Marine rotor fault diagnosis method of frequency spectrum during maximum overlapping discrete wavelet packet transformation

The invention relates to a marine rotor fault diagnosis method of a frequency spectrum during maximum overlapped discrete wavelet packet transformation. The method comprises the following steps: 1) acquiring a signal: acquiring a rotor vibration signal x (t); 2) signal noise reduction: carrying out singular value difference spectrum noise reduction on the rotor vibration signal x (t); 3) signal reconstruction: decomposing the denoised signal x '(t) by using a maximum overlapping discrete wavelet packet transform method to obtain a reconstructed signal x' '(t); 4) constructing a data set: obtaining a two-dimensional time-frequency spectrogram of the reconstructed signal x ''(t) by adopting Hilbert transformation, and forming a rotor training data set M and a test data set N; 5) establishing a pre-training fault diagnosis model; 6) finely adjusting network model parameters; and 7) verifying the model: inputting the test data set N into the trained AlexNet network, and diagnosing the working state and the fault