Você está na página 1de 19

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225


Published online 13 September 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/nag.951

Stabilization procedures in coupled poromechanics problems:


A critical assessment

Matthias Preisig∗, † and Jean H. Prévost


Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ 08544, U.S.A.

SUMMARY
Numerical solutions for problems in coupled poromechanics suffer from spurious pressure oscillations
when small time increments are used. This has prompted many researchers to develop methods to overcome
these oscillations. In this paper, we present an overview of the methods that in our view are most promising.
In particular we investigate several stabilized procedures, namely the fluid pressure Laplacian stabilization
(FPL), a stabilization that uses bubble functions to resolve the fine-scale solution within elements, and
a method derived by using finite increment calculus (FIC). On a simple one-dimensional test problem,
we investigate stability of the three methods and show that the approach using bubble functions does not
remove oscillations for all time step sizes. On the other hand, the analysis reveals that FIC stabilizes the
pressure for all time step sizes, and it leads to a definition of the stabilization parameter in the case of the
FPL-stabilization. Numerical tests in one and two dimensions on 4-noded bilinear and linear triangular
elements confirm the effectiveness of both the FPL- and the FIC-stabilizations schemes for linear and
nonlinear problems. Copyright 䉷 2010 John Wiley & Sons, Ltd.

Received 8 February 2010; Revised 15 April 2010; Accepted 10 May 2010

KEY WORDS: poromechanics; stabilized finite elements; fluid pressure Laplacian; finite increment
calculus; bubble functions

1. INTRODUCTION

In coupled poromechanics problems, finite elements sometimes exhibit pressure oscillations espe-
cially near draining boundaries. These oscillations can be caused by (a) the incompressibility
constraint and can be attributed to the violation of the LBB-condition [1, 2] or (b) by violation
of the minimum time step criterion [3, 4]. Methods that work well for incompressible elasticity,
such as Bbar [5] or elements that satisfy the LBB-condition, do not guarantee oscillation-free
results in the case of poromechanics. The oscillations can be overcome by locally refining the
mesh especially near the draining boundary. In practical applications this is, in general, not
feasible because of the resulting computational cost. An alternative solution is to use stabi-
lized methods, which allow the use of smaller time steps without the need for refining the
mesh. We focus on methods that are applicable to most commonly used finite elements in two
dimensions: the four-node bilinear quadrangle and the three-node linear triangle with equal-order
interpolants.
The problem of oscillating pore fluid pressure was identified almost simultaneously about
30 years ago by Vermeer and Verruijt [4] and by Gresho and Lee [3] for the mathematically

∗ Correspondence to: Matthias Preisig, Department of Civil and Environmental Engineering, Princeton University,
Princeton, NJ 08544, U.S.A.
† E-mail: mpreisig@princeton.edu

Copyright 䉷 2010 John Wiley & Sons, Ltd.


1208 M. PREISIG AND J. H. PRÉVOST

equivalent transient heat conduction problem. Their conclusion was that one had to refine the
mesh until the minimum time step constraint is satisfied. Since then, many authors have worked
on developing stabilized methods that allow obtaining accurate solutions using coarser meshes.
With so many different approaches having been published, very few are found to be (a) practical,
and (b) stable and accurate. In this paper we try to give an overview of the existing stabilization
schemes, apply them to coupled poromechanics and investigate the ones that in our opinion
are most promising (and practical) with respect to their stability and accuracy. We introduce
a novel approach for assessing stability of a wide variety of stabilized procedures in a unified
fashion. On the one-dimensional case, the coupled poromechanics equations can be simplified
by decoupling the continuity equation from the equation of conservation of momentum. This
allows investigating directly the instability in the pressure field that occurs at very small time
steps. Stability can, for this case, be established analytically. Further, stability and accuracy of the
obtained schemes are analyzed and confirmed numerically in two dimensions. All of the methods
that work in the decoupled one-dimensional problem also produce oscillation-free results in two
dimensions.
Out of the many approaches proposed by other researchers, some are only mentioned briefly due
to their difficulty in being used in practical applications. Discontinuous Galerkin methods [6, 7],
although they are expected to produce stable results, are not considered because of the additional
computational cost caused by the increase of degrees of freedom they require. A coupling between
discontinuous and continuous Galerkin methods [8] can, however, reduce the computational penalty.
Another very promising class of methods is generally referred to as residual methods. The idea
is to add to the weak form a residual-term that will vanish as the approximate solution tends
to the exact solution. In the method developed in Wan et al. [9], the residual of the momentum
equation is used for stabilization. The necessity to compute recovered stresses with high accu-
racy makes this method impractical in general for practical applications due to its computational
burden.
Mira et al. [10] and White and Borja [11] propose methods that decrease the amplitude of
the pressure oscillations, but fail to completely remove them near the draining boundary. These
methods fail to stabilize the problem of coupled poromechanics.
In Tchonkova et al. [12], a stabilization procedure based on the minimization of a least-squares
functional is presented. Results for a relatively large initial time step seem to be accurate at the
beginning, but degrade at later times.
The work done in the field of time splitting procedures for solving the Navier–Stokes equations,
based on characteristic-based split (CBS) procedures [13], has been extended to poromechanics by
Pastor et al. [14] and Salomoni and Schrefler [15]. In their derivation of the CBS-type formulation
they include the acceleration of the fluid in Darcy’s law, which generally would be neglected. This
can be seen as a way to include inertial effects that might be important at very early times near
the boundary.
Truty and Zimmermann [16] derived a method based on the stabilization of the Stokes
problem proposed by Brezzi and Pitkäranta in [17]. The stabilization parameter in this formula-
tion, called stabilization through fluid pressure Laplacian (FPL), is derived from the minimum
time step requirement proposed by Vermeer and Verruijt [4]. Aguilar et al. [18] use the
same stabilization as proposed by Truty and Zimmermann [16], which they fail to refer-
ence, and define a stabilization parameter that does not depend on the time step size. This
shortcoming causes inaccurate results at larger time steps. In this paper, we propose a stabi-
lization parameter that is consistently derived from the decoupled pressure equation in one
dimension.
Variational multiscale methods using bubble functions to approximate the fine scale in each
element have been proposed in Nakshatrala et al. [19] for the Stokes problem. The approach is
based on the work by Hughes [20]. We analyze this method in one dimension and show that the
form of the stabilization matrices produced by it is almost identical to one of the FPL method,
depending on the form of the selected bubble function. However, the stabilization parameters of
the two approaches differ substantially. Further, the analysis will reveal for the first time that the

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1209

bubble function approach does not lead to a useful stabilization in one dimension, which we believe
is necessary for the method to work in higher dimensions.
The last method that we investigate is a stabilization based on finite increment calculus (FIC),
developed by Oñate and coworkers [21, 22]. In this approach, stabilization terms are derived by
expanding the residual of the governing equation in a Taylor series. We will apply FIC to the
equations of coupled poromechanics for the first time. We show analytically in one and numerically
in two dimensions that the FIC-stabilization is unconditionally stable and produces accurate results.
The remainder of the paper is organized as follows: In Section 2, the governing equations of
coupled poromechanics are presented and the weak form is derived. Section 3 introduces the FPL-
stabilization and derives the stabilization terms based on a quadratic and a piecewise linear bubble
function. A stability analysis in one dimension provides the stabilization parameter for the FPL,
whereas it reveals only conditional stability for the bubble function approach. A similar analysis
is performed for the FIC-stabilization. In Section 4, the FPL- and FIC-stabilizations are tested on
one- and two-dimensional linear and nonlinear benchmark problems.

2. POROELASTICITY EQUATIONS

The detailed derivations of the poromechanics equations can be found in Coussy [23]. We consider
a fully saturated medium of porosity , solid mass density s and fluid mass density f . The mass
density of the saturated porous medium is  = [(1−)s +f ]. Conservation of momentum can
then be written as
∇ ·tot +g = 0 (1)
 
where g is the gravity acceleration. tot is the sum of effective stress  and pore fluid pressure p:
  
tot =  −bp (2)
  
where  is the identity tensor. Biot’s coefficient b depends on the compressibility of the solid
matrix:b = 1− K /K s where K is the bulk modulus of the dry porous material and K s is the bulk
modulus of the grains.
The equation of continuity relates the fluid pressure to the divergence of the fluid velocity q f
and the divergence of the solid velocity v s : 

1
ṗ +∇ ·q f +b∇ ·v s = 0 (3)
M  
with
1 b − 
= + (4)
M Ks Kf
In this work we focus on incompressible fluids, the case that represents most difficulties. In this
case, the fluid bulk modulus K f becomes infinity. For very small time steps the behavior of the
saturated, porous medium is quasi-incompressible as the fluid is given no time to evacuate.
Darcy’s law relates the fluid velocity to the pressure gradient ∇ p and the body force f g :

q f = −(∇ p −f g ) (5)
 
 is the mobility and is defined by  = k/f , where k is the permeability and f the dynamic
viscosity of the pore fluid. Assuming that the body force f g is constant we can substitute Darcy’s

law into the continuity equation, leading to the following system of equations:
∇ · −b∇ p +g = 0 (6)
 
1
ṗ −∇ 2 p +b∇ ·v s = 0 (7)
M 

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1210 M. PREISIG AND J. H. PRÉVOST

2.1. Weak form


The weak form of Equations (6) and (7) is expressed as
B(w, q; v s , p) = L(w , q) (8)
  
with
  
 1
B(w, q; v s , p) = (w) :  d− ∇ ·wbp d+ q ṗ d
        M
 
+ ∇q ·∇ p d+ bq∇ ·v s d (9)
  
 
L(w, q) = f w · g d+ w ·t d (10)
     

3. STABILIZED FORMULATIONS

We analyze stability of the stabilized formulations on a one-dimensional problem, for which the
coupled poromechanics equations can be simplified. This is achieved by expressing ∇ ·v s in terms
of ∇ ṗ. First, we take the time derivative of Equation (2): 
 
 4
˙ tot
xx = 0 = 
˙ xx −b ṗ ⇒ b ṗ = K + ˙xx (11)
3
Here  is the shear modulus of the solid material. Since yy = zz = 0, we can write:
b
∇ ·v s = ˙xx = ṗ (12)
 4
K+
3
Using the above in Equation (7) yields a one-dimensional pressure equation
c ṗ −∇ 2 p = 0 (13)
where  = xx and
1 b2
c= + (14)
M 4
K+
3
Stabilized formulations are obtained by adding terms to the weak form (Equation (8)):
B(w, q; v s , p)+ B(q; p)stab = L(w, q) (15)
  
B(q; p)stab is a sum of elemental contributions integrated over the interior e of each element e.

3.1. FPL-stabilization
In the FPL-stabilization, the rate of the residual of Equation (13) integrated over the element
interiors e is added to the weak form:
n el 

B(q; p)stab = q(c p̈ −∇ 2 ṗ) d (16)
e=1 e

is a stabilization parameter and will be specified later. The second time derivative of the pressure
p̈ is equal to zero, since we use a first-order time integration scheme. Integration by parts on the

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1211

second term in B(q; p)stab leads to a term of the same form as in [16]:


n el
B(q; p)stab = ∇q ·∇ ṗ d (17)
e=1 e

3.1.1. Stability analysis of the FPL-stabilization. Stability of the formulation is analyzed on the
one-dimensional problem. Therefore, we use the form where the fluid velocity has been eliminated
from the continuity equation (see Equation (13)):
 
cq ṗ d+ ∇q ·∇ p d+ B(q; p)stab = 0 (18)
 

Using linear finite elements, this equation leads to the following discrete matrix form:

C ṗ + K p = 0 (19)
 
The global matrices are assembled as follows:

   
he 2 1 1 −1
C= A#elements
e=1 C e, Ce =c + (20)
   6 1 2 h e −1 1

 
1 1 −1
K = A#elements
e=1 K e, Ke = (21)
   h e −1 1

h e is a characteristic element length.

Remark 1
In this analysis, we use a consistent mass matrix because the contribution from the divergence
of the solid velocity (Equation (12) and the second term in Equation (14)) cannot be lumped in
general.

Remark 2
Note that with this form of stabilization, a lumped mass matrix formulation (in one dimension)
can be retrieved by selecting the stabilization parameter to be = (ch 2e /6)(1/). This has first been
observed in the context of bubble functions by Franca and Russo [24].

For a stability analysis, we assemble two elements. The global matrices C and K then become:
 

⎡ ⎤
he he
⎢ c 3 + h e 6
−
c
he
0⎥
⎢ ⎥
⎢ h ⎥
⎢ e he he ⎥
C = ⎢ c − 2c +2 c − ⎥ (22)
 ⎢ ⎢
6 he 3 he 6 he ⎥

⎣ he he ⎦
0 c − c +
6 he 3 he
⎡ ⎤
1 −1 0
1 ⎢ ⎥
K =  ⎣ −1 2 −1⎦ (23)
 he
0 −1 1

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1212 M. PREISIG AND J. H. PRÉVOST

Let us introduce the following finite difference formula for discretization in time:
p n+1 − p̃
ṗ = n+1 (24)
 n+1
t
p̃ = p n +(1−
)t ṗ (25)
n+1  n
where
∈ [0.5, 1]. This leads to the fully discrete matrix equation at time tn+1 :
 
1 1
C + K p n+1 = C p n+1 (26)

t   
t  

with initial conditions p0(k) = 0 at all nodes k except at the boundary ( p (1) = − p0 ). This is a linear
problem; therefore, the absolute value of the pressure does not matter. Thus, for simplicity, we
shift the initial pressure field by − p0 such that at all interior nodes p = 0 initially, and such that
p = − p0 at the draining boundary. The second equation, which is identical for all interior nodes
i = 2, . . . , n −1, takes the form Ap (i−1) + Bp (i) +C p (i+1) = 0. For constant coefficients A, B and
C, the solution is of the form p (i) = r k . Substituting this into the equation for node i yields

−B ± B 2 −4AC
r= (27)
2C
The first condition for a solution to exist is that the discriminant has to be greater than zero.
This results in the following condition on :
h 2e 1
>−c −
t (28)
12 
For the series to be non-oscillatory, we need r >0. Because the numerator is always smaller than
0, the series does not oscillate if C<0:
c he   
− 1+ <0

t 6
t h e
he  
c − <
t (29)
6 he he
h 2e
>c −
t
6
The first condition is always satisfied if the second is satisfied.
Note that this condition indicates that can be negative. Numerical tests, however, clearly show
that negative do not work. Therefore we introduce a switch that assures accuracy at large time
steps, where no stabilization is needed. We propose the following stabilization approach:
⎧ 2

⎪ h h2 c h2
⎪ c e −
t for
t< e = e
⎨ 4 4  4cf
= (30)

⎪ h 2e c h 2e

⎩0 for
t =
4  4cf
The 4 in the denominator is chosen in order to make satisfy Equation (29). With this we account
for the fact that the definition of the characteristic element length h e in multi-dimensions is not
as straightforward as in one dimension. The numerical tests in Section 4 will confirm our choice.
The diffusivity coefficient, which will be used later in this paper, is defined as [23]
4
K+ 
cf = M 3 (31)
4
K +b2 M + 
3

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1213

3.2. Bubble function stabilization


Using the concept of variational multiscale methods an expression for the fine-scale solution
can be found, which can be approximated using a bubble function. The original work by
Nakshatrala et al. [19] for nearly incompressible elasticity is extended to poromechanics. The
detailed derivation is presented in the appendix. The stabilization terms take the following form
(Equation (A5)):
 
B(q; p)stab = −2 c∇q (x)∇ ṗ d− c∇q∇ (x) ṗ d
 

− 2 ∇q∇ 2 (x)∇ p d (32)


Two forms of bubble functions are investigated: A quadratic and a piecewise linear ‘hat’-function.
For both forms, it turns out that the three stabilization terms cancel themselves mutually, resulting
in B(q; p)stab = 0.

3.3. Finite-increment calculus


A third stabilization method can be derived using FIC, a framework for deriving stabilized methods
proposed by Oñate et al. in [21, 22]. The FIC-approach is based on the idea that the governing
equations can be expressed on a discrete domain by the standard equations plus the first-order terms
of a Taylor-expansion. The continuity equation (Equation (7)) is therefore written as a residual:

1
ṗ −∇ 2 p +b∇ ·v s = r (33)
M 
h
r −  ·∇r = 0 (34)
2

Equation (34) is the finite-increment form of the continuity equation. h = {h x h y h z }T is a vector of


 We use the following vector
characteristic element lengths in the corresponding directions in space.
instead: h = h e {1 1 1} , where h e is a homogenized characteristic element length. Substituting
T

Equation(33) into Equation (34) leads to:


 
1 h 1
ṗ −∇ p +b∇ ·v s −  ·∇
2
ṗ −∇ p +b∇ ·v s = 0
2
(35)
M  2 M 
The third spatial derivative of the pressure can be dropped from the equation. To avoid having to
project ∇ ·v s onto the nodes in order to compute its gradient, we approximate its contribution by
 (12), which is valid in one dimension. The contribution of the stabilization to the
using Equation
weak form is:

ch
B(q; p)stab = − q  ·∇ ṗ d (36)
e 2

The validity of this approximation will be confirmed through numerical results in Section 4.2.1.

3.3.1. Stability analysis of the FIC-stabilization. In one dimension, the FIC-stabilization leads to
a matrix that is added to the mass matrix as follows:
   
he 2 1 ch e 1 −1 1
Ce =c − (37)
 6 1 2 2 2 −1 1

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1214 M. PREISIG AND J. H. PRÉVOST

Note that the FIC-stabilization matrix is non-symmetric. The coefficients of Equation (27) become

5ch e 
A= −
12
t h e
2ch e 2
B= +
3
t h e
ch e 
C =− −
12
t h e
For non-oscillatory solutions, all roots of the equation have to be positive. This is true for all
t>0.

4. NUMERICAL RESULTS

In our implementation of coupled poromechanics equations (Equations (8), (9) and (10)), we solve
the resulting non-symmetric finite element algebraic equations by using a backward Euler time
integration (
= 1) and a non-symmetric direct solver [25].
In this section, we assume that the bulk modulus of the matrix is much smaller than the bulk
modulus of the grains (K K s ). In that case we have b = 1. The effect of gravity is neglected
(g = 0).

4.1. One-dimensional consolidation problem
In the first test problem, a layer of a poroelastic medium between two permeable, parallel and
infinite plates is analyzed. The distance between the two planes is 2L. Owing to symmetry, only
half of the width of the layer is modeled. The boundary condition at x = L is p = 0, the initial
pressure everywhere except at x = L is equal to p0 . The one-dimensional problem, from which
the solid velocity has been eliminated (Equation (13)), has an analytical solution, which is given
by Carslaw and Jaeger [26]:
4 p0 
∞ (−1)n 2n +1
e−bn T cos bn x
2
p(x, T ) = with bn = (38)
n=0 2n +1 2L

where T = tcf /L 2 .
Figure 1 shows the pressure after one time step near the draining boundary, computed analytically,
and using linear finite elements with and without FPL-stabilization. The results show that stable
solutions can either be obtained using a very fine mesh (tcf /h 2 = 1) or by using the stabilized
formulation. This is confirmed by an analysis with a single, extremely small time step at T =
tcf /L 2 = 10−10 . The pressure at the first node adjacent to the draining boundary is given in
Table I for unstabilized, FPL and FIC schemes. At later time the solution does not degrade
noticeably, as can be seen in Figure 2 for both FPL and FIC schemes. This is in contrast to the
residual-based method presented in Wan et al. [9], where the stabilization introduces a significant
error unless a modified formulation requiring a considerable computational effort is adopted.
Slightly better accuracy is observed for the FPL-stabilization scheme.

4.2. Two-dimensional problems


The same stabilization procedure has been implemented in two dimensions using bilinear four-node
finite elements with 2-by-2 Gauss quadrature. Some results obtained with triangular elements are
also provided.
The characteristic element length h e is taken to be the length of the element in direction of flow
(side length of element in direction perpendicular to the draining surface).

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1215

1.6
1.4
1.2
1

0
p/p
0.8 unstabilized
stabilized
0.6
Δ t/(h2/cf)=10–2
0.4
Δ t/(h2/cf)=1
0.2
analytical
0
0.95 0.96 0.97 0.98 0.99 1
x/L

Figure 1. One-dimensional consolidation: non-dimensionalized pressure after a


single time step at T = tcf /L 2 = 10−6 . FPL-stabilization.

Table I. One-dimensional consolidation: non-dimensionalized pressure at


x/L = 0.99 after single time step of tcf /h 2 = 10−6 at T = tcf /L 2 = 10−10 .
Method p(x/L = 0.99)/ p0
Exact 1
Unstabilized 1.95
FPL 1.0000000
FIC 0.997

Figure 2. One-dimensional consolidation: non-dimensionalized pressure after


10 000 time steps at T = tcf /L 2 = 10−3 (left) and T = tcf /L 2 = 10−1 (right).

The following problems were analyzed assuming the fluid to be incompressible (1/M = 0, b = 1).
The diffusivity coefficient therefore becomes:
 
4
cf =  K + (39)
3

4.2.1. Mandel problem. A layer of width 2L is compressed vertically. The top and bottom bound-
aries are fully impermeable, whereas the lateral boundaries drain ( p = 0 imposed at x = ±L). The
distributed load applied to the top surface is q = −2. Material parameters are E = 104 , = 0 and
 = 10−4 . The fluid is incompressible ( u = 0.5). Thus, we have c = 1/E and cf = 1. The initial
pressure due to the load q is p0 = − 13 (1+ u )q = 1.
Symmetry allows modeling of only half of the width (see Figure 3). The top surface is rigid,
all its nodes therefore have the same vertical displacement. The horizontal displacements are
unconstrained on the top, right and bottom boundaries.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1216 M. PREISIG AND J. H. PRÉVOST

Figure 3. Mandel problem.

Figure 4. Mandel problem: pressure along horizontal axis using FPL- and FIC-stabilizations. Results are
shown for three different non-dimensionalized time step sizes.

An analytical solution, developed by Mandel [27], is given in [23]:

p(x, T ) ∞ cos(
x/L)−cos(
)
n n
=2 sin(
n ) exp(−
2n T ) (40)
p0

n=1 n −sin(
n ) cos(
n )

where T = tcf /L 2 is the dimensionless time and


n is determined from the equation:

tan
n 1−
= (41)

n 1− u

The result is purely one-dimensional; the pressure does not vary along vertical lines. Figure 4
shows the horizontal pressure profile at two different times, for three different non-dimensional
time step sizes. Note that for tcf /h 2 = 1, FPL-stabilization is switched off. Meshes with 32 and
100 four-noded elements along the horizontal axis are used. The results using the FPL- or the
FIC-stabilization scheme do not differ much, with the exception of a slightly better accuracy of
FPL at small t. Results obtained with triangular elements are almost identical and are therefore
not shown here.
Figure 5 shows the result of a convergence analysis for the FPL-stabilization. The numerical
results are compared with analytical results at T = 10−6 . It can be seen that for small time step
lengths and for sufficiently fine meshes, quadratic convergence is obtained.

4.2.2. Cryer problem (axially loaded sphere). A sphere of radius a is loaded concentrically with
an evenly distributed surface load p0 (Figure 6). The results are compared with an analytical
solution by Cryer [28], which gives the pressure at the center of the sphere

p(0, T )  ∞ −8
+2(4
−s )/ cos s
n n
= exp(−sn T ) (42)
p0 n=1 sn −12
+16
2

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1217

Figure 5. Mandel problem: h-convergence of the pressure error (results shown


after first step). FPL-stabilization.

Figure 6. Axially loaded sphere (coarsest mesh shown).

where the dimensionless time is defined as T = cf t/a 2 and −sn are the complex roots of the
transcendental equation:
√ √ √
(s +4
) sinh s −4
s cosh s = 0 (43)
The dimensionless coefficient
is defined by
 1−2

= = (44)
+2 2(1− )
The sphere is modeled by discretizing a quarter of a circle and using axial symmetry along x = 0
and plane symmetry along y = 0. The properties of the poroelastic material are: Young’s modulus
E = 103 , solid Poisson’s ratio = 0 (thus shear modulus  = E/2) and mobility  = 10−3 .
Figure 7 shows the pressure along a radial line at time T = 5×100−6 after one time step, obtained
using 4-node elements with FPL- and FIC-stabilizations. In the analysis, four meshes with 61, 379,
2611 and 16 651 nodes are used. A result using triangular elements with FPL-stabilization is also
included and shows almost an identical pressure distribution as for 4-noded elements. Refining
the mesh results in better approximation of the analytical result at the center of the sphere. No
oscillations are visible for neither stabilization methods. A view of the pressure fields together
with the discretizations for the 379-node mesh is shown in Figure 8.

4.2.3. Load on infinite halfspace. The analytical solution of a load on the surface of an infinite
halfspace is available in McNamee and Gibson [29]. The pore pressure is given by
 
*S
p = 2 −e (45)
*z

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1218 M. PREISIG AND J. H. PRÉVOST

1.4 1.4

1.2 1.2

1 1

0.8 –5 0.8
Δ t =9.9 10
p/p0

0
p/p
nd

0.6 Δ t =1.1 10–2 0.6


nd
Δ t =4.1 10–2
nd
0.4 0.4
Δ tnd=8.9 10–2
–2
0.2 Δ t =1.1 10 (triangles) 0.2
nd
analytical solution at r = 0
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r/a r/a

Figure 7. Axially loaded sphere: pressure along a radial line at T = tcf /a 2 = 5×10−6 . Left: FPL-stabi-
lization, right: FIC-stabilization. The non-dimensional time step size is defined as tnd = tcf /h 2 .

1 1
1

0.8 0.8
0.8

0.6 0.6
0.6

0.4 0.4
0.4

0.2 0.2 0.2

0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Figure 8. Axially loaded sphere: pressure on mesh using 4-node (left) and triangular
finite elements (right). FPL-stabilization.

where  is a function of Poisson’s ratio :  = (1− )/(1−2 ),

 ∞ x 
*S 
2 = K ,  e−z
*z 2−1 0 r
  
√ −1 −(2−1)/2 2 T −1 √
× 1+erf( T )+ e erfc  T d (46)
 

and
   
1 ∞ x z− z

2 e = K ,  e erfc √ − T
2−1 0 r 2 T
 
−1 −(2−1)/2 2 T +(−1/)z −1 √ z
+ e erfc  T+ √ d (47)
  2 T

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1219

Figure 9. Load on infinite halfspace: finite element mesh.

The dimensionless time is defined as T = cf t/a 2 . The term K represents the loading

  ⎪ 2
x ⎨ cos (x) sin  plane strain
K ,  =  (48)
r ⎪
⎩ J (r )J ()
0 1 axisymmetric

where J I are Bessel functions of order I .


The analyzed geometry is illustrated in Figure 9. The height and width of the model are L = 5a.
On the top surface, zero pressure is imposed, whereas on all other boundaries the motion of the
solid matrix in perpendicular direction is set to zero. A surface load q is applied as indicated over
the length a. In the subsequent analyses the solution is captured at T = 0.8×10−4 , with both the
FPL- and the FIC-stabilization schemes. Four meshes with 331, 1801, 10 981 and 77 281 nodes
are analyzed in plane strain (Figures 10 and 11 left) as well as axisymmetric form (Figures 10
and 11 right). All analyses show very good agreement with the analytical solution for sufficiently
fine meshes. A small deviation close to the surface is probably due to the truncation of the infinite
halfspace at L. Both stabilization schemes perform well, no clear distinction as to which method
is more accurate can be made.

4.3. Bearing capacity of strip and circular footings


Special attention is required in nonlinear analyses involving elastic–plastic constitutive material
behavior. For many materials, plastic deformation occurs at constant volume. This constraint can
lead to locking even if the solid matrix and the pore fluid both are compressible. For quadrilateral
finite elements, this can be overcome by using the Bbar approach [5]. Commend et al. [30] showed
that stabilization is an efficient way of remediating locking for the dry, unsaturated case, for
low-order quadrilateral as well as triangular finite elements. This approach, however, requires a
mixed formulation for the geomechanics equations. In the following analysis, we show that the
proposed FPL- and FIC-stabilizations effectively stabilize pore pressures in a coupled bearing
capacity analysis at very high loading rate, a situation that is sometimes referred to as ‘undrained’.
In the analyses, the Bbar formulation is used to relate strains to displacements.
For the analysis, the same geometry as shown in Figure 9 is used. The footing is assumed
flexible in order to avoid stress concentrations around the corner of the footing. The number of
nodes has been reduced from 10 981 to 5025 by moving the right and lower boundaries closer to

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1220 M. PREISIG AND J. H. PRÉVOST

0 0

0.2 Δ tnd = 0.001 0.2


Δ tnd = 0.0058
0.4 Δ tnd = 0.036 0.4
Δ tnd = 0.26
z/a

z/a
analytical sol.
0.6 0.6

0.8 0.8

1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
p/q p/q

Figure 10. Load on infinite halfspace, FPL-stabilization: pressure along a vertical


profile at x = 0, T = tcf /a 2 = 0.8×10−4 . Left: strip load, right: circular load. The
non-dimensional time step size is defined as tnd = tcf /h 2 .

0 0

0.2 Δ tnd = 0.001 0.2


Δ tnd = 0.0058
0.4 Δ tnd = 0.036 0.4
Δ tnd = 0.26
z/a

z/a

0.6 analytical sol. 0.6

0.8 0.8

1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
p/q p/q

Figure 11. Load on infinite halfspace, FIC-stabilization: pressure along a vertical


profile at x = 0, T = tcf /a 2 = 0.8×10−4 . Left: strip load, right: circular load. The
non-dimensional time step size is defined as tnd = tcf /h 2 .

the footing. The length L in these analyses is equal to 5. The problem is analyzed both in plane
strain (strip footing) and in axisymmetry (circular footing). The elastic limit qe and the ultimate
bearing capacity qult for the strip footing are given by Prandtl [31] as a function of the cohesion c:

qe = c (49)
qult = (2+ )c (50)
The elastic limit and the bearing capacity for the circular footing are given by Eason et al. [32] for
a Tresca yield criterion. Using a Von Mises criterion, fitted to the apexes of the Tresca criterion,
gives an upper bound to the Tresca bearing capacity:
qe = 3.46c (51)
qult = 5.24c (52)

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1221

6 6

5 5

4 4

load q

load q
3 3
strip load circular load
2 2
FPL
1 FIC 1 FPL
analytical analytical
0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3
vertical displacement vertical displacement

Figure 12. Load–displacement curves for strip and circular footing.

0
FPL
FIC
–0.1 no stab.
depth

–0.2

–0.3

–0.4
–2 0 2 4 6 8
pressure

Figure 13. Pore fluid pressure along the axis of symmetry below the strip footing at T = 2×10−9 .

Young’s modulus E = 100, Poisson’s ratio = 0.25 and mobility k/f = 3.2×10−10 were selected
in the analyses, resulting in a diffusivity coefficient cf = 4×10−8 . The load corresponding to the
elastic limit qe is applied in a single time step of non-dimensional size tcf / h 2 = 10−6 . Then the
load is increased in steps of 0.005qe until the ultimate load qult , where failure occurs. Figure 12
shows the load–displacement curves for the strip and the circular footing. Both results are free
of oscillations. The last converged results before failure occurs are obtained at 1.007qult for the
strip footing (FPL and FIC results are almost identical) and at 0.999qult for the circular footing.
Figure 13 shows the pore fluid pressure along the vertical axis underneath the footing. The
oscillations in the case where no stabilization is used are effectively remediated by FPL- and
FIC-stabilizations.

5. CONCLUSIONS

In this paper, we investigated the problem of pressure oscillations at small time steps when
solving the coupled poromechanics equations. An overview of existing methods is presented. FPL-
stabilization, a stabilization derived in a variational multiscale framework with bubble functions
to resolve the fine scale, and a stabilization derived using FIC are analyzed in detail.
In one dimension, the analysis shows that the stabilization parameter derived using two types
of bubble functions does not give stable solutions for small time step sizes. The same analysis
for the FPL-stabilization scheme provides a stabilization parameter that leads to unconditionally
stable solutions. A procedure is presented in which the FPL-stabilization is activated only for

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1222 M. PREISIG AND J. H. PRÉVOST

small reduced time step sizes, this in order to avoid negative stabilization parameters. The stability
analysis shows that FIC leads to unconditionally stable solutions independent of the time step size.
FIC-stabilization introduces a non-symmetric stabilization matrix, whereas the matrix resulting
from FPL is symmetric.
Numerical examples in one and two dimensions assess the accuracy of the FPL- and FIC-
stabilizations and show that both methods effectively eliminate all oscillations in the pressure field.
Both methods work for 4-node as well as 3-node triangular elements (results are not shown for
FIC). Slightly better accuracy is observed for the FPL method. Both stabilization methods, in
conjunction with the Bbar formulation for quadrilateral elements, are also applicable to problems
with nonlinear material behavior. Preliminary results show that the stability in two dimensions
extends to brick elements in three dimensions, and we think it is safe to assume that this also holds
for tetrahedra.

APPENDIX A: STABILIZATION USING BUBBLE FUNCTIONS

We start by writing the weak form of Equation (13):


 
cq ṗ d+ ∇q∇ p d = 0
 
Now we split the pressure test and trial functions p and q into a coarse and a fine scale part:
p = p + p  , q = q +q  . The weak form can then be divided into an equation for the coarse scales
(•) and one for the fine scales (•) .
The coarse scale
   
cq ṗ d+ ∇q∇ p d+ cq ṗ  d+ ∇q∇ p  d = 0 (A1)
   
and the fine scale
  
cq  ṗ  d+ ∇q  ∇ p  d = − q r (x) d (A2)
  
where r (x) is the coarse-scale residual:
r (x) = c ṗ −∇ 2 p (A3)
From the strong form of Equation (A2), we see that
c ṗ  −∇ 2 p  = −r (x)
The analytical solution of this equation can be expressed with a Green’s function:

p  (x) = − G(x, y)r(y) dx +I.C.

In the following derivations, we consider only the initial time step. Initial conditions can therefore
be dropped.
We approximate the solution with a single bubble function:
p  (x) = be (x) , q  (x) = be (x)
Substituting these functions into Equation (A2) yields
  
e ˙
cb (x)b (x) d+ ∇b (x)∇b (x) d = − be (x)r (x) d
e e e
  
Using arbitrariness of 
  
c(be (x))2 ˙ dx + ∇be (x)·∇be (x) dx = − be (x)r (x) dx
  

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1223

and discretizing in time ( 1 =


t ˙ 1 at t = t1 )
   
˙ 1 c(be (x))2 +
t∇be (x)·∇be (x) dx = − be (x)r (x) dx
 

Let De be a scalar constant for each element:



D e = [c(be (x))2 +
t∇be (x)·∇be (x)] dx


then we can solve for ˙ 1 :



˙ 1 = − be (x)r(x) dx(D e )−1


and the solution writes



p1 = −
tbe (x) be (y)r (y) dy(D e )−1 ≈ − (x)r (x) (A4)


where

(x) =
tb (x) e
be (y) dy(D e )−1


Remark 3
The approximation in Equation (A4) assumes that the residual r (x) is constant within an element
Nakshatrala [19].

The fine-scale solution remains to be introduced into the coarse-scale weak form (Equation (A1)):
 
cq ṗ d+ ∇q∇ p d+ B(q; p  )stab = 0
 

where
 
 
B(q; p )stab = cq ṗ d+ c∇q∇ p  d
 

Substituting p  and ṗ  from Equation (A4):


 
B(q; p)stab = − cq (x)ṙ (x) d− ∇q∇( (x)r(x)) d
 

Now the coarse-scale residual (Equation (A3)) is put back into the above equation. Since we
eliminated all fine-scale variables, we can drop the bar for simplicity:
 
B(q; p)stab = − c q (x) p̈ d+ cq (x)∇ 2 ṗ d
2
 
   
− c∇q∇( (x) ṗ) d+ 2 ∇q∇ (x)∇ 2 p d
 
In the above equation, we can drop second-order time derivatives. Integration by parts is performed
on the second term (the boundary terms vanish because the bubble function is zero on element
boundaries). Combining terms we obtain:
 
B(q; p)stab = −2 c∇q (x)∇ ṗ d− cq∇ (x)∇ ṗ d
 
 
− c∇q∇ (x) ṗ d+ 2 ∇q∇( (x)∇ 2 p) d
 

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
1224 M. PREISIG AND J. H. PRÉVOST

Expanding the fourth term yields the following:


 
2 ∇q∇( (x)∇ 2 p) d = 2 ∇q((∇ (x)∇ 2 p + (x)∇ 3 p)) d
 

The derivative on the pressure can be reduced to second order by integration by parts. Further
integration by parts would require the gradient of (x) to be zero on element boundaries. But since
we use linear finite elements, we make the argument that the second-order derivatives on ṗ and q
vanish and therefore we drop the entire term. Finally we get
 
B(q; p)stab = −2 c∇q (x)∇ ṗ d− c∇q∇ (x) ṗ d
 

− cq∇ (x)∇ ṗ d (A5)


ACKNOWLEDGEMENTS
We gratefully acknowledge useful discussions with A. Truty and Th. Zimmermann. This work is supported
by Swiss National Science Foundation under contract grant number PBELP2-127849 and Carbon Miti-
gation Initiative (http://cmi.princeton.edu) sponsored by BP and Ford.

REFERENCES
1. Babuška I. Error bounds for finite-element method. Numerische Mathematik 1971; 16:322–333. DOI: 10.1007/
BF02165003.
2. Brezzi F. On the existence, uniqueness and approximation of saddle-point problems arising from Lagrangian
multipliers. Revue Française d’Automatique Informatique et Recherche Operationnelle 1974; 8:129–151.
3. Gresho PM, Lee RL. Don’t suppress the wiggles—they’re telling you something! Computers and Fluids 1981;
9(2):223–253.
4. Vermeer PA, Verruijt A. An accuracy condition for consolidation by finite elements. International Journal for
Numerical and Analytical Methods in Geomechanics 1981; 5:1–14.
5. Hughes TJR. The Finite Element Method. Dover Publications: Mineola, NY, 2000.
6. Phillips PJ, Wheeler MF. A coupling of mixed and discontinuous Galerkin finite-element methods for poroelasticity.
Computational Geosciences 2008; 12(4):417–435. DOI: 10.1007/s10596-008-9082-1.
7. Phillips PJ, Wheeler MF. Overcoming the problem of locking in linear elasticity and poroelasticity: an heuristic
approach. Computational Geosciences 2009; 13(1):5–12. DOI: 10.1007/s10596-008-9114-x.
8. Liu R, Wheeler MF, Dawson CN, Dean RH. On a coupled discontinuous/continuous Galerkin framework and an
adaptive penalty scheme for poroelasticity problems. Computer Methods in Applied Mechanics and Engineering
2009; 198(41–44):3499–3510. DOI: 10.1016/j.cma.2009.07.005.
9. Wan J, Durlofsky LJ, Hughes TJR, Aziz K. Stabilized finite element methods for coupled geomechanics-reservoir
flow simulations. SPE Reservoir Simulation Symposium, Houston, TX, 2003; number SPE 79694.
10. Mira P, Pastor M, Li T, Liu X. A new stabilized enhanced strain element with equal order of interpolation for soil
consolidation problems. Computer Methods in Applied Mechanics and Engineering 2003; 192(37–38):4257–4277.
DOI: 10.1016/S0045-7825(03)00416-X.
11. White JA, Borja RI. Stabilized low-order finite elements for coupled solid-deformation/fluid-diffusion and their
application to fault zone transients. Computer Methods in Applied Mechanics and Engineering 2008; 197(49–
50):4353–4366. DOI: 10.1016/j.cma.2008.05.015.
12. Tchonkova M, Peters J, Sture S. A new mixed finite element method for poro-elasticity. International Journal
for Numerical and Analytical Methods in Geomechanics 2008; 32(6):579–606. DOI: 10.1002/nag.630.
13. Zienkiewicz OC, Taylor RL, Nithiarasu P. The Finite Element Method for Fluid Dynamics (6th edn). Elsevier
Butterworth-Heinemann: Oxford, England, 2005.
14. Pastor M, Li T, Liu X, Quecedo M, Zienkiewicz OC. Fractional step algorithm allowing equal order of interpolation
for coupled analysis of saturated soil problems. Mechanics of Cohesive-Frictional Materials 2000; 5(7):511–534.
DOI: 10.1002/1099-1484(200010) 5:7<511::AID-CFM87>3.0.CO;2-S.
15. Salomoni VA, Schrefler BA. A CBS-type stabilizing algorithm for the consolidation of saturated porous media.
International Journal for Numerical Methods in Engineering 2005; 63(4):502–527. DOI: 10.1002/nme.1275.
16. Truty A, Zimmermann Th. Stabilized mixed finite element formulations for materially nonlinear partially
saturated media. Computer Methods in Applied Mechanics and Engineering 2006; 195(13–16):1517–1546. DOI:
10.1016/j.cma.2005.05.044.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag
STABILIZATION PROCEDURES IN COUPLED POROMECHANICS 1225

17. Brezzi F, Pitkäranta J. On the stabilization of finite element approximations of the Stokes equations. In Efficient
Solutions of Elliptic Systems, Hackbusch W (ed.). Notes on Numerical Fluid Mechanics, vol. 10. Vieweg:
Braunschweig, Germany, January 1984; 11–19.
18. Aguilar G, Gaspar FJ, Lisbona F, Rodrigo C. Numerical stabilization of Biot’s consolidation model by a
perturbation on the flow equation. International Journal for Numerical Methods in Engineering 2008; 75(11):
1282–1300. DOI: 10.1002/nme.2295.
19. Nakshatrala KB, Masud A, Hjelmstad KD. On finite element formulations for nearly incompressible linear
elasticity. Computational Mechanics 2008; 41(4):547–561. DOI: 10.1007/s00466-007-0212-8.
20. Hughes TJR. Multiscale phenomena: Green’s functions, the Dirichlet-to-Neumann formulation, subgrid scale
models, bubbles and the origins of stabilized methods. Computer Methods in Applied Mechanics and Engineering
1995; 127(1–4):387–401.
21. Oñate E. Derivation of stabilized equations for numerical solution of advective–diffusive transport and fluid flow
problems. Computer Methods in Applied Mechanics and Engineering 1998; 151(1–2):233–265.
22. Oñate E, Rojek J, Taylor RL, Zienkiewicz OC. Finite calculus formulation for incompressible solids using linear
triangles and tetrahedra. International Journal for Numerical Methods in Engineering 2004; 59:1473–1500. DOI:
10.1002/nme.922.
23. Coussy O. Poromechanics. Wiley: Chichester, England, 2004.
24. Franca LP, Russo A. Mass lumping emanating from residual-free bubbles. Computer Methods in Applied Mechanics
and Engineering 1997; 142(3–4):353–360.
25. Prévost JH. DYNAFLOW: A Nonlinear Transient Finite Element Analysis Program. Department of Civil and
Environmental Engineering, Princeton University, Princeton, NJ, 1981; last update 2009. Available from: http://
www.princeton.edu/∼dynaflow/.
26. Carslaw HS, Jaeger JC. Conduction of Heat in Solids (2nd edn). Clarendon Press: Oxford, England, 1959.
27. Mandel J. Consolidation des sols (étude mathématique). Geotechnique 1953; 3(7):287–299.
28. Cryer CW. A comparison of the three dimensional consolidation theories of Biot and Terzaghi. Quarterly Journal
of Mechanics and Applied Mathematics 1963; 16(4):401–412.
29. McNamee J, Gibson RE. Plane strain and axially symmetric problems of the consolidation of a semi-infinite
clay stratum. Quarterly Journal of Mechanics and Applied Mathematics 1960; 13(2):210–227.
30. Commend S, Truty A, Zimmermann Th. Stabilized finite elements applied to elastoplasticity: I. Mixed
displacement–pressure formulation. Computer Methods in Applied Mechanics and Engineering 2004; 193 (33–35):
3559–3586.
31. Prandtl L. Über die Härte plastischer Körper. Nachrichten von der Königlichen Gesellschaft der Wissenschaften
zu Göttingen 1920; 74–85.
32. Eason G, Shield RT. The plastic indentation of a semi-infinite solid by a perfectly rough circular punch. Zeitschrift
für Angewandte Mathematik und Physik 1960; 11(1):33–43.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2011; 35:1207–1225
DOI: 10.1002/nag

Você também pode gostar