Sunteți pe pagina 1din 9

______________________Basics of the Finite Element Method Applied in Civil Engineering

CHAPTER 3

DISCRETE APPROXIMATIONS

The FEM is a procedure for obtaining a numerical approximation to the so


called “exact” solution of a boundary problem expressed over a domain D.
This domain is replaced by the union of disjoint and adjacent sub domains
De (elements). Generally, the geometry of D is only approximated by that of
the union of individual sub domains De. The unknown function is locally
approximated over each element by an interpolation formula expressed in
terms of the values taken by the function (and possibly its derivatives) at a
set of nodal points located on the element’s boundary.

This chapter is dedicated to the approximation techniques used to replace a


continuous physical system by an equivalent discrete one. A few simple
examples will describe the nodal approximation of a function on the domain
D, leading to the finite element approximation.

A simple but suggestive way to illustrate the idea of a continuum


approximation by discrete parts is to calculate the circle perimeter (or
length) L = πd by replacing it with the perimeter of an inscribed polygon
with n sides. The polygon’s perimeter P = nln will become closer to the
circle length by increasing the number of sides n (concurrently with the
reduction of side length ln). The difference between the real length of the
circle L and the sum of polygon’s sides P is the approximation error (see
figure 3.1),

e = L – P.

Enlarging the number of sides the error decreases. The real length of the
circle is reached for n → ∞.

25
Chapter 3 Discrete Approximations___________________________________________

1
ln
8 2
α

d 7 3
α = 2π/n
ln = d sin
6 4

5
Fig. 3.1 Approximation of circle’s length by an n-sided polygon

Let’s suppose that a physical system can be described by a mathematical


model in terms of a so called exact variable (or function) u(x) which is
replaced on the system’s domain D by an approximate function ua(x), thus
the difference (or the error)
e(x) = ua(x) - u(x)

is small enough for the desired purpose. To define the approximate function
ua(x) the following steps should be followed:

- to chose a convenient function depending on n parameters ai

u(x, a1, a2, …, an);

- to find the ai parameters which satisfy the previous condition (task


that can be accomplished by assigning a null value to the error e(x)
in n different points of the domain).

The function u(x, a1, a2, …, an) is usually chosen so that the integral or
differential operations should be easily applied, leading to polynomial
expression or to a solution of a differential equation.

Because the approximate function ua(x) is usually linear regarding to ai, the
general form may be

26
______________________Basics of the Finite Element Method Applied in Civil Engineering

u a ( x) = P1 ( x)a1 + P2 ( x)a 2 + ... + Pn ( x)a n

 a1 
a 
 2
or u a ( x) = [P1 ( x) P2 ( x) ... Pn ( x)] .  = P ⋅ a
 
.
a n 

where P1, P2, … , Pn are linear independent functions and


a1, a2, … , an are the approximation parameters.

Because the parameters a1, a2, … , an don’t usually have a physical


meaning, they can be assigned as values of the exact function u(x) in n
different points (called nodes) where the function’s values are known.

ua(x1) = u(x1) = u1
ua(x2) = u(x2) = u2
……………
ua(xn) = u(xn) = un

Thus, the approximate function will be

u ( x) = N1 ( x)u1 + N 2 ( x)u 2 + ... + N n ( x)u n =


 u1 
u 
 2
[N1 ( x) N 2 ( x) ... N n ( x)] .  = N ⋅ u
 
.
u n 

The nodal approximation has two fundamental properties:

0 when i ≠ j
• because ua(xi) = ui the interpolation functions Ni = 
1 when i = j

27
Chapter 3 Discrete Approximations___________________________________________

• the approximation error defined by e(x) = ua(x) - u(x) is null in all


the nodes xi, e(xi) = 0.

In the following examples the unknown functions are one-dimensional


(depending on a single variable x) but the procedure can be extended to the
n-dimensional space.

3.1 EXAMPLE 1 Approximation of a physical parameter

Supposing that a particular physical parameter can be measured in only 3


points along the one-dimensional domain between x = 0 and x = 1. To know
the parameter’s value in any point between the boundaries, an
approximation function ua(x) should be found, based on the known values
corresponding to the measuring points.

Measuring points x 0 0.5 1


Function values u(x) 10 18 12

Choosing a polynomial approximation of second degree,

u(x) ≈ u(x, a1, a2, a3) = a1 + a2 x + a3 x 2

u(x = 0) = ua(x = 0) = a1 = 10

u(x = 0.5) = ua(x = 0.5) = a1 + 0.5 a2 + 0.25 a3 = 18

u(x = 1) = ua(x = 1) = a1 + a2 + a3 = 12

yields the solution a1 = 10


a2 = 30
a3 = –28

The physical parameter’s value in a certain point, x = 0.75 for instance, will
be

ua(x = 0.75) = a1 + 0.75 a2 + 0.5625 a3 = 16.75

28
______________________Basics of the Finite Element Method Applied in Civil Engineering

3.2 EXAMPLE 2 Approximation of a differential equation solution

Let’s find o function u(x) satisfying the differential equation

d 2 u ( x)
= f ( x) for 0 ≤ x ≤ 1
dx 2

with the boundary conditions u(x) = 0 when x = 0 and x = 1, while f(x) is a


known function, thus f(x = 0.25) = 1 and f(x = 0.75) = 0.25.

An approximate function which satisfies the boundary condition is the


following:
u ( x) ≈ u a ( x) = a1 sin(πx) + a 2 sin(2πx)
with
d 2u
2
= − a1π 2 sin(0.25π ) − 4a2π 2 sin(0.5π ) = g ( x1 ) = 1
dx x
1

d 2u
= − a1π 2 sin(0.75π ) − 4a 2π 2 sin(1.5π ) = g ( x 2) = 0.25
dx 2 x2

5 1 3 1
This leads to a1 = ; a2 = and the final expression of the
4 2π
2
32 π 2
approximate function

5 1 3 1
u ( x ) ≈ u a ( x) = − 2
sin(πx) − sin(2πx)
4 2π 32 π 2

For a certain point, for instance x = 0.25,

23 1
u a ( x = 0.25) = − = −0.0728
32 π 2

Note: In this example the approximation has replaced a differential


equation by two algebraic equations with a1 and a2 as unknowns.

29
Chapter 3 Discrete Approximations___________________________________________

3.3 EXAMPLE 3 Lagrange type, 4 - nodes approximation

Let’s consider an arbitrary function u(x) known in only 4 points,


approximated by

u ( x) = N1 ( x)u1 + N 2 ( x)u 2 + N 3 ( x)u3 + N 4 ( x)u 4

where Ni are Lagrange polynomials of 3rd order, written in a general form

4 (x − x j )
N i ( x) = ∏
j =1 ( xi − x j )
j ≠i

which are satisfying the previous fundamental properties.

If x1 = 1, x2 = 2, x3 = 5 and x4 = 7, the N1(x) function’s graph gives:


1
N1 = − ( x − 2)( x − 5)( x − 7)
24

x 1 1.5 2 3 4 5 6 7
77 1 1 1
N1 1 0 − − 0 0
192 3 4 6

N1

x1 x2 x3 x4 x
1.0 2.0 5.0 7.0

u u(x)
u1 u2 u4
ua uai(x)
u3
e(x)

x1 x2 x3 x4 x
Fig. 3.2 Graphs of polynomial N1, exact function u(x), approximate function ua(x)
and error distribution along the domain D ∈ [x1, x4]

30
______________________Basics of the Finite Element Method Applied in Civil Engineering

The exact function u(x), the approximate function ua(x) as well as the error
e(x) are represented in figure (3.2).

The nodal approximation method for a one-variable function can be


extended to multi-variable functions. For 3 variables

u(x, y, z) = u(x), where x = [x y z], x∈D

The approximate function, written ua(x), should verify the equalities

ua(xi) = u(xi) = ui, where

x = [xi yi zi], i = 1, 2 , … n are the node’s i coordinates.

The previous examples render evident the difficulty to build an approximate


function u(x) for an increasing number of nodes n and the correspondent
parameters ui. The problem becomes even harder for a domain D with a
complex shape, if the approximate function u(x) should also satisfy some
boundary conditions.

The nodal approximation method on sub domains De enables finding the


approximate functions, being also suitable for computer programming.
When the sub-domains are called finite elements, the method becomes the
finite element approximation method.

Two steps should be followed:

- to identify the sub-domains (called elements) De ∈ D, ∑De ≈ D;


- to define the approximate function ua(x)e on each element, by
applying the nodal approximation method.

This approximation method has the following characteristics:

- each ua(x)e function is defined only by the nodal variables attached


on the element De;
- the approximate functions ua(x)e are defined to be continuous on
their sub-domain De; also they satisfy some continuity condition
between adjoining elements.

31
Chapter 3 Discrete Approximations___________________________________________

3.4 EXAMPLE 4 One dimensional approximation using finite elements

Elements geometry definition nodes: 1, 2, 3, 4


nodal coordinates: x1, x2, x3, x4
complete domain: D: x1 < x < x4
elements: D1 : x1 < x < x2
D2 : x2 < x < x3
D3 : x3 < x < x4

Approximate functions uae(x) known nodal values: u1, u2, u3, u4


ua(x)e linear on each element

u
u2 u4
ua u1 u(x)
ua 2(x
)
3 (x )
1 x) u3 ua
ua ( uai(x)
1 x1 2 x2 3 x3 4 x4 x
D1 D2 D3

Fig. 3.3 Graphs of exact and approximate functions along the domain D

Element 1 (D1) ua1(x)= N1 u1 + N2 u2 x1 ≤ x ≤ x2

x − x2
where N1 = N1(x1) = 1 N1(x2) = 0
x1 − x2
x − x1
N2 = N2(x1) = 0 N2(x2) = 1
x2 − x1

Element 2 (D2) ua2(x)= N1 u2 + N2 u3 x2 ≤ x ≤ x3

x − x3 x − x2
where N1 = ; N2 = etc.
x2 − x3 x3 − x2

Element 3 (D3) ua3(x)= N1 u3 + N2 u4 x3 ≤ x ≤ x4

32
______________________Basics of the Finite Element Method Applied in Civil Engineering

x − x4 x − x3
where N1 = ; N2 = etc.
x3 − x4 x4 − x3

The approximate functions uae(x) and the shape functions Ni(x) are different
on each element De. The sum of functions ua1(x), ua2(x), ua3(x) gives the
approximate function on the complete domain D.

u a1( x ) u2
u1

x
x1 x2 x3 x4

Fig. 3.4 Graph of approximate function ua1(x) along the domain D

Remarks

1. Notice the continuity of different functions uai(x) in the nodal points


xi; although adjacent functions have the same values in each nodal
point, the “slope” of the graphs differs (achieved function continuity
without slope continuity);
2. The approximate function can be sought on the whole domain D as a
3 order polynomial,

u ( x) ≈ u ( x, a1 , a 2 , a3 , a 4 ) = a1 + a 2 x + a3 x 2 + a 4 x 3

according to the known values of the exact function u(x); the same
steps should be followed as in Example 1, this time looking for four
ai parameters. Notice that the same approximation problem can be
solved by using either simple linear functions on discrete domains
(i.e. a larger number of functions) or a single polynomial function of
higher order on the whole domain.

33

S-ar putea să vă placă și