Finite Element Modelling for Advanced Processing

A crash-course in theoretical and computational Solid Mechanics

WMME013-05

Francesco Maresca

Lecture 2 / April 21st 2021

Summary: previous week and today’s lecture

LAST WEEK Von Mises

1. Small deformations solid mechanics

1D tension

2. Elasto-plasticity:
i. Yield surface
ii. Consistency condition Tresca

iii. Hardening
Plastic loading

Elastic unloading
TODAY: FEM in small deformations

WMME013-05 \ FEM for Advanced Processing 2

FEM in small deformations

3
Weak form of equilibrium equations

Recall the equilibrium equations (equilibrium of momentum)

+ boundary conditions (BCs)
in V on displacements and tractions
Perform the following manipulation
1. Take the dot product with (admissible displacement “variation”)
in V, ( where displ. are assigned).

2. Integrate over the whole domain V

Includes assigned tractions at boundary

3. Use divergence theorem

and chain rule:
WMME013-05 \ FEM for Advanced Processing 4
Weak form of equilibrium equations
Point 3 in detail: rework

Compact notation Index notation

(Divergence theorem
plus chain rule)

(Divergence theorem) ,

(Chain rule) , ,
,

WMME013-05 \ FEM for Advanced Processing 5

Weak form of equilibrium equations

“Strong” form of equilibrium (SE):

in V Displacement (Dirichlet) and traction (Neumann) BCs

“Weak” form of equilibrium (WE):

1) admissible
(zero where displacement BCs in (SE))
2) equal to traction BCs in (SE)

Internal, “virtual” work External, “virtual” work

WE = Principle of Virtual Works (PVW) Equilibrium. WE satisfied by a larger class

of functions (PC1) than SE.
“Strong”: 1st derivative of stress ~ 2nd derivatives of displacement
“Weak”: no stress derivatives ~ 1st derivative of displacement  requirements on derivative
“weaker” in WE than SE.

WMME013-05 \ FEM for Advanced Processing 6

Weighted residuals approach

We want to solve the equilibrium equations, i.e. find s.t.

in V +BCs What if we cannot find analytically?

Approach:
1. Take a trial function , list of (unknown) parameters
 (in general), residual

2. Find such that the following relation holds:

weighting function
(or test function)
Weighted residuals

WMME013-05 \ FEM for Advanced Processing 7

Weighted residual methods
Weighted residuals: minimize the error
in weighted integral sense (meaning depends on )

Examples of weighted residual methods:

1. Collocation method
, Dirac’s delta, list of points

1. The specific weighted residual method
2. Least square method defined by choice of weighting fct.

Error is minimized when 2. The accuracy of the solution

 ( ) depends by class of trial fct.

3. Galerkin’s method (used in FEM)

: test fct. approximated like trial function

WMME013-05 \ FEM for Advanced Processing 8

Discretization: why is it needed?
trial function approximating the (unknown) solution in V.

Example 1: interpolate displacement field with 1st order polynomial:

where ,
 is defined in the whole domain.
Once is determined with weighted residual method, the strain

will be only constant in (independent of ).

 Exact solution only for problems with homogeneous deformation

E.g: uniaxial tension

WMME013-05 \ FEM for Advanced Processing 9

Discretization: why is it needed?
trial function approximating the (unknown) solution in V.

Example 2: interpolate displacement field with 2nd order

(complete) polynomial:

where ,
 is defined in the whole domain.
Once is determined with weighted residual method, the strain
Coenen et al., IJNME 90:1

will be only linear in .

Strain localization

 Increasing order of polynomial  Solving problems with higher complexity.

But generally inefficient! No strain/homogeneous strain

WMME013-05 \ FEM for Advanced Processing 10

Finite Element Discretization
FEM strategy:
1. Write the analytical weak form (WE);
2. Partition the domain into subdomains (finite elements);
3. Define interpolation functions with local support (element-wise);
4. Assemble the elements;
5. Solve the algebraic weighted residual/WE equation.

(exact solution)

(discrete solution, using

linear interpolations element-wise)

Note: both the domain

and the solution are approximate

WMME013-05 \ FEM for Advanced Processing 11

FEM Example: 1-D bar. From SE to WE
In 1-D linear elasticity, the equilibrium equations reduce to

We consider the boundary conditions (BCs):

“Essential” (Dirichlet) BC

“Natural” (Neumann) BC

1. Weak form (WE)

Recall integration by parts:

WMME013-05 \ FEM for Advanced Processing 12

FEM Example: 1-D bar. FE partitioning and interpolation

(WE)

2. Subdomain partitioning Connectivity matrix

1 2 3 elements (m) Element (m) Node i ( ) Node j ( 𝒋 )
1 1 ( ) 2 ( )
1 2 3 4 nodes (i,j)
2 2 ( ) 3 ( )
coordinates
3 3 ( ) 4 ( )

3. Per-element interpolation
Example: linear interpolation
m
gives
i j ℓ
(local support) The 1st eq. gives
ℓ

WMME013-05 \ FEM for Advanced Processing 13

FEM Example: 1-D bar. FE interpolation
3. Per-element interpolation (continued)
Example: linear interpolation
m

i j
(local support)

Linear interpolation rewritten in terms of nodal displacements

Lagrange polynomial

vector of shape (interpolation) fuctions

Properties of shape functions

i j
WMME013-05 \ FEM for Advanced Processing 14
FEM Example: 1-D bar. Assembly

(WE)

4. Assembly 3
1 2
Discretized weak form: 1 2 3 4
WE discrete sum
(WE)  of per-element WEs

Recall: in the Galerkin method, the admissible test fct. has same degree of approximation as the trial function

arbitrary displacement variations at the nodes

Since scalar function, (enables to get final system of algebraic equations)

(only shape functions depend explicitly on )

WMME013-05 \ FEM for Advanced Processing 15

FEM Example: 1-D bar. Assembly
4. Assembly (continued)

WE discrete sum
(WE) 
of per-element WEs

m subscript:
per-element quantities

global array of nodal displacement variations (here: entries). Note: ,

global array of nodal forces.

Take node values

out of integrals:

Element stiffness matrix

(WE) 

WMME013-05 \ FEM for Advanced Processing 16

FEM Example: 1-D bar. Assembly
4. Assembly (continued)

Perform the sums = assembly

Recall the connectivity matrix

Element (m) Node i ( ) Node j ( 𝒋 )
1 1 ( ) 2 ( )
2 2 ( ) 3 ( )
3 3 ( ) 4 ( )

Global stiffness matrix Global vector of nodal displacements

(WE) 

(WE)  Algebraic linear This is the nodal equilibrium equation

system of equations: The solution is the list of nodal
displacements

WMME013-05 \ FEM for Advanced Processing 17

FEM Example: 1-D bar. Assembly
4. Assembly (continued)
Recall:

Assumption (simplification): (constant)

Recall the connectivity matrix m=1

m=2
Element (m) Node i ( ) Node j ( 𝒋 )
m=3
1 1 ( ) 2 ( )
2 2 ( ) 3 ( )
3 3 ( ) 4 ( )

WMME013-05 \ FEM for Advanced Processing 18

FEM Example: 1-D bar. Solution
5. Solve algebraic equations

in this example

unknown
Recall BCs: This is

known  Only , and are unknown!

System partitioning and solution: Linear algebraic systems to solve

1. This gives

2. By using the solution

is found

WMME013-05 \ FEM for Advanced Processing 19

FEM Example: 1-D bar. Solution
5. Solve algebraic equations
System partitioning

1.

FE solution

2.

1 1 2 2 3 3 4

WMME013-05 \ FEM for Advanced Processing 20

Summary and next lectures
1. Equilibrium in strong (SE) and weak (WE) form
2. Weighted residuals
3. FEM for solids in 1-D:
i. Write WE
ii. Subdomain partitioning
iii. Shape functions (interpolation)
iv. Finite Element System assembly
v. Solution of the algebraic WE (with partitioning)
1 1 2 2 3 3 4

Next: Practical on Friday 23rd April: Implement 1D FEM (MATLAB)

Lecture on Wednesday 28th April: 3D FEM in small deformations

WMME013-05 \ FEM for Advanced Processing 21