Abstract Predicting surface topography in Wire Electrical Discharge Machining (WEDM) remains challenging due to complex electrical and thermal interactions during the process. Traditional methods relying on machining parameters and simplified surface finish indicators often fail to capture the full complexity of surface topography. This study introduces a novel approach that leverages cutting power data, an accessible proxy for discharge parameters, to predict and reconstruct 3D surface topography in WEDM. The method decomposes surface topography into roughness, waviness, and form components using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Sample Entropy (SampEn). Each component is modeled separately: roughness as a random process following a normal distribution, waviness as a uniform sinusoidal function, and form as a cubic B-spline. A multi-task Convolutional Neural Network (CNN) is developed to simultaneously predict these components based on features extracted from cutting power data. Experimental validation was conducted on a 6061 aluminum (AL 6061) workpiece machined under various cutting conditions. The proposed model outperformed benchmark models in both frequency and spatial domain metrics, accurately capturing key surface characteristics such as roughness variations, waviness patterns, and shape deviations. These results demonstrate an enhanced reliability in surface topography predictions. By integrating cutting power data with advanced decomposition technique and AI model, this study provides a framework for real-time surface topography prediction in WEDM, contributing to improved process optimization and control.
Liu et al. (2025) studied this question.