Fault Diagnosis of Train Bogie Bearing Based on AP-Tensor Clustering Under Multichannel Data

Due to rapid development of measurement and sensing technology, there is more monitoring data on the train bogie, and the data obtained is no longer a single channel. Meanwhile, due to the high safety requirements of trains, the fault samples and frequency are low, which makes it challenging to obtain enough samples by supervised learning. Therefore, this paper proposes an unsupervised learning approach that directly deals with multi-channel data to solve the above problems. First, the time and frequency features construct the second-order multi-channel tensor samples. Based on the tensor samples, an AP-tensor clustering method is proposed for unsupervised recognition. The typical train bearings were used for the experiment. The results show the advantage of the proposed approach in other comparison techniques.