An intelligent approach for multi-response optimization: a case study of non-traditional machining process

Abstract

The present work proposes an intelligent approach to solve multi-response optimization problem in electrical discharge machining of AISI D2 using response surface methodology (RSM) combined with optimization techniques. Four process parameters (factors) such as discharge current (Ip), pulse-on-time (Ton), duty factor (τ) and flushing pressure (Fp) and four important responses like material removal rate (MRR), tool wear rate (TWR), surface roughness (Ra) and circularity (r1/r2) of machined component are considered in this study. A Box-Behnken RSM design is used to collect experimental data and develop empirical models relating input parameters and responses. Genetic algorithm (GA), an efficient search technique, is used to obtain the optimal setting for desired responses. It is to be noted that there is no single optimal setting which will produce best performance satisfying all the responses. In industries, to solve such problems, managers frequently depend on their past experience and judgement. Human intervention causes uncertainties present in the decision making process gleaned into solution methodology resulting in inferior solutions. Fuzzy inference system has been a viable option to address multiple response problems considering uncertainties and impreciseness caused during judgement process and experimental data collection. However, choosing right kind of membership functions and development of fuzzy rule base happen to be cumbersome job for the managers. To address this issue, a methodology based on combined neuro-fuzzy system and particle swarm optimization (PSO) is adopted to optimize multiple responses simultaneously. To avoid the conflicting nature of responses, they are first converted to signal-to-noise (S/N) ratio and then normalized. The proposed neuro-fuzzy approach is used to convert the responses into a single equivalent response known as Multi-response Performance Characteristic Index (MPCI). The effect of parameters on MPCI values has been studied in detail and a process model has been developed. Finally, optimal parameter setting is obtained by particle swarm optimization technique. The optimal setting so generated that satisfy all the responses may not be the best one due to aggregation of responses into a single response during neuro-fuzzy stage. In this direction, a multi-objective optimization based on non-dominated sorting genetic algorithm (NSGA) has been adopted to optimize the responses such that a set of mutually dominant solutions are found over a wide range of machining parameters. The proposed optimal settings are validated using thermal-modeling of finite element analysis.