Fault diagnosis and location method of wind turbine bearing based on feature fusion under various working conditions1

Aiming at the problem of unbalanced fault data of rolling bearing and difficulty in extracting complementary features, a fault diagnosis and location method for wind turbine bearing based on feature fusion under various working conditions is proposed. This method takes the original vibration signal and current signal as the model input to avoid missed diagnosis and misdiagnosis caused by the insensitivity of vibration characteristics to faults. First, the Convolutional Block Attention Module (CBAM) is used to optimize the BilSTM for fault information timing feature extraction. At the same time, in order to solve the problem of manually setting 1DCNN super parameters, the goose swarm algorithm (GOA) is used to optimize 1DCNN for feature extraction, and the first level feature fusion is performed. Secondly, the one-dimensional fault data are transformed into a two-dimensional graph through the Markov Transition Field (MTF), and the 2DCNN is used to extract fault features, which are fused with the previously extracted fault features for secondary feature fusion. Finally, Softmax is used for fault classification. The experimental verification is carried out through the Paderborn dataset in Germany, and compared with the unoptimized 1DCNN, 2DCNN and BiLSTM to verify the superiority of the proposed method.