Industrial IoT: Predicting Failures Before They Happen
Project Overview
SmartGrid Italia manages a fleet of 500+ industrial power generators distributed across remote locations. Unplanned downtime was costing them over €1.2M annually in emergency repairs and service level agreement (SLA) penalties.
The Solution
We developed an end-to-end IoT platform that ingests telemetry data (vibration, temperature, voltage) and uses Machine Learning to predict component failures up to 7 days in advance.
Predictive Engine
The core innovation was a Random Forest classification model trained on 3 years of historical maintenance logs. The model runs at the edge (on-device) to minimize bandwidth costs, transmitting only anomalies and health heartbeats to the cloud.
Digital Twin Visualization
The web platform features Mapbox integration for geospatial tracking and a Three.js 3D viewer that allows engineers to inspect the virtual "Digital Twin" of each generator to pinpoint problematic components remotely.
Implementation Details
- Edge: Python script running on Raspberry Pi industrial modules.
- Protocol: MQTT over cellular networks (NB-IoT).
- Data Lake: Azure Data Explorer for petabyte-scale time-series storage.
- ML Ops: MLflow for model versioning and deployment.
- App: Next.js dashboard with server-side rendering for fast initial loads.
Business Impact
- 40% reduction in unplanned downtime.
- 25% decrease in maintenance costs by switching from scheduled to condition-based maintenance.
- Remote Diagnostics: 30% of issues resolved without a truck roll.