Electronic ebook

Structural
simulation
gets real
MSC Nastran the original and best

Authors: Al Robertson, Hugues Jeancolas,

Sarah Palfreyman and Dr. Keith Hanna
Table of contents
Foreword
Product development cycle
accelerated by 10x faster MSC
Nastran performance
Daimler Trucks, USA
Aeroelastic Co - Simulation flight
loads toolkit
Stirling Dynamics, UK
03 Leveraging MSC Nastran 26
embedded fatigue significantly
increases result precision
04 Husqvarna Group, Sweden
Efficient rotordynamic analysis 30
using the superlement approach
for an aircraft engine in MSC
07
Nastran
Pratt & Whitney Canada, Canada

Medical innovation one step 10 MSC Nastran based fatigue 34

further through MSC Nastran
MARC, MSC Software, USA
10x productivity gain with MSC 14
Apex - MSC Nastran workflow
MSC Software, USA
Volkswagen AG increase agility 15
and collaboration in engineering
lifecycle predictions at Navistar
Tech Mahindra, India
Tips and tricks - open source post 38
processing of MSC Nastran results
using HDF5
MSC Software, USA

development using MSC Nastran

MSC Nastran speeds forward 40
modules
Volkswagen, Germany
using graphics processing units
NVIDIA, USA
MSC Nastran and AVL EXCITETM 18
integration delivers better, faster Author profiles 45
and more efficient powertrain
simulations
MSC Software, USA

Co-simulation - Breaking the back 20

of multiphysics CAE simulation
MSC Software, USA
 Foreword
I could give you many reasons why structural engineers around the world rely
on MSC Software, but our best advocates have always been our customers.

In this eBook structural engineers from many industries share insights into how
they have addressed challenges by leveraging our solutions and heritage in
serving the structural engineering community for over 50 years.

You can learn, for example, how Pratt and Whitney leveraged powerful
model reduction techniques to obtain accurate and reliable rotordynamic
solutions for its flagship PT-6 engine without the computational costs of
traditional 3D models.

Or be inspired by Stirling Dynamics, a fast-growing aerospace company, which

was able to certify aircraft using aeroelastics through a partnership between
MSC Software and BAE Systems.

Discover how Daimler Trucks achieved a 10x performance boost in durability workflows over comparable FEA
tools, encouraging others to similarly strive for higher levels of productivity leading to more design variants
which leads to more durable vehicles.

Automotive engineers worldwide depend on MSC Nastran to address durability and fatigue, and in this eBook
the engineers at Volkwagen and Tech Mahindra illustrate why.

MSC Software and NVIDIA engineers have been working together for several years to ensure our customers
benefit from high performance computing by using GPUs to accelerate the sparse direct solvers in MSC
Nastran. Discover how Volvo has leveraged GPUs for dynamic analysis and achieved substantial savings in run
time, which are as much as 50% faster for high frequency runs.

Our customers also turn to us for validation and verification (V&V) expertise, which is the point where
simulation gets real. Read how Husqvvarna, which makes forestry and gardening tools, trusted MSC experts
on V&V when building new durability workflows for their power tools given MSC Nastran’s accuracy in
matching fatigue data.

You’ll also find out why the medical community is trusting MSC Software experts and solutions to help design
devices that can withstand vibration, non-linear stress, and thermal loading, all while meeting stringent
regulatory requirements.

Our simulation products continue to evolve, giving our customers new opportunities to innovate. MSC’s
Dr Keith Hanna writes about how co-simulation is pivotal to accurate predictive data. In particular, he
discusses how structural analysis will play a critical role in tying together everything from materials, design &
engineering, through to manufacturing and inspection, bringing the virtual into the real world.

We hope you enjoy this glimpse into how leading structural engineers are using MSC Nastran to solve real-world
problems more quickly, cost-effectively and accurately.

Ready to turn your challenges into the next exceptional story, sincerely,

Sarah Palfreyman
Product Marketing Manager for Structures Center of Excellence
MSC Nastran | The Original and Best in Class for Finite Element Analysis

Sarah.Palfreyman@hexagon.com

www.mscsoftware.com

3
Daimler trucks accelerates
product development cycle
with 10x faster MSC Nastran
performance
By Tim Rzesnitzek, Senior Analysis Engineer, Daimler Trucks North America
Daimler Trucks manufactures and sells commercial trucks
and buses as well as performs services and maintenance
of their products. Daimler strives for efficiency and
innovation, and the Daimler Trucks North America
development team is applying computer-aided engineering
(CAE) techniques to the truck development process to
help the company achieve these goals.
Truck manufacturers must ensure the safe carriage of
payloads while simultaneously lightweighting their vehicles
to improve fuel efficiency and associated emissions,
and the ability to predict the fatigue life of the vehicle
in the early stages of the design and development cycle
enhances product life, reduces testing and prototype
costs and accelerates time to market.
4
MSC Nastran is the crucial link in our durability analysis
workflow to perform finite element analysis. Our team
has selected MSC Nastran because we conducted
a market evaluation and found its performance was
consistently 10x faster compared to competing finite
element software products.
We perform multibody simulation in order to accurately
replicate the dynamic and non-linear bushing behavior and
to capture the road input by using sophisticated, elastic
tire models together with 3D road profiles. All durability
events of the physical test program are simulated in full
length according to the duty cycle. Different duty cycles
are used dependent on the vehicle type and the targeted
customer profile (e.g. highway vs. extreme duty application).
Workflow:
Multibody Simulation Time-domain
Forces Extraction Loading
Figure 1. Durability Analysis Workflow at Daimler Trucks NA
Figure 2. Simulation of all durability events of physical test according to duty cycle
Finite element methods are used in MSC Nastran to compute
the stress distribution in the vehicle substructures, like
chassis, cab or hood. In the FE simulation, the substructure is
excited in all interface locations with the loads retrieved from
the Multi-Body simulation.
The stress results of the FE analysis are used to calculate
damage for almost every metal component and selected
plastic parts. Finally, we receive a distribution of the
combined damage for a complete duty cycle as well as the
damage contribution of every durability event.
The FE analysis of a particular substructure is approached
in two steps. At first, we perform a normal mode analysis
in order to compute real modes, residual vectors and
modal stresses. In addition to the usual result data we
keep the MASTER/DBALL and IFPDAT files. This analysis
has to be executed once for a specific model.
In the second step we run a transient modal analysis
for every durability event, which is set up as a restart
run based on the normal mode analysis. A ‘restart’ is a
logical way of continuing from a previous run without
having to start from the beginning. In dynamic analysis,
the calculation of normal modes is, in general, the most
expensive operation in terms of computation time and
cost. Using the restart option, the normal modes do not
have to be re-calculated for every durability event. This
does not only save computation time but also ensures
5
Finite Element Analysis
Fatigue Analysis
Stress/Strain
consistency regarding the modes of the eigenmode
analysis and the participation factors of the several
durability events.
The main result of the second step is the computation
of the participation factors for every durability event.
Additionally, we also export the interface loads which
are applied to the FE model as well as the displacements
of selected nodes. The interface loads and the nodal
displacements are used to animate the time-dependent
deformation in response to the applied interface loads.
Once the transient response analysis is performed, we
use MSC Nastran results to derive stress time-history by
superposition of modal stress and participation factors
in nCode. E-N analysis for metal parts and S-N analysis
for plastic parts is performed for every durability event.
Through these analyses we achieve an accurate prediction
of the combined fatigue damage according to the duty
cycle. This completes our workflow.

Figure 3. Substructure of a truck - with input loads from MBS simulation
SOL103
• Computation of eigenmodes, residual vectors and
according modal stresses
• To be performed for once for a specific FE
Solution Time Using MSC Nastran and
Competitor Finite Element Solvers
When we compared the solution time for normal modes
analysis using MSC Nastran and the competing finite
element solver we had been using, we observed that
MSC Nastran can solve this model in 5 hours whereas
our previous finite element analysis product took
over 50 hours to perform the same analysis using the
samemachine environment. Normal Modes Analysis (SOL
103) is 10 times faster using MSC Nastran.
The performance issues using the former finite
element solver were mainly related to the modal stress
computation of second order solid elements.
The validation of a full vehicle incorporates the analysis
of the chassis, cab and hood structure. By performing
these analyses with MSC Nastran we reduce the
overallturnaround time by a couple of days. With that, the
use of the MSC Nastran solver is crucial to perform our
durability analyses.
Figure 3. Simulation of sub-structures with input loads from
MBS simulations
SOL112
• Restart based on SOL103
• NO re-computation of eigenmodes
Keep DBALL
• Computation of modal participation factors only
MASTER • To be performed for every durability event (track)
IFPDAT
Summary
The Durability process at the Daimler Trucks North
America has been outlined above. The durability process
has enabled identification of high stress locations more
effectively. We found the overall process to be nearly ten
times faster than the previous method. With fewer and
smaller intermediate files, the bookkeeping is also much
simpler. We have successfully validated this methodology
based on several parameters and it has exhibited a good
level of correlation with our physical test data. The Daimler
Trucks North America team see several opportunities
in running similar simulations in the near future. MSC’s
robust offering of MSC Nastran solver, continuous support
from field team as well as continuous investments in
the development has made a significant difference in
simplifying and speeding up our durability analysis.
6
Aeroelastic Co-Simulation
flight loads toolkit
By Martijn Renooij, Stirling Dynamics, UK
A part of the Expleo Group of companies, Stirling
Dynamics is a fast-growing, advanced engineering
company that delivers a range of complex systems and
technical services to the aerospace and marine markets.
Trading since 1987, the company has accumulated a wealth
of knowledge on over 70 different aircraft types and
delivered to both civil and military programmes around
the globe Recognised as leading experts in the areas of
aircraft loads, fluid dynamics and aeroelastics, Stirling
Dynamics is approved to the global aerospace quality
standard AS9100 as well as being a member of the ITAR
Approved Community.
7
Introduction
Stirling Dynamics has developed a non-linear aeroelastic
toolkit for aircraft loads (Reference 1) as part of a UK
NATEP (National Aerospace Technology Programme)
initiative in collaboration with MSC Software UK and
supported by the end-user, BAE Systems. Aircraft
loads assessments (like gusts and manoeuvre loads)
are typically performed using linear aircraft models
and, although this is generally considered to be an
acceptable means of analysis, including non-linear terms
improves themodelling accuracy and reliability. Non-
linear aeroelasticsolutions are generally only available to
the major aerospace OEMs who have their own tailored
toolkits. The current processes used by most aircraft
companies, excluding the two largest OEMs, is based on
linear assumptions and this has been accepted as means
of compliance for generating gust and manoeuvre loads
for aircraft design by the certification authorities. This
puts smaller OEMs into a disadvantageous position as
the linear models are generally considered to beoverly
conservative. The project objective for Stirling Dynamics
was to develop their own in-house toolset. In parallel, as
part of the same programme, tool development at MSC
Software UK was aimed at developing a commercially
available product. A more detailed description of the MSC
development is covered in the following sections.
Figure 1. Generic UAV Model for this study (courtesy of BAE Systems)
Figure 2. Generic UAV Model inside the Toolkit user interface
8
MSC Software Co-Simulation CFD-FEA
Coupling
A key feature of the Aeroelastic CFD Manoeuvres Toolkit
is that it is based on the widely used MSC Nastran for FEA
structural analysis, and scFLOW from Cradle for CFD, plus
inputs from the end-user, BAE Systems. The tool allows for
increased fidelity of the nonlinear aeroelastic effects that
contribute to the loads experienced by an aircraft over a
wide designof- experiment (DoE) design space. The tool
takes nonlinear aerodynamic effects into account from
shock movements or flow separations around an aircraft,
improving the accuracy and simplicity of loads modelling
on flexing structures such as aircraft wings. MSC Software
provided a robust and reliable commercial CFD and FEA
solver co-simulation toolset to enable this. The toolkit also
automates the simulation process very substantially. The
toolkit includes new methods and it provides for:
• Extraction of aeroelastic loads from multiple scFLOW
CFD analyses
• Application of the fluid load to an aeroelastic MSC
Nastran model for various trim conditions
• Coupling of the aerodynamic loads in all 6 degrees of
freedom (DOFs) to the structural FEA model In tandem
with BAE Systems, a generic Unmanned Aerial
In tandem with BAE Systems, a generic Unmanned Aerial
Vehicle (UAV) demonstration was created for BAE Systems
to showcase the tool (Figure 1). The toolkit user interface
(Figure 2) automates the direct mapping of CFD results
from scFLOW onto finite element models from MSC
Nastran to then predict and visualise aeroelastic effects.
Summary

As part of a UK NATEP (National Aerospace Technology
Programme) initiative in collaboration with MSC Software
UK and supported by the end-user, BAE Systems, Stirling
Dynamics has developed a non-linear aeroelastic toolkit
for aircraft loads for use in their engineering programmes.
In parallel and part of the same programme, MSC Software
has been able to develop Dr Simon Hancock, R&D Manager,
Stirling Dynamics an innovative non-linear structural
analysis-CFD solution with MSC Software simulation tools
that will lead to more optimised aircraft aeroelastic models,
and higher simulation fidelity resulting in the reduction of
uncertainties in fluid-structure analysis predictions.
The Aeroelastic CFD Manoeuvres Toolkit means
thatuncertainties in the aircraft modelling process
will be reduced, which results in increased accuracy in
the CAE models produced, less conservative aircraft
designs and lighter future aircraft. The key benefit of
lighter aircraft will be less fuel burn and therefore more
fuel-efficient flights and significant cost savings to
aircraft manufacturers and users; this ultimately has
environmental benefits for us all.
A demonstration of this toolkit by way of a generic
UAV model was produced for BAE Systems in the UK.
This toolkit will be valuable to aid in future aircraft
certification requirements for emerging and small OEM
aircraft manufacturers.
Reference
UK NATEP Website Projects: http://www.natep.org.uk/
projectdesign-modelling and http://www.natep.org.uk/
documents/ADS%20Natep%20210%20x%20210%20
Supplement%20v5.pdf

Stirling has a long history of developing independent aircraft design tools.

Working on this NATEP project together with MSC Software and BAE Systems
has been an excellent opportunity to develop our capabilities further and to
use these new tools for future aircraft design.”

Dr Simon Hancock,
R&D Manager, Stirling Dynamics

9
Medical innovation
one step further through MSC
Nastran and Marc Simulation
By Matthew Kokaly, Product Manager, MSC Nastran and MARC
In this article, we are showcasing some interesting case
studies performed using MSC Nastran and Marc to
see how the finite element analysis is leveraged in the
medical industry and taken innovation one step further
using simulation.
Hearing is one of the five sensations critical to the
perception of the surrounding and communication.
The overall function of the ear is to convert physical
vibrations to nervous impulses. In other words, the
vibration from sound is transduced to electric signals in
the ear, where they are processed by the central auditory
system in the brain. Structural analysis is a crucial part of
any design, implementation, and maintenance process in
vehicles, machine tools, spacecraft, buildings and even
medical devices.
10
Bone-Anchored Hearing Aid, which is commonly referred
to as BAHA provides patients with a higher level of user
satisfaction compared to other hearing aids. With the help
of computational mechanics such as the Finite-Element
Analysis (FEA), the performance of BAHA can be improved
before actual costly physical models are built.
Lena Kim, Masters Candidate at Chalmers University
of Technology, built a 3D FEA model of Human Head to
perform a study which leverages MSC Nastran’s FEA results
to correlate with the physical tests which in-turn helps the
cost-reduction of the Bone-Anchored Hearing Aid.
In this study a valid 3D FEA model of the human head is
developed, which is used to investigate and simulate the
vibration pattern of bone-conducted sound using MSC

Figure 1. Structure of ear: Outer, middle and inner ear
Figure 2. Schematic overview of BAHA implant in function
Nastran. FEA models enable us to investigate the factors
that affect the bone conduction pathway, find the correct
position to produce the vibration level for the hearing
sensation, and further optimize the device in a patient-
specific way.
Structural analysis- specifically modal analysis is the first
step and plays an important role in the sound and vibration
analysis. By performing modal analysis, one can find the
system’s natural frequency and mode shapes (shape of
vibration) without external force and damping. The results
of modal analysis characterize the basic dynamic behavior
of the structure and indicate how the structure will
respond under dynamic loading.
11
Figure 3. BAHA device, meshed anatomical data of skin and 3D mesh of
the skull
FE analysis approximates the geometry or original model
in several different types of elements. The behavior of the
structure of interest is obtained by analyzing the overall
behavior of the elements. The schematic in Figure 4
illustrates the process from the real model to visualization
of result. Each of these procedures requires several sets
of commercial software which includes MSC Nastran for
structural analysis, Beta CAE pre and post-processor
ANSA and Meta-post.
The dynamic frequency response was simulated using
a commercially available structural analysis software,
MSC Nastran. For this model, loads, frequency range,
analysis output, and damping coefficients were assigned
in the form of Nastran codes. Two types of analysis were
performed in this study: normal mode and frequency
response analysis. The response of the skull surface of the
surrogate was simulated and the output was the velocity
for mechanical point impedance. The test data was
compared against results using point mass method using
MSC Nastran and the results show good agreement with
test data.
To conclude, using MSC Nastran frequency response
analysis of head simulator model was performed at
mechanical point impedance (MPI). The results were
remarkably consistent with physical test data in both Non-
strutural mass (NSM) and Fluid-Structure (FS) models with
anti- resonance frequency at approximately 75-90 Hz; there
was only a 5% difference in magnitude level. This study
proved that FEA results are close to the physical test and
hence they result in cost-reduction of the BAHA device.
Prosthetic feet using carbon fiber Real model
composites: Design, simlation & testing
CAD or CT scan
Prosthetic limbs represent an area of advanced
biomedical device technology where artificial limbs meet
with advanced aerospace-grade composite materials.
The development of prosthetic feet using carbon fiber Discretized finite element model by pre-processor
composite materials has paved the way for many amputees
to resume an active and athletic life-style. By combining Different types of elements, Loads Boundary conditions,
Material properties ... etc
advanced materials, an understanding of design and
stiffness tailoring of composite materials, and aerospace
manufacturing techniques, the end result is prosthetic
limbs with realistic flexure, “springiness”, and strength. Simulation in Solver
These lifelike prosthetics are a far cry from the “wooden
leg” or “stump” associated with injured pirates of old! The Normal modes analysis, Frequency response analysis
technology has advanced so far that that runners with
composite feet and legs are qualified to run in the Olympics!

The active users demand for “springy” but, strong Calculation

prosthetic legs. Major challenges to create these type
of Prosthetic legs include Fatigue Durability, balance Mode shapes, displacements, velocity, acceleration
between strength, stiffness and weight amongst many
others. The demand for a prosthetic foot that is suitable
for all terrain, with lightweight, exceptional shock
absorption and great energy return is very much needed. Visulization by post-processor
Composites play a huge role for lightweighting. Dr. T. Kim
Parnell from Parnell Engineering and Consulting used Graphs, Videos ... etc
MSC Marc to perform Finite Element Analysis on the heel
design, test different material properties and come up Figure 4. Finite element process
with an optimal design.

Model description was done using MSC Marc/Mentat,

where the contact bodies are defined such that Heel,
Urethane are flexible contact bodies and Overload,
Keel are assumed rigid and materials are defined as
composites with Tsai-Wu failure criteria and the results of
2 Urethan Configurations are compared in order to come
up with the final design.

After performing analysis using Marc, it was concluded

that, Short Urethan Heel, 1/8” thick causes increased heel
bending and it is more flexible whereas, Long Urethane
Heel, 1/16” thick, has reduced heel bending due to more
uniform heel contact, stiffer response due to earlier
contact with Keel.

Figure 5. Mechanical point impedance magnitude at BAHA position in

After comparing different results of both heel types, NSM model. The red line represents the simulation result with NSM
it wasconcluded that the Proposed 1/16” Long Heel is model and the blue line indicates the experimental data from Cochlear.
stiffer due to earlier secondary contact with Heel and
earlier contact with Keel. 1/16” Long heel gets progressive
stiffening and the stress results for heel improved even for
same displacement and thus higher applied load.

To conclude, simulation using MSC Marc/Mentat

with physical testing help to better understand the
delamination failure mode.

12
Figure 6. Carbon Fiber Prosthetic Foot - Components: Heel, Keel, Overload, Urethan (Toe and Heel)

Figure 7. Proposed 1/16” Long Figure 8. Initial 1/8” short

Dr. T. Kim Parnell from Parnell Engineering and Consulting

used MSC Marc to perform Finite Element Analysis on the
heel design, test different material properties and come up
with an optimal design.

13
10x Productivity gain with MSC
Apex –MSC Nastran workflow
By Zanlang Yin, Product Marketing Manager, MSC Software

In recent years, OEMs in the aerospace, automotive,
and shipbuilding industries are being confronted with
a variety of challenges. Stricter safety requirements,
increasing operation cost, and higher user expectations,
are driving OEMs to innovate, especially in product
development to gain continuous competitive advantages
in the market. These strategic initiatives also bring new
requirements for CAE tools. User experience, smart
workflows, simulation democratizationyou name it! No
matter what buzzwords are out there, the core need of
CAE users is the same: productivity.
MSC Apex Iberian Lynx is the ultimate CAE productivity
booster for MSC Nastran based users. Beginning with
Iberian Lynx, the interoperability between MSC Apex and
MSC Nastran enables the new “MSC Apex – MSC Nastran
– MSC Apex” workflow. Users can now combine the world-
class user experience of MSC Apex with the power of MSC
Nastran, the world’s most trusted multidisciplinary Finite
Element Analysis (FEA) solver, to facilitate their product
development. Such interoperability enables existing MSC
Nastran users to achieve an up to 10x productivity gain in
several different aspects.

Though determining go-to-market of new products, the

productivity of product development depends heavily
on CAE simulations, where innovations are fertilized. In
current workflows, however, CAE engineers are doing
chores instead of innovations due to their software. They
need to transfer model files across the pre-processor,
FEA solver, and post-processor to perform a single round
of analysis. In most cases, CAE engineers have to wait
for a modified model from design engineers to begin the
next round analysis. Such a process generally repeats
multiple times until a final design is frozen. Productivity is
often further constrained with poor user interfaces, non- MSC Nastran

intuitive instructions, and redundant manual operations

of the software. As a consequence, CAE engineers
spend 50%~70% of their time in model preparations.
The engineering teams are in desperate need of a
next-generation simulation tool to enable productive
simulation-driven workflows.

Figure 1. Apex – MSC Nastran – Apex Workflow

14
Volkswagen AG increase
agility and collaboration in
engineering development using
MSC Nastran modules
By Dr. Bruns Dr.-Ing. Jürgen Bruns, CAE-Methods, Volkswagen AG
As we all know in the finite element world, traditional FEA
model build-up contains multiple parts and substructures
which are then organized in finite element codes (such as
‘include’ files in MSC Nastran) for easy exchangeability.
While using these include files, a strict numbering
guideline needs to be followed to avoid ID collisions
between INCLUDEs.
It is required in MSC Nastran that a complete model
is builtup without overlapping of ID’s. This sort of build
provides independent INCLUDE files to handle which
are easy to reuse and interchange whenever the vehicle
design changes. However, with this method, it is hard to
manage these files as the users must follow an elaborate
numbering scheme. This method needs high discipline,
15
monitoring effort as well because it needs to be adapted
regularly due to new requirements.
Volkswagen has enjoyed a long-standing relationship
with MSC Software for many years. Volkswagen not only
uses the advanced technologies from MSC Software
for their analyses, they also drive the requirements for
the new developments in MSC Nastran. One of the most
recent developments is “Modules” in MSC Nastran.
As computational resources are becoming cheaper,
models are getting larger. These larger models require
efficient model management. With this new technology
called “Modules” in MSC Nastran, whose development
was driven by Volkswagen, the model management now
becomes very accessible.
Modules are introduced in MSC Nastran as an assembly
management tool; it is a standalone Bulk Data section
that typically contains a part or component of an overall
assembly. Modules are analogous to part super elements
without reduction. MSC Nastran can also create multiple
instances of a primary MODULE with a process referred to
as MODULES Instantiation.
Now, there is also a need to connect these different
modules, for which manual, automatic and semi-automatic
methods are available for connecting modules.
Each module has its own ID numbering scheme for grid
points, elements, properties and so on. The ID numbers do
not have to be unique across Modules.

The key benefits of modules are that they reduce reliance
on a pre-processor for creating copies of, translating,
rotating, or mirroring an existing primary MODULE – this
also helps reduce modeling time. Additionally, MSC
Nastran’s MODULE uses similar definitions as some other
codes – this allows for quick translating of a MSC Nastran
BDF for other simulations such as crash simulation with
PAM-CRASH.
“The modules approach helps us save time and ensure
our high-quality results. We can use the decrease effort in
model build-up, exchange, as well as pre-post processing
for NVH, Durability, and Nonlinear Analysis,” says Dr. Bruns
from Volkswagen.
A module is a standalone Bulk Data section that typically
represents a component of an overall assembly, e.g., an
automobile wheel or fender. Modules allow the users to
assemble multiple Bulk Data sections to form a single MSC
Nastran input file. Each module is delimited by the BEGIN
MODULE and END MODULE Bulk Data entries.

MSC Nastran

Figure 1. Traditional Model Build Up

The modules approach helps us save time and ensure our high-quality results.
We can use the decrease effort in model build-up, exchange, as well as pre-
post processing for NVH, Durability, and Nonlinear Analysis.”

Dr. Bruns, Volkswagen

16
Figure 2. Model Build up at Volkswagen

Just like in part super elements, each module may have its
ID numbering scheme for grids, elements, properties, and
the ID numbers do not have to be unique across modules.
Post-processing results are available in the HDF5,
F06,punch, and op2 files, which are partitioned by module.

So, to conclude, MSC Nastran Modules have the

samecharacteristics as INCLUDEs, but without any ID
collisions. Modules reduce the “numbering” work in
the model assembly processes. Modules enable the
automotive OEMs to easily exchange the models with
other analysis groups (such as statics, crash, NVH). Finally,
userscan use switch variants easily to try out different
designs in the vehicle development.

Figure 3. MSC Nastran Module Overall Bulk Data Section

17
MSC Nastran and AVL EXCITE™
integration delivers better,
faster and more Efficient
powertrain simulations
By Bernhard Loibnegger, AVL Paul Herster, AVL Excite Product Manager
Originally developed for NASA engineers, MSC Nastran
is the most trusted FEA solver for design and analysis of
aerospace and automotive structures. Industries have
commonly used the proven solutions of MSC Nastran for
the past 50 years. To build upon the same practice, MSC
Nastran provides streamlined workflows for powertrain
simulations, improved verification options for contact
models and enhanced HPC methods for simulation of
large-scale problems.
Through a direct coupling between MSC Nastran and
the AVL EXCITE, engineers can perform the analysis of
dynamics and acoustics of powertrain systems faster and
more accurate.
AVL EXCITE is a multi-body dynamic simulation software
that provides rigid and flexible solutions for the analysis
of powertrain systems. The robust solutions of AVL
18
EXCITE family are equipped with advanced technology
to calculate dynamics, strength, vibration, and acoustics
of conventional and electrified powertrains. Engineers
can create accurate models that account for all physical
properties of materials and lubricants to predict contact
behavior, friction, and wear in slider bearings, pistons,
piston rings and gears.
With the new direct coupling, AVL EXCITE can now be used
with MSC Nastran to increase the accuracy of Multi-Body
simulations by providing a better representation of the
flexible components. Similarly, the FEA engineers can
benefit from this integration to examine each component
with realistic boundary conditions and dynamic loads.
Direct export of AVL EXCITE input (.EXB) files from MSC
Nastran makes data transfer much easier and faster,
eliminating the need for creating and translating extensive
intermediate data.
Figure 1. MSC Nastran SOL 103 to AVL Excite to SOL 111 Workflow
The workflow allows engineers to obtain a full
characterization of structural response by leveraging the
nonlinear solution of MSC Nastran (SOL 400) to include
contact nonlinearities as well as preloaded conditions in
their powertrain simulations. Additionally, MSC Nastran
can read back output results from AVL EXCITE for further
dynamic and acoustic analysis of sound radiations
generated from Internal Combustion Engines and other
components. This technology is also beneficial for vehicle
durability simulations, taking advantage of the award-
winning embedded fatigue capabilities of MSC Nastran to
obtain fatigue life and damage responses of components.
This functionality will be of benefit to the automotive
industry by providing the analyst that uses MSC Nastran
and AVL EXCITETM and workflow to perform various
analysis (like dynamic, vibrational, acoustical) - a direct,
convenient and efficient method to exchange data.
The enhanced MSC Nastran-AVL EXCITETM Interface with
MSC Nastran has multiple advantages over the currently
used DMAP alters method:
Figure 2. MSC Nastran SOL 400 to AVL Excite to SOL 400 Workflow
19
1. Direct: Creation of EXB file from MSC Nastran rather
than exporting multiple op2/op4 files from MSC
Nastran which are converted by another standalone
proprietary program into an EXB file.
2. Simple: Only two files are created during step-1
(EXB and external superelement op2) as opposed to
multiple op2/op4 files (GEOMUNIT, EIGNUNIT, PSTUNIT,
GPSRUNIT, MKAAUNIT, MFFUNIT, TRANUNIT).
3. Fast: For step-1, shared memory parallelization
(SMP) would be enabled to make the calculation
of computationally intensive higher order mass
invariants faster. For step-2, EXTSEOUT feature
is leveraged to minimize data storage and enable
efficient data recovery in MSC Nastran.
4. Convenient: Allows for quick export of EXB file to AVL
and quick import of AVL results back into MSC Nastran.
The direct, efficient, fast, and convenient method
to exchange data between the two solvers enables
significant time savings, resulting in up to 3x performance
gain for analysis of powertrains and engine systems.
Co-simulation - Breaking the back
of multiphysics CAE Simulation
By Dr. Keith Hanna, Vice President Marketing, MSC Software
Over the last 20 years or so, one of the ‘holy grails’ of
Computer-Aided Engineering (CAE) has been ‘multiphysics’
simulations, i.e. co-simulations between different physics
simulation types (Reference. 1). Multiphysics, even though
an ill-defined term, kind of feels intuitively right to us
engineers from the perspective of real-world engineering
simulation. Fluids don’t usually exist in isolation from
structural effects, or acoustics, or dynamics in the real
world we inhabit, yet for many years past, engineers have
simulated these subdomains of physics as isolated point
simulation solutions in CAE. I have to add a rider here that
my Physics teacher at High School, Mr Copeland, would
have taken me to task for using the word ‘multiphysics’ at
all because strictly it is wrong. There is actually only one
discipline of physics, as there is for chemistry and biology.
Within physics, there are subdisciplines like Newtownian
Laws of Motion (and fluids),
20
Maxwell’s Laws of Electromagnetics, Structural Stress-
Strain laws, etc. This ‘multiphysics’ word that was coined
in the 1990s is actually a consequence of a failure of
the CAE industry to solve the fundamental underlying
physics equations in a combined way, in my opinion. Partly,
this is due to practical problems that we find because
of different mathematical techniques. For instance,
Finite Volume Methods (best for fluids), Finite Element
Methods (best for structures and acoustics) mean
that for more efficient solver convergence in real world
engineering problems, one or the other is chosen as the
best methodology. And partly, software vendors have
struggled to grapple with the multiphysics challenge to
deliver usable engineering simulation tools because the
mathematical approaches don’t tend to gel well together
when combined.
Figure 1. MSC Software’s Unique Multiphysics Co-Simulation Capabilities

What was very striking about MSC Software when I joined
earlier this year, apart from its 50+ year history stretching
back to the early NASA moon landing project, and the
desire for accuracy in software solutions, is a largely
understated history of endeavoring to provide world-class
CAE co-simulation solutions to its customers as well as
engineers everywhere. Everything from acoustics to Multi
Body Dynamics (MBD), to CFD, to structural analysis, and
explicit crash dynamics can be connected together in
MSC Software’s (see Figure 1). I have gathered together
a multiplicity of multiphysics co-simulation applications
that can be done with MSC software tools today, both
in two-product couplings, as well as in toolchains of
productsimulations that were mere pipe dreams a few
years ago (see Table 1).

Co-Simulation Multiphysics area Products involved Industries Applications

Nastran / Marc + Aeroflutter, Value opening, MEMs, VIV, suspension

Fluid-structure Interaction (FSI) All
scFLOW/scSTREAM loads, thermo-mechanial stress...
Nastran / Marc +
Structural & Aeroacoustics All Cabin noise, door rattle, noise & vibration
scFLOW + Actran
Large particle movement, vechicle side wind events,
Multi-Body Dynamics & Fluids Adams + scFLOW All
vehicle running over a puddle...
Automous vehicles, ADAS validation real time vehicle
Virtual Drive & Vehicle Dynamics Adams + Vires VTD Automotive
driving simulator...
Adams + EDEM, EDEM + Car stability on a surface, filtration, bulk material
Particlates & MBD & CFD Auto, Aero Chem& Proc
scFLOW handling...

Door sag & closing, vehicle extreme load cases (eg.

MBD & Nonlinear FEA Adams + Marc Automotive hitting a kerb), running over an obstacle, battery pack
deformation...

Adams + Easy5 / Matlab

Robot arms, machinery, landing gear system, vehicle
1d Systems & MBD & Controls Simulink / Maplesoft / GT All
ABS, ESC, traction control...
Suiteetc

Table 1. A Cross Section of Co-Simulation CAE applications connected to the MSC Software Solution Suite

21
Multi-Body Dynamics Led Co-Simulation

When one digs down, one soon discovers that the market
leading Adams MBD product is being used extensively for
industrial grade co-simulations around the world today
as a matter of course, and is open to connections with all
sorts of commercial and open-source software (see Figure
2). Some notable recent innovations are connections to
the market leading Discrete Element Modeling software,
EDEM, from DEM Solutions, as illustrated by Figure 3,
VIRES Virtual Test Drive (VTD) for ADAS and autonomous
vehicle drive simulation (VIRES has been part of Hexagon
| MSC Software since 2017), and of course MATLAB & Figure 4. MSC’s Adams MBD Co-Simulation Capability Ecosystem
SIMULINK for 0D/1D systems and controls modeling (from
MathWorks). Adams coupled with Marc can solve wiper
blade movements, hydraulic actuators, rubber door seals,
and full vehicle bush modeling. I will go into more detail
below about the exciting new co-simulation multiphysics
capabilities that the CFD software acquisition of Software
Cradle by MSC Software in 2016 is now beginning to open
up for CFD and CAE engineers alike. A very complex and
challenging dynamic co-simulation is shown in Figure 4
where a non-linear bending and twisting metallic flap in
a cross flow was simulated in an scFLOW - Marc - MSC
Nastran toolchain.

Acoustics Led Co-Simulation

Actran is recognized to be the market leader in acoustics
simulation with a twenty-year history behind it and a
blue chip user base worldwide. It has many multiphysics
acoustic applications like the classic one of coupling
with CFD for aeroacoustic analysis. Actran provides
today a complete endtoend solution from mechanical
load prediction to noise and mechanical load prediction
to noise and vibration assessment. in a recent Webinar,
(Reference 2) my colleagues Romain Baudson and Yijun
Fan looked at acoustic predictions for electric vehicles
with Actran-led toolchains.
Figure 5 shows the two toolchains they employed in their
cosimulation analyses. Multibody simulation with Adams
provides the structural loads when the conventional
powertrain is operating, then MSC Nastran was used
for vibration analysis. Finally, noise radiation is then
Figure 5. MSC’s Nastran – scFLOW - Marc Co-Simulation of a flap bending
and twisting in a crossflow
computed with Actran with high accuracy in a fully
automated process. By comparison, for Electric Vehicles,
electromagnetic simulation results from various third
party software (for example, JMAG, MagNet, Maxwell…)
can be integrated with MSC Software’s FEA tools for
electric motor noise prediction. In doing so, Actran
provides accurate noise predictions for both the electric
motor and the gear reducer. Finally, a relatively recent
acoustics and fluids co-simulation is illustrated in Figure 6
between Actran and scFLOW from Software Cradle where
an automotive exhaust and muffler system was modeled in
CFD and coupled with Actran for acoustic noise prediction.
CFD Led Co-Simulation

Computational Fluid Dynamics (CFD) has long been a

passion of mine and it is genuinely quite exciting to see
what is happening in Japan with Software Cradle’s co-
simulation CFD efforts. They have devised a remarkably
robust and open software platform for co-simulation (see
Figure 7). Software Cradle’s CFD products – scFLOW (that
employs polyhedral meshes), scSTREAM (with Cartesian
meshes) and SC/Tetra (with tetrahedral meshes) have a
long history of use by large OEMs and leading Japanese
companies and come with a strong technical pedigree.

They are technology leading in free surface and overset

meshing techniques with fast, accurate and robust
Figure 3. MSC’s Adams MBD Co-Simulation Capability with EDEM from approaches. Physical quantities can be passed between
DEM Solutions third party and MSC Software’s CAE software and Cradle’s

22
Figure 6. MSC’s Adams – Nastran and Actran Conventional Automotive Powertrain Co-Simulation Toolchain (left) and E-Mag – Nastran – Actran
e-Powertrain Toolchain for Acoustics (right)

Figure 7. MSC’s Cradle CFD and Actran Acoustics co-simulation for an Automotive Exhaust and Muffler Aeroacoustics Prediction

CFD tools using its FMI (Functional Mockup Interface).

The FMI supports general physical quantity settings,
user-defined functions, and script languages. With
Cradle | scFLOW

Adams in particular, this co-simulation approach can do

large particulate flows on free surfaces (Figure 8), vane
pumps, washing machine drum vibration, and MBD fuel
tank sloshing (Figure 9) for instance. A really exciting
MBD application is that of a vehicle driving through a
large puddle with suspension effects taken into account
by Adams or the same vehicle’s dynamics in a cross
wind (Figure 10). Co-simulation with Marc means that
Cradle CFD couplings can do aircraft fuel tank baffles,
liquid quenching, flexible plates / membranes / valve
seals (Figure 11), and Sirocco fans. When coupled to MSC Cradle | scFLOW

Nastran, scFLOW can do FSI of ship’s propeller with

cavitation effects included for example (see Figure 12).

It’s the really cool co-simulation CFD applications that

catch your eye the most, however. Software Cradle
have coupled Adams with scFLOW to look at dynamic
aircraftwing and flap deployment aerodynamic predictions
(see Figure 13).

Another exciting new aerodynamic cosimulation Adams

application between scFLOW and MSC Nastran / Marc is

non-linear panel flutter of a supersonic plate (Reference Figure 8. Software Cradle’s Open Coupling CFD Platform (Left) and FMI
3), involving a highly unusual coupled undulating vibration Interfaces (Right)
of the flat plate’s surface at high speeds resulting in
moving shocks on the surface (see Figure 13). It was
accomplished by my colleague Fausto Gill di Vincenzo in
Italy who employed our seamless MSC Nastran - scFLOW.

CFD coupling technology for an efficient and fast solution.

The CFD domain deformation prediction was done with
sliding mesh capabilities and then the FEA was performed
by MSC Nastran and Marc. The CFD elements used could
be tets, hexas, or polyhedral elements.

23
Figure 9. Adams – scSTREAM prediction of large logs in a channel
free surface flow

Finally, another area of interest of mine is

thermomechanical stress predictions for electronic
components on chips and PCBs (Printed Circuit
Boards). This application too is quite difficult to do as
a cosimulation by most commercial vendors. In Figure
14, an electronic chip resistor has been simulated that
undergoes repeated heat generation by its own on-off
actions (Reference 4). This will eventually cause the
breakage of its solder joints over the lifetime of the
component and ultimately product failure. Being able
to predict it in advance and to locate high stress areas
is very important. Temperature distribution predictions
from scSTREAM were mapped onto the mesh of the
structural analysis solver (e.g. MSC Nastran or ANSYS
Mechanical). The stress on the solder connection was
then predicted. This is a very powerful way of predicting
failure mechanisms and their likely locations in the
consumer electronics industry and is relatively easy to do
in scSTREAM as a co-simulation.
Figure 10. Adams - scFLOW prediction of dynamic sloshing of liquid in a tank

Summary

Classical ‘multiphysics’ challenges for CAE include fluids

and structures (FSI, VIV), structures and acoustics,
structures and dynamics, and fluids and dynamics.
Multiphysics simulation may sound easy in principle yet
many commercial CAE vendors have had point physics
solutions for decades now but have failed to implement
usable coupled solutions for industrial grade engineering
application (either loose- or closelycoupled CAE). So,
‘multiphysics’ is still one of the ‘Holy Grails’ of CAE solution
even after 40 to 50 years of point simulations in all the
physics sub-disciplines. MSC Software has grappled with Figure 11. Adams – scSTREAM prediction of a dynamic vehicle
this co-simulation conundrum and come up with many suspension movement through a puddle
usable solutions and tool chains for real world engineering
applications. I would encourage you to explore what we
offer and use MSC One credits to try them out. Ask your
local MSC account manager or support engineer if you
want to know more. We offer both the technologies and
the professional services that can help you solve your
industry CAE co-simulation challenges.

24
Figure 12 FSI analysis using Marc and scFLOW of a Valve with an Elastic Figure 15. Adams - scFLOW dynamic prediction of wing flap deployment
Membrane deformed by flow aerodynamics (above top) and a vibrating supersonic plate shock
structure co-simulation in MSC Nastran – scFLOW CFD (above bottom)

Figure 13. MSC Nastran - scFLOW prediction of propeller with free

surface and cavitation

Figure 16. scSTREAM thermo-mechanical stress prediction of electronic

component solder stresses

References
1. “Back to the Future; Trends in Commercial CFD” by J. Parry
& K. Hanna, NAFEMS World Congress, Boston, USA, 2011:
http://s3.mentor.com/public_documents/whitepaper/
resources/mentorpaper_90428.pdf

2. “Noise Prediction for Electric Powertrain” Webinar,

by R. Baudson and Y. Fan, August 2018: https://www.
mscsoftware.com/events_assets/Webcasts/2018_
Webinars/noise-prediction-for-electric-vehicle powertrain.
html

3. “An Efficient Fluid Structural Workflow using the Seamless

Coupling Technology” Webinar, August 2018: http://www.
mscsoftware.com/events_assets/Webcasts/2018_
Webinars/an-efficient-fluidstructuralworkflow-using-the-
seamless couplingtechnologywithin-msc-nastran-marc-
and-sc-tetra.html

4. “Predicting Durability of Junctions for Mounted Parts”

Cradle CFD Case Study: https://www.cradle-cfd.com/
images/applications/electronics/11ST_MountedParts
Durability_email

Figure 14. Adams – scSTREAM prediction of a dynamic vehicle

suspension movement through a puddle

25
Husqvarna Group - Leveraging
MSC Nastran embedded
fatigue significantly increases
result precision
By Marcus Fälth, CAE Technology Specialist, Research and Development Department,
Husqvarna AB

Figure 1. Husqvarna Handheld Power Cutter

Durability analysis, including material fatigue life prediction,
is a very important part of Computer Aided Engineering. It
provides valuable information about the product service
life, which highly influences customer experience and
satisfaction. Husqvarna Group, global leading producer of
outdoor power products, count on MSC Software’s Nastran
Embedded Fatigue (NEF) as the heart of an optimized,
efficient process for fatigue analysis.
Fatigue in any part of the product is caused by vibrations
from the two-stroke engine, which drives for example a
chainsaw or a handheld power cutter (figure 1). A clear
understanding of this phenomena is needed before going
into production. That is why virtual prototyping is used to
assess where the material could fail and to strengthen
the part structural integrity accordingly. For virtual
prototyping, a system model of the device is created
consisting of the engine, housing and surrounding parts
like fuel tank and covers. Boundary conditions represent
the handle and how the force is transferred to the
operator’s hands.
This article focuses on showing how we are able to
count on MSC Software’s Nastran Embedded Fatigue as
the heart of an optimized, efficient process for fatigue
analysis. Optimized Nastran Embedded Fatigue process
created considerably smaller amount of data and hence
the large intermediate results files, with their tedious re-
import, are no longer needed.

26
Figure 2. Traditional Fatigue Analysis Workflow
Traditional Fatigue Analysis Process
Structural vibration analysis with MSC Nastran has a long
tradition at Husqvarna. It has been constantly improved,
for example by increasing the complexity of the models.
But fatigue analysis initially posed some difficulties for the
engineers.
Conventional strength and fatigue analysis consists
ofseveral steps:
1. Finite Element model definition
2. Structural Vibration analysis
3. Analysis output which may contain large results files
4. MSC Nastran Results file import into a separate
fatigue software, which is very time-consuming
5. Many read and write operations that can lead to
memory problems
6. Postprocessing of fatigue life results
27
This is a tedious process which does not easily allow
design optimizations. We were only able to optimize short
time intervals, more precisely, just one engine revolution.
The limitation is caused by the extraction of the stress
data, which means large results files and therefore long
run times.
The MSC Nastran – Adams process yields the
modaldeformations, stress time histories and stress
peaks. Actually, we wanted to analyze a period of 1 second,
but could only manage one stroke. We had to reduce the
number of output steps to 40 and already reached 40
GB of results file size, giving the example of a crankcase
analysis. Moreover, the resolution was not high enough to
ensure to hit the maximum of the stress history.
Analyzing longer periods of time is important to assure
that a specific system mode of interest is excited by any
multiple of the engine revolution speed. This governs a
more robust analysis regarding modeling uncertainties.
The workflow had already been significantly accelerated
by the close coupling of the motion solver (Adams) and
the Finite Element Analysis tool (MSC Nastran). Using NEF,
Husqvarna was able to enhance the process even further.
At the beginning of the systematization of the NEF process
at Husqvarna, first there was no clearly defined workflow
for durability analysis. This changed when MSC‘s Nastran
Embedded Fatigue became available. After a year of
intensive evaluation, we decided to use NEF in productive
operation for vibration analysis.
Nastran Embedded Fatigue – Innovation
and Disruption
The difference between Nastran Embedded Fatigue (NEF)
and the traditional process is that NEF does not need to
export and import large binary results files, as the fatigue
analysis is done directly in MSC Nastran - a considerable
gain efficiency. (Figure 3)
To illustrate that, let’s look again at the example of the
above-mentioned crank case: today it consists of 250,000
shell elements, 200 strokes, 30,000 output steps and a
universal output results file of just 50 MB size. In contrast
to the traditional process, which had 250,000 shell
elements, only 1 stroke, and 40 output steps, but a 40 GB
output results file. A longer time, for example 200 strokes
– about 1 second – can now be analyzed with ease, leading
to a significant increase in result precision.
Figure 1. Husqvarna Handheld Power Cutter
28
Fatigue Material Data and Reality
Of course we know well that our modelling is not perfect.
Imprecisions must be taken into account. For example,
the analysis assumes a uniform temperature, though
there may be a significant temperature distribution in a
component. Another simplification is the modelling of
attachments. Thus, the calculated Eigen frequencies may
deviate from the true values. Also, some assumptions
are made concerning the engine revolution speed. On the
other hand, this gets easier because a longer time can be
simulated now, which helps to even out fluctuations.
The fatigue life result is improved by calibrating the input
fatigue material data (the S-N curve) to real life testing. We
modelled a physical fatigue test and the stress range was
transformed via the Goodman relation for adjustments of
the S-N curve. The Goodman relation (Haigh diagram) is
used to quantify the interaction of mean and alternating
stresses on the fatigue life.
What was Achieved?
The optimized Nastran Embedded Fatigue process
creates a considerably smaller amount of data. The large
intermediate results files, with their tedious re-import, are
no longer needed.
“We can analyze the whole load cycle, not only the
immediate surrounding of load peaks,” Fälth says. The
process has proven to be robust to changes in the
excitation. Data transfer takes much less time, because
there are only MB of data instead of GB. NEF has useful
functionality to automatically generate S-N curves from a
smaller amount of input data.
First there was no clearly defined
workflow for durability analysis.
This changed when MSC‘s Nastran
Embedded Fatigue became
available. After a year of intensive
evaluation, we decided to use
NEF in productive operation for
vibration analysis.”
Marcus Fälth
 + =
Figure 4. Nastran Embedded Fatigue Process: Structural Analysis Results + S-N Data Plot = Fatigue outputs such as stress, Damage, Life

Also, the system model is now created by a black-box

approach. Adams runs in a scripted environment. This
means a democratization of analysis and simulation: Far
simulation engineers than before can do the high-end
durability analysis, as they don’t need to understand the
exact details of the process. The geometry of Husqvarna
products can be optimized in very efficient iteration cycles.

When cutting through hard materials like concrete, the

professional power cutter experiences unsteady exterior
forces. The influence of this behavior on durability and
potential fracture of the power cutter can be evaluated
in much more detail with the help of NEF. The analysis is
comprised of 250 strokes (2 s), 140 force pulses acting on
the structure, and 60 000 stress output steps.

About Husqvarna

Husqvarna Group is a global leading producer of outdoor

power products for forest, park and garden care. Products
include chainsaws, trimmers, and robotic lawn mowers
and ride-on lawn mowers.

The Group also produces machines for the construction

and stone industries. The Group’s products and solutions
are sold under brands including Husqvarna, Gardena,
McCulloch, Poulan Pro, Weed Eater, Flymo, Jonsered,
Diamant Boart, and RedMax.

29
Efficient rotordynamic analysis
using the superelement approach
for an aircraft engine in
MSC Nastran
By Yves Fournier, Pratt & Whitney Canada Corp., Canada

In response to an increasing demand for rapid prototyping,
aerospace and automotive industry leaders are taking
advantage of new simulation techniques to deliver
high quality products that are safe, reliable, and cost
efficient. Simulation plays an important role in product
development, helping engineers make early design
changes and avoid any negative surprises down the road.
This is especially true for design and analysis of many
aerospace and automotive structures, including those
that consist of rotating components i.e. jet engines, turbo
machinery, centrifuges etc. Exploration of new aero-
engine configurations drives unseen and complex dynamic
behavior which requires advanced simulation techniques
that provide high levels of accuracy. As the original and
most trusted FEA solver that has been widely used by
major aerospace companies, MSC Nastran provides highly
accurate and reliable Rotordynamic solutions to examine
the behavior of rotating machinery.

30
 Exploration of new aero-engine
configurations drives unseen and
complex dynamic behavior which
requires advanced simulation
techniques that provide high
levels of accuracy.”

Due to complexity of the parts, the modeling approach

chosen for rotating components depends on various
variables such as the computational resources available,
required accuracy, and the type of elements used for
the rotor geometry. Traditional methods (reference 1)
are based on a simplified representation of the model
through use of 1D beam elements for shafts and point
masses for disks (see Figure 1). While limited in accuracy,
capturing the effects of large flexible disks and the
complex geometry of rotors while properly representing
the models, they are still common in the industry today.

A better approach that has recently been used by Pratt

and Whitney is to analyze large engine models with high-
fidelity two-dimensional (2D) axisymmetric harmonic and
three-dimensional (3D) shell and solid elements. In spite
of staggering advances in computation, however, the
so-called FE rotor models are still considered to be quite
large. For example, when transitioning from the traditional
1D to the fully 3D rotor model, model size can increase
by three orders of magnitude. Even with today’s modern
computational systems, the amount of time required to
solve linear Rotordynamic models can stretch to hours and
days (see Table). Furthermore, a detailed Rotordynamic
study generally covers a series of analyses to determine
stability (complex eigenvalues), quantify unbalance loads
(frequency response), and also predict the performance of
bearings / dampers (nonlinear transient response)modes.
Each superelement can be viewed as a standalone model
and can be processed independently.

31
Figure 1. The four types of Rotordynamic structural modeling
approaches being used today
Figure 2. Explanation of Superelements in the context of an Aircraft
Assembly
Taking advantage of the SE technology in MSC Nastran
leads to reduced computational cost, the ability to
solve very large problems, the protection of intellectual
property, and the enablement of a modular model
description that promotes parallel processing (through
utilization of High Performance Computing, HPC).
In this project (Reference 2), engineers at Pratt & Whitney
Canada and MSC Software worked as a team to analyze a
realistic aircraft engine in which its rotating structures are
modeled with high-fidelity 3D solid/shell elements. This was
a simplified PT6 engine model used for FEA Verification &
Validation (see Figure 3). A Rotor is attached to an engine
casing through bearings at B1*, B2* and B3* and includes
two sets of turbines located at T1 and T2. For this analysis,
linear stiffness and viscous damping coefficients were
prescribed for these three bearings. Turbine Blades were
not modeled in the analysis and the blade assembly was
treated as a point mass. While the rotor model itself was
symmetric, the casing’s assembly which is connected to
ground at the engine mount locations C1 and C2, was not
symmetric. The dynamics of the engine assembly were
solved using modal frequency response solutions inside
MSC Nastran. The ultimate goal of this project was to
reduce wall time for the simulation and improve efficiency
while maintaining a high level of accuracy. Both rotor and
stator assemblies were reduced adequately and their
representation error was controlled by the number of
dynamic modes specified in the CMS method.
32
Figure 3. PT-6 Engine, 3D CAD model, Rotor FEA Model and Engine
Casing Model in MSC Nastran
In this project, the rotor and rotor support were modeled
and analyzed individually as a straightforward assembly
and as a whole engine system for result validation
purposes. We ensured that the bearing node pairs B1R-
B1C, B2R-B2C and B3R-B3C coincide. The complete engine
model of the Direct approach was reduced to a number of
real modes using Real eigenvalue analysis. These modes
were then used for the subsequent rotordynamic analysis
that included damping and skew-symmetric rotor speed
dependent terms. External SE of the Rotor and Casing
Model were analyzed separately and reduced to their
physical boundary points and dynamic modes. Assembly
of the separate external Rotor and Casings SEs created
earlier were analyzed. In all cases, the error was less than
expected (0.1% baseline error of the External SEs). This
approach allowed the team to better understand the
individual behavior of components as well as the coupling
effects of the whole Rotordynamic system.
The MSC Nastran model for the complete engine
contained a 3D rotor with 91,979 Degrees of Freedom. The
time taken to generate the SEs for this rotor model with
100 modes reduced by 3 to 5 orders of magnitude to 1.5
minutes when using MSC Nastran 2017 on a Linux Machine
with 4 GB Ram. We also used a 20 node, 4 core HPC Cluster
with 120 GB RAM for the Casing simulations. In comparison
to traditional 1D elements, 3D elements increase the
model size by three to four orders of magnitude. Solving
such large 3D rotor models is not conducive to easy
design and parametric studies, especially when it comes
to performing nonlinear simulations. As an alternative,
the External SE approach utilizing CMS method has been
shown to be an effective way to improve performance
without loss of accuracy. Once the External SE for a
component is created, it can be reused in different
solutions with very little computational overhead. A Modal
solution with 100 modes produced the best set of results
for this case (see Figure 5).
 Figure 4. PT-6 Engine results in MSC Nastran for Bending Motions

Summary and Conclusions

In comparison to traditional 1D elements, 3D elements

Figure 5. PT-6 Engine results in MSC Nastran for Bending Motions
increase model size by three to four orders of magnitude
in aerospace rotordynamics MSC Nastran simulations.
Large computational time associated with 3D models are
not conducive to design and parametric studies or for
nonlinear analysis. An external SE approach utilizing the
CMS method has been shown to be an effective way to
improve performance without loss of accuracy for a PT-6
engine. Once the External SE for a component is created,
it can be reused in different solutions with very little
computational overhead. Simulation results revealed that
the baseline error of the External SEs amounts to less
than 0.1% with associated wall clock time reductions of 3
to 4 orders of magnitude.

References
The time taken to generate the Super
“Rapid Reconfiguration of Engines for Dynamics Simulation,”
Elements for this rotor model with Kumar, D., Juethner, K., and Fournier, Y., SAE Technical Paper
100 modes reduced by 3 to 5 orders of 2016-01-2017, 2016, doi:10.4271/2016-01-2017

magnitude. “Efficient Rotordynamic Analysis using the Superelement

Approach for an Aircraft Engine”, Y. Fournier, D. Kumar & K.
Juethner, Proceedings of ASME Turbo Expo 2017: Turbomachinery
Technical Conference and Exposition, June 26-30, 2017, Charlotte
NC, USA

33
MSC Nastran based fatigue
lifecycle predictions at
Navistar-Tech Mahindra
By Chinmay Pawaskar, Principal CAE Analyst, Tech Mahindra; Pravin Kulkarni,
Lead CAE Analyst, Tech Mahindra, and Stefano Cassara,
Manager - Vehicle Dynamics Simulation, Navistar

Navistar International Corporation manufactures and

sells commercial and military trucks, diesel engines, and
school and commercial buses; and provides service parts
for trucks and diesel engines worldwide. Tech Mahindra
offers engineering solutions in design & styling, product
development, validation, testing and system integration to
global automotive players. At the Navistar- Tech Mahindra
Global Engineering Center, the team applies computer-
aided engineering techniques to the truck product
development process.

With an increasing focus on the safe carriage of payload

& light weighting, the ability to predict the fatigue life
of vehicle components has gained a lot of importance.
Predicting the fatigue life accurately in the early stages
of design and development cycle enhances product life,
reduces testing and prototype costs and ensures greater
speed to market.

Physical tests are used to assess the fatigue life of a

vehicle (see Figure 1). This is done using laboratory shaker
tests or fullvehicle testing on a proving ground. These
tests are costly and time consuming, so to avoid an
iterative design-test-redesign loop, we need an accurate
virtual testing methodology that can be used to generate
designs that will meet durability requirements that first
time around.
Figure 2a. Development of MBD-based Virtual Prototypes
1. A Virtual Cab Shaker Test Rig where laboratory inputs
areused to drive a virtual cab and cab suspension
model. The accelerations, forces are used to validate
and optimize the cab design.

Our Adams virtual prototypes models have now evolved
to accurately predict ride and handling performance.
Correlation studies on strains also continues to improve.
Figure 2 outlines key models in the evolution of our Multi-
Body Dynamics (MBD) virtual prototypes:
2. A Virtual Test Track in Adams where a hi-fidelity full
vehicle model is driven over realistic rough roads.
Accelerations, force, strains are used to predict the
structural adequacy of the vehicle components. In
the past, we have used two processes to estimate
fatigue life:

Figure 1. The traditional physical process for fatigue life prediction

34
Figure 2b. Development of MBD-based Virtual Prototypes
• A transient stress recovery in Adams followed by cycle
counting in Hyper Graph (see Figure 3)
Figure 3. Virtual Transient Stress Recovery Lifetime prediction process
using Adams
• A Nastran linear-static inertia-relief - This method
uses time-slices of loads corresponding to the
instance of peak-stress identified in the MBD
simulation (see Figure 4)
Figure 4. Virtual Static Inertia Relief Lifetime Prediction Process using
Nastran
Figure 5. Comparison between the current fatigue life prediction methods. Left: Stress contour from Nastran. White spots indicate top hot-spots
from Adams. Right: Nastran (red curve) and Adams (blue curve) for the same component under test
35
Both methods produce comparable results (see Figure
5). Adams accurately predicts the stress for the entire
duration of the event, but the calculation is limited to
the regions of stress recovery. On the other hand, the
Nastran method gives us a full picture of the stress in the
FE model but only at a few time instances. One proviso in
this process is that for dynamic events, this method may
underestimate the stresses by ~1.2 times for the frame rail
and ~2.5 times for the chassis mounted components.
We wanted to have a process that uses the entire time-
history of loads, calculates the fatigue life, and can display
results of the entire component. For this, we turned to
MSC’s new offering – Nastran Embedded Fatigue (NEF).
This would entail exporting the entire Adams time history
of force and moments acting on a component and feed
it into NEF (see Figure 6). In NEF, we would add fatigue
specific cards into the Nastran bulk data file and reference
the loads calculated by Adams. With this information, we
can predict where damage occurs as well as the lifetime of
the components.
Results for the Adams-NEF process are typically
presented using two metrics: Contours of Damage and
Log Life. Figure 8 shows these metrics plotted for the
Cross Member and End Bracket of a Light Truck traversing
a rough road. Contours of Damage represent the total
damage to the structure. Damage ranges from 0 to 1.0.
When D reaches 1.0, failure occurs. Cases in which the
stress range is greater.
Nastran Embedded Fatigue Procedure
With NEF, we have two methodologies available to
us:A Quasi-Static Approach (SOL 101) and a Modal
Superposition Approach (SOL 103 & SOL 112) shown
pictorially in Figure 7. Both SOL 103 (modal analysis) and
SOL 112 (modal transient analysis) use identical methods
to compute fatigue life. For SOL 103, the modal responses
are provided and associated to each node.
Then the cyclically varying stress responses are obtained
through a linear superposition. SOL 112 computes the
modalresponses automatically. Loads are exported from
Adams via DAC files, a main script block of the NASTRAN
deck is set up with an appropriate calculation flow. The
NEF code is then run to produce output NEF files. This
run will produce an OP2, F06, F04, and log files (for large
models, the output may spill over into several files). All
files can be viewed in Hyper View. than the YS or UTS, D
may be greater than 1.0. The log file contains the number
of cyclic repeats the structure can sustain before failure.
The life of the part is inversely proportional to the damage,
reported in log terms. In this case, the chassis was excited
at 52 attachment points with 3 DOFs per attachment point
in a free-free inertia relief analysis.
Figure 10 shows the results for damage for two designs: an
old and an optimized cross member & end bracket.
In another case study, we studied the trade-off between
ride clearance and the frame’s structural durability. We
used Adams to compute the loads acting on a chassis as

Figure 6. Virtual Lifetime Prediction Process using Nastran Embedded

Fatigue

Figure 8. NEF Fatigue Model predictions of damage contours (left) and

life prediction (right) to a cross member and end bracket assembly

Figure 9. NEF Fatigue Model predictions of contours of damage to an

old (left) and new (right) cross member and end bracket assembly after
testing

Figure 7. NEF Fatigue Model SOL 101 and SOL 103/112 approaches

36
the vehicle traversed the virtual test track. The loads i.e.
forces and torques acting in 6 directions at 55 different
Figure 10: NEF Fatigue Model predictions of contours of damage in a
Frame Rail analysis
locations were used to compute a NEF-based fatigue
analysis of the chassis. In this exercise, we used the SOL
101 Quasi-Static Linear Superposition approach. Figure 11
shows the contours of Damage for various ride clearances.
The turn-around time for this study was 1 day.
Finally, we compared the NEF-based virtual life prediction
to our existing virtual life prediction. As shown in Figure
12, we extract the von Mises elemental stress from NEF
(in blue squares) and compare them to the transient von
Mises nodal stress prediction from Adams (in red circles).
Figure 11. NEF Fatigue Model predictions of absolutevon Mises Stresses
in Adams and NEF for a transmission cross member
In Figure 12, we see that results from NEF (blue line)
approaches the results from the Adams simulation (red
line). Thus, NEF provides the same accuracy as that of
Adams. More importantly, it provides a complete picture
of the structural performance of the chassis.
37
Figure 13. NEF Fatigue Model predictions of von Mises Stresses in Adams
and NEF over time for a transmission cross member
In Summary
The Virtual Fatigue Life Prediction process at the Navistar-
Tech Mahindra Engineer Centre has been outlined
above. The NEF process has enabled identification of
high stress locations more effectively. We found the
overall process to be nearly three times faster than the
previous method. With fewer and smaller intermediate
files, the bookkeeping is also much simpler. The load time
history (DAC/RPC files) can be used directly without any
conversion. In addition, fatigue can now be included as a
parameter in an optimization study. We have successfully
validated this methodology based on several parameters
and it has exhibited a good level of correlation with our
physical test data. The Navistar-Tech Mahindra team see
several opportunities in running similar simulations in the
near future. MSC’s new offering has made a significant
difference in simplifying and speeding up our fatigue
lifecycle predictions.
Tips & tricks HDF5: A very useful
enhancement for MSC Nastran
and Patran
Let’s face it, as FEA engineers we’ve all been in that
situation where the finite element model has been
prepared, boundary conditions have been checked and
the model has been solved, but then it comes time to
investigate the results. It’s not unusual on a large project
to produce different outputs for different purposes:
an XDB or OP2 for Patran, a Punch file to pass to Excel,
and the F06 to read the printed output. In addition, your
company may have programs that read one of these files
to perform special purpose calculations. Managing all
these files becomes a project of its own.
Well let me tell you, there is a better way to do this!! As a
premier FEA solver that is widely used by the Aerospace and
Automotive industries, MSC Nastran now takes advantage
of the HDF5 file to manage and store your FEA data. This
can simplify your post processing tasks and make data
management much easier and simpler (see below).
So what is HDF5? It is an open source file format, data
model, and library developed by the HDF Group.
38
HDF5 has three significant advantages compared to
previous result file formats: 1) The HDF5 file is smaller than
XDB and OP2, 2) accessing results is significantly faster
with HDF5, and 3) the input and output datablocks are
stored in a single, high precision file. With input and output
in the same file, you eliminate confusion and potential
errors keeping the input BDF with its associated output.
In addition, you can minimize output to the F06 file, so it
will be smaller as you only need warnings and errors. This
means you gain better traceability in comparison to using
xdb and op2. It’s as simple as it gets!
HDF5 is designed for storing and managing large and
complex data. It can be used in any field: biology, physics,
and of course engineering. Note, the HD in HDF does not
stand for high definition! it stands for ‘hierarchical data
format’. The hierarchical format is designed to manage
all sorts of data objects including tabular data, graphs,
images, and documents.
Implementing HDF5 in MSC Nastran has been an ongoing
project that began with the 2016 release. The latest
 MSC Nastran

MSC Nastran

Before HDF5 After HDF5

release of MSC Nastran and Patran have the most
complete library of output data blocks for HDF5 files.
Some examples include: nonlinear results, optimization
results, fatigue life and damage results and rotordynamics
results. MSC continues to extend HDF5 coverage with
each new release of our software.
One of the nice things about HDF files is that you don’t
need a lot of documentation to know what’s in it … The
HDF5 file is self-describing! It’s easy to figure out what
information is stored; just open and view the hierarchy with
HDFView. In addition, MSC delivers the schema definition
with each release.

The HDF5 file produced by MSC Nastran is a binary file and Which one would you rather use to manage your MSC
the fastest way to access results in Patran. It definitely Nastran output data? Whether you deal with small or large
improves ‘Results Attach’ performance, especially when scaled MSC Nastran models, HDF5 output within MSC
using large models with a million degrees of freedom. Nastran is the ideal choice for analyzing your complex FEA
Although HDF5 is a binary file, you can use the HDFView models, so do take advantage of it.
tool to view your MSC Nastran data in a browser. In
addition, you can copy and paste data from HDFView to
other applications like Excel, or export a data table or a
csv format for plotting purposes.

HDF5 data can be accessed with a variety of programing

languages. The HDF Group provides native bindings for C/
C++, Fortran, and Java. Third party extensions are available
for Python, .Net, and many other languages. MSC provides
HDF5 examples in C++, Python and Java. In addition, for
those writing DMAP Alters, the new DMAP modules for
writing HDF5 files are documented for your use.

HDFVIEW showing Nodal Information from the MSC Nastran Input file.

39
MSC Nastran speeds forward
using NVIDIA graphics
processing units
By Niveditha Krishnamoorthy, Development Relations Manager, NVIDIA

Analysts in today’s environment are working on
increasingly larger and complex models. With advances
in electric car manufacturing and quieter electric
motor engines, high-frequency Noises and Vibrations
are becoming more prominent, which makes these
calculations exponentially more expensive.
As computing technology improves, the industry is leaning
towards analyzing real-world scenarios more accurately
with the use of Finite Element (FE) Analysis to support
these high-fidelity models.
Current trends in High-Performance Computing (HPC)
are moving towards the use of multicore processor
architectures to achieve speedup through the extraction
of a high degree of fine-grained parallelism from the
applications. This hybrid computing trend is led by
NVIDIA Graphics Processing Units (GPUs), which have
been developed exclusively for computational tasks as
massively parallel co-processors to the CPU. Today’s
GPUs can provide memory bandwidth and floating-point
performance that are several factors faster than the
latest CPUs. To exploit this hybrid computing model and

40
the massively parallel GPU architecture, application
software will need to be redesigned. MSC Software and
NVIDIA engineers have been working together on the use
of GPUs to accelerate the sparse direct solvers in MSC
Nastran for the last several years. We will dive deeper
into the recent GPU computing developments in MSC
Nastran in this article, including the support of NVH
solutions with Fast Frequency Response (FASTFR), and
Matrix Multiply and Add (MPYAD) modules development.
Representative industry examples will be presented to
demonstrate the performance speedup resulting from
GPU acceleration.
The resulting rapid CAE (Computer-Aided Engineering)
simulation capability from GPUs has the potential to
transform current practices in engineering analysis and
design optimization procedures, enabling manufacturing
OEMs to deliver their products to market at a much faster
rate than is currently possible with CPUbased solutions.
Working jointly with NVIDIA provides MSC Software
acomprehensive portfolio of enterprise-grade, high-
performance systems, and software, with high-value
services to help manufacturers implement MSC
Nastran throughout the value chain and product
lifecycle. Many of these systems are already certified
by MSC Software and deliver unparalleled productivity,
performance, and flexibility.
GPUs have been a significant part of HPC for decades,
and MSC Nastran has been on the market for more than
50 years. There had been attempts to utilize GPUs in MSC
Nastran in the past, which have not led to much success.
One of the reasons was the architecture of MSC Nastran,
which relied heavily on the disc storage of the data and
having a minimum memory footprint.
Another reason was the memory limitations and the
overhead caused by the data transfer between the
host and the GPUs. In the last few years, GPUs have
accelerated performance – NVIDIA has added more
Leveraging the GPUs for dynamic analysis using MSC Nastran 2019 at Volvo
opens for substantial savings in run time (up to 50% faster) for the runs
extending to high frequency.
Andrzej Pietrzyk,
Method Development for NVH CAE at Volvo Car Corporation
41
cores and memory specifically for the FE applications.
This has helped specifically with NVH simulations. To
accommodate the GPU usage, MSC Nastran has evolved
to maximize memory usage, and the GPUs have become
increasingly faster, and their onboard memory has
reached the tens of gigabytes mark, they are finally ready
for successful symbiosis.
Why Use GPU?
Most customers ask why they should buy a GPU. It is
generally beneficial when models are mainly composed
of solid elements, and the simulation is dominated by
matrix factorizations, not like in Normal Modes Analysis
(SOL 103), where there are lots of modes, and FBS
dominates. Furthermore, each user should evaluate
whether multiple CPU cores can give them the same
benefit as a GPU while keeping in mind that GPU usage
can free up some of the CPU cores for use by other
applications/users.
A primary use of GPU is to offload massive computational
tasks from CPU. GPUs will not be useful for small models
that operate on a small amount of data, do a large
number of I/O operations, or iteratively solve small sets
of equations. Every time a GPU is invoked, it comes at a
fixed cost of, typically, a fraction of a second, 0.1 – 1 sec.
So, if you are trying to use a GPU to compute a product
of two 100 x 100 matrices, which takes 0.01 seconds on
the CPU, you will pay a 0.1 – 1-second penalty for just
using the GPU before you even start transferring the
data between the host memory and the GPU. This will not
benefit your simulation.
However, if you are dealing with a large model that
requires operations on matrices with dimensions of the
order of 10000 x 10000 or more, then the GPU will come
in very handy. In fact, for dense matrix operations, every
additional GPU is roughly equivalent to having an extra
CPU with 20 cores. By splitting the workload between the
CPU and one GPU, you can achieve approximately a factor
2x speedup. When splitting the workload between the
CPU and two GPUs – factor 3x speedup, and so forth. For
sparse matrix operations, every GPU is roughly ten times
faster than the CPU, so all sparse BLAS operations should
be handled by GPUs alone.

Similar to the CPUs, GPUs have multiple cores that can

be used for parallel computations, and they also have
parallel “streams” that can perform parallel data transfers
between the memory of the host machine and the GPU
memory. Every time a GPU is invoked in a computation,
all the input matrices need to be copied from the host
memory into the GPU memory, and the resulting matrix
needs to be copied from the GPU back to the host. By
using multiple streams in parallel, the cost of these data
transfers can be significantly reduced. However, it would Figure 2 represents performance improvement for a dense
be best if you considered that each GPU stream could only matrix multiplication benchmark: 50,000 x 50,000 dense
communicate to a dedicated CPU core. So, when you are matrices. Blue bar: 20-core Intel Xeon CPU; red bar: 20-
performing some computations involving dense matrices, core Intel Xeon CPU + two NVIDIA Tesla K40 GPU; green
precise balance must be reached between how many bar: 20-core Intel Xeon CPU + NVIDIA Tesla K40 + Quadro
CPU cores are included in calculations, and how many are GV100 GPU; purple bar: 20-core Intel Xeon CPU + two
dedicated to communicating with the GPU streams. Nvidia Tesla V100 GPU.

Rule of the thumb is: the number of CPU threads/cores
should always be larger than the number of GPUs you are
trying to use. For example, in a scenario where you try to
compute something using two GPUs and just four CPU
cores, the workload will be divided between the GPUs
because the number of the CPU cores is too small to
compete with the GPUs. In a scenario where you try to use
two GPUs and 20 cores, four cores will be assigned to each
GPU for parallel streaming of data, and the remaining 12
cores will be used for the computation.
MSC Nastran GPU acceleration is user-transparent,
meaning jobs launch and complete without additional
user steps. Also, the schematic of a CPU with an attached
GPU accelerator is such that CPU begins/ends job, GPU
manages heavy computations. This gives unparalleled
productivity and performance with NVIDIA GPUs.
It is possible to use several different GPU models on the
same host system, with varying sizes of memory and
efficiencies. Every time Nastran uses GPUs for BLAS
operations, it will check the elapsed times on each GPU
and CPU and balance the workload in a way that would
minimize the elapsed time across all devices. This way, the
fastest GPU/CPU gets most workload, while the slowest
GPU/CPU gets the least workload.
The current implementation of GPU support is using CUDA
Toolkit by NVIDIA and only works with NVIDIA GPU devices.
The threshold matrix size is set to 4000 for dense matrices
and 500 for sparse matrices. Matrices are partitioned
among the CPU and GPUs and, depending on the memory
availability, may be processed in one or multiple passes on
each device.

Examples

When you deal with sparse matrices and sparse BLAS

operations, the work partitioning logic is simpler, since
all work will be handled by GPUs only. In case you are
using one GPU, a typical large sparse BLAS operation
will be about 3x – 10x times faster. Since most of these
operations are scalable, the more GPUs you use, the
greater the speedup. With two GPUs, you get 6x – 20x
speedup, and so on. Specific speedup values are strongly
dependent on the hardware, on the size of the input
matrices, and the type of operations.

Figure 1 represents an example of performance improvement

for densesparse matrix multiplication by adding one or two
GPUs. Input matrix parameters: matrix A 56,000 x 192,000, Figure 1
100% dense; matrix B 92,000 x 192,000, 1.5% dense. Blue bar:
20-core Intel Xeon CPU; red bar: single NVIDIA Tesla K40 GPU;
green bar: two NVIDIA Tesla K40 GPUs.

42
Below, we shall consider several examples from the
automotive industry to highlight the usage of GPU
accelerate the solution. All models are SOL111 with varying
number of degrees-of-freedom (DOF), frequency range,
number of eigenmodes, and load cases. Note how the
performance improves with an increasing number of
frequencies and eigenmodes. The reason for this is that
Figure 5. Performance improvement for model from
Figure.4 (1500 frequencies, 50,000 eigenmodes) by adding
one or two GPUs. Blue bar: 20-core Intel Xeon CPU; red
bar: 20-core Intel Xeon CPU + NVIDIA Tesla K40 GPU; green
bar: 20-core Intel Xeon CPU + two NVIDIA Tesla K40 GPUs.

Figure 2 Figure 3

most of the computationally intensive operations are
proportional to the number of eigenmodes to the third
power and the number of frequencies. This is where the
GPU makes the impact. In all the simulations shown below,
the GPUs were actually used only in a couple of modules.
Yet they managed to cut down the entire simulation
elapsed times by 15-35%, or between 0.4 – 4 hours.
Figure 6. Performance improvement for model from
Figure.4 (2000 frequencies, 73,600 eigenmodes) by adding
two GPUs of different types. Blue bar: 20-core Intel Xeon
CPU; red bar: 20-core Intel Xeon CPU + two NVIDIA Tesla
K40 GPU; green bar: 20-core Intel Xeon CPU + two NVIDIA
Tesla V100 GPU.

Figure 3 shows an example of SOL111 performance

improvement by adding a single GPU. The model: 20.5M
DOF, 2050 frequencies, 30,000 eigenmodes, 4 righthand
sides. Blue bar: 20-core Intel Xeon CPU; red bar: 20-core
Intel Xeon CPU + NVIDIA Tesla K40 GPU.

Figure 4 represents the car-body model with 43.5M DOF

provided by Volvo Car Corporation used for benchmarking.
Andrzej Pietrzyk, Method Development for NVH CAE at
Volvo Car Corporation, says, “Leveraging the GPUs for
dynamic analysis using MSC Nastran 2019 at Volvo opens
for substantial savings in run time (up to 50% faster) for
the runs extending to high frequency. This is definitely a
highly promising development path for MSC Nastran.” Figure 4

43
 Key Takeaways

GPUs allow speeding up large computationally intensive

MSC Nastran operations by factors between 2 and 20,
compared to the elapsed time on CPU alone. Depending
on the percentage of these operations in the total
elapsed time of the entire simulation, the results
obtained can be several times faster than a simulation
that is run only on CPUs.

The larger the simulation, the larger the speedup. For the
same model, the frequency response analysis over the
frequency range of 1500 Hz and 50,000 modes improves
from 11.3 hours to 7.2 hours by adding 2 GPUs. With the
frequency range increased to 2000 Hz and 73,000 modes,
the elapsed times on CPU alone and with two Tesla K40
GPUS are 30.3 and 18.1 hours, respectively.

Newer GPU models are several times faster than older

models. For dense BLAS operations 5-year old Tesla
K40 model is roughly equivalent to adding an extra CPU
Figure 5 to the host machine. Adding a newer Quadro GV100 or
Tesla V100 model is approximately equal to adding CPU
that is 2-3 times faster. It is possible to split the workload
dynamically between different GPU models and get the
best performance out of all available devices. For sparse
BLAS operations adding a Tesla K40 GPU leads to a
speedup of 10x on average. Adding a Tesla V100 leads to a
speedup of ~20x– 30x compared to the elapsed times for
the same operations handled by Intel MKL kernels on CPU.

Acknowlegdgement

We would like to thank Illia Silin who contributed greatly

towards this article.

Figure 6

44
Author profiles
Al Robertson
Product Manager, MSC Nastran, MSC Software

Al Robertson is the product manager for MSC Nastran at MSC Software, and is based
in Newport Beach, California. Al has over 30 years of experience within the simulation
software development industry in various roles including product marketing, product
management, training and support as well as several years within the aerospace industry as
a CAE specialist and airworthiness engineer.

Hugues Jeancolas
Vice President, Product Management

With over 15 years working in Strategy, Product Management, and Product Marketing roles
in North America and Europe, Mr. Jeancolas brings extensive domain knowledge of Virtual
Product Prototyping solutions, Product Life-cycle Management (PLM), Computer Aided
Design (CAD), and Computer Aided Engineering (CAE). Mr. Jeancolas received a Mechanical
Engineering Diploma from the University of Technology of Compiegne, France in 2000 and
a Master of Science in Mechanical Engineering from the University of Houston in 2001. He
started his career at HKS, Inc renamed Abaqus, Inc, later acquired by Dassault Systemes and
joined MSC Software in 2012. As the Vice President for Product Management, he is responsible
for MSC’s product portfolio and the company’s strategic product planning process.

Sarah Palfreyman
Product Marketing Manager, Structures Center of Excellence

Sarah Palfreyman is the Product Marketing Manager for MSC Software’s Structures Center
of Excellence covering MSC Nastran, Patran, and Marc. With a M.S. in Computational and
Mathematical Engineering from Stanford University she has over 10+ years of experience
leading Product Marketing, Product Management and Business Development activities at
major CAE companies. Sarah and her family live in Colorado where she enjoys balancing
technical pursuits with the great outdoors.

Dr. Keith Hanna

Vice President, Marketing

Dr. Keith Hanna is the Vice President, Marketing of MSC Software. Dr. Hanna brings over
25 years of experience in the CFD, CAE, EDA and PLM industries, spanning a wide range of
global technical and marketing roles inside Siemens PLM, Mentor Graphics Corp., ANSYS
Inc. and Fluent Inc. His career prior to engineering simulation included practical experience
of the metallurgical and mining industry at Br. Steel and De Beers. He has both BSc and
PhD engineering degrees from the University of Birmingham in England and is a respected
commentator on the CFD/CAE industry, a pioneer of CFD in sport, and a former member of
the Executive Committee of the International Sports Engineering Association.

45
Hexagon is a global leader in sensor, software and autonomous solutions. We are putting data to work to boost efficiency,
productivity, and quality across industrial, manufacturing, infrastructure, safety, and mobility applications.
Our technologies are shaping urban and production ecosystems to become increasingly connected and autonomous –
ensuring a scalable, sustainable future.
MSC Software, part of Hexagon’s Manufacturing Intelligence division, is one of the ten original software companies and
a global leader in helping product manufacturers to advance their engineering methods with simulation software and services.
Learn more at mscsoftware.com. Hexagon’s Manufacturing Intelligence division provides solutions that utilise data from design
and engineering, production and metrology to make manufacturing smarter.
Learn more about Hexagon (Nasdaq Stockholm: HEXA B) at hexagon.com and follow us @HexagonAB.

© 2021 Hexagon AB and/or its subsidiaries and affiliates. All rights reserved.