针对代理模型优化技术在结构耐撞性优化中的应用问题,开展了径向基函数和响应面的预测精度检测和适用性研究,并针对径向基函数中散布常数取值机制不明确的问题进行数值分析,提出基于优化计算的选取策略。具体实施为,先使用遗传算法(Genetic algorithm,GA)直接进行搜寻优化,得到的结果可认为是最优结果,并作为代理模型优化方法的参照;通过实例证实代理模型精度检测的必要性,并检验均匀试验设计、中心复合试验设计、析因试验设计在碰撞问题中的适用性;在构造径向基函数时将散布常数设为变量,再以几个随机样点处的数值结果为目标进行模型修正,得到修正的散布常数,代入构造的径向基函数中。结果表明该方案得到的散布常数能够使得径向基函数更好地表征结构耐撞性问题。
The prediction accuracy and applicability of the radial basis function (RBF) and the response surface method (RSM) are studied, and an optimization idea is put forward to solve the problem of selecting an applicable shaping parameter of the RBF model when the surrogate model technology is employed in structure crashworthiness optimization. The genetic algorithm (GA) is used to optimize the crashing energy absorption of a thin-walled structure and find the global optimum which can be viewed as the target of the surrogate model. The necessity of testing surrogate model’s prediction accuracy via numerical examples is demonstrated and the applicability of several methods of design of experiments is tested. Thaping parameter of RBF model is treated as a variable, and a few random sample points are used to update radial basis function model suitable. The results show that the updated radial basis function model will be more suitable for structure crashworthiness problem.