The Study On Global Sensitivity Analysis And Dynamic Metamodel Of Multiple-parameters Nonlinear System

This subject studies is carried out the National Natural Science Fund Project(51275164)" The robust optimization for human injury based on global sensitivityanalysis and metamodel fitting basis function in the vehicle collision".Multiple-parameter nonlinear system has two notable feature, numerousparameter and strong interaction effect between parameters. Tiny change of aparameter can lead to big change in output. Therefore, the optimization design ofmultiple-parameter nonlinear system involves huge workload and is difficult to getthe optimal solution. So that the high efficiency metamodel technology and the globalsensitivity analysis which can screen the important parameters have become the keyresearch content of the optimization design.Based on the privious research work, this paper further studies the globalsensitivity analysis and the metamodel, and the two technologies are applied to thefield of automobile occupant restraint system. The main research contents and resultsare as follows:The fitting ability of different number variables and nonlinear level is measuredamong the response surface model, Kriging model and radial basis function, from thethree aspects that precision, computational efficiency and fitting features.This paper calculates the importance and gets the main effect of each parametersthrough the global sensitivity analysis based on metamodel. A four dimensionalnonlinear test function is used to illustrate the calculation principle and analysismethod of the global sensitivity analysis. At last, the effectiveness of the globalsensitivity analysis based on metamodel is verified.When the model has a lot of design parameters and their value range is very wide,the traditional metamodel cannot fully guarantee the local precision of the wholedesign space and traditional space reduction method is easy to mistakenly deleted theglobal optimal solution of the problem. In order to solve this problem, this paperproposes dynamic metamodel optimization strategy of design space adaptiveadjustment based on main effect information of global sensitivity analysis. Anexample function and a reducer optimization design examples are used to verify thesuperiority of this paper’s optimization strategy.By MADYMO software, the occupant restraint system simulation model of the frontal crash is built and design parameters are selected. Though the global sensitivityanalysis, the strong sensitivity parameters are indentified. The proposed optimizationstrategy is applied to the important parameters’ optimal design of the occupantrestraint system. The optimization results show that the occupant injury is reducedsignificantly and the optimization strategy improves the security of a vehicle occupantrestraint system obviously. The optimization strategy has obvious reference in thefrontal crash of automobile occupant restraint system concept design.