A Defect Detection Method for Rail Fasteners and Fishplates Based on Deep Learning

Abstract

Among the components of railway tracks, Fasteners and Fishplates are more censorious and vulnerable components that need a regular inspection. Fasteners are used to impose the rail onto a fixed position of crosstie and Fishplates join two different sections of rail. Like other components of railway track Fasteners and Fishplates inevitably produce defects as a result of long-term train operations and direct revelation to the natural surroundings which have a crucial effect on the safety of train operation. Several missing fasteners and defective fishplates cause gage widening, wheel climb, and train derailments consequently. In this work, a multi-object detection method based on a deep convolutional neural network that can perform automated detection of missing fasteners and defective fishplates alongside non-defective ones is proposed. Among state of art object detection algorithms, several YOLOv5 architectures are trained on a dataset consisting of images collected from the Bangladesh Railway. Experimental tests are conducted on the ballast railway track images and videos collected from the field to demonstrate the effectiveness and robustness of the method. Experimental results from multiple contemporary state of art object detection algorithms show that YOLOv5 not only detects fasteners and fishplates defects precisely but also outperforms others through several evaluation metrics in all stages. Keywords: Railway Fasteners and Fishplates, Defection Detection, Automatic Inspection, Deep Learning, Convolutional Neural Network, YOLOv5.