Professional Documents
Culture Documents
Finite Element Solution Via Standard Galerkin's Method
Finite Element Solution Via Standard Galerkin's Method
Introduction
The exact solutions to the governing equations derived in the last chapter are generally not available. Hence, numerical solutions are introduced to obtain approximate
solutions. The finite element method is a valuable approximation tool for the solution
of Partial Dierential Equations (PDE) when the analytical solutions are dicult or
impossible to obtain due to complicated geometry or boundary conditions. Such a
method uses a spatial discretization and a weighted residual formulation to transform the governing PDE (strong form) into an integral equation (weak form) that
upon variational treatment yields to the solution of a system of matrix equations. It
turns out that this mathematical/numerical process yields an approximate solution
to the original boundary value problem (Zienkiewicz[105]; Cook, 1981[106]; Hughes,
1987[107]; Zienkiewicz and Taylor, 1991[108]). In this chapter, Galerkins method
of approximation is used to discretize the resulting governing equations in order to
demonstrate the applicability of the proposed model to address sand production in
real boundary value problem context. Considering the complexity of this problem,
it is of interest (and yet without losing generality) to first examine two-dimensional
conditions and restrict the problem to pure erosional aspects, i.e. the deformation
76
c
c
c
+ + qx
+ qy
= 0.
t
t
x
y
t
(4.1)
(1 )c qx2 + qy2 = 0.
t
(4.2)
qx qy
+
= 0.
x
y
(4.3)
p
= 0.
x
p
(1 )2 qy + k0y
= 0.
y
(1 )2 qx + k0x 3
(4.4)
(4.5)
where
= (1 c)f + cs
(4.6)
There are five governing equations in total with five unknowns which are c, ,
p, qx , and qy referring to fluidized solid concentration, porosity, fluid pressure, fluid
flux in x direction, and fluid flux in y direction respectively.
4.2
Semi-discrete Galerkin approximation is used in order to discretize main field variables c(x, t), (x, t), p(x, t), qx (x, t), and qy (x, t), i.e.
77
c(x, t) = Nj (x)
cj (t)
(4.7)
k (t)
(x, t) = Nk (x)
pl (t)
p(x, t) = Nl (x)
qxm (t)
qx (x, t) = Nm (x)
qym (t)
qy (x, t) = Nm (x)
p , pp , qxp , qyp , and Np are respectively fluidized solid concentration, poroswhere cp ,
ity, fluid pressure, fluid flux in x direction, fluid flux in y direction, and interpolation
function at node p, for p = 1 to nh , the total number of nodes. It is again recalled
that Einstein index notation is used with repeated indices implying summation and
the index p is dummy.
Applying Galerkins method of weighted residual (with weighting functions equal
to interpolation functions) over the entire domain to each one of Equations (4.14.5) in turn together with discretizing time derivatives by standard finite dierence
formula and also linearizing time variables, a system of five non-linear equations is
obtained with its generic form
Wn+1 (cn+1 , n+1 , pn+1 , qxn+1 , qyn+1 ) = Hn (cn , n , pn+1 , qxn , qyn )
(4.8)
in which W and H are functionals which originate from Equations (4.1-4.5) and
subscripts n and n + 1 refer to time stations tn and tn+1 respectively. Vectors c, ,
j , pj , qxj , and qyj
p, qx , and qy contain nodal values of respective unknowns cj ,
with j = 1 to nh .
78
Wn+1 Wn
W
=
t
t
(4.9)
k
k+1 k+1
k+1
k+1
k+1
k
k
k
k
k
Wn+1
(cn+1 , k+1
n+1 , pn+1 , qxn+1 , qyn+1 ) = Hn+1 (cn+1 , n+1 , pn+1 , qxn+1 , qyn+1 )
k
k
W
W
k
: cn+1 +
: kn+1
+
c n+1
n+1
k
k
W
W
k
+
: pn+1 +
: qkxn+1
p n+1
qx n+1
k
W
+
: qkyn+1 + 0(c2 , 2 , ...)
qy n+1
= Hn (cn , n , ...)
Hence, the new estimates at iteration k + 1 and time step n + 1
79
(4.10)
k
k
ck+1
n+1 = cn+1 + cn+1
k
k
k+1
n+1 = n+1 + n+1
k
k
pk+1
n+1 = pn+1 + pn+1
(4.11)
k
k
qk+1
xn+1 = qxn+1 + qxn+1
k
k
qk+1
yn+1 = qyn+1 + qyn+1
ckn+1
kn+1
k
k
k
k
k
Jn+1 (c) Jn+1 () Jn+1 (p) Jn+1 (qx ) Jn+1 (qy ) pkn+1
qkxn+1
qk
yn+1
= H(cn , n , ...) W(ckn+1 , kn+1 , ...)
(4.12)
where the Right Hand Side (RHS) is known, and the Left Hand Side (LHS) involves
the evaluation of a Jacobian J such as
Jkn+1
W
=
c, , p, qx , qy
(4.13)
n+1
Successive iterates are performed until the convergence criteria is satisfied, i.e.
80
k+1
where
k+1
pn+1 pkn+1 <
1,
2,
3,
4,
and
1;
3;
k+1
k+1
(4.14)
2
4;
k+1
0
[A1]44 [A2]44
[B1]
0
44 [B2]44
0.
0
0
[D1]
44 [D2]44 [D3]44
k
[A3]44 [A4]44
[B3]44 [B4]44
[C1]44 [C2]44
[D4]44
0
0
[E4]44
n+1(2020)
k
c
n+1
n+1
k
pn+1
qk
xn+1
k
qyn+1
201
(4.15)
[A1]ij =
i
Ni Nj Nk h n+1
n d
2k + (1 2)
k
t
Z
n+1
Nj
n
+ (1 )
qmx
Nm 2 qmx
+ Ni
d
x
Z
n+1
Nj
n
Nm 2 qmy
d
+ (1 )
qmy
+ Ni
y
81
(4.16)
Z
Ni Nj Nk
Ni Nj
n+1
n
[A2]ij =
2
ck + (1 2) ck d
d
t
t
Z
Nm
[A3]ij =
Ni Nj 2 cn+1
cnm d
m + (1 )
x
Z
Nm
n
[A4]ij =
Ni Nj 2 cn+1
+
(1
)
c
m
m d
y
Z
h
i
n+1 + (1 )
n ) Ni Nj d
[B1]ij = Q 1 Nk (
k
k
Z
[B2]ij =
[B3]ij =
Nj Ni
d +
t
QNi (
cn+1
+ (1 )
cnk )Nk Nj d
k
h
i
n+1
n
1 Nk (k + (1 )k )
cn+1
+ (1 )
cnp Ni Np
p
(4.17)
(4.18)
(4.19)
(4.20)
(4.21)
(4.22)
n+1
n
2 [
qxm
+ (1 )
qxm
] Nm Nj
d
2Q
h
i
n+1
n
cn+1
+ (1 )
cnp Ni Np (4.23)
[B4]ij = 1 Nk (k + (1 )k )
p
n+1
n
qmy
Nm Nj
2
qmy + (1 )
d
2Q
Z
Nj
(4.24)
Ni d
[C1]ij =
x
Z
Nj
[C2]ij =
Ni d
(4.25)
y
Z
h
i
n+1
n 2 n+1
n
qxm
qxm + (1 )
Nm Ni d
[D1]ij = (s f )Nj 1 Nk (k + (1 )k )
(4.26)
[D2]ij =
n+1
2 f + Np
cnp (s f )
cp + (1 )
i
h
n+1
n
n+1 + (1 )
n ) 1
qxm
+ (1 )
qxm
(4.27)
Nj Nm Ni
Nk (
k
k
h
i2
n+1 + (1 )
n ) Nj Nl pn+1 + (1 )
pnl Ni d
+3k0x Nk (
k
k
l
x
82
[D3]ij =
[D4]ij =
n+1
n 3 Nj
k0x (Nk (k + (1 )k ))
Ni d
x
n+1
n+1
cnp f + Np
cnp s (4.29)
1 Np
cp + (1 )
cp + (1 )
i2
n
n+1 + (1 )
1 Nk
Nj Ni d
k
[E1]ij =
h
i2
n+1
n
n+1 + (1 )
n )
(s f )Nj 1 Nk (
+
(1
)
q
q
k
k
ym
ym Nm Ni d
(4.30)
[E2]ij =
[E4]ij =
(4.28)
n+1
2 f + Nl
cnl (s f )
cl + (1 )
i
h
n+1
n
n+1 + (1 )
n ) 1
q
+
(1
)
q
(4.31)
Nk (
ym
ym Nj Nm Ni
k
k
i2
h
n
n+1 + (1 )
n ) Nj Nl pn+1 + (1 )
p
+3k0y Nk (
l Ni d
k
k
l
y
Z
N
n+1
n
j
3
))
k0y (Nk (
Ni d
+ (1 )
(4.32)
[E3]ij =
k
k
y
n+1
n+1
cnm f + Nm
cnm s (4.33)
1 Nm
cm + (1 )
cm + (1 )
i2
n+1
n
1 Nk k + (1 )k
Nj Ni d
4.3
4.3.1
Numerical examples
One-dimensional example
c (x, t = 0) = c0 = 0.0002
(x, t = 0) = 0 = 0.25
p (x = 1, t) = 10 MPa
p (x = 0, t)= 0
y
flux qx
0.2
0.6
0.4
0.8
1.0 x (m)
1-D Mesh
Figure 4.1: Finite element mesh for one dimensional representation of wellbore
84
85
0.25
peak
0.2
0.15
2-D mesh
0.1
1-D mesh
(Vardoulakis et. al, 1996)
0.05
0
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
time (days)
1
0.9
0.8
0.7
0.6
0.5
2-D mesh
inflection point
0.4
1-D mesh
(Vardoulakis et. al, 1996)
0.3
0.2
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
time(days)
0.5
4.5
4
0.4
0.3
3
3
0.2
2.5
2
2-D mesh
0.1
1-D mesh
(Vardoulakis et. al, 1996)
1.5
1
0
0
0.2
0.4
0.6
0.8
time (days)
1.2
1.4
88
1
0.08 day
0.22 day
0.32 day
0.54 day
1 day
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0
0.2
0.4
0.6
0.8
x (m)
89
In order to further illustrate the capability of the model, one quarter of a section
of a well of inner radius r0 = 0.1 m is examined under anisotropic permeability
conditions, i.e. k0x = 1.5 k0y . The outer boundary of the reservoir extends to 1 m
at which a pressure of 10 MPa is imposed while the inner boundary at wellbore is
kept at 0 MPa so as to induce drawdown. The simulation parameters such as initial
fluidized sand concentration c0 , porosity 0 , and erosion coecient are listed in
Table 4.2.
Figure 4.6 shows the finite element mesh composed of 850 nodes and 792 bilinear
elements with each node having five degrees of freedom, i.e. c, , qx , qy , and p.
Notice the placement of finer elements near the wellbore where nonlinearities are
expected to be severe due to high erosional activity. The pressure contours at time
1.0 day are shown in Figure 4.7 where depletion is biased along the direction of
higher permeability, i.e. x-direction.
In order to appreciate the erosional process as fluid is drawn down inside the
wellbore, it is of interest to look at the evolution of both fluidized sand concentration
and porosity with time. Initially at t = 0.35 day, there is a small zone of high
sand concentration (c 0.14) spreading laterally, see Figure 4.8. This represents
approximately a twenty eight fold increase in fluidized sand concentration from its
90
850 nodes
792 elements
0.9
0.8
c, p and calculated at
each node
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
Figure 4.6: Finite element mesh for a two dimensional representation of the wellbore
1.0
9.38E+06
8.75E+06
8.13E+06
7.50E+06
6.88E+06
6.25E+06
5.63E+06
5.00E+06
4.38E+06
3.75E+06
3.13E+06
2.50E+06
1.88E+06
1.25E+06
6.25E+05
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
x( m)
1
1.44E-01
1.35E-01
1.25E-01
1.15E-01
1.06E-01
9.62E-02
8.67E-02
7.71E-02
6.75E-02
5.80E-02
4.84E-02
3.88E-02
2.93E-02
1.97E-02
1.01E-02
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
92
1
9.00E-01
7.50E-01
7.00E-01
6.50E-01
6.00E-01
5.50E-01
5.00E-01
4.50E-01
4.00E-01
3.50E-01
3.00E-01
2.50E-01
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
1
9.00E-01
7.50E-01
7.00E-01
6.50E-01
6.00E-01
5.50E-01
5.00E-01
4.50E-01
4.00E-01
3.50E-01
3.00E-01
2.50E-01
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
1
9.00E-01
7.50E-01
7.00E-01
6.50E-01
6.00E-01
5.50E-01
5.00E-01
4.50E-01
4.00E-01
3.50E-01
3.00E-01
2.50E-01
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Cavity
0
0.1
0.2
0.3
0.4
0.5
x(m)
0.6
0.7
0.8
0.9
0.5
0.4
0.3
0.2
0.1
P3
P2
perforations
0
P1
0.1
0.2
0.3
0.4
x(m)
0.5
extends to 5m
0.5
0.850
0.813
0.776
0.739
0.701
0.664
0.627
0.590
0.553
0.516
0.479
0.441
0.404
0.367
0.330
0.4
0.3
0.2
0.1
0.1
0.2
0.3
x(m)
0.4
0.5
0.5
0.900
0.859
0.819
0.778
0.737
0.696
0.656
0.615
0.574
0.534
0.493
0.452
0.411
0.371
0.330
0.4
0.3
0.2
0.1
0.1
0.2
0.3
x(m)
0.4
0.5
98
0.5
0.850
0.813
0.776
0.739
0.701
0.664
0.627
0.590
0.553
0.516
0.479
0.441
0.404
0.367
0.330
0.4
0.3
0.2
0.1
0.1
0.2
0.3
x(m)
0.4
0.5
0.5
0.850
0.813
0.776
0.739
0.701
0.664
0.627
0.590
0.553
0.516
0.479
0.441
0.404
0.367
0.330
0.4
0.3
0.2
0.1
0.1
0.2
0.3
x(m)
0.4
0.5
x(m)
0.5
0.4
0.3
0.2
0.1
0.1
0.2
x(m)
0.3
0.4
0.5
101
4.4
The numerical results pertaining to one- and two-dimensional conditions show that
sand production can be captured within a continuum framework. While configurations investigated are quite simple, the numerical results give physical insights
that will aid in developing a more complete model that accounts for sand matrix
deformations.
However, it was found that numerical results were sometimes corrupted with
instabilities in the form of node-to-node oscillations or wiggles when applying severe conditions such as imposing highly anisotropic permeability conditions, and
simulating realistic perforation pressure or flux boundary conditions on the inner
wellbore. It was realized that a standard finite element discretization combined with
a Newton-Raphson method could not achieve stable solutions in trying to simulate
realistic boundary value problems. Thus, a numerical scheme must be devised to
overcome numerical instabilities associated with the computation of field variables
with strong gradients. Furthermore, from a computational eciency view point, it
is impractical to solve for five degrees of freedom (fluidized concentration c, porosity
, mixture flux qi , and pressure p) at each node in the current finite element scheme.
For example, a small domain with a radius 0.5 m was simulated due to limitations
imposed by current computer resources. It was also critical to find a good equation
solver and storage scheme for solving eciently the large sparse matrix structure
that arises from the proposed model.
102