Water turbine fault diagnosis method based on multilayer time-frequency image recognition

The invention discloses a water turbine fault diagnosis method based on multilayer time-frequency image recognition. The method comprises the following steps: collecting a vibration signal and a throw signal of a water turbine, and carrying out noise removal processing and segmentation on the vibration signal and the throw signal; converting the signal into a time-frequency image by adopting Wigner distribution transformation, and carrying out size compression and adjustment on the image; acquiring 1-3-order two-dimensional intrinsic mode components of the time-frequency image by adopting a two-dimensional empirical mode, and performing splicing operation to obtain a spliced image; selecting sample data of a to-be-identified working condition and a fault working condition to construct a convolutional neural network model; and performing fault diagnosis application on an unknown sample of the unit by using the established model. According to the invention, the monitoring work intensity of hydropower station op