Robust Parameter Optimization For Multi-Response Using Response Surface Methodology

This dissertation mainly studies the robustness of the responses to the fluctuation of controllable factors in multi-response robust parameter optimization. Considering the optimization and robustness for multi-response problem simultaneously, this dissertation mainly focuses on obtaining a compromised robust optimum condition on which multiple responses are simultaneously optimized and insensitive to small changes of input variables. Aiming at this objective, several multi-response robust optimization models are constructed based on Response Surface Methodology (RSM). With the introduction of basic theories and traditional methods of response surface methodology and robust parameter design, robust optimization methods for multi-response are proposed in this dissertation as follows.First, the robustness of responses to the fluctuations of input variables is discussed. Considering that input variables are hard to be fixed, and the responses will change corresponding to the fluctuation of input variables in a small region, two measures of robustness are presented, and the technical approach to the robust parameter optimization of multi-response is proposed in this dissertation.Second, with the improvement of traditional desirability function, a new robustness desirability function is proposed to measure the robustness of responses to the fluctuations of controllable variables. Considering the optimization and robustness of a process, a new robustness optimization desirability function is presented with the tradeoff between traditional optimization desirability function and robustness desirability function mentioned above. Case study shows that the proposed method is feasible to solve the robustness problem of multi-response optimization.Third, considering the correlations among several responses for the robust optimization of multi-response problem, corrected generalized distance function and robustness optimization overall generalized distance function are introduced. A compromised robust optimum can be obtained by finding conditions that minimize the improved generalized distance. Cases study shows the feasibility and effectiveness of these methods.Last, with the selection of appropriate robustness matrix, a new robustness function is constructed. Based on process economy and robustness, a new robustness loss function is proposed to reduce cost and improve robustness of process parameter optimization.