Bearing Fault Diagnosis Based On Signal Sparse Representation

Rolling bearing is the basic component in manufacturing machine,its working condition is directly related to the health condition of the whole mechanical equipment.At the same time,with the continuous upgrading of mechanical equipment,the requirements for bearing rotation accuracy and reliability are also constantly improved.But bearing operating environment is becoming more and more demanding.It usually works in high temperature,high pressure,overload and/or other extreme conditions,prone to varying degrees of failure.In this paper,the signal sparse representation theory is applied in the field of bearing fault diagnosis.Based on the signal sparse representation theory,feature components of bearing fault are extracted to realize bearing fault diagnosis.The signal sparse representation mainly includes two parts:1)sparse dictionary construction,and 2)sparse recovery algorithm.Through the dynamic analysis of bearing fault,fault mechanism is revealed,and an analytical dictionary is constructed based on this.Meanwhile,for different types of faults,there are significant differences in fault feature frequency.This domain knowledge can provide a basis for bearing fault diagnosis.The following parts of the paper mainly focus on two types of sparse optimization problems:7)₀-norm constraint and7)₁-norm constraint.For the l₀-norm constrained problem,Stagewise Orthogonal Matching Pursuit（St OMP）algorithm is used to extracted bearing fault feature components.Then,Cluster-contraction Stagewise Orthogonal Matching Pursuit（Cc St OMP）algorithm is proposed to solve the pathological equation in weight determination due to the decrease in atomic matching accuracy.The proposed algorithm adds the clustering contraction mechanism in the atomic matching process thus improves the behavior of the support set.Both simulation and experimental studies have verified that the proposed Cc St OMP can extract bearing fault feature more precisely,so as to achieve bearing fault diagnosis accurately.For the l₁-norm constrained problem,Iterative Shrinkage Thresholding（IST）algorithm is applied to reconstruct the fault feature components.Due to noise interference,the reconstructed signal is prone to over-fitting.Considering that bearing fault signal has obvious Sparse Within/Across Group（SWAG）property,Local-enhanced Iterative Shrinkage Thresholding（Le IST）algorithm is proposed in this paper.By enhancing the local sparse constraints in the sparse optimization function,this algorithm has better sparse recovery capabilities.Both simulation and experimental studies have verified that the proposed algorithm shows good anti-noise capabilities and can realize more accurate diagnosis of bearing fault.