NUMERICAL METHODS AND GEOMECHANICS

Lecture #5. Introduction to Finite Difference Method

5.1 Fundamental background of FDM 5.2 Numerical formulation of FLAC 5.3 Analysis example

5.1 Fundamental background of FDM A SIMPLE MECHANICAL ANALOG

Lecture #5. Introduction to Finite Difference Method

Newtons Law of Motion

F = ma = m

du dt

, u u, u
F(t)

For a continuous body, this can be generalized as where = mass density,

i ij du = + gi x j dt

xj = coordinate vector (x,y) ij = components of the stress tensor, and gi = gravitation

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.1 Fundamental background of FDM STRESS-STRAIN EQUATIONS

Lecture #5. Introduction to Finite Difference Method

In addition to the law of motion, a continuous material must obey a constitutive relation - that is, a relation between stresses and strains. For an elastic material this is:

In general, the form is as follows: where

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC What is FLAC?

Lecture #5. Introduction to Finite Difference Method

quasi-static problems. This has advantages for problems that involve physical instability, such as collapse, as will be explained later.

FLAC & FLAC3D solves the full dynamic equations of motion even for

To model the static response of a system, a relaxation scheme is used in which damping absorbs kinetic energy. This approach can model collapse problems in a more realistic and efficient manner than other schemes, e.g., matrix-solution methods.

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.1 Fundamental background of FDM Basic Explicit Calculation Cycle For all gridpoints (nodes)
velocities

Lecture #5. Introduction to Finite Difference Method

Gauss theorem

Equilibrium Equation (Equation of Motion) i ij du = + gi x j dt For all zones (elements) Stress - Strain Relation (Constitutive Equation)

nodal forces

Fi = ij n j L

strain rates

new stresses

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.1 Fundamental background of FDM A GENERAL FINITE-DIFFERENCE FORMULA

Lecture #5. Introduction to Finite Difference Method

In the finite difference method, each derivative in the previous equations (motion & stress-strain) is replaced by an algebraic expression relating variables at specific locations in the grid. The algebraic expressions are fully explicit; all quantities on the righthand side of the expressions are known. Consequently each element (zone or gridpoint) in a FLAC grid appears to be physically isolated from its neighbors during one calculational timestep. This is the basis of the calculation cycle:

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC

Lecture #5. Introduction to Finite Difference Method

FLACs grid is internally composed of triangles. These are combined into

quadrilaterals. The scheme for deriving difference equations for a polygon is described as follows:

Overlaid Elements

Triangular element Nodal force vector with velocity vectors

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC FLAC: For all elements...

Gauss theorem,

Lecture #5. Introduction to Finite Difference Method

f n fdS = S i A xi dA

(b ) ui
b

nodal velocity S a

The outward flux of a vector field through a closed surface is equal to the volume integral of the divergence of the region inside the surface is used to derived a finite difference formula for elements of arbitrary shape. For a polygon the formula becomes

i( a ) u

nodal velocity

1 f f ni S xi A S
i u 1 (a) (b ) u u + ( i i ) n j S x j 2 A S j i u 1 u eij = + t 2 x j xi
NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

This formula is applied to calculating the strain increments, eij, for a zone:

5.2 Numerical Formulation of FLAC FLAC: For all gridpoints...

Once all stresses have been calculated, gridpoint forces are derived from the resulting tractions acting on the sides of each triangle. For example,

Lecture #5. Introduction to Finite Difference Method

Then a classical central finite-difference formula is used to obtain new velocities and displacements:

( in large strain mode)

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC

Lecture #5. Introduction to Finite Difference Method

Overlay & Mixed-Discretization Formulation of FLAC:

+
Each is constant-stress/constant-strain: Volume strain averaged over and separately

/2 =
. Deviatoric strain evaluated for

(Mixed discretization procedure)

Solution is Updated Lagrangian (grid moves with the material), and explicit (local changes do not affect neighbours in one timestep )
NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC

Lecture #5. Introduction to Finite Difference Method

combined into hexahedral elements. The scheme for deriving difference equations for a tetrahedron is described as follows:

FLAC3Ds grid is internally composed of tetrahedral elements. These are

8 node zone with two overlays of 5 tetrahedra in each overlay

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC

Lecture #5. Introduction to Finite Difference Method

Various degenerate forms of elements are also available in FLAC3D to enable the grid to conform to required geometry of boundaries:

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC FLAC3D: For all elements... Gauss theorem,
Face unit normal vector

Lecture #5. Introduction to Finite Difference Method

vi( l )
ni( f )

where vi vi , j x j is used to derived a finite difference formula for elements of arbitrary shape -

vi vi , j x j

Hence,
Volume V
Average face velocity vi
(f)

This formula is applied to calculating the strain rates, ij , for a zone:

Face area

S( f )
NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC FLAC3D: For all gridpoints...

Once all stresses have been calculated, gridpoint forces are derived from the resulting tractions acting on the sides of each tetrahedron. For example,

Lecture #5. Introduction to Finite Difference Method

for each triangular face. Then, contributions from each face and each tetrahedron (within a hexahedron) are summed -

Sum of contribution at global node of all tetrahedra meeting at that node

Then a classical central finite-difference formula is used to obtain new velocities and displacements:

( in large strain mode)

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC

Lecture #5. Introduction to Finite Difference Method

Overlay & Mixed-Discretization Formulation of FLAC3D :

+
Each is constant-stress/constant-strain.

/2 =
Forces from 2 overlays are averaged at gridpoints gives symmetry

Volume strain averaged over:

. Deviatoric strain evaluated for:

(Mixed discretization procedure) gives accurate plastic flow. Solution is Updated Lagrangian (grid moves with the material), and explicit (local changes do not affect neighbours in one timestep )
NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC Methods of solution in time domain

F
stress
numerical grid

Lecture #5. Introduction to Finite Difference Method

displacement u force F

EXPLICIT
All elements:

u x

IMPLICIT
element {F } = [ K ]{u} global } + [ K ]{u} = {F } [ m]{u

{F } = f ({u} , )
(nonlinear law)

Assume (u) are fixed Assume (F) are fixed Correct if x t < min Cp
p-wave speed

All nodes:
F } = {u t m

Repeat for n time-steps No iterations within steps

Solve complete set of equations for each time step Iterate within time step if nonlinearity present

Information cannot physically propagate between elements during one time step

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC Methods compared

Explicit, time-marching 1. Can follow nonlinear laws without internal iteration, since displacements are frozen within constitutive calculation. 2. Solution time increases as N3/2 for self-similar problems. 3. Physical instability does not cause numerical instability. 4. Large problems can be modeled with small memory, since matrix is not stored. 5. Large strains, displacements and rotations are modeled without extra computer time.

Lecture #5. Introduction to Finite Difference Method

Implicit, static 1. Iteration of the entire process is necessary to follow nonlinear laws 2. Solution time increases with N2 or even N3. 3. Physical instability is difficult to model. 4. Large memory requirements, or disk usage. 5. Significantly more time needed for large strain models.

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC DYNAMIC RELAXATION

Lecture #5. Introduction to Finite Difference Method

In dynamic relaxation gridpoints are moved according to Newtons law of motion. The acceleration of a gridpoint is proportional to the out-of-balance force. This solution scheme determines the set of displacements that will bring the system to equilibrium, or indicate the failure mode. There are two important considerations with dynamic relaxation: 1) Choice of timestep 2) Effect of damping

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC TIMESTEP

Lecture #5. Introduction to Finite Difference Method

In order to satisfy numerical stability the timestep must satisfy the condition:

xmin t < Cp
where Cp is proportional to 1 /mgp. For static analysis, gridpoint masses are scaled so that local critical timesteps are equal ( t = 1 ) which provides the optimum speed of convergence. Nodal inertial masses are then adjusted to fulfill the stability condition:

Note that gravitational masses are not affected.

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC DAMPING

Lecture #5. Introduction to Finite Difference Method

Velocity-proportional damping introduces body forces that can affect the solution. Local damping is used in FLAC --- The damping force at a gridpoint is proportional to the magnitude of the unbalanced force with the sign set to ensure that vibrational modes are damped:

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.2 Numerical Formulation of FLAC LOCAL DAMPING

Lecture #5. Introduction to Finite Difference Method

Damping forces are introduced to the equations of motion:

t i = [ Fi | Fi |sgn(u i ) ] u m
where Fi is the unbalanced force

The damping force, Fd is:

i ) Fd = Fi sgn(u

In FLAC the unbalanced force ratio (ratio of unbalanced force, Fi , to the applied force magnitude, Fm) is monitored to determine the static state. By default, when Fi / Fm < 0.001, then the model is considered to be in an equilibrium state.
NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.4 FLAC analysis for footing

JOB TITLE : .
(*10^1)

Lecture #5. Introduction to Finite Difference Method

FLAC (Version 4.00)

1.400

LEGEND 8-Aug-02 11:02 step 0 -9.751E-01 <x< 2.098E+01 -5.975E+00 <y< 1.598E+01 Material model mohr-coulomb Grid plot 0 Fixed Gridpoints X X-direction B Both directions 5E 0

B X X X X X X X X X B

X X X X X X X X X X

1.000

0.600

0.200

-0.200

Itasca Consulting Group, Inc. Minneapolis, Minnesota USA

0.200 0.600 1.000 (*10^1) 1.400 1.800

;Project Record Tree export ;Title:Strip Footing Testsnew ;Branch 1:st1.sav config extra 5 grid 20,10 model elastic group 'clay' notnull model mohr notnull group 'clay' prop density=2000.0 bulk=5000000.0 shear=2300000.0 cohesion=100000.0 & friction=0.0 dilation=0.0 tension=1.0E10 notnull group 'clay' fix x y i 1 21 j 1 fix x i 21 j 1 11 fix x i 1 j 1 11 save st1.sav

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.4 FLAC analysis for footing

;Branch 2:st2.sav fix x y i 1 4 j 11 set echo off call load.fis load set echo off call err.fis err history 1 disp history 2 err history 3 load history 4 sol save st2.sav restore st2.sav ;Branch 0:st4.sav initial yvelocity -0.0010 i 1 4 j 11 history 999 unbalanced cycle 3000 save st4.sav restore st2.sav

Lecture #5. Introduction to Finite Difference Method

;Branch 0:st6.sav set gravity=9.81 free x y i 2 4 j 11 free y i 1 j 11 history 999 unbalanced solve fix x y i 1 4 j 11 initial yvelocity -0.0010 i 1 4 j 11 ;Branch 0:st5.sav cycle 3000 model mohr notnull group 'clay' save st6.sav prop density=2000.0 bulk=5000000.0 shear=2300000.0 restore st2.sav cohesion=100000.0 & ;Branch 0:st7.sav friction=0.0 dilation=0.0 initial yvelocity -0.01 i 1 4 j 11 tension=0.0 notnull group 'clay' initial yvelocity -0.0010 i 1 4 j 11 history 999 unbalanced cycle 1000 ;Branch 0:st3.sav history 999 unbalanced save st7.sav initial yvelocity -1.0E-4 i 1 4 j 11 cycle 3000 history 999 unbalanced save st5.sav cycle 30000 restore st2.sav save st3.sav restore st2.sav
NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS

5.5 FLAC analysis for slope and tunnel

Lecture #5. Introduction to Finite Difference Method

NUMERICAL NUMERICAL METHODS METHODS AND AND GEOMECHANICS GEOMECHANICS