Experimental investigations and artificial neural network based prediction of kerf parameters due to wire electro-discharge machining of straight and exponential curve profiles

Wire Electro-Discharge Machining (WEDM) is a widely used machining process for cutting electrically conductive metals using the intense thermal energy of a spark. This study experimentally investigates the kerf characteristics of straight and exponential curve profiles machined on stainless-steel sheets using Computer Numerical Control (CNC) WEDM. Experiments were designed using the Central Composite Design (CCD), and the resulting kerf parameters were evaluated to examine the influence of profile geometry on dimensional accuracy. A feed-forward backpropagation Artificial Neural Network (ANN) with a single hidden layer of eight neurones (4–8–5) was developed in MATLAB R2018a using the Levenberg–Marquardt algorithm to predict the kerf parameters. The ANN demonstrated satisfactory predictive capability, with average percentage errors of 5.45%, 3.18%, 17.04%, 11.67% and 3.82% for average top kerf width (AKW _T ), average bottom kerf width (AKW _B ), average kerf taper (AKT), average top kerf deviation (AKD _T ) and average bottom kerf deviation (AKD _B ), respectively, for the straight profile, and 5.45%, 3.81%, 9.70%, 11.67% and 3.77%, respectively, for the exponential profile. The results further revealed that the exponential profile consistently exhibited higher kerf deviation and taper than the straight profile, highlighting the significant role of profile geometry in governing dimensional accuracy during WEDM. These findings show the applicability of ANN-based modelling for predicting kerf characteristics and provide insight into improving dimensional accuracy in the WEDM of complex curved profiles.