Machine learning and statistical approach in modeling and optimization of surface roughness in wire electrical discharge machining

Abstract

The current work implements machine learning techniques such as artificial neural network (ANN), support vector machine (SVM), and genetic algorithm (GA) to model and optimize the surface roughness during wire electrical discharge machining (WEDM) of Inconel 718. For this, surface roughness values were obtained from real-time WEDM experiments conducted under the different levels of control factors such as pulse on time, pulse off time, peak current, servo voltage, and wire feed rate. The optimum ANN model architecture was identified as 5-10-10-1 and SVM parameters were tuned with the help of the grid search technique. The ANN and SVM models’ predictions were compared with response surface methodology (RSM) predictions and performance was evaluated based on correlation coefficient (R-value) between experimental and model predictions. The SVM predictions were accurate among all the models studied, as determined from the R-value of 0.99998 with experimental results and the least mean absolute percentage error (MAPE) of 0.0347%. Further, the GA approach was implemented using the developed RSM equation as the fitness function and led to 61.31% improvement in the surface roughness. The proposed SVM and GA approach would help quick and high accurate prediction and optimization of surface roughness during WEDM of Inconel 718.