Vous êtes sur la page 1sur 6

Date May 27, 2007 Memo Number STI:07/01

Author Sheldon Imaoka ANSYS Revision 11.0


Subject Sheldons ANSYS.NET Tips and Tricks: CMS in Workbench Simulation
Keywords CMS, Flexible Dynamics, Rigid Dynamics, Joints, Superelements

1. Introduction:
At Workbench 11.0, the ANSYS Rigid Dynamics add-on module enables users to
model complex kinematic assemblies, whereby system-level performance as well as
accurate force loading on components can be obtained.
Although the parts in the assembly may initially be considered rigid, an interesting,
related method is the use of component mode synthesis (CMS) to account for the
flexibility of parts. This memo will attempt to cover a way in which users can
incorporate CMS in their multibody analyses in Workbench Simulation.

2. Background on System-Level Analysis Methods:
With the ANSYS solver, users can solve a wide array of problems. For system-level
analyses (e.g., transient analyses), one may be more interested in the overall system
performance and force distribution rather than detailed stress analyses, so the flexibility
of the parts may be approximated with various methods, as summarized below:

Part Representation # DOF Nonlinear Inertial Force Temp Contact Joints
Solid/Shell Mesh Many All x x x x x
Superelement Few NLGEOM + + + + +
Beams Few All x x x + +
Spring/Mass/Damper Few NLGEOM + + +
Rigid (Mass) 6 NLGEOM x + + x
Loading Interactions


In the above table, an x denotes functionality available directly in the Workbench
Simulation 11.0 GUI while an + indicates ANSYS functionality that would need to be
accessed with Commands objects. (Defining Joints in the Workbench Simulation GUI
requires the use of the ANSYS Rigid Dynamics add-on license.)
Each method has its pros and cons, based accuracy and efficiency considerations. For
example, modeling an entire system with spring/mass/dampers (i.e., lumped parameters)
can be computationally efficient at the expense of approximating the part flexibilities.
The ability to model rigid and flexible parts in Workbench Simulation
1
is very
attractive since the rigid parts are modeled with mass and rigid links (i.e., rigid contact
elements), providing load transfer capabilities with little CPU cost. However, the
solution time is dictated by the size of the flexible mesh, and in some cases, the rigid
assumption may not be accurate enough for users needs.
A better balance is the use of CMS superelements, where the flexibility of the model
is retained, yet the number of DOF is reduced, thereby providing very efficient solutions.

1
While Workbench 11.0 introduced the Rigid Dynamic analysis capability, the discussion on rigid
bodies in this memo applies to the Flexible Dynamic analysis capability in Workbench Simulation 11.0
and to the procedure outlined in the Multibody Analysis Guide in the ANSYS help manual. This is because
the Rigid Dynamic solver only applies to rigid bodies and is different from the regular ANSYS solver.
3. Background on Component Mode Synthesis (CMS):
A superelement has reduced number of DOF compared with the full model, yet it
can still accurately model the flexibility of structures. Superelements can be created by
regular substructuring or by component mode synthesis, where retained DOF u
m
are
designated by the user as those to keep while removed DOF u
s
are eliminated by
condensation of the matrices. In doing so, a more complex element stiffness (and mass)
matrix is created that can be a more computationally efficient representation of the part,
as long as the number of retained DOF u
m
is not too large. Also, since this matrix
condensation occurs once, linear material behavior is assumed within the superelement.
For regular substructuring, only interface nodes
2
are retained for the FE solution:
{ }
[ ] [ ] [ ][ ] [ ]
[ ] [ ] [ ][ ] [ ] [ ][ ] [ ] [ ][ ] [ ][ ] [ ]
[ ]{ } [ ]{ } [ ]{ } { } F u K u C u M
K K M K K K K M M K K M M
K K K K K
u
u
u
m m m
sm ss ss ss ms sm ss ms sm ss ms mm
sm ss ms mm
s
m

1 1 1 1
1
= + +
+ =
=

& & &



The stiffness matrix condensation is exact, so this provides an accurate stiffness
representation between the interface nodes. However, the mass representation is
approximated and is lumped at the interface nodes, so this is not a preferred method for
dynamic simulations.
3

For component mode synthesis, besides the interface nodes, additional generalized
coordinates y

are added to the superelement. Generalized coordinates are the mode


shapes of the structure
4
, and these provide information on its dynamic response:
{ } [ ]
[ ] [ ] [ ][ ]
[ ] [ ] [ ][ ]
[ ]{ } [ ]{ } { } F u K u M
T M T M
T K T K
y
u
T
u
u
u
T
T
m
s
m

= +
=
=

=
& &


Consequently, CMS provides accurate static and dynamic representations of the
substructure at minimal cost.

Since CMS superelements can be used in large-deflection nonlinear analyses, they are
especially attractive for nonlinear transient problems because of the low number of DOF
(i.e., interface nodes and generalized coordinates) and because of the accurate dynamic
representation. CMS superelements, however, can also be used in static, modal,
harmonic, and response spectrum analyses as well.

2
Technically speaking, it is master DOF that are retained in substructure generation, so a subscript m is
used, but for the sake of simplicity, the terminology interface nodes will be used. The subscript s is used
for the condensed (removed) DOF.
3
One can add more master DOF inside of the part, but having enough master DOF to characterize the
dynamic content of a regular substructure requires some user experience.
4
Depending on the method used, the transformation matrix [T] will differ. For example, in the fixed-
interface method, the normal modes are those obtained with the interface nodes fixed.
If contact is used with superelements (regular substructure or CMS), the number of
interface nodes increases dramatically, depending on the size of the contacting region.
Hence, CMS superelements should be used in Flexible Dynamic analyses in Workbench
Simulation with Joints. Joints are defined at a single node, so the number of interface
nodes used in CMS is reduced to a bare minimum.

The number of modes and the transformation matrix used determine whether high-
frequency dynamic responses are captured accurately. Consult Chapter 11, Component
Mode Synthesis in the Advanced Analysis Techniques Guide and Chapter 5, Using
Component Mode Synthesis Superelements in a Multibody Analysis in the Multibody
Analysis Guide in the ANSYS documentation for more details.

4. Procedure for including CMS in Workbench Simulation:
CMS is not directly supported in the Workbench Simulation 11.0 GUI, but it can be
easily added in a number of ways. Because it may be common for different users to work
on different models, the author will describe one approach assuming different databases
for the CMS superelements with APDL Commands objects.
Basic workflow in Workbench Simulation 11.0:
1. Flexible Dynamic analysis of system is defined, using rigid parts and Joints
2. Superelement(s) of individual part(s) are generated with interface nodes located
at the same position and orientation as the Joints connected to those part(s). Note
that interface node numbers should not overlap between superelement parts.
3. In the Flexible Dynamic analysis of system, the rigid parts are replaced by
superelements, and the Flexible Dynamic simulation is performed.
4. The deformation and joint results of interest are reviewed from the Flexible
Dynamic Analysis (note that deformations are reported at the mass CG, so these
values will be lower than the flexible representation). Depending on the users
needs, results at one or all points can be retrieved for each part, and the
deformation, stresses, and strains of the individual part(s) can be postprocessed
separately.

The CMS representation can provide a much more efficient solution with a low
number of interface nodes. Hence, the author prefers using CMS substructures not for
Contact but only for Joints.
If CMS superelements are to be connected to other parts via Joints, the interface
nodes need to be defined, as noted in step 2. This can be simply done by adding an
appropriate Coordinate System that is the same location and orientation as the Joint
Reference Coordinate System. Then, add a Remote Displacement support with rigid
behavior referencing the surface(s) and the coordinate system used in the Joint can be
defined. When the CMS superelement is generated, the remote displacement node should
be treated as the retained interface node.
5


5
The interface nodes (nodes of the Remote Displacement support) can be selected by imposing a 0
rotational constraint in Workbench Simulation, then selecting nodes with 0 rotation constraint with the
NSEL command. Alternatively, an undocumented variable create cms components can be set in the
Workbench Simulation Variable Manager with a value of 1. Then, create a Named Selection of the body
such as INTERFACE. The interface nodes will be in a nodal component called INTERFACE_master.
5. Example Case 1:
A sample Autodesk Inventor assembly was used for the first
example. A Flexible Dynamic analysis was set up with all nine
parts as rigid, as shown on the right.
Interaction between the parts was defined by Joints using rigid
behavior for the associated surfaces. Two Joint Conditions were used
for loading of the sample engine.

In a separate model, the connecting rod was set up in order to create a CMS
superelement. The figure on the left shows the original system-level analysis with the
Joints. The individual model needs to reproduce the Joint Reference Coordinate System
location and orientation with a Remote Displacement support, as shown in the center.
With that completed, the model can be meshed, and a superelement can be created.
To accomplish the generation of the superelement, a Commands object is inserted that
selects constrained nodes (from the Remote Displacement support), deletes the
constraints, and designates those nodes as master DOF. A CMS generation pass is
performed with a user-specified number of requested modes, and the resulting file.sub
is the superelement file needed for the system-level analysis.
A sample macro cms_generate.mac is included in this memo that automates the
above steps, with the number of modes supplied as ARG1. There are some additional
features the author included in the macro, such as (a) renaming and copying the resulting
superelement file partX.sub to the parent directory, (b) creating an Excel spreadsheet
listing of the attachment node locations, and (c) creating a PNG image of the element plot.
This procedure is repeated for each rigid part that will be converted to a superelement.
Note that since multiple superelements may exist, each .sub superelement file should
have a unique name (the author chose to use partX.sub, where X is an integer). Lastly,
the user should ensure that the same unit system is selected as the system-level assembly
since generated superelement files are only usable in the unit system they were created in.

Once the superelements are generated, they can be incorporated into the system-level
model. The author prefers doing this by adding two Commands objects, one under the
part to be replaced and another under the Flexible Dynamic branch. The first has a
single line, PART1_ = MATID allowing the user to reference the element type ID a
priori. The second Commands object deletes all rigid links of the rigid part except one,
adds the superelement, then couples coincident nodes together this is automated with
the supplied macro cms_use.mac.
The rigid body is connected to Joints via rigid links (i.e., rigid contact elements). The
reason why the author prefers deleting all but one rigid link is so that the user can
postprocess as usual in Workbench Simulation, the original rigid mass is used to track
the position of the part, so if only one of the rigid links is left, the mass element moves
with the connected joint, and the user can still visualize the response of the system. Note,
however, that if the part undergoes appreciable straining, the part will not look
completely connected to the other bodies since a rigid representation is not very accurate
for a deformable body. However, the author feels that the ability to postprocess and
visualize the assembly in Workbench Simulation outweighs this small discrepancy in
solving parts as flexible but postprocessing them as rigid. Of course, results for Joints
and Springs can be probed as normal with this approach.
The charts below show relative angular velocity at the crankshaft-connecting rod
Joint for both the rigid assembly as well as another assembly with the piston, connecting
rod, and crankshaft modeled as CMS superelements. For relatively low loading, as
illustrated on the left, both results match well, as expected. For higher loading where
there is some relative deformation, as shown on the right, the rigid-only and CMS cases
start to show differences due to the flexibility of the parts.
It is important to note that the solution times for both system-level runs were relatively
quick. The rigid-only case ran 303 iterations in 19 seconds on a 3.2 GHz PC. The
assembly with 3 CMS components ran for 1340 iterations in 191 seconds. The CMS
assembly case required more iterations because the flexibility of 3 parts were included.
Although the CMS model solved the time-dependent loading in about 3 minutes, the
CMS generation required 1240 seconds for the connecting rod (82,500 nodes), 5500
seconds for the crank-shaft (184,000 nodes), and 1580 seconds for the piston (91,000
nodes). These were purposely made to be a fine mesh, so it took about 2.3 hours total to
generate the CMS superelements. However, different engineers could generate these
CMS superelements in parallel (1.5 hours max elapsed), and only parts that change their
mesh need to regenerate a CMS superelement. Hence, for multiple analyses, the CMS
approach is extremely efficient. The solution times for running 1340 iterations for this
same , full model having 1 million DOF would not even be comparable!


6. Example Case 2:
A simpler example of two beams with a revolute joint and an applied moment is
provided with this memo. There are three Models:
Simplified Representation > Rigid Bodies this model assumes both parts rigid
317 iterations in 13 seconds
Full Representation > Flexible Bodies this model has both parts as flexible
1270 iterations in 13500 seconds
CMS - Use > CMS Use Pass this model has both parts as CMS superelements
949 iterations in 49 seconds (+20 seconds total to generate superelements)
In all three cases, the revolute joint between the beams has stops defined at 45 via a
Commands object.
Animations of the above three cases are included, and one can see that the deformable
nature of the parts is quite important to achieve accurate results. However, the full
model takes about 270 times longer than the CMS model! This is due to the fact that, per
iteration, solving the full FE equations is more computationally expensive than the CMS
model, which has much fewer DOF.

One point(s) of interest are determined, the input file cms_expand_one.inp or
cms_expand_all.inp can be used to expand the results. This means that the
deformation, stresses, and strains for the individual parts will be retrieved. This should
be done in a different directory, where the ANSYS solution is solved in batch. Once the
ANSYS run is completed (this expansion generally doesnt take too long), the entire
folder can be read in via Tools menu > Read ANSYS Result Files. Postprocessing
for the individual parts can then be performed as normal. (If only one result is of interest,
the author recommends setting the Environment branch to Static Structural whereas if
all results are required, set the Environment branch to Flexible Dynamic with the same
Analysis Settings as the original system-level model.)
For this example, the results near TIME=0.020298 (0.020309) seconds is shown for
the CMS part as well as for the Full representation, and the stress values and
distribution agree very well.


7. Conclusion:
This memo covered one method in which flexible bodies can be represented very
efficiently and accurately in system-level analyses via CMS.