Research On Mechanical Compound Faults Diagnosis Based On Non-negative Matrix Factorization

In the fault diagnosis and condition monitoring of rotating machinery,there are often multiple sources of faults.Due to the mutual interference between multiple fault source signals and the noisy signal acquisition environment in the real field,it increases the difficulty of fault diagnosis and identification.At the same time,the blind separation of multi-source coupling fault signals under sensor-limited underdetermined conditions is also particularly difficult.Therefore,the research on mechanical equipment fault diagnosis based on non-negative matrix factorization was carried out,and applies it to the fault diagnosis of multi-source coupling signal separation.The main contents are presented as follows:(1)An underdetermined blind source separation algorithm based on EVMD-LNMF is proposed.The implementation of the VMD algorithm is analyzed.The energy convergence factor is constructed for the shortcomings of the VMD algorithm.The optimal number of modal components in the VMD algorithm is adaptively determined,which solves the problem that the number of modal components is difficult to determine.Meanwhile,the objective function of the NMF algorithm is studied.According to the characteristics of coupling signals,the LNMF algorithm with locality is selected.For the selection of feature dimension r in LNMF algorithm,the proximity eigenvalue dominant method is used to obtain the LNMF algorithm,and the optimal decomposition dimension is used to estimate the number of coupling fault signal sources,which realizes the dimensionality reduction of multi-modal components accurately.Finally,the above two are combined,and the advantages of two algorithms is used to separate and extract the feature information of the compound faults.(2)A signal separation method based on constrained enhancement sparse non-negative matrix factorization algorithm is studied.Firstly,the basic theory of short-time Fourier transform is studied,and the important influence parameter about window function is analyzed.According to the characteristics of window function,the better Sinebell window function for processing compound fault signals is selected.Secondly,a constraint reference vector is introduced in the sparse non-negative matrix factorization algorithm,which can be generated adaptively according to the source signal.The output separated signal is constrained by the vector,and the vector will update according to the feedback of the separated signal.The redundancy of the mixture signal will be reduced during the constantly updating of the vector.Finally,the compound fault features can be separated and extracted automatically by using the sparse property of the improved sparse non-negative matrix factorization algorithm,and realize the compound faults diagnosis of mechanical equipment.(3)A model of the multi-faults diagnosis method based on multi-constrained non-negative matrix factorization algorithm is constructed.According to the characteristics of the signal generated when multiple faults occur,the corresponding constraints are introduced in the traditional non-negative matrix factorization algorithm to form a multi-constrained non-negative matrix factorization algorithm model.By utilizing the respective advantages of various constraints,different source signal feature components can be separated,and intelligent diagnosis of multiple faults in the rotating mechanical equipment can be realized.The results about the simulation and the experimental signal of the rolling bearing show that the multi-constraint non-negative matrix factorization algorithm can solve the separation and extraction of the signal information of multiple fault sources effectively,and realize the problem of blind source separation of multi-source coupling signals.