Anda di halaman 1dari 26

September 25, 2002

Probabilistic Design
Extra! Extra!
Training
How can PADT Help?
ANSYS Focus Archives
Subscribe / Unsubscribe
In the Next Focus

September 25, 2002


Probabilistic Design System
by Rod Scholl, PADT

Overview
At some point, most ANSYS users have had to vary some parameter in their model to
investigate impact on results. Perhaps you’ve done this through the GUI and multiple runs – or
maybe you’ve written a macro that Do Loops through a few input values. ANSYS
Probabilistic Design System (PDS) automates the process.

Using simple menu picks (or commands) you can specify many input variables and their
variation in statistical terms (Gaussian, Weibul, etc.) and then ANSYS manages the many runs
that sample the “terrain”. You can choose the sampling method (Random, Monte Carlo, etc,).
Finally ANSYS builds an .html report (via the Automatic Report Generator) that graphs the
specified output variables. It will even evaluate correlations between inputs and outputs! This
is the best application of the Report Generator that I’ve encountered.

An Ideal-World Case Study:


Your materials person gives you Gaussian distributions for your properties. Your dimensions
are specified with tolerances as well as distributions based on SPC generated data at the
vendor. (This is why this is titled “Ideal-World” case…). You also receive boundary
conditions that are in terms of means and standard deviations assuming a Gaussian distribution.

Tolerance
Distributions
Baseline Model
Report

Mat. Props

σy
ANSYS PDS

Load / Temp
Distributions

September 25, 2002


You set up the model as usual with a single case, but then specify inputs statistically in the
PDS module. ANSYS runs the model 1,000 times (I’ve seen people do as much as 40,000)
and summarizes the results in a report. You can then provide a distribution of peak stress for
your lifing person. Unless you’re the lifing person, and then you can specify a probability of
life rather than a discrete number of cycles!

A Real World Case Study:


You are given the model of a flange with fixed dimensions and material properties. However,
the deflections are unknown at the inner and outer radii. Non unusually, you must know
characterize the stress before the deflections are finalized.

The Procedure:

1. Build a database that solves for a single case and verify results.
2. Change the database to have the BC’s applied using variables.
3. Specify statistical distributions of the input variables using PDS.
4. Specify desired output variables using PDS.
5. Instruct PDS to solve the model multiple times.
6. Review the report generated by PDS.

1. Build a database that solves for a single case.


(You should cut and paste this text to a file then name it FLANGE.INP) You can then read in
the file into a clean database using Utility Menu > File > Read Input From…

/PREP7
/title,Axisymmetric Flange (PDS by PADT)
UIMP,1,EX, , ,11.13948e6, !Sets Material Properties
UIMP,1,NUXY, , ,.3326,
K,1,6,2 !Draws Outline
k,2,6,0
k,3,4.5,0

September 25, 2002


k,4,4.5,.1
k,5,5.95,.1
k,6,5.95,2,
LSTR, 2, 3 !Makes Lines
LSTR, 2, 1
LSTR, 1, 6
LSTR, 5, 6
LSTR, 5, 4
LSTR, 4, 3
LFILLT,1,2,.05, ,
LFILLT,5,4,0.1, ,
al,5,8,4,3,2,7,1,6 !Makes Areas
et,1,82 !Meshes
keyopt,1,3,1
esize,.0175
amesh,all

/solu
dl,6,,ux,.05 !Apply Deflection Loads
d,346,,.025,,,,ux
d,2,,0,,,,uy
solve

/post1
PLNSOL,s,eqv
lsel,s,,,6 !Store Output Variables
nsll,s,1
nsort,s,eqv

*get,INNRSTRS,sort,0,max

lsel,s,,,8
nsll,s,1
nsort,s,eqv
*get,CORNSTRS,sort,0,max
allsel

1. Cont’d – Verify Results – They should look as follows:

September 25, 2002


2. Change the database to have the deflections applied as variables.

Original
Modified
/solu /solu
dl,6,,ux,.05 !Apply Deflection Loads INNERDF=.05
d,346,,.025,,,,ux OUTERDF=.025
d,2,,0,,,,uy dl,6,,ux,INNERDF !Apply Deflection Loads
solve d,346,,OUTERDF,,,,ux
d,2,,0,,,,uy
solve
You should be able to rerun your macro and get the same results!!!

3. Specify Statistical Distribution of the Input variables using PDS.


3.1) Open the Probabilistic Design Module

September 25, 2002


3.2) Specify Random Input Variables

Choose OUTERDF
(Notice ANSYS “found” your
Variables.)

Accept the default


Gaussian distribution

Use These Values

September 25, 2002


3.2 Cont’d) Do the same for the variable INNERDF. Rather than Guassian, choose
Gamma:

Use these
Use These Values
Values

September 25, 2002


4) Specify Desired Output Variables. Note that we had to define these FIRST using *GET’s at
the end of our original macro -- PDS will not do this for you!

Add INNRSTRS also.

September 25, 2002


5) Instruct PDS to solve the model multiple times. PDS is designed to distribute these runs
over multiple computers in your company. It will compile all the results back on your own PC.
This distributed computing is not too difficult to set up – but for this example we will run them
all on our own machine.

Use the Method Wizard…

• Skip past the first screen with the clock hitting NEXT
• Accept default of No by hitting NEXT
• Select the 1st three boxes as shown below then hit NEXT

• Skip past the last screen by hitting FINISH

There are several other dials to turn, but these defaults will work in our case. The other bells
and whistles are fairly self-explanatory; just browse around someday when you are bored…

September 25, 2002


5) Setting up the PDS runs Cont’d…

You must “Assign” to ANSYS PDS what macro file to execute. This is the input file we
named FLANGE.INP and then modified to have variables control the deflection.

DON’T FORGET THIS NEXT STEP!!!!


You must comment out(using the !)the variable definitions in your macro.
Or else, PDS will run the same version 100 times!

Original Modified
/solu /solu
INNERDF=.05 ! INNERDF=.05
OUTERDF=.025 ! OUTERDF=.025

(or you could have just deleted the lines of code…)


(don’t forget to save the file!)

Another Note – if you used any /CLEAR’s in a macro – you must get rid
of them

Yet Another Note – if you used any RESUME’s in a Macro you must
preface it with PARSAV and PARRES to help ANSYS track the PDS
variables.

September 25, 2002


5) Now you are ready to go… Just a few more clicks…

September 25, 2002


6) Generate Report
Use whatever settings you like, I chose these just for kicks…

September 25, 2002


Below are a few samples from the automatically generated report. You can view
the entire report online. Also, there are other files stored such as parameter
history for each run, etc., which are not addressed here. See the ANSYS
documentation for details.

EXAMPLE REPORT:

Axisymmetric Flange (PDS by PADT)


By
Date: September 18, 2002; Time: 15 : 44 : 38

An ANSYS Probabilistic Analysis Report


Generated by the ANSYS Probabilistic Design System

Introduction
Deterministic Model Information
Probabilistic Model Information
Probabilistic Analysis Summary
Regression Analysis Summary
Probabilistic Analysis Results

I. Introduction
The ANSYS Probabilistic Design System (PDS) analyzes a component or a system involving
uncertain input parameters. These input parameters (geometry, material proerties, boundary
conditions, etc.) are defined in the ANSYS model. The variation of these input parameters are

II. Deterministic Model Information


The analysis file containing the deterministic model is stored in FLANGE.INP . ANSYS
executes or "loops through" multiple times during the probabilistic analysis.

September 25, 2002


Table 2 . Material Properties
Modulus of Elasticity 1.1139E+07

Poisson's Ratio 0.3326

III. Probabilistic Model Information

Random Input Variables


The deterministic model has 2 parameters that are regarded as random input variables. A list of
these random input variables, the distribution they are subjected to, and their distribution
parameters are provided in Table 3 . The probability density functions and cumulative
distribution functions are shown in Figures 2 to 3 .
Table 3 . Random Input Variable Specifications
No. Name Type Par1 Par2 Par3 Par4

1 OUTERDF GAUS 3.50000E-02 2.50000E-02

2 INNERDF GAMA 2.0000 75.000

September 25, 2002


Figure 2. PDF & CDF of Input Random Variable OUTERDF

VI.1 Statistics of the Probabilistic Results

Result Set FLANGE-BY-PADT


Table 5 . Result Set FLANGE-BY-PADT - Statistics of the Random Input Variables

Standard
Name Mean Skewness Kurtosis Minimum Maximum
Deviation

-7.5428E- -2.5259E-
OUTERDF 3.4971E-02 2.4709E-02 -0.2193 9.3599E-02
03 02

2.6657E-
INNERDF 1.8893E-02 1.354 2.388 1.7564E-03 0.1037
02

Table 6 . Result Set FLANGE-BY-PADT - Statistics of the Random Output Parameters

Standard
Name Mean Skewness Kurtosis Minimum Maximum
Deviation

CORNSTRS 4.7380E+04 3.3403E+04 1.358 2.396 3756. 1.8358E+05

INNRSTRS 7.7384E+04 5.4640E+04 1.352 2.382 5338. 2.9996E+05

September 25, 2002


Figure 11. Result Set FLANGE-BY-PADT - Histogram of Input Variable INNERDF

Figure 13.
Result Set FLANGE-BY-PADT - Histogram of Output Parameter INNRSTRS

September 25, 2002


Result Set FLANGE-BY-PADT

Table 7 . Result Set FLANGE-BY-PADT - Linear Correlation Coefficients Between Input


And Output Variables

Out\Inp OUTERDF INNERDF

CORNSTRS -0.060 1.000

INNRSTRS -0.061 1.000

Table 8 . Result Set FLANGE-BY-PADT - Spearman Rank Order Correlation


Coefficients Between Input And Output Variables

Out\Inp OUTERDF INNERDF

CORNSTRS 0.013 1.000

INNRSTRS 0.012 1.000

Acknowledgement: Thanks to Stefan Reh, from


whom’s presentation I swiped some graphics!
Please contact Rod Scholl with questions

Back to Table of Contents

September 25, 2002


Extra! Extra!
What's New . . .
à Enhanced Technical Support
à Heal Method
à ANSYS Offers Probabilistic Design
Seminars
à Advanced Analysis Applications: Impact,
Metal Forming, HF EMAG, and MEMS
à The Next Focus
Archives

ANSYS Focus is a periodical publication provided by Phoenix


Analysis & Design Technologies (PADT).

September 25, 2002


Enhanced Technical Support
by Rod Scholl, PADT
A part of my day I most enjoy is supporting “Enhanced Technical Support” agreements.
We provide this service to a host of customers under other names like “Mentoring”,
“Customization”, etc. I suppose I enjoy it, because I usually get to help people in a minor
jam. I hear Tow Truck drivers have similar experiences.

Are you are ever working on ANSYS and think, “It would save me a lot of effort if only
someone would stop by my office and ____

__ help me get this thing to converge…


__ help me get results by 3 PM today…
__ help me modify this macro…
__ write a generic macro—then I can pick it up and tweak it from there…
__ teach me how to use this new feature, X…
__ help me check for errors…
__ put another set of eyes on this, just to be sure…?”

Many of us have resources like that in our own companies (and maybe you’re one of them) –
of course, these folks are usually quite popular and often it is hard to get their time. I feel
lucky working at an ANSYS reseller, because I always have an expert nearby to help with my
own ANSYS analyses:

Daniel Larusso (Karate Kid): Yeah, but you knew karate.


Mr. Miyagi: hmph . . . someone always know more.

Some customers have enough users that we make regular visits, drop by
offices and sit with people. Other customers purchase a chunk of time
like 10 hrs, and we use it over several visits, or phone calls as the
demand arises. The best benefit of all this, is that “Enhanced Technical
Support” not only helps you achieve your goal, it prioritizes your
learning. We want to make you the expert.

Combining your understanding of your analysis goals, process and


skills, with PADT’s knowledge of ANSYS, analysis methods and
automation, allows us to realize huge time-savings. Often a 15-minute
discussion, explanation, script debugging, macro-writing, etc. will save someone an
afternoon of irritation. Set this service up before you run into trouble – then we’re
only a phone call away!

Back to List of Articles


September 25, 2002
~Heal Method
by Rod Scholl, PADT

There have been some changes in the relationship between


ANSYS, Inc. and Cadfix (Transcendata). Some results of
these changed are:
• “Cadfix For ANSYS” is no more.
• Automatic Geometry Healer (AGM) is a
new product sold by ANSYS, Inc.
• CADFix is sold by Transcendata
• “Cadfix for ANSYS” customers will
receive licensing for both CADFix and the
Automatic Geometry Healer (AGM)
through the end of their TECS agreement. (Contact your ANSYS ASD for
details).

The Automatic Geometry Healer (AGM) is invoked by a single command ~HEAL. There is
no menu selection (but should be in ANSYS 7.0). When ~HEAL is issued, ANSYS will
import the file and attempt to repair using the AGM. The AGM is based on “CADFix for
ANSYS” technology, but it is all run in batch mode. There is no user interaction.

~HEAL, Name, Extension, Path, Method

The key to the ~HEAL command is in the Method option. You have the choice of either
Method=ANSYS or Method=Cadfix. If you’re geometry is being imported via a connection
product (Pro/E, Catia, UG, Para, Sat, Ideas, etc.) then you will use Method=ANSYS. If you’re
file is an .IGES file, you will use Method=Cadfix.

Method=ANSYS
Upon issuing the ~HEAL command with the Method set to ANSYS, the file will be
imported using your connection product for which you need a license. Then the
resulting .anf file is processed by the Automatic Geometry Healer (AGM), for which
you need a license. NOTE: You must have CADFix 5.0 installed locally, even though
you don’t need a license for it.

Method=Cadfix
Upon issuing the ~HEAL command with the Method set to Cadfix, the file will be
repaired using the Automatic Geometry Healer (AGM), for which you need a license.
NOTE: You must have CADFix 5.0 installed locally, for which you need a license!

I have tested the Method=ANSYS on a Pro/E model. The AGM was unable to remove slivers,
sew surfaces, repair poor Cad-tolerancing problems, etc; albeit was a fairly poor model.

Perhaps it would also not have been repairable with Cadfix for ANSYS. The nice part was that
without any options to set, it only takes one try with the AGM to fail! The bad part was that it
September 25, 2002
took several hours for the repair and several more hours to read in the geometry; albeit it was a
somewhat complex 3D model. I have not tested the Method=Cadfix option.

Recommendations
If you use a connection product (Pro/E, Catia, UG, Para, Sat, Ideas, etc.), we recommend that
you test out your class of CAD generated geometry using the AGM (~HEAL). If you do not
have success, you may want to evaluate a seat of CADFix from Transcendata.

If you do not use a connection product, and simply use .IGES, hopefully your geometry comes
in clean. If it does not (big surprise), then you will need a CADFix license to repair geometry,
regardless of whether you use AGM. If you can take advantage of the automation provided by
AGM, compared to the learning curve to operate CADFix alone, you may want to purchase a
seat of AGM as well.

Back to List of Articles

September 25, 2002


Whether it’s one of our regularly-scheduled classes in our training room or a customized class
tailored to your specifications and location, you can be assured that the training you receive
will have immediate, positive results on your capabilities in design, analysis, and product and
process improvement.

See the upcoming schedule below. Or, learn more about how training at PADT can save you
time and money. (Or drop an e-mail to Ted Harris!)

Month Start End Course Description Location


Oct '02 3-Oct 3-Oct 103 DesignSpace 6.0 Irvine, CA
7-Oct 9-Oct 101 Introduction to ANSYS, Part I Tempe, AZ
10-Oct 11-Oct 301 Heat Transfer Tempe, AZ
15-Oct 17-Oct 201 Basic Nonlinearities, Advanced Tempe, AZ
Contact, and Bolt Pretension
21-Oct 23-Oct 101 Introduction to ANSYS, Part I Los Angeles, CA
24-Oct 25-Oct 801 Introduction to ANSYS Tempe, AZ
Customization
28-Oct 29-Oct 701 Design Optimization and Tempe, AZ
Probabilistic Design
30-Oct 1-Nov 601 Introduction to FLOTRAN Irvine, CA
Nov '02 4-Nov 6-Nov 101 Introduction to ANSYS, Part I Tempe, AZ
7-Nov 8-Nov 301 Heat Transfer Irvine, CA
11-Nov 13-Nov 201 Basic Nonlinearities, Advanced Los Angeles, CA
Contact, and Bolt Pretension
14-Nov 15-Nov 501 ANSYS/LS-DYNA Tempe, AZ
18-Nov 20-Nov 101 Introduction to ANSYS, Part I Irvine, CA
21-Nov 22-Nov 203 Dynamics Tempe, AZ
25-Nov 26-Nov 602 FLOTRAN II Los Angeles, CA

Back to List of Articles

September 25, 2002


Subscribe / Unsubscribe
The ANSYS Focus is a periodical published by PADT. It contains information
about ANSYS products and services provided by PADT. Freely distribute the
ANSYS Focus in its entirety.
Subscribe to ANSYS Focus by sending an e-mail to
focus@padtinc.com with Subscribe as the subject.

Unsubscribe to ANSYS Focus by sending an e-mail to


focus@padtinc.com with Unsubscribe as the subject.

For administration questions please e-mail Rod Scholl.

Contribute
Send in your contribution! Articles, and information helpful to the user are welcomed and
encouraged. E-mail Rod Scholl.

ANSYS Focus is a periodical publication provided by


Phoenix Analysis & Design Technologies (PADT).

Back to Table of Contents

September 25, 2002


Next in Focus

ANSYS Focus is a periodical publication provided by


Phoenix Analysis & Design Technologies (PADT).

Back to List of Articles

September 25, 2002


The Nitty-Gritty

PHOENIX ANALYSIS AND DESIGN TECHNOLOGIES


(PADT) MAKES NO REPRESENTATIONS ABOUT THE
SUITABILITY OF THE INFORMATION CONTAINED IN
THESE DOCUMENTS AND RELATED GRAPHICS FOR
ANY PURPOSE. ALL SUCH DOCUMENTS AND
RELATED GRAPHICS ARE PROVIDED "AS IS"
WITHOUT WARRANTY OF ANY KIND AND ARE
SUBJECT TO CHANGE WITHOUT NOTICE. THE
ENTIRE RISK ARISING OUT OF THEIR USE REMAINS
WITH THE RECIPIENT. IN NO EVENT, INCLUDING
INACCURATE INFORMATION, SHALL PADT BE
LIABLE FOR ANY DIRECT, CONSEQUENTIAL,
INCIDENTAL, SPECIAL, PUNITIVE OR OTHER
DAMAGES WHATSOEVER (INCLUDING WITHOUT
LIMITATION, DAMAGES FOR LOSS OF BUSINESS
PROFITS, BUSINESS INTERRUPTION, OR LOSS OF
BUSINESS INFORMATION), EVEN IF PADT HAS BEEN
ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
THE VIEWS EXPRESSED IN THE ANSYS FOCUS ARE
SOLEY PHOENIX ANALYSIS AND DESIGN
TECHNOLOGIES AND NOT THOSE OF ANSYS, INC.
Designed by: Rod Scholl
Prepared by: Julie Canepa

ANSYS Focus is a periodical publication provided by


Phoenix Analysis & Design Technologies (PADT).

Back to List of Articles

September 25, 2002

Anda mungkin juga menyukai