Developing a Flexible Methodology for Modeling and Solving Multiple Response Optimization Problems

Simultaneous optimization of multiple quality characteristics (responses) of a product or process is required in many real-world problems. Multiresponse optimization (MRO) techniques tries to solve such problems; the ultimate objective of which is to adjust control factors that provides most desired values for the responses. Regression techniques are most commonly used methods to identify and estimate relationships between control variables and responses. Due to the industrial advances and hence the complexity of processes and systems, many relationships between input variables and quality characteristics have become much more complex. In such circumstances, classic regression techniques encounter difficulties to create a well-fitted model which can be easily optimized. The alternative approach proposed in this study is a regression tree method called CART, which is a data mining method. Since the output of CART consists of several if-then terms, NSGA-II algorithm was considered to solve the model and achieve the optimal solutions. Finally, we evaluate performance of the proposed method with a real data set about modeling and improvement of automotive engines.