0% found this document useful (0 votes)
70 views5 pages

Storage & Solution-1

This document discusses techniques for storing and solving large finite element models. It describes banded matrix storage, which stores only the nonzero elements within a bandwidth of the matrix diagonal to save memory. Direct solution methods like Gauss elimination and iterative methods are covered. For very large sparse matrices, skyline storage tracks the nonzero pattern. Wavefront solvers and substructuring are introduced to decompose large models into smaller subproblems to improve computational efficiency.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
70 views5 pages

Storage & Solution-1

This document discusses techniques for storing and solving large finite element models. It describes banded matrix storage, which stores only the nonzero elements within a bandwidth of the matrix diagonal to save memory. Direct solution methods like Gauss elimination and iterative methods are covered. For very large sparse matrices, skyline storage tracks the nonzero pattern. Wavefront solvers and substructuring are introduced to decompose large models into smaller subproblems to improve computational efficiency.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

M.

Vable

MEEM 5170: Storage and Solution

Fall 2015

Storage and Solution Techniques


Banded Matrix
Each node has two degrees of freedom (u and v)
Element matrix is 6 x 6
LTE= Linear Triangle Element

LTE

LTE

LTE

Half Band Width =NB

LTE

LTE

LTE

10

11

12

13

14

15

16

10

11

12

13

14

15

16

Only the upper banded form of the global stiffness matrix is stored in a
rectangular matrix of size NxNB

M. Vable

MEEM 5170: Storage and Solution

Fall 2015

Banded Matrix
6

7
LTE

LTE
LTE

LTE

LTE

LTE

Half Band Width =NB


1

10

11

12

13

14

15

16

10

11

12

13

14

15

16

The bandwidth is solely dictated by the difference between the highest and
the lowest node number on the elements.

M. Vable

MEEM 5170: Storage and Solution

Fall 2015

Equation Solving [K]{d}={R}


Direct Methods: Matrix inversion, Gauss Elimination, Gauss-Jordan, Choleskys
Method.

Iterative Methods: Jacobi, Gauss-Seidel, Successive Overrelaxation.


Gauss Elimination
A lower triangular matrix has zero coefficients above the diagonal [L]
A upper triangular matrix has zero coefficients below the diagonal [U]
Any matrix [K] can be decomposed as [K]=[L][U]. The algebraic equations
become: [L][U]{d}={R}. If we substitute [U]{d}={x} we obtain [L]{x}={R}. The
solution is obtained in two steps:
1. Forward substitution: Solve for {x} from: [L]{x}={R}
2. Backward substitution: Solve for {d} from: [U]{d}={x}

If the algebraic matrix is symmetric and positive definite then the


decomposition is done using Choleskys method to obtain [U]=[L]T.
Forward substitution is done in increasing node number order and
backward substitution is done in decreasing node number order.
Pros and Cons
Direct methods are better for multiple load vectors.The computation
2

time is proportional to NN B .
Computation times for iterative methods depend upon the starting
guess solution. In time dependent problems that are solved in small
time steps the initial guess vector is known from previous time step.
Non-linear problems which are solved as linear problems iteratively
also use less time in iterative methods.

M. Vable

MEEM 5170: Storage and Solution

Fall 2015

Large Systems
Skyline storage scheme
The stiffness matrix of very large structures may have mostly zero terms, even
inside the band width. Such matrices are called sparse matrices.
Hi= The height of the column above the diagonal of the ith row beyond
which there is no non-zero term.
1

10

11

12

13

14

15

16

10

11

12

13

14

15

16

Hi

Skyline algorithms are used for sparse matrices.

M. Vable

MEEM 5170: Storage and Solution

Fall 2015

Wavefront (Frontal) equation solver:


Assembly and equation solving proceeds simultaneously.
Computation of coefficients in a stiffness matrix and load vector is
complete only when contribution of all elements that share the node
have been added.
A complete coefficient can be processed in the solution process.
Wavefront number is a measure of number of coefficients being
manipulated in the solution process at a given step.
Substructure (Matrix partitioning)
[A]

[B]

d1

R1

[C]

[D]

d2

R2

Algebraic equations can be written as:

The two sets of matrix equations can be solved as shown below.


1
A d 1 + B d 2 = R 1 or d 1 = A R 1 B d 2

C d1 + D d2 = R2
C A

R1 B d2 + D d2 = R2

D C A

B d2 = R2 C A
KS d2 = RS

R1

In substructuring {d2} are the nodal displacements on the interface of two substructures and {d1}are the displacements of nodes on either of the substructure.

Substructuring is used for large structures. Each substructure can be


used as a superelement if a substructure repeats in a structure.
Substructuring could be used for creating meshes with different densities either for creating original mesh or for mesh refinements.
In non-linear analysis such as in elastic-plastic analysis. The plastic
zone is made a part of a substructure.

You might also like