Rail Fatigue Crack Classification Based on Improved DPC Algorithm Using Acoustic Emission Technology

The effective classification of crack, as a key aspect in maintaining rail safety, plays a significant role in ensuring the reliable operation of high-speed railway. There are mainly supervised and unsupervised algorithms for classification problems. Supervised learning algorithms require a large amount of labeled data and have limited generalization capability, and unsupervised learning algorithms have low classification accuracy. To address these shortcomings, a multi-center density peak clustering based on the multi-layered weight density (MDPC-MWD) is proposed to classify rail cracks more efficiently using acoustic emission technology. The information of different rail cracks is comprehensively reflected by studying the entropy feature of the signals from the perspective of singular spectrum. Then, an improved multi-layered weight density is proposed to address the uneven density distribution of the crack datasets. Finally, the classification results are obtained through micro-cluster self-recognition and self-merging strategies. The method is demonstrated in the rail fatigue experiments, and the results show that the proposed MDPC-MWD method achieve outstanding classification performance.