Structured neural network models to improve robust design solutions

Abstract

The primary objective of robust design (RD) is to identify the ranges of input factors by reducing the process bias (i.e., the deviation between the mean and the desirable target value) and variance to be as small as possible to improve product quality in the manufacturing process offline stage. Well estimated models for the process mean and variance (or standard deviation) in RD have traditionally involved several error assumptions. However, the functional relationships between input and out output factors can be well estimated without these error assumptions by using a neural network (NN). In this paper, we therefore propose an NN-based estimation method as a RD modeling approach. The modeling method based on a feedback NN approach is first integrated in the RD response functions estimation. Two new feedback NN structures are then proposed. Next, the existing recurrent NNs (i.e., Jordan-type and Elman-type NNs) and the proposed feedback NN approaches are suggested as an alternative RD modeling method. Simulation studies are also performed to verify the potential of the proposed new NN-based estimation method. A case study is conducted to determine the effectiveness of the proposed method. Comparative studies between the proposed NN structure-based methods and existing methods (i.e., conventional NN structures, statistical LSM, IP, and RSM) were conducted. The results of these particular simulation and case studies results represents that the proposed approaches may provide slightly better solutions.