Wheel Defect Detection Using a Hybrid Deep Learning Approach

Area of Science:

• Railway Engineering
• Machine Learning
• Vibration Analysis

Background:

• Defective railway wheels compromise operational safety and performance, leading to increased vibrations, noise, and track damage.
• Early detection of wheel defects is critical for safety, comfort, and cost-effective maintenance.
• Onboard detection is challenged by ambient vibrations, causing false alarms and inaccurate defect identification.

Purpose of the Study:

• To develop a hybrid deep learning approach for accurate onboard detection of railway wheel defects.
• To enhance the accuracy of wheel defect detection while maintaining cost-effectiveness and efficiency.
• To address the limitations of current detection methods caused by environmental vibrations.

Main Methods:

• A realistic railway wheelset simulation model was developed to generate vibration data under various conditions (non-faulty and six faulty scenarios).
• Simulations were conducted at different speeds and track conditions, generating 200,000 data points per iteration.
• A hybrid deep learning model combining a multi-layer perceptron (MLP) for feature extraction and several machine learning models (SVM, Random Forest, Decision Tree, KNN) for classification was trained and evaluated.

Main Results:

• The hybrid MLP-RF model achieved 99% accuracy, and the MLP-DT model achieved 98% accuracy in detecting wheel defects.
• The proposed deep learning models demonstrated high effectiveness in accurately classifying and predicting wheel defects.
• The study evaluated sensor layout effectiveness and applied deep learning for improved wheel flat detection.

Conclusions:

• The hybrid deep learning approach offers a highly accurate and effective solution for onboard railway wheel defect detection.
• The developed simulation model provides a valuable tool for generating comprehensive datasets for training and evaluating detection algorithms.
• This research contributes to enhancing railway safety and operational efficiency through advanced defect detection techniques.