INTELLIGENT FAULT DETECTION IN RAILWAY POINT MACHINES USING TRANSFER LEARNING AND MOTOR CURRENT SIGNATURE ANALYSIS

Ensuring the reliability of railway point machines is crucial for the safe and efficient operation of rail networks. This study introduces an intelligent fault detection approach for S۷۰۰ K point machines using motor current signal analysis. By integrating deep learning and discrete wavelet transform (DWT), we developed a robust classification framework for diagnosing healthy and faulty machine conditions. The methodology involves preprocessing current signals using different wavelet functions to extract distinctive features, followed by training deep neural networks for fault classification. Two convolutional neural network (CNN) architectures —AlexNet and EfficientNet-B۰—were evaluated to assess their ability to detect faults under both familiar and unseen conditions. Additionally, the impact of wavelet selection and decomposition level on model performance was investigated. The study further explores the effect of adjusting data distribution in training, particularly by weighting normal operating conditions to improve fault detection accuracy and model generalization. This research highlights the potential of combining advanced signal processing techniques with deep learning for predictive maintenance in railway systems. The proposed approach enhances fault detection efficiency, contributing to more reliable and cost-effective railway infrastructure management.

کلیدواژه ها:

Railway point machines ، fault detection ، transfer learning ، discrete wavelet transform (DWT) ، convolutional neural networks (CNN) ، predictive maintenance