Predictive maintenance on industrial sensor data
Anonymised engagement: TechieYan built a predictive maintenance pipeline on multi-sensor shop-floor data — reducing unplanned downtime and false alarms for a manufacturing operator.
Stack: Python · XGBoost · LSTM · MLflow · Grafana · MQTT · Edge gateway
Challenge
Legacy PLCs and heterogeneous sensor formats produced noisy time-series data. The client needed models that operators would trust — not black-box alerts.
Solution
TechieYan engineered a feature store on tabular + vibration data, trained gradient-boosted and LSTM ensembles, and deployed inference on an edge gateway with Grafana alerting tied to maintenance workflows.
Outcomes
- Documented 35% reduction in false-positive maintenance alerts
- Sub-200ms inference on edge gateway hardware
- Reproducible training pipeline with MLflow model registry
- Operator dashboard integrated with existing MES exports
Want similar results for your programme?
Discuss your project →