A quantitative detection method for wheel polygonization of heavy-haul locomotives based on a hybrid deep learning model

Abstract

Wheel polygonization poses severe threat to the operational safety of railway vehicles. To improve running safety, condition monitoring of wheel polygonization based on axle box acceleration (ABA) signals is an effective method. However, the signals usually contain strong noise, making it difficult to diagnose polygonal wheels quantitatively. To address these issues, this paper presents a quantitative detection method for wheel polygonization based on a deep learning (DL) model that integrates Convolutional Neural Network (CNN) with Long Short-term Memory (LSTM). The CNN is responsible for automatically extracting the sensitive spatial features from the original input and it can retain effective signal components. The LSTM is introduced to fuse the local features and conduct sequence modeling for polygonal wear amplitude estimation. Both dynamics simulations and field tests are carried out to verify the feasibility of the proposed method. The detection results indicate that the proposed model can accurately estimate the wear amplitudes.