Penalty Based
Contact
Formulation
©©2010
2010ANSYS,
ANSYS,Inc.
Inc.All
Allrights
rightsreserved.
reserved. 1 ANSYS, Inc. Proprietary
Overview
A. Basic concept of contact
B. Contact Formulations
C. Penalty based stiffness and penetration properties
D. Tips for achieving convergence
E. Questions & Answers
© 2010 ANSYS, Inc. All rights reserved. 2 ANSYS, Inc. Proprietary
A. Basic Concepts
Contact:
• When two separate surfaces touch each other such that
they become mutually tangent, they are said to be in
contact.
• In the common physical sense, surfaces that are in
contact have these characteristics:
– They do not interpenetrate.
– They can transmit compressive normal forces and
tangential friction forces.
– They often do not transmit tensile normal forces.
• They are therefore free to separate and move away from each
other.
• Contact is a changing-status nonlinearity. That is, the
stiffness of the system depends on the contact status,
whether parts are touching or separated.
© 2010 ANSYS, Inc. All rights reserved. 3 ANSYS, Inc. Proprietary
... Basic Concepts
How compatibility is enforced in a contact region:
• Physical contacting bodies do not interpenetrate. Therefore, the
program must establish a relationship between the two surfaces
to prevent them from passing through each other in the
analysis.
– When the program prevents interpenetration, we say that it
enforces contact compatibility.
– Workbench Mechanical offers several different contact
formulations to enforce compatibility at the contact interface.
F Penetration occurs when contact
compatibility is not enforced.
Target F
Contact
© 2010 ANSYS, Inc. All rights reserved. 4 ANSYS, Inc. Proprietary
B. Contact Formulations
• For nonlinear solid body contact of faces, Pure Penalty or Augmented
Lagrange formulations can be used:
– Both of these are penalty-based contact formulations:
Fnormal knormal x penetration
– Here, for a finite contact force Fnormal, there is a concept of contact
stiffness knormal. The higher the contact stiffness, the lower the
penetration xpenetration, as shown in the figure below
– Ideally, for an infinite knormal, one would get zero penetration. This is not
numerically possible with penalty-based methods, but as long as
xpenetration is small or negligible, the solution results will be accurate.
Fn
xp
© 2010 ANSYS, Inc. All rights reserved. 5 ANSYS, Inc. Proprietary
... Contact Formulations
• The main difference between Pure Penalty and
Augmented Lagrange methods is that the latter
augments the contact force (pressure) calculations:
Pure Penalty: Fnormal knormal x penetration
Augmented Lagrange: Fnormal knormal x penetration l
• Because of the extra term l, the augmented
Lagrange method is less sensitive to the magnitude
of the contact stiffness knormal.
• All things being equal, Augmented Lagrange should
produce less penetration then Pure Penalty, but
might take more iterations to converge.
© 2010 ANSYS, Inc. All rights reserved. 6 ANSYS, Inc. Proprietary
... Contact Formulations
• The aforementioned options relate contact in the normal
direction. If friction or rough/bonded contact is defined, a
similar situation exists in the tangential direction.
– Similar to the impenetrability condition, in the tangential direction,
the two bodies should not slide relative to each other if they are
“sticking”
– Pure penalty formulation is always used in the tangential direction
– Tangential contact stiffness and sliding distance are the
analogous parameters:
If “sticking”: Ftangential ktangential xsliding
where xsliding ideally is zero for sticking, although some slip is
allowed in the penalty-based method.
– Unlike the Normal Contact Stiffness, the Tangential Contact
Stiffness cannot directly be changed by the user.
© 2010 ANSYS, Inc. All rights reserved. 7 ANSYS, Inc. Proprietary
... Contact Formulations
• Another available option is Lagrange multiplier Formulation:
– The Normal Lagrange Formulation adds an extra degree of
freedom (contact pressure) to satisfy contact compatibility.
Consequently, instead of resolving contact force as contact
pressure) is solved for explicitly as an extra DOF. Fnormal
stiffness and penetration, contact force (contact
DOF
• Enforces zero/nearly-zero penetration with pressure DOF
• Does not require a normal contact stiffness (zero elastic slip)
• Requires Direct Solver, which can be more computationally expensive
© 2010 ANSYS, Inc. All rights reserved. 8 ANSYS, Inc. Proprietary
... Contact Formulations
• Chattering is an issue which often occurs with Normal Lagrange
method
– If no penetration is allowed (left), then the contact status is either
open or closed (a step function). This can sometimes make
convergence more difficult because contact points may oscillate
between open/closed status. This is called chattering
– If some slight penetration is allowed (right), it can make it easier
to converge since contact is no longer a step change.
Contact Status Contact Status
Open Open
Penetration Gap Penetration Gap
Closed Closed Penetration
Normal Lagrange Method Penalty-Based Method
© 2010 ANSYS, Inc. All rights reserved. 9 ANSYS, Inc. Proprietary
... Contact Formulations
• For the specific case of “Bonded” and “No Separation” type of
contact between two faces, a multi-point constraint (MPC)
formulation is available.
– MPC internally adds constraint equations to “tie” the
displacements between contacting surfaces
– This approach is not penalty-based or Lagrange multiplier-
based. It is a direct, efficient way of relating surfaces of contact
regions which are bonded.
– Large-deformation effects also are supported with MPC-based
bonded contact
© 2010 ANSYS, Inc. All rights reserved. 10 ANSYS, Inc. Proprietary
... Contact Formulations
• One additional note worth mentioning is that contact is detected
differently, depending on the formulation used:
– Pure Penalty and Augmented Lagrange Formulations use
integration point detection. This results in more detection points
(10 in this example on left)
– Normal Lagrange and MPC Formulation use nodal detection
(normal direction from Target). This results in fewer detection
points (6 in the example on right)
– Nodal detection may handle contact at edges slightly better, but a
localized, finer mesh will alleviate this situation with integration
point detection.
Integration Point Detection Nodal Detection
© 2010 ANSYS, Inc. All rights reserved. 11 ANSYS, Inc. Proprietary
C. Contact Stiffness and
Penetration
• Although “Pure Penalty” is the default in
Workbench-Mechanical, “Augmented
Lagrange” is recommended for general
frictionless or frictional contact in large-
deformation problems.
– Augmented Lagrange formulation adds
additional controls to automatically reduce
penetration
• The “Normal Stiffness” is the contact
stiffness
knormal explained earlier, used only for “Pure
Penalty” or “Augmented Lagrange”
– This is a relative factor. The use of 1.0 is
recommended for general bulk deformation-
dominated problems. For bending-dominated
situations, a smaller value of 0.1 may be
useful
if convergence difficulties are encountered.
© 2010 ANSYS, Inc. All rights reserved. 12 ANSYS, Inc. Proprietary
... Contact Stiffness and
Penetration
• The Normal Contact Stiffness knormal is the most important parameter
affecting both accuracy and convergence behavior.
– A large value of stiffness gives better accuracy, but the problem may
become more difficult to convergence.
– If the contact stiffness is too large, the model may oscillate, with
contacting surfaces bouncing off of each other
F
F
F
Fcontact
Iteration n Iteration n+1 Iteration n+2
© 2010 ANSYS, Inc. All rights reserved. 13 ANSYS, Inc. Proprietary
... Contact Stiffness and
Penetration
• Example showing effect of contact stiffness:
Formulation Normal Stiffness Max Deform Max Eqv Stress Max Contact Pressure Max Penetration Iterations
Augmented Lagrage 0.01 2.84E-03 1% 26.102 1% 0.979 36% 2.70E-04 2
Augmented Lagrage 0.1 2.80E-03 0% 25.802 0% 1.228 20% 3.38E-05 2
Augmented Lagrage 1 2.80E-03 0% 25.679 0% 1.568 2% 4.32E-06 3
Augmented Lagrage 10 2.80E-03 0% 25.765 0% 1.599 4% 4.41E-07 4
Normal Lagrange - 2.80E-03 0% 25.768 0% 1.535 0% 3.17E-10 2
• As is apparent from the above
table, the lower the contact
stiffness factor, the higher the
penetration. However, it also
often makes the solution
faster/easier to converge
(fewer iterations)
© 2010 ANSYS, Inc. All rights reserved. 14 ANSYS, Inc. Proprietary
D. Tips for achieving successful
convergence
• The default Normal Stiffness is automatically
determined by WB-Mechanical.
– The user may input a “Normal Stiffness Factor” (FKN)
which is a multiplier on the code calculated stiffness.
The lower the factor, the lower the contact stiffness.
• Default FKN =10 (for Bonded and No Separation behaviors)
• Default FKN=1.0 (for all other behaviors)
• Some general guidelines on selection of Normal
Stiffness for contact problems:
– For bulk-dominated problems: Use “Program
Controlled” or manually enter a “Normal Stiffness
Factor” of “1”
– For bending-dominated problems: Manually enter a
“Normal Stiffness Factor” of “0.01” to “0.1”
© 2010 ANSYS, Inc. All rights reserved. 15 ANSYS, Inc. Proprietary
... Tips for achieving successful
convergence
• Also, the contact stiffness can
be automatically adjusted during
the solution.
– If difficulties arise, the stiffness
will be reduced to enhance
convergence.
© 2010 ANSYS, Inc. All rights reserved. 16 ANSYS, Inc. Proprietary
... Tips for achieving successful
convergence
• Adding an offset at a contact region is sometimes necessary
(i.e. To simulate an interference fit between parts).
This can have a negative effect on convergence, by dramatically increasing the force
generated at the interface.
• “Add Offset, Ramped Effects” - Applies the interference gradually over several substeps
within a load step, thereby stabilizing the convergence pattern for challenging
interference problems.
© 2010 ANSYS, Inc. All rights reserved. 17 ANSYS, Inc. Proprietary
... Tips for achieving successful
convergence
• By plotting the force residuals (available in Solution Information
branch), user can identify contact regions causing instability.
– Lowering the contact Normal Stiffness at these locations and refining
the mesh, the problem can be solved.
© 2010 ANSYS, Inc. All rights reserved. 18 ANSYS, Inc. Proprietary