An Unsupervised Learning Approach for Wayside Train Wheel Flat Detection.

Area of Science:

• Mechanical Engineering
• Railway Engineering
• Signal Processing

Background:

• Wheel flats are a common cause of rail damage, leading to increased maintenance costs and potential component failures.
• Accurate detection of wheel flats is crucial for ensuring railway safety and operational efficiency.

Purpose of the Study:

• To compare the effectiveness of four feature extraction methods (AR, ARX, PCA, CWT) for automatic detection of wheel flats.
• To evaluate the performance of these methods using rail acceleration data from freight vehicles.

Main Methods:

• Feature extraction using auto-regressive (AR), auto-regressive exogenous (ARX), principal component analysis (PCA), and continuous wavelet transform (CWT).
• Feature normalization via a latent variable method.
• Data fusion to improve defective wheel recognition sensitivity.
• Outlier analysis for damage detection.

Main Results:

• AR and ARX methods proved more efficient for wheel flat detection compared to CWT and PCA.
• A single rail sensor is sufficient for identifying defective wheels across most features.
• AR and ARX methods can distinguish between left and right-side defective wheels.
• The ARX method showed robustness in detecting wheel flats using accelerometers placed solely on sleepers.

Conclusions:

• AR and ARX methods are superior for automated wheel flat damage detection in railway systems.
• The findings suggest that simplified sensor configurations can achieve reliable defect identification.
• The ARX method offers a robust solution for detecting wheel flats, even with limited sensor placement.