汽车碰撞安全基础 (17).pdf

2010 中国汽车安全技术国际研讨会 -1-Establishment of Adaptive Occupant Restraint System Optimization Platform Yi Huang,Cong Wang,Yong Xia,Qing Zhou1 2 1.State Key Laboratory of Automotive Safety and Energy,Beijing,100084,China 2.Department of Automotive Engineering,Tsinghua University,Beijing,100084,China y- Abstract:A versatile optimization platform for adaptive restraint system investigation was developed.The platform used MADYMO/Scaler to generate occupant models with different statures.Scaling method was used to obtain injury assessment reference values(IARV)which were used as injury thresholds for the occupant model.A driver side adaptive occupant restraint system model with 9 variables was built as a MADYMO/Exchange project,enabling completely automatic model adjustment and generation.The optimization process was controlled by modeFRONTIER.Including 2 impact velocities and 5 occupant models,10 crash scenarios were analyzed.Genetic algorithm was used to obtain optimal restraint configuration for each scenario.The total computation time for each scenario was less than 30 hours.The robustness of the optimal restraint configurations was inspected.Keywords:Occupant Restraint System,Optimization,Adaptive 1.Introduction The traditional occupant restraint systems(ORS)are designed and optimized according to very limited themes of real-life accidents,which are mostly defined in the related regulations.Limited attention was paid to the variations in the traffic accidents,for example,the variations in crash severity and occupant characteristic.Researches showed that these variations have big influence on the performance of the restraint system1.The future trend of occupant restraint system is to make it adaptive to different crash situations and occupants involved,to achieve tailored protection 2.The adaptive occupant restraint system(AORS)is highly configurable and includes many parameters.Consequently,large amount of computer simulations are needed to investigate the relationship between system performance and parameters.In this study,a platform with the features of rapid computer simulation and automatic model adjustment was developed,so that modern optimization methods can be utilized in this platform to investigate the optimal restraint system configurations for different crash scenarios.2.Occupant Model and Injury Assessment Reference Value To study the influence caused by the variations of occupant,occupant models with different sizes should be used in simulations.Existent occupant models used in regulation related tests cannot fulfill this requirement,because only three versions of occupant models with different statures are available for frontal impact(Hybrid III 5th percentile female,50th percentile male and 95th percentile male).To obtain occupant models with various sizes,scaling method should be adopted.The scaling method has already been applied in the development of the Hybrid III 5th percentile female and 95th percentile male dummy models3 4 5.In this study,occupant models with different heights were generated with the MADYMO/Scaler software,which uses the database Generator of Body(GEBOD)to obtain anthropometric data6.Figure 1 shows the height distribution of the GEBOD database.A good coverage of the population was obtained by selecting the following five different heights:5th%ile Female,1.52 m 2010 中国汽车安全技术国际研讨会 -2-50th%ile Female,1.62 m 95th%ile Female,1.72 m 50th%ile Male,1.77 m 95th%ile Male,1.88 m Figure 2 shows the scaled occupant models alongside with the standard Hybrid III 50th percentile male dummy model.0204060801001.41.51.61.71.81.92Percentile%Height m FemaleMale Figure 1.GEBOD Height Distribution Figure 2.Scaled Occupant Models The Injury Assessment Reference Value(IARV)should be scaled before being applied to the scaled occupant models.The scaling was based on the anthropometric data of the occupants.The reference model was Hybrid III 50th percentile male dummy model.The IARV for the reference model comes from FMVSS 208 regulation.Table 1 shows the scaled IARV for the 5 occupant models 7 8.Hybrid III 50th%ile 5th%ile Female 50th%ileFemale 95th%ileFemale 50th%ileMale 95th%ile Male 2010 中国汽车安全技术国际研讨会 -3-Table 1.Scaled IARV Injury Value IARV Hybrid III50th%ile 5th%ileFemale50th%ileFemale95th%ileFemale50th%ile Male 95th%ileMale HIC36 1000 1123 1090 1059 993 972 3ms Chest Acceleration g 60 72.0 69.1 66.5 60.5 58.5 Chest Compression mm 63 55.5 57.3 59.0 59.4 61.0 Flexion Moment in Neck Nm 310 217.4 235.3 254.2 345.2 363.3 Extension Moment in Neck Nm 135 94.7 102.5 110.7 150.3 158.2 Tension Force in Neck N 6806 5371.4 5662.9 5962.0 7311.7 7565.6 Compression Force in Neck N 6160 4861.6 5125.4 5396.1 6617.7 6847.5 Peak Tension in Neck N 4170 3291.0 3469.6 3652.9 4479.8 4635.4 Peak Compression in Neck N 4000 3156.9 3328.2 3504.0 4297.2 4446.4 Femur Axial Force N 10000 6866.6 7385.1 7922.3 8351.4 8886.8 3.Sled Model The model representing the driver side restraint system was modeled on the basis of a mid-class vehicle(Figure 3).Several adaptive components were built into the restraint system,including seatbelt D-ring,seatbelt pretensioner,two-level load limiter and airbag.The variables of the components are shown in Table 2.Figure 3.Sled Model Table 2.Design Space Variable Range/ValueSeatbelt D-ring height 0.60.9 m Pretensioner trigger time 10 ms Pretension speed 515 m/s Pull-in stroke of pretensioner 100250 mmLoad limiting level 1 of load limiter37 kN Load limiting level 2 of load limiter16 kN Shift stroke of load limiter 0 mm Trigger time of airbag 10 ms Massflow scaler of airbag inflator 0.651.35 In this study,two crash speeds were adopted(40,56 km/h)to represent the low and high crash severities(Figure 4).The 56 km/h crash pulse comes from an actual US NCAP crash test.The 40 km/h crash pulse is scaled from the 56 km/h one.The whole model was established as an MADYMO/Exchange project,which serves as a model generator.Interfaces to adjust the variables of the system were created.With the help of MADYMO/Exchange,occupant model replacement,model positioning and seatbelt fitting can be completed automatically without any manual intervention.Consequently,massive scale optimization work becomes feasible.2010 中国汽车安全技术国际研讨会 -4-1000100200300400500020406080100120Deceleration m/s2 Time ms 56km/h40km/h Figure 4.Crash Pulses 4.Optimization Platform In current study,the optimization software modeFRONTIER was adopted to build the optimization platform,because of its convenient interface with MADYMO/Exchange and strong feature of solving non-linear optimization problem.Figure 5 illustrates the work flow of the optimization platform.Once the variables,optimization strategy,constraints and objective are set,the optimization process will be carried out automatically.The injury values recorded in the occupant models were divided by the corresponding IARV to get unified injury indicators.To give a comprehensive evaluation to the performance of the occupant restraint system,a Weighted Injury Criteria(WIC)was used9 10.WIC assigns different weights to the injury values according to the probability of the injury distribution from the traffic accident statistics.360.2730.070.530.1336_23_2_ijHICa msCcompFlFrWICNHICTha msThCcompThFlThFrTh Figure 5.DOE/Optimization Workflow a Simple Version for Demonstration 2010 中国汽车安全技术国际研讨会 -5-5.Optimization Approach and Result In total,10 crash scenarios(5 occupant models,2 crash pulses)were analyzed.The relationship between system variables and outputs of injury value were investigated.The initial design space was sampled with the Sobol method.Uniform sample distribution over the design space was achieved,which is very suitable for heuristic optimization algorithms11 12.The genetic algorithm:MOGA-II provided by modeFRONTIER was adopted here to solve the problem13.Each generation had a population of 20.The whole process evolved at least 50 generations.The objective of the optimization was to minimize WIC for each crash scenario.The constraint was that the unified injury indicators should be lower than 0.8.The results showed that this approach had a relatively high efficiency in obtaining an optimal point(Figure 6).Figure 6.Optimization Process with MOGA-II Algorithm The values for each variable were recorded during the whole optimization process to calculate the mean value and standard deviation.The mean value standard deviation formed the optimal variable range for the crash scenario.To ensure the robustness of the optimal configurations obtained here,a validation process was carried out.The optimal variable range was treated as new design space for robustness inspection.200 points were sampled from the design space with Sobol algorithm.Variations in responses were examined.The results showed a small bandwidth of responses,though some responses exceeded the constraints defined in the previous process because of the noise of this highly nonlinear system.Figure 7 shows the obtained optimal restraint system configuration for each crash scenario.The interval between the upper and lower boundaries of the optimal range showed the sensitivity of the variable.For example,the optimal ranges of load limiting level 2 were greater than those of load limiting level 1.It implied that the protection performance was more sensitive to the change in load limiting level 1.Consequently,level 1 should be paid more attention to.Considerable differences in the configurations for cases with different combinations of occupant model and crash speed were observed.The initial kinetic energy,which was dominated by crash speed,had big influence on the optimal configurations.Higher crash speed leads to higher initial kinetic energy,which imposed greater loading on the restraint system.As shown in Figure 7(b)and(c),the load limiting levels and the pretensioner pull-in amounts for cases under 56 km/h crash speed were significantly greater,which were very close to their upper limits.It was attributable to the overloading to the seatbelt caused by large initial kinetic energy.Large amount of energy was absorbed within the pretensioning phase,preventing the head using up available travel distance at a high velocity relative to the interior.2010 中国汽车安全技术国际研讨会 -6-Otherwise,the airbag would be bottomed-out or a much stiffer airbag would be needed.Either of these two cases was not favorable for preventing head injury.Figure 7(g)shows that a slightly stiffer airbag was needed for a case under higher crash speed to prevent the head from hitting on the steering wheel.A large occupant also applied greater load to the restraint system.For the cases under the crash speed of 40 km/h,larger load limiting level 1 was needed for a larger occupant.Besides,the load-limiter shift stroke increased as larger occupant model was used.It means that the seatbelt absorbed more energy before the load-limiter switched to the second load limiting level.It should be noticeable that the position of the D-ring changed in different crash scenarios.Lower D-ring positions were favorable for larger occupant models,which was more obvious in the cases under higher crash speed(Figure 7(a).Low D-ring position helps reducing the initial effective length of the seatbelt,making the pretensioning more efficient.0.00.20.40.60.81.05%F50%F95%F50%M95%MPositionofDringinZDirectionmOccupantCategory40km/h56km/h0501001502002503005%F50%F95%F50%M95%MPretensionerPullinAmountmmOccupantCategory40km/h56km/h(a)Position of D-ring in Z Direction(b)Pretensioner Pull-in Amount 0100020003000400050006000700080005%F50%F95%F50%M95%MLoadLimitingLevel1NOccupantCategory40km/h56km/h0200400600800100012001400160018005%F50%F95%F50%M95%MLoadLimitingLevel2NOccupantCategory40km/h56km/h(c)Load Limiting Level 1(d)Load Limiting Level 2 0204060801001205%F50%F95%F50%M95%MLoadlimiterShiftStrokemmOccupantCategory40km/h56km/h02468101214165%F50%F95%F50%M95%MAirbagFiretimemsOccupantCategory40km/h56km/h(e)Load-limiter Shift Stroke(f)Airbag Firetime 2010 中国汽车安全技术国际研讨会 -7-00.20.40.60.815%F50%F95%F50%M95%MAirbagMassflowScalerOccupantCategory40km/h56km/h(g)Airbag Massflow Scaler Figure 7 Optimal Restraint System Configurations 6.Discussion An optimization platform was built to investigate adaptive restraint system.Occupant models generated with scaling method were adopted to represent occupant population with more precise anthropometry.The platform was built with MADYMO/Exchange,with the feature of automatic model adjustment and generation,enabling large scale optimization with the help of modern optimization tool.The highly non-linear problem of restraint system optimization can be solved with modern heuristic optimization algorithm efficiently.In this study,the genetic algorithm(MOGA-II)was proved to have good performance in solving this problem.For each crash scenario,the optimization process including 1000 simulations(20 samples for each generation and 50 generations in total)can be completed in less than 30 hours(with a modern personal desktop computer,2 jobs in parallel).The optimization results showed that different restraint system configurations were needed under different crash scenarios to lower injuries,implying that adaptive restraint system was beneficial.The optimal range for each variable was robust and reliable.It can be used as guidance for the design of modern restraint system with adaptive feature.The platform was modularized and could be easily extended to include more variables or switched to be suitable for different crash types.The next step would be to make full use of this platform to investigate a complex restraint system with more variables to develop some new concept/configuration of adaptive occupant restraint system.Acknowledgments Initiation of this work was based on the discussion with Toyota Motor Corporation.The authors would like to express thanks to the sponsorship from the Toyota-Tsinghua Technical Center joint project.Thanks also to National Science Foundation of China(NSFC).Part of this work is supported by the NSFC project(project number:50975156).We also appreciate the educational license of MADYMO provided by TASS.References 1 Happee,R.&Haaster,R.V.Optimization of vehicle passive safety for occupants with varying anthropometry Proceedings of the 16th International Technical Conference on the Enhanced Safety of Vehicles,1998 2 Cuerden,R.;Hill,J.;Kirk,A.&Mackay,M.The potential effectiveness of adaptive restraints Proceedings of the IRCOBI Conference,International Research Council on the Biomechanics of 2010 中国汽车安全技术国际研讨会 -8-Impact,2001,323-336 3 Mertz,H.J.;Irwin,A.L.;Melvin,J.W.;Stalnaker,R.L.&Beebe,M.S.Size,weight and biomechanical impact response requirements for adult size small female and large male dummies SAE International Congress and Exposition,SAE International,Warrendale,Pennsylvania,USA,1989 4 Mertz,H.J.A procedure for normalizing impact response data SAE Government Industry Meeting and Exposition,1984 5 Mertz,H.J.;Irwin,A.L.&Prasad,Priya Biomechanical and scaling bases f