An integrated probability-based approach for multiple response surface optimization.

摘要

Nearly all real life systems have multiple quality characteristics where individual modeling and optimization approaches can not provide a balanced compromising solution. Since performance, cost, schedule, and consistency remain the basics of any design process, design configurations are expected to meet several conflicting requirements at the same time. Correlation between responses and model parameter uncertainty demands extra scrutiny and prevents practitioners from studying responses in isolation. Like any other multi-objective problem, multi-response optimization problem requires trade-offs and compromises, which in turn makes the available algorithms difficult to generalize for all design problems. Although multiple modeling and optimization approaches have been highly utilized in different industries, and several software applications are available, there is no perfect solution to date and this is likely to remain so in the future. Therefore, problem specific structure, diversity, and the complexity of the available approaches require careful consideration by the quality engineers in their applications.;The purpose of this dissertation is to suggest strategies in order to improve the quality of processes and products with multiple quality characteristics. An integrated probability-based approach will be applied in the modeling and optimization of the problem, which will utilize strengths of probability-based and desirability approaches. A conformance probability metric is the most commonly used optimization criterion for probability-based approaches and it will be shown that particularly when conformance probability is high it can prematurely stop the search process and give biased solutions in mean response values. Another concern is when the number of responses increases a feasible solution set may not exist due to the response constraints. Therefore, penalization of infeasible solutions can help to identify near feasible solutions and also help decision makers articulate their preference information efficiently in order to find compromising solutions.;The proposed approach is coded in MATLAB by the help of readily available tools in the MATLAB Toolbox. Several cases from published literature are implemented and simulations are conducted to show the quality of proposed and existing methodologies. The results showed that, operating conditions obtained by the proposed approach are always superior in terms of mean targets, and almost equally good in terms of conformance probability.