Research On Feature Extraction Of Gear Vibration Signal Based On Compressed Sensing And Extended Bispectrum

It is very necessary to monitor the status of the key parts such as gears and detect fault accurately and timely according to the state data in modern industrial production.But the traditional signal acquisition system based on Nyquist sampling theory will generate massive running state data and there are various kinds of noise.Therefore,the existing fault diagnosis methods are faced with the problem of huge raw data and difficult to extract signal characteristics.Compressed sensing theory developed in recent years is based on the sparse characteristics of signals to achieve compressed sampling,which is a new idea to solve the huge amount of data.At present,most of the research on the application of compression sensing is to compress the signal first,then reconstruct the original signal for analysis.However,the existing reconstruction algorithms have many problems,such as high computational complexity,large resource consumption,and the accuracy and probability of reconstruction are affected by many factors,especially the poor effect of signal reconstruction including noise,and the reconstruction is not the direct purpose of signal processing.Therefore,this paper discarded the reconstruction process and studied the characteristics of the compressed signal directly,and tried to establish the corresponding relationship between the signal features of the compressed domain and the fault states,so as to improve the speed and efficiency of state monitoring and fault diagnosis.Firstly,based on compressive sensing theory and the sparse characteristics of gear vibration signals in frequency domain,the compression sampling model of modulation bandwidth converter suitable for gear vibration signal is studied.The principle of the model and the corresponding characteristics in frequency domain are analyzed.There are various kinds of noise in the gear vibration signal under the actual working condition,so the bispectrum with good noise suppression effect is studied,and the suppression effect of bispectrum on additive noise with random phase distribution in(-?,?] is analyzed theoretically.In this paper,the characteristics of compressed signal in frequency domain are studied in depth.Combined with the advantages of compressed sensing low-dimensional sampling and high-order spectral noise suppression,an extended bispectral method is studied to extract the phase coupling characteristics of compressed signal directly,and its effectiveness is verified by simulation.The effect of the noise of the compressed signal in the extended bispectrum analysis is theoretically analyzed.The extended bispectral method is compared with the traditional reconstruction method in terms of the time of the algorithm and the influence of noise with different intensity.The existing problems in practical application are studied.The extended bispectrum is extended from time domain to angle domain,and order analysis is introduced.The concept of order extended bispectrum is put forward to solve the problem of fluctuating speed and non-stationary running state in actual operation.Aiming at the problem that the signal feature is blurred by the irrelevant noise component,the angular domain synchronization averaging method is introduced to suppress and weaken the non-synchronous component which is not related to the rotational speed.Finally,the algorithm structure is improved to meet the requirement of low frequency sampling,and the improved algorithm is verified by the actual gear vibration data.