You are on page 1of 32

12

Mesh Free Methods


Introduction Nico van der Aa
19th November 2003

/ department of mathematics and computer science

1/31

12
Contents
What is a Mesh Free Method? Mesh Free Method Why do we like to have Mesh Free Methods? How can we use Mesh Free Methods? Node generation Shape functions Conclusions Questions

/ department of mathematics and computer science

2/31

12
What is a mesh?
A mesh is a net that is formed by connecting nodes in a predened manner. Other words are grid, volumes, cells or elements.

/ department of mathematics and computer science

3/31

12
What is mesh free?
Mesh Free Methods use a set of nodes scattered within the problem domain as well as sets of nodes scattered on the boundaries of the domain to represent (not discretize) the problem domain and its boundaries. No mesh implies no information on the relationship between the nodes is required.

/ department of mathematics and computer science

4/31

12
When is a method a Mesh Free Method?
minimum requirement: A predened mesh is not necessary (at least in eld variable interpolation). ideal requirement: No mesh is necessary at all throughout the process of solving the problem of a given arbitrary geometry governed by partial dierential system equations subject to all kinds of boundary conditions. The Mesh Free Methods developed so far are not really ideal.

/ department of mathematics and computer science

5/31

12
Why the need for Mesh Free Methods?
For an engineer the concern is more the manpower time, and less the computer time, Large deformations can deteriorate the accuracy because of element distortions, Specic physical problems, such as breakage of a solid and stress calculations, require more exibility.

Mesh Free Methods are answers to the problems of the Finite Elements Method!

/ department of mathematics and computer science

6/31

12
How do Mesh Free Methods work?
Step 1: Domain representation
The problem domain is dened.

/ department of mathematics and computer science

7/31

12
How do Mesh Free Methods work?
Step 1: Domain representation
The problem domain is dened. A set of nodes is chosen to represent the problem domain and its boundary. In general, the density of the nodes is not uniform.

/ department of mathematics and computer science

7/31

12
Support Domain
Denition The support domain for a point x is a sphere of a certain radius that relates to the nodal spacing near the point x. Reason Determines the number of nodes to be used to approximate the function value at x. Restriction The nodal density does not vary drastically in the problem domain.

/ department of mathematics and computer science

8/31

12
Step 2: Field interpolation
The eld variable u at any point x within the problem domain is interpolated using the values of this eld at all the nodes within the support domain of x. Mathematically,
n

u(x) =
j=1

i(x)ui

/ department of mathematics and computer science

9/31

12
Step 3: Formulation of system equations
The equations of a mesh free method can be formulated using the shape functions and a strong or weak form system equation. Result in global system matrices for the entire problem domain. The procedures of forming system equations are slightly dierent for dierent MFree methods.

/ department of mathematics and computer science

10/31

12
Step 4: Solving the global MFree equations

Numerical method depends on the kind of equations! Choices: Direct solver vs. iterative solver; Implicit method vs. explicit solver; Required accuracy; Speed of method; Stability of method; ...

/ department of mathematics and computer science

11/31

12

/ department of mathematics and computer science

12/31

12
Features of a nodal generation
The nodes must represent both problem domain and boundary; The nodes can be chosen arbitrary within reason; The node distribution can be uniform or not.

/ department of mathematics and computer science

13/31

12
Support domain vs Inuence domain
An inuence domain is the domain that a node exerts an inuence upon. It goes with a node.

inuence domain support domain node point uniform distribution of nodes nonregularly distributed nodes

/ department of mathematics and computer science

14/31

12
Properties of Mesh Free Shape Functions
Partition of unity (compulsory condition)
n

i(x) = 1
i=1

Linear eld reproduction (preferable condition)


n

i(x)xi = x
i=1

Kronecker delta function property (preferable condition)

i(xj ) =

1 i=j 0 i=j
15/31

/ department of mathematics and computer science

12
Requirements of a good method of shape function construction:
1. arbitrary nodal distribution 2. stability 3. consistency 4. compact support 5. eciency 6. delta function property 7. compatibility

/ department of mathematics and computer science

16/31

12
Methods to construct shape functions:
Finite integral representation methods: Smoothed particle hydrodynamics (SPH) method, ...
x2

f (x) =
x1

f ()W (x )d

Finite series representation methods: Moving Least Squares (MLS) methods, Point Interpolation Methods (PIMs), ...

f (x) = a0 + a1p1(x) + a2p2(x) + ...


Finite dierential representation methods:

f (x) = f (x0) + f (x0)(x a) +

1 f (x0)(x a)2 + ... 2!


17/31

/ department of mathematics and computer science

12
SPH method (nite integral method)
The integral representation of the eld function u(x):

u(x) =

u()(x )d

In SPH:

u(x) =

u()W (x , h)d

This integral representation is valid and converges when the weight function satises certain conditions: 1. W (x , h) > 0 over ; 2. W (x , h) = 0 outside ; 3.

W (x , h)d = 1;

4. W is a monotonically decreasing function; 5. W (s, h) (s) as h 0.

/ department of mathematics and computer science

18/31

12
SPH: Discretised form
n n

uh(x) =
i=1

W (x xi)ul Vl =
i=1

i(x)ui

The shape function i are dened as:

i(x) = W (x xi)Vl
Remarks: Kronecker delta function is not always satised (ul = uh (x)); Compatible in eld function approximation; Consistent in the interior of the domain, not on the boundary; Reproductivity in the interior of the domain, not on the boundary; The shape functions need to be scaled to satisfy the condition of unity.

/ department of mathematics and computer science

19/31

12
SPH: consistency
C 0 consistency is ensured by the condition of unity imposed on a weight function. C 1 consistency is not immediately available on the boundary.

/ department of mathematics and computer science

20/31

12
SPH: most commonly used shape function
Cubic spline weight function (W1): Quartic spline weight function (W2): Exponential weight function (W3);
Weight functions 1.2 W1 W2 W3 1

0.8

0.6 W

0.4

0.2

SCALING IS STILL NECESSARY


0.8 0.6 0.4 0.2

0.2 1

/ department of mathematics and computer science

0 d

0.2

0.4

0.6

0.8

21/31

12
MLS (Moving Least Squares)
m

u (x) =
j=0

pj (x)aj (x) = pT (x)a(x)

with

pT (x) = 1, x, x2, ..., xm pT (x, y) = 1, x, y, xy, x2, y 2, ..., xm, y m pT (x, y, z) = 1, x, y, z, xy, yz, xz, x2, y 2, z 2, ..., xm, y m, z m
Features: nite series representation; ensures the approximated eld function is continuous and smooth in the entire problem domain; capable of producing an approximation with the desired order of consistency.

/ department of mathematics and computer science

22/31

12
MLS: weighted residual
uh(x, xi) = pT (xi)a(x)
Weighted residual:
n

J=
i=1 n

W (x xi)[uh(x, xi) u(xi)]2 W (x xi)[pT (xi)a(x) ui]2


i=1

In MLS approximation, at an arbitrary point x, a(x) is chosen to minimize the weighted residual.

/ department of mathematics and computer science

23/31

12
MLS: minimization condition requirement
J = 0 A(x)a(x) = B(x)U s a
with

A(x) =

n i=1

Wi(x)p(xi)pT (xi)
T

Wi(x) = W (x xi) B i = Wi(x)p(xi)

B(x) = [B 1, B 2, ..., B n] U s = {u1, u2, ..., un}


Solution

a(x) = A1(x)B(x)U s

/ department of mathematics and computer science

24/31

12
MLS: shape function
n

u (x) =
i=1

i(x)ui

with the shape functions


m

i(x) =
j=0

pj (x)(A1(x)B(x))ji = pT (x)A1(x)B i

Remarks about the MLS shape function: Kronecker delta function is not satised; possesses mth order consistency; is reproductive; scaling is not necessary to satisfy the unity requirement; compatible.

/ department of mathematics and computer science

25/31

12
PIM (Point Interpolation Method)
n n

uh(x, xQ) =
i=1

Bi(x)ai(x) =
i=1

pi(x)ai(xQ) = pT (x)a(xQ)

Determination of a: enforce for every node i

ui = pT (xi)a,
or in matrix natotation:

i = 1, ..., n

U s = P Qa a = P 1U s Q

/ department of mathematics and computer science

26/31

12
PIM: shape function
n

uh(x) =
i=1

i(x)ui = (x)U s

The matrix of shape functions is dened as

(x) = pT (x)P 1 Q
Note that P 1 does not have to exist! Q

/ department of mathematics and computer science

27/31

12
PIM: properties of the shape function
Shape functions have nth order consistency; are linearly independent; possesses the Kronecker delta function property; are partitions of unity; possess reproducing properties; use no weight functions in their construction; are not compatible.

/ department of mathematics and computer science

28/31

12
3. Conclusions
Mesh Free Methods are a respons to the limitations of Finite Element Methods. Mesh Free Methods do not use meshes, only nodes. The implementation of Mesh Free Methods diers from Finite Element Methods only in the shape function construction and node generation. The ideal Mesh Free Method is not found yet.

/ department of mathematics and computer science

29/31

12
References Mesh free methods:
Moving beyond the nite element method Gui-Rong Liu CRC Press LLC 2003

/ department of mathematics and computer science

30/31

12

Questions ?

/ department of mathematics and computer science

31/31

You might also like