A contribution to reliable rail transport: AI-powered real-time wheel defect detection

Area of Science:

• Artificial Intelligence
• Machine Learning
• Railway Engineering

Background:

• Railway transport is crucial for economic growth and regional integration.
• Wheel defects are a significant safety risk and cause financial losses in railways.
• Real-time monitoring is essential for early detection and prevention of railway wheel failures.

Purpose of the Study:

• To develop and evaluate an AI-based framework for real-time railway wheel defect detection.
• To address challenges in dataset creation, including class imbalances and annotation.
• To demonstrate the practical deployment of the AI model in operational railway environments.

Main Methods:

• Development of a custom wayside imaging system to capture high-resolution images.
• Creation of the FaultSeg dataset for training and evaluation.
• Evaluation of advanced You Only Look Once (YOLO) models (v5-v12) and a Real-Time Detection (RTD) Transformer model.
• Extensive hyperparameter tuning for optimal model performance.

Main Results:

• The YOLOv5-seg model achieved superior performance with 91% precision, 90% recall, and 92% mAP@0.5.
• Real-time processing was achieved at 30 FPS with latency under 30 ms.
• The optimized model was successfully deployed on an edge device for operational use.

Conclusions:

• The AI-based framework enables effective real-time detection of railway wheel defects.
• The developed system enhances predictive maintenance strategies and improves overall railway safety.
• This work contributes a publicly available dataset and demonstrates practical AI application in railway condition monitoring.