You are on page 1of 11

Chapter 29: Rapid Road Response Optimization of a Camaro Model using Automatic External Superelement

Optimization

Rapid Road Response


Optimization of a Camaro
29 Model using Automatic
External Superelement
Optimization


Summary 487

Introduction 488

Requested Solutions 489
 Optimization Solutions 490

Modeling Tip 495

Input File(s) 496
CHAPTER 29 487
Rapid Road Response Optimization of a Camaro Model using Automatic External Superelement

Summary
Title Chapter 29: Rapid Road Response Optimization of a Camaro Model using Automatic
External Superelement Optimization, AESO
Features • Grids 23K
• Total degrees of freedom 137K
• Degrees of freedom in residual 7K
• Elements 37K
• Subcases 2
• Frequencies 3
Geometry

Material properties Mild Steel (E = 2x107 Psi, nu = 0.28, rho = 7.835x10-5 lbf-s2/in4)
Analysis type Modal/Direct Frequency Analysis
Boundary conditions See the asm file, aeso9.asm, containing boundary connection data
Element type CQUAD4, CTRI3, CROD
Loads Random inputs applied on left and right suspension, including cross-correlation (see
Figure 29-2)
FE results
5.0E -0 3

4.0E -0 3
S u m m ed A c celeratio

3.0E -0 3
S UM Init
S um fina l
2.0E -0 3

1.0E -0 3

0.0 E +0 0
4 6 8 10 12 14
F req u en cy (H z )
488 MD Demonstration Problems
CHAPTER 29

Introduction
The purpose of the example is to illustrate how to run an Automatic External Superelement Optimization, AESO, job
and to demonstrate significant performance gain can be achieved with AESO. Learn more about the capability, consult
MD Nastran Design Sensitifity and Optimization User’s Guide. It is assumed that the reader is experienced in
performing modal frequency analysis. The discussion of the analysis modeling is kept to minimum.
The Camaro model is provided by General Motor Corp (Figure 29-1). Random inputs are applied on left and right
suspension, including cross-correlation (Figure 29-2). The road response optimization task is solved by varying spring
constants of the engine mount to achieve maximum ride comfort. Both a regular (or a single shot) optimization run
and an AESO run are performed. The efficiency and accuracy of the solutions are compared between two approaches.

Figure 29-1 Camaro Model

1.60E-02

1.20E-02
Input Spectra

LEFT SUSP
8.00E-03
RIGHT SUSP
REAL L/R
4.00E-03
IMAG L/R

0.00E+00
4 6 8 10 12 14
-4.00E-03
Frequency

Figure 29-2 Input Load Power Spectra


CHAPTER 29 489
Rapid Road Response Optimization of a Camaro Model using Automatic External Superelement

Requested Solutions
The task will be solved in three design scenarios that are described in detail in the Optimization Solution section. Each
of three cases is solved by a single run approach and the AESO run approach. Then, the results and performance data
are compared between two approaches. It has been observed that the single shot run may fail with signal = 11 message
in the log file when design cycle is greater than 1 due to some modeling issue. However, this behavior does not show
up in the AESO runs. In this document, the results from good single shot runs will be presented but the input file is
not included.
The AESO approach should demonstrate that
• accurate and very efficient solution can be obtained
• the reduced model allows to perform re-analyses and/or optimization tasks many times rapidly
• much larger performance gain is achieved with Analysis=DFREQ
490 MD Demonstration Problems
CHAPTER 29

Optimization Solutions

Case A
This design case is to minimize the sum of RMS acceleration at driver’s seat and passenger’s seat while limiting the
PSD response at steering column by varying nine spring constants of the engine mount. Listing 1 shows the required
design model set up for Case A.

Listing 1 Design model set up for Case A

...
DESOBJ = 1020
DESSUB = 101
...
BEGIN BULK
$ design model set up
DESVAR 11 K5307 1.0 0.01 3.0
......
DESVAR 24 K5018 1.0 0.01 3.0
DVCREL1 5307 CELAS2 5307 K
11 1246.3
......
DVCREL1 5018 CELAS2 5018 K
24 1120.
FREQ1 4 6.0 0.1 60
$ LHS - Acceleration at Driver's seat
DRESP1 1033 ACC1033 RMSaccl 3 620 1033
$ RHS - Acceleration at Passenger's seat
DRESP1 2033 ACC2033 RMSaccl 3 620 2033
$
$ sum of RMS accelerations at Driver's and Passenger's seats
DRESP2 1020 sumrms 1020
dresp1 1033 2033
DEQATN 1020 object(driver,pass) = driver + pass
$
DRESP1 9105 ACC9005 PSDACCL 620 3 MAX 9005
DCONSTR 101 9105 2.5e-3
DOPTPRM DESMAX 20 P1 1 P2 15 conv1 5.e-3

Each AESO job requires two separate runs: an AESO creation run and an AESO assembly run.
To activate an AESO creation run, you need to add the following user input to a regular optimization job (bold face in
Listing 2): 1) an FMS ASSIGN statement that specifies the file name for the assembly run that will be generated from
the AESO creation run and 2) a keyword on DOPTPRM entry, autose = 1 that activates an AESO creation run.
CHAPTER 29 491
Rapid Road Response Optimization of a Camaro Model using Automatic External Superelement

Listing 2 Required user inputs for activating AESO creation run

assign aeso='test9_2.dat'
.....
begin bulk
doptprm desmax 5 p1 1 P2 15
delx 0.2 delp 0.8 autose 1

After the creation run is complete, search for the user information message 9181 in the f06 file that indicates a
successful run.

^^^
^^^ USER INFORMATION MESSAGE 9181 (FEA)
^^^ THE JOB IS TERMINATED FOR AN AUTO EXTERNAL CREATION RUN
^^^

The input file for the assembly run (aeso9_2.dat) is automatically generated from the creation run. Its Bulk Data
section contains the residual model (or the design model) while the Control Section is the identical copy from the
original optimization job. Some special contents in an assembly run are shown in bold face in Listing 3. The FMS
ASSIGN statement references the Nastran Master database file and the DBLOCATE statement identifies the data block
that contains various boundary matrices. The INCLUDE statement includes an assembly file that include boundary
connection data. Notice that the AUTOSE = 1 request on the DOPTPRM entry added for the creation run has been
changed to AUTOSE = 0.

Listing 3 Special contents in an assembly file

nastran buffsize= 65537


nastran rseqcont=1
assign se1= './test9.MASTER'
dblocate datablk(EXTDB) logical=se1, CONVERT(SEID=1)
SOL 200
CEND
......
BEGIN BULK
include './test9.asm'
DOPTPRM DESMAX 5 P1 1 P2 15
DELX 0.2 DELP 0.8 AUTOSE 0

Figure 29-3 shows that the sum of RMS is reduced from the initial value of 0.154 to the final of 0.130 by the road
response optimization. Table 29-1 compares the accuracy of the results and performance in terms of Clock time
between the regular approach and the AESO approach and clearly shows that the AESO is able to obtain the same final
design but with one fifth of the time spent by a single shot run.
492 MD Demonstration Problems
CHAPTER 29

5.0E -0 3

4.0E -0 3
S u m m ed A c celeratio

3.0E -0 3
S UM Init
S um final
2.0E -0 3

1.0E -0 3

0.0 E +0 0
4 6 8 10 12 14
F req u en cy (H z )

Figure 29-3 Sum of RMS Reduced from 0.154 to 0.130

Table 29-1 Results and Performance Data for Case A


Initial Final Init. Max Init. Max # Design Clock Time
Case A OBJ OBJ Const Const Cycle (Minute)
Single Shot Run 0.1534 0.0639 0.1329 -0.2102 9 37

AESO Creation 5
Run
AESO Assembly 0.1534 0.0639 0.1319 -0.2102 9 1
Run
ASEO Total 6

Performance 6
Ratio
CHAPTER 29 493
Rapid Road Response Optimization of a Camaro Model using Automatic External Superelement

Case B
This case minimizes the RMS acceleration at Driver’s seat and maintains frequency dependent limits on PSD
acceleration at driver’s seat by varying nine spring constants of the engine mount. Listing 4 shows the required design
model set up for Case B.

Listing 4 Design Model Set up for Case B

...
DESPBJ = 1033
DESSUB = 101
...
BEGIN BULK
$ design model set up

$ Desin model set up


$
DESVAR 11 K5307 1.0 0.01 3.0
......
DESVAR 24 K5018 1.0 0.01 3.0
DVCREL1 5307 CELAS2 5307 K
11 1246.3
......
DVCREL1 5018 CELAS2 5018 K
24 1120.
$ LHS - Driver's seat to floor (Response for Objective to be minimized)
DRESP1 1033 ACC1033 RMSaccl 3 620 1033
DRESP1 1133 ACC1033 PSDACCL 620 3 1033
DCONSTR 101 1133 1133
DOPTPRM DESMAX 20 P1 1 P2 15 conv1 5.e-3
TABLED1 1133
0.0 1.0e03 6.0 1.0e-3 7.0 1.7e-3 8.0 1.7e-3
12.0 2.0e-4 endt

Notice that in Case B, the design objective now is to minimize an RMS acceleration at Driver's seat while limiting
maintaining the frequency dependent limits on the PSD acceleration at Driver seat. The rest of the analysis model is
kept the same. Therefore, the outputs from the creation run for Case A can be reused here except replacing the
objective and constraints for Case A (Listing 1) with that for Case B formulation (Listing 4).
Figure 29-4 shows that the RMS acceleration at Driver's seat is reduced from the initial of 0.071 to the final of 0.058.
Table 29-2 compares the accuracy of the results and performance dat between the regular approach and the AESO
approach. Again, AESO achieves the same final design as the single shot run. Since no creation run is required because
it can reuse the results from the Case A's creation run, the speed up by the AESO run vs. a single shot run for Case B
is a factor of 33.
494 MD Demonstration Problems
CHAPTER 29

3.0E-03
P asse n g er A cce lera

2.5E-03

2.0E-03
2033 Init
1.5E-03
2033 Final
1.0E-03

5.0E-04

0.0E+00
4 6 8 10 12 14
Frequency (Hz)

Figure 29-4 RMS Reduced from 0.071 to 0.058

Table 29-2 Results and Performance Data for Case B


Initial Final Init. Max Final Max # Design Clock Time
Case B OBJ OBJ Const Const Cycle (Minute)
Single Shot Run 0.0713 0.0586 0.2855 -0.0025 14 33

AESO Creation 0
Run
AESO Assembly 0.0713 0.0584 0.2855 -0.0201 9 1
Run
ASEO Total 1

Performance 33
Ratio

Case C
This case is exactly the same as Case A except the frequency response is solved by the Direct Frequency Analysis
Solver. Specifically, the ANALYSIS=MFREQ Case Control command in Case A is replaces by ANALYSIS=DFREQ
command in Case C.
Therefore, the same discussions presented for Case A can be directly applied here. Table 3 compares the results and
performance data between a single shot run and shows the relationship to Case C. Again, the final design from AESO
agrees well with that from a single shot run. However, the performance gain by AESO is a factor of 40.
CHAPTER 29 495
Rapid Road Response Optimization of a Camaro Model using Automatic External Superelement

In fact, the assembly run could be run directly by assessing the database file and asm file and the assembly run file
generated from the creation run for Case A since these files are identical if ANALYSIS=MFREQ or ANALYSIS=DFREQ.
Therefore, the performance gain would be a factor of 244 (i.e. 244=244/1) assuming the time spent by the assembly
run for Case B is still five minutes.

Table 29-3 Results and Performance Data for Case C


Initial Final Init. Max Init. Max # Design Clock Time
Case A OBJ OBJ Const Const Cycle (Minute)
Single Shot Run 0.1535 0.1327 -0.0631 -0.2073 9 244

AESO Creation 5
Run
AESO Assembly 0.1534 0.1327 -0.0636 -0.2062 9 1
Run
ASEO Total 6

Performance 40
Ratio

Modeling Tip
This section provides some guideline or modeling tips for performing AESO tasks:
• Only database option is supported in AESO. No op2 or punch option is supported.
• The nondesigned part is treated as a single part component and can not be further partitioned.
• The performance gain achieved by an AESO job depends on the size of the analysis model, the ratio of the
design model size vs. the analysis model size and number of boundary points shared by designed part and
nondesigned part. A general rule of thumb is that the relative ratio should be less than 10%. The smaller the
ratio, the more performance gain can be achieved.
• The UIM 7824 from the creation run lists the size of your analysis model and design model (in terms of
number of the grid points). DRATIO may be adjusted for a larger or smaller residual model.

• Submit the AESO creation run with SCR=NO option to store the Nastran database. An assembly run does not
require that option.
• It is recommended to use Matrix domain based domain decomposition (domain solver acms(partopt=dof) for
large scale normal modes or model frequency tasks, say the total number of degrees of freedom is half million
or higher.
496 MD Demonstration Problems
CHAPTER 29

• ASSIGN AESO=’fn’ is required in the creation to define a file name of the assembly file. Directly assigning
the original job name to filename should be avoided. A good practice is to add some suffix to the original file
name such as myjob_2nd.dat where myjob is the original file name.
• General guidelines or limitations to the manual External Superelement analysis also apply to AESO.
• Refer to the MD Release Guide for more guidelines and limitations.

Input File(s)
Case A
File Description
nug_29.dat BDF for an AESO run of Road Response Optimization

Case B
File Description
nug_29b.dat BDF for an AESO run of Road Response Optimization.

Case C
File Description
nug_29c.dat BDF for an AESO run of Road Response Optimization