A Fault Diagnosis Method Of Rotating Machinery Based On GP-Singular Spectrum Decomposition Time-Frequency Diagram And CNN

Rotating machinery is a key component in mechanical equipment,and the fault detection and diagnosis of rotating machinery are critical to the good operation of mechanical equipment.The vibration signal of a rotating machine is generally a non-linear and non-stationary signal,and the existing fault diagnosis methods of rotating machinery have some defects for the analysis of non-stationary and nonlinear signals.Therefore,the Singular Spectrum Decomposition(SSD)is used for processing non-stationary and nonlinear signals.The basic theory and application of the method are studied in this paper.An improved singular spectrum decomposition method named GP-Singular Spectrum Decomposition(GP-SSD)is proposed,and it's used for fault diagnosis of rotating machinery.For fault diagnosis of rotating machinery under variable working conditions,the selection of features is highly critical when extract fault features directly.The time-frequency diagram is an information representation that can simultaneously reflect the time domain and frequency domain information for the mechanical vibration signals.Therefore,a fault diagnosis method of rotating machinery based on the GP-singular spectrum decomposition time-frequency diagram is proposed in this paper.Based on the characteristics of the vibration signal of rotating machinery,a CNN model of the rotating machinery is proposed.The combination of the two method is applied to the fault diagnosis of the rotating machinery in variable working conditions.The research content of the thesis is carried out in the following aspects:(1)A GP-singular spectrum decomposition method is proposed.The GP-singular spectrum decomposition method is proposed for the shortcomings in the singular spectrum decomposition method which lacks of theoretical support for the selection of the embedded dimension.The singular spectrum decomposition method has the disadvantage which is the subjective selection of embedding dimension.The GP algorithm is used to automatically select the embedding dimension based on the relationship between fractal dimension and embedding dimension.The simulation and experimental signals are also used to verify the superiority of GP-singular spectrum decomposition method.(2)A time-frequency diagram named GP-singular spectrum decomposition time-frequency diagram is proposed.The GP-singular spectrum decomposition method is used to decompose the original signal,and then Hilbert transform is used to obtain its instantaneous frequency and instantaneous amplitude to form the time-frequency diagram of GP-singular spectrum decomposition.Compared with the time-frequency diagram of Short-Time Fourier transform,the time-frequency diagram of Wavelet Transform and the time-frequency diagram of HHT,it is found that the time-frequency diagram of GP-singular spectrum decomposition has better performance in frequency resolution and time-frequency aggregation.(3)A structure of convolutional neural network based on the characteristics of rotating mechanical signals is proposed,which is the CNN model of the rotating machinery.Based on the compound fault signals of rotating machinery,the structures of convolutional neural network like convolution kernel size,number,pooling mode and others are selected to obtain the best performance.(4)A fault diagnosis method of rotating machinery based on the time-frequency diagram of GP-singular spectrum decomposition and the CNN model of rotating machinery is proposed.The GP-singular spectrum decomposition method is used to construct the time-frequency diagram of GP-singular spectrum decomposition,and then the time-frequency diagrams are used as the inputs of the convolutional neural network to classify the fault category of the rotating machinery.The method can achieve good results when it is used in the fault diagnosis of gears and bearings in variable working conditions.