Industry Spotlight:

Consumer
Products
Page 16

Hardware Platforms... How

fast can they handle typical
ANSYS workloads?

Run CAE applications

efficiently to get large
jobs done quickly

Build 3-D crack meshes

CFX Update is now more easily with specialized
modeler
part of ANSYS Solutions!
 Contents
Industry Spotlight Departments
16 Consumer Products Editorial 2
A continuing series on the value
of engineering simulation in Industry News 3
specific industries
CFD Update 5

Features CAE Community 13

20 Benchmarks for Simulation at Work 15

Hardware Platforms
How fast can they handle Software Highlights 32
typical ANSYS workloads?
Guest Commentary 34

22 Tuning Unix Systems for Tech File 36

High Performance
Make sure your operating system Tips and Techniques 39
runs CAE applications efficiently
to get large jobs done quickly

26 3-D Crack Analysis About the cover

Build fracture meshes more
There are many examples of
easily with specialized successful consumer product Industry Spotlight:

companies using ANSYS

modeler that acts as a front simulation technology to
end to ANSYS develop winning products.
Our cover article describes
how Allsteel, Inc. used
simulation to shorten the
29 Managing Computing Resources design cycle by orders of
Software coordinates computing workload magnitude for a new-from-
the-ground-up office chair.
using a set of resources distributed across
a network.

For ANSYS, Inc. sales information, call 1.866.ANSYS.AI, or visit www.ansys.com on the Internet
To subscribe, go to www.ansys.com/subscribe.

Editorial Director Designers Ad Sales Manager Editorial Advisor

John Krouse Miller Creative Group Ann Stanton Kelly Wall
jkrouse@compuserve.com info@millercreativegroup.com ann.stanton@ansys.com kelly.wall@ansys.com

Managing Editor Art Director Circulation Manager CFD Update Advisor

Jennifer L. Hucko Paul DiMieri Elaine Travers Chris Reeves
jennifer.hucko@ansys.com paul.dimieri@ansys.com elaine.travers@ansys.com chris.reeves@ansys.com

ANSYS Solutions is published for ANSYS, Inc. customers, partners, and others interested in the field of design and analysis applications.

The content of ANSYS Solutions has been carefully reviewed and is deemed to be accurate and complete. However, neither ANSYS, Inc., nor Miller Creative Group guaran-
tees or warrants accuracy or completeness of the material contained in this publication. ANSYS, DesignSpace, ANSYS DesignModeler, ANSYS DesignXplorer VT, ANSYS
DesignXplorer, ANSYS ProFEA, ANSYS Emax, ANSYS Workbench environment, Multi-field, CFX, AI*Environment, CADOE S.A. and any and all ANSYS, Inc. product names
referenced on any media, manual or the like, are registered trademarks or trademarks of subsidiaries of ANSYS, Inc. located in the United States or other countries. All other
product names mentioned are trademarks or registered trademarks of their respective manufacturers.
POSTMASTER: Send change of address to ANSYS, Inc., Southpointe, 275 Technology Drive, Canonsburg, PA 15317, USA ©2004 ANSYS, Inc. All rights reserved.

www.ansys.com ANSYS Solutions | Spring 2004

 Editorial

2
Welcome Back
Redesigned print version returns better than ever
Welcome back to the print In this same vein, a link between CFX computational
version of ANSYS Solutions fluid dynamics and ANSYS Multiphysics enables a seam-
magazine, with a redesigned less transfer of CFD mechanical and thermal loads to
look and layout delivered in a ANSYS for stress analysis. Don’t miss the CFD Update,
hard copy you can easily “What’s New in Computational Fluid Dynamics,” on page 5.
carry around, leaf through, This special section is a new addition to the magazine,
pass around, and file away for covering this fascinating and rapidly emerging engineering
future reference. Of course, simulation technology.
all the content for this issue, Also covered in the Software Highlights article are two
as well as past editions, will more new solvers directed at parallel processing: the Dis-
By John Krouse still be posted in portable tributed Preconditioned Conjugate Gradient solver (DPCG)
Editorial Director document format (PDF) at and the Distributed Jacobi Conjugate Gradient solver
www.ansyssolutions.com to (DJCG). These solvers enable users to more fully take
ANSYS Solutions
take advantage of the speed advantage of distributed computing resources, a topic also
jkrouse@compuserve.com
and convenience of the Inter- addressed in the article “New Solvers in ANSYS 8.0 Push
net. We’ll continue to publish news, overview articles, the Envelope” on page 32.
user stories, software A wide range of other
reviews, product informa- ENGINEERING SIMULATION TOOLS advancements are included in
tion, application tips, and ANSYS 8.0, including Compo-
other material directed
SUCH AS ANSYS 8.0 DO MORE THAN nent Mode Synthesis for large
toward design engineers SAVE TIME AND MONEY — THEY FACILITATE dynamic problems, enhanced
using simulation as part of contact capabilities, paramet-
product development, INNOVATION, IMPROVE QUALITY AND PUT ric mesh solutions, and greater
professional analysts ap- BRAND VALUE INTO PRODUCTS. integration and automation
plying the technology on a features in Workbench. Log
daily basis, and engineering managers overseeing these onto www.ansys.com for further details. When you take a
operations as part of overall product development. look at software such as this coming out of ANSYS, Inc.,
We want to be responsive and flexible in deciding it’s easy to see why the company is doing so well while
how our editorial content can best serve your needs, so others in this market appear to be struggling.
we invite your feedback. Send your comments and con- Theoreticians undoubtedly find such advances math-
tributions to me at jkrouse@compuserve.com. Together, ematically intriguing, and market-watchers will hail this as
we can continue building a publication that will make a another big step forward in the industry. From a practical
valuable contribution to the engineering community and perspective, analysts and engineering professionals using
provide a resource we hope you will find useful, issue the software can see significant gains in productivity, with
after issue. some tasks being done in minutes instead of hours or days.
As always, the publication is focused on software It lets companies tackle larger, more complex problems
applications for mechanical simulation and engineering faster than ever and reduce costly physical testing.
processes to help readers understand and apply finite- But engineering simulation tools such as ANSYS 8.0
element analysis (FEA) and other simulation technology do more than save time and money — they facilitate inno-
developed and supported by ANSYS, Inc. In this issue, vation, improve quality, and put brand value into products.
be sure to take a look at the Software Highlights article, In this way, engineering simulation translates directly into a
“New Solvers in ANSYS 8.0 Push the Envelope,” on page competitive advantage for companies with enough sense
32 for a good insight into the way the software develop- and foresight to make the necessary investment in the
ers have come up with new tools and blended them into technology, as well as the skilled people in their organiza-
the overall mix of FEA capabilities. tion putting these tools to work. ■
The new Multi-Field solver provides a framework for
handling coupled-field problems where interdependent
physics must be taken into account to come up with a
realistic simulation: complex applications where struc-
tural and thermal behavior interact in electronic circuits,
for example, or fluid flow and acoustics in products rang-
ing from hydraulic valves to automobiles.

www.ansys.com ANSYS Solutions | Spring 2004

Industry News
Recent Announcements
ANSYS Announces Channel Partner Relationship
with Texas Engineering
Texas Engineering Systems, which provides engineer-
ing software and services in Texas, Arkansas, and
Louisiana, will offer the ANSYS suite of engineering
simulation products to companies throughout the
region. ANSYS and Texas Engineering Systems will
work together to satisfy the needs of companies
requiring solutions in areas like up-front simulation,
complex nonlinear analysis, and coupled physics for
any of the major CAD systems utilized today.
For more information, visit www.texaseng.com.
ANSYS CFX Participates in FLOMANIA Project
ANSYS CFX is an active participant in the FLOMANIA
Project. The primary purpose of this project is to
support the European aeronautics industry in the
area of flow-physics modeling by bridging the gap
between the most advanced research results in the
area of turbulence modeling and their application in
daily industrial practice.
The project is considered to be a unique attempt to
group the best experts in research and CFD develop-
ers in Europe to transfer their advanced knowledge to
the aeronautical industry in a controlled, objective-ori-
ented way.
Proceedings are available from the FLOMANIA pro-
ject’s recent workshop on detached-eddy simulation
(DES) turbulence modeling under Downloads on
the FLOMANIA Web site at cfd.me.umist.ac.uk/
flomania/index2.html.
ANSYS Research Version Available for
Educational Users
ANSYS has introduced ANSYS Research Multiphysics
for educational users who require more node or ele-
ment capability or who receive grants from commercial
companies or government agencies to do complex
studies. The version provides five tasks per license
www.ansys.com
with restrictions. Usage must be for information that
will be available in the public domain and is strictly lim-
ited to nonprofit applications; it cannot be used for
consulting work. Users of the ANSYS Research ver-
sion are required to submit a case study to ANSYS
Corporate Marketing on a yearly basis.
The Multiphysics application provides Mechanical,
including Structural Nonlinear, Structural Dynamics,
Structural Buckling and Thermal, CFD, and Electro-
magnetic for low- and high-frequency analysis.
Additional functionality includes the ability to solve
using coupled physics and prep/post-processing,
solvers, and a report writer.
For information on pricing and capabilities, please
contact John Kiefer, Education Sales Coordinator, at
724.514.3079 or john.kiefer@ansys.com, or call your
ANSYS Channel Partner.
Upcoming Events
ANSYS Seminar Series — Come Out and Learn!
With the aim of providing ongoing education to exist-
ing and future customers, ANSYS and its partners are
conducting a seminar series focusing on ANSYS tech-
nologies as well as best practices.
Now Showing:
• CFD Modeling & Post-Processing Featuring
ICEM CFD
• Coupled-Field Simulations Featuring ANSYS
Multiphysics
• Engineering Simulations for Six Sigma
• Up-Front Design and Optimization Featuring
ANSYS DesignSpace & DesignXplorer VT
• Nonlinear Analysis Featuring ANSYS Structural
For more information, visit www.ansys.com/seminars
or contact your local ANSYS sales office.
ANSYS Solutions | Spring 2004
 Industry News

ITI Manta Announces Durability and
Fatigue Seminars
4 ITI Manta, a subsidiary of International TechneGroup
Incorporated (ITI), is offering a series of Durability and
Fatigue seminars in the Greater Cincinnati area in
Spring 2004. The courses will provide an introduction
to modern theories of fatigue and their practical appli-
cation. Attendees will learn about material stress/
fatigue, optimizing designs to reduce fatigue, and
technologies available to analyze models predicting
areas of concern.
Durability by Design
Date: March 29, 2004
Fee: No charge
This seminar is designed to illustrate problems that
now can be solved using durability analysis software
for FEA models. A technology update in key areas of
fatigue analysis such as multiaxial fatigue, effects of
temperature, notch sensitivity, and residual stresses
will be provided.
Design Engineers responsible for product durability in
the ground vehicle, and aerospace and manufacturing
industries should attend. A basic knowledge of fatigue
or attendance in the "Fundamentals of Modern Metal
Fatigue Analysis" listed above is recommended.
The seminar will cover:
• Introduction to biaxial fatigue
• Biaxial fatigue using equivalent stress/strain
• Critical plane analysis
• Fatigue lives from finite element models
• Statistics and reliability
• Crack growth analysis
• Fatigue of welded joints
• Fatigue lives from the PSD
• Fatigue test and design
• Additional advanced topics
For more seminar information or to register, visit
www.safetechnology.com/events.html or contact ITI
Manta’s Rick Aveline at 513.248.5224, 800.783.9199,
or rick.aveline@iti-global.com.

The seminar will cover:

• An introduction to modern methods of metal SAE 2004 World Congress
fatigue analysis March 8 – 11, 2004
• Theory overview of uniaxial and multiaxial fatigue Detroit, MI
• Case studies and tools demonstrations www.sae.org/congress
• An introduction to using fe-safe with FE software
• Highlights of advanced and unique capabilities in Lockheed Martin Conference
fe-safe such as a fully integrated signal processing March 17 – 19, 2004
capability, high temperature and creep fatigue, fe- Sunnyvale, CA
rotate, analysis of complex loading histories
2004 International ANSYS Conference
Fundamentals of Modern Metal Fatigue Analysis Profiting from Simulation: Business and Technical
Date: March 30 – 31, 2004 Leadership through Simulation-Driven Design
Fee: $495 (attend all four Modern Metal Fatigue May 24 – 26, 2004
Analysis days for $895) Pittsburgh, PA
www.ansys.com/conf_2004
Design and Test Engineers responsible for product
durability in the ground vehicle, aerospace and manu-
facturing industries should attend. No previous knowl-
edge of fatigue is necessary.

The seminar will cover:

• Introduction to fatigue
• Local stress-strain fatigue
• Materials data
• Signal analysis for fatigue
• Using stress-life data
• Examples of fatigue failures and fatigue tests
• Geometric effects
• Fatigue analysis in practice

Advanced Modern Metal Fatigue Analysis

Date: April 1 – 2, 2004
Fee: $495 (attend all four Modern Metal
Fatigue Analysis days for $895)

www.ansys.com ANSYS Solutions | Spring 2004

CFD Update: What’s New in Computational Fluid Dynamics

CFX Simulation for Predicting Aircraft Drag

Accurate results from a fully coupled formulation and
algebraic multi-grid solver
By Florian Menter and Robin Lantry, ANSYS CFX

AIAA (American Institute of Aeronautics and All of the grids were of the low-Re type with y+
Astronautics) has held two workshops in the last three values of the order of one. Transition was specified in
years to evaluate the ability of modern CFD methods the numerical tests according to the transitions strips
to accurately predict lift and drag of complete aircraft on the wind-tunnel models. The grid refinement study
configurations. The first workshop was held in 2001 showed that the main solution parameters did not
and had an active participation of 18 CFD groups. A change between the medium and the fine grid. All sim-
DLR-F4 wing-body configuration test case was used ulations shown are computed on the medium grid.
with structured and unstructured grids provided by the
organization and by ANSYS ICEM CFD. The results of
this first comparison showed a wide spread between
the results produced with different codes and turbu-
lence models. Even different codes with the same tur-
bulence model produced significantly different results.
As a consequence of this outcome, it was decided to
hold a second workshop on the same topic, but with
different test cases.
The second workshop was held in Orlando on June
21 and 22, 2003. Due to experience gained by participa-
DLR-F6 configuration used as test case at the 2003 AIAA Drag
tion in the EU aeronautics project FLOMANIA, CFX from Prediction Workshop. Left: wing body (WB); Right: wing body-
ANSYS was able to compute the matrix of specified test nacelle-pylon (WBNP)
cases. The geometries selected were the DLR-F6
wing-body (WB) and DLR-F6 wing-body-nacelle-pylon
(WBNP) configuration. The goal of the comparison was
to compute lift, drag and pitching moments of both con-
figurations for a series of seven angles of attack.
To ensure the quality of the numerical results, par-
ticipants also were required to perform a grid refine-
ment study for each configuration. The structured grids
provided by ANSYS ICEM CFD were:

Grids Coarse Medium Fine

WB 3,450,000 5,820,000 10,130,000
WBNP 4,890,000 8,430,000 13,690,000

Table 1: Number of nodes for different grids for WB and WBNP

Lift vs. angle of attack for WB test case. Angles of attack were
configuration -3°, -2°, -1.5°, -1.0°, -0.0°, 1.0°, and 1.5°. Modification to the SST
model, which mimics the adverse pressure gradient performance of
the S-A model, produced an appropriate shift as shown by red triangle.

www.ansys.com ANSYS Solutions | Spring 2004

 CFD Update: What’s New in Computational Fluid Dynamics
Numerical Method
All simulations were computed with CFX-5.6. CFX-5 is
a finite, volume-based CFD method for unstructured
hybrid meshes. It features some of the most advanced
CFD techniques available today, namely a fully coupled
6 formulation in combination with an algebraic multi-grid
solver. For a pressure-based method, this combination
has proven essential to obtain high-quality converged
results for the current transonic test cases. CFX-5 is
highly optimized in its physical modeling. The current
simulations have been carried out using the SST turbu-
lence model in combination with the CFX-automatic
wall treatment.
Lift vs. angle of attack α for WBNP test case
Drag polar for WB and WBNP test case
Wall pressure distributions at two span-wise locations for WBNP test case (yb/0.377 and yb/0.847) show the shock at the upper surface.
www.ansys.com
Results
The computed lift, Cl, vs. the angle of attack, a, for the
wing-body test case (WB) is in very good agreement
with the experimental data over the entire a-range.
Surprisingly, it was found that the simulation of the lift
curve proved difficult for most other codes in the com-
parison. In many cases, the lift curves were shifted by
▲ α~0.25-0.5°, which is a significant deviation in aero-
nautical simulations. As the majority of these simula-
tions have been computed with the Spalart-Allmaras
(S-A) one-equation turbulence model, the sensitivity of
the simulations to the turbulence model formulation
has been investigated after the workshop using CFX.
It was found that a modification to the SST model,
which mimics the adverse pressure gradient perform-
ance of the S-A model, did indeed produce a shift of
the expected magnitude.
The results for the lift-curve for the WBNP case
are again in good agreement with the experiments,
except for negative angles of attack. Compared
to most of the other codes, CFX was still quite close
to the data. All simulations showed a tendency of
over-predicting the lift at negative angles of attack.
The main reason seems to be an over-prediction of
a separation at the wing-pylon intersection, which
affects the circulation around the wing. This issue is
currently being investigated in more detail by several
of the participating groups.
For an airplane, the drag is one order of magni-
tude smaller than the lift. It is therefore a notoriously
difficult quantity to compute. The agreement of the
CFX results with the experimental data for the drag-
polar for both configurations is therefore surprisingly
good for all angles of attack. It was possible to clearly
identify the additional installation drag due to pylon
and nacelle. The CFX results could therefore satisfy
the main demand to accurately predict lift and drag of
complete aircraft configurations using CFD posed by
the workshop.
The shock on the upper surface, which is cap-
tured quite well by the simulations, can clearly be
identified in the chart of surface pressure distributions
at different span-wise locations on the wing of the
WBNP case.
ANSYS Solutions | Spring 2004

Separation details on top-side of wing. Left: experiment with oil flow. Right: CFX-5 simulation. Region of reverse flow can be identified.
In the CFX analysis of flow details on the upper
side of the wing for the WBNP configuration, small
regions of flow reversal, which are in good agreement
with experiment, can be identified.
Numerical Performance
One of the main limiting factors in CFD simulations for
complete aircraft configurations is the relatively high
computing requirements. It is therefore important that
the nonlinear iterative CFD procedure converges with
high efficiency. For the lift and the drag force for the
WB case, CFX typically converged for all cases in
about 100-150 iterative steps. This was one of the
lowest iteration counts in the workshop comparison,
Convergence of forces for WB test case
www.ansys.com
7
although the comparison is not straightforward due
to the many different numerical schemes involved.
Due to the scalable parallelization of CFX-5, simula-
tions for a grid of about 6 million nodes can be con-
verged overnight on a modern 16-process PC cluster.
Conclusions
CFX from ANSYS has participated successfully in
the validation study of AIAA. The studies showed
that pressure-based methods are very well suited
for the simulation of aeronautical flows at transon-
ic conditions. Of the 25 participants, CFX software
produced one of the most consistent sets of
results in good general agreement with the experi-
mental data. ■
Web sites for More Information
Links
Results of First AIAA CFD Drag Prediction Workshop,
Anaheim, CA June 9–10, 2001
aaac.larc.nasa.gov/tsab/cfdlarc/aiaadpw/
Workshop1/Final_Schedule_and_Results.html
FLOMANIA
cfd.me.umist.ac.uk/flomania/
Results of Second AIAA CFD Drag Prediction Workshop,
Orlando, CA June 21–12, 2003
aaac.larc.nasa.gov/tsab/cfdlarc/aiaadpw/
Workshop2/Final_Schedule_and_Results.htm
ANSYS Solutions | Spring 2004
 CFD Update: What’s New in Computational Fluid Dynamics

8
Improving Ventilation at
Grand Central Terminal
CFX simulation helps study
air flow in one of Manhattan’s
largest underground structures.
Grand Central Terminal (GCT) serves as the principal
hub of the MTA Metro-North Railroad, the second
largest commuter railroad in the United States, with
approximately 240,000 customer trips each weekday
and some 70 million trips per year. When GCT opened
for business in 1913, its Beaux-Arts terminal was one
of two major stations in New York City for long dis-
tance rail travel. But as rail travel in the United States
decreased in the 1950s, the private railroad that
owned the terminal began to neglect upkeep and main-
tenance, and its condition deteriorated rapidly. In 1997,
a $175 million restoration project was completed.
Years of grime were removed from the ceilings and
stairs and the station was restored to its original beauty.
Many new shops and eateries have since opened.
The GCT concourses lead to the “trainshed,” one
of the largest underground structures in Manhattan.
The trainshed has 30 platforms on two levels.
Ventilating the trainshed has become increasingly
difficult over the years as widespread use of air-
conditioned equipment increasingly added waste heat
into a facility designed long before the age of
air-conditioning. Ventilation is currently provided
by sidewalk grills and a few, small vent shafts. feet occupied by the trainshed and the fact that out-
Temperatures at ground level in the trainshed are side air can only be reached in limited areas because
typically about 15 degrees Fahrenheit above ambient of the dense construction above ground. Because of
during the summer months. Improving passenger the unique design challenges, several changes that
comfort is but one of several reasons why efforts are had been made in the past at considerable expense
being made to improve ventilation, although passen- ended up having no major positive impact. To avoid
gers generally only move through the platform for a repeating that experience, Hatch Mott McDonald, an
few minutes before they enter the air-conditioned engineering firm based in New York, was contracted
trains. Another concern is improving the comfort of to conduct a preliminary study using computational
the MTA employees who spend even more time in the fluid dynamics (CFD) to explain the fluid and thermo-
trainshed. A primary concern is increasing the service dynamic processes that drive environmental condi-
life of apparatus on the trains that must contend with tions in the trainshed.
high ambient temperatures. “The initial study with CFX highlighted the advan-
Computer simulation is being used to predict the tages of computer simulation by correlating well with
benefit of a wide range of proposed changes to the existing conditions and confirming the effects of
ventilation system in the Grand Central Terminal train- recent ventilation system changes,” said Norman
shed, making it possible to improve conditions in a Rhodes, project manager for Hatch Mott McDonald.
cost-effective way. The trainshed, located in midtown “As a result, we won a much larger contract to design
New York City, experiences high air temperatures a new ventilation system. We are currently evaluating
during the summer months. Improving ventilation is a the cost and benefits of a wide range of potential
difficult challenge because of the 2.5 million square design improvements, which will make it possible to

www.ansys.com ANSYS Solutions | Spring 2004


Grand Central Terminal in New York City serves as the principal hub of the
MTA Metro-North Railroad, the second largest commuter railroad in the
United States, with approximately 240,000 customer trips each weekday.
obtain the most benefits for the money that is avail-
able to improve the ventilation system.”
Initial Design Study
Hatch Mott MacDonald is an award-winning, full-
service consulting engineering firm offering public and
private clients multi-disciplined expertise and compre-
hensive capabilities in planning, environmental
assessments, studies and analysis, design, procure-
ment, construction engineering and inspection,
project, program and construction management and
facility maintenance and operations. The company ini-
tially won a contract for a preliminary study during
which global lumped-parameter and 3-D CFD models
of the trainshed were developed and applied to under-
stand the current ventilation conditions and the impact
of changes that have been made in recent years. A
lumped-parameter model uses coefficients that are
assumed to apply over the entire region being mod-
eled. Its advantage is computational simplicity. CFD
provides far more accuracy and is the primary tool for
evaluating the various design options.
www.ansys.com
9
“With CFX, we can
be confident that we
will deliver a solution
that not only works
but gives our client
the biggest bang
for the buck.”
Rhodes and his team used the background
drawings of the trainshed as-built conditions as
source for the CFD model. To reduce the computa-
tional requirements, he created two separate models
for the upper and lower levels of the trainshed and
treated the results of each as a boundary condition for
the other. The greatest challenges were the excep-
tionally large area covered by the model and the many
construction details involved. Rhodes selected CFX
software as the CFD modeling tool for this project.
“We have found CFX accuracy, robustness, and
scalability to be ideally suited to the large and chal-
lenging modeling problems that we address on a daily
basis,” Rhodes said. “For example, CFX provides an
extremely comprehensive user subroutine capability
that allows us to create fully programmable user func-
tions. We used this capability extensively on the project
to improve the accuracy of our model in simulating flow
through the grills and similar features.
ANSYS Solutions | Spring 2004
 CFD Update: What’s New in Computational Fluid Dynamics
10 “During this study we were able to explain to our
client various observations that had been made but
were difficult to understand,” Rhodes said. “For exam-
ple, fans in an office building directly above the train-
shed that were previously used to supply air were
changed to extract air during the early 1990s. Some of
the people in the building had the perception that the
environment was better when the fans were set to
supply. The model explained that perception by accu-
rately predicting that the change lost the benefit of
a region of cooler inlet air. The excellent correlation
of the phase one model helped to generate confi-
dence in computer simulation and led to the award of
a $2.3 million contract for the design of a new venti-
lation system.”
Working under the current contract, Rhodes and
his team greatly increased the detail of the model to
provide the accurate predictions required for making
decisions. One of the most critical areas was found to
be the sidewalk grilles, which present a challenge
because, at a width of about three feet, they are on a
very small scale in relation to the rest of the model. Yet
their impact is great because they represent such a
large proportion of the available venting. Hatch Mott
McDonald analysts also refined the heat sources in
the model, which consist of the trains themselves, the
air conditioners on the trains, and the thermal inertia of
the buildings above the trainshed.
Critical Design Challenges
“Once we completed the model, our next step was to
validate it by comparing its predictions to temperature
and humidity measurements along the platform,”
Rhodes said. “We also compared the air currents pre-
dicted by the model to observations of the smoke flow
when small trash fires occur in the trainshed. The
model correlated very well to the actual physical
CFX analysis shows the temperature difference in the trainshed at 46th Street under original conditions (base 46th) and
if grates were added (grates46th).
www.ansys.com
results, so we are now in the process of using it to
assess the impact of a wide range of design changes.
The critical design challenge is optimizing the design
by providing the most ventilation efficiency per dollar
expended. While we are still in the midst of our design
study, it’s already clear that the most effective
approach will be one that combines more grille space
with active ventilation. From a big-picture standpoint,
the main trade-off is adding additional sidewalk grilles
versus putting in more ventilation. We have found that
the sidewalk grilles have a very positive impact —
temperatures are generally lower to the north of the
grille. On the other hand, we also have looked at
the impact of increasing air supply, which generally
has a smaller impact but affects a larger area.
An important principle that we have established is the
need to move air from the south end to the north end
of the trainshed so that it can flow out of the
sidewalk grilles.”
The combined effects of these changes are so
subtle and complex that many different model itera-
tions will be required in order to obtain an optimized
design. “There is no cheap fix to this problem,”
Rhodes concluded. “We need to take advantage of
every possible tool at our disposal. If our funding were
unlimited, this would be easy. The big challenge is
determining which combination of changes will have
the greatest impact on environmental conditions at
the lowest possible cost. With conventional engineer-
ing calculations and physical experiments alone, we
would have had to essentially guess at the answer.
CFX makes it possible to base our design decisions
on facts rather than opinions. We are in the process of
evaluating every reasonable configuration, estimating
the cost of each and determining the impact on train-
shed conditions. With CFX, we can be confident that
we will deliver a solution that not only works but gives
our client the biggest bang for the buck.” ■
ANSYS Solutions | Spring 2004
CFX for Complex
Nuclear Safety Analysis
A pioneering study uses CFX 11
to evaluate reactor cooling
By Attila Aszodi and Gabor Legradi
Budapest University of Technology and Economics

Three-dimensional CFD codes are expected to play an essential role in the

safety analysis of nuclear reactors, similar to the role of one-dimensional
system codes currently used. In the Institute of Nuclear Techniques of the
Budapest University of Technology and Economics, our CFX team has
examined many of the components of the VVER-440 type nuclear reactors,
including the fuel assembly, the fuel assembly head, and the reactor pres-
sure vessel itself.
A pioneering CFD study recently addressed a comprehensive safety
analysis of a complete nuclear system. In this study, CFX was used to exam-
ine the natural circulation processes between the reactor pressure vessel
(RPV) and the cooling pond of VVER-440 type reactors under the special
circumstances that arise during maintenance. During annual maintenance,
the RPV and the cooling pond form a connected flow domain, and some The modeled geometry. After a calculation for the
of the fuel assemblies are transported underwater from the reactor to the normal operational state, the Inlet II and Outlet II
cooling pond. Because of radioactive decay, a significant amount of heat is boundaries were closed.
generated by the fuel assemblies located in the reactor and the cooling
pond. The reactor is cooled by natural circulation through one or two steam
generator heat exchangers. The cooling pond has its own cooling loops. The
study was performed to determine whether it would be possible to cool the
reactor if the steam generators failed and other emergency systems were
not available.
In large water tanks, intensive circulation can be caused by even very
small temperature differences. The main goal was to investigate whether this
natural circulation would be sufficiently strong to allow the cooling system of
the pond to cool the entire system. The transient calculation that was per-
formed showed that the cooling pond system as currently designed would
not be capable of removing the heat produced in the RPV. Modifications
were also studied and proposed to ensure that the cooling pond system
would be capable of removing the heat generated in the reactor under these
circumstances.
Using CFX, natural circulation was investigated in an extremely large
domain and complex geometry. Typically, the calculations covered more Velocity isosurface of 10 cm/s velocity during the
transport of the fuel assemblies using the discharge
than 10,000-second-long transients. This study has shown clearly that CFX pipe chunks as outlet. The isosurface is colored
is a very effective tool for performing safety analyses of complex nuclear by the axial velocity components, so the red shows
systems. ■ up-flow, the blue down-flow.

Temperature values in the top of the transfer pond

(Left) Temperature field during the transport of the fuel assemblies with heat and the reactor core during the transport, and the
exchanger failure. transient calculated by the assumption that the
(Right) Temperature field during the same event using the “discharge pipe chunks” coolant outlet was replaced by the discharge pipe
as outlet. chunks at the bottom of the transfer pond.

www.ansys.com ANSYS Solutions | Spring 2004

 CFD Update: What’s New in Computational Fluid Dynamics
Superior
12
Technology
Continuing advancements in the CFX
family of products make the best
technology accessible to customers.
By Michael Raw
Vice President of Product Development,
Fluid Business ANSYS CFX
CFX-5 software from ANSYS, Inc. contains the best
computational fluid dynamics (CFD) technology avail-
able today. CFX-5 was created and continues to be
enhanced to meet the needs of industry. It incorpo-
rates the flexible models of CFX-4, the reliable solver
and state-of-the-art rotating machinery capabilities of
CFX-TASCflow, and years of experience from industry
professionals who understand the requirements of
CFD software in the real world. CFX-5 is used in a
broad range of industries including aerospace,
automotive, HVAC, manufacturing, power generation,
chemical processing, turbomachinery, electronics,
and biomedicine.
The core technology in CFX-5 is a superior cou-
pled multigrid solver that provides speed, robustness,
accuracy, and scalability by solving the full hydro-
dynamic system of equations simultaneously across
all grid nodes. For large meshes — which are increas-
ingly required in today’s engineering simulations — this
translates into significant and predictable reductions in
run times and faster project turnarounds. Such pre-
dictability and speed are critical for efficient integration
of CFD into design optimization systems. The solver
also allows for exceptional parallelization that is highly
scalable, providing faster answers with fewer comput-
ing resources.
The ability of the solver to perform when
interequation coupling is significant allows for
interoperability of nearly all models, delivering the
most complete and accurate results possible.
In addition to progressive models, including
multiphase, heat transfer, radiation, combustion, and
real fluid, CFX-5 delivers innovative solutions for
turbulence modeling — the most fundamental and im-
portant physical model for many industrial flows. The
Shear Stress Transport (SST) model available within
CFX-5 has been extensively validated and provides
automated near wall treatment. It is as economical as
the k-ε model, but offers much higher fidelity, giving
excellent answers on a wide range of flows. CFX-5
also offers full Reynolds Stress models based on the
ω-equation with automatic wall treatment and a newly
developed zonal Detached Eddy Simulation (DES)
formulation based on the SST model.
www.ansys.com
The flexibility of the CFX system can adapt easily
to the preferred geometry creation and meshing
solution of the user and might include ANSYS Design-
Modeler or ANSYS-ICEM CFD products. The new
CFX-Pre physics pre-processor streamlines analysis
setup through a self-guiding interface, the ability to
read one or more meshes from a variety of sources,
options for assigning domains, and a customizable
Model Library of physics definitions. CFX-Post pro-
vides a full suite of tools for quantitative and qualitative
CFD post-processing. Throughout CFX-5 are tools to
enable integration into design processes, faster com-
pletion of repetitive tasks, customization, and easier
convergence.
The superior performance of CFX-5 is not based
on any single feature alone. Rather, it is the combina-
tion of proven, leading-edge technology in all elements
of the software that provides the speed, reliability,
accuracy and flexibility that companies trust to make
them successful. ■
Grid Generation for Rotating Applications
CFX-TurboGrid is a powerful and user-friendly grid genera-
tion software tool for rotating applications. It creates high-
quality hexahedral mesh using an automatically created
customized mesh topology for the blade geometry being
studied. This allows accurate and fast CFD analysis.
Designing and Evaluating
Turbomachinery Blades
CFX-BladeGen provides turbomachinery designers with a
complete system for three-dimensional blade design and
evaluation. It is easily coupled to lower-level analysis tools,
3-D structural analysis software, or optimization technology.
The combined offering provides a complete solution for
design and analysis of rotating machinery components.
ANSYS Solutions | Spring 2004
2004 International ANSYS
Conference Preview
Want to profit from simulation?
Want to use your ANSYS solutions more efficiently and effectively?
Want to make product development more automated,
more integrated and more innovative?
Engineering simulation is changing fast. Every day,
it’s becoming more automated, more integrated
and more innovative. Simulation has the ability to
drive product development, reduce costs and get
products to market more quickly.
Today, companies can profit from this new wave of
simulation — if they use the right technologies at the
right time.
To thrive in the marketplace, it’s essential to attend the
2004 International ANSYS Conference, Profiting from
Simulation: Business and Technical Leadership
through Simulation-Driven Design, May 24-26, at the
Pittsburgh Hilton and Towers, Pittsburgh, Pa.
As the only global conference dedicated to the true
simulation of multiphysics, the 2004 International
ANSYS Conference is a unique opportunity for
attendees to learn about ANSYS’ strategy for
the future of engineering design simulation and
analysis. Plus, attendees have the chance to gain an
nderstanding of ANSYS, ANSYS ICEM CFD and CFX,
and how the integration of these technologies offers
an unprecedented level of capability in one solution.
www.ansys.com
13
■ Gain insight into ANSYS vision and strategy and how
the integration of ANSYS, ANSYS ICEM CFD and
CFX offers an unprecedented level of capability in one
solution
■ Learn about product development best practices
from ANSYS, Inc., LMS International, Hewlett-
Packard Company, Herman Miller, Inc. and other
prominent companies
■ Discover how to improve products and processes
by using automated and integrated simulation to
drive design
■ Listen to more than 100 customer presentations
from industries including automotive, aerospace,
consumer products, electronics and many more
■ Exchange ideas with ANSYS, ANSYS ICEM CFD
and CFX professionals from companies throughout
the world
■ Join other managers, process-owners and decision-
makers during the management track to learn how
leading companies are unlocking the potential of
computer simulation technology
Lead your company into the next generation of
simulation-driven development. You can’t afford to
miss this event!
Register at www.ansys.com/conf_2004/solutions.htm
today.
ANSYS Solutions | Spring 2004
 Agenda

2004 International ANSYS

Conference Preview
14

Venue General Session Speakers

Hilton Pittsburgh and Towers Brad Butterworth (Keynote Speaker)

600 Commonwealth Place Tactician, Team Alinghi
Gateway Center
Jim Cashman
Pittsburgh, PA 15222
President and CEO
T 412.391.4600
ANSYS, Inc.

Hotel rooms are provided at the Hilton Pittsburgh and Larry Larder
Towers at a discounted rate of $125 for single and dou- Director-Engineering Services
ble rooms. Special accommodation rates include $160 Herman Miller
for a towers single room, and $300 for a suite. Rates
apply until May 3, 2004. Bruce Toal
Director of Marketing and Solutions
Note: All currency is in U.S. Dollars for High Performance Technical Computing Division,
To Make Reservations at the Hilton Pittsburgh and Hewlett-Packard Company
Towers, visit www.ansys.com/conf_2004/venue.htm
Dr. Urbain Vaneurzen
Chairman and Chief Executive Officer
LMS International

Michael Wheeler
Vice President and General Manager,
Mechanical Business
ANSYS, Inc.

Visit www.ansys.com/conf_2004/solutions

for a complete listing of technical sessions,

management tracks and more!

www.ansys.com ANSYS Solutions | Spring 2004

Simulation at Work
ANSYS Analyses Help Stanley
Works Speed Reengineered
Impact Socket Wrenches to Market
When market researchers at the Stanley Mechanics Tools Group realized that
mechanics and assembly line workers wanted slimmer sockets for their powered
hand tools, engineers at this Dallas, Texas, unit of The Stanley Works, New Britain,
Connecticut, solved the challenges with ANSYS.
The problem was that existing socket sets were too
fat for use in the cramped access spaces of many new
products. In brief, Stanley’s marketing research found
that customers wanted impact socket wrenches with
exterior diameters no larger than the sockets used
with hand socket wrenches. Hand-tool sockets were
always slimmer because muscle-powered tools exert
far lower stresses on the sockets than power-driven
tools.
Stanley engineers found that with redesigned
geometry, the reduced exterior dimensions—thinner
walls and thus the sockets’ outside diameter—led
to stress increases of less than 10%. “That left us
with a more than adequate margin of safety,” said
A.J. Garg, senior project manager. He used ANSYS
sensitivity analyses on the impact stresses due to
velocity, mass, energy and the tool’s air pressure.
“Mechanics in the study groups also told our
marketing people they prefer impact sockets with the
longest possible life in addition to the thinnest possi-
ble walls and narrowest outside diameter,” said Garg.
Although produced in the tens of millions a year
by manufacturers around the globe, apparently no
one had ever bothered to analyze sockets for socket
Figure 1. Stress Simulation of an Impact Socket
www.ansys.com
15
wrenches thoroughly. Searches of the engineering
literature turned up no significant analyses of impact-
wrench sockets.
A key initial remedy was to reduce the socket’s
O.D. from 1.625 to 1.5 inches, a difference of 7.7%.
That gave mechanics an additional clearance of just
under 0.063 inch on either side. Despite this
seemingly small gain, the mechanics reported they
were pleased with the additional tight places they
could reach into.
However, reduced diameter led to reduced
structural stiffness in the socket and resulted in high-
er impact stresses. “Theoretically, the impact velocity
at the fillets could be reduced with tighter tolerances
in the manufacturing tools,” Garg said. “But the addi-
tional cost of the tooling plus increased needs for its
maintenance, downtime and rework were not cost-
effective.”
That meant the strains of the increased operating
velocity on the fillets in the thinner socket walls had to
be accommodated with an improved design.
“The primary analyses were nonlinear, contact
and impact,” Garg said. “Twenty different socket sizes
were analyzed, thanks to the speed of the parametrics
in the ANSYS Parametric Design Language (APDL).
The ANSYS models were correlated with strain-gauge
testing of an impact socket in Stanley’s physical-
testing facilities.
“Without these ANSYS tools, we would have
been forced to use a lot more trial-and-error,” Garg
said. “We would have had to build many more proto-
types and conduct a lot more physical tests. We have
the facilities to do those tests here, but they are time-
consuming and prototypes are costly. The power
hand tools business is very competitive in both price
and time-to-market. We wanted to make sure we
seized the advantages of being first to market with a
new product.
“Also, physical tests cannot tell you anywhere
near as much as FEA,” he added. “That means ANSYS
greatly reduced our risks in starting initial production
as quickly as we did.” ■
ANSYS Solutions | Spring 2004
 Industry Spotlight

16
Industry Spotlight:
Consumer Products
A continuing series on the value of engineering
simulation in specific industries.

Think about all the products purchased in retail sales transactions

at stores, showroom floors, by phone, on the Web. All the many home
appliances, hand tools, toys, furniture, telephones, pagers, personal
computers, electronic gizmos, TVs, stereos, lawn and garden products, and sports and recreation equip-
ment from golf clubs to tennis rackets. These and other manufactured items comprise the giant consumer
products market, one of the world’s largest industry segments and the pounding engine that drives the
global economy.
These products are essential to our modern lifestyle, and hardly a week goes by that we all don’t
contribute substantially to the billions of dollars spent annually in this growing market. Manufacturers vie
aggressively for these dollars, and the rapidly shifting market can be outrageously lucrative for a few and
unforgivingly brutal to most others. Quality products that are innovative, first to market and affordable, and
that hit the mark on customer expectations, can be gangbusters, grabbing the lion’s share of the market
and building brand value that carries through to future offerings. Those that are second-rate in any one
of these areas generally turn out to be ho-hum performers and often fade into obscurity soon after they’re
unveiled. Others invariably fall by the wayside because they fail to measure up to strict standards
monitored by organizations and agencies such as the Consumer Product Safety Commission.
The stakes are high in consumer product industries, with few second chances and the survival
of companies at risk with each new product launch. Companies at the forefront of this market put a
heavy emphasis on product development, using some of the most advanced simulation tools to evaluate
alternatives, weed out bad ideas, spot and fix mistakes in the early conceptual stages of development,
and optimize designs so products can sail through testing and production to get to market quickly.
There are many examples of successful consumer product companies using ANSYS simulation
technology to develop winning products. Other case studies can be found at www.ansys.com about
projects at companies such as 3M, Maytag, Rubbermaid, Segway, Speed Queen, Wilson Sporting Goods
and Whirlpool.
These and other companies remain leaders in the consumer product industry in large measure
because they value the expertise of their technical professionals and put engineering simulation to
good use in leveraging the corporate intellect of their product development teams. Proof of this is their
continued strength and market success in one of the most competitive industries in the world.

www.ansys.com ANSYS Solutions | Spring 2004

Case-in-point:
Office Furniture Company Shortens Design Process,
Quickly Completes Project
The Allsteel #19 office chair, with 18 major components, considers the human to be part number 19.
The chair was analyzed extensively with ANSYS Mechanical to ensure style, function and reliability.
Executive Summary
Challenge
To develop the “perfect” office chair that
exceeds standards for strength and durability
Solution
Use ANSYS Mechanical to thoroughly
analyze all critical components
Benefits
• Shortened design process by at least
three orders of magnitude
• New-from-the-ground-up project was
completed on time
• Chair met all design objectives for style,
function, and reliability
• Chair passed all tests with minimal
redesign
www.ansys.com
17
By Robert Mills
for ANSYS, Inc.
For Allsteel, Inc., the mission was nothing less than
to create the perfect office chair. It’s called #19, as
it has 18 major components, with the human as
part number 19. The chair, introduced in March
2002, was a new-from-scratch effort to not only
create a product but also to uniquely define the
Allsteel brand.
Despite the name, the Allsteel #19 chair
is made primarily of cast recycled aluminum.
Indeed, the chair has the feel of a finely crafted air-
craft. Half of its weight comes from its aluminum
frame, a skeleton of sorts onto which the seat, back
and armrests are attached. Even the armrest sup-
ports and base are made of aluminum. The chair,
which starts at $1,350, has achieved critical
acclaim in the industrial design arena.
As with many projects at the company, the
use of ANSYS Mechanical software from ANSYS,
Inc. of Canonsburg, Pennsylvania, was a given
from the start. “We are not new kids on the block
when it comes to ANSYS,” said Emad Tanbour, a
senior project manager at HON Technology Inc., a
division of HON Industries, the parent company to
Allsteel. He said the company uses ANSYS on a
daily basis. He roughly estimates that in the last
five years, the company has used ANSYS for 3,000
different simulations. Tanbour has a Ph.D. in engi-
neering and is also an adjunct professor in
mechanical engineering at the University of Iowa in
Iowa City. He is co-authoring a textbook on FEA.
ANSYS Solutions | Spring 2004
 Industry Spotlight
Challenge
18
The chair design started from a clean slate.
The team included people from management,
engineering, quality assurance, manufacturing,
marketing and sales. The chair’s lead designer
was Marcus Koepke, the founder of Marcus Cur-
tis Design, Indianapolis. Since 1991, he has been
developing self-balanced, weight-activated chairs
(known as passive ergonomic chairs). The team
used rapid prototyping to evaluate many solu-
tions to design problems.
For the computational analysis group, their
part of achieving the team’s mission was to ensure
that the chair was strong and reliable. “For the
first time, we are offering a lifetime warranty, so
the chair has to be reliable,” Tanbour said.
Chair design may not appear to require
extensive finite element analysis (FEA). But then
again, #19 is no ordinary chair. “It’s a very beau-
tiful chair with a lot of complicated geometry,”
Tanbour said. “There isn’t a straight edge on it.”
The base, which is a variation of the classic five-
leg star base, is one example. In addition to their
sweeping curves, the arms of the chair’s base are
somewhat U-shaped in cross section and include
a complex ribbing pattern to strengthen them.
What’s more, #19 has several complex
mechanisms, resulting in more variations that
the analysis had to take into account. In one
version of the chair, the mechanism for raising
and lowering the seat uses a two-cylinder design
to accommodate a wider range of users. Most
chairs use only one cylinder.
The tilt mechanism isn’t a simple hinge and
spring. Instead, when the sitter leans back, a
four-bar mechanism tilts the rear of the seat up
so that the sitter’s eyes remain at the same level.
A mechanism in the armrests allows them to be
raised and lowered with a push of a button. The
armrests also swivel laterally in and out in multiple
positions. Even the armrests themselves are a
mechanism of sorts, in that they are cantilevered
to flex with the sitter’s arms.
These were not all of the project’s chal-
lenges. “Every component is custom designed,
even the casters and fasteners,” Tanbour said.
“There isn’t an off-the-shelf component in the
whole chair.” The effort was also under time pressure.
www.ansys.com
“We faced the challenge of getting to market fast.
Our company introduces new products to market
at an outstanding pace,” Tanbour said.
For the computational analysis team, the
proof would be in the testing. The chair had
to meet strict standards set by the Business and
Institutional Furniture Manufacturer’s Association,
or BIFMA. This organization demands two types
of testing — function loading and proof loading.
What’s more, Allsteel requires that its products
exceed the BIFMA standards by 10 percent. The
tilt mechanism, for instance, underwent 200,000
cycles of front-to-back tilt testing, the equivalent
to tilting the chair every six minutes for 10 years.
A heavy-bag drop-test involves dropping a 300-lb.
weight from six inches above the seat.
Solution
Because of the demands of the project and the
complexity of the chair, Tanbour ran more than
300 analysis cases on the chair. “Some projects
that involved assembly analysis spanned a week
to four weeks,” he explained, while others were
done in a day. The different cases were run for
various reasons, such as to evaluate different
design solutions or for quality assurance. “No part
was tested before we analyzed it,” Tanbour said.
Most of the analysis was static, linear, although
some nonlinear analysis was done on nonmetallic
parts.
For this project, especially, Tanbour and his
team focused on doing numerous analysis runs to
optimize the design. Using a direct connection
to Pro/ENGINEER, the company’s CAD system,
the analysis team brought the geometry directly
into ANSYS. Then, they used ANSYS’ smart
meshing capability to create the finite element
mesh. “We do not invest a lot of time on meshing
our designs until later stages of analysis,”
Tanbour said, adding that the smart meshing
function works well.
Using the direct connection, Tanbour said,
results in a better-quality model and requires less
time for geometry repair. Later, when the design
started to become more defined, Tanbour’s group
also did manual meshing. Tanbour said some of
the FEA models for the chair required more than
300,000 elements, while smaller parts were modeled
with 20,000 to 30,000 elements.
ANSYS Solutions | Spring 2004
“We trust ANSYS and the theory behind it. No doubt about it.
It’s been in the market for 30 years. There’s a lot of confidence
in ANSYS. It’s a very well utilized asset in our company,”

Some parts were particularly challenging.
For instance, the seat cushion is made of a
patent-pending technology that is a combination
of foam, Technogel (a special material made by
a German company), and a microporous textile
upholstery. A solid shell was needed to support
this cushion and meet strength requirements.
Tanbour used ANSYS to optimize the seat shell
design, taking into account geometrical variations
such as ribbing. He even used ANSYS to help
choose the proprietary material used in the shell.
“Typical stress analysis projects can be
implemented daily due to the impressive solvers
at ANSYS and our state-of-the-art hardware,”
Tanbour said. He currently uses a Compaq Evo
W8000 with dual 2.2 GHz XEON processors and
four gigabytes of RAM, running under Windows
2000. He also is able to dial into the analysis
from his home computer to keep runs moving
overnight. “That saves us a lot of time,” he said.
“We trust ANSYS and the theory behind it.
No doubt about it. It’s been in the market for 30 19
years. There’s a lot of confidence in ANSYS. It’s a
very well utilized asset in our company,” Tanbour
said.
The extensive analysis even played a role in
marketing the Allsteel #19 chair. For the product
launch, a virtual reality show was created to
demonstrate the chair and how it was designed,
engineered and manufactured. Animations from
the ANSYS analysis — showing deflections — and
CAD models were incorporated into this produc-
tion, which was created by VizTek Inc., Iowa City,
Iowa. The ANSYS animations were included in
a DVD made to promote the product as well,
showing how the chair was thoroughly analyzed
and tested. ■

Benefits
The effort helped minimize dependence on
prototyping and repeated testing of the Allsteel
#19 chair. “We rely heavily on ANSYS stress
analysis and use it extensively before investing in
prototyping or testing,” Tanbour said.
Don’t talk to Tanbour about how the project
might have progressed had it not been for
ANSYS; it’s almost as if it’s too difficult for him to
consider. “You just couldn’t design some of these
components by hand,” he said as he explained
how ANSYS was used to make the chair’s base
strong and light. As such, Tanbour doesn’t know
specifically how much time using ANSYS on this
particular project saved. But in general, Tanbour
said, “Using ANSYS shortens the design process
by at least three orders of magnitude.”
“The use of FEA on this project is a happy
story,” Tanbour continued. He explained that
using an outside consultant on a project of this
magnitude would be impractical. “In essence, we
did millions of dollars’ worth of FEA on this
project. In-house is the only way to go for smart
companies.”

The challenge to the #19 chair analysis was the

unique curved shape of the frame.

www.ansys.com ANSYS Solutions | Spring 2004

20

The speed at which a computer solves various simulation

problems has a significant impact on project deadlines and
user productivity. More broadly, analysis-processing speed
also can influence how the technology is implemented within
a company’s product development processes, since fast
simulations are required for running routine “what-if”
simulations and seamlessly integrating analysis into the
design process. The type of platform selected to run these
applications is thus a critical issue in building a company’s
overall CAE hardware/software solution.

www.ansys.com ANSYS Solutions | Spring 2004

 Because of the complexity and specialized Standard
nature of engineering simulation solutions, however,
For the Standard benchmark tests, benchmark data
basic processor specifications may not always pro-
was obtained from the HP rx5670* Intel Itanium 2
vide a good indication of how long these problems
processor @ 1.5 GHz/6mb L3 — HP-UX* 11i V2 serial
actually take to solve. To provide a way to more accu-
performance, and the Sun Blade* 2000 USIII
rately measure this speed for typical ANSYS solutions,
@1.015GHz — Solaris* 8. 21
the company developed benchmarks consisting of a
Figure 2 demonstrates serial performance of the
suite of ANSYS models and their respective files
Standard set of workloads for the ANSYS application.
designed to show the performance of the various
The performance presented is normalized to the
hardware platforms that ANSYS supports. These
Itanium 2–based platform, and larger scores are better
models are supplied to ANSYS partners so that they
(faster). ■
may be run on the latest, fastest hardware.
_______________________________________________
The test suite is designed to cover a wide range
of ANSYS usage. Since ANSYS Workbench uses the
This article is an excerpt from the Intel Itanium Platform
ANSYS solver for its solution, benchmark results are Performance white paper. For more information on Intel and
valid for both the ANSYS and DesignSpace families of its products, visit the company’s Web site at www.intel.com.
products.
There are two sets of workloads: Standard and *Other names and brands may be claimed as the
Platinum. The Standard workloads are smaller prob- property of others.
lems (up to 500,000 Degrees of Freedom or DOF) and
the Platinum workloads are much larger in scope (1 to
10 million DOF) and generally need the richest
possible system configurations to run. DOF is an
ANSYS 7.1 Platinum Performance ■ 1P
indicator of problem size and complexity. HP rx5670* 1.5GHz Itanium 2 processor ■ 2P
The particular set of recent benchmark results ■ 4P
shown in the accompanying charts evaluates the per- 2.5
(Higher is better)

formance of the ANSYS application with the Standard 2

Speedup

and Platinum sets of workloads on multiple platforms. 1.5

These results show that the Intel Itanium 2 processor- 1
0.5
based platform provides superior performance and
0
scaling for ANSYS. Details on the hardware config-
bm-slsb-a0x
bm-slsb-p0x
bm-slsb-s0x
bm-slsb-a1x
bm-slsb-p1x
bm-slsb-s1x
bm-slsb-a2x
bm-slsb-p2x
bm-slsb-s2x
bm-slsb-a3x
bm-slsb-p3x
bm-slsb-s3x
bm-slsb-a7x
bm-slsb-p7x
bm-slsb-s7x
urations and performance data are available at
www1.ansys.com/cgi-bin/HardwareSupport/
ANSYSHW.exe.
Performance tests and ratings are measured
using specific computer systems and/or components Figure 1. Performance of the ANSYS Platinum workloads.
and reflect the approximate performance of Intel prod- Longer bars indicate better performance.
ucts as measured by those tests. Any difference in
system hardware or software design or configuration
may affect actual performance. Buyers should consult
other sources of information to evaluate the perform-
ANSYS 7.1 Standard Serial Performance
ance of systems or components they are considering
Speedup Relative to HP rx5670*

purchasing. For more information on performance 1.2

tests and on the performance of Intel products, refer- 1
(Higher is better)

ence www.intel.com/procs/perf/limits.htm or call 0.8

(U.S.) 1-800-628-8686 or 1.916.356.3104. 0.6
0.4
0.2
Platinum 0
For Platinum benchmark tests, Intel Itanium 2 proces-
bmslsbal

bmslsbax

bmslsbfm

bmslsbpl

bmslsbpm

bmslsbpx

bmslsbsm

bmslsbsx

bmsmbblm

bmttnbfm

bmttnbpm

bmttnbsm

sor data was obtained by Intel with HP rx5670* - 4 x

1.5GHZ processors, 6M L3 cache, Red Hat AS 2.1.
Figure 1 demonstrates the scalability of
the Itanium 2–based platform for the ANSYS 7.1
application for the Platinum set of workloads. ■ HP rx5670 @ 1.5 GHz – HP-UX* 11i v2
These are described in further detail at ■ Sun Blade* 2000 USIII @ 1.015 GHz – Solaris* 8
http://www.ansys.com/services/hardware_sup
port/platinum_descriptions.htm.
Speedup is measured as the ratio of elapsed Figure 2. Performance of the ANSYS Standard workloads. Results are
time on one CPU and elapsed time on n=1, 2, 4 standardized to the Itanium 2–based platform. Longer bars indicate
number of CPUs. better performance.

www.ansys.com ANSYS Solutions | Spring 2004

 Tuning UNIX Systems
for High Performance
22
Make sure your operating system runs
CAE applications efficiently to get large
jobs done quickly
By Lynn Lewis
CAE Segment Manager
Hewlett-Packard Company

An operating system manages system resources. For each step

away from the processor, a system spends orders of magnitude of
processor cycles to get data. It takes 10s of cycles to fetch data from
cache, 1000s of cycles to reload cache from RAM, and 100,000s of
cycles to read and write to disk. A good operating system manages
these cycles to maximize system efficiency.

Remember, getting data to a processor is the point of Disk I/O

a system. The disk, memory and operating system all
work together to support that “beast.” If an untuned
system is used for ANSYS jobs, the user is wasting
processor speeds and feeds.
Think about Unix as if it had been written by
hippies: young socialistic idealists working in a com-
mune. This is not true, but it gives a humorous way to
look at the operating system that covers salient design
points: equality of access and a small system heritage.
The purpose of this article is to change the perception
of your Unix systems attitude from an easygoing
hippie focused on equality to a hard-charging yuppie
focused on getting large jobs done quickly.
The meat of the article is based on HP-UX 11i
from Hewlett-Packard. It applies at the highest level to
any operating system, particularly Unix-based sys-
tems such as HP-UX, IRIX, Tru64, and Solaris with
only the kernel variable names changed; but also to
AIX and Linux.
Beginning away from the processor and working
toward it, disk I/O and file system configuration are
important to consider when configuring systems to
run CAE codes. This is because high-speed disks and
tuned file systems can improve job run time dramati-
cally, especially with structural analysis jobs. CFD and
crash simulations are not nearly as I/O intensive,
but depending on the particular dataset also may ben-
efit from striped and tuned file systems. Tune for the
worst case.
If ANSYS is the only application to be run on the
system and there are multiple users running at once,
HP suggests setting buffer cache to a low setting.
ANSYS has its own I/O buffering system which, when
used with buffer cache set low, can help performance
substantially in a throughput environment.
If there is a heterogeneous group of applications
to be run, then a lot will depend on what the other
applications are. But, most likely, a buffer cache per-
centage set as low as possible, along with the ANSYS
I/O buffering scheme, is the way to go.
ANSYS CONFIG70.ANS File and
Parameters
In the ANSYS online manuals, users will find a
description of the config70.ans file and the parame-
ters for it. The two that affect I/O performance are
“size_bio” and “num_bufr.” Depending on the I/O
requirements of the job, the user may want to adjust
these values. The size_bio parameter sets the size of
the I/O buffer, and num_bufr is the number of buffers.
Note that when these are set above the default values,
the memory may need to be increased for the job.
Experience with running ANSYS benchmarks
shows that these two parameters have the second
greatest impact on job performance, with the greatest
factor being system configuration settings. Note espe-
cially that this is critical for multiprocessor machines
where ANSYS may be competing for I/O resources
with another application.

www.ansys.com ANSYS Solutions | Spring 2004

Striping Given enough disks and the right tunings, VxFS
on a SuperDome running 11.x has been measured at
When striping across multiple disks, the striped volume
providing over 2GB/sec read and write on a single file
size cannot exceed the capacity of the smallest disk
system. It is capable of getting the most out of hard-
multiplied by the number of disks used in the striping.
ware. Install these HP-UX patches:
Best performance results from a striped logical
volume that spans similar disks. The more closely one • HP-UX 11i (11.11) PHKL_23240 or later
matches the striped disks in terms of speed, capacity, and PHKL_25022 23
and interface type, the better the performance. So, for • HP-UX 11.0 PHKL_23773 or later
example, when striping across several disks of varying
speeds, performance will be no faster than that of the Building a VxFS file system is simpler than HFS.
slowest disk. There are fewer options. First, build the file system itself
If one has more than one interface card or bus to and then mount it.
which disks can be connected, distribute the disks as
evenly as possible among them. That is, each inter-
face card or bus should have roughly the same num- Knowing Your Operating System
ber of disks attached to it. The best I/O performance
will be achieved when using more than one bus and Introduced with HP-UX 11i, an Operating System Environment
interleaving the stripes of the logical volume. For consists of the basic HP-UX 11 operating system and software
example, when having two buses with two disks on to provide a complete user environment for commercial and
each bus, the disks should be ordered so that disk 1 technical systems. For technical computing systems such as
is on bus 1, disk 2 is on bus 2, disk 3 is on bus 1 and compute or file servers, there are two operating environments:
Minimal Technical Computing Operating Environment (MTCOE)
disk 4 is on bus 2. This can be accomplished by
and Technical Computing Operating Environment (TCOE, also a
adding the disks to the volume group in the order
superset of MTCOE). For a complete content list of the various
desired and stripe as in the above example. As an
operating environment offerings, see the Hewlett-Packard doc-
alternate method, users can create the logical volume umentation Web site at http://docs.hp.com/.
without giving it a size, by not using the -L option, and
then using lvextend to give the size of the logical vol- Due to the various naming references to the operating system
ume and to specify which disks to use. Please see the it is sometimes confusing as to which operating system is actu-
Unix manual pages for these commands for more ally running on a system. Using “uname -a” will return all the
details. information in a single string. Please see the manual pages for
other options and further details.
File Systems
Systems that are used to run CAE applications should always be
Though this section refers to HFS and VxFS, there are
running a 64-bit version of the OS. If the OS is 32-bit, then the
other file systems used for high performance; most application will not have the advantage of the increased
are journal file systems because of boot speed and addressing capability provided by 64-bit systems.
reliability as well as performance.
HFS is the High-Performance File System format Occasionally, patches for HP-UX may be required for your
used with the HP-UX operating system. As of HP-UX application. Patches can be found at us-support.external.
11i, all file systems on TCOE-based machines default hp.com (for Americas and Asia) or europe-support.external.
to VxFS, except for the stand partition on machines hp.com (for Europe).
using PA-RISC processors, which is HFS.
The Technical Marketing Group at HP recom-
mends that file system space be striped across multi-
ple dedicated disks using large 64K blocks and 8K
fragments and, if possible, multiple controllers. HP
recommends one disk controller for every four or five
disks. Do not use the directory/tmp for temporary or
scratch files and do not confuse striping which is RAID
0 with RAID 5. Using RAID 5 disks for temporary stor-
age will severely degrade performance. The large
stripes that the Technical Marketing Group recom-
mends are not default and must be created at config-
uration time.

Information on VxFS 3.x

In the fall of 2000, Hewlett-Packard and Veritas
released a new version of the VxFS file system for
HP-UX. Once tuned, this file system outperforms
all other file systems on HP-UX systems. However,
if left untuned, performance will be worse than its HP supports two basic configurations for high-performance technical
customers to scale-up (SMP) or scale-out (cluster): The Integrity
predecessor.
Superdome (left) is a 64-cpu SMP. The cluster of rx5670, four CPU
nodes (right) is capable of supporting larger configurations.

www.ansys.com ANSYS Solutions | Spring 2004


Tuning and Mounting Options for
File Systems
VxFS has a large number of options to tune. HP can-
not recommend the best solution for every applica-
24 tion; however, HP has come across a very good
middle ground that provides excellent performance
for most applications. The mount options used for
scratch file systems in HP’s own technical server are
available. Remember, this is for a scratch file system
and not for a file system that needs to be stable over
time.
Having older disks or disks with different
characteristics (slower spindle speed or different
on-disk cache), one may have to do some experimen-
tation. The two main settings that will have to be
adjusted are read_nstream and write_nstream. These
should be set to the number of disks in the stripe.
Refer to the vxtunefs manual page for a detailed
explanation of each parameter.
Memory
The goal is to have enough memory for the entire job
to reside in memory and still have room for the OS and
buffer cache. To determine there’s enough memory,
run /opt/perf/bin/glance and an application at the
same time. A 60-day trial version of glance is included
on the HP-UX application CD-ROM. Select the appli-
cation process in glance, and monitor the memory
usage while working. There are four things to watch
for: large (U)ser memory, very small (B)uffer cache
memory, high (V)irtual memory disk usage and the
ratio of RSS (resident set size) to VSS (virtual set size).
The RSS is the amount of physical memory
(RAM) currently being used by a process. VSS is that
amount of virtual memory allocated to a process.
Ideally, the RSS and VSS should be the same. A large
difference combined with Virtual Memory Disk activity
means there is not enough RAM and the application
has to use swap space.
Swapping
Nothing will kill performance faster than swapping.
Benchmark experience has shown that adding
additional RAM to memory-starved applications can
improve performance by as much as 500 percent.
If swapping becomes necessary, use striped,
dedicated, raw swap partitions that are on different
physical disks than the application data. But either buy
more memory or redesign numerical experiment to
match resources.
Processors and Batch Jobs
For best results, use a basic queuing system, and do
not overload the system with large numbers of jobs all
demanding all the resources. With additional proces-
sors, consider allocating one CPU for system tasks.
Allowing more than one compute thread per CPU can
lead to cache-thrashing and excessive context
switching. HP used Platform’s LSF.
www.ansys.com
Adjusting Kernel Parameters
During the review process of the source document for this
sketch, it was reported by our reviewers that there were
problems on HP-UX using SAM to set certain kernel
parameters to the values.
If you encounter a problem with any systems GUI adminis-
tration tool to set kernel parameters, try building a kernel
using the manual process outlined in the respective system
manual.
With any system, always save a backup kernel. Then, if the
new kernel fails to boot, boot the system from the backup
kernel, and repeat the process of creating a new kernel.
Review the systems administration guide for further details
regarding the process of building a kernel and rebooting
the system.
Use the recommended CAE parameters as a starting point
for kernel parameter settings for both workstations and
servers.
A chart listing recommendations for almost forty different
kernel parameters for HP-UX can be obtained from HP by
emailing lynnlewis@hp.com with the subject “Request
CAE Kernel Parameters.”
Kernel Parameters
In the past, several HP-UX kernel parameters
needed to be modified for optimal performance.
The HP factory has recognized this and now ships
all servers and Visualize workstations pre-configured
for most technical users. In addition, a kernel
parameter template has been added to HP’s GUI sys-
tem administrative tool, SAM, to make changes easier.
The templates are:
• CAE/ME/General Eng. Workstation 64-bit
and 32-bit
• EE Engineering Workstation 64-bit and 32-bit
The exact values for these templates can be
found in the HP-UX 11.0 release notes at
http://docs.hp.com. They should be reviewed in con-
junction with the system’s configuration guide and
kernel parameter descriptions.
To conserve memory consumption and simplify
setting of the parameters, HP is defining three
sizes of systems: small, medium and large,
depending on the amount of RAM in the system.
Those parameters affecting buffer cache will be
set according to the size of the system and will
not require user intervention. ■
Have comments on this article or questions about Hewlett-
Packard products? Inquiries are welcome and should be
directed to CAE Segment Manager Lynn Lewis at
lynnlewis@hp.com.
This article is based on documentation edited by
John Cowles, Optimization Technology Manager of
Hewlett-Packard Company.
ANSYS Solutions | Spring 2004
www.ansys.com ANSYS Solutions | Spring 2004
26

3-D Crack Analysis

Build 3-D crack meshes more easily with specialized
modeler that acts as a front end to ANSYS

By Greg Thorwald,
Computational
Mechanics Engineer
and Ted Anderson,
President
and Founder
Structural Reliability
Technology
Figure 1 (above). Surface
crack in a flat plate; the
mesh is back half-symmetric.
Fracture mechanics analysis is beneficial for safety, for reliability, and
for saving time and money. Because fracture can occur at stress
levels well below the yield strength, it is important to have an analysis
method that is capable of evaluating cracks. Finite element analysis
can be used to obtain J-integral or stress intensity values, which allow
cracks to be evaluated as safe or unsafe for given conditions. For
example, the critical crack size can be determined for a particular
loading, material, and geometry. The expected lifetime of a component
subjected to cyclic loading can be evaluated by a fatigue crack growth
analysis. The number of loading cycles for a crack to grow from a small
initial size to a final critical size can be predicted and used to set an
efficient inspection schedule.

www.ansys.com ANSYS Solutions | Spring 2004

 Incorporating cracks into 3-D finite element
meshes can increase the usual modeling effort by an
order of magnitude. General-purpose 3-D modeling
and meshing software is not well suited to this task.
There are many challenges to meshing a 3-D crack,
including the crack shape not naturally fitting within
the geometry, the requirement of using brick elements
in the focused mesh pattern, having collapsed brick
elements along the crack front, listing the crack front
nodes in the input file, correctly constraining crack
plane nodes for symmetry, and cracks following
curved surfaces. A parametric case study to examine
the effect of different crack sizes, crack locations,
loads, materials, and geometry requires a new mesh
for each case; the modeling effort must be repeated
for each case. Brick elements are required to permit
the J-integral — an energy release rate value that
uniquely characterizes the crack front stresses and
strains — to be computed at the crack front nodes.
The collapsed brick elements at the crack front give a
strain singularity and help improve numerical accuracy.
In a finite element crack analysis, the J-integral can be
computed as a volume integral in the concentric rings
of elements around the crack front. The J-integral from
each ring of elements can be compared to check the
integral’s path-dependence, a check for numerical
accuracy. For elastic materials, the J-integral results
can be used to obtain the stress intensity value, KI,
used in fracture and fatigue calculations.
FEA-Crack Software
To reduce the time and effort of building 3-D crack
meshes, Structural Reliability Technology (SRT) has
created FEA-Crack software, which can quickly gen-
erate crack meshes for a variety of crack shapes and
geometries.
FEA-Crack acts as a front end to ANSYS for the
purpose of 3-D crack analysis; the crack mesh gener-
ator creates the complete and ready-to-run ANSYS
input file. Using the built-in library of geometric shapes
requires only a few input values to generate the crack
mesh. A user-defined geometry option allows a crack
to be added to virtually any solid model. If desired,
FEA-Crack can be used as a driver program to run
multiple ANSYS input files in batch mode. The post-
processing module in FEA-Crack uses stress and
strain values from the ANSYS results file to compute
the J-integral along the crack front. These results are
displayed in a series of x-y plots. The post-processing
module also includes complete 3-D graphics capabil-
ities for viewing the stress results and deformed
shape.
Figure 2. Stress results and deformed shape for a surface
crack in a nozzle; the mesh is half-symmetric.
www.ansys.com
FEA-Crack can be used to analyze stationary
(non-growing) cracks in order to evaluate the driving
force for brittle fracture. FEA-Crack can also perform 27
fatigue crack propagation analyses by means of a
re-meshing crack growth feature. A ductile crack
growth capability will be available in a future release.
Sample Meshes
Figure 1 shows a model of a semi-elliptical surface
crack in a flat plate. This model was created in FEA-
Crack in approximately 2 minutes. The crack face ele-
ments are colored light blue to highlight the crack
location. The mesh is half-symmetric, with a symme-
try plane passing through the crack plane. Taking
advantage of mesh symmetry when possible reduces
the mesh size and speeds up the analysis time. The
concentric rings of elements around the crack front
are visible at each end of the crack on the top mesh
surface. Higher local mesh refinement along the crack
front is desirable to better capture the high stress and
strain gradients near the crack and to improve the J-
integral results. FEA-Crack gives the user complete
control over the level of mesh refinement, both at the
crack and remote from the crack.
Figure 2 shows a nozzle geometry with a surface
crack at the toe of the fillet weld connecting the noz-
zle neck to the spherical head. The color map shows
the stress results, and the deformed shape shows the
crack opening due to internal pressure loading. The
local region of high stress around the crack front is
characteristic of the stress pattern at the tip of a crack
(shown by the reds and yellows in the color map). The
stress intensity at the crack front, due to the higher
local stresses, is the driving force causing fracture,
even at low load levels. This nozzle mesh is half-sym-
metric and includes half of the crack length; the sym-
metry plane bisects the surface crack. Since the crack
face follows the curved connection intersection, both
the front and back crack faces must be included in
this model.
ANSYS Solutions | Spring 2004
 Figure 3. Stress results for a corner crack in a plate with a
hole; the mesh is back half-symmetric.

28

crack front. Figure 4 shows the J-integral results plot

for the six cracks; as the crack size increases, the
J-integral value increases the most at the bottom of
the crack (crack angle of π/2 radians). The crack
analysis would continue by using the J-integral values
to get stress intensity values for use in a fracture
assessment or a fatigue analysis.

Parametric 3-D Crack Analysis Conclusion

Bolted and riveted connections are important in civil Being able to quickly generate 3-D crack meshes
and aerospace structures. Cracks can form on the makes crack analysis easier and can lead to improved
edge of the rivet or bolt holes perpendicular to the safety and reliability of structures that have or may
primary tensile stress direction, leading to fracture develop cracks. The examples shown here provide a
problems in the connection. The corner crack shape look at some of the typical crack mesh features and
opens on the top and left edges, where the previous characteristic results. Using FEA-Crack to quickly
surface crack opens only along the top edge of the generate crack meshes and post-process results
crack. A parametric analysis can examine the effect of adds a general 3-D crack analysis capability to
the crack size on structural integrity. ANSYS. ■
In this example, three crack lengths at two crack
depths give a total of six cracks for analysis. Ratios of
the crack length to crack depth, and the crack depth
to the thickness, are often used to set the parametric Greg Thorwald, Ph.D., is a computational mechanics engineer
at Structural Reliability Technology (SRT) and is the lead devel-
crack sizes. The crack length to depth ratios are 1.0,
oper of the company’s FEA-Crack software. Ted Anderson,
2.0 and 4.0 (increasing crack length), and the crack
Ph.D., is the president and founder of SRT, and is the author of
depth to thickness ratios are 0.2 and 0.6 (increasing the textbook Fracture Mechanics: Fundamentals and
crack depth) for this example. The parametric analysis Applications published by CRC Press. For more information on
could be extended to include many more crack size SRT’s software, go to www.srt-boulder.com.
ratios, different hole sizes, other material values, etc.,
for dozens or hundreds of crack analysis cases.
The first step in the crack analysis is to enter the
basic model dimensions, loading, and material prop-
erty values. The crack meshes for each crack size are
then built using FEA-Crack, and the finite element Flat plate with a hole, six corner cracks at the hole,
analyses are run with ANSYS. The post-processing J-integral results along each crack front.
module computes the J-integrals for each crack size.
When post-processing the results, it is useful to view
the stress color map and the deformed shape to
check that the boundary conditions were correctly
applied and that the mesh is deforming in the
expected direction.
Figure 3 shows the stress color map and
deformed shape of the largest corner crack (crack
number six); the crack is being pulled open by the
applied tensile stress. There are again local high
stresses (red to green on the color map) along the

Figure 4. J-integral plots for the six crack sizes in the plate
with a hole parametric analysis example.

www.ansys.com ANSYS Solutions | Spring 2004

 29

Software coordinates computing workload using a

set of resources distributed across a network
By Tony Iams
Vice President and Research Director
D.H. Brown Associates, Inc.

Grid computing stands as a metaphor for treating computing resources in the same manner that a utility
grid treats resources such as electricity. Just as consumers do not care whether the electrical power they use
has been generated from a nuclear power plant, a coal-fired generator, or a hydroelectric dam, users of Grid
computing infrastructures gain access to computing resources without worrying about the configuration
or location of the server hardware used to run their applications. When fully implemented, Grid computing
allows IT organizations to maximize their total computing capacity, while enabling rapid completion of
compute-intensive applications when necessary. The core of Grid computing infrastructures is usually a set
of Distributed Resource Management (DRM) software (which may also be referred to as Scheduler, Queuing,
or Job Management software), to which users submit workloads. The Grid software schedules the execution
of each workload using a set of resources that are distributed across a network, and returns the results.

Grid computing has emerged as a promising tech-
nique for enabling the virtualization of computing
resources that decouple workloads from details about
the underlying hardware. It offers the potential for com-
puting resources to be provisioned with far more flexi-
bility than traditional deployment methods. Because
computing resources that are virtualized on a Grid can
be allocated precisely to meet the needs of particular
workloads, administrators gain better utilization of
these resources. Grid computing also carries the
potential to improve collaboration by providing an
infrastructure that ties together all of the tools and
processes needed for geographically distributed users
to share computing resources and data.
What are the benefits of Grid computing, and what
can the IT community expect? Expectation manage-
ment represents one of the most critical issues facing
planners of Grid infrastructures today. Grid computing
offers a tremendous appeal, but note that the amount
of work that still lies ahead in implementing the full
range of Grid capabilities can be easily underestimated.
At this stage, savvy IT managers will realize that Grid
computing technology may be oversold, risking a
backlash if the expectations of users (and managers)
are not met. The resulting disappointment may slow
down future deployment of Grid computing in areas
where it makes sense, given its capabilities at the time.
It is thus important for everyone involved in the adop-
tion of Grid computing to be fully aware of what can
and cannot be done today.
Using Scattered Computing Resources
Until now, many users could not take advantage of the
Grid’s potential. Most existing Grids are deployed in
research and development environments, where the
user base tends to have some expertise in the use of
advanced computing systems. While the computing
techniques underlying Grid computing are rooted in
traditional clustering and massively parallel processing
(MPP) approaches that have been available for many
years, these systems traditionally required consider-
able expertise, including an in-depth understanding of
UNIX command-line interfaces (CLIs) for submitting
jobs and monitoring job progress.
Users of existing Grids typically interact with them
using CLIs and hand-crafted batch scripts. However,
as Grid computing broadens its base of applications,
less experienced users will increasingly need to take
advantage of Grid services. These users may bring
deep expertise in a particular application area, but
they will not necessarily command the computer skills

www.ansys.com ANSYS Solutions | Spring 2004

30
necessary to access today’s Grid. New types of inter-
faces will be needed that allow less-experienced users
to take advantage of Grid services in their day-to-day
activities.
Some organizations are implementing web-based
Grid Portals to make harnessing of the Grid’s power
easier for users. These Portals allow users to take
advantage of Grid computing using simple GUI-based
(Graphical User Interface) techniques such as pull-
down menus, while making the Grid accessible over
the Web from any location within the organization.
In Web parlance, the most basic definition of a
“portal” is a GUI that is implemented entirely within the
scope of a Web browser and aggregates information
from multiple sources. Advanced portals may include
a Web server, integrated mechanisms for security and
identity management, authentication via a variety of
methodologies, access to both business and technical
applications, and mechanisms for content aggregation
and personalization. Portals typically represent the var-
ious information sources as “topics” or “channels,”
which may be separated through mechanisms such as
tabbed pages or multiple panels on a single page.
Because the Portal interface is fully useable through a
Web browser, users are able to access the Portal’s
functions from any location worldwide via the Internet.
Grid Portals apply this technique for simplifying
the use and management of Grid computing infra-
structures. Grid Portals offer the following benefits:
• Simplification: The Grid Portal GUI makes it
possible for users without extensive computer
expertise to submit jobs, monitor job progress,
and capture processing results in a Grid envi-
ronment. This potentially expands the base of
users who can take advantage of Grid comput-
ing resources.
• Global Access: Users can securely access
Grid computing resources from any location
worldwide. This provides a convenience for
regular Grid users, and enables teams of users
who are geographically distributed to share a
set of Grid resources.
• Front-End Integration: If an organization has
made a commitment to implement a global
Portal architecture for integrating various infor-
mation resources, Grid computing resources
can be added to an organization’s information
systems’ front end relatively seamlessly.
• Manageability: If the Grid Portal GUI includes
management functions, it enables less sea-
soned administrators to employ a familiar inter-
face for configuring the Grid infrastructure.
Administrators also gain the ability to manage
the Grid infrastructure from any physical
location.
www.ansys.com
Commercial Offerings
A variety of methods exist for implementing Grid
Portals. In the traditional technical domain, pioneering
Grid users often employ homegrown systems based
on public domain toolkits. D.H. Brown Associates,
Inc.’s (DHBA) forthcoming study on Grid Portals
examines three commercially supported solutions
highlighted in this tech trend:
• Altair Engineering’s e-compute interface for its
Portable Batch System Pro scheduler
• Platform Computing’s Web interface for
Platform LSF
• Sun’s Grid Engine Portal
Each of these solutions provides a framework that
administrators can use to build custom interfaces for
their Grid infrastructure. All three systems support a
complete set of functional capabilities needed to
develop the Grid Portal interface, as a secure interme-
diate layer that resides between users and the
Distributed Resource Manager used to schedule Grid
resources. Because Grid applications are very diverse,
with a variety of requirements for inputting and out-
putting data, and often requiring a variety of parame-
ters from users to specify how processing should take
place, most Grid Portal solutions require customization
for the environments in which they are deployed. They
effectively serve as toolkits that administrators use to
build custom interfaces for their Grid infrastructure.
Altair Engineering’s e-compute is a Web-based
GUI designed to simplify the submission and moni-
toring of computationally intensive jobs to Altair
Engineering’s PBS Pro DRM. Although e-compute is
optimized to work with PBS Pro, it is also designed to
work with other DRM systems such as Platform
Computing’s LSF and Cray’s Network Queuing
Environment (NQE), and it can be customized to work
with other load management products as well. E-com-
pute supports XML, SOAP, J2EE, HTML, and CSS1
standards. Altair has invested particularly in the devel-
opment of functional areas such as security to make
sure that it could guarantee secure access to comput-
ing resources.
Platform Computing is one of the most estab-
lished vendors in the Grid computing market. Platform
provides a Web-based GUI that is designed to enable
explicit interaction with its LSF software. The Web GUI
has been included with Platform LSF since Release 5,
which began shipping in 2002, and there is no addi-
tional charge for its use. Platform LSF includes a rea-
sonably elegant Web-based interface that works out of
the box, but Platform is not trying to provide a full-
function Grid Portal solution. Rather, the Web GUI is
fully integrated with all of the functions of Platform LSF
for invocation by Web-based applications. To work
ANSYS Solutions | Spring 2004
 31
31
with other DRMs, Platform has developed an open
source “metascheduler” framework called the
Community Scheduler Framework (CSF), to contribute
to future versions of the Globus Toolkit. CSF is an
extensible framework for implementing metasched-
ulers, which can negotiate with heterogeneous work-
load execution software such as Platform LSF, as well
as other resource managers such as Sun Grid Engine
and PBS. However, Platform LSF offers strong support
for web services standards and protocols such as XML
and SOAP. Platform LSF potentially simplifies integra-
tion with other enterprise IT systems using Web serv-
ices, and enables interoperability with emerging indus-
try standards for Grid computing. These include the
newest release of the Globus Toolkit, which is based munity such as standards bodies, and the gradual
on XML/SOAP and the latest version of Open Grid enabling of its software products for Grid computing.
Services Architecture (OGSA), and the Open Grid SGI has narrowed its focus exclusively on the
Services Infrastructure (OGSI). requirements of technical computing users, and adds
Sun’s GEP is the only studied product that truly value to Grid computing in three ways: high-end SMP
integrates the control of DRM functions into a Portal servers, visualization, and distributed storage manage-
environment. That is, the Grid simply appears as one ment. Sun is attempting to redefine the economics of
of many channels within the Portal, residing alongside Grid computing by driving large volumes of its
other channels representing different sources of infor- GridWare software into the market.
mation and computing tasks. Sun is also pricing its Sun is also enhancing the usability of Grid com-
Grid solution aggressively, as part of its effort to rede- puting infrastructures with its Grid Engine Portal (GEP),
fine the economics of Grid computing by driving large a secure, web-based GUI that combines a traditional
volumes of its Grid software into the market. However, Portal interface with a set of point-and-click tools for
the Sun ONE Portal Server currently runs only on the controlling GridWare.
Solaris operating system. DHBA’s forthcoming study on the system vendors
reviews their efforts in staking out positions for Grid
System Vendors computing. ■
Grid computing deployments are inherently compre-
hensive undertakings that touch many levels of IT
infrastructure, including hardware, networks, systems
software, middleware, applications, development D.H. Brown Associates, Inc. collaborates with software ven-
tools, and services. Thus, the leading system vendors dors and leading end users to produce in-depth research
reports on trends in engineering, manufacturing, and technical
will play a major role in enabling the success of the
computing. The company also hosts ongoing teleconferences
Grid computing model. Indeed, the vendors have
on Grid computing and prepares reports on latest trends in the
responded to growing user interest in Grid computing
technology. For more information, call 914.937.4302 or visit
with a variety of development and marketing initiatives
www.dhbrown.com.
in an effort to establish leadership for providing Grid
computing solutions.
HP is taking several business initiatives to make
sure it remains at the forefront of the Grid computing
movement. Key Grid middleware is supported on HP
systems. The company provides consulting services
for designing and implementing Grids with its consult-
ing and integration organization.
IBM is attempting to capture the lead in driving
Grid computing out of the traditional High
Performance Technical Computing (HPTC) domain
into more mainstream applications by seeking new
ways to make it more generally applicable to a variety
of workloads. IBM’s efforts include the aggressive
pursuit of a leadership role in areas of the Grid com-

www.ansys.com ANSYS Solutions | Spring 2004

 Software Highlights

New Solvers in
32 ANSYS 8.0
Push the Envelope
Latest release includes new Multi-Field and parallel performance solvers.

The continuing challenge in engineering analysis is to provide

realistic simulations quickly for an expanding range of applications
where models are increasingly complex, behavior is often
nonlinear, and the physics of problems sometimes involve fields.

With the launch of version 8.0, ANSYS strength-
ens its leadership position in this market by announc-
ing new solvers that met this challenge, enabling users
to solve problems faster than ever and to tackle new
classes of applications where previously there have
been no solutions available: a new Multi-Field solver
for handling coupled physics problems, and two new
parallel performance solvers for simulation solutions
using distributed computing resources.
Handling Multiple Fields
In recent years, there has been an increase in the
number of industries and applications needing to
solve coupled-physics problems that often involve
structural, thermal, electomagnetics, fluid flow, and
acoustics.
Often a user needs to include more than two
physics to achieve a realistic simulation, and there is a
need to facilitate coupling to external analysis codes.
While ANSYS pioneered multiphysics technology in
the 1980s, it continues to expand and enhance this
powerful capability with its new Multi-Field solver.
These capabilities reach a broader base of users
and applications in a wide range of industries includ-
ing automotive, aerospace, telecommunications, elec-
tronics, consumer products, and industrial equipment.
ANSYS 8.0’s Multi-Field solver provides an easy-
to-use framework to solve coupled-field problems in
many new markets and applications where previously
there have been no solutions. With this new version,
ANSYS offers a general-purpose, automated sequen-
tial coupled-physics solver applicable across all
physics available in ANSYS Multiphysics. With the
Multi-Field solver, each physics can have totally inde-
pendent meshes and solution settings. This allows the
multiphysics problem to be divided up easily amongst
single physics experts (both within a company or to
external consultants).
Additionally, the recent acquisition of the CFX
division of AEA Technology has resulted in a joint
product collaboration to develop a link between the
CFX-5 product and ANSYS Multiphysics. This
advancement enables seamless transfer of mechani-
cal and thermal loads from a CFD analysis in CFX-5 to
ANSYS for stress analysis.
With ANSYS 8.0, companies are no longer
required to depend solely on “super-users” to han-
dle all advanced physics capabilities of ANSYS
Multiphysics. Sharing the multiphysics load across
several physics experts, with the Multi-Field solver,
results in higher quality and more accurate work com-
bined with the benefits of vastly compressed analysis
process timescales. One key feature that enables
independent users to build and set up their physics
analysis is the dissimilar mesh interface allowed
between each of the physics fields.
Parallel Performance
ANSYS 8.0 now extends distributing processing into
the next generation, building on its world-class serial
solvers. The ANSYS 8.0 add-on module (Parallel
Performance for ANSYS) now includes four solvers —
Algebraic Multi-Grid solver (AMG), a Distributed
Domain Solver (DDS), and two new solvers, the
Distributed Pre-conditioned Conjugate Gradient
solver (DPCG) and the Distributed Jacobi Conjugate
Gradient solver (DJCG).
The new DPCG and DJCG solvers directly
target distributed parallel processing and lead the
market in speed, memory efficiency, and analysis
types. These new solvers will enable ANSYS
customers to complete more and larger analyses
in less time than was previously possible.

www.ansys.com ANSYS Solutions | Spring 2004


The DPCG solver preserves all of the merits of
the classic PCG solver and can be run on either
shared memory or distributed memory machines with
superior scalability to the PCG solver. Compared to
the DDS solver, the DPCG solver is more robust and
uses less memory with similar scalability at low (less
than 16) numbers of processors. The total sum of
memory used by the DPCG (summed total over the
network or all processors) is about 30% more than the
classic PCG solver. In addition to the limitations of the
classic PCG solver, the DPCG solver does not support
subspace eigensolver (Powerdynamics), MSAVE and
PRECISION commands, or p-elements.
The DJCG solver is based on an enhanced theory
for the classic JCG solver. Scalability of this solver is
superior to the JCG solver with little additional memory
required. The DJCG solver is available only for STATIC
and TRANS (full method) analyses where the stiffness
is symmetric. This solver does not support the fast
thermal option (THOPT).
The new solvers complement the other two exist-
ing ANSYS parallel performance solvers: AMG and
DDS. AMG is based on a multi-level method and is
intended for use on shared-memory, multiprocessing
servers. It scales well for up to eight processors, yet
maintains excellent performance levels with as few as
two processors. It also works well with models typical-
ly difficult for other types of iterative solvers including ill-
conditioned models resulting from large-element
aspect ratios or models with shell and/or beam ele-
ments attached to solid elements. DDS is a scalable
solver designed for use in a distributed-memory envi-
ronment or a combination of both distributed- and
shared-memory environments. It is intended for large
static or full transient analyses with symmetric matri-
ces that do not involve pre-stress, inertia relief, cou-
pling, or constraint equations.
Learn More about ANSYS 8.0
In addition to new Multi-Field and parallel performance
solvers, ANSYS 8.0 features improved capabilities for contact
formulation, automated simulation, integration, parametric
mesh capabilities, and other technological advancements. For
a detailed overview, visit www.ansys.com.
www.ansys.com
33
Key Features of the Multi-Field Solver
• Each physics field is treated as a field with independent
model and mesh
• Any number of physics fields can be sequentially
coupled
• Each field is defined by a group of element types
• Surface and volumetric loads transfer across
physics fields
• Sequential (load vector) coupling is between fields,
and each field may have:
– Different analysis types (transient, static, or harmonic)
– Different solvers and analysis options
– Different mesh descretization
• Unidirectional coupling is between CFX by ANSYS and
ANSYS Multiphysics
• Each field can be imported from an external solver
(e.g., CFX-5)
• Automated morphing of nonstructural elements
• Material and geometric nonlinearity
• Independent results files for each physics field
The licensing of the Parallel Performance add-on
module for ANSYS is straightforward. To perform any
ANSYS analysis requires that one license of the
appropriate level of functionality of the software be
checked out from the license server. To run a single
analysis in parallel only requires the addition of one
Parallel Performance for ANSYS license. So in total
there would be one ANSYS license and one Parallel
Performance for ANSYS license.
Note that the licensing is not CPU-based. In
other words, the number of CPUs involved is not con-
sidered in the cost of the license. Users need only
have one Parallel Performance license for each analy-
sis they want to run in parallel, in addition to the
license for core ANSYS, of course. ■
ANSYS Solutions | Spring 2004
 Guest Commentary
Overcoming Barriers
to Effective
34
Quality
Assurance
in Finite Element Analysis
Part 1 of 2
By Vince Adams
Director of Analysis Services
IMPACT Engineering
Solutions, Inc.
A recent article published by
D.H. Brown Associates notes
that CAE analysts and designers
today often manage data in very
unstructured ways, making it dif-
ficult to reuse data, locate information, and create
archives that can facilitate product development deci-
sions for future generations of products. The article
further states that CAE early in the design cycle
requires quality assurance mechanisms to support the
management of performance information for making
decisions.
While only a few years ago the subject of QA in
FEA was limited to analysis-centric organizations such
as NAFEMS or internal groups at large organizations
such as John Deere or Ford, the proliferation of the
technology, its potential impact on the bottom line, and
the growing awareness that bad FEA is worse than no
FEA at all has forced the topic into the mainstream of
product development discussions. However, based on
our work with companies ranging in size from global
engineering organizations to local teams of five or six
technical people, too few companies have taken seri-
ous steps toward controlling the quality of the finite
element usage.
The reasons for this resistance or failure to adapt
QA programs are as varied as the products these com-
panies develop. However, two reasons tend to surface
most often when the topic surfaces. The first reason,
which we will explore in the following commentary, is
that it’s a great idea — for other companies — but we
are doing just fine. The second, to be covered in the
next installment, is that implementing a QA program is
too complex for a technology that has no formal place
or plan in the product development process to begin
with. Examining these two barriers to implementation
will shed some light on the structuring of a QA plan
that can overcome these barriers.
www.ansys.com
Fear of Self-Assessment
A few years ago, I had the opportunity to observe an
engineering organization — an ANSYS user —that had
one of the most inefficient analysis implementations I
have seen to date. Upon completing my audit and
submitting my report, this company resisted making
any changes and stated that they felt they were on top
of things.
A post mortem of this study suggested a handful
of reasons for this resistance. These included cost,
inertia, arrogance, and ignorance — all common
human traits that can be used to explain a variety of
behaviors. However, further discussion brought to light
a final and more likely cause (in our opinion) — fear.
The threat of being “exposed,” the threat of learning
that mistakes were made and the natural extension of
these mistakes — fear of losing employment — may
have played the most significant part. We don’t for a
minute believe that the managers of the groups we
met with huddled in a closed conference room with the
lights off and discussed their fears. It also is unlikely
that many of them even were conscious of their con-
cern. We do believe — and have observed this at other
companies — that many people think that an honest
assessment of a comfortable situation poses a threat
to the status quo that to this point hasn’t been identi-
fied internally as a problem.
The CEO of IMPACT Engineering Solutions uses
the analogy of a smoker to emphasize this point. There
is sufficient information in the public domain — and
widespread access to this information — that smoking
will adversely affect one’s health. However, it is clear
that most smokers forge right ahead despite the dan-
gers. We can assume an attitude of “it’s not hurting me
today” or “I’ll quit when it starts affecting my health.”
It’s likely that these same smokers will consciously or
subconsciously avoid a visit to a doctor because they
don’t want to hear what they don’t intend to act upon.
Many engineering organizations share this characteris-
tic. One other fear simply may be that engineers or
managers have too much on their plates already, so
why open another can of worms? Their current FEA
ANSYS Solutions | Spring 2004

implementation appears to work, and there are many
other problems that have a higher profile and greater
visibility requiring attention. All this suggests that a QA
program for FEA needs to be structured so that it is
entirely non-threatening. It must be viewed as a posi-
tive growth experience for all parties involved. It must
be presented in a way that builds on existing practices,
instead of rejects them. Finally, it must be simple and
unintrusive enough to not threaten project schedules
and workloads.
Everything Here Is Working Fine
The other likely scenario for resistance is over-confi-
dence in skills and methods. We recently met with a
group of analysts at a major corporation with the
explicit intent of presenting skills assessment and a QA
program for several of their divisions. Interestingly, the
feedback we received from the coordinator of the pro-
gram was that the users thought that they were
advanced enough in their utilization of FEA that the
investment would be wasted.
One attendee went so far as to state that their
lack of a test case as an example of FEA being per-
formed incorrectly was proof enough that their
processes were just fine. It is possible that this user
wasn’t paying attention when several group leaders
stated plainly that there was no standardized reporting
or test cross-referencing and that each group devel-
oped their own techniques and procedures. While
“validation by correlation” is a noble and potentially
viable QA procedure, the fact of the matter is that too
few companies actually correlate their FE models. It’s
too easy to confuse product testing after FEA with cor-
relation to FEA.
After the session, I met with one of their analysts
who shared with me a contact model in which results
matched strain gage data perfectly. I pointed out that
there were huge discrepancies in the geometry, the
bolt preload method, the bolt preload calculations, the
placement of the strain gage, and resolution of the
model that all pointed to an incorrect model. He
acknowledged the modeling errors but fell back on the
fact that the results matched the test. For him,
confirmation. For me, a lucky combination.
Keeping this in mind, a QA program must include
components to rectify this gap in the analysis process
with testing designed to validate modeling methods as
well as product performance. Additionally, the QA pro-
cedure must include an assessment of the skills and
methods of the users and the level of responsibility
that each user should be allowed — again, in a
non-threatening manner. This assessment should not
only identify areas of weakness but also provide feed-
back on improvement options to bolster skill levels.
www.ansys.com
35
Finally, the QA program should provide cross-depart-
mental best practices and guidelines, such as pre- and
post-analysis checklists and report formats, so that
users can learn from the work of others and a compa-
ny’s intellectual capital is used most effectively.
Lofty Goals, But No Plans
When a company only uses FEA to double-check a
design that was developed using traditional methods,
the risk associated with a lack of FEA QA is minimal.
However, many companies are espousing goals to
drive design decisions directly from FEA data without
having a clear idea of how the technology will be inte-
grated into the design process or what type of QA is
required. Achieving the goal of reduced testing or vali-
dation through prototypes must be preceded by a plan
to target specific problems and challenges through
analysis. That plan must include a QA program to
smooth out the inconsistencies that will arise from
sporadic, intermittent, and/or uncontrolled use.
The successful implementation of a QA program
should be accompanied by the education of the man-
agement team as to the capabilities and limitations of
the technology so that expectations can be set prop-
erly. Similarly, a comprehensive study of current tools
and processes should precede and drive the develop-
ment of such a plan so that goals and targets can be
set with proper, company-specific metrics put in place.
Finally, a company must perceive that the develop-
ment of a QA plan for FEA (a technology whose role in
the company is not always clear to management or to
the engineering staff) will not interfere with higher visi-
bility directives and programs. The availability of an
existing yet customizable QA program from an outside
source can alleviate the pressure a manager might
otherwise feel in allocating precious internal resources
to this task. ■
Part 2, appearing in the next issue of ANSYS Solutions, will dis-
cuss step-by-step ways to best implement a quality assurance
program.
Vince Adams is co-author of the book Building Better Products
with Finite Element Analysis and the inaugural chairman for the
NAFEMS North American Steering Committee. He currently
serves as Director of Analysis Services at IMPACT Engineering
Solutions, Inc. (www.impactengsol.com), a consulting firm pro-
viding design and analysis services and support to industrial
clients in a wide range of industries around the world. Vince can
be reached at vadams@impactengsol.com.
ANSYS Solutions | Spring 2004
 Tech File

36

Interpreting Your
Analysis Results
Spend time reviewing the answers to
understand what they really mean
By John Crawford
Consulting Analyst

You’ve spent three days building a model, a day applying boundary

conditions, and, after an overnight solution, you’re ready to post-
process the results and move on to your next analysis. What’s wrong
with this picture? The problem is that we often don’t spend enough
time reviewing the answers and gaining insight into the problem to
offer an enlightened interpretation of what the results really mean.

A lot of variables are involved in using finite element
analysis to arrive at an accurate evaluation of a given
problem. Every problem has a range of loads,
material properties, geometry configurations, and
modeling methods that can have a profound effect on
the magnitude and accuracy of the results. It is up to
the analyst to look at the problem, the techniques
used in modeling it, and the results of the analysis and
determine what he thinks the real-world answer might
be. After reviewing the initial results from an analysis,
one might decide to change the element density,
modify some boundary conditions, increase contact
element stiffness, use a different convergence criteria,
or change any number of other factors in an attempt
to arrive at an answer that is sufficiently accurate for
what’s needed.
Here’s an example: Mesh density has always
played a significant role in obtaining accurate results.
The next time you post-process some results, try
arriving at the true answer by checking six items:
■ Averaged nodal result value (PLNS)
■ Unaveraged nodal result value (PLES)
■ Estimated error (PLES,SDSG or PLES,TDSG)
■ When using tetrahedral elements, compare
the PLNS values using Powergraphics to
those obtained when using full graphics
■ When using Powergraphics to view results
from elements with midside nodes, set
/EFACET to 2 and see if the midside node
results are higher than the results at the cor-
ner nodes
■ Look at the SBMN and SBMX values listed
on the full graphics plot

www.ansys.com ANSYS Solutions | Spring 2004

 To demonstrate this, the simple beam shown in
Figure 1 is fixed at one end and has a pressure load
on the free end. Figure 2 shows how the stress in the
fillet varies as mesh density increases. (See Table 1 for
the mesh densities and results.) With over 450,000
nodes in the model, the stress curves have yet to con-
verge to a common answer. Mother Nature knows that
there is only one answer, but which of the answers
from these analyses is the right one? Are any of them
right, or does the real answer lie somewhere in
between? Should you average the results of the vari-
ous plotting methods, or should you take a more con-
servative path and pick the highest one? This is for
you to decide, using your knowledge of the problem,
the way you chose to model it, and how your results
will be used.
Contact element stiffness is an artifact of finite
element analysis that does not exist in the real world. 37
We have to pick a contact element stiffness that
seems reasonable for our problem. If contact element
stiffness is too low, there may be excessive penetra-
tion that might cause significant error in our results.
If it’s too stiff, the analysis may not converge. How
do we know that the contact element stiffness used in
our analysis is appropriate? Like so many other
aspects of using finite element analysis, it depends on

Figure 1: A simple stepped cantilever beam with a fillet demonstrates how mesh density can influence
the accuracy of the answer.

Nodes Elements Powergraphics Powergraphics Powergraphics Estimated Error Full Graphics Full Graphics Full Graphics Full Graphics
PLNS,S,EQV PLES,S,EQV /EFACE,2 PLES,SDSG PLNS,S,EQV PLNS SBMX PLNS,S,EQV PLES SBMX
PLES,S,EQV
981 496 3614.52 4707.34 3614.52 1678.07 3304.25 4639.77 4707.34 6042.86
6020 3736 3620.28 4411.15 3984.66 1875.62 3325.5 4193.97 4411.15 5189.52
17747 11611 3975.48 4774.48 3975.48 1658.37 3624.26 4320.74 4774.48 5674.08
39309 26555 4040.1 4513.39 4040.1 983.59 3935.24 4466.05 4513.39 5011.23
85406 59407 4126.12 4269.61 4126.12 853.96 4040.6 4495.02 4269.61 4714.84
134319 94501 4115.29 4539.43 4115.29 909 4001.47 4364.92 4539.43 4994.15
214605 152459 4109.44 4342.61 4109.62 845.03 4013.97 4322.72 4342.61 4747.63
276101 195999 4180.16 4342.98 4204.41 789.89 4099.74 4408.77 4342.98 4671.72
468605 337203 4106.3 4220.36 4123.12 560.99 4072.92 4305.73 4220.36 4500.17

Table 1: Fillet stresses obtained by varying mesh density in a stepped cantilever beam.

www.ansys.com ANSYS Solutions | Spring 2004

www.ansys.com ANSYS Solutions | Spring 2004
 Tech File
the problem. You can look at the amount of penetra-
tion and see if it looks like it’s significant, but do you
really know how much penetration is acceptable and
38 how much is too much? One foolproof way of deter-
mining the best contact element stiffness is to make
several runs with increasing contact element stiffness
and compare the results. As contact stiffness
increases, the change in the results should decrease.
At what point do you declare that the answer is “good
enough”? It depends on the goals of the analysis and
how the answers will be used.
There are lots of other issues that we must think
about when reviewing our results. How does the prob-
lem respond to varying boundary conditions? If it’s
very sensitive to boundary conditions and there is
some variability in what the boundary conditions can
be, should you analyze the worst case or the average
case? How do you know which case is the worst with-
out analyzing several of them and reviewing the
results? Not all problems are simple enough for it to be
intuitively obvious which combination of boundary
conditions is the most severe.
Geometry can play a significant role in the vari-
ability of the answers we get. Sometimes the geome-
try we analyze isn’t an accurate representation of the
actual hardware. CAD/CAE geometry has perfectly
straight lines, fillets of constant radius, and is usually
the nominal shape. But if you look closely enough at
real-world geometry, you’ll see that it is never straight
and may not be nominal, either. Tolerances in manu-
facturing and assembly can have a significant effect
on how actual components behave. Are these incon-
sistencies significant for your problem? Which ones
should you include in your analysis?
Finite element models are rarely refined enough
to take into account the effects of surface finish, case
Figure 2: Von Mises stress results for the fillet region of a stepped cantilever beam.
www.ansys.com
hardening of materials, and other factors that can sig-
nificantly effect fatigue life calculations and other
downstream applications for your results. How do you
accommodate these factors in the utilization of your
results?
In an ideal world, we would run a number of solu-
tions to help develop a feel for a problem and allow us
to determine within a tolerable and quantifiable error
what we believe the answers to our problem really are.
However, in the real world, we usually have schedules
to meet and frequently don’t have the time to perform
the repeated analyses needed to gain a comprehen-
sive understanding of the problem. This makes the
analyst’s judgment a crucially important part of the
analysis process, since chances are good that you’ll
only have one chance to analyze the problem and
determine what the answer is likely to be.
Some of the variability in our results can be quan-
tified using the probabilistic design tools and varia-
tional technology capabilities that have recently
become available in ANSYS, but due to present-day
limits in computing power, we can only apply these
tools to a subset of the problems we are interested in
solving. As computer systems grow more powerful,
we will be able to use these technologies more and
more in our daily work and quantify some of the vari-
abilities in the problems we solve. However, these
capabilities will not relieve us of our responsibility to
carefully review results and make sound engineering
judgments regarding their validity and accuracy.
Rather than just report analysis results, we have the
responsibility to review them and interpret them in a
manner that is appropriate for the problem we are
solving. Mother Nature always knows exactly what the
answer will be. How often do we? ■
ANSYS Solutions | Spring 2004
 Tips and Techniques

Choosing an Element Technology SHELL181

Avoid problems by using the right elements
for each application
By ANSYS, Inc. Technical Support
PLANE182 39

The newer ANSYS elements (SHELL181, PLANE182,
PLANE183, SOLID185, SOLID186, and SOLID187) have a
choice of element technologies and material constitutive
behavior. The material and element technologies have been
separated, giving multiple combinations of behaviors with a
smaller element toolset.
The element technologies for the 18X series of ele-
ments are alternate approaches to conventional displace-
Shear locking causes the elements to be too stiff in
bending (also known as parasitic shear), especially in thin
PLANE183
members, or in elements with large aspect ratios. Shear
strains (the change in the element angle at a node) for bend-
ing loads are zero, but because a lower-order element can
only have straight edges, the element angle at each node
can’t be maintained when the nodes deform, inducing an arti-
ficial strain. A higher-order element, however, has curved

ment-based elements, which are not perfect for all analy-
ses. This article discusses the various element technologies
and gives tips on when to use them.
Conventional Displacement-Based Elements
Finite element analysis solves for the displacements at the
nodes, but the stresses and strains in the element are cal-
culated at the integration points, locations inside the ele-
ments (also called Gauss points). The number of integration
points depends on the dimensionality, the order of the ele-
ment and the element formulation chosen. Conventional
displacement-based elements use Gauss quadrature rules;
these rules prescribe the best sampling locations for evalu-
ating an integral numerically rather than analytically. A
4-node quad element has a 2x2 integration order, and an
8-node quad element has a 3x3 integration order. The inte-
gration points are arranged as such: a 4-noded quad ele-
ment has four integration points, each one a small distance
in from the nearest node, and an 8-noded quad has nine
integration points, one near each corner and midside
nodes, and one in the element center.
Gauss’s theorem is that if you use the number of Gauss
points in the rule, your integral will be accurate, so if you
use full integration, your strain energy calculation will be
exact. But, element accuracy does not guarantee solution
accuracy. These elements tend to be too stiff, and shear
and volumetric locking can occur.
Some guidelines on element selection are listed
in the accompanying chart.
SOLID185
edges, so it can maintain the element angles, and no shear
strain is induced.
Volumetric locking is an overly stiff behavior that is due
to incompressibility of the material. Incompressibility occurs
when plastic deformation is dominant, when the material is
hyperelastic or simply when Poisson’s ratio is at or near 0.5
for elastic materials. The bulk modulus of linear elastic mate-
SOLID186
rial, used to calculate the volumetric stresses, is equal to
E/(3(1-2*nu)). As Poisson’s ratio (nu) approaches 0.5, the bulk
modulus approaches infinity, and the element will lock.
The 18X series of elements has three element technolo-
gies to tackle these problems: B-Bar, Enhanced Strain and
URI to which can be added the Mixed U-P Formulation. All
are effective when dealing with nearly incompressible plastic-
SOLID187
ity and hyperelasticity, but Mixed U-P is required for fully
incompressible material, and only Enhanced Strain and URI
are effective in dealing with shear locking. Each technology is
explained below, with their pros and cons.
B-Bar, or Selective Reduced Integration
The strain at each integration point is calculated by multiply-
ing the displacement by the strain-displacement matrix, rep-
resented as B in finite element equations. The strain equations
can be divided into two parts: the volumetric (due to change
in volume) and the deviatoric (due to change in shape). By
evaluating the volumetric portion with the average over the
element while still evaluating the deviatoric portion at all four

18x Element Pros Cons When to Use

Formulation
B-Bar Efficient for nearly incompress- May be susceptible to shear locking Recommended for most
ible, bulk-deformation problems. analyses, especially bulk-
B-Bar with Efficient for fully incompressible, Same considerations as above, plus deformation problems.
Mixed U-P bulk-deformation problems. direct solvers required due to Mixed U-P
Enhanced Will handle bending dominated, Extra CPU time required for Recommended if problem is
Strain nearly-incompressible problems. Enhanced Strain terms (condensed bending-dominated. If problem
out at element level) has some bending, use when
Enhanced Will handle bending dominated, Same considerations as above, plus accuracy is an issue.
Strain with fully-incompressible problems. direct solvers required due to Mixed U-P
Mixed U-P
URI Will handle shear locking and Because of hourglass modes, Recommended choice only for
nearly incompressible problems. not recommended choice compatibility with LS-DYNA in
URI with Will handle shear locking and Same considerations as above, plus implicit-to-explicit or explicit-to-
Mixed U-P fully incompressible problems. direct solvers required due to Mixed U-P implicit simulations.

Note: For higher-order elements, URI is used by default. The only choice is whether to include the fully incompressible, Mixed U-P formulation.

www.ansys.com ANSYS Solutions | Spring 2004

www.ansys.com ANSYS Solutions | Spring 2004
 Tips and Techniques
integration points, the element is softer and overcomes vol-
umetric locking. However, it is still fully integrated for the
deviatoric portion, so it is still susceptible to shear locking.
This element formulation is only available for lower-order
40 elements, and is suited for nearly incompressible materials,
not fully incompressible. It is efficient for bulk deformation
problems, and recommended for most analyses, and is the
default for ANSYS elements PLANE182 and SOLID185.
Reduced integration with second-order elements is
frequently an excellent choice because the stresses are
highly accurate and there are no hourglass modes (except
in a one element model).
In the ANSYS GUI, B-Bar is referred to as “full integra-
tion” because it is the fullest set of integration points avail-
able for these elements.
Uniform Reduced Integration
URI is provided primarily for compatibility with LS-DYNA.
The other element technologies are usually preferable.
This formulation uses an integration rule one order
lower than full integration. Thus, a 4-noded quad will have
a 1x1 integration, an 8-noded quad will have 2x2, and a 20-
noded brick will have 2x2x2. Different from B-Bar, with URI
both the volumetric and deviatoric components use
reduced integration. This formulation makes the element
less stiff, eliminating the problem of shear and volumetric
locking. These elements may exhibit hourglass modes,
which would result in excessive displacements because
certain deformation patterns produce zero strain energy.
A small amount of stiffness, called the hourglass stiff-
ness, is applied to the element to control this. This value
can be changed, but it is recommended not to use too high
a value, as it has no physical meaning. Hourglassing can
also be avoided by not using point loads, or by refining the
mesh.
URI formulation can be used with lower-order or higher-
order elements, and are useful for counteracting shear lock-
ing or problems with nearly incompressible plasticity and
hyperelasticity. However, because fewer integration points
are used, more elements may be needed to capture stress
gradients. ANSYS higher-order elements (PLANE182,
PLANE183, SOLID186) use URI by default. It is generally
recommended to use B-Bar or Enhanced Strain for lower-
order elements instead of URI.
PLANE182 with Ogden
Material Model
www.ansys.com
Enhanced Strain Formulation
The Enhanced Strain Formulation (also known as
Incompatible Modes) adds additional degrees of freedom to
lower-order elements to allow curvature of the elements’ sides
and to counteract volumetric locking. These elements’ are
called Incompatible, because the curved edges lead to gaps
and overlaps in the mesh (there is no shared mid-side node to
keep the edges compatible). If the element is rectangular, it
models pure bending exactly, regardless of aspect ratio, but,
as the element becomes more trapezoidal, the elements do
not perform as well.
This element formulation is excellent for bending-domi-
nated problems, and is also useful for nearly incompressible
materials. This element formulation will require extra CPU time
to calculate the additional degrees of freedom. PLANE182
and SOLID185 have this formulation as one of their keyopts
(higher-order elements do not have shear locking, so they
don’t need this). Most other lower-order PLANE, SOLID and
SHELL elements support a subset of enhanced strain, called
“extra displacement shapes” or “bubble functions.” This for-
mulation adds additional internal degrees of freedom to allevi-
ate shear locking only, so it is still susceptible to volumetric
locking.
At 8.0, a Simplified Enhanced Strain Formulation has
been added for PLANE182 and SOLID185. This is similar to
the “extra displacement shapes” for PLANE42, SOLID45 and
other non-18X elements.
Mixed U-P
The Mixed U-P (also called Hybrid, or Herrmann) formulation
solves the problem of incompressible material behavior by
solving the hydrostatic (volumetric) as an additional degree of
freedom. The stiffness matrix is broken up into displacement
and pressure terms, and instead of solving just for the
displacements, ANSYS also solves for pressure degrees of
freedom for each element. Because this is solved for inde-
pendently, the hydrostatic pressure is no longer dependent on
Poisson’s ratio. This additional degree of freedom is solved for
an internal node generated by ANSYS that can’t be accessed
or seen by the user.
This formulation will cause the stiffness matrix to have
zero diagonals when the material is fully incompressible.
Iterative solvers can’t handle matrices with zero diagonals,
so a direct solver (Frontal or Sparse) must be used with this
element formulation. Also, if the number of pressure degrees
of freedom is greater than the number of unconstrained dis-
placements, the elements will lock. This problem can be over-
come by refining the mesh. Mixed U-P can be combined with
all three of the element technologies described above.
At 8.0, Mixed U-P has been extended for nearly incom-
pressible hyperelastic problems. It was for fully incompress-
ible hyperelasticity and nearly incompressible elasto-plastic
analyses prior to 8.0. It should be considered when Poisson’s
ratio is very high, such as greater than 0.4999.
Another new feature at 8.0 is activated by the ETCON-
TROL command. It will suggest and even modify element
formulation options for the 18x elements during solution. ■
For More Information
Additional information on element technologies is available at the following links:
http://www.ansys.com/industry/nonlinear/presentations/elements-18x.pdf
http://www.ansys.com/industry/nonlinear/presentations/180xAdvantages.pdf
ANSYS Solutions | Spring 2004
www.ansys.com ANSYS Solutions | Spring 2004