Challenge
Fleet operators were reacting to failures rather than preventing them, resulting in costly unplanned downtime and expedited parts procurement.
Solution
- CAN/J1939 telemetry ingestion at 1 Hz per asset.
- Feature engineering: rolling statistics, fault-code frequency, duty-cycle patterns.
- Gradient-boosted survival model predicting failure probability over a 30-day horizon.
- Alert dashboard with maintenance scheduling integration.
Results
- Unplanned downtime: −18%
- Maintenance cost per unit: −11%
- Model precision at 7-day horizon: 84%