Case Studies

Rolling Stock Predictive Maintenance

Technologies

AI
Predictive maintenance
Condition monitoring
Cloud AI

Client’s Need

Monitor and predict anomalies at carriage, train and fleet level with focus on doors, propulsion and air conditioning systems. Solution must be able to collect data in cloud, perform data analytics, predict future anomalies, and update the prediction logic based on recently collected data.

Approach

Azure Cloud infrastructure development and setup by containers, considering data lake interfacing with SAP, data elaboration (Databricks), data storage and data visualization with Power BI. A specialised container leverages Spark to perform data elaboration, AI-powered predictive inference and retraining.

Main Outcomes

  • Structured data collection pipeline allowing detailed data analysis and monitoring of distances, down time, fault categories and occurrences, fault warning grouped by carriage type, train, fleet and travel route.
  • Predictive Maintenance algorithm for fault detection and prediction over variable time horizons, exploiting field data, weather data and environmental data. The solution also implements features for decision making support, for example suggesting the best maintenance scheduling time.
  • Power BI dashboard with different levels of access, where main statistical data are shown and alerts are propagated. User and Operator are able to perform custom analyses.