Optimization Design Method of Non-linear Complex System Based on Global Sensitivity Analysis and Dynamic Metamodel

Abstract

Multi-parameter optimization design of complex nonlinear system with high nonlinearity involves huge computer, and the optimal solution in theory is also difficult to obtain. A new method is proposed that combines the global sensitivity analysis with dynamic metamodel. By using the variance-based Sobol global sensitivity analysis method, the complex system model is simplified, and the sensitivity parameters are defined to construct the multi-objective metamodel, which is solved by using the genetic optimization algorithm of NSGA-II to obtain the contemporary optimal solution. In the process of optimization, the metamodel and the searching space are continually updated, and the accuracy of the solutions in the optimal zone is improved gradually till the optimization iterations terminate with the convergence criteria satisfied. This method is used to concept optimization design of a vehicle occupant restraint system and good results are achieved with the head injury criterion(HIC), chest acceleration C3 ms and chest deflection D reduced by 11.3%, 11.8% and 9.4% respectively. The optimization results and the precision of the near optimal solution is better than that of static metamodel and the NSGA-II genetic optimization algorithm, which has proved the validity of the proposed new method.