ME 478 FINITE ELEMENT METHOD

Chapter 8. Other Cool Stuff

LINEAR ELASTO-DYNAMICS
&& + Ku = fε o + fb + ft + P
Mu
where
u ( x, y, t ) = Nu
ε ( x, y, t ) = Bu
σ ( x, y, t ) = D(Bu − ε o )
and the stiffness and mass matrices are:
K = ∫∫∫BT DBdv M = ∫∫∫NT ρNdv
V V
Note that M and K have the same form (due to connectivity)

MODAL ANALYSIS
We will start by looking at the free vibration problem
Mu&& + Ku = 0 (f = 0)
) )
u (t ) = u sin(ωt + θ ) where u
let are the mode shapes, then
by substitution, we have:
)
(K − ω 2M )u = 0
for a nontrivial solution to exist, we require
det(K − ω M ) = 0
2

from which we can solve for the frequencies and then the
corresponding mode shapes.

8.1
EXAMPLE: 1-D BAR 2
Recall that we had:
u 
u ( s ) = [N1 N 2 ] 1 
u2 
where
1− s 1+ s 1
N1 = N2 =
2 2
and the strains are related to the displacements through:
 − 1 1   u1 
εs =     = Bu
 L L  u2 
And the stress/strain relation as:
σ s = EB u
We apply Newton’s second law of motion to the 2 nodes
F = ma
∂ 2u1
f1ext − f1int = m1 2
∂t
∂ 2u 2
f 2 ext − f 2 int = m2 2
∂t
AL AL
where 1 m = ρ m = ρ
2 and 2 (called the lumped mass)
1

Writing out the equations, we have:

 f1ext   f1int  m1 0  u&&1 
 = +   u&& 
 2 ext   2 int  
f f 0 m2  2 

Consistent Mass matrix

Starting with d’Alembert’s principle
Χe = −ρu
&&(x, y)
8.2
 e
where Χ is the effective body force due to the mass of the
element. We can then convert this body force to nodal forces
through
fb = ∫∫∫NT Xe dv
V

Making the substitution for Χe and knowing that &&(x, y) = Nu

u && ,
we have
fb = ∫∫∫NT ρNdvu
&&
V
ELASTODYNAMICS INCLUDING INELASTICITY
Starting with:
&& + Ku = f
Mu
s
where
s = ∫∫∫ B T σ dv
V

Now including rate effects σ = Dε + Vε&

Therefore
s = ∫∫∫ B T DB dvu + ∫∫∫ B T VB dvu&
V V
So we have
&& + Cu& + Ku = f
Mu

8.3
SPECIAL CASES
1) No damping-Linear elasto-dynamics
&& + Ku = f
Mu
2) No inertial effects-elasto-statics
Ku = f
3) Quasi-static-Visco-elasto statics
Cu& + Ku = f
4) Viscous flow (Newtonian fluids)
Cu& = f
RHEOLOGICAL MODELS
σ Hookian σ = Eε
ε

σ Newtonian σ = ηε&
ε&

Parallel Arrangement (Voigt-Kelvin Model)

σ ε

t t

ε σ

t t

Serial Arrangement (Maxwell model)

σ ε

t t

ε σ

8.4
During the creep response, the material represented by these
models undergoes
1. Initial instantaneous response
2. Non-linear delayed elastic response
3. Instantaneous elastic recovery
4. Delayed elastic-viscoelastic recovery
5. Permanent deformation

For the Maxwell model, the relationship between the deflection and
the applied load is:
P& P P ( 0)
u& = + , u ( 0) =
k η k
The resulting creep function for a unit step is:
1 1 
c (t ) =  + t U (t )
k η 
and the relaxation function is:
r (t ) = ke − ( k /η ) tU (t )
For the Voigt model (slightly different than above), the relationship
between the deflection and the applied load is:
P = ku + ηu& , u ( 0) = 0
The resulting creep function for a unit step is:
1
( )
c (t ) = 1 − e −( k /η ) t U (t )
k
and the relaxation function is:
r (t ) = ηδ (t ) + kU (t )

Viscoelastic Overstress – Delayed Elasticity

This element is associated with some threshold condition
σy(εp)

8.5
 ELASTO-PLASTICITY
(we will look at this in more detail later)

Total Format (Henke)

E ε = εe + ε p

σy(εp) Rate Format (St Venant VonMises Prandtl-Reuss)

ε& = ε& e + ε& p

And our stress-strain relation becomes: σ& = D &⋅ ( ε& − ε& p )

σ

E
ε
εp εe
ELASTO-VISCO-PLASTICITY
σ ET ∂σ E Ep
ET = =
∂ε E + E p
E
ε

σ Ep ∂σ
Ep =
∂ε p
εp
This is a combination of the three rheological models. Solution
procedure is much more complicated than for linear elasticity

8.6
 Example One Dimensional Viscoplasticity
Consider the basic one-dimensional
viscoplastic model. The total strain n the
model can be expressed as the sum of the
elastic and the viscoelstic components as:
ε = ε e + ε vp
The applied stress is related to the elastic
strain by:
σ e = σ = Eε e
The stress in the dashpot is related to the
viscoplastic strain by:
σ d = ηε& vp
And the stress in the friction slider is:
σ p =σ if σp <Y
σ p =Y if σp ≥Y
where Y represents the threshold stress which is a function of
some yield stress and some strain hardening as:
Y = σ y + H ' ε vp
Prior to the onset of viscoplastic yielding, ε vp = 0 giving σd = 0
thus σ p = σ . Combining stresses in the dashpot and the friction
slider gives:
σ = σ y + H ' ε vp + ηε&vp
Using ε vp = ε − ε e andσ e = σ = Eε e gives
H ' Eε + ηEε& = H ' σ + E (σ − σ y ) + ησ&
which is a first order ODE, Rearranging, we get

8.7
 σ& 1
ε& = + [σ − (σ y + H ' ε vp )]
E η
or
ε& = ε&e + ε&vp
Considering the case when we apply a constant stress to the model
H ' Eε + ηEε& = H ' σ A + E (σ A − σ y )
The solution is:
σA (σ A − σ y )
ε (t ) = + [1 − e −H 't / η ]
E H'
Note: the solution to ODE’s of the form y '+ p y = r is
y (t ) = e − h ∫ e h rdt + C where: h = ∫ pdt
In the case of a perfectly viscoplastic material, ( H ' = 0 ) then we
have (by applying L’Hopital’s rule):
σA (σ A − σ y )t
ε (t ) = +
E η
Viscoplasticity is a transient phenomena, thus the solution involves
taking a time incremental (time-stepping) approach. The simplest
approach is to use Euler’s rule where we extrapolate the value at
some time tn+1 in terms of the quantities at time t.

Using this approach, we can define the viscoplastic strain

increment over time step ∆t n = t n +1 − t n as:
∆ε vpn = ε&vpn ∆t n
The change in length of the element due to the strain increment is:
∆u n = ∆ε vpn L

8.8
 and adding this to the change in length due to the applied loading
gives:
L
∆u n = ∆ε vpn L + ∆Pn
AE
and rewriting in matrix form gives:
∆u1n 
∆u =  n  = K −1∆Fn
n

∆u 2 
Where
1
∆Fn = AEε& ∆t n   + ∆Pn
n

− 1
vp

and
EA  1 − 1
K=
L − 1 1 
Rewritting in the standard form
K = ∫ B T DBdv
V

∆Fn = ∫ B T Dεdv + ∆Pn

V
The updated displacements at obtained as:
u n +1 = u n + ∆u n
The stress increment is:
∆σ n = E∆ε en = E ( ∆ε n − ∆ε vpn )
 ∆u1n − ∆u 2n 

= E − ε vp ∆t n 
& n

 L 
The stress at time tn+1 is:

8.9
 σ n +1 = σ n + ∆σ n
The viscoplastic strain is:
ε vpn +1 = ε vpn + ∆ε vpn
And lastly, the viscoplastic strain rate is:
1
ε&vpn +1 = [σ n +1 − (σ y + H ' ε vp
n +1
)]
η
The out of balance residual forces as expressed as the sum of the
applied nodal loads and the nodal forces equivalent to the
elemental stresses are:
1
Ψn +1 = Aσ n +1
  + Pn +1
− 1
These residual forces are added to the pseudo forces to give the
forces for the next time increment as:
1
∆Fn +1 + = AEε vp ∆t n +1   + ∆Pn +1 + Ψn +1
& n +1

− 1
This step is repeated until the solution for the desired time duration
or steady state conditions are achieved (when the viscoplastic
strain rate becomes negligible).

There is a limit on the time step that one can use for the
viscoplastic solution. For the one-dimensional case considered
here, the limiting value is (Cormeau proposed this one and there
are many different values that have been proposed):
ησ y
∆t ≤
E

8.10
 Computational Implementation of 1-D
Viscoplasticity
STEP 1 At time t=tn, we compute that following quantities using the
standard approach
σ n, ε n, ε vpn , f n , and u n known
and compute the viscoplastic strain rate for each element as:
1
ε&vpn = [σ n − (σ y + H ' ε vp
n
)]
η
STEP 2 Compute the displacement increment according to:
∆u1n 
∆u =  n  = K −1∆Fn
n

∆u 2 
where
1
∆Fn = AEε&vpn ∆t n   + ∆Pn
− 1
and for each element, the stiffness matrix is
EA  1 − 1
K=
L − 1 1 
STEP 3 Compute the stress increment and the viscoplastic strain
increment for each element as:
 ∆u1n − ∆u 2n 
∆σ = E 
n
 − ε vp ∆t n 
& n

 L 
∆ε vpn = ε&vpn ∆t n

8.11
STEP 4 Determine the total displacements, stresses and the
viscoplastic strains
u n +1 = u n + ∆u n
σ n +1 = σ n + ∆σ n
ε vpn +1 = ε vpn + ∆ε vpn
STEP 5 Compute the Viscoplastic strain rate for each element for
the next time step
1
ε&vpn +1 = [σ n +1 − (σ y + H ' ε vp
n +1
)]
η
STEP 6 Evaluate the residual forces by applying equilibrium
correction to each element
n +1 1
Ψn +1 = Aσ   + ∆Pn +1
− 1
and add these into the vector of pseudo nodal loads to be used in
the next time step
1
∆Fn +1 = AEε&vpn +1∆t n +1   + ∆Pn +1 + Ψn +1
− 1
STEP 7 Check if the viscoplastic strain rate in each element has
become negligibly small. If so, then steady state conditions are
said to have been reached. If not, return to STEP 1 and repeat the
entire process for the next time step. One way to check for
convergence is as follows:

∑ (∆ε )
M
n
vp i
i =1
× 100 ≤ TOLERANCE
∑ (∆ε )
M
1
vp i
i =1

8.12