Modified variable return to scale back-propagation neural network robust parameter optimization procedure for multi-quality processes

Abstract

Selecting the optimum process parameter level setting for multi-quality processes is cumbersome. Previous methods were plagued by complex computational search, unrealistic assumptions, ignoring the interrelationship between responses and failure to select optimum process parameter level settings. The methods of variable return to scale (VRS) back-propagation neural network (BPNN) previously adopted were limited by the use of weak models, poor discriminatory tendency and an inability to select the optimum parameter level setting. This study applied a modified VRS-adequate BPNN topology model in the robust parameter procedure to solve this problem. Here, standard VRS models are allowed to self-assess, leading to partitioning. The upper bound of the free variable of the VRS model is restricted and the VRS penalization coefficient is adopted to determine the optimum process parameter level setting. The effectiveness of the proposed model measured by the total anticipated improvement yielded the highest total improvement over the existing methods.