You are on page 1of 15

Part I1

Goals of this lecture


Understand the relative complexity between time-
dependent FEA and static FEA.

Part II Time Integration Schemes


Understand the difference between explicit and implicit time
integration.
Fundamental! All commercial codes are either explicit or
implicit (most can do both). They are fundamentally for
modelling very different problem types.
Associate time integration scheme with problem type.
Understand, based on the problem you are trying to solve, when
you should choose an explicit code, and when you should choose
an implicit code.

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 2


Discretizing Time
Just as we discretize space into finite elements, so must we
discretize time into discrete timesteps.
We do this so that we can determine nodal displacements at
different time increments.
Stepping through time in a finite element simulation is called time
integration.
There are two types of direct time integration:
Explicit time integration
Implicit time integration

A simple time discretization scheme with equal length time steps is:
= 0 , 0 + , 0 + 2, , 0 +
where is called the timestep

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 3


Explicit Time Integration
For explicit time integration, we can define values explicitly as functions of known parameters.
For explicit schemes the equation of motion is evaluated at the old timestep,
Lets develop this idea with an example:
Velocity
= ,
Using simple time discretization scheme (i.e. equal timesteps) from the previous slide, we can say:
= ,
Lets use the following simple scheme (Eulers Method) to approximate a derivative:
+1
=


Since = , we can say:

+1
=

We can now solve explicitly for the unknown +1
+1 = +

To put this equation to use, we need initial conditions, 0 = and 0 = :


0 : 0 = , 0 =
1 : 1 = 0 + 0
2 : 2 = `1 + 1

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 4


Implicit Time Integration
For implicit time integration:
For implicit methods the equation of motion is evaluated at the new timestep +1 .
Like before, lets develop this idea with an example:
Velocity (Again)
= ,
Using simple time discretization scheme (i.e. equal timesteps) from the previous slide, we can say:
= ,
Lets use the following simple scheme (Backward Eulers Method) to approximate a derivative:
1
=


Since = , we can say:

1
=

Lets reindex to get:
+1
= +1

Solving for +1 gives:
+1 = +1 +
Notice that +1 appears on both sides of the equation (i.e. +1 on the left and its time-derivative, +1 , on
the right). I.e. this is an implicit equation.
Putting this equation to use may require advanced solution techniques; depending on the problem type.

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 5


Stability
Stability refers to the response of the system to
disturbances.
If the response is characterized by unbounded oscillations,
it is unstable.
If it maintains steady-state, or monotonically decays, it is
stable.
Explicit time integration schemes are only stable for very
small .
The particular stable value of will be discussed later.
Implicit time integration schemes are unconditionally stable.
Which means large are okay.
That doesnt mean that large are accurate

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 6


Example for Gravitational
Acceleration
To illustrate these concepts, lets do an example thats easy to validate,
gravitational acceleration.
Question: What is the position and velocity of an object dropped from = 0
at = 0.1 []. Note: positive direction is down; initial velocity is zero.
Analytical solution for position:
1
= 0 + 0 + 2
2

Apply initial conditions: 0 = 0, 0 = 0, = 9.81 2

1
= 0 + 0 0.1 + 9.81 0.1 2 = 0.04905
2
Analytical solution for velocity:
= 2 0 + 02

Apply initial conditions: 0 = 0, 0 = 0, = 9.81
2
m
= 2 9.81 0.04905 0 + 0 = 0.981
s
Why is the following incorrect?
- We know = 0.04905
- We know = 0.1
0.04905
- = = 0.1 = 0.4905 [m/s]
Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 7
Explicit Euler Method Solution Explicit Method
t x xdot xdotdot

We set up a table like this 0.0005


0 0 0
0 0.004905
9.81
9.81
0.001 2.45E-06 0.00981 9.81
0.0015 7.36E-06 0.014715 9.81
Its common to start with the highest unknown derivative
Velocity in this case (as acceleration is constant and given)
Since acceleration is the derivative of velocity, we can use the
Euler Method as defined before, except for and (instead
of and ), so:
+1 = +
And as before, for displacement, we use:
+1 = +
We then choose a , and calculate however many iterations
we need to get to = 0.1

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 8


Implicit Backward Euler Method
Solution Implicit Method
t
0
x
0
xdot
0
xdotdot
9.81
We set up a table like this 0.0005 2.45E-06 0.004905
0.001 7.36E-06 0.00981
9.81
9.81
0.0015 1.47E-05 0.014715 9.81

Again, start with the highest unknown derivative


Velocity in this case (as acceleration is constant and given)
Again, similar we use and (instead of and ), but
this time with the Backward Euler Method so:
+1 = +1 +
And as before, for displacement, we use:
+1 = +1 +
We then choose a , and calculate however many
iterations we need to get to = 0.1

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 9


Compare Results
Simple Implicit Method Results Simple Implicit & Explicit Numerical and
Analytical Solutions to Gravitational
Consistently overestimates Acceleration of a Body
displacement. Implicit Method Analytical Solution Explicit Method

Error increasing over time 0.00016

0.00014

Simple Explicit Method Results 0.00012

Consistently underestimates

Displacement [m]
0.0001

displacement. 0.00008

Error increasing over time 0.00006

0.00004

0.00002

0
0 0.001 0.002 0.003 0.004 0.005 0.006
Time [s]

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 10


Better Explicit Method
The Central Difference Method
Explicit Method
Acceleration and Velocity
+1 1
=
2

Acceleration and Displacement


+1 2 + 1
=
2

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 11


Calculating for the Central
Difference Method
Theoretically, the timestep for the Central Difference Method must be less than:
2


3
Practically, it must be at most, of this value, such that:
4
3 2

4
It is common for Explicit analysis to calculate the stable timestep in terms of the mesh size and the
material properties.

=

where is the shortest straight path through an element
is the speed of sound of the material

E.g. for a steel rod of length, = 0.01


207 9
, = = = 5135.1
7850
0.01
= = 1.95 6
5135.1

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 12


Courant-Friedrichs-Lewy Criterion:
Explicit Time Integration
Maximum Stable Timestep,
, , ,
= = =

3 1 2

where: is the maximum stable timestep for an explicit time integration solution
, is the smallest element edge length in the entire finite element mesh
is the speed of sound in a particular materal

=

where: is the elastic bulk modulus of the material
is the density of the material

=
3 12
where: E is the Youngs modulus of the material
is the Poissons ratio of the material

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 13


Better Implicit Method
Newmark Method
+1 = + 1 + +1
2
1
+1 = + + + +1
2
where and are user chosen parameters
Newmark parameter guidance
1 1
= , = Yields unconditionally stable implicit method.
2 4

Choosing a timestep of = usually yields good results.
20
Where is the period of vibration of the behaviour you wish to
model
1
= , = 0 Yields the explicit central difference method
2

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 14


End of Lecture Part 2

Winter 2017 ENGI 9094 - Advanced Numerical Structural Analysis 15

You might also like