Fault diagnosis method of gearbox

The invention relates to the technical field of mechanical fault detection, and discloses a fault diagnosis method of a gearbox. The fault diagnosis method comprises the following steps: collecting a vibration signal, carrying out the spectral analysis on the signal, and obtaining a spectrogram of the sampling signal in four working states; computing energy of a plurality of frequency bands; regarding the energy of the frequency bands as an input of a neural network, and obtaining an ideal output of the neural network, wherein the ideal output corresponds to the four working states; programming and outputting data of the frequency bands in matlab, and determining the neural network when an obtained target output is coincident with the ideal output in result; collecting the vibration signal of the gearbox to be detected, outputting the vibration signal to the neural network, and obtaining the working state, when the signal is collected, of the gearbox. By the utilization of the BP neural network, the four typical working states of a gear are trained, studied, simulated and analyzed, and the fault of the gearbox can be accurately distinguished through the analysis result.