AISpare (Industry)
Long Title: AI-Driven Spare Parts Demand and Warranty Risk Prediction for Light Rail Vehicles
Short:
Reliable forecasting of spare parts demand is a central challenge in the rail industry, as it ensures fleet availability, reduces life-cycle costs, and supports long-term service contracts. Forecasts must consider both planned demand resulting from wear, inspections, and regulatory replacement intervals as well as stochastic corrective demand caused by unexpected failures. Unplanned failures of large or complex components with long lead times are particularly critical because they can cause downtime and contractual penalties. Existing approaches such as condition-based or predictive maintenance rely on sensor data to detect failures but often neglect historical demand information and are therefore limited in long-term planning horizons ranging from two to more than 30 years.
This project therefore develops an AI-driven approach for forecasting spare parts demand based on historical field data from light rail vehicles. The objective is to systematically identify technical and operational influencing factors, compare heuristic and algorithmic forecasting approaches, and derive reliable predictions for spare parts consumption and warranty risks. By modelling relationships between usage, component characteristics, and failure patterns, failure probabilities can be quantified and used as a basis for informed pricing and planning decisions. This reduces manual and error-prone calculations during the bidding phase and enables consistent, data-based decision support.
Results:
The project results first include the creation of a structured and cleaned dataset in which preventive and corrective maintenance events are systematically classified. Using Failure Mode and Effects Analysis (FMEA), statistically significant influencing variables such as component age, mileage, and usage cycles are identified and prioritized to establish a reliable basis for predictive modelling.
Based on this, multiple data-driven forecasting approaches are developed and evaluated using appropriate performance metrics. The analysis investigates to what extent algorithmic models outperform experience-based heuristic estimations in terms of accuracy, applicability, and practicality. The resulting models enable the estimation of failure probabilities and future spare parts consumption at fleet level.
Finally, a conceptual framework is provided that integrates the predictive insights into the bidding phase and applies them to extended warranty premium calculation. This allows risks to be quantified, pricing strategies to be improved, and long-term service contracts to be managed more economically.
Partner: ALSTOM S.A., opens an external URL in a new window
Academic Integration
Within the scope of the project, a Master’s thesis was completed in cooperation with ALSTOM:
Markus Haftel, Data-Driven Spare Parts Demand Forecasting for Light Rail Vehicles
Contact Details:
Email: andreas.steiner@tuwien.ac.at