A Study on Wheel Member Condition Recognition Using Machine Learning (Support Vector Machine)

Area of Science:

• Railway Engineering
• Machine Learning Applications
• Predictive Maintenance

Background:

• Current railway wheel management relies on post-event inspections after issues arise.
• This reactive approach poses risks to operational safety and efficiency.
• There is a need for proactive methods to detect and predict wheel abnormalities.

Purpose of the Study:

• To develop an advanced method for the early prediction of railway wheel abnormalities.
• To enhance the performance of machine learning algorithms for wheel condition classification.
• To improve the safety and reliability of railway operations through predictive maintenance.

Main Methods:

• Collected real-time operational data from sensors on railway vehicles (Busan Metro Line 4).
• Analyzed key factors and performed data distribution and correlation analyses to identify critical parameters for classification.
• Applied machine learning algorithms, including Support Vector Machine (SVM) with Linear and RBF Kernels, and Random Forest, using acceleration data.

Main Results:

• Identified the z-axis of acceleration as a significant factor for classifying wheel conditions.
• Achieved high accuracy in classifying railway wheels as in-service or defective using machine learning models.
• The SVM (Linear Kernel) model demonstrated the highest recognition rate at 98.70%.

Conclusions:

• The proposed method effectively predicts railway wheel abnormalities using acceleration data.
• Machine learning, particularly SVM (Linear Kernel), offers a highly accurate solution for real-time wheel condition monitoring.
• Implementing this predictive approach can significantly enhance railway safety and operational efficiency.