A real-time IoT-based predictive maintenance system for industrial pumps. Sensor data streams live to Firebase, an ML model predicts failure probability and remaining useful life, and a Flask dashboard visualises everything as it happens — no page refresh needed.
Pump Predictive Maintenance is a full-stack ML pipeline built around a real industrial problem: knowing when a machine is about to fail before it actually does. Rather than batch predictions on historical data, it operates continuously — a sensor simulator pushes live readings every 2 seconds, a listener runs them through a trained scikit-learn classifier, and a browser dashboard reflects the results instantly without any page refresh.
Built during an industry internship at IOCL Guwahati, this project bridges the gap between ML experimentation and operational tooling — turning a trained model into something an engineer could actually watch and act on.
The core idea was to close the loop between data collection and decision-making as tightly as possible. Every component — the Firebase pipeline, the ML listener, the live Flask dashboard — was designed to keep latency low and the output immediately readable. The system outputs three things per reading: a failure probability score, a health status (HEALTHY / WARNING / CRITICAL), and an estimated remaining useful life in hours, giving operators exactly what they need to act before something breaks.
