benchmark january 2014b
TRANSCRIPT
BENCHMARK
Bubbling UnderModel Validation for Industry
January 2014 issue . . .
Improving the Simulation of Bird Strike on Plastic Windshields
Reverse Engineering Made Simple
Icons of CFD - Prof. David Gosman
To Infinity & Beyond?
�THE INTERNATIONAL MAGAZINE FOR ENGINEERING DESIGNERS & ANALYSTS FROM NAFEMS
from your
editorDavid [email protected]@benchtweet
It’s 2014, and we’re one year closer to the next. As ever, I dohope you managed to have some time to yourself over theholiday period, and that any over-indulgence is now a distantmemory. It’s going to be a massive year for NAFEMS onceagain, with our regional conference programme taking shape,
as well as preparation for the announcement of our 2015 WorldCongress, and we’ve hit the ground running once again with acomprehensive and wide-reaching seminar and training programmefor the first part of the year. If you’ve so far resisted the temptation toengage with our events and training courses, 2014 is the ideal yearto rectify that. I’m certain you won’t be disappointed.
That being said, it’s often commented that the only thing that iscertain is life is that life itself is uncertain. Well, aside from deathand taxes, there’s a lot to be gained from that statement. I oftenthink that everything we do in the analysis and simulation world isfocussed around reducing the level of uncertainty about how thingswill behave to a more tolerable level. Or at least being a little moreinformed about what will happen when the unforeseen occurs. Untilwe know everything there is to know about everything that exists anddoesn’t exist, there will always be space for the uncertain to deraileven the best-laid plans. And this is where that fun feeling of “buthow do we know what is unknown” gives you that dizzy feeling –much like the feeling you get when trying to comprehend the size ofthe universe.
In this issue, we look (as ever, some may say) at varying levels ofuncertainty and unknowns. Yes, we can be certain that every year acertain number of pigeons will be inexplicably unable to see that 747looming into their airspace and end-up almost embedded in thewindshield, but how can we make sure the resulting damage to theglass isn’t so bad that the pilot and passengers are exposed to adangerous situation mid-flight? We can also be certain that bubbleswill form when certain gases are introduced to certain liquids, buthow can we be sure that they form in the most effective way when weneed them to? And in this ever-expanding landscape of seeminglyinfinite computing power and storage capacity, do we need toredefine our concept of infinite, or do we just need to stop using theterm incorrectly?
These are all questions which we can’t answer with completeaccuracy (well, most of them), but which are covered in this issue ofbenchmark. I suppose when we can answer everything definitively,then we’ll all be out of a job. Or will we? How will we know whenthere are no questions left?
I may go for a lie-down in a darkened room now, but in themeantime, I hope you enjoy this issue.
Editorial
EditorDavid Quinn
Deputy EditorNicola McLeish
Design/Productiond2 print
AdvertisingPaul Steward
SubscriptionsChristine Bell
MembershipFor information on membership of NAFEMS, contact Paul Steward on
ISSN 0951 6859
Publisher
NAFEMSBeckford Business Centre
Beckford StreetHamilton, Lanarkshire
ML3 0BTUK
t +44(0)1355 225688 f +44(0)1698 823311 e [email protected]
Errors & Omissions: While every care is takenin compiling benchmark, neither the Editors,nor NAFEMS, can be held responsible for theconsequences of any errors in, or omissionsfrom, the contents. The views expressed by
contributors are their own and all informationis accepted in good faith as being correct at the
time of going to press. The Editors will notaccept any advertisement considered by themto be misleading or otherwise unsuitable for
inclusion in benchmark, however, the presenceof any advertisement should not be consideredto convey or imply any form of commendation
by NAFEMS.
1
Regional Conferences 2014 Call for PresentationsNAFEMS will once again host a series of ground-breakingconferences across the globe. Bringing together theanalysis and simulation community, the 2014 regional conferences provide an open forum to shareexperiences, explore future trends and examine cutting-edge applications.
The 2014 Conferences will increase awareness andprovide a discussion forum for topics that are vital toengineering industrialists and academics, offeringattendees an unrivalled combination of industrialknowledge, expertise and forward-thinking to aid theirdeployment of CAE now and in the future.
The international simulation community are invited toshare their experiences by submitting their presentationfor their local event:
13-14 May 2014 | NORDIC | Göteborg, SwedenAbstract deadline – 7 February20-21 May | Deutschsprachige| Bamberg, GermanyAbstract deadline – 7 February28-30 May | Americas | Colorado Springs, CO, USAAbstract deadline – 29 January4-5 June | France | Paris, FranceAbstract deadline – 3 February 10-11 June | UK | Oxford, UKAbstract deadline – 10 January
Full details of each event, including abstract submissionguidelines can be found at www.nafems.org/2014.
500+ Complete COGAN Industry Needs SurveyThe COGAN Industry Needs survey was launched at theend of last year and more than 500 geotechnicalengineers gave their input. The results of the survey willhelp shape the deliverables of the project and ensurethat they meet the needs of the industry.
Aiming to improve competency in geotechnical analysis,the COGAN project follows on from the exceptional workdone during the CCOPPS and EASIT2 projects,transferring the innovative outputs to the geotechnicalsector of the construction industry. The project willachieve this by preparing a framework for defining andrecording numerical analysis skills in geotechnicalengineering as well as E-Learning modules in keytechnical areas.
NAFEMS is the co-ordinator for this EU-funded Leonardoda Vinci Transfer of Innovation project which will rununtil September 2015.
Keep up to date on the project atwww.cogan.eu.com.
Launch of Corporate & Flexipass SolutionsNAFEMS has expanded its popular e-learning trainingproduct to include customised corporate options, as wellas multiple-seat purchasing options for corporations andindividuals.
NAFEMS e-learning has been providing world-classtraining in FEA, CFD and related technologies in a virtuallearning environment for the past five years. Over 3,000individuals from around the globe have been trained inthat time, from all ability levels across every industry.
The e-learning course programme has developed into anoffering of numerous stand-alone topics, coveringsubjects as diverse as“Basic FE Analysis” through“Essentials of Fluid Mechanics” to“Advanced Dynamic FEAnalysis", “CFD for Structural Engineers”, “SimulationData Management”, and many more.
'Flexipass' introduces the option for both individuals andcompanies to buy multiple 'seats' on NAFEMS e-learning,at a discounted rate. These seats can then be used onany e-learning course for the period of one year.
Corporate e-learning comes in two levels - gold andplatinum. Many corporations have very specific trainingrequirements for their staff, and some may find thattheir needs would be better met by a combination of thestand-alone e-learning topics, into a seamless flowproviding a truly unique training course, specificallydesigned to meet their needs, at their pace, and at theirlevel of experience.
NAFEMS has addressed this need by offering two levelsof corporate e-learning providing organisations with theultimate is customized online training which is tailoredspecifically to customers’ needs.
Full details can be found at www.nafems.org/e-learning
New Publications from NAFEMSA number of new publications have recently beenpublished by NAFEMS. These have been sent out tomembers and are now available on the NAFEMS websitefor members to purchase additional copies and for non-members.
Finite Element Analysis for Engineers - A Primer (Order ref: R0110)
Validating Numerical Modelling in GeotechnicalEngineering (Order Ref: R0111)
The NAFEMS Simulation Capability Survey 2013 (Order ref: R0113)
Generation and Propagation of Sound in Solids andFluids – Modern Analysis Methods in Acoustics (Order ref: R0115)
2
news....
IMPROVING TH OF BIRD STRIK WINDSHIELDS
10
10 IMPROVING THESIMULATION OF BIRDSTRIKE ON PLASTICWINDSHIELDS
26 REVERSE ENGINEERINGMADE SIMPLE
36 TO INFINITY & BEYOND?
26
ReverEnginMade
18
18 MIXING PROCESS BYGAS BUBBLINGAN EXAMPLE OFMODEL VALIDATIONFOR INDUSTRY
32 ICONS OF CFDPROF. DAVIDGOSMAN
Computational Fluid Dynamics is about solving difficultengineering problems, using expensive software,enormous computing resources and highly trainedengineers. If the problems weren't difficult, or veryimportant, then it is doubtful that anyone would devote somuch effort, time, and money at solving them. From theperspective of a modern engineer, it would be easy toassume that this desire to apply simulation technologycomplex problems is a recent concern; that only today arewe able to contemplate solving tough industrial problems,armed with a complex array of multi-physics simulationtools.
This is a misconception. Forty years ago, CFD was born froma desire to solve difficult problems involving turbulence,heat-transfer, and combustion, based on the vision of asmall group of pioneering researchers who were able to seebeyond the meager computing resources available at thetime, and to develop the techniques and methods thatwould ultimately revolutionize engineering.
Prof. David Gosman is one of those pioneers. As a memberof Prof. Spalding’s Imperial College CFD research groupfrom the beginning, he played a pivotal role in developingsimulation methodologies that could cope with the complexgeometries of real industrial problems, many of which areemployed in all commercial CFD codes today. He alsopioneered the use of CFD for combustion in reciprocatingengines and methodologies and software that he developedhave been applied to investigate the design of almost everyautomotive engine designed since the early 1990s.
Prof. Gosman arrived at Imperial College in theautumn of 1962 have recently graduated
from the University of British
Columbia, to study for his PhD under Prof. Brian Spalding.In the early 1960s the focus of the Spalding’s research wasthe development of a 'universal method' for computingturbulent flows, using momentum integral methods for two-dimensional shear flows, and designed to account for freeflows and wall jets. Although these techniques provedmoderately successful for the prediction of “parabolic”boundary layer type flows, they were not applicable to moregeneral "elliptic" type problems (with strong pressuregradients, separation, recirculation and impingement).
Since the solution of "industrial" type problems, especiallythose including combustion, required the solution of elliptictype problems, Prof. Spalding and his team eventuallyabandoned the 2D-parabolic approach in favor of adiscretized “stream-function-vorticity" approach, thatsolved the two-dimensional Navier-Stokes equations (castin terms of stream function and vorticity) using a finite-volume approach and upwind differencing. Although Prof.Gosman's mainly experimental PhD did not directly involvethe development of these methods, he soon becameentangled in their development, to such an extent that thepublication of his thesis was delayed by a number of years.It was this diversion that was to ultimately define his wholecareer.
The culmination of the stream-function-vorticity approachwas the publication of the 1969 book "Heat and MassTransfer in Recirculating Flow”[1], for which Prof. Gosmanwas the editor, and which included the source code for theCFD tool called ANSWER, developed by Runchal andWolfshtein. This book marked a turning point for CFD,demonstrating for the first time that industrially relevant
flow problems could besolved using numericalsimulation, and
Icons CFDProf. David Gosman
32
36
regulars NEWS I WHAT’S ON I CAE GUY I INDUSTRY EVENTS
3
what’s onElements of Turbulence Modeling14 January 2014 2 week e-Learning Course
Successful application of turbulence modeling requiresengineering judgment depending on physics of the flow,accuracy, project requirements, turnaround time, andcomputational resources available. This course offers theattendees the practical knowledge for using turbulencemodeling for complex engineering applications. Through asimple and moderately technical approach, this coursedescribes why we need turbulence modeling and how thesemodels represent turbulent flows. Various approaches andnumber of popular turbulence models will be discussedalong with advantages and disadvantages of these models.
Fatigue & Fracture Mechanics in FEAnalysis15 January 2014 4 Week e-Learning Course
The objective of this course is to break down the fatigueanalysis process into clearly defined steps, giving anoverview of the physics involved and showing how tosuccessfully implement practical solutions using FiniteElement Analysis.
This course is aimed at practicing engineers who wish tolearn more about how to apply finite element techniques tofatigue analysis in the most effective manner. Ideally astudent should have some experience of FEA analysis, butthis is not essential.
Practical Modelling of Joints andConnections16 January 2014 3 Week e-Learning Course
Most structures involve some form of jointing or connection.Traditional fabricated structures have used many thousandsof bolts and rivets to connect components together in acontinuous manner, in the case of ships and aircraft thetotal can run into millions. The engineer is faced with anoften difficult decision when attempting to simulate suchconnections and joints within a Finite Element Analysis. Inmany cases the details of each individual connection can beignored if an overall stiffness or strength assessment is tobe made and the connection is assumed reasonably
continuous. However there may be doubts about the localflexibility and load paths developed with this assumption.The objective of this course is to review the variousconnection and joint technologies in use, give an overview ofthe physics involved and show how to successfullyimplement practical solutions using Finite Element Analysis.
Einfuehrung in die praktischeAnwendung der Finite-Elemente-Methode (FEM)22 January 2014 Training Course |Wiesbaden, Germany
The course, delivered in German, provides practical andsoftware-independent knowledge for the successful andefficient use of the Finite Element Method. Following arefresher of mechanical structure based knowledge that isfundamental for the understanding and expert evaluation ofFE calculations, FE programs are discussed in an easy-to-understand manner. Numerous application-specificexamples from industry are provided.
Verification and Validation Masterclass11 February 2014 Training Course | Frankfurt, Germany
This two-day masterclass provides an in-depth overview of:V&V concepts and methodologies to ensure the necessarylevel of confidence required by virtual simulation; issues tobe addressed and best practices for implementation of V&V,especially with regard to the synergy between physical andvirtual tests and the optimisation of demonstrations ofcompliance; major criteria for V&V planning (technicalissues, benefits, costs, risks etc.) and considerations forsetting up V&V business cases.
This course is perfect for engineers and senior analysts incharge of simulation activities or preparing to take newresponsibility in the management of simulation, especiallywith regard to V&V responsibility, managers in charge ofengineering simulation teams and willing to improve theirknowledge in V&V and in the relevant processes andprogramme managers who need to make critical decisionsbased on engineering simulation results and that wish toincrease their understanding and visibility of the requiredV&V activities.
4 nafems.org/events
CFD for StructuralDesigners and Analysts
12 February 2014 2 Week e-Learning Course
Structural engineers often need to resort to moresophisticated thermal fluid simulations to obtain boundaryconditions, loading, performance, etc. for their designs andanalyses. Additionally, many engineering problems includestrong interaction between structure and fluid.
This course aims to introduce the essential principles offluid dynamics, important flow phenomena, and basics ofCFD process to structural engineers and how CFD can bebeneficial for their multidisciplinary problems.
The CFD for Structural Designers and Analysts courseprovides the principles of fluid dynamics, turbulence, andheat transfer relevant to structural analyses. The materialinclude simple and practical examples and case studies.
Fluid Dynamics Review for CFD
6 March 2014 2 Week e-Learning Course
The reduced need for expert knowledge in mathematics andcomputer science to use CFD, in addition to proliferation ofCFD, ensures that more organizations and newer analystswould embark on using CFD in their design and analysisprocesses. It is imperative that these new organizations andanalysts know and understand the physical principles of fluiddynamics in addition to acquiring working knowledge ofCFD.
The Fluid Dynamics Review for CFD course provides anoverview of the essential principles of fluid dynamics forpractical application of CFD for real-life engineeringproblems. This course introduces and discusses thefundamentals of fluid dynamics to help understand thephysical principles behind CFD for correct and effective useof CFD.
Introduction to CFD Analysis: Theory and Applications19 March 2014 Training Course | Wiesbaden, Germany
The course offers excellent guidance on how to judge whichapproximations are acceptable and appropriate for solving a
wide range of practical problems. Of equal importance is themanner in which the results are interpreted. Advice isprovided which allows the correct decisions to be taken,based on results which are known to be reliable. Interactionis encouraged throughout the course, with the planning anddesign of a complete CFD project and examples of simplehand calculations, mesh designs and solution designs beingset for the class to complete.
Méthode des Éléments Finis pour leDimensionnement et la Vérification dePièces et Structures25 March 2014 Training Course | Paris, France
This French –language training course is the first step ofNAFEMS education for beginners in analysis. It aims to giveto designers and technicians the theoretical knowledge andbest practice information regarding FEM and FEA to safelyand efficiently model and analyse parts and structures.While independent of any software, the course containsseveral exercises to ensure sound knowledge.
NAFEMS Regional Conferences 201413-14 May 2014 | NORDIC| Göteborg, Sweden20-21 May | Deutschsprachige| Bamberg, Germany28-30 May | Americas | Colorado Springs, CO, USA4-5 June | France | Paris, France 10-11 June | UK | Oxford, UK
Gain an insight into future trends and cutting-edgeapplications with the only truly independent and vendor-neutral forums dedicated to analysis and simulation with our2014 Regional Conference Programme.
NAFEMS will once again host a series of ground-breakingconferences across the globe, bringing together the analysisand simulation community in an open forum to shareexperiences, explore future trends and examine cutting-edgeapplications.
The 2014 Conferences will increase awareness and provide adiscussion forum for topics that are vital to engineeringindustrialists and academics, offeringattendees an unrivalled combination ofindustrial knowledge, expertise and forward-thinking to aid their deployment of CAE nowand in the future. 5
NEW
NEW
7
NAFEMS welcomes users of all CFD software productsto contribute their papers. Further information about thejournal can be found on the NAFEMS website at www.nafems.org/about/tech/cfd/activities/journal/
Papers for consideration should be sent [email protected]
Deadline is 31st January 2014
We thank all the authors in advance for theircontribution.
This call for papers is for a special issue of the NAFEMS International Journal of CFD Case Studies to showcaseCFD used by designers.
Numerical simulation is now firmly entrenched in the design process. Previously, simulations followed the designprocess to confirm the functionality. Now it is being used to kick off the design process often at the draft or conceptdesign stage.
Engineering flow simulations are non-linear in nature and often have challenges due to modelling the flow regiongeometry (rather than the solid part that is modelled by CAD), turbulence (which usually must be modelled ratherthan directly simulated), coupled physical processes and many others. These aspects are not easy to handle andhave prompted much software development to improve usability and ensure CFD is as widely used as possible.Software packages aimed at the non-CFD specialist or design engineer can, for example, automatically chooseturbulence models and perform meshing for the user. Simulations performed by the designer are often used togive early understanding of relative performance of design iterations and to test out new ideas early in the designprocess. They may subsequently result in additional calculations by a CFD specialist, for example where a greaterlevel of accuracy or prediction confidence is required or additional, more complex physics is present.
This special issue of the NAFEMS International Journal of CFD case studies will show how designers and other non-CFD specialists are using CFD tools and how the results are interpreted for engineering design purposes. Paperswill promote the benefit of numerical simulation to and by non-CFD specialists and show how designers, byunderstanding further the behaviour of fluid flow and heat transfer via their simulations, can produce more effectivefirst designs.
call for papers
industry events
2014jan-jun
March
07March
17March
25April
03-18
April
23May
12
May
14May
19May
21June
04June
17June
24
LMS Nordic Users Conference
Conference I Gothenburg, SwedenMSC User Meeting 2014
Conference I Munich, Germany
STAR Global Conference 2014
Conference I Vienna, AustriaSIMULIA Community Conference 2014
Conference I Providence, RI, USA
3rd Buckling and PostbucklingBehaviour ConferenceConference I Braunschweig, Germany
ESI Global Forum 2014
Conference I Paris, France
ANSYS Conference & 32nd CADFEMUsers' MeetingConference I Nürnberg, Germany
CIMdata’s PLM Market & Industry ForumsConferences I Various
FLOW-3D European UsersConferenceConference I Vienna, Austria
Engineering Simulation Show 2014
Conference I Derby UK
Altair European TechnologyConferenceConference I Munich, Germany
modeFRONTIER International UserMeeting 14Conference I Trieste, Italy
8 nafems.org/industryevents
REGIONALCONFERENCES
2014
NAFEMS
NORDICGöteborg, Sweden 13-14 May
DEUTSCHSPRACHIGEBamberg, Germany 20-21 May
AMERICASColorado Springs, USA 28-30 May
FRANCEParis, France 4-5 June
UKOxford, UK 10-11 June
nafems.org/2014 9
IMPROVING TH OF BIRD STRIK WINDSHIELDS
10
Aircraft and other vehicle windshields are exposed to a widevariety of loads, one of the most dangerous being the impact ofsolid objects at high velocities. For aircraft of all kinds, a relevant
load case is bird strike, happening mostly during the landing andtake‐off phase of flight. It is vital that such events do not completelydestroy the windshield, which can be made of different materials, forexample glass, PMMA, polycarbonate and combinations of thesematerials joined by rubbery interlayer materials (see Figure 1).
To ensure safety, such glazing needs to fulfil a wide variety of tests, one ofwhich is a bird strike test where an impactor (e.g. dead chicken or gelatineblock) is shot onto the windshield using a large pneumatic cannon. Thesetests are time consuming and expensive; therefore FEM is used to optimizeparameters like material type and layer thickness. This should result in onlya final homologation test and no other costly test series.
In a government�supported project, the Institute for product and productionengineering (IPPE) at the University of Applied Sciences and ArtsNorthwestern Switzerland (FHNW), Mecaplex Ltd as manufacturer of suchaircraft structures and Aerofem GmbH as design and calculation company,have joined forces to increase the prediction accuracy of such simulations.To achieve this goal it was necessary to conduct extensive material testingas well as examine a variety of FE material models.
Parts: Mecaplex Ltd, Grenchen, Switzerland Simulation: Aerofem GmbH, Ennetbur̈gen, Switzerland Project: University of Applied Sciences and Arts Northwestern Switzerland FHNW/ Institute of Product and Production Engineering, Windisch, Switzerland
HE SIMULATION KE ON PLASTIC
11
Important issuesFinite element programs, such as LS‐Dyna whichwas used in this project, offer a wide range ofmaterial models, from simple linear‐elastic tocomplex nonlinear with damage models and so on.But the answer to the important question, whichone is best suited to the task at hand, needs to befound by the users themselves. To find a suitablemodel, knowledge about the material andunderstanding its loading conditions arenecessary in order to know the capabilities thatneed to be included in the material model.
The first task therefore was to give some thought tothe loading of the parts and the consequences ofthis, in order to ensure a reliable FEM simulation.
For example these are:� High impact velocites lead to high strain rates
in the plastic materials used. Their behaviour is
strongly strain rate dependent, meaning theirstiffness behaviour at high rates of deformationdiffers from ones measured at low rates.
� The impact leads to a bending deformation ofthe windshield, resulting in tensile andcompressive stresses. The materials behavedifferently in each mode, not only regarding thestiffening but also the failure behaviour.
� As aircraft fly in a wide variety of weatherconditions, materials need to work from low tohigh temperatures. Plastics properties alsochange with altering temperatures.
There are many more considerations like this. Tobegin with, we concentrated on analysing thestrain rate and load condition dependency of themost commonly used materials and on how torepresent these characteristics correctly in FEMsimulations of the birdstrike.
Figure 1: Bird strikedamaged windshield,example of materiallayup in aircraftwindshield (below)
“To ensure safety, such glazing needs to fulfil awide variety of tests, one of which is a birdstrike test where an impactor is shot onto thewindshield using a large pneumatic cannon.
12”
Glass
Adh. Interlayer
PMMA
Adh. Interlayer
PMMA
Material model calibration:An exampleAs already mentioned, layers of glassy polymerscan be combined using an interlayer material,which can for example be a thermoplasticpolyurethane (TPU). These thin films ofthermoplastic elastomers show a highly nonlinearelastic behaviour, completed by strain ratedependency and being nonsymmetric regardingtension and compression loads.
To find a suitable modelling method, we firstconducted a thorough study of the literature to geta better impression of the necessary behaviours inFE simulation of these materials. While it wouldalso be feasible to model such interlayers by usingspecial contacts (*Tiebreak in LS‐DYNA) orcohesive elements, we chose to use continuumelements for the interlayer. Only with the complexmaterial models available there was it possible toinclude all relevant material behaviours. Twocandidates were chosen as material model:
� *Mat_ Plasticity_Compression _Tension(Mat_124), an elasto‐plastic material modeloffering the possibility to define different basecurves and strain rate dependencies for
tension/compression. Through a Maxwell‐typeviscoelasticity included with a Prony Series alsothe elastic part can be influenced. Onedrawback is that only base curves of true stressversus plastic strain can be inputted which arethen scaled for different strain rates, but thisdoes not allow for a change of the curve shapee.g. with increasing strain rate.
� *Mat_Simplified_Rubber (Mat_181), describedas being a “quasi”‐hyperelastic rubber model.The “quasi”‐term is necessary, because there isno real strain energy function used todetermine the stresses (as with "true" hyperelasticity), but only the tangent stiffness isderived as if an energy function was present.Using the model, one needs to keep in mindthat, with the chosen approaches, effects likecreep and stress relaxation cannot be modelledin a correct way, as it lacks the capability torepresent hysteresis in a visco‐elastic sense.
Having chosen the FE modelling method, the nextstep is to get the right input data.
Figure 2: Test Setupfor high speedtensile testing (farleft), strainmeasurement (left,resultingstress/strain curvesat different strainrates in tension andcompression
13
High speed camera
Cold light lamps
Upper clamp and catch
Specimen
Load cell
Support
To investigate the strain rate dependency, apurpose‐built measurement system wasdeveloped at the FHNW (Figure 2). Using a fasthydropulser, waterjet‐cut specimens with a specialshape are torn at speeds up to 4m/s allowing forstrain rates up to 200/s. Forces are measured usinga piezoelectric load cell and strains are recordedby digital image correlation with pictures from ahigh‐speed camera.
Compressive tests were conducted using roundsamples of stacked foil, compressed on a standarduniversal test machine at low strain rates. Theresults were then extrapolated to higher strainrates using the tensile results and comparableliterature data from e.g. split Hopkinson pressurebar testing.
Further testing then included cyclic loading, shear,confined compression and dynamic mechanicalanalysis (DMA), as well as comparisons betweenin‐ plane and transverse data. After thiscomprehensive test program, the material can beconsidered well enough known under bird strikeconditions.
Material model verificationThe tensile and compression stress‐strain curves(Figure 2) are the basis for both FE materialmodels. Simulations of these basic tests showedthat the material models represent the tests whichthey are based on well. Now, as there are manymore modes in a real impact than just tension andcompression, it was interesting to see how thematerial models behave under other loadingconditions. For this Arcan type shear tests (Figure3) as well as instrumented pendulum impact tests(IPIT, Figure 4) were used. To gain useful results,for some of the tests it was necessary to addcovering layers to the rather soft interlayermaterial. Using PC or PMMA would have beenideal, but as these materials also pose simulationproblems themselves, in order to ease comparisonbetween simulation and test of the interlayer only,thin aluminium sheets were used as top andbottom layer (see e.g. Figure 4).
...the most promisingcombination ofmaterial models hasbeen put to the test bysimulation of a fullcanopy FE model
“
”14 Figure 3: Arcan bulk shear test setup, test and simulationwith stress comparison, force vs displacement curvesfor different material models
For the other materials used in windshields asimilar research and verification process has beenconducted, to achieve correct material models forall of them. Where necessary, parameteridentification also made use of optimizationmethods to achieve the best accord between testand simulation.
Impact simulation validationHaving identified the material parameters of allindividual materials, the next step is the validationof combinations of materials in one layered plate.Here an impact machine developed at the FHNWcomes into operation, allowing for tests of squaresheets of material under impact load with a steelimpactor, providing measurement of impact forceversus time. The measurement is performed bymetering the acceleration of a large,air‐suspended mass serving as base plate for theimpacted plate. Results of such a simulation (in thisexample shown with layered‐shell elements) thetest setup and comparison of measured andsimulated forces are shown in Figure 5.
As for the material models, LS‐DYNA offers a widerange of possibilities for contacts, furthermore therefinement of geometrical modelling as well asmesh densities and other parameters besides thematerial model itself need to be correct. Onlywhen all these components match can a goodcorrelation between test and simulation beachieved, even in the simple impact setup shown.
Finally, the most promising combination ofmaterial models has been put to the test bysimulation of a full canopy FE model. Thissimulation (Figure 6) was compared to resultsproduced with models prior to this project andwith bird strike tests. It could be shown that severalparameters like deformation during impact, failureand plastic deformation after test cannow be simulated with much higheraccuracy.
...the risk offailure during thehomologation tests isminimized, therebygiving a competitiveadvantage to theparticipatingcompanies.
“
”
15
Figure 4: IPIT test setup, tested sample and simulation,force vs. displacement curves for different materialmodels
In Figures 3 and 4 it can be seen that MAT_181 iswell suited to also represent these loads whichwere not directly used as input to the materialmodel. Note that, to achieve these results, nofurther modifications of the material cards weremade; only the data from tension/compressiontesting were used.
MAT_124 is generally acceptable, but fails when itcomes to unloading: This is only represented bylinear elasticity, whereas in MAT_181 an unloadingload curve is defined.
Conclusions and OutlookBy using newly developed or improvedmeasurement methods an enhancedunderstanding of the behaviour of the differentplastics used for aircraft windshields wasachieved. This data allowed to select the bestmaterial models and to identify their parametersto simulate bird impact on these parts. It couldbe shown that the new models providesignificantly enhanced prediction quality. Thisnow allows a further optimization of the glazingsregarding thickness and material types to
minimize weight. Furthermore the risk of failureduring the homologation tests is minimized,thereby giving a competitive advantage to theparticipating companies.
Further work on the topic will focus on thematerials in a temperature range between ‐40°and 80°C under high strain rate loading. A facilityunder construction is an instrumented test rigallowing for small scale birdstrike tests byshooting a birdlike material like ballistic gelatineat high velocities.
16
Figure 5: Impact test (left), results of test andsimulation (middle), test and simulation curvesof force versus time (right)
Figure 6: FE‐Simulation of bird strike on canopy
References: Fritzsche P. et al: A procedure for the simulation of failure inthermoplastic composites; Composite Structures 2007Wyss I.: Bird Impact on Multilayer Aircraft Transparencies:Investigating the Interlayer Simulation; Master Thesis 2012Ramakrishnan K.: Low Velocity Impact Behaviour ofUnreinforced Bi‐layer Plastic Laminates; Master Thesis 2009
Rinaldi R. et al: Modeling of the mechanical behaviour ofamorphous glassy polymer based on the quasi‐point defecttheory— Part II: 3D formulation and finite element modelingofpolycarbonate; Int. J. of Non‐Linear Mechanics 2011
/nafems
/nafems
@nafems
nafems.org
18
Mixing Processby Gas BubblingAn Example of ModelValidation for IndustryOlivier Geoffroy, Hervé Rouch (INOPRO, France)
19
20
Bubbling may not be seen as animportant industrial technology orapplication. Especially in France:“bubbling” in common Frenchlanguage means having a goodtime, or doing nothing.
“”
Based on the paper which won theaward for the Best Practical Use ofSimulation Technology, sponsored byAutodesk, at the NAFEMS WorldCongress 2013, this article fromHervé Rouch and Olivier Geoffroy atInopro is an example of modeldevelopment and validation forindustrial mixing using air bubblingin a liquid tank.
Bubbling may not be seen as an importantindustrial technology or application. Especially inFrance: “bubbling” in common French languagemeans having a good time, or doing nothing. Ourengineers had fun with some virtual bubbles butthey did much more than “nothing” to achieve theresults presented here. Such technology is mainlyused in research fields for mixing (material withspecific rheology, high temperature...). Themodelling technology itself, fast transient VOF(Volume Of Fluid) two-phase fluid dynamics, is muchmore commonly used and may be applied for someother process modelling. This example was chosenbecause it is much more than standard consultingwork. It is a good example of model and softwarebenchmarking, model development and validation.
The model will be presented, and then the meshsensitivity is described as it is an important part ofthe work. Finally some results are compared toexperiments of air bubbling in oil for one and twocolumns of bubbles.
GeometryThe geometry of the results presented here is asimple geometry for experimental test andsimulation validation. It is a 510mm diametercylindrical tank filled with 350mm of oil. The airinjection is a thin tube at the bottom face of thetank. The inner diameter of the injection is 3mm
The geometry of the simulations is exactly same,modelled in axi-symmetry or in 3D for a quarter ora half of the cylinder (one or two symmetry planes).
ModelsThe results were obtained using OpenFoam(www.openfoam.com). This choice was made fortwo reasons:
• The code is “open”, therefore all informationabout model, solver, and boundary conditionsare known and may be modified,
• As an alternative to other commercial CFDcodes, with faster solving due to otherinterface solving in the VOF model.
Interfoam was used to solve the VOF model withinterface compression techniques (interfacecapturing methodology) [3]. The VOF model [4] is atwo fluids model solving an average conservation
Figure 1: Geometry of the experiment. The 3D sim same geometry in full 3D or with one or tw
equation. Following this type of two phase model,the transport models are the same for bothphases. In the simulated conditions, the liquid flowaround bubble columns is laminar but it is turbulentin the injection region. The air flow inside thebubbles is turbulent, especially when taking thermaleffects into account. Then we choose the k -omegaSST model for turbulence [2]. It may be that thebubbling phenomenon needs transient solving, andthat the turbulence vortex size is large due to lowenergy of the turbulence. Then a LES type modelmay be used without significantly increasing solvingtime and memory. Such models have been testedand give similar results.
Following these choices, the solved conservationequations are as follows:
• Mass conservation of the mixed fluids:
(1)
where U is the velocity of the fluids and ρ thevolume mass.
• Momentum conservation:
(2)
where m is the fluids viscosity, s is the surfacetension at the interface between the two phases, k
is the interface curvature, and g is the indicatorfunction which is the gas volume fraction. Forimmiscible fluids modelled by VOF methodology, it isequal to 0 in the liquid phase and to 1 in the gasphase, and between 0 and 1 if the interfacecrosses the mesh volume.
(3)
where k is kinetic energy of turbulence, ui, uj is thevelocity component relatively to the directions xi xj.For the detail of other terms you may see Menterarticle [2] orhttp://turbmodels.larc.nasa.gov/sst.html.
(4)
where ω is the energy dissipation (see samesources for details).
All these conservation equations are similar to theone solved for one phase flow. For two phase flowsolved by VOF methodology, the indicator functionis solved. In the present case it is solved using acompression technique of the interface. This is themost original part of the method.
(5)
where Ur is the compression velocity [2]
Figure 2: Example of a 3D mesh for full 3D simulation
mulation is done in exactly wo symmetry plane.
21
The transient solution procedure is very common,and is referred as PISO [1].
Boundary conditions have to be applied to thecylinder wall, in liquid and in air, at the air inlet insidethe injection tube, and at the air outlet on top of thecylinder. Common wall conditions for turbulencemodels with so called wall functions are used onwalls (cylinder tank walls and tube outside) [5]. Thevalues of velocity, turbulence energy and dissipationare fixed from the wall function. Zero gradients arefixed for the gas fraction and pressure.
At the air inlet the velocity is fixed from the flowrate of 980l/h. A non-constant value is chosen: theaverage value corresponds to the chosen flow rate,and a periodic function is added to allow bubbles toappear (see sensibilities section). Pure air isinjected. The turbulence parameters are chosen forthe stabilized flow inside the injection tube.
At the outlet, the pressure is fixed to 0 (relativepressure) and all other unknown gradients are fixedto zero.
The mixed fluid properties are calculated from thegas and liquid properties following:
(6)
(7)
where the volume mass and the viscosity indices bya and b are the liquid and the gas properties.
Dimensionless ApproachThe properties of the fluids are the following for thegas (air):
• Volume mass : � = 1,225 kg/m3
• Dynamic viscosity : � = 1,789e-5 Pa.s.
And for oil:
• Volume mass : � = 978 kg/m3
• Dynamic viscosity : � = 3.8729 Pa.s. Thenkinematic viscosity is 3960 cSt. Sensibilityhas been done for 2000 cSt et 8540 cst.
• Surface tension at oil/air interface : 0.021N/m²
In the liquid phase, the velocities are between 0.2and 0.5m/s, corresponding to a Reynolds number of50. In the air phase, the velocity may be 30m/s nearthe injection tube extremity or when the bubbles areclosed. Then it could be as high as 8000. That is whya turbulent model was chosen despite the smalldimensions. The bubble size does not depend on theinjection size, but on the turbulence of thecontinuous phase. It also appears that thecoalescence phenomenon significantly increasesbubble velocity.
It's a good example of model and softwarebenchmarking, model development and validation.“ ”
22Figure 3: Results obtained with too much skewed mesh, with Fluent in 2D (left and OpenFoam in 3D(right) Also the mesh size has to be smaller than the size of the bubbles divided by 20. The figure 4
shows a bad interface calculation with a larger mesh size.
23Figure 6: Simulated results from 3D simulation (left) and experimental results (right)
Figure 4: Results obtained with too large mesh size
Figure 5: Result with constant flow rate (top, no bubbles), and modified flow rate (bottom): high frequency sinusoidal perturbation flow rate.
Figure 9: Velocity magnitude 110mm over tank bottom,from simulation (yellow) and measurement (orange). Xaxis is the horizontal distance from bubbling injection.
Figure 10:Vertical component of velocity, 110mm over tankbottom, from simulation (yellow) and measurement (orange).X axis is the horizontal distance from bubbling injection.
The dimensionless approach gives the followingvalues:
• Surface tension effects: The Weber numberis the ratio of inertial forces over surfacetension. The value is 300, inertial effects arepredominant.
• The Froude number is the ration velocityforces over weight. It is equal to 0,2.
• The Ohnesorge number characterizes theshape of bubbles and is the ratio of viscous
forces over inertial forces and surfacetension. Its value is 4.3, showing that theviscous forces are a little greater thansurface tension and inertia.
• The Bond number is the ratio of gravity oversurface tension, and if this value is 730,then gravity is predominant.
In conclusion, the main effects are gravity andinertia, which are higher than viscous and surfacetension effects.
24
Figure 7: Velocity magnitude 50mm under the surface, fromsimulation (yellow) and measurement (orange). X axis is the
horizontal distance from bubbling injection.
Figure 8: Vertical component of velocity, 50mm under thesurface, from simulation (yellow) and measurement
(orange). X axis is the horizontal distance from bubblinginjection.
ResultsFirst results are on mesh and injection sensibility.Figure 3 gives an example of a result obtained on askewed mesh. The skew may be less than 2 and thegeometrical progression of the mesh size less than1.1.
The bubble formation is also consistant withinjection flow rate stability. With a constant flowrate, bubbles are not created. Then we use a highfrequency sinusoidal function with unchangedaverage value.
For one column of bubbles the results fromexperiments and simulations are compared in figure6.
This qualitative validation was confirmed byquantitative validation by comparing the simulatedliquid velocities with the measured velocity usingthe PIV method. Figure 7 shows good agreementbetween the measured and simulated liquidvelocities.
These calculations have been performed with a 3Dmesh of 111000 elements. With fixed timestepping, the CPU time is 16 hours per physicssecond, allowing a calculation of 5 seconds ofmixing (needed to get pseudo stationary solution)in half a day using only 8 cores.
This method has been used to simulate two columnbubbling. The bubbles of the two columns interactand sometimes coalesce as in figure 11.
ConclusionsBubbling is primarlily used for mixing in industrialapplications. A VOF model has been used tosimulate one and two column bubbling withappropriate mesh and injection boundaryconditions. The results have been validated bycomparison to measurements for air bubbling in oil.Qualitative comparison has been carried outbetween experimental and simulated results forthe bubble shape and coalescence. Quantitativecomparisons have been done for liquid velocity. Thevalidity of the model and the short computer timeneeded allow efficient simulation in industrialapplications.
References[1] Issa R. I., 1986, “Solution of the implicitly discretisedfluid flow equations by operator splitting.” J. Comp.Phys., 62(1):40 65.
[2] Menter. F. R., 1994, “Two-Equation Eddy-ViscosityTurbulence Models for Engineering Applications.” AIAAJournal, 32(8):1598-1605.
[3] Rusche Henrik, 2002, “Computational fluid dynamicsof dispersed two-phase flows at high phasefractions”, Ph-D thesis, Imperial College of Science,Technology & Medecine, London RU.
[4] Hirt C. W. and Nicols. B. D. 1981 “Volume of fluid(VOF) method for the dynamics of free boundaries”. J.Comp. Phys., 39:201-225.
[5] Launder B. E. and Spalding D. B., 1974, “Thenumerical computation of turbulent flows.” Comp.Meth. Appl. Mech. Eng., 3:269 289.
Figure 11: Simulation of two column bubbling 25
26
ReverEnginMade
27
rseneering Simple
Dipl.-Ing.(FH) Alexander KraußProf. Dr.-Ing. Uwe FischerWest Saxon University of Applied Sciences of Zwickau
28
The dimensioning of thin walled parts withinautomotive construction is highly influenced by FiniteElement Analysis and Optimization. The common targetsare to improve stiffness or eigenfrequencies. The FEModel which is based on the concept CAD geometry ofthe designer is mostly created by a specializedcomputational engineer. The data exchange is done withuniversal non-parametric CAD data formats, like IGES. Atthis stage a direct update of the FE geometry, bychanging the CAD geometry and vice versa, isimpossible, so the necessary variation of the shape isachieved by deforming the FE mesh. These deformationsare called beads in technical language, and they must berebuilt in CAD geometry by hand. This so called “reverseengineering” is very time consuming for the designer.Within the research project “Automation of optimizationtasks inside the construction process of thin walledparts”, at the University of Applied Sciences Zwickau(Germany), in association with FES Zwickau andVolkswagen Kassel, we created a new workflow whichallows for more simple reverse engineering. The processwill be demonstrated in this article by using anautomotive body part, namely a car tailgate.
Initial SituationTo increase the fourth eigenfrequency of the tailgate, a FEshape optimization was performed using the commercialsoftware OptiStruct® (Altair®). The result is a distortedfinite element mesh, where necessary changes are shown bythe red areas (figure 1). They have to be transferred fromthe computational engineer to the designer.
Process OverviewOur aim is to support the designer in his daily workflow andto reduce time consuming steps, but not to create a one-click solution. Therefore a step by step process is used tocreate parametric CAD geometry out of distorted finiteelement shapes. The method is focused thereby onproducibility. The process can be divided into the followingsteps, as shown in Figure 2. The individual actions will bediscussed later in further detail.
Step 1: Data TransferWithin this first step only the necessary red areas will betransferred to CAD. With HyperView®, the postprocessor forOptiStruct® results, a so-called “ISO-Surface” can begenerated (Figure 1). Using the “export as STL-File”function, a triangulated mesh will be created, which is thenimported into CATIA V5® R19 (Dassault Systems®). Dealingwith such data in CATIA® is not easy, since extra licensedworkbenches are normally necessary. To give the designerthe ability to use this STL-file within the Generative-Shape-
Design workbench, we programmed an STL-To-IGES-Converter. The converter is accessed by a VBA Macro andallows a direct import into CAD, while using the dataformat IGES. The designer only chooses the given STL-Fileand a geometrical set, where the optimization surface willbe inserted. But the converter also runs as standalonesoftware with GUI. Therefore it can be used with other CADor FE optimization software. The result is a single, non-parametric surface which can be seen in the structure tree(Figure 3).
Step 2: Data Filtering and SurfaceDisassemblyFor further editing the “join” surface has to be“disassembled” into its domains. Small pieces which remainafter the transfer process also have to be filtered out.Therefore we developed another VBA Macro which combinesdisassembling and filtering. With an interactive preview theuser can define the size of the filter. Each single domain isnow able to be selected. The process is shown in Figure 4.
Step 3: Parametric Feature CreationThe construction of a bead is mostly a compromise betweenproducibility and approximation of the optimization results.The bead geometry is mainly characterized by its topsurface and contour. These are represented by thetransferred optimization surfaces. We created CATIA® User-Features for some standard cases in bead constructions. Theso-called Bead Construction Toolkit is an easy to usefeature catalogue. It allows the designer to pick the rightfeature for his case.
Figure 1: Results of FE shape optimization and created ISO-Surface
Figure 2: Process overview
User-Feature 1: Line BeadIn some special cases the optimization surface shows adistinct direction and a nearly constant profile (Figure 5).The designer creates a guideline which approximates themiddle curve of the optimization surface. He can thenchoose a feature from the catalogue which fits best intothe profile, e.g. a semi-circle or trapeze bead. Afterselecting the guideline and the base surface, the featurewill be created. With just a few parameters, like the heightand width of the bead, the geometry can be easilycontrolled and optimized, but normally a distinct directionis not provided. Therefore the next two features have beencreated to suite a wider range of cases.
User-Feature 2: Custom Bead Contour, Basedon Original Surface This feature uses a custom contour which is drafted by thedesigner. The easiest way to do this is to place a sketch onthe plane of the local inertia axis system of theoptimization surface (Figure 6). In this case a feature isalso provided. The sketch is then transferred to the originalsurface by the feature and cuts out the top surface of thebead. This piece is then translated along the draw direction
and so the bead is created. If the original surface changes,the bead will follow. But this assumes a high qualityoriginal surface, which applies to the most parametricshapes created using CATIA®.
User-Feature 3: Bead contour and Surface,Based on Optimization Surface To approximate the optimization results in a more accurateway, the last feature is based directly on the triangulatedsurface. The designer is now independent from the qualityof the original shape. With a feature integrated surfacerecognition, the top surface of the bead is created, withreference to the optimization surface. Using the parameter“precision of approximation”, the shape can be controlledas shown in Figure 7. Besides this, the height of the surfaceand thus the height of the bead can be changed.Furthermore, the discontinuous border is adapted withcontrol points for a continuous spline. If the automaticapproximation is not suitable for production, the contourcan be easily modified. The designer can do this byremoving or adding control points to the spline. Inconnection with other construction elements, the bead iscreated with a Power Copy.
29
Our aim is to support the designer in his dailyworkflow and to reduce time consuming steps, butnot to create a one-click solution. Therefore a stepby step process is used to create parametric CADgeometry out of distorted finite element shapes.
“”
Figure 3: Data transfer OptiStruct® – CATIA® V5
Evaluating the Results The most interesting question is: how close can we come tothe optimization results? CATIA® V5 has the ability tocalculate and import results from multipurpose FE-Solverslike MSC.Nastran® and ABAQUS® with special workbenches.An import of the FE mesh created by the computationalengineer is also possible in some cases. To get a quickevaluation of the results, a Hybrid FE model can be used.This consists of an imported FE-mesh and remeshed areaswhich are based on the CAD geometry which has beenchanged by the designer. After building up a finite elementmodel and FE analysis, the designer can evaluate theeffectiveness of his beads and correct them if necessary,e.g. modify the height to improve stiffness. If there aresmall changes in CAD geometry, the designer can try torestore the original results with a parameter optimization ofthe beads. So there is no need for FE shape optimization inevery case, which is a big advantage. If a new FE shapeoptimization was performed, the new ISO Surface can beeasily used to update the bead feature.
The remeshed part can also be used to evaluateproducibility. Therefore, we created a CAD integratedtemplate for deep drawing simulation which gives a surfaceimpression of critical areas on the part.
Summary and DiscussionThe process shown gives an overview of how to createparametric CAD geometry out of non-parametric FE shapeoptimization results by using a CATIA® User-Feature. Themethod focus on producibility, where the designerthemselves can define what the bead looked with respect tooptimization results. The biggest advantage of this methodis a much more efficient design cycle in productdevelopment. With result evaluation using CAD integratedFEA systems, the designer can calculate the effectiveness ofthe created beads. The computational engineer is then ableto concentrate on more specific tasks. But our research isnot finished at this point. There are some problems likemulti-domain beads, surface borders and holes that have tobe solved. Besides this, the whole process is only reallyefficient when the computational engineer and designerwork together as closely as possible. 30
Figure 4: Disassembly and filtering of the join surface with a CATIA® Macro
Figure 5: Creation of Line Beads
AcknowledgementsThe editors take this opportunity to gratefully acknowledgethe German Federal Ministry of Education and Research(BMBF) for sponsoring this project (Number: 17N2011). Ourvery special thanks go to Christoph Schleicher and RonnyKubik for the support by programming the converter. Forcareful reading and correction of the text a special thanksgo to Tina Singer and Tristan Lodge.
Contact:Dipl.-Ing. (FH) Alexander KraußE-Mail: [email protected]: +49 (0) 375/5361778
Dr.-Friedrichs-Ring 2A08056 ZwickauGermany
31
The biggest advantage of this method is a muchmore efficient design cycle in productdevelopment. With result evaluation using CADintegrated FEA systems, the designer cancalculate the effectiveness of the created beads.
“”
Figure 6: Bead with sketched top surface contour
Figure 7: Bead with surface approximation
Computational Fluid Dynamics is about solving difficultengineering problems, using expensive software,enormous computing resources and highly trainedengineers. If the problems weren't difficult, or veryimportant, then it is doubtful that anyone would devote somuch effort, time, and money at solving them. From theperspective of a modern engineer, it would be easy toassume that this desire to apply simulation technologycomplex problems is a recent concern; that only today arewe able to contemplate solving tough industrial problems,armed with a complex array of multi-physics simulationtools.
This is a misconception. Forty years ago, CFD was born froma desire to solve difficult problems involving turbulence,heat-transfer, and combustion, based on the vision of asmall group of pioneering researchers who were able to seebeyond the meager computing resources available at thetime, and to develop the techniques and methods thatwould ultimately revolutionize engineering.
Prof. David Gosman is one of those pioneers. As a memberof Prof. Spalding’s Imperial College CFD research groupfrom the beginning, he played a pivotal role in developingsimulation methodologies that could cope with the complexgeometries of real industrial problems, many of which areemployed in all commercial CFD codes today. He alsopioneered the use of CFD for combustion in reciprocatingengines and methodologies and software that he developedhave been applied to investigate the design of almost everyautomotive engine designed since the early 1990s.
Prof. Gosman arrived at Imperial College in theautumn of 1962 have recently graduated
from the University of British
Columbia, to study for his PhD under Prof. Brian Spalding.In the early 1960s the focus of the Spalding’s research wasthe development of a 'universal method' for computingturbulent flows, using momentum integral methods for two-dimensional shear flows, and designed to account for freeflows and wall jets. Although these techniques provedmoderately successful for the prediction of “parabolic”boundary layer type flows, they were not applicable to moregeneral "elliptic" type problems (with strong pressuregradients, separation, recirculation and impingement).
Since the solution of "industrial" type problems, especiallythose including combustion, required the solution of elliptictype problems, Prof. Spalding and his team eventuallyabandoned the 2D-parabolic approach in favor of adiscretized “stream-function-vorticity" approach, thatsolved the two-dimensional Navier-Stokes equations (castin terms of stream function and vorticity) using a finite-volume approach and upwind differencing. Although Prof.Gosman's mainly experimental PhD did not directly involvethe development of these methods, he soon becameentangled in their development, to such an extent that thepublication of his thesis was delayed by a number of years.It was this diversion that was to ultimately define his wholecareer.
The culmination of the stream-function-vorticity approachwas the publication of the 1969 book "Heat and MassTransfer in Recirculating Flow”[1], for which Prof. Gosmanwas the editor, and which included the source code for theCFD tool called ANSWER, developed by Runchal andWolfshtein. This book marked a turning point for CFD,demonstrating for the first time that industrially relevant
flow problems could besolved using numericalsimulation, and
Icons CFDProf. David Gosman
32
providing a tool with which to do it. The techniquesadvocated by this publication were subsequently used toprovide the first examples of CFD applied to a recirculatingflow exhibiting combustion.
Having demonstrated that the stream-function-vorticityapproach could be used to simulate low-speed two-dimensional flow problems, Prof. Spalding and his teambegan to investigate the extension of the methods to three-dimensions. However, they quickly realized that thesolution of Navier-Stokes Equations in three dimensionsrequires the solution of six equations when cast in terms ofstream-function and vorticity, but only four equations if castin terms of the primitive variables of velocity and pressure.
This realization development was followed quickly by twoimportant developments: the 1976 introduction of Patankarand Spalding's SIMPLE algorithm variants of which were toform the backbone of almost every CFD code that followed;a year later, Launder and Spalding published the standardk-epsilon model, which provided the first practical methodof modeling turbulence without invoking an arbitrary lengthscale. With these essential ingredients in place, Prof.Spalding’s group were now free to start developing problemspecific CFD codes that were capable of addressing realengineering problems. Although simple by contemporarystandards, these codes could easily be considered an earlyprototype for all that followed, performing important role inestablishing the credibility of the new discipline ofComputational Fluid Dynamics and directly inspiring all ofthe commercial CFD codes that would eventually follow.
Prof. Gosman's own contribution during this period was atwo-dimensional code called TEACH, which he originallydeveloped (together with Dr W.P. Pun) as an educationaltool for the post-experience courses in CFD that Spalding’steam were beginning to offer. With TEACH, Prof. Gosmanpioneered the use of CFD in the classroom, introducingnumerical simulation into the curriculum for undergraduatemechanical engineering students, authoring the first courseto make practical use of CFD as a learning tool for fluidmechanics and heat transfer, and publishing the first textbook [2]. This showed tremendous foresight; although CFDtechniques were beginning to be tentatively examined insome "high technology" sectors of industry, the release ofthe first general purpose commercial CFD codes was stillhalf a decade away, and practical fluid mechanics wasalmost entirely dominated by experimental methods.
Although TEACH was originally conceived as a teachingtool, by publishing the source code (as a 1000 lineFORTRAN program at the back of a text book), Prof. Gosmanmay have inadvertentlypioneered the opensource CFD
movement. TEACH was subjected to much modification andextension, and was probably the most widely used CFDcode in the pre-commercial world [3,4].
The original motivation for Spalding had to been to developsimulation methods for problems involving heat-transferand combustion, which unlike pure fluid mechanicsproblems involved recirculation zones that were not easilyaddressed by either existing theoretical methods orexperimental investigation. It was now, in the late 1970s,that some of these ambitions started to be realized, withthe first practical simulations of combustion in gas turbinesand stationary combustors.
Despite this success, the most significant combustionproblem of all, that of the automobile engine, remainedunaddressed. Unlike other combustion problems, engineflow combustion processes are non-stationary; taking placein a solution domain that has a complex geometry andmoving boundaries. The accurate simulation of enginecombustion would also require the development of multi-phase models to account for fuel-sprays and films, as wellas ignition, combustion and turbulence models. Of course,considering the unsteady nature of the problem, thecomplex physics and the large mesh sizes required, thesolution of engine combustion problems also necessatedthe development of a robust and efficient solutionalgorithm to perform the large number of time-stepsnecessary to achieve a credible solution using the limitedcomputing resources available.
Prof. Gosman published the first axisymmetric CFDsimulation of cold flow in a reciprocating engine in 1978[5],before dedicating much of the next decade to developingthe techniques that would allow the simulation of a fully-detailed engine combustion process in three-dimensions.To account for the movement of pistons and valves, hedeveloped a novel Eulerian-Lagrangian moving meshmethodology, which eventually included cell-layer additionand removal to prevent numerical problems that can occurin high aspect ratio cells. In the field of fuel spray modeling,he also co-developed the Huh-Gosman model for sprayatomization and the Gosman-Bai model for wallimpingement. To address time-step and stability concerns,Prof. Gosman implemented the non-iterative PISO algorithmdeveloped by Imperial colleague Dr Raad Issa, whichallowed the computationally efficient solution of unsteadycompressible flows using relatively large time-steps.
The combination of this work, with many otherdevelopments, resulted in the CFD code SPEED, which wasdeveloped as a semi-commercial collaboration betweenProf. Gosman’s Imperial team and a number of industrialpartners.
33
34
Prof. Gosman’s other significant research interest was indeveloping simulation methodologies that could cope withthe complex geometries of real engineering problems. Thecommercial CFD codes of the time relied almost entirely onfully structured cartesian computational grids, which dealtwith complexity using a crude “stair-step” approach, whichin effect led to large inaccuracies in any geometry thatcould not be represented as a combination of cylinders andboxes. A decade spent trying to model the geometries ofcomplex engine combustion chambers and inductionsystems had convinced Gosman of the need to develop amore robust methodology for simulation using body-fittedmeshes, not just for engines, but also for all types ofindustrial CFD problems. He therefore set out tosystematically find a way of producing flexible meshmethodology that would fit all geometries, howevercomplex. After significant investigation of manyalternatives, he finally settled on an approach based on co-located Cartesian velocities inspired by work of Rhie andChow and then generalized this to partially and then fullyunstructured CFD meshes, including those with slidinginterfaces.
By the middle of 1980s Prof. Gosman’s team had assembleda formidable set of simulation tools, many of which were farin advance of the commercial CFD codes that had begun toemerge, particularly in the field of complex geometryhandling. The experience of testing and supporting SPEEDhad convinced Prof. Gosman that academia was not an idealenvironment from which to develop a CFD code and so,together with Dr Raad Issa, he formed ComputationalDynamics Ltd as a commercial venture, with the aim ofdeveloping an unstructured body-fitted industrial CFD code.
Incorporated on Monday December 19th 1987, the dayknown to the rest of the world as “Black Monday”,Computational Dynamics faced an uneasy birth.Unsurprisingly, as the world’s stock markets crashedaround them, Gosman and Issa initially struggled to findinvestors willing to fund their start-up company, in whatwas still a relatively obscure corner of the technologymarket. That funding would eventually come from adapco; aNew York based structural engineering consultancycompany, which had been performing structural analysis ofengine cylinder heads. adapco had recently turned to CFDsimulation as a mechanism for providing more accurateheat-transfer coefficients boundary conditions for their FEAsimulations, but had been frustrated by the fact that noneof the commercial codes offered the body-fittedmethodologies required to provide results with enoughaccuracy. adapco’s President Steve MacDonald wasintroduced to Prof. Gosman by a mutual contact at the FordMotor Company, and quickly determined that Gosman’s CFDcode would not-only solve his heat transfer coefficientproblems, it would also provide a useful tool for the water-jacket flow-balancing simulations that some of hiscustomers were demanding.
With adapco’s backing, Computational Dynamics set aboutproducing a commercial version of their body-fitted CFDcode named STAR-CD (which stands for SimulatingTransport in Arbitrary Regions). The first version was block-
structured but, by its second release in 1991, STAR-CD hadbecome the first truly unstructured commercial code,offering engineers the ability to construct meshes from anycombination of hexahedral, tetrahedral and prismatic cellsand thereby providing unparalleled geometrical flexibility.Technology that had been developed for SPEED also madeits way into STAR-CD, and it quickly became the default CFDcode for the simulation of engine combustion problems.
More than 25 years after its first release, STAR-CD is stillgoing strong, and continues to occupy a leading position inthe engine simulation market. The vast majority of enginesdeveloped since the early 1990s have been designed andnumerically tested with the aid of STAR-CD, providinginsight that has allowed engine manufacturers tosignificantly reduce both fuel consumption and emissions.Computational Dynamics and adapco now jointly tradeunder the name CD-adapco, and collectively employ morethan 800 people in developing and supporting STAR-CD andtheir next-generation CFD tool STAR-CCM+.
Despite the success of this commercial venture, Gosmanremained dedicated to his academic work, and wasappointed Professor of CFD at Imperial College in 1988,eventually publishing over 200 papers on CFD. In the sameway that Spalding’s group eventually spawned multiple CFDcodes including TEACH, the current leading Open SourceCFD code FOAM (now OpenFOAM) was developed by DrHenry Weller during his time in Prof. Gosman’s researchteam.
Until recently Prof. Gosman could proudly confess that,despite a whole career spent pioneering, developing andeducating with CFD tools, he had never actually performeda CFD calculation using a commercial CFD code. However,he was recently observed participating in "STAR-CCM+ forbeginners" training class. Maybe this is his greatest legacy:after more than 40 years of development CFD tools are nowso accessible that even a CFD Icon can learn how to usethem.
References[1] Gosman A D (Editor), 1969 “Heat and Mass Transfer in
Recirculating Flows”, ISBN 0122919505
[2] Gosman A D, Launder B E, Reece GJ, 1985, “Computer-aidedEngineering, Heat Transfer and Fluid Flow”, ISBN0853128669
[3] Runchal A K, 2008, “Brian Spalding: CFD and reality”, Proc.CHT-08, ICHMT - International Centre for Heat and MassTransfer, International Symposium on Advances inComputational Heat Transfer, May 11 ss Transfer,International
[4] Hirschell E H (Editor), 2009, “Notes on Numerical FluidMechanics': 40 Years of Numerical Fluid Mechanics andAerodynamics in Retrospect”, ISBN 3540708049
[5] Gosman A D and Johns R J R, 1978, "Development of apredictive tool for in-cylinder gas motion in engines" SAEInternational Congress, Detroit, paper 7803l5
14 January 2014Elements of Turbulence ModelingOnline
15 January 2014Fatigue & Fracture Mechanics in FE AnalysisOnline
16 January 2014Practical Modelling of Joints and ConnectionsOnline
22 January 2014Einfuehrung in die praktische Anwendung der Finite-Elemente-Methode (FEM)Wiesbaden, Germany
11 February 2014Verification and Validation MasterclassFrankfurt, Germany
12 February 2014CFD for Structural Designers and AnalystsOnline
6 March 2014Fluid Dynamics Review for CFDOnline
19 March 2014Introduction to CFD Analysis: Theory and ApplicationsWiesbaden, Germany
25 March 2014Méthode des Éléments Finis pour le Dimensionnement et laVérification de Pièces et StructuresParis, France
upcoming datesnafems.org/training
36
37
To infinity& beyond?
There’s always goingto be a limit, whichmeans it’s finite. So,the phrase “Infinitecomputing” isessentially amarketing slogan.And a good one.
“
”
I recently attended Autodesk University (AU), a userconference held annually in that most common yetdisturbingly dissonant of places, Las Vegas.
Autodesk is perhaps best known for its AutoCADdraughting product, but in recent years the companyhas made massive strides in establishing itself as aplayer in the simulation field.
This has been made possible through acquisition,which has seen the integration of a wide range ofsimulation technologies and products.
From Solid Dynamics’ kinematics tools, through Algorand into the more specialist areas such as Moldflowand Firehole (for composites analysis), it’s clear thatthis is no longer the ‘AutoCAD’ company, but a wholenew beast.
One of the key themes of this year’s AU keynotesession, given by Autodesk’s chief technology officerJeff Kowalski, was the potential for cloud-computing,which the company always tags with the term “infinitecomputing”.
Now, I’m perfectly aware that I’m talking to aneducated audience with experience of large scalecomputing.
After all, the simulation field has been a key adopterof cluster and server-based compute farms - if not akey driver in their development in the first place. Andchances are you have the same reaction to this term“Infinite computing” as I do.
Namely, “bullshit.”
Off-site and out of mind?The idea of using a hardware stack that allows largerscale computation of studies is nothing new in thesimulation space.
In fact, one might argue that it’s standard practice,particularly in those advanced areas of automotive,motor-sport and aerospace.
Often these facilities are accessed from remotelocations, over the internet or by VPN, which isessentially what the cloud is.
Banks of server racks that you rent and use as you seefit. The only real technical difference is that theservers aren’t on your premises and owned by you.
So with that in mind, the use of the term “Infinite”seems misleading: No matter the actual size of thedata centre and the number of racks available, it’s notinfinite, is it?
There’s always going to be a limit, which means it’sfinite. So, the phrase “Infinite computing” isessentially a marketing slogan. And a good one.
It’s certainly a hell of a lot snappier than “Loads ofserver racks, we can’t count how many, but trust us,there’s… loads.”
Speeding up the ProcessWith that off my chest, let’s consider the train ofthought that this term “Infinite Computing” ensued. Ifwe assume that the term is real and that we do indeedhave limitless computing potential at our fingertips,how would that affect the simulation process?
Or to put it another way: “If we have a limitlessamount of computation resources at our disposal,how does it affect the questions that we ask?”
We’re all used to working with assumptions in theworld of simulation. Whether they’re big or small, thewhole process relies upon them. The reason for themis to get around the limitations in either the softwareor the hardware used to compute the simulations.
So, for instance it might be abstracting a set ofgeometry to remove small and seeminglyinconsequential features (fillets, small holes etc) tomake meshing more efficient or it might be using anidealised physics model to get the simulation to solvemore quickly.
Examples that spring to mind are the use of beam andmid-plane modelling techniques. I’m sure there arebetter examples around but the facts remain, that toget the best out of both the code and the computationresources, we’re used to this type of idealisationwork.
What if we had limitless (or at least, near limitless)computation capability, would this work still be done?
38
Would we be able to take the geometry from thedesign and engineering process, rich with all thefeatures that exist, and run with that? Would we nothave to spend hours de-featuring a model or creatingan accurate mid-plane model from scratch?
There’s also the question of iterative solves: Whetherconvergence-based techniques or design ofexperiments-based optimisation, many simulationstudies require not just a single solve, but rather aseries.
When setting the solve characteristics for these typesof jobs, we’re often forced into the decision ofaccuracy or reliability of results versus the timeavailable.
The design of experimental type optimisation oftenrequires that a specific number of iterations orvariation of parameters is set. After all, having anoptimised design is one thing, but if it takes six yearsto solve, it’s no good to man nor beast.
Shifting the BottleneckThen, of course, into this you can also throw thewhole multi-physics thing. Multiple solvers workingon quite different tasks, passing data back and forth,to achieve a comprehensive result takes a LOT ofprocessing time.
So, imagine a world where none of this matters,where we could solve without worry about thehorsepower needed to get a usable result.
Would we change the questions that we ask? Wouldwe look to have fully detailed models that are solvedin multiple physics domains? Would we look for theabsolute optimised design, where every singleparameter and goal is used to drive that optimisation?
Design and engineering and the simulation process isalways a time constrained process. Removing thecomputation barrier, we would essentially be doingnothing to improve our working processes.
Yes, the argument is that a much richer, more detailedunderstanding of a products’ performance andbehaviour could be garnered, that cannot be denied.
But for the majority, how much extra would weactually learn? Answers could be gained faster, butwould that extra time be spent adding moreinformation.
Would the bottleneck shift from the calculation stageto the pre-processing? Would the time we save ongaining quick answers be negated by the need tospend hours upon hours setting up each solve? I’ve asuspicion that it would.
Technology used to be cold and efficient, but whenyou take a look at how it’s used, it’s incredibly fluid.Parkinson’s Law, says: “Work expands so as to fill thetime available for its completion”.
Given the potential to vastlyincrease computation powerwe’d be raising the bar forwhat can be achieved.
That I’m sure would elevatethe game in terms of thecomplexity of oursimulation studies, butthen we’d probably moanthat things are grinding toa halt again.
Al Dean is Editor in Chief and Co Founder of DEVELOP3D
...the simulation field has been a key adopter of cluster andserver-based compute farms - if not a key driver in theirdevelopment in the first place. And chances are you havethe same reaction to this term “Infiinite computing” as I do.
“”
CAE has certainly benefitted from computertechnology. While laptops have replaced desktops andprojectors have replace viewgraphs, as we all know, itis faster and cheaper processors that have increasedthe CAE horsepower, thereby allowing more accurateand quicker CAE solutions. I remember running fullscale automotive CFD models on one CPU andagonizing over the hand-built mesh so that it wouldsolve a fast as possible. Let’s face it, we all know thatfaster computers have directly benefitted CAE;however, technology comes in subtler forms as well. Ishould point out that I see technology as a tool to get amessage across. Therefore, I confess to being surprisedat the results of a relatively new feature of a Googlesearch: the auto complete function, which was added awhile ago to leverage common search phrases withpage views, thereby aiding in your search. Go aheadand fire up Google, then in the search line, type in“engineers are”[1].
I’ll wait.
Now that is quite ... enlightening, because, in my mind,this simple bit of technology – the auto completefunction – is direct feedback on how the public (rightlyor wrongly) perceives engineers. We are: arrogant(well, sure, sometimes probably because we have atendency to be in continual “problem solving” mode),people too (that’s nice), boring (possibly, my 14 yearold daughter certainly thinks so), weird (probably),dumb or smart (hmmm, make up your mind), nerds(true), underpaid (also true, but then again mostpeople think they fall into this category, even CEOs askfor more compensation independent of how theircompany is doing[2]), and/or cool (finally, someonethinks we part of the “in crowd”).
At least engineers have a better reputation thanscientists (idiot, liberals, and liars) and
mathematicians (crazy, lazy, and stupid) or evenlawyers (evil) and CEOs (sociopaths). Furthermore,this simple auto complete function does more than justshow the perception of engineers, scientists, orwhatever – it is a direct path into the public’sconsciousness. I will confess I spent way too much timewith different “is” and “are” auto complete searches oncareers, politicians, celebrities, athletes, policies, etc. Iam certainly aware of the benefits and drawback ofofficial public opinion polls[3], and while I agree thatan internet search may (OK, does) include an aspect ofthe extreme, I cannot think of an easier way to gaugepublic opinion.
I have always said the engineers do not get enoughcredit for what we do[4]. Auto complete seems toconfirm that. However, as I also said in my last column,the potential of CAE rests on convincing others that itis, in fact, useful. In this case, it means convincing thepublic that computer aided engineers are not arrogant,stupid (or smart), etc. Here is what I would like to seefor an auto complete function in the (hopefully near)future:
• engineers are important • engineers are leaders• engineers are creative• engineers are easy to understand• engineers are making the things that make the
world go around
But it has to start with us as we are the only ones thatcan change how others perceive us.. This is my NewYear’s resolution: to take a small step towardsexpanding the public knowledge of engineering ingeneral and hopefully with an emphasis on computeraided engineering.
What are your thoughts on this? What could we do tobetter inform the public of our usefulness? Send me ane-mail at: [email protected].
-The CAE Guy
[1] I got this idea from the Bad Astronomy blog atSlate: “Scientists are …”, Phil Plait,www.slate.com, December 4t, 2013.
[2] “Are We Paying Our CEOs Enough?”, DavidFutrelle, Time (online), March 28, 2012,http://business.time.com/2012/03/28/are-we-paying-our-ceos-enough/
[3] “The CAE Guy”, Benchmark, NAFEMS, January2013.
[4] “The CAE Guy”, Benchmark, NAFEMS, January2008.
The CAEGuy
40
