Business Objective
Our client is the maintenance planning department of a leading operator in the North American railroad industry with a fleet of 200,000+ railcars and an annual planning budget of about USD 1 B.
Wheel failure is a major maintenance activity for the client and estimating wheel replacement costs are crucial for accurate budget planning. Accurate forecasting is needed for optimum wheel inventory planning. The client wanted to improve budget planning and wheel inventory planning by forecasting wheel failures by the wheel size and the fleet type.
Challenges
- Large data volume with up to 30 years of historical wheel failure data
- Difficulty in training model using historical data, which includes data of wheels that have not failed yet
- Difficulty in accounting for weather changes for more accurate failure forecasting – Weather is different over the years (e.g. cold versus warm winters)
Solution MethodologyÂ
- Utilized data that included wheel attributes (wheel size, type, age, anticipated future mileage, location in car, geographic location) and weather data (month of the year, US vs Canada weather)
- Performed data exploration to look at both historical wheel failure data and active wheels that had no failure, and adjustment of weather variations across the US and Canada
- Developed Wheel Survival Model to arrive at Survival Distribution Function i.e.
Probability of wheel survival at a given mileage. Developed a Wheel Diagnostics dashboard - Calculated wheel failure forecast by summing the failure probabilities of all wheels by wheel size & fleet type
Business Impact
- Highly accurate wheel failure predictions with error margins < 2%
- Cost savings of USD 3 million recurring annually
