Research on Bearing Cross-domain Fault Diagnosis Based on Invariant Subspace Learning with Tensor Alignment

Rolling bearings are very important in the research of fault diagnosis of rotating machinery. With the continuous development of big data technology, fault diagnosis has ushered in a data-driven era. This paper starts with high-dimensional rolling bearing fault data and proposes a tensor invariant subspace learning technique (TAISL). By converting the vibration signal formula into a third-order tensor in order to directly adapt the source domain and target domain in the case of tensor representation, the TAISL method is introduced into the tensor representation method to narrow the distance between the source domain and the target domain and then the 2DCNN method is used to identify bearing faults. Through experiments under various conditions, the proposed method was fully verified. The results show that the proposed method can effectively shorten the distance between different distributions reduce noise interference, and effectively classify fault samples even when the data distribution is inconsistent and the data dimension is high.