top of page
Image by SJ Objio
Title: Condition-Based Maintenance in Mine Railway Transportation Systems Using Big Data Analytics
Organization

Vale - The University of Genova

Period

February 2019 - July 2022

Status

Completed

Est. Budget

USD 1.6M

Est. Business Benefit

USD 9.5M

Indicative Value-to-cost

5.9x

Keywords

Condition-based maintenance, RUL, OL-SVR, streaming data, rail asset health, predictive maintenance

Benefit basis

Estimated value from avoided rail service disruption, reduced bearing failure risk, optimised maintenance scheduling, and improved corridor reliability.

Business Challenges 
​Mine railway transportation systems operate under high payload, long-haul, and demanding environmental conditions. Failures in components such as axle bearings can create safety risks, disrupt logistics corridors, and cause significant production delays. The key challenge was to convert high-volume streaming sensor data into actionable condition-based maintenance decisions while maintaining computational efficiency for continuous monitoring.
Expanded Technical Solution
The project developed a condition-monitoring and predictive-maintenance framework for railway assets, with a focus on train axle bearing health and Remaining Useful Life (RUL). Streaming sensor data was cleaned, synchronised, filtered, and converted into health indicators. An Online Support Vector Regression (OL-SVR) approach was used to update RUL estimates as new data arrived. The technical innovation was the balancing of prediction accuracy against computational time and model-maintenance effort, making the approach more suitable for operational deployment in a live rail environment.
Technical Work Packages/Methods
  • Designed a streaming data architecture for high-frequency sensor feeds such as temperature, vibration, wheel/axle condition indicators, speed, load, and operating context.
  • Applied signal processing, outlier detection, missing-data treatment, event segmentation, and sensor-quality scoring before model training.
  • Extracted degradation features including rolling-window statistics, trend indicators, rate-of-change metrics, spectral signatures, and operating-regime variables.
  • Implemented OL-SVR for adaptive RUL prediction and compared model accuracy against computational cost.
  • Defined alert thresholds, confidence bands, and maintenance decision rules for maintenance planners.
Key Deliverables
  • CBM analytical framework for mine railway assets.
  • Sensor-data quality and feature-engineering pipeline.
  • OL-SVR RUL prediction methodology validated on real collected datasets.
  • Maintenance decision-support logic for intervention timing and risk ranking.
Results and Value Created
The completed methodology demonstrated how real sensor streams could be transformed into predictive maintenance intelligence for rail assets. It improved the technical basis for early fault identification, helped maintenance teams move from calendar-based maintenance to condition-based intervention, and reduced the likelihood of unplanned service disruption. The estimated benefit reflects avoided rail interruptions, reduced component failure risk, lower emergency maintenance costs, and improved asset availability.
bottom of page