摘要
在不同乘员损伤评价标准下,汽车碰撞安全约束系统优化可能得到不同的匹配参数。基于实验设计的方法选取一定数量的设计点,分别通过Polynomial SVD、Kriging、神经网络算法建立约束系统的代理模型,用随机样本验证模型的拟合精度。选取神经网络代理模型进行遗传算法优化,发现代理模型在一些接触边界上存在失准的情况,因此采用多刚体MADYMO模型分别以多目标优化、加权综合评价指标WIC和C-NCAP评分3种损伤评价标准,通过MOGA-Ⅱ遗传算法计算得到不同标准下的优化结果。仿真结果对比发现:由于约束系统属于多输入变量的多目标优化问题,在有限的计算时间内难以得到具有代表性的Pareto非劣解集,加之高维的解集空间难以直接得出具有工程参考价值的约束系统匹配参数,而采用两种归一化标准则得出了两组不同的匹配参数,体现出WIC以权重为优先级的重要部位重点保护和C-NCAP以各部位伤害值达标为目标的优化特点。
Based on different occupant injury and trauma criteria of car safety,for a restraint system may be got after optimization. The experimental design the design points and surrogate models were constructed with Polynomial SVD,Kriging and neural net.T h e n , the precision of three models were tested by several random samples. The neural net model wasoptimized by genetic algorithm. There are a large discrepancy between surrogates and M A D Y M Omodels on the contact boundary＊ Multi-propose optimization , weighted injury criteria , China-new carassessment program, were discussed, respectively, and results were iterated b y M O G algorithm under three criteria. Comparing the simulation results shows that the optimized restraintsystem through the multi-objective optimization is difficult to reach the non-inferiority Pareto solutionin the limited calculation time. Moreover , it is hard to search the reference values for a restraintsystem in the solution set of high dimensional space. O n the other hand , the different matchingparameters were got based on two normalized standards , which shows that W I C focuses on importantbody parts depending on their weights and C - N C A P aims to reach the injury
theshild if esch part.
作者
胡远志
甘顺
刘西
蒋成约
任立海
HU Yuanzhi, GANShun, LIUXi, JIANG Chengyue, RENLihai(Key Laboratory of Advanced Manufacture Technology for Automobile Parts,Ministry of Education, Chongqing 400054,Chin)
出处
《重庆理工大学学报（自然科学）》
CAS
北大核心
2018年第5期1-11,共11页
Journal of Chongqing University of Technology：Natural Science
基金
国家自然科学基金资助项目(51405050)
2015年重庆市重点产业共性关键技术创新专项(cstc2015zdcyztzx60010)
2015重庆市基础与前沿研究计划项目(cstc2015jcyj A00048)
