An effective framework using identification and image reconstruction algorithm for train component defect detection

doi:10.1007/s10489-021-02981-4

Return

An effective framework using identification and image reconstruction algorithm for train component defect detection

2022-01-11

PRE

DOI：10.1007/s10489-021-02981-4

Original

Original request for help

Save

Abstract

En 中文

Under long-term high-speed movement, the precision components of trains are extremely prone to defects, which could potentially endanger the safe operation of the train. However, there are many types of precision train components prone to defects; they are small in size and difficult to locate accurately. The defects themselves are also highly uncertain and diverse, making it impossible to establish an effective defect database. Meanwhile, a detection algorithm should have a high processing speed to ensure timely maintenance. Within the above context, in this paper an effective framework for multi-type train component defect detection based on an identification and image reconstruction algorithm is proposed. The framework is composed of a component identification stage and a defect diagnosis stage. The component identification method based on a component pre-location algorithm focuses attention on key areas of the train and ensures the visual integrity of the detected components. The component defect diagnosis method is based on an image-similarity generative adversarial network, which allows unsupervised reconstruction of template images to participate in defect diagnosis, thus coping with the diversity of component types and defect conditions effectively. The evaluation results on a CR400BF electric multiple unit series image dataset show that the framework has good robustness in complex environments and better performance in defect detection of train components.

Keywords:

Automatic defect detection

High-speed railway

Visual inspection

ISGAN

Unsupervised learning