Business Objective
Our client is a leading operator in the North American railroad industry with a fleet of 200,000+ railcars. The client had WILD (Wheel Impact Load Detector) System – a wayside sensor that detects faulty wheels. Flooding, weather, or mechanical failure can cause faults in these detectors. A faulty detector could cause the removal of 200-500 healthy wheels, resulting in unnecessary repair expenses.
Hence, the client wanted to:
- Build a wheel health monitoring system for WILD detectors
- Develop an interactive tool for early identification of faulty detectors
- Develop a reporting system for periodic review of detectors
Challenges
- Large data volume – 100GB data over a 3-year period
- Seasonal variations in traffic and weather patterns pose estimation challenges
- Solution integration with existing infrastructure for quick and actionable results
Solution Methodology
- Reduced ~100GB data to 100MB by converting raw data into hourly alerts and summarizing hourly alerts to a weekly level
- Clustered detectors into homogenous groups by location, climate, load, car type, and speed
- Created metric alerts per wheel (APW) and normalized traffic using a 3-year average APW value for each site
- Identified outliers leveraging control charts and leveraged Z-score based detector scoring system to identify faulty detectors
- Refreshed benchmark every week based on new data
Business Impact
- Saved over USD1 M in annual wheel replacement costs resulting from faulty WILD detectors
- Developed a comprehensive system that delivers weekly reports via email and has a Tableau-based interactive UI with drill-down capabilities
