Artificial neural network coupled with machining models for accurate prediction of machining response of magnesium alloy AZ31B: a hybrid model

Abstract The significance of forecasting cutting forces in machining has increased in recent years to reduce expensive experimental trials. This research introduces a hybrid model integrating an empirical model with an artificial neural network to anticipate machining responses. Turning experiments on wrought magnesium alloy AZ31B, where cutting and thrust forces during orthogonal machining were recorded, serve as the focus. The hybrid approach combines physically grounded empirical models with feed-forward neural networks to forecast machining forces, bridging disparities between theoretical machining models and practical operating conditions. The artificial neural network (ANN) calculates the discrepancy between the empirical model and the resulting machining forces. The outcomes of this hybrid model exhibit remarkable accuracy in predicting cutting forces. Specifically, the hybrid model predicts experimentally obtained cutting and thrust forces with Mean Absolute Percentage Error (MAPE) values of 1.80% and 0.88%, respectively, in contrast to MAPE values of 4.043% and 1.79% achieved by the standalone feed-forward network model, and 11.355% and 9.623% achieved by the standalone least square regression model in predicting AZ31B cutting and thrust orthogonal machining forces. In addition to a 98.99 and 99.07% R-squared (R 2 ) achieved using the hybrid model for predicting experimentally collected cutting and thrust forces respectively. This novel residual-learning hybrid empirical ANN framework applied AZ31B alloy machining, provides both physical interpretation and superior predictive accuracy over existing standalone models.