Detection of Wheel Polygonization Based on Wayside Monitoring and Artificial Intelligence

Area of Science:

• Engineering
• Artificial Intelligence
• Data Science

Background:

• Wheel polygonization is a critical issue affecting rail vehicle dynamics and track integrity.
• Accurate detection of wheel defects is essential for ensuring operational safety and reducing maintenance costs.
• Existing methods for wheel defect detection may lack robustness against environmental and operational variations.

Purpose of the Study:

• To develop and evaluate an artificial intelligence-based approach for detecting wheel polygonization.
• To compare the effectiveness of Autoregressive Exogenous (ARX) models and Continuous Wavelets Transform (CWT) for feature extraction.
• To assess the impact of data normalization, data fusion, and outlier analysis on defect detection accuracy.

Main Methods:

• Simulated train-track interaction to generate dynamic responses from a Laagrss-type rail vehicle.
• Unsupervised feature extraction using Autoregressive Exogenous (ARX) models and Continuous Wavelets Transform (CWT).
• Data normalization via Principal Component Analysis (PCA) and data fusion using Mahalanobis distance.
• Outlier analysis for distinguishing healthy from defective wheels, complemented by sensitivity analysis on sensor configurations.

Main Results:

• The proposed AI methodology successfully detects wheel polygonization using dynamic responses.
• Both ARX and CWT demonstrated capability in feature extraction for wheel defect analysis.
• PCA and Mahalanobis distance effectively normalized data and enhanced defect recognition sensitivity.
• Outlier analysis provided a reliable method for classifying wheel conditions.

Conclusions:

• The developed AI approach offers a promising solution for automated wheel polygonization detection.
• The combination of advanced signal processing and machine learning techniques enhances the accuracy and robustness of the system.
• Sensitivity analysis highlights the importance of sensor placement and number for optimal system performance.