A Fault diagnosis method for planetary gearboxes based on Bi-LSTM and feature screening by two-sample z test

For the fault diagnosis of planetary gearboxes, a method combined Bi-LSTM and two-sample z test is proposed. First, the two-sample z test method is used to calculate z-values between different faults, of each characteristic index under the same working condition. Then, the minimum z-values and the mean z-values of each characteristic index, are selected to form two characteristic screening matrices. Next, the first three maximum z-values of each row in the two matrices are selected respectively. And the characteristic indexes corresponding to the selected z-values, are selected as more discriminative characteristic indexes for various faults, to form a new characteristic set. The Bi-LSTM model is used for richer characteristic information, to perform classification of fault characteristic sequences. This method can decrease the degree of characteristic redundancy. Therefore, the computer memory consumption is greatly reduced, and the efficiency of model training gets higher. In addition, the experiments show that, the influence of local fault characteristic sequences aliasing has been effectively lessened. And the robustness and accuracy of fault diagnosis results have been improved.