13e rencontre SMAI math ‐ industrie Physically correct and timely simulations: the challenge towards Virtual Prototyping Argiris Kamoulakos, ESI Group 22/03/2011 SMAI ‐ HPC 1
Introduction • Today's industry trend is towards reduction of physical prototyping through simulation driven design in order to reduce drastically the time ‐ to ‐ market of new products. This is called Virtual Prototyping. However, it relies heavily on the capacity to reproduce through simulation the correct physics and in a timely way to respect the industrial development process deadlines. 22/03/2011 SMAI ‐ HPC 2
ESI Group Offering Software & Services for End ‐ to ‐ End Virtual Prototyping 22/03/2011 SMAI ‐ HPC 3
Enjeux / problématique industrielle • Crash Failure Analysis as an industrially affordable solution • Must be performed “over-night” 22/03/2011 SMAI ‐ HPC 4
A typical problem: automotive side ‐ impact B ‐ Pillar model 22/03/2011 SMAI ‐ HPC
The Wilkins (EWK) metal rupture model - Mathematical details - A cumulative strain damage model Non proportional loading Plasticity based Triaxiality dependence Lode angle dependence 22/03/2011 SMAI ‐ HPC
Importance of "Size effect" A severe condition imposed on model element size in order to be predictive Size effect: • critical volume to be saturated with damage in order to start a “crack” • Typical values • Aluminum: (0.15mm)**3 • ~6 solid elements per mm • High Strength Steel: (0.05mm)**3 • 15 ‐ 20 solid elements per mm 22/03/2011 SMAI ‐ HPC 7
High fidelity modelling of the B ‐ Pillar ~1,000,000 solids ~4,000 shells 22/03/2011 SMAI ‐ HPC
The industrial challenge The industrial challenge Crash Failure Analysis as an industrially affordable solution Example Must be performed “over-night” 22/03/2011 SMAI ‐ HPC
Towards extreme local refinement Towards extreme local refinement Car Model: Subframe (520k-Model): 520,000 TET10-Elements 879,000 Elements Time Step: 1 μ s Time Step: 0.05 μ s + Time step ratio : R t = 20 Local model element fraction : f L = 59 % Problem: It is not just a matter of “size”, but algorithmic, too. (Disproportionate distribution of solution effort)
Casted parts pose a special issue Gas Porosity, Shrink Porosity, Micro Porosity no porosity with porosity SMAI ‐ HPC 22/03/2011
Very detailed models are needed for the prediction of the grain structure Uniform Solidification in Ni Turbocharger CAFE (Cellular Automata FE) was developed in collaboration with: Howmet, Snecma (France), ABB, EPFL (Switzerland), AETC, Rolls Royce (UK), PCC (USA) SMAI ‐ HPC 22/03/2011 12
In Aeronautics the problem of Birdstrike is particularly accute SPH Courtesy of EC project CRAHVI 22/03/2011 SMAI ‐ HPC
Incorporation of details is critical for performance prediction You cannot predict what you do not model for ! Courtesy of EC project CRAHVI The rivets and their effect in rupture have to be sufficiently represented by the model. SMAI ‐ HPC 14 22/03/2011
Les points de blocage • Mesh size – huge meshes needed • Computational algorithm – Dissimilar meshes in size and type (FE / meshless) • Initial state of structure – Manufacture (filamentary composites) – Connectors (spotwelds, rivets etc.) 22/03/2011 SMAI ‐ HPC 15
• Manufacture of Fillamentary Composites 22/03/2011 SMAI ‐ HPC 16
New Manufacturing methods for advanced composites 17
New Manufacturing methods for advanced composites 18
Virtual Properties prediction (through virtual coupon testing) Particle methods for the resin SMAI ‐ HPC 19 22/03/2011
Virtual Properties prediction (through virtual coupon testing) Particle methods for the resin SMAI ‐ HPC 20 22/03/2011
• Modelling connectors (Spotwelds, Rivets etc.) 22/03/2011 SMAI ‐ HPC 21
Effect of connectors in automotive crash Spotewelded beam modelling Deformed Deformed spotwelds beam 22 22/03/2011 SMAI ‐ HPC
Effect of connectors in automotive crash Spotewelded beam modelling 23 22/03/2011 SMAI ‐ HPC
The challenge of Self Piercing Rivets: knowing their shape and state 22/03/2011 SMAI ‐ HPC
The challenge of Self Piercing Rivets: knowing their shape and state 22/03/2011 SMAI ‐ HPC
Les points de blocage • Mesh size – huge meshes needed • Computational algorithm – Dissimilar meshes in size and type (FE / meshless) 22/03/2011 SMAI ‐ HPC 26
• Modelling substructures and systems 22/03/2011 SMAI ‐ HPC 27
Application to Battery Modelling Pr Prob oblema lematics: ma tics: material issues terial issues Mic Microscale oscale structur cture ( e (20 0 μ m) m) Discontin Discontinuous uous Nano-par Nano-particles les, nano-f nano-fibers Modeling Modeling Continuous mec Contin uous mechanics anics Issues Iss Contin Continuous mec uous mechanics anics pr proper operti ties es Interpreta Inter tation tion Calibr Calibration tion Ca Cath thodes Separat rators rs Anod Anodes SMAI ‐ HPC 22/03/2011 28
Application to Battery Modelling Structur ctural la al layout: ut: shaping: bending, stretching (?), drawing (?) 200 layers 4mm 20 µm plastic covers clamping: pressure Requirements: Re Str Stresses sses and str and strains in eac ins in each h la layer Result : 200 sult : 200 elements/mm elements/mm A 20cm x 20cm ensemb A 20cm x 20cm ensemble le giv gives s 32 Million 32 Million elements elements SMAI ‐ HPC 22/03/2011 29
Ultimate goal: be predictive at the system level EC project HUMOS demonstrator of real car crashes HUMOS Bullet car speed = 48 km/h Target car speed = 24 km/h HUMOS 50% male model: Driver of target car Hands attached on steering wheel Right foot on breaking pedal Left foot on the floor Restraint devices: belts Injury assessment SMAI ‐ HPC 22/03/2011 30
Marine applications are the next frontier Motion in Waves: Regular head waves of 8 meter height and a wave length of 294 meter 22/03/2011 SMAI ‐ HPC
• What are the answers ? • What needs to be assessed / improved ? 22/03/2011 SMAI ‐ HPC 32
Multiscale ??? 22/03/2011 SMAI ‐ HPC
Parallelisation ??? (DMP etc.) 4 3.5 3 2.5 Time (Hours) 2 Series1 1.5 1 0.5 0 0 20 40 60 80 100 120 140 Number of CPUs 8 processors : a bit 128 processors : less more than 3.5 hours than half hour 22/03/2011 SMAI ‐ HPC
Infrastructure ??? 1 ms 22/03/2011 SMAI ‐ HPC
GPUs ??? For FE and mesheless !!! 22/03/2011 SMAI ‐ HPC 36
Methodology ??? FE versus meshless for fracture or XFEM etc. 22/03/2011 SMAI ‐ HPC 37
Ultimate goal: be predictive at the system level EC project HUMOS demonstrator of real car crashes HUMOS Mid ‐ sagittal section SMAI ‐ HPC 22/03/2011 38
Towards virtual prototyping at large scale … Courtesy of EADS ‐ CASA 39
…and "full system" predictability Model 1 Model 2 32 processes 128 processes 22/03/2011 SMAI ‐ HPC 40
22/03/2011 SMAI ‐ HPC 41
Recommend
More recommend
