Enhancement 10/15/2018 Jason Wang jason@lstc.com Outline - PowerPoint PPT Presentation

Recent Airbag CPM Enhancement 10/15/2018 Jason Wang jason@lstc.com

Outline • Enhancement – *MAT_ADD_AIRBAG_POROSITY_LEAKAGE – Inflator orifice area – Deflection coefficient defined by a FPRIC vs pressure curve – Vent options (*DEFINE_CPM_VENT) • Internal vent with cone angle – R9 scalability and repeatability – CPM performance improvement – Special decomposition • Conclusions

*MAT_ADD_AIRBAG_POROSITY_LEAKAGE *MAT_ADD_AIRBAG_POROSITY_LEAKAGE $ mid flc fac ela fvopt x0 x1 • Similar options as the 3 rd card of *MAT_34 • If this is used with *MAT_34, the options in this card has higher priority than *MAT_34 • It can be used with any *MAT to define leakage • fvopt=7 or 8 for CPM (pressure vs. velocity)

*MAT_ADD_AIRBAG_POROSITY_LEAKAGE Porous velocity unloading Seam leakage, fac for *DEFINE_FUNCTION Loading P max Absolute pressure • “ fac ” can be defined as a factor, a load curve under *MAT_34 and *MAT_ADD • For CPM, “ fac ” can be defined as a *DEFINE_FUNCTION under *MAT_ADD card – User can control Loading/unloading with different porosity velocity curves – Input: part absolute pressure, time Output: velocity – Extended values, upon request

*MAT_ADD_AIRBAG_POROSITY_LEAKAGE *MAT_FABRIC $# mid ro ea eb ec prba prca prcb 19 8.7600E-7 0.300000 0.200000 0.300000 0.200000 0.200000 0.200000 $# gab gbc gca cse el prl lratio damp 0.040000 0.040000 0.040000 1.000000 0.060000 0.350000 0.100000 0.200000 $3.0000000 1.000 -17420 $# aopt flc fac ela lnrc form fvopt tsrfac 3.000000 ------ *MAT_ADD_AIRBAG_POROSITY_LEAKAGE $ mid flc fac ela fvopt x0 x1 19 1.0 -99 7 1.0 *DEFINE_FUNCTION 99 float fac(float p, float time) { float x, y, pa=101325.15, pmax, frcf; pmax=3.0*pa; frcf=471; x = (p*1e9 - pa)/1e6; y=0.013*x*x+0.47*x+1.3; return frcf*y; }

*AIRBAG_PARTICLE *AIRBAG_PARTICLE ….. $Orifice cards $ NID AN i Vd i CA i INFO i • 𝐵𝑂 𝑗𝑜𝑔𝑝 = ∑𝐵𝑂𝑗 • 𝐵𝑂 𝑗 /𝐵𝑂 𝑗𝑜𝑔𝑝 is used to distribute the mass among orifice • AN i > 0, the value is the orifice area • AN i < 0, the abs(AN i) is load curve ID for orifice area vs. time. The mass will be adjusted based on 𝐵𝑂 𝑗 (t) /𝐵𝑂𝑗𝑜𝑔𝑝(𝑢) during run time.

*DEFINE_CPM_VENT Internal vent with cone angle VANG=-1,-1 VANG=-1,-2(local system) VANG=10,30 Default Particles tends to have very uniform space distribution passing the internal vent which loss the “jetting” behavior observed in the tests

*DEFINE_CPM_VENT Internal vent with cone angle This new option greatly helps to improve the correlation between tests and simulations 1. Cone angle is defined by using above keyword card. 2. Additional option VANG=-1 will allow code to adjust the release based on the vent condition H. Ida, M. Aoki, M. Asaoka, K. Ohtani,"A Study of gas flow behavior in airbag deployment simulation",24th International Technical Conference on the Enhanced Safety of Vehicles(ESV). No. 15-0081, 2015.

Benchmark DAB Models ALE CPM: NO VANG CPM: VANG=-1 Estimated gas flow VANG=-1 NEW VANG redirection angles (45 0 ) FUNCTION In some airbags an inner fabric structure is used to redirect the gas flow to inflate the airbag in a certain way.

*DEFINE_CPM_VENT VANG=-2 VANG=-2 VANG=-1 ALE VANG=- 2: user defines a local coordinate system for ‘jet’ to follow .

Scalability and repeatability R9 • Baseline airbag models created by JSOL/Arup for demo/research purposes. – CAB = curtain airbag, DAB = driver’s airbag, PAB = passenger airbag • All models have typical size, shape, inflator & fabric. • All have been developed to be robust (insensitive, repeatable, not prone to error) and inflate with no issues. • All models are tested with different number of cores. 2hrs 7.5hrs 5hrs 30min 1.2hrs 1hr Courtesy of: Richard Taylor, Arup

Scalability and repeatability R9 • The DAB model has two external vents, fabric and seam line porosity, all affected by contact blocking. Despite this results are very similar for all analyses. Press ssur ure Volu lume me Intern ernal al-Ene Energy rgy The slight difference in internal energy is due to different levels of vent contact blocking by different crease patterns. Unblocked ocked Area of Left and Rig ight Vent nts 32cpu 64cp 32cp u u 64cpu 16PP PPN N and 8PPN resu sults s are identica cal

Improvement of execution speed • Major cost for CPM airbag simulation airbag self contact particle to particle contact (p2p) particle to fabric contact (p2f) p2p • nbody collision • equal space to equal nbody p2f • node to surface bucket sort • More efficient neighborhood search • Improvement on communications

Performance – Tank test (64 processors) 25000 R9 R10 20000 15000 10000 5000 0 Total CPM p2f p2p Chamber Pressure Vent rate

Scalability - CAB CAB OpenMP Enabled 12 10 8 6 R9 R10 4 2 0 12 24 48 96 192 96x2

CPU time improvement on CAB 10000 9000 8000 7000 6000 5000 CPM p2f p2p 4000 3000 2000 1000 0 R7.1.3 R9.3.0 R10 R11 Dev

CPU time of CAB by features 25000 Contact CPM Element Processing Elapsed time (Seconds) 20000 15000 10000 5000 0 R7.1.3 R9.3.0 R10 R11 Dev CAB  Performance was measured with 96 processors  CPM is about 3x faster from R7 to R10  Self contact about the same  The overall speed up is about ~20% for bag, ~5% for full car

*CONTROL_MPP_DECOMPOSITION_ARRANGE_PARTS Runtime reduced from 7hrs 5min to 5hrs 53min. 20% faster! • Bags in parallel • Bag self contacts in parallel • Contacts between bag and dummy in parallel • set 16 for the DAB • Set 48 for the PAB DAB & PAB 64cpu, Default DAB & PAB 16/48 Distributed Courtesy of: Richard Taylor, Arup

Improvement of execution speed • New algorithm is set as default method for R10.2(later), R11 and Dev binaries • There is NO input change needed • If one would like to test the old scheme, input needs to be modified • New algorithm is about 3x faster for moderate processor counts (16 -64 cores). It will be even better with large core counts (>64 cores) from improved message passing. • OpenMP enabled • Due to other features in the model, the overall speed up is about 20%.

Conclusions  For simulation with more than 1 bag, please consider distribute each bag, its casing, dummy into particular group of processors  HYBRID enabled - performance improvement on general and OpenMP features  Heat sink/source effects are developed last week  All new features are developed closely with customers and validated with tests. If you have any idea in mind, please share with me.  jason@lstc.com Thank you

Development During Customer’s Visit • Enhancement – Heat transfer between gas and srounding structure – Time dependent Cd_ext – SFFDC can be defined for each bag instead as global variable in *CONTROL_CPM – JT effect considered while switching CPM to UPM

*AIRBAG_PARTICLE *AIRBAG_PARTICLE ….. $Card 5 $ IAIR NGAS NORIF NID1 NID2 NID3 CHM CD_EXT • CD_EXT>0: external drag coefficient • CD_EXT<0: Load curve ID for time dependent drag coefficient

*AIRBAG_PARTICLE *AIRBAG_PARTICLE ….. $ Heat convection part set card; NPDATA>0 $ SIDH STYPEH HCONV FPROC SDFBLK KP INIP CP • CP: Specific heat for structure part • If specific heat of the structure part is given, the initial temperature of this part has the initial air temperature and the part mass (M) is automatically calculated. Heat transfer between gas and structure is based on the convection equation and then the new part temperature is updated. The time history of part temperature is stored in the abstat_cpm database under the field of part_temp . • 𝐹 = 𝐵𝑓𝑠𝑏 × 𝐼 eff × 𝑈 gas − 𝑈 part (𝑢 − 1) • 𝑈 part 𝑢 = 𝑈 part 𝑢 − 1 + 𝐹/(𝑁 × 𝐷𝑄)

Joule-Thomson Effect intermolecular forces  The Ideal Gas Law assumes the existence of a gas with no volume and no intermolecular interactions with other molecules.  The approximation is often good enough to describe real gases, except at very high pressures and very low temperatures.  The intermolecular forces play a greater role in determining the properties of the real gas at very high pressures and very low temperatures.  Joule-Thomson effect occurs when a real gas expands from high to low pressure at constant enthalpy.

Joule-Thomson Effect CPM includes negative Joule-Thomson effect which may have significant effect for pressurized Hydrogen, Helium and Neon. This feature is extended for CPM to UPM switching. *DEFINE_CPM_GAS_PROPERTIES ID Xmm C p0 C p1 C p2 C p3 C p4 µ t0 µ t1 µ t2 µ t3 µ t4 *DEFINE_CPM_VENT ID C23 LCTC23 LCPC23 ENH_V PPOP C23UP JT IDS1 IDS2