Mechanical fault diagnosis method based on improved PSO-SOM-BPNN

The invention discloses a mechanical fault diagnosis method based on improved PSO-SOM-BPNN. According to the scheme, a signal processing technology and two types of neural networks are connected in series and combined to carry out rapid and accurate identification on the state of a rotating machine. The method comprises the following steps: for a diagnosed rotating machine, after a vibration signal is collected, de-noising is immediately carried out on the vibration signal through wavelet analysis, and energy characteristics are extracted through wavelet packet decomposition; during the period, the primary function and the decomposition layer number of the two are optimized; and then the extracted energy features are used as SOM-BPNN input. According to the invention, an inertia weight and a learning factor of a PSO algorithm are adjusted in a self-adaptive mode, a speed item is abandoned to optimize SOM-BPNN parameters, and the self-adaptive speed-item-free particle swarm optimization-based SOM-BPNN mechanic