Applying Hierarchical Genetic Algorithm Based Neural Network and Multiple Objective Evolutionary Algorithm to Optimize Parameter Design with Dynamic Characteristics

Many soft computing techniques were used to resolve Taguchi's parameter design problems. These methods consist of two major steps where neural networks are first adopted to find the functional relationship between the desired responses and control factor values and then simulated annealing or genetic algorithm is applied to determine an optimal combination of control factors. However, neural networks tend to trap the error function in a local minimum when one tries to find the parameters of the network. Besides, the sensitivity measure and variability measure need to be optimized simultaneously in a dynamic system. In this paper, we integrate a hierarchical genetic algorithm (HGA) and a multiple objective evolutionary algorithm (MOEA) to optimize the dynamic parameter design problem. The proposed method applies a HGA based neural network to derive the relationship between the input factors and corresponding outputs, β and SN ratio. Then a MOEA is applied to obtain the non-dominated solution of predicted SN ratio and β. Finally, in the confirmation phase, confirmation experiments are conducted to determine the best parameter setting. An industry case of injection molding process is demonstrated to show the effectiveness and its applicability to other industries.