lecture7 fe modeling topics part ii€¦ · mae 323: lecture 7 modeling topics: part ii 2011 alex...
TRANSCRIPT
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 1MAE 323 Lecture 7 FE Modeling Topics: Part 2
Modeling Topics
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 2MAE 323 Lecture 7 FE Modeling Topics: Part 2
Constraints
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 3MAE 323 Lecture 7 FE Modeling Topics: Part 2
•So far, we have seen how boundary conditions of the form ui=0, vi=0, θi=0 may
be applied to solve algebraic structural system equations.
•The setting of nodal DoFs to prescribed values in this manner falls under a
broader category of operations referred to generally as constraints
• A basic feature of FEM is that the system equations as generated do not
automatically contain such constraints within them*
•Frequently, it becomes necessary to specify more complicated constraints than
those we have already seen. So let’s start by seeing what these might look like
*In the mathematical derivations of FE solutions to differential equations, this difficulty is
overcome by specifying that the shape functions must be equal to boundary values at
boundaries.
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 4MAE 323 Lecture 7 FE Modeling Topics: Part 2
•In general, constraints involve assignments to, or relations
among DoFs which must not be violated during solution.
•Such a relation might look like this:
2 34 8 8u u− =
•We might have a constraint that “ties” a rotational DoF of a
beam to a solid, for example:
beam
Solid
L L
1 2
1 2 2 sinz
u u L θ− = ⋅
u1 u2
θz
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 5MAE 323 Lecture 7 FE Modeling Topics: Part 2
•All such relations have the same basic form:
⋅ =C u q (1)
•In building finite element models, such relations are extremely
useful. So, we’d like some general procedure of handling such
relations once we have assembled the global system equations:
⋅ =K u f
•There are three common methods of handling such relations
in a general way in FEM. They are:
1. Direct Constraint Elimination
2. Methods Based on Lagrange Multipliers
3. Penalty-Based Methods
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 6MAE 323 Lecture 7 FE Modeling Topics: Part 2
[ ] r
r c
c
=
u 0C C
u 0
Here ur and uc are ,respectively, DoFs to be retained and DoFs
to be eliminated. Because there are as many DoFs of uc as
there are independent equations of constraint, matrix Cc is
square and non-singular. Solution for uc yields:
c rc r= ⋅u C u
Where
1
rc c r
−= − ⋅C C C
(2)
(3)
(4)
Consider the common case where q=0:
1. Direct Constraint Elimination
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 7MAE 323 Lecture 7 FE Modeling Topics: Part 2
We now combine the identity ur=ur and (4):
r
r
c
= ⋅
uT u
uwhere
rc
=
IT
C(5)
Now, if we write the constraint-
reduced equations as:
' ' '⋅ =K u f
Then the transformation , T (5) may be applied
to the original partitioned system:
rr rc r r
cr cc c c
⋅ = = =
K K u fK u f
K K u f
1. Direct Constraint Elimination
(6)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 8MAE 323 Lecture 7 FE Modeling Topics: Part 2
where:
'
'
T
T
= ⋅ ⋅
= ⋅
K T K T
f T f
•After (6) is solved for ur, Equation (3) yields uc. If q is not equal
to zero in (1), additional terms appear in the RHS of (6).
1. Direct Constraint Elimination
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 9MAE 323 Lecture 7 FE Modeling Topics: Part 2
•If Eq (2) simply sets certain DoFs (uc) to zero, then Cr=0 and
Cc=I. Hence Crc =0 is this case and Eq (6) is equivalent to
discarding rows and columns associated with uc
•Otherwise, the choice of which DoFs to place in uc is not
unique, so the choice of Cc is not unique. One might then
define Cc to be the last c linearly independent columns of C
•It is possible to apply constraints (1) serially without
partitioning and applying equation (6). All DoFs are retained
in such an approach, but the resulting system may not be
positive definite
1. Direct Constraint Elimination
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 10MAE 323 Lecture 7 FE Modeling Topics: Part 2
Example 1
x
y
u2,F2
u3,F3
k2
k1
k3
1
2
3
Let’s take the three-spring system introduced in Chapter
2 and apply equal forces to all three nodes
u1,F1
•In addition, let k1=k2=k2
•Apply forces at nodes 1, 2, and 3
such that F1=F2=F3=F
1 1
2 2
3 3
2 1 0
1 2 1
0 1 1
u F
k u F
u F
−
− − = −
From Chapter 2, we know the original
governing equation is:
•Now, let’s impose the
constraint: u2=u3(7)
1. Direct Constraint Elimination
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 11MAE 323 Lecture 7 FE Modeling Topics: Part 2
…so that:
Example 1
3c u=u
[ ]1
2
3
0 1 1
u
u
u
− = ⋅
C u
[ ]0 1
1
r
c
=
= −
C
C
…and:
[ ]1 0 1rc c r
−= − ⋅ =C C C
(8)
(9)
(10)
1. Direct Constraint Elimination
Let
…which results in the partition:
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 12MAE 323 Lecture 7 FE Modeling Topics: Part 2
At this point, we could either plug into the resultant
form (see Chapter 7) explicitly, OR carry out the
calculation:
Example 1
•The transformation, T is given by:
Either way, we get the reduced system:
(11)
(12)
1 0
0 1
0 1rc
= =
IT
C
'
'
T
T
= ⋅ ⋅
= ⋅
K T K T
F T F
2 1'
1 1
1'
2
k
F
− = −
=
K
F
1. Direct Constraint Elimination
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 13MAE 323 Lecture 7 FE Modeling Topics: Part 2
Example 1
•We can now solve for ur:
13
' '5
r
F
k
− = ⋅ =
u K f
…and we can use (3) to solve for uc:
[ ]3
30 1 5
5c rc r
F Fu
k k
= = ⋅ = ⋅ =
u C u
…and the original constraint condition u2=u3 is
immediately verified by inspection.
1. Direct Constraint Elimination
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 14MAE 323 Lecture 7 FE Modeling Topics: Part 2
2. The Method of Lagrange Multipliers
•The method of Lagrange multipliers is simple enough to state in
words: In this method, any constraints placed on a system (such
as (1)) must not contribute to the total potential energy of the
system
•To see how this statement is implemented, we start by writing
both the original governing equation AND the constraints
0
0
⋅ − =
⋅ − =
K u f
C u q
(13)
•Now, use (13) to express the total potential energy of the
constrained system:
( ) 0T T T
pΠ = ⋅ ⋅ − ⋅ + ⋅ ⋅ − =u K u u f λ C u q (14)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 15MAE 323 Lecture 7 FE Modeling Topics: Part 2
2. The Method of Lagrange Multipliers
…where λλλλ represents unknown coefficients with units of force
required to enforce (14).
•Equation (14) may be interpreted as stating that the
requirement that constraints not change the total potential
energy (do no work) entails that there be forces, λλλλ associated
with such constraints . Furthermore, these forces must be solved
explicitly
•We must find a linear form the equations embodying the energy
expression (14) which also includes the variables, λλλλ. This can be
done by applying a generalized form of Castigliano’s Second
Theorem:
p
p
∂Π=
∂
∂Π=
∂
0u
0λ
(15)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 16MAE 323 Lecture 7 FE Modeling Topics: Part 2
•Applying (15) to (14) yields:
T =
u fK C
λ qC 0(16)
•It is apparent that applying (16) increases the matrix
system size by as many rows and columns as there are
constraints
•However, there are situations where application of (16) is
more advantageous than direct constraint elimination. One
such situation is when a single DoF is constrained to many
DoFs. Other situations involve the application of nonlinear
constraints*
2. The Method of Lagrange Multipliers
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 17MAE 323 Lecture 7 FE Modeling Topics: Part 2
2. The Method of Lagrange Multipliers
Example 2
•In this example, let us re-use the model in example
1—applying the same loads, boundary conditions and
constraints– but this time, we will solve the problem
using the method of Lagrange Multipliers
•From equation (9), we have:
[ ]1
2
3
0 1 1
u
u
u
− = ⋅
C u
•Thus, we have all the information we need
to plug values directly into (16) :
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 18MAE 323 Lecture 7 FE Modeling Topics: Part 2
…where the force (λ=-F) may be interpreted as the constraint
force associated with a rigid link between nodes 2 and 3
2. The Method of Lagrange Multipliers
Example 2
1
2
3
2 0 0
2 1
0 1
0 1 1 0 0
T
k k u F
k k k u F
k k u F
λ
− − − = = = − − −
u fK C
λ qC 0
•Plugging in values yields:
(17)
•Solution of (17) yields:
1
2
3
3 /
5 /
5 /
u F k
u F k
u F k
Fλ
= −
(18)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 19MAE 323 Lecture 7 FE Modeling Topics: Part 2
•One may interpret this substitution as replacing the constraint force
with an equivalent stiffness matrix based on the penalty parameter
3. Penalty-Based Methods
•A third method of handling constraints may be derived by an
approximation of the method of Lagrange Multipliers. Namely, if
one introduces the approximation:
= ⋅ ⋅λ α C u
where αααα is a diagonal matrix of “penalty” numbers with units of
stiffness, then this expression may be substituted in the potential
energy expression (14):
( )( ) 0T T T
pΠ = ⋅ ⋅ − ⋅ + ⋅ ⋅ ⋅ ⋅ − =u K u u f α C u C u q
(18)
(19)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 20MAE 323 Lecture 7 FE Modeling Topics: Part 2
3. Penalty-Based Methods
•Once again, taking the derivative with respect to u (Castigliano’s
Second Theorem) yields:
( )T T+ ⋅ ⋅ ⋅ = + ⋅ ⋅K C α C u f C α q
•Thus, if α=0, there is effectively no constraint. The Lagrange
Multiplier result is retrieved only in the limit as α approaches
infinity. Numerically, this is not possible and so the penalty
method is always approximate. In practice, the value of α must
be such that CT⋅αααα⋅C is greater than K, but not so great as to cause
numerical difficulties
(20)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 21MAE 323 Lecture 7 FE Modeling Topics: Part 2
3. Penalty-Based Methods
Example3
•The model used previously will be re-used a third
time to demonstrate the use of the penalty method
of applying constraints
•As before, the constraint (9), C is re-used:
[ ]1
2
3
0 1 1
u
u
u
− = ⋅
C u
•In this case, let α1=α2=α3. Upon plugging into (20), this
yields:
1
2
3
2 1 0 0 0 0
1 2 1 0 1 1
0 1 1 0 1 1
u
k u
u
α
−
− − + − = − −
F
F
F
(21)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 22MAE 323 Lecture 7 FE Modeling Topics: Part 2
3. Penalty-Based Methods
Example3
•The solution to (21) is:
1
2
3
3 /
5 /
5 6
( )
u F k
u F k
u F Fk
k k
α
α
= + +
•Thus we see that when α=0, u3=6F/k (the same as if no constraint
existed between node 2 and 3). As α→∞, u3→5F/k, as expected (the
same as previous solutions)
•Although the Penalty Method is approximate, it does not increase the
number of DoFs in the system of equations (although it may increase
matrix bandwidth), and if the penalty values are chosen carefully, the
resulting equations may have desirable convergence properties in
numerical solutions of nonlinear problems
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 23MAE 323 Lecture 7 FE Modeling Topics: Part 2
Summary
•We have reviewed the three main methods of applying constraint
conditions to the numerical form of problems arising in structural
mechancs. We might summarize the pros and cons of each as follows:
1. Direct Constraint Elimination: This method splits a system of
equations into constrained and retained degrees of freedom.
Athough it may seem that the resulting system to be solved is
smaller, this is not true. In fact, direct constraint elimination
may be cumbersome and inefficient to implement in real
applications. This makes it suitable for only for linear problems
in which the total number of constrained DoFs is relatively
small. The advantage of the technique, however, lies in its
accuracy and ability to simply handle constraints of arbitrary
complexity
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 24MAE 323 Lecture 7 FE Modeling Topics: Part 2
Summary
2. The Method of Lagrange Multipliers: This method is simpler to
implement numerically, but increases the number of DoFs to be
solved (one for each DoF constrained). It’s advantage over direct
constraint elimination lies in its ability to handle nonlinear problems
more easily. However, special nonlinear routines must be used in
such applications (such as constraint optimization routines) and
convergence may be difficult. In small linear problems, this method
may be generally preferable to direct constraint elimination
(commercial codes seem divided over this).
3. The Penalty Method: This method is approximate, and so is not the
best candidate for small linear problems where accuracy is
important. However, it’s straightfoward implementation in nonlinear
problems makes it a top choice for such applications
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 25MAE 323 Lecture 7 FE Modeling Topics: Part 2
Some Practical Examples
•Direct Constraint Elimination is often used in cases where one wants to
connect DoFs in some fashion within a purely linear analysis (it becomes
very expensive for nonlinear analysis)
•A rather iconic example from structural analysis would be the modeling
of fasteners with beams connected to solid elements (see slide 4)
•Another example would be bonded contact between two surfaces, but
we will cover this later when we discuss contact
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 26MAE 323 Lecture 7 FE Modeling Topics: Part 2
Some Examples From ANSYS Workbench
• Moment Loading :
– For solid bodies moments can be applied on a surface only.
– If multiple surfaces are selected, the moment load is evenly distributed.
– Vector or component method can be employed using the right hand rule.
– For surface bodies a moment can be applied to a vertex, edge or surface.
– Units of moment are in Force*length.
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 27MAE 323 Lecture 7 FE Modeling Topics: Part 2
Some Examples From ANSYS Workbench
• Remote Force Loading :– Applies an offset force on a surface or edge of a body.
– The user supplies the origin of the force (geometry or coordinates).
– Can be defined using vector or component method.
– Applies an equivalent force and moment on the surface.
– Example: 10 inch beam with a 1 lbf remote force scoped to the end of the beam. Remote force is located 20 inches from the fixed support.
20”
F=1 lbf
Moment Reaction
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 28MAE 323 Lecture 7 FE Modeling Topics: Part 2
Some Examples From ANSYS Workbench
• Bolt Pretension:– Applies a pretension load to a solid cylindrical section or beam using:
• Pretension load (force)
• OR
• Adjustment (length)
– For body loading a local coordinate system is required (preload in z direction).
– For sequenced loading additional options are available (see next page).
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 29MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural
Nonlinearities
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 30MAE 323 Lecture 7 FE Modeling Topics: Part 2
•In structural mechanics, a problem is nonlinear if the stiffness matrix
OR the load vector depend on the displacements. Nonlinearities can be
classified as material nonlinearity (associated with changes in material
properties, as in plasticity), or geometric nonlinearity (associated with
changes in configuration, as in large deflections of a slender elastic
beam).
•In thermal problems, nonlinearity can arise from temperature-
dependent conductivity, convection, or heat flux. An example of a
nonlinear heat flux would be that associated with radiation
⋅ =K u f
•To put more precisely: in linear problems, both K and f are
considered independent of the primary variable u. In nonlinear
problems, either K or f is dependent on u
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 31MAE 323 Lecture 7 FE Modeling Topics: Part 2
Material Nonlinearity: Example—Plasticity
•When a ductile material is stressed beyond it’s elastic limit, it will
yield, acquiring permanent deformation
•The response of the material to loads beyond this limit is classified
as plastic
•Plasticity is brought about at the microstructural level by crystalline
planes slipping past one another. This behavior is mostly driven by
shear, so that volumetric strains are small if present at all
ε
σ
Yield Strength σy
Elastic Plastic
Unloading
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 32MAE 323 Lecture 7 FE Modeling Topics: Part 2
Geometric Nonlinearity: Example—Large Deflection
•If a structure experiences large deformations, its changing geometric
configuration can cause the structure to respond nonlinearly. This happens
as a result of either changing load magnitudes and orientations (relative to
the deforming material), or changing stiffness, or both
•In the figure below*, the tip deflection of a fishing pole under a constant
weight is a nonlinear function of the rod’s flexure
*taken with permission from ANSYS 13.0 help documentation ( 2010 SAS IP, Inc)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 33MAE 323 Lecture 7 FE Modeling Topics: Part 2
Solving Nonlinear Problems – The Newton-Raphson Method
•Since the applied load or stiffness is no longer independent of the primary
(displacement) variable, nonlinear problems are usually solved incrementally
as progressively changing linear problems – each subsequent iteration
depending on the result of the last
•For static structural problems, the most common method of doing this is
the Newton-Raphson Procedure
•In this procedure, one equilibrium iteration may be written as:
T A nr
i i i⋅ ∆ = −K u F F
•Displacements are calculated incrementally at iteration, i
according to:
1i i i+ = + ∆u u u
(22)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 34MAE 323 Lecture 7 FE Modeling Topics: Part 2
Solving Nonlinear Problems – The Newton-Raphson Method
In equation (22):T Tangent Stiffness matrix at iteration i
Vector of restoring loads corresponding to element internal loads
i
nr
i
=
=
K
F
•Both and are evaluated based on the values given by ui.
The RHS of (22) is called the residual or out-of-balance load vector
(the amount the system is out of equilibrium. A single solution
iteration is depicted graphically below for a one DOF model.
T
iK
nr
iF
•The steps performed in the Newton-Raphson procedure are
described below:
1. Assume u0. This is usually 0
2. Compute the updated tangent matrix and restoring load
from ui
3. Calculate ∆ui from (22)
4. Add ∆ui to ui in order to obtain next iteration ui+1
5. Repeat steps 2 thru 4 until convergence is obtained
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 35MAE 323 Lecture 7 FE Modeling Topics: Part 2
•In the preceding, convergence (step 5) is determined based on a residual
criterion. In principle, this should be zero, but in practice, small positive
numerical values are used based on numerical considerations for efficiency
•Below left, the first iteration of a NR run is shown. Below right, after 2
iterations
Solving Nonlinear Problems – The Newton-Raphson Method
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 36MAE 323 Lecture 7 FE Modeling Topics: Part 2
Nonlinear Solution Controls in ANSYS Workbench
• The “Analysis Settings” details provide general control over the solution process:
• Step Controls:
– Manual and auto time stepping controls.
– Specify the number of steps in an analysis and an end “time” for each step.
– “Time” is a tracking mechanism in static analyses (discussed later).
• Solver Controls:
– Two solvers available (default program chosen):
• Direct solver (Sparse solver in ANSYS).
• Iterative solver (PCG solver in ANSYS).
– Weak springs:
• Mechanical tries to anticipate under-constrained models.
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 37MAE 323 Lecture 7 FE Modeling Topics: Part 2
Nonlinear Solution Controls in ANSYS Workbench
• Analysis Data Management:
– Solver Files Directory is the location where analysis files will be stored if a project has not yet been saved.
– Future Analysis: indicates whether a down stream analysis (e.g. pre-stressed modal) will use the solution. This is set automatically when coupled analyses are configured in the project schematic.
– Scratch Solver Files Directory: temporary directory used during solution.
– Save MAPDL db.
– Delete Unneeded Files: may choose to save all files for future use in Mechanical APDL.
– Solver Units: Active System or manual.
– Solver Unit System: if the above setting is “manual”, you may choose 1 of 8 possible solver unit systems to insure consistency when data is shared with Mechanical APDL (does not affect results/load displays in the GUI).
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 38MAE 323 Lecture 7 FE Modeling Topics: Part 2
Nonlinear Solution Controls in ANSYS Workbench
• Step Controls:
– Multiple steps allow a series of static analyses to be set
up and solved sequentially.
– For a static analysis, the end time can be used as a
counter/tracker to identify the load steps and substeps.
– Results can be viewed step by step.
– Load values for each step can be entered in the “Tabular
Data” section provided.
The time and load value are
displayed in the graphics
window
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 39MAE 323 Lecture 7 FE Modeling Topics: Part 2
Nonlinear Solution Controls in ANSYS Workbench
• Results for each individual step can be viewed after the solution by selecting the desired step and RMB >“Retrieve This Result”.
Select desired step
and RMB to
retrieve result
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 40MAE 323 Lecture 7 FE Modeling Topics: Part 2
Contact Problems
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 41MAE 323 Lecture 7 FE Modeling Topics: Part 2
( )e e
ee T e
SV V
dV wdV wdSδ = +∫ ∫ ∫ε C ε b F
•First, recall Equation (5) from Chapter 5. This is the general
governing equation for an elastic continuum in weak form for
a single element (hence the superscript e):
( ), ,
1
2ij i j j i
w wδε = +where:
•Also recall that terms involving δ refer to the “variation” which we
are trying to minimize (in this case, by using a trial function for
displacement, w). In what follows, we’ll keep everything in terms of
δ to maintain consistency (we wont’ try to solve these equations—
just use their form to explain contact problems)
•Let’s move everything to the LHS to show that this equation
represents total potential energy, and sum over all elements:
( )e e
ee T e
p
e SV V
dV dV dSδ δ δ Π = − −
∑ ∫ ∫ ∫ε C ε b u F u (23)
Structural Contact Problems
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 42MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
•Now, structural contact problems involve a new energy term
which represents the interaction between two or more bodies. So
(23) becomes:
( ) "contact term"e e
ee T e
p
e SV V
dV dV dSδ δ δ Π = − − +
∑ ∫ ∫ ∫ε C ε b u F u
•This “contact term” is most easily expressed in terms of the
constraints we have already seen. However, such constraints are
typically nonlinear (this means that Direct Constraint Elimination is
usually not used).
•Most commercial codes give the user a choice (or perhaps choose
one for you) between the following three popular options:
•Penalty Methods
•Lagrange Methods
•Augmented Lagrange Methods
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 43MAE 323 Lecture 7 FE Modeling Topics: Part 2
•To keep things simple, consider contact between two bodies
(labeled 1 and 2). The new contact term would produce a stiffness
matrix connecting the two bodies (because it involves a constraint
relating the DoFs between the two bodies)
K1
K2
K1
K2
Kcontact
Structural Contact Problems
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 44MAE 323 Lecture 7 FE Modeling Topics: Part 2
•The Penalty and Lagrange Methods are the same as we’ve already
seen. Many commercial codes offer a hybrid combination between the
two, however, called the Augmented Lagrange Method (This is to
combine the best of both approaches). Below we summarize the form
of the contact term for each of the three methods*:
Structural Contact Problems
Penalty Method:
( )"contact term"= n n n t t t
A
g g dAα δ α δ+∫ g g
Lagrange Method:
( )"contact term"= n n t t
A
g dAλ δ λ δ+∫ g
Augmented Lagrange Method:
[ ] [ ]( )"contact term"= n n n n t t t t
A
g g dAλ α δ λ α δ+ + +∫ g g
*This description is taken from CADFEM online training documentation
F
gn
gt
αn αt
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 45MAE 323 Lecture 7 FE Modeling Topics: Part 2
•Here, gn and gt are the normal and tangential contact “gaps” between
the two bodies. The tangential component is only needed for frictional
contact, and it should be noted that this is a tensor containing a static
and kinematic term (to include a stick and slip condition).
•Let us simplify the presentation further by focusing only on frictionless,
or normal contact
Structural Contact Problems
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 46MAE 323 Lecture 7 FE Modeling Topics: Part 2
x
y
u1=0 k1
1
u4=0
k2
4
3u2,F2
Structural Contact Problems
Example 1: Penalty Method
2
u3,F3
g0
•To see how the Penalty Method
works in a problem, consider the 2-
spring system at left. The springs are
fixed at nodes 1 and 4. The following
summarizes the properties of the
system (the units are arbitrary):
1
4
2
3
0
1 2
0
0
5
0
1
2
u
u
F
F
g
k k k
=
=
=
=
=
= = =
•g0 is the “initial” geometric gap
between nodes 2 and 3
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 47MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•We can use our knowledge of the matrix equation of a 1-dimensional
(canonical) spring element, together with the Direct Stiffness Method
of matrix addition (Chapter 2) to derive the basic matrix equation:
1
2 2
3 3
4
1 1 0 0 0
1 1 0 0
0 0 1 1
0 0 1 1 0
u
u Fk
u F
u
−
− = − −
2 2
3 3
1 0
0 1
u Fk
u F
=
•We can eliminate the rows and columns associated with the fixed DoFs
(1 and 4). This leaves us with the unconstrained equation. Note,
because of the gap, DoFs u2 and u3 are not “connected”. These are, in
effect two different, simultaneous but unrelated spring problems
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 48MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•Now, to apply the constraint, we first go back to the “contact
term”, which includes it – remembering to eliminate the frictional
term:"contact term"= n n n
A
g g dAα δ∫
•Here, the normal gap, gn is given by:
( ) [ ] 2
0 3 2 0
3
1 1n
ug g u u g
u
= − − = − −
•Since there is no contact area (the contact region is just a point),
we don’t’ have to perform the integral in (24).
•We see now that gn is the form C⋅u-q. The constraint we want to
enforce is actually gn≥0.
(24)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 49MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•Upon finding the first variation of (24), we have a stiffness term and a
residual (In other words, after taking the derivative with respect to δu.
Note that this yields an equation identical to equation (20)):
0
Stiffness Term =
Residual (force) Term = -
T
Tg
α
α
⋅ ⋅
⋅ ⋅
C C
C
•Here we are assuming a single scalar penalty stiffness, α.
These terms may now be added to the unconstrained
equations in a Newton-Raphson scheme:
( )0
T Tgα α+ ⋅ ⋅ ⋅ = − ⋅ ⋅K C C u f C
•The contact stiffness and residual terms only get added to the
system equations when: 0
ng ≤
(25)
(26)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 50MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•In fact, applying (25) when condition (26) is satisfied makes this a nonlinear
constraint. It ensures that gn≥0 within numerical bounds determined by the
numerical properties of the system equations with the inclusion of α
•Applying these constraints to our system equations results in
2 2
0
3 3
1 0
0 1
u Fk g
u F
α α α
α α α
− − + = −
−
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 51MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•Let‘s see how this system behaves qualitatively. When gn >0
(when u3-u2<g0), the solution behaves the same as if no constraint
is present (α=0):
2 2
3 3
/
/
u F k
u F k
=
•But when gn≤0 (u3-u2≥g0), we have:
2 01
3 02
1
( 2 )
F gu k
F gu kk k
αα α
αα αα
++ = −++
•We can easily see that when α=0, (27) is recovered, but as α→∞,
(28) becomes singular. Well before this, the matrix becomes too ill-
conditioned to solve with finite precision on digital computers
(27)
(28)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 52MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•If we plot u2 vs. F2, we see that there is a “knee” in the curve at
F2=2 . This is when u1 equals the initial gap. At forces higher than
this value, the constraint is engaged and both springs resist F2. This
bilinear curve represents one the simplest kinds of nonlinearity we
can encounter in structural mechanics.
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 53MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•Now, let’s see how the system behaves quantitatively by plugging
in the values from earlier:
1
4
2
3
0
1 2
0
0
5
0
1
2
100
u
u
F
F
g
k k k
α
=
=
=
=
=
= = =
=
•Since F2/k = 5/2 > g0, equation (28) is used.
In most commercial systems, this check
would be performed in the first iteration of
a nonlinear Newton-Raphson routine
•Plugging values at right into (28) yields:
2
3
1.7574
.74257
u
u
=
gn=-0.01485
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 54MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 1: Penalty Method
•So, with a penalty value of 100 (50 times the spring stiffness), we
have a final gap of roughly -0.015 (a penetration of 1.5 percent the
original gap value)
•If we increase the value to 1000, we get:
2
3
1.7508
.74925
u
u
=
gn=-0.001499
•This behavior is characteristic of the penalty method. Namely, one needs
fairly large penalty values to enforce the constraint with high accuracy. The
inaccuracy is manifested by an unavoidable penetration. If one wants
increased accuracy, the Lagrange or Augmented Lagrange Method must be
used.
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 55MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 2: Lagrange Method
•We will demonstrate the Lagrange Method on the same
spring system as before. Again, we ignore friction:
"contact term"= n n
A
g dAλ δ∫
( ) [ ] 2
0 3 2 0
3
1 1 0n
ug g u u g
u
= − − = − − ≥
•Once again, we want to apply the constraint:
•We re-write this as:
[ ] 2
0
3
1 1u
gu
− ≥ −
(29)
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 56MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 2: Lagrange Method
•This time, we’ll save a little effort and note that we have
everything we need to plug into equation (17) (just as we could
have plugged directly into equation (20) in the previous example)
2 2
3 3
0
0 1
0 1
1 1 0
Tk u F
k u F
gλ
− = = = − −
u fK C
λ qC 0(30)
•However, we should note a subtlety. Equation (29) involves an
inequality, which we have simply ignored in (30). In real
implementations of the Lagrange Method, we can’t do this. And we
can’t just add in the terms C and CT when the gap reaches a threshold
with many solvers. This difficulty reflects the fact that this is really a
constraint optimization problem, and some more sophisticated
techniques are required to enforce (29). In any case, we will ignore this
for now and pretend that we already know the gap is closed
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 57MAE 323 Lecture 7 FE Modeling Topics: Part 2
Structural Contact Problems
Example 2: Lagrange Method
•Solving (30) yields:
2
3
1.75
0.75
3.5
u
u
λ
=
•The λ variable gives us the constraint force, which we know
must act opposite the applied load. This reveals a benefit of
the Lagrange Method: It automatically provides some useful
reaction force information. For example, we immediately
know that that there is a 3.5 lbf tensile load and 1.5 lbf (5-3.5)
compressive load acting on nodes 2 and 3:
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 58MAE 323 Lecture 7 FE Modeling Topics: Part 2
ANSYS Workbench Implementation
• When importing assemblies of solid parts, contact regions are automatically created between the solid bodies.
– Contact allows non-matching meshes at boundaries between solid parts
– Tolerance controls under “Contact” branch allows the user to specify distance of auto contact detection via slider bar
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 59MAE 323 Lecture 7 FE Modeling Topics: Part 2
ANSYS Workbench Implementation
• In Mechanical, the concept of contact and target surfaces are used for each contact region:
– One side of a contact region is referred to as a contact surface, the other side is referred to as a
target surface.
– The contact surfaces are restricted from penetrating through the target surface.
• When one side is designated the contact and the other side the target, this is called asymmetric contact.
• If both sides are made to be contact & target this is called symmetric contact.
• By default, Mechanical uses symmetric
contact for solid assemblies.
• For ANSYS Professional licenses and
above, the user may change to
asymmetric contact, as desired.
Symmetric
ContactAsymmetric
Contact
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 60MAE 323 Lecture 7 FE Modeling Topics: Part 2
ANSYS Workbench Implementation
• Five contact types are available:
– Bonded and No Separation contact are linear and require only 1
iteration.
– Frictionless, Rough and Frictional contact are nonlinear and
require multiple iterations.
• Nonlinear contact types allow an “interface treatment”
option:
• “Add Offset”: input zero or non-zero value for initial adjustment
• “Adjusted to Touch”: ANSYS closes any gap to a just touching position
(ANSYS Professional and above)
Contact Type Iterations Normal Behavior (Separation) Tangential Behavior (Sliding)
Bonded 1 No Gaps No Sliding
No Separation 1 No Gaps Sliding Allowed
Frictionless Multiple Gaps Allowed Sliding Allowed
Rough Multiple Gaps Allowed No Sliding
Frictional Multiple Gaps Allowed Sliding Allowed
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 61MAE 323 Lecture 7 FE Modeling Topics: Part 2
ANSYS Workbench Implementation
• Interface treatment options:
Add offset: contact surface is numerically
offset a given amount in positive or negative
direction (offset can be ramped on).
Adjusted to touch: offsets contact surface
to provide initial contact with target
regardless of actual gap/penetration.
TCC T
Modeling Topics: Part IIMAE 323: Lecture 7
2011 Alex Grishin 62MAE 323 Lecture 7 FE Modeling Topics: Part 2
ANSYS Workbench Implementation
• Advanced options (see chapter 3 for additional details on the pinball region):– Pin Ball Region:
• Inside pinball = near-field contact
• Outside pinball = far-field contact
• Allows the solver to more efficiently process contact calculations.
• For ANSYS Professional licenses and above, mixed assemblies of shells and solids are supported as well as more contact options.
In this case, the gap between
the two parts is bigger than the
pinball region, so no automatic
gap closure will be performed.