Predictive Maintenance Platform for Industrial Equipment

Instrumented 14 production lines with 280 vibration, temperature, and current sensors, transmitting data via MQTT to a central broker.

Built a time-series anomaly detection model using LSTM networks trained on 2 years of historical sensor data and maintenance logs.

Developed a maintenance scheduling engine that generates work orders 72 hours before predicted failures, integrated with the client's SAP ERP.

Created a plant operations dashboard in Grafana showing real-time equipment health scores, predicted failure probabilities, and maintenance backlog.

Conducted a 6-week shadow mode validation where predictions were logged but not acted upon, confirming 91% prediction accuracy before go-live.