Study On The Sparse Representation Method For Fault Diagnosis Of Gear Transmission System

Gear transmission systems of rotating machinery widely used in industrial machinery are widely used in industrial production.However,the bearings and gears in the gear transmission systems are prone to failures,which will cause problems or even failures in the operation of the entire transmission system.Therefore,it is necessary to diagnose the faults of the bearings and gears in the gear transmission system as soon as possible,and find faults and deal with them in time.It is the most common method for fault diagnosis to process and analyze the vibration signals of gears and bearings.During the process of processing the vibration signal,whether there is a fault or the type of the fault can be well diagnosed by extracting and classifying the fault characteristics of the faulty bearings and gears.In this paper,systematic research is carried out on the fault feature extraction and classification method of two important components of gears and bearings.The basic research on sparse representation theory such as basis pursuit,matching pursuit,dictionary construction and K-SVD has been carried out.The following major research results have been achieved.Firstly,a transient characteristic detection method based on improved Morlet wavelet and adaptive iterative threshold algorithm is proposed.An improved kurtosis index is designed to better search the best sparse parameters.An improved Morlet wavelet that strictly satisfies the compatibility conditions is constructed.Combining them with Fourier dictionary and using adaptive iterative threshold shrinkage algorithm(AITSA)solve the parameter optimization problem under sparse representation and can optimally detect transient components.Secondly,an impulsive fault feature extraction method based on adaptive OMP algorithm and improved K-SVD algorithm is proposed.According to the morphological characteristics of harmonics and modulated components,using the over-complete Fourier dictionary and adaptive spark,the OMP algorithm can adaptively and accurately separate harmonic components from the observed signals.Then,the circulating shift and main period segmentation method are applied to establish the signal matrix,and the time-domain average is used in the signal matrix to obtain an efficient initial transient dictionary construction method.This method avoids local optimization,speeds up the operation,improves the detection accuracy of the transient characteristics and lays a key foundation for extending the K-SVD algorithm to 2D signal processing.Finally,based on adaptive OMP algorithm and 2-dimensional signal adaptive K-SVD algorithm,a fault identification method based on LC-KSVD for gear transmission system is proposed.The characteristics of the signal processed based on the adaptive OMP algorithm and the improved K-SVD algorithm are calculated and used as the input of the LC-KSVD for dictionary learning and classification.The method is applied to the mechanical vibration signal of the bearing fault test rig.The results show that the method has a good classification effect.In summary,the methods proposed in this paper have good fault feaure extraction characteristics and fault feature classification characteristics.