MODULE 3

CIV5142Z

Introduction to Finite Element Modelling

in Structural Analysis
Advanced Finite Element Modelling

Prof Sebastian Skatulla

Computational Continuum Mechanics Group (CCM)

Department of Civil Engineering
University of Cape Town

2024

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems

Non-linear initial boundary value problem

A non-linear solid mechanics problem consists of a coupled system of non-linear partial
differential equations:
in the Lagrangian form

Kinematics: C = FT F
Governing equation: ρ0 ẍ = Div (FS) + b in B
∂ψ
Constitutive equation: S=2
∂C
Boundary conditions: Pn = t̂(n) on ∂BN , u = û on ∂BD .
in the Eulerian form
Kinematics: b = FFT
Governing equation: ρ ẍ = div σ + bt in Bt
∂ψt
Constitutive equation: σ = 2b
∂b
(ν)
Boundary conditions: σν = t̂t on ∂Bt,N , u = û on ∂BD,t .

This problem representation is referred to as the strong form.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Non-linear initial boundary value problem
In general, it is not possible to find an analytical solution of this differential
problem.
The problem is solved numerically in an approximate sense, e.g. using the finite
element method, the finite difference method or the meshfree method.
These numerical methods are based on solution techniques such as the principle
of virtual work, variational principles or the weighted residual method.

Weighted residual method

In the weighted residual method the governing equations - here the equilibrium
equations - are cast into integral form and multiplied with a test function/weight
function:
Z   Z
FSn − t̂(n) · δu dA = 0


Div (FS) + b · δu dV +
B ∂BN

since δu is arbitrary, the equilibrium equations in the integrand must hold.

Further mathematical manipulations which involve integrating by parts and
assuming δu = 0 on ∂BD , we arrive at:
Z Z Z
FS : δF dV − b · δu dV − t̂(n) · δu dA = 0
B B ∂BN

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Weighted residual method cont’d

The final integral statement is referred to as weak form, because the original
2nd -order differential problem (in the primary variable u) has been transferred to
a 1st -order one.
In the weak form the force equilibrium is not directly enforced any more and it
poses less strong requirements to the continuity/differentiability of the solution,
i.e. the approximation of u must be only once differentiable.
Reducing the order of continuity of the approximation of the primary unknown,
which is usually displacement field, from twice to only once differentiable
(so-called C 1 -continuous) is particularly important in FEM. Albeit, the FEM
approximation can achieve higher than C 1 continuity locally on element level, it is
globally only C 0 continuous.
Deformation-dependent traction force loading such as pressure refers to the
deformed configuration Z
p ν · δu da
∂BN ,t

can be transferred to undeformed configuration via Nanson’s formula

Z
p JF−T n · δu dA .
∂BN

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Principle of virtual work

The principle states that "the total virtual work done by all the forces acting on a
system in static equilibrium is zero for a set of infinitesimal virtual displacements
from equilibrium":
δΨ(int) − W(ext) = 0
where the internal virtual (elastic) strain energy in the Lagrangian form reads as
Z Z
δΨ(int) = S : δE dV = FS : δF dV
B B

and the external virtual work as

Z Z
W(ext) = b · δu dV + t̂(n) · δu dA .
B ∂BN

The virtual work is the work done by the virtual displacements, which can be
arbitrary, provided they are consistent with the constraints of the system, i.e.
δu = 0 on ∂BD .
Clearly, the solid mechanics problem is equivalently formulated via the principle of
virtual work and the weighted residual method.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Displacement-based variational principle
In case of a hyper-elastic material, a so-called strain energy function, ψ(E), exists
which describes the elastic energy stored in a solid body.
The equilibrium equation is fulfilled for this displacement field u which makes out
of all possible solutions the following functional stationary:

F = Ψ(int) + Ψ(ext) = stationary

where the internal elastic potential in the Lagrangian is expressed as

Z
Ψ(int) = ψ(E) dV
B

and the external potential as

Z Z
Ψ(ext) = − b · u dV − t̂(n) · u dA .
B ∂BN

This leads to minimization of the functional involving a directional derivative, the

so-called first variation of F :
∂F
δF = · δu = δΨ(int) + δΨ(ext) = 0
∂u

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Displacement-based variational principle cont’d

The evaluation of the variation of internal and external potentials gives:
Z Z Z
δF = S : δE dV − b · δu dV − t̂(n) · δu dA = 0
B B ∂BN

∂ψ ∂ψ
with δψ = : δE and S = .
∂E ∂E
Equivalently, if the strain energy is formulated in terms of the right Cauchy-Green
deformation C:
Z Z Z
1
δF = S : δC dV − b · δu dV − t̂(n) · δu dA = 0
B
2 B ∂BN

∂ψ ∂ψ
with δψ = : δC and S = 2 .
∂C ∂C
Note: In this context the definition
∂(•)
δ(•) := δu
∂u
refers to the variation with respect to primary variable, u.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Linearization of displacement-based non-linear variational principle
For the general case of finite strain and non-linear material behaviour, the
previously introduced variational principles need to be solved incrementally and
iteratively.
For this, at each iteration step, i, of an incremental loading or time step, n, the
variational functional δF (u) is linearized using a first-order Taylor expansion in
the vicinity of some known solution of the displacement field, ui−1n which yields



∂(δF )
δF uin = δF ui−1
n + ∆u = δF ui−1
n + | i−1 ∆u ≈ 0 ,
∂u un
where ∆u is the incremental displacement field such that the unknown
displacement field in the current iteration step

uin = ui−1
n + ∆u .
Similar to the variation of stress and strain we define
∂(•)
∆(•) := ∆u
∂u
and the linearized variational formulation reads as
Z Z
∂S
δF = S : δE dV + ∆E : δE + S : ∆δE dV − Wext = 0 .
B B
∂E

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Linearization of displacement-based non-linear variational principle cont’d
For computational purposes the variation and the linearization of the Green strain
tensor as well as the linearization of the second Piola-Kirchhoff stress tensor
needs to be evaluated in index notation:
1 1
δEij = ∆ (Fri Frj ) = (δur ,i Frj + Fri δur ,j )
2 2
1
1
∆Ekl = ∆us,k Fsl + Fsk ∆us,l , ∆δEij = (δur ,i ∆ur ,j + ∆ur ,i δur ,j )
2 2
∂Sij ∂ψ
∆Sij = ∆Ekl = ∆Ekl = Hijkl ∆Ekl
∂Ekl ∂Eij Ekl

Then, the linearized variational principle is given in index notation:

Z Z
δF = Fri Sij δur ,j dV + Hijkl Fsk ∆us,l Fri δur ,j + Sij ∆ur ,i δur ,j dV
B B
Z Z
(n)
− bi δui dV − t̂i δui dA = 0 .
B ∂BN

making use of the symmetry of S and the Hessian matrix /tangent operator:

Hijkl = Hjikl = Hjilk = Hijlk .

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Analysis of non-linear solid mechanics problems cont’d

Linearization of displacement-based non-linear variational principle cont’d
Clearly, at the outset of the incremental solving process where u = 0, we find
F = 1, E = 0, ψ(E) = 0 and S = 0 and thus, the linearized variational principle
reduces to:
Z Z Z
(n)
δF = Hijkl ∆uk,l δui,j dV − bi δui dV − t̂i δui dA = 0 .
B B ∂BN

If we consider only small strains, i.e.

1 1 1
Eij ≈ (ui,j + uj,i ), δEij ≈ (δui,j + δuj,i ), ∆Ekl ≈ (∆uk,l + δul,k ), ∆δEij ≈ 0
2 2 2
and linear elastic material
1
Sij = Hijkl Ekl ≈ Hijkl (uk,l + ul,k ) = Hijkl uk,l
2
then we get essentially the same as above:
Z Z Z
(n)
δF = Hijkl (uk,l + ∆uk,l ) δui,j dV − bi δui dV − t̂i δui dA = 0
B B ∂BN

which is linear functional in the displacement and its derivatives.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

FEM approximation of a displacement-based variational principle cont’d
For the purpose of discretizing the variational principle we consider FEM
approximations of the displacement field, its increment, the virtual displacement
field as well as their spatial derivatives:
nen nen nen
X X X
ui ≈ N I uiI , δui ≈ N I δuiI , ∆ui ≈ N I ∆uiI
I I I
nen nen nen
X ∂N I X ∂N I X ∂N I
ui,j ≈ uiI , δui,j ≈ δuiI , ∆ui,j ≈ ∆uiI .
∂Xj ∂Xj ∂Xj
I I I

Note: The derivatives of FEM interpolation functions are computed with respect
to the undeformed configuration.
The element-wise evaluated linearized variational formulation is rewritten as
XZ ∂N I
δF = Fri Sij δurI dV
Be
∂Xj
e
XZ ∂N J ∂N I ∂N J ∂N I
+ Dijkl Fsk ∆usJ Fri δurI + Sij ∆urJ δurI dV
Be
∂Xl ∂Xj ∂Xi ∂Xj
e
XZ XZ (n)
− bi N I δuiI dV − t̂i N I δuiI dA = 0 .
e Be e ∂Be
N

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

FEM approximation of a displacement-based variational principle cont’d
The integration of the discretized/approximated variational principle results in a
so-called discrete equation system of the following form for each element:
 i  i
[K]in [∆u]in = f(ext) − f(int) = [r]in ,
n n

where the external force vector, the internal force vector and stiffness matrix are
expressed as Z Z
(n)
[I·3+i] =
e
f(ext) bi N I dV − t̂i N I dA
Be ∂Be
N
Z
∂N I
e
f(int) [I·3+i] = Fri Sij dV
e ∂Xj
ZB
∂N J ∂N I
[I·3+r ][J·3+s] =
e
KM Hijkl Fsk Fri
e ∂Xl ∂Xj
ZB
∂N J ∂N I
KGe [I·3+r ][J·3+r ] = Sij dV ,
Be
∂Xi ∂Xj

respectively, and [r] denotes the residual vector.

Rows and columns of the discrete equation system refer to the nodal numbers
I, J = 1, num. of nodes and their 3 nodal displacement degrees of freedom each.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

FEM approximation of a displacement-based variational principle cont’d

The stiffness matrix/tangent can be additively split into material and geometric
stiffness parts, Ke = KeM + KeG with:
Z
∂N J ∂N I
KeM = Hijkl Fsk Fri dV
e ∂Xl ∂Xj
ZB
∂N J ∂N I
KeG = Sij dV
Be
∂Xi ∂Xj

where the geometric stiffness arises from the non-linear strain-displacement

relations, i.e. E.
The stiffness matrix resulting from this standard non-linear variational
formulation is symmetric.
However, the incorporation of the pressure loading provides additional
contributions to the stiffness matrix renders the latter unsymmetric. In order to
be able to still use a linear equation solver for symmetric matrices one can to
symmetrize the stiffness matrix:
1

K= K + KT .
2

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

Incremental and iterative solving of non-linear variational principles

Fig.: Non-linear load-displacement diagram.

For a general non-linear mechanics problem of the form
   
f(ext) − f(int) = 0
the load-deformation or stress-strain solution path is non-linear and potentially
also path-dependent.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

Incremental and iterative solving of non-linear variational principles cont’d
The linearization of the previously stated non-linear problem statement
   
[K] [∆u] = f(ext) − f(int) = 0
only convergences to a solution, if the latter is close enough to a previous
equilibrium satisfying solution such that the linearization applies.
Consequently, the external loads P ∈ {b, t̂(n) } must be ramped up in very small
increments.
The loading magnitude at a loading step or time step n is therefore expressed as

Pn = λn P (0)

where λn denotes the current loading factor associated with the set of
reference/target loads P (0) .
The loading factor is incrementally updated via

λn = λn−1 + ∆λn ,

where ∆λn is the loading factor increment for the current loading step.
Also, for small loading factor increments ∆λn , the non-linear nature of the
problem necessitates an iterative solution procedure to find a converged solution.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

Incremental and iterative solving of non-linear variational principles cont’d

Fig.: Newton-Raphson method.

The Newton-Raphson method is the fastest converging iterative solution

procedure, if the initial solution is close to the unknown solution, i.e. within the
zone of attraction.
In this case, the rate of convergence is up to a quadratic order as illustrated
schematically above.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

Incremental and iterative solving of non-linear variational principles cont’d

Thus, the discrete equation system takes the following form:
m  h ij 
(0)
X i
λjn f(ext)

[K]in [∆u]in = − f(int) = [r]in ,
n
j

considering m sets of external loads, each of them associated with a

(0)
corresponding reference external force vector f(ext) referring to P (0) .
The stiffness matrix, the internal force and the residual vector are evaluated at
the last known/solved displacement field:

[u]n−1 , for i = 0 (converged solution of the previous loading step)
[u]in =
[u]i−1
n , for i > 0 (solution of previous iteration step)

The loading application is usually prescribed via a given loading function and the
loading factor increments ∆λn are therefore known for each loading step n and
held fixed while the displacement field degrees of freedom uin are iteratively
determined.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

Incremental and iterative solving of non-linear variational principles cont’d

Accordingly, the discrete equation system is solved for an unknown displacement
increment or correction, ∆uin , for each iteration step i via

−1
[∆u]in = [K]in [r]in .

The displacement field vector [u]in is then additively updated at the end of each
iteration step using
[u]in = [u]i−1
n + [∆u]in .

The iterative procedure is terminated when the norm of the recomputed residual
vector using the updated displacement field || [r]in || < ϵs where ϵs is a
user-defined threshold.
The general objective in this iterative approach is to choose the loading factor
increments ∆λn small enough that a converged solution can be found within a
few iteration steps. This implies that the unknown solution is in the close vicinity
of the last known solution.

Prof S. Skatulla CIV5142Z - Introduction to FEM

 MODULE 3 Nonlinear Finite Element Method

Numerical analysis of non-linear solid mechanics problems cont’d

Tab.: Flow chart of incremental iterative solving procedure.

i=0
• Set the initial nodal displacement DOF: [u]n=1 =0
• Loop over all incremental loading/time-steps n = 1, . . .
• Loop over Newton-Raphson iterations i = 1, . . . for current loading or time step n
 
1 Assemble the external force vector f(ext) .
n
 i
2 Assemble the internal force vector f(int) .
n
3 Assemble the stiffness matrix [K]in .
   i
4 Compute the residual vector [r]in = f(ext) − f(int) .
n n
5 Assemble the discrete equation system of the whole system [K]in [∆u]in = [r]in and
solve it for the displacement correction [∆u]in .
6 Update the nodal displacement DOF.
7 Check for convergence, i.e. terminate iteration if ∥ [∆u]in ∥ < ϵs and proceed with
the next incremental loading step or continue with next iteration step (resume with
1.)

Prof S. Skatulla CIV5142Z - Introduction to FEM