DERIVING FINITE ELEMENT EQUATIONS USING
WEIGHTED RESIDUAL METHOD - LEAST SQUARES
APPROACH
Presented by
A. Sabiha Khathun
M.TECH (Machine Design)
16481D1501
Subject: Finite Element Method
GUDLAVALLERU ENGINEERING COLLEGE
Sheshadri Rao Knowledge Village, Gudlavalleru,PIN:521356
Introduction
The finite element method (FEM) is a numerical method for
solving problems of engineering and mathematical physics.
STARTED: for solving linear problems in solid mechanics
CURRENT: for solving non linear problem, dynamic, transient
problems
Domain is descreitized into finite element method
Application of residual and variational method
Differential equation converted into algebraic equation using
integral approximations.
Approaches To Solve A Engineering Problem
General process
a) Assume a trail solution
Discretizie the domain in to different elements
Approximate the field variable over the domain (interpolation
functions)
b) Satisfy the boundary conditions
c) Satisfy the differential equation with trail solution(error is residual)
d) Minimize the residual over the whole domain to obtain equations.
Engineering Problems & Governing Differential
Equations:
Integral formulation: The equation derived using an energy
principle, such as the principle of minimum potential energy, or a
weighted residual method, such as Galerkin Method, are usually of
a weak form.
Three different approaches are being used when formulating an
FEM problem. They are:
1. Direct Approach
2. Variational Approach
3. Weighted Residual Method
Weighted Residual Methods
Weighted residual method is a generic class of method developed
to obtain approximate solution to the differential equations of the
form
L (φ) + f = 0
where φ(x) is the dependent variable and is unknown and f(x) is a
known function.
i.e., L (ψ) + f ≠ 0 or
L (ψ) + f = R
where R(x) is a measure of error commonly referred to as the
residual.
UNIFORM ROD, UNIFORM AXIAL LOAD
Governing differential equation is
Continued…
With regards to the selection of weight function, we have
several choices. Hence, depending upon nature of weight
function, we have different types of weighted residual methods.
Some of the standard methods are:
1. Point Collocation Method
2. Subdomain Collocation Method
3. Least Square Method
4. Galerkin Method
Objective:
Deriving finite element equation using weighted residual method-
least square method.
a) Problem formulation
b) Solution to given differential equation subject to the boundary
conditions.
Method of least squares
In this method, the integral of the weighted square of the
residual over the domain is required to be a minimum;
i.e.,
The necessary conditions for minimizing the integral can be
expressed as
c1and c2 are unknown constants & f1(x) and f2(x) are trail
functions
Consider the beam deflection problem. The governing
differential equation is
y(0) = 0,y(L) = 0.
The trigonometric trial function is given by
where
and the residual is
and its derivative
The weighted residual equation can now be written as
On integrating above equation, it gives
Solving for A, we have
and thus, the approximate solution is
Example 1.
Finding solution for the following differential equation
Subject to the boundary conditions
using the least squares method.
Approach
Assume an approximate solution satisfying the boundary
conditions which two unknown constants.
Use the conditions to minimize the integral of the square of
the residue to evaluate the constants.
Differential equation is
The approximate solution satisfying the boundary conditions is
taken as
Where c1and c2 are unknown constants .
using this solution the residue can be expressed as
R2 = c12(-2+x-x2)2+c22(2-6c2+x2-x3)2+x2
On simplifying, we get
R 2 = C12(4-4x+x2-2x3+x4)+c22(4-24x+40x2-16x3+13x4-
2x5+x6)+2c1c2(-4+14x-16x2+9x3-2x4-x5)-2c1x+c1x2-
c1x3+2c2-6c2x2+c2x3-c2x4)
On integrating with respect to x with limits o to 1, we get
dx = 0.033c12+3.74c22-2(0.65)c1c2-0.91c1+0.05c2
The weighting function w is usually taken as unity in this method.
The application of least squares method leads to following
equations:
-0.65c 1+3.74c 2+0.05=0
0.066c 1-1.3c 2-0.9166=0
On solving the two equations we get c1=-5.62and c2=-0.188
The approximate solution of the equation is
Φ(x)= -5.62x(1-x)-0.188x2(1-x)
CONCLUSION
More widely used than collocation and Subdomain,
linear systems are easily assembled.
linear systems are relatively well conditioned.
least-square functional provide an easily computable residual
error indicator .
the treatment of general boundary conditions, including no
homogeneous ones, is greatly facilitated because boundary
condition residuals can be incorporated into the least-square
functional.
