Online inspection system for the automatic detection of bolt defects on a freight train

Inspecting the condition of the key components of freight trains is an important task in the rail industry. Bolts on the wheel bearings are key components of a bogie, and bolt defects, such as missing or broken bolts, can lead to serious accidents. To improve the traditional manual inspection procedure, which is both laborious and inflexible, a novel method of automatic image recognition for bolt defects is proposed in this paper. The main procedures are as follows. When a freight train drives through the inspection station, images of the train’s wheels are captured by cameras installed alongside the track. Based on the local binary pattern descriptor, a support vector machine classifier is trained to distinguish between bolt and non-bolt images. The classifier is then combined with a rotate-and-slide window method to localize the three bolt regions in the wheel image. Specifically, a self-updating method is proposed in the training phase to automatically capture the various different situations experienced by bolts in real-life scenarios. After localization, we distinguish defective bolts from normal bolts based on whether there is a hexagonal shape in the bolt region. As demonstrated by real-life experiments, our proposed method can guarantee to find bolt defects and further work will be devoted to reducing the false alarm rate.