Machine Learning–Based Anomaly Detection in Edm Process Signalsin E-Commerce
Abstract
Neha Dinesh Rangdal, Pooja Honna, Sujith Chand, Pravardhan Prasad, Priyanshu Thakur and Gajanan M Naik
Electrical Discharge Machining (EDM) stability is crucial but often threatened by anomalies like short-circuits and arcing, which degrade surface quality. To solve this, we developed a powerful hybrid machine learning framework that analyzes real-time EDM signals (current, voltage, acoustic) using a multi-stage approach: it applies sophisticated signal preprocessing, extracts features via time–frequency methods, uses denoising/variational autoencoders for ro- bust representation learning, and employs a fusion classifier (CNN-LSTM/ Isolation Forest) for detection. This system achieved an F1-score of 0.94 for anomaly detection and high localiza- tion accuracy, proving its practical utility with sub-60 ms latency deployment on an NVIDIA Jetson Xavier NX, making it an immediately viable solution for industry-grade process moni- toring.