Abstract
Abstract
Unexpected malfunctions in industrial rotating machinery frequently result in decreased output and higher operating expenses. To achieve real-time anomaly identification and remaining useful life estimation, this work presents a novel predictive maintenance architecture that combines IoT-based multi-sensor, edge inference, and an interpretable AI model. For thorough data collection, the pro posed system uses an ESP32 board interfaced with the various sensors. The edge computing node is a Raspberry Pi 5, which runs the proposed hybrid Random Forest algorithm for remaining useful life prediction and fault categorisation. With a mean absolute error of 1.58 h, an inference latency of 38 ms per cycle, and an accuracy of 99.0%, F1-Score of 99.3%, and Remaining Useful Life (RUL) prediction with R2 = 0.9995, trained and evaluated on 1200 samples collected from a single induction motor testbed. Email notifications, IoT-based alerting via Blynk, and real-time visualization via a Tkinter dashboard interface prompt a response to crucial circumstances. Decision trans parency was achieved through the use of SHapley Additive exPlanations (SHAP) attribution analysis. The results suggest suitability for Industry 4.0 deployment in small enterprises.