Intelligent Fault Prediction System for Smart Manufacturing using Random Forest and Real-Time Deployment

Unplanned equipment downtime in smart manufacturing causes significant production losses and maintenance costs. Predictive fault detection using machine learning can anticipate failures before they occur, yet many solutions remain confined to offline analysis or lack practical real time deployment. This paper presents an intelligent fault prediction system built around a Random Forest classifier, integrated into a real time monitoring pipeline suitable for Industry 4.0 environments. Vibration, temperature, and acoustic emission data were collected from a CNC milling machine under normal and fault conditions (tool wear, bearing fault, misalignment). A set of 42 statistical and frequency domain features was extracted from sensor windows. The Random Forest model, trained on 12,000 labeled samples, achieved 97.8% accuracy, 0.975 precision, 0.982 recall, and a 0.994 AUC in detecting four operational states. The model was exported using ONNX and deployed on an edge device (NVIDIA Jetson Nano) with a Flask based API and a dashboard for real time alerts. Compared to support vector machine and logistic regression baselines, the Random Forest demonstrated superior robustness and inference latency under 12 ms per prediction, meeting real time constraints. The system successfully predicted faults an average of 18 minutes before critical failure in a live test, demonstrating the viability of lightweight ensemble methods for industrial fault prediction.