An explainable machine learning framework for railway predictive maintenance using data streams from the metro...

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Public transportation systems generate vast sensor data crucial for operational efficiency.
• Predictive maintenance in Intelligent Transportation Systems (ITS) can significantly improve quality and productivity.
• Current ITS maintenance often lacks real-time, data-driven predictive capabilities.

Purpose of the Study:

• To develop a real-time, data-driven predictive maintenance solution for Intelligent Transportation Systems (ITS).
• To implement an online processing pipeline for fault prediction with explainability.
• To validate the system's performance and practical applicability in railway operations.

Main Methods:

• A novel online processing pipeline integrating sample pre-processing, incremental Machine Learning classification, and outcome explanation.
• Development of a dedicated pre-processing module for on-the-fly feature engineering (statistical and frequency-related).
• Integration of an explainability module for natural language and visual fault prediction insights.

Main Results:

• Achieved over 98% F-measure and 99% accuracy on the MetroPT dataset.
• Demonstrated high performance and reliability even with class imbalance and noisy data.
• The explainability module effectively reflects the decision-making process of the fault prediction.

Conclusions:

• The proposed pipeline offers a methodologically sound and practically applicable approach for proactive maintenance in railway operations.
• High accuracy and F-measure are critical for maximizing service availability, reducing costs, and enhancing safety.
• The system empowers decision-makers with early failure detection and clear explanations for swift, informed actions.