A new robust design method from an inverse-problem perspective

摘要

This paper develops a new approach to robust design (RD). Current RD techniques, including the Taguchi approach, and enhanced approaches using response surface methodology (RSM), can process limited amounts of statistical and engineering information. The primary objective of this paper is to view RD from Bayesian perspectives by incorporating inverse problem (IP) concepts in order to relax the basic assumptions of the least-squares method. The practical benefits of applying IP to RD is the ability of estimation, by treating each model parameter as a random variable, and the flexibility of forwarding and inversing the estimated model. A numerical example is shown, and a comparative study, using conventional response surface robust design (RSRD) models and the proposed IP-based RD (IPRD) models, is present for verification purposes. The numerical example demonstrates that the proposed IPRD models provide significantly better RD solutions than the conventional RSRD models reported in the literature.