Bi-Objective predictive modelling and optimization in micro-electrical discharge machining process using an artificially intelligent system

Electrical Discharge Machining (EDM) is a highly complex, non-linear system whose process modelling using normal techniques such as regression is not much favorable. Hence, soft computing technique such as artificial neural network (ANN), is applied for modelling. Numerous issues are encountered in conventional machining of Nimonic 80A and hence EDM is one of the favorable techniques for machining. Based on obtained value of mean square error, rigorous checks were done to obtain the best performance and it was found that ANN model trained by employing gradient descent with momentum and adaptive learning rate (GDX) algorithm outperformed others due to its exhaustive training and fine tuning. Reasonable accuracy of the ANN model was indicated by its prediction results. Considering the conflicting nature of the objectives, an evolutionary algorithm, nondominated sorting genetic algorithm-II (NSGA-II) is utilized for performing multi-objective optimization. The distance of the non-dominated pareto optimal solutions from the ideal point, is proposed as a solution for selection of optimum point. Confirmatory experiments were carried out and the error obtained was under acceptable limits. Scanning Electron Microscopy (SEM) images indicated that at the optimal setting, considerable finer hole quality, lower value of overcut, less dense re-cast layer and less dense heat affected zone (HAZ), were seen. This efficient intelligent method may serve as a decision support system through its improved prediction accuracy with less dependency on the experimental data and simultaneous optimization of the process conditions.