Surface roughness decides how a machined part behaves in service. It affects fatigue life, wear, sealing, friction and, in many cases, whether a part is accepted at inspection at all. Non-traditional machining (NTM) processes such as electrical discharge machining, wire EDM, electrochemical machining, laser beam machining, abrasive water jet machining and ultrasonic machining are now routine choices for hard, brittle and geometrically complex parts. Their surface texture, however, comes from thermal, chemical or erosive mechanisms that are stochastic and strongly coupled, so it is hard to predict with classical physics-based models or with simple regression equations. Over the last two decades, machine learning (ML) has become a practical alternative. Artificial neural networks, support vector regression, tree-based ensembles, fuzzy and neuro-fuzzy systems, and more recently deep learning have all been used to map process parameters and sensor signals to roughness values.

This paper reviews that body of work. It first outlines the main NTM processes and the way roughness arises in each, then summarises the factors that influence it and the traditional prediction approaches (empirical, statistical and analytical). The discussion then turns to the ML techniques most often applied, a survey of representative studies, a side-by-side comparison of the techniques, and the benefits that ML brings. The review also looks honestly at the obstacles: small datasets, limited transferability, poor interpretability and inconsistent reporting. From this analysis, the main research gaps are identified and directions for future work are proposed, including hybrid physics-informed models, transfer learning, explainable AI and real-time in-process prediction.

1.  Keywords

Surface roughness; non-traditional machining; machine learning; artificial neural network; support vector regression; random forest; electrical discharge machining; wire EDM; laser machining; abrasive water jet; predictive modelling; Industry 4.0.