Open Circuit Fault Diagnosis Method Of 24 Pulse Wave Rectifier Based On Compressed Sensing Theory

At present,Nyquist sampling theorem is applied to fault diagnosis of rectifier output signals.In order to ensure the accuracy of diagnosis results,a lot of signal acquisition is needed,which leads to problems such as large amount of data processing,data storage and transmission difficulties.Compressed Sensing(CS)theory is born,it collects data signals at a frequency lower than Nyquist's,thus reducing the amount of sampled data.In addition,CS theory integrates the two processes of traditional signal data sampling and data compression,and uses the non-adaptive projection of the signal to preserve the important information of the original signal,and then selects the appropriate reconstruction algorithm to reconstruct the original signal,which can alleviate the pressure of data transmission and reduce the demand for data storage space.In this thesis,based on CS theory,CS theory and BP neural network are combined to compress and sample the voltage signal of 24 pulse wave rectifier,and finally realized fault diagnosis and recognition.Firstly,the compressed sensing theory is introduced in detail to construct a 24 pulse rectifier model and analyze the sparsity of the original signal.By analyzing the original signal sparse transform reflected signal characteristics,using discrete cosine transform(DCT)to the original signal is sparse representation,and then use the triangle coefficient weighted method to optimize the measurement matrix,to reduce the correlation among the sparse and measure matrix,to strengthen the sampling efficiency of low frequency signal,finally to make the compression performance of the sample must be promoted.Secondly,after studying the reconstruction algorithm of matching tracking class,the traditional inner product similarity matching criterion was replaced,and a piecewise orthogonal matching tracking algorithm using Dice coefficient similarity measurement criterion was proposed to optimize the support set.After experimental comparison with other traditional matched tracking reconstruction algorithms,the algorithm proposed in this thesis can effectively distinguish the two approximate atoms,and can better select the atoms that are more matching with the residual value,so as to improve the signal reconstruction quality and reconstruction performance to a certain extent.Thirdly,BP neural network is applied to fault diagnosis of rectifier.After the compression sampling and reconstruction of the output voltage signal of the 24 pulse wave rectifier,the fault feature extraction of the sparse coefficient vector to be solved was conducted directly,and then the extracted fault feature vector was input into the BP neural network to realize fault diagnosis and recognition of the rectifier.The experiment shows that compared with the traditional method,the length of signal data to be processed by this method is greatly reduced,the fault diagnosis accuracy is improved,and the fault diagnosis speed is accelerated.