Business Context
MachineryCo, a manufacturing firm with 200+ machines operating across multiple plants, aims to reduce unexpected equipment failures that lead to costly downtime. The engineering team seeks a predictive maintenance model to anticipate failures based on sensor data and operational metrics, thereby improving maintenance scheduling and reducing disruptions.
Dataset
| Feature Group | Count | Examples |
|---|
| Sensor Data | 15 | temperature, vibration, pressure |
| Operational Metrics | 10 | run_time, load, maintenance_history |
| Environmental Data | 5 | humidity, ambient_temperature, dust_level |
- Size: 500K records, collected over 3 years
- Target: Continuous — time until next failure (in hours)
- Class balance: Not applicable (regression task)
- Missing data: Approximately 10% missing in sensor data due to equipment malfunctions
Requirements
- Build a regression model to predict time until the next failure based on the provided features.
- Achieve a mean absolute error (MAE) of less than 10 hours on the validation set.
- Provide insights on feature importance to guide maintenance strategies.
- Address missing data appropriately during preprocessing.
- Explain your choice of model and evaluation strategy.
Constraints
- The model must run in real-time to provide alerts for maintenance.
- It should be interpretable enough for engineers to understand the predictions.
- The model should be retrained quarterly with new data to adapt to changes in machine behavior.
