy silt 16.64 0.35 1.86 7 24.8
Gray mucky silty clay 31.37 0.45 1.76 14.1 13.5
Gray sandy silt 68.74 0.35 1.86 8.4 19.4
Gray mucky clay 31.35 0.45 1.75 12 15.3
Gray silty clay 40.42 0.45 1.72 12.5 7.4
Gray sandy silt 56.74 0.40 1.76 9.8 10.2
Fig. 13. Maximum relative displacement between adjacent tunnel segments: Fig. 14. Maximum cable tension at eight flexible joints: (a) seismic wave along
(a) seismic wave along the transverse direction; (b) seismic wave along the the transverse direction; (b) seismic wave along the longitudinal direction.
longitudinal direction.
Table 4
Compression measuring results of GINA gaskets and isolation bearings
[12] Hallquist JO, Wainscott B, Schweizerhof K. Improved simulation of thin- Fracture Mechanics 2002;38:165–75.
sheet metal forming using LS-DYNA3D on parallel computers. Journal of [16] Jones N. Structure impact. Cambridge: Cambridge University Press;
Materials Processing Technology 1995;50(1–4):144–57. 1989. p. 211–47.
[13] Oettl G, Stark RF, Hofstetter G. A comparison of elastic-plastic soil [17] Harari I, Barbone PE, Montgomery JM. Finite-element formulations
models for 2D FE analysis of tunneling. Computers and Geotechnics for exterior problems: Application to hybrid methods, non-reflecting
1998;23:19–38. boundary conditions, and infinite elements. International Journal for
[14] Hallquist JO. LS-DYNA keyword users manual version 970. Livermore: Numerical Methods in Engineering 1997;40(15):2791–805.
Livermore Software Technology Corporation; 2003. [18] Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media.
[15] Batra RC, Ching HK. Energy release rates in a constrained epoxy disc Journal of the Engineering Mechanics Division ASCE 1969;95(EM4):
with Hookean and Mooney–Rivlin materials. Theoretical and Applied 859–77.