Research On Fault Classification Method Of Rolling Bearings Based On Supervised Sparse Representation

Mechanical equipment plays a vital role in various industries,such as aerospace,petrochemical transportation,high-speed railway,and deep ocean exploration.Its running condition directly affects the national economic industry.Rolling bearing is widely used in mechanical transmission system,and it is an important component of rotating machinery.Its running state is closely related to the overall safety of the whole mechanical equipment.Given the harsh operating conditions of mechanical equipment,it is susceptible to fatigue failure of rotating parts,which can lead to catastrophic equipment failure.Therefore,the implementation of fault diagnosis and classification of these key components and the development of maintenance strategies according to the equipment status can not only effectively avoid sudden accidents,but also greatly reduce the production and maintenance costs of enterprises,which is of great significance to ensure the stable and safe operation of mechanical equipment.This research was supported by the National Natural Science Foundation of China(Project No.51905218).This paper focuses on rolling bearings and aims to classify single fault classification and compound fault classification of rolling bearings.To achieve this goal,we propose a fault classification method based on supervised sparse representation,and conduct the theoretical analysis and application research on related problems.In this paper,the fault modes of rolling bearings are described,the waveform characteristics of vibration signals of rolling bearings under fault conditions are analyzed,and the sparse characteristics of vibration signals are explained.Furthermore,the paper elaborates on the sparse representation theory of signals,including the construction of sparse representation models,the construction of overcomplete dictionaries,and the solution of sparse representation coefficients.Aiming at single fault classification of rolling bearings,an enhanced label consistent K-SVD fault classification method based on supervised sparse representation was proposed.The traditional label consistent K-SVD method lacks of adaptive feature extraction and is greatly affected by different data features.To address this issue,the proposed method constructs an adaptive feature extraction objective function,which uses the measure feature index of Euclidean distance within class and interclass distance,and selects adaptive features according to the score of the objective function.Compared with common feature extraction methods,the superiority of the proposed adaptive feature extraction method is verified.In order to solve the problem that improper setting of traditional over-complete dictionary can easily lead to the local optimal solution of learning,the adaptive feature training set is initialized and the learning dictionary of adaptive feature is generated,which not only ensures the match between the dictionary and the signal,but also ensures the adaptability of the dictionary.The sparse coefficient matrix was calculated by the orthogonal matching tracking algorithm,and the input signals of the test set were classified by the trained dictionary and classification parameters.The effectiveness and superiority of the proposed method are compared with common statistical features used in feature extraction,PCA dimension reduction,and other feature extraction methods for single fault classification of rolling bearings and bearings in a gearbox.Aiming at compound fault classification of rolling bearings,a dual dictionary label consistent K-SVD fault classification method based on fault information was proposed.Based on two different kinds of fault information,the fault feature sets for the inner and outer ring faults were constructed respectively,and the double dictionaries representing different fault features were learned.The objective function of the double dictionary label consistent K-SVD algorithm was established by using the fault information.The optimal dictionary and sparse coefficient obtained from the double dictionary learning based on the fault information were used as the classification basis to realize the classification of different compound faults.The experimental results of self-made experimental bench data show that this method has a good effect on classification accuracy,and to some extent solves the problem that the single dictionary supervised sparse representation classification method cannot match different fault features due to the limitation of dictionary number.A comparison experiment was carried out between the single dictionary LC-KSVD,which is used for feature extraction by common statistical features,and the single dictionary LC-KSVD algorithm,which is used for enhanced adaptive feature extraction,to verify the accuracy of the proposed method in the classification of rolling bearing compound faults.In summary,based on supervised sparse representation,this paper formulated feature extraction strategies for single and compound fault classification of rolling bearings respectively,and realized fault classification of rolling bearings by combining adaptive learning dictionary.This research can effectively prevent sudden accidents,but greatly reduce the production and maintenance costs of enterprises,with practical significance for fault feature extraction and fault classification of rotating machinery equipment.