Research On Fault Diagnosis Methods Of The Non-stationary Multi-component Signal Of Gearbox

The safety and reliability of the mechanical system in the industrial process determine the quality of products.Whether the fault can be identified and classified in time is the key to ensure the safe operation of the system and restrain the deterioration of faults.With the rapid development of manufacturing digitization,in the face of massive data,how to quickly extract signal features,find sensitive feature sets and accurately identify and classify faults is the key to efficiently discover mechanical system faults and avoid serious damage.In industrial production,as the main component of transmission power,the health of gearbox determines the mechanical system can work efficiently.The gearbox is assembled from gears,bearings and other components,and often operates under variable load and variable speed.The vibration signal of gearbox with complex components is characterized by non-stationary,non-Gaussiality and nonlinearity,and contains vibration information of each component.For intelligent and efficient fault diagnosis of complex vibration signal,it is necessary to separate the signal effectively,extract information of features effectively and classify the fault accurately.To solve these difficulties,a series of fault diagnosis methods for non-stationary multi-component signals of the gearbox should be proposed based on the fault mechanism and the vibration features of the gearbox,the advantages of signal processing algorithms and the operation characteristics of the gearbox.The main content and final results of this paper include the following parts:(1)An adaptive multi-ridge extraction method for peak detection based on synchro-squeezing wavelet transform(SWT-AMRE)is proposed.This method can effectively reduce the running time of the algorithm and can be used for real-time extraction.It lays a good foundation for the extraction of time-frequency ridges.Multiple ridges in the time-frequency matrix of synchro-squeezing wavelet transform can be extracted adaptively.Under different noise intensity,this method can keep high extraction precision.The overall mean relative error between SWT-AMRE and the theoretical value is only 2.92%.And compared with the traditional peak detection,the mean relative error is reduced about 50%.(2)A VKF signal separation and feature extraction method based on DSWT-IMRE is proposed.In this method,DSWT-IMRE is used as the high-precision adaptive IF estimation method and the high-precision IF estimation is used as the instantaneous frequency parameter of VKF filter,so that the complex multi-component non-stationary signals can be separated directly in the time domain.It is transformed into a combination of multiple stationary single component signals and residual information signals.The corresponding IF estimation of each component signal is taken as the reference frequency of the order tracking analysis,and the order tracking analysis of each component signal is carried out respectively.The diagonal slice of bispectrum analysis of the signal residual is carried out,so as to suppress the Gaussian noise and effectively separate and extract the fault features in the vibration signal.In signal reconstruction,the relative error of this method is reduced by about 9%compared with the reconstruction method of VKF based on traditional peak detection.(3)A cyclic GMM-FCM pattern recognition method based on VKF joint time-domain features is proposed.The joint time-domain features based on VKF can effectively reduced the dimension of the feature set.This method makes it as the feature set of pattern recognition for clustering analysis.The method can detect the global and local outliers and make the classification boundary more reasonable.In this method,the number of FCM categories is determined adaptively by cyclic GMM algorithm to make the classification result more accurate.The classification accuracy reached 98.77%.(4)A power flow gearbox test bed is built and vibration signals of the gearbox in various states,such as faultless state,pitting transition state,tooth surface pitting state and composite fault state,is collected.The feasibility,validity,superiority and practicability of the proposed method are verified by simulation data and experiment data,which provide research direction for fault diagnosis of non-stationary multi-component signals in gearbox.