A Machine Learning Approach for Predicting Wire Breakage in WEDM with Edge Deployment Considerations

One of the key challenges in wire electrical discharge machining (WEDM) is the occurrence of wire breakage, which interrupts machining continuity and degrades product quality. To address this challenge, recent studies have increasingly applied machine learning (ML) techniques to analyze manufacturing parameters and predict the possibility of wire breakage. This study proposes a lightweight neural network model for wire breakage prediction, using gap voltage (GV), feed rate (FR), and water resistance (WR) as input features. The model is designed to predict whether wire breakage will occur within the next 10 s, achieving a test prediction accuracy of 83%. In response to the growing demand for real-time process monitoring in intelligent manufacturing, the model is designed with edge computing deployment in mind. Compared to cloud-based systems, edge computing offers reduced latency and increased responsiveness but poses strict memory constraints, computation power, and energy consumption. The model was optimized with practical deployment guidelines to ensure edge compatibility, keeping the model size under 100 KB and inference latency below 1ms. The resulting model balances computational efficiency and predictive performance, making it well-suited for integration into intelligent WEDM systems. This research contributes a viable artificial intelligence (AI) solution for predictive maintenance and real-time edge-device decision-making in modern manufacturing.