Intelligent Unsupervised Defect Detection of Rail Surface via Generative Adversarial Networks

Abstract

One of the most important factors affecting the safety of the train during the operation of the railway system is rail surface defects. Therefore, detecting these defects at an early stage is very important for rail safety. Defect detection approaches using supervised learning are based on the use of healthy and defective data in training and the evaluation of the trained model during testing. In this study, an adversarial generative network-based unsupervised defect detection system is proposed to detect defects in a situation where only healthy rail surface data is available. The performance of the proposed method was also evaluated with rail detection and image enhancement studies. The proposed approach has been tested with healthy and defective data taken in different light conditions, and defects have been detected with an accuracy rate of over 98%. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.