A Survey on Data-Driven Predictive Maintenance for the Railway Industry.

Area of Science:

• Industrial Engineering
• Data Science
• Artificial Intelligence

Background:

• Predictive Maintenance (PdM) leverages sensor data from industrial equipment for monitoring and fault detection.
• Machine Learning (ML) and Deep Learning (DL) are increasingly applied to PdM, but performance depends on method selection.
• Current progress in ML/DL for PdM faces challenges, necessitating careful consideration of techniques.

Purpose of the Study:

• To survey existing Machine Learning (ML) and Deep Learning (DL) techniques for Predictive Maintenance (PdM) specifically within the railway industry.
• To provide a structured overview of current approaches, categorizing them by task type, methods, evaluation metrics, equipment, and datasets.
• To identify challenges and suggest future research directions in railway PdM.

Main Methods:

• Systematic literature review of ML and DL techniques applied to railway Predictive Maintenance (PdM).
• Taxonomic classification of surveyed methods based on task, employed algorithms, evaluation metrics, specific railway equipment/processes, and datasets.
• Analysis of current trends and limitations in the application of ML/DL for railway PdM.

Main Results:

• The survey categorizes various ML and DL approaches used for PdM in the railway sector.
• It highlights the diversity of techniques and their application to different railway equipment and processes.
• Key challenges and performance dependencies on method choice are identified.

Conclusions:

• The application of ML and DL in railway PdM is promising but still developing, with significant challenges remaining.
• Appropriate selection of ML/DL methods is crucial for effective PdM performance in the railway industry.
• Future research should focus on addressing identified challenges to advance railway predictive maintenance.