Sunteți pe pagina 1din 4

INTRODUCTION TO FINITE DIFFERENCE METHOD

Dr. K. Rajagopal
Professor, Civil Engineering Department, IIT Madras

The governing differential equations are converted to difference form in this method. The continuous derivatives
are transformed to discrete form by using the difference functions in discrete form. This is accomplished by
using the Taylor series expansion of functions.

Taylor series expansions to determine the value of a function f(x) in the neighbourhood x+x and x-x are,
f(x+x) = f(x) + x f(x) + (x)2 f(x)/2! + (x)3 f(x)/3! + ……… (1)
f(x-x) = f(x) - x f(x) + (x)2 f(x)/2! - (x)3 f(x)/3! + ……… (2)

in which f(x) is the first derivative of the function f(x) and f(x) is the second differentiation of the function and
f (x) is the 3rd derivative of the function f(x) and so on….

Forward-difference scheme:
from Eq. 1, f(x) = [f(x+x)-f(x)]/x + E(x) (3)

Backward difference scheme:


from Eq. 2, f(x) = [f(x)-f(x-x)]/x + E(x) (4)

Central difference scheme:


by subtracting Eq. 2 from Eq. 1, f(x) = [f(x+x)- f(x-x)]/2x + E(x2) (5)

E(x) in Equations 3 and 4 is the error of the order of x. As the order of error in central difference scheme is of
the order x2, as compared to x in forward/backward difference schemes, the solution in the former scheme
converges faster than the latter as x value is decreased.

By adding Eq. 1 and 2, we can get a central difference expression for the second order differential quantity as
follows,

f(x) = [f(x+x) – 2 f(x) + f(x-x)]/x2 + E(x2) (6)

The order of error in the second difference term is also of the same order as in the first difference term.

Finite difference form of the 1-d consolidation equation

u 2
 u
 cv 2 (7)
t z
in which u is the pore pressure, cv is the coefficient of consolidation, t is the time and z is the space variable.
The above is an initial value problem in which both boundary and initial conditions are specified. For exact
solution, we need two boundary conditions on the pore pressures and also know the pore pressure values at time
t=0. You may recall that the classical Terzaghi’s solution for 1-d consolidation assumes the initial pore pressures
to be constant over the full depth, equal to the applied pressure. The boundary conditions may be the pore
pressure is zero at the ground surface (z=0) and permeable or impermeable boundary at the bottom surface
(z=h). Permeable boundary means that the pore pressure at that boundary is zero. The impermeable boundary
condition means that there is no flow across the boundary. Darcy’s equation can be used for velocity and the
flow quantity can be calculated as q=vA=kiA, where “i” is the gradient of flow. For the flow to be zero, the
gradient must be zero, i.e. the variation of pore pressure with respect to z should be zero at z=h.

The governing differential equation for consolidation can be converted to finite difference form using forward
difference scheme in time and central difference scheme in space as follows:

1/4
utzt  utz  utz z  2 utz  u zz z 
 cv   (8)
t  z 2 
The above equation is called the explicit equation because there is only one unknown term at time (t+t) which
can be expressed in terms of all other known parameters.

t  t
uz
cv t t

 u z  ( 2 ) u z z  2 utz  utz z
t
 (9)

z
The above equation can be easily solved for pore pressures at the next time step (t+t) knowing the pore
pressures at the current time “t”. The above formulation is called EXPLICIT formulation because the unknown
quantity on the left side of the equation is expressed in terms of known quantities on the right hand side. The
scheme is only conditionally stable for selected values of t and z. These values are to be selected carefully
such that the quantity in the round brackets in Eq. 9 is less than or equal to 0.50.

Solution process starts with discretising the given soil depth into certain number of divisions with a suitable z
as shown below:

1 U1=0, t>0
t  t
u 2  u 2   (u1  2 u 2  u3)
t t t t
2
  cv 2  0.5
t
3
z z
4 t  t
u 7  u 7   ( 2 u 7  2 u 6)
H t t t

6
7u 
u 6  u 7
 0  u 6  u 7
z 7 2z
7
Imaginary point
Figure 1. Schematic for finite difference analysis
In the above, the height of soil is divided into 6 divisions and hence z=H/6. The solution process is easy to start
because all pore pressures (u1…u7) are known at time t=0 from the initial conditions. The impermeable
boundary condition at the bottom of soil layer (at node 7) is imposed as explained in the figure. The scheme is
stable only for  values less than or equal to 0.50.

The finite difference scheme can be written in implicit form by writing the RHS of Eq. 7 in terms of pore
pressures at time (t+t) as shown below:

utzt  utz  ( 2
) 
cv t t t
u z  z  2 u t  t
z  u t  t
z  z 
z
The above system cannot be solved as easily as earlier. A system of simultaneous equations is obtained at each
time step which needs to be solved by matrix methods. This matrix will be a tri-diagonal with only three terms

2/4
(maximum) on any row. The solution in this case is a bit tedious but solution converges for any value of .
Hence higher values of t can be chosen to accelerate the solution process until the desired degree of
consolidation is achieved.

Two computer programs, consexp.bas and consimp.bas are available for student’s use to test the above scheme.
The executables of these programs are consexp.exe and consimp.exe. Both programs are interactive and require
minimum input from the user. At the program prompt, appropriate input data may be given. Two files are
generated, one with the degree of consolidation at various elapsed times and time factors and another with pore
pressures at various grid points.

APPLICATIONS TO DYNAMIC PROBLEMS

The dynamic equilibrium equation that combines the inertia, damping and elastic forces can be written as
follows:

  C U
MU  KU  R (10)

in which M, C and K are the mass, damping and stiffness matrices; R is the force vector and U, U  and U  are
the displacement, velocity and acceleration vectors respectively. The above equation is nothing but a second
order linear differential equation. Closed form analytical solution to the above equation can be obtained for
some simple problems. However, as the number of degrees of freedom in the system increases, deriving
analytical solution to the above equation becomes difficult. In such cases, numerical solutions are preferred.

Using the central difference scheme, acceleration and velocity at time t can be written in terms of known
displacements at time (t+t), t and (t-t) as follows,

  1  U
U t t - t - 2 U t  U t  t 
t 2
(11)
  1 U
U t t  t - U t - t 
2 t

Substituting the above two quantities in the governing differential equation (10) and simplifying, we get

 1 1  2 1 1
 2 M C  U t  t  R t - (K - 2 M)U t - ( 2 M - C)U t -t (12)
 t 2t  t t 2 t

The LHS of the equation is similar to [K]{u} and the RHS is the force vector that is a function of displacements
at older time steps, mass and damping of the system.

The solution of Ut+t is based on using the equilibrium equations at time “t”. The examination of the above
equation shows that the solution at time t requires the solution at time -t, i.e. at a time even before the start of
the system. The solution at that time can be obtained conveniently from the Taylor series expansion as follows:

  t U
2
U -t  U o - t U  (13)
o o
2
The time step t should be less than critical time step length tcr that is equal to Tn/ in which Tn is smallest
period of the assemblage. If the time step length is larger, the solution diverges from the actual solution.

Once the displacements at the current step (t+t) are determined, the accelerations and velocities may also be
determined using the standard Taylor series expressions.

Numerical Example:

Consider the response of the following undamped system subjected to initial displacement and velocity of 0.0.

3/4
2 0 U1   6  2 U 1   0 
0 1      2 4  U   10 (A)
  U 2    2   

The fundamental periods of the system in first and second mode of vibration are T 1=4.45 sec, T2=2.8 sec. Use
Central difference scheme to calculate the response of the system with t=T2/10 and 10T2.

As the displacements and velocities are given at time t=0, the corresponding accelerations can be obtained
directly from Eq. A.

2 0 U1   6  2 0  0 
0 1      2 4  0  10
  U 2      

hence, U ={0 10}T


o

if t=0.28, displacements at time t=--t ,

U 1-t  0 0 0.28 2 0  0 


 2      0.28     
U -t  0 0 2 10 0.392
The effective stiffness matrix on LHS is,

1 1 1 2 0 1 0 0  25.5 0 
[K̂]  M C  
t 2
2 t 0.28 2 0 1 2  0.28 0
 
0  0 12.8

The effective load vector on RHS is,

 0   6  2 2  2 0   U 1   1  2 0 1 0 0   U 1 
{R̂}t               0 1 2  0.28 0 0 U 

10    2 4  0.28
2 2
 0 1  U 2  t  0.28      2  t t
R̂ t
 0  45
  
2 

25.5
Ut  
0 
12.8
U t - t
10  2 21.5  0

Hence, we need to solve the recursive equation,

25.5 0 
 0 U t  t  R̂ t
 12.8

Solving the above equation recursively over the different time steps,

Time t 2t 3t 4t 5t 6t 7t 8t 9t 10t 11t 12t
Ut 0.0 0.0307 0.168 0.487 1.02 1.70 2.40 2.91 3.07 2.77 2.04 1.02
0.392 1.45 2.83 4.14
5.02 5.26 4.90 4.17 3.37 2.78 2.54 2.60

Now, consider the case with t=28 which is much larger than the critical time step length. Using exactly the
same calculation methodology, the result obtained is as follows which clearly shows the divergence of the
solution:

Time t 2t
Ut 0.0 3.03106
3.83103 -1.21107

Some EXCEL spread sheet programs are available to solve for dynamic response of single degree of freedom
systems subjected to different types of impulse loads.

4/4

S-ar putea să vă placă și