An Effective Simulation-Based Design Optimization Algorithm Using Sequential Regularized Multiquadric Regression

This paper presents an effective simulation-based optimization algorithm that can be used for optimal design of large scale, computationally expensive systems. The initial response surface model of the proposed method is constructed using the Regularized Multiquadric (RMQ) regression based upon the simulation results at a set of the sampling points that has the uniform-space filling property. The quasi-Monte Carlo (QMC) Sampling using the Halton Sequence is selected for the implementation of the optimization algorithm. The regression analysis is performed on the objective and constraint responses to generate the RMQ models. The optimization process is then performed based on the approximate functions using a Sequential Quadratic Programming (SQP) algorithm to obtain a predicted optimal design satisfied the constraint requirements. The finite element model analysis is evaluated at the predicted design to validate the response attributes (such as structural weight, internal energy etc.). The convergence tolerance is checked in each of the iterations. The DOE matrix is augmented by the designs obtained in both steps. The process is repeated sequentially until the solution converges. The proposed method is demonstrated on a rectangular tube crush optimization problem to improve the energy absorption efficiency. Numerical results show that the proposed approach converged quickly to a feasible design with no constraint violation. The total number of finite element simulations was significantly reduced and the total computational cost was reduced by 60~70% compared to the traditional RSM methods. The proposed method is effective and has the potential to handle large-scale engineering problems with large number of design variables.

並列關鍵字

Response surface method ； Multiquadric regression ； Design optimization

國際替代計量

全文下載

主題瀏覽