Research On Rolling Bearing Fault Diagnosis Method Based On Sparse Dictionary And Sparse Regularization

With the development of society and the progress of science and technology,the machinery and equipment has been developing rapidly towards large-scale,automation and intelligence.Rolling bearing is a key transmission component in mechanical equipment,which is widely used in large mechanical equipment such as aeroengine,wind turbine,high-speed train,steam turbine,etc.Due to its long-term operation under complex conditions such as variable load and variable speed,it is easy to be damaged.The lesser consequence is to cause the stoppage of production and affect the economic benefit;More serious hazards may cause major safety accidents and endanger life and property.Therefore,it is of great significance to study the condition monitoring and fault diagnosis methods of rolling bearings to ensure the safe and reliable operation of machinery and equipment.And how to effectively extract fault features from vibration signals of rolling bearings is the key to realize fault diagnosis of rolling bearings.Due to the change or instability of load and speed in the actual operation of the equipment,the vibration signal of rolling bearings often presents non-stationary and non-Gaussian characteristics.In view of the complex structure,serious noise interference,non-stationary characteristics and other problems of rolling bearing vibration signals in the background of strong noise,this paper takes sparse representation as the theoretical basis,and conducts in-depth research on rolling bearing feature extraction and fault diagnosis based on sparse dictionary construction and sparse regularization,providing theoretical and technical support for equipment operation and maintenance.The main research contents of this paper are as follows:（1）Aiming at the periodicity and impact characteristics of rolling bearing fault signals,a rolling bearing feature extraction method based on improved Chirplet is proposed.A sinusoidal frequency modulation factor is introduced into Chirplet atom,and quantum particle swarm optimization is used to optimize the parameters of the improved linear frequency modulation wavelet,and then the signal sparse reconstruction is realized by orthogonal matching tracking algorithm.Finally,the wavelet time-frequency parameters and the time domain statistical feature parameters of the reconstructed signal are used as sensitive feature sets to input the classifier for fault type identification,and the effectiveness of the proposed method for rolling bearing feature recognition is verified.（2）Aiming at the problem that it is difficult to extract weak fault features accurately under strong background noise,an incoherent and shift-invariant dictionary learning method based on adaptive multiple time domain synchronous averaging is proposed.A new learning mode of shift invariant dictionary is constructed by making full use of the periodic characteristics of fault signals and the structural characteristics of cyclic matrix.By resampling the noisy signal,the period of interest is quickly determined based on the time-domain characteristics and correlation analysis.And the correlation is used to reduce the number of cycles,and the shift invariant dictionary basis function is determined.Then,the alternating projection method is introduced to deal with the dictionary incoherently to improve the sparse expression ability of the dictionary.Through theoretical and experimental analysis,the effectiveness of the proposed method for weak fault feature extraction is verified,and then fault diagnosis is realized.（3）Aiming at the sparse underestimation problem of l₁norm regularization,a sparse model based on generalized log-multivariate non-convex penalty is designed.This method combines convex analysis and maximal monotone operator splitting techniques to solve the inverse problem of sparse regularization.The proposed near neighbor operator corresponding to the non-convex penalty function can asymptotically approximate the identity function,which is conducive to improving the problem of the underestimation of high amplitude features by the l₁norm.The convexity condition of the non-convexity model is studied,and the optimal solution is obtained by the proximal method.The purpose of reducing parameter dependence and improving the applicability of the sparse model is achieved by analyzing the parameter constraint relationship in the model.By constructing a sparse representation dictionary that matches the fault signal,the generalized multivariate regularization method is used to improve the sparse estimation accuracy of the fault signal,thereby improving the reliability of fault diagnosis.