Proposal of the “Total Error Minimization Method” for robust design

Abstract We propose a new approach for the robust design of a product with single or multiple performances/outputs without axioms. When there are no theoretical models, experimental methods (e.g., the popular and robust Taguchi and Nakazawa design methods) with an orthogonal array are important. Meanwhile, if mathematical or physical models are available, other robust design approaches using, for example, a genetic algorithm, are applicable. The proposed approach is compatible with both experimental and theoretical approaches. The approach is proposed through the minimization of the total difference (i.e., error) between the required specification and experimental output for different variables with noise. We applied the approach to an electric circuit problem to obtain the required specification and target cost. The optimal results are the same as those obtained using the Nakazawa method. Additionally, a genetic-algorithm-based design with a proposed metric, namely the optimal design under contentious variables, is presented. The obtained performance/output better matches the required specification when using the proposed approach than when using the discrete approach. Employing the proposed method, we expect a united robust design without an axiom.