Research On Single Channel Blind Source Separation And Application In Fault Diagnosis For Pumping Unit

Single channel blind source separation is used widely in practical engineering field,and it is research hotspot in single processing field.Aiming at the problem that current single channel blind source separation algorithm cannot determine number of independent components in single channel and could not realize online separation,based on empirical mode decomposition and nonnegative matrix/ tensor factorization,this paper proposes two novel single channel blind source separation algorithms.Combined with Gaussian process modeling and particle filter,the online estimation for sources is achieved.A fault diagnosis system for pumping unit based on single channel blind source separation is designed and achieved.The principle work of this paper is described as follows:1.Mathematical model for single channel blind source separation is studied and a method using empirical mode decomposition and nonnegative matrix factorization to solve SCBSS is proposed.Single channel is decomposed into multiple intrinsic mode functions firstly,then eigenvalue method is used to determine independent component number in single channel signal.Multiple channel signals are reconstructed and underdetermined blind source separation problem is transformed into well-posed problem.Finally,nonnegative matrix factorization is used to recover original signals.2.A blind source separation algorithm based on nonnegative tensor factorization is realized.The concept and decomposition form of tensor is studied,and nonnegative tensor decomposition is proposed to applied in multi-channel signals separation.Information in time-domain and frequency-domain are fully used to obtain better separation performance.And a new single channel blind source separation algorithm is proposed.3.An online single channel blind source separation algorithm is proposed and achieved.Combined with original signals from SCBSS algorithm,aiming at the modeling for big data,state transition equation and measurement equation for single channel blind source separation are modeled using sparse pseudo-input Gaussian process modeling method.Then the particle filter algorithm is used to estimate original signals online according to observation.4.Based on the proposed single channel blind source separation algorithm,a fault diagnosis system for pumping unit has been achieved.The acoustic emission is collected using single sensor.Then single channel signal is processed using single channel blind source separation algorithm.And feature extraction and detection are realized at last.The effectiveness and practicability of proposed algorithm is validated through field test.A multichannel mapping method based on empirical mode decomposition is proposed and achieved.The method is combined with nonnegative matrix/ tensor factorization,achieving two single channel blind source separation algorithms.Meanwhile,the two methods provide basis for the building of state-space model,Gaussian process and particle filter are combined to implement the online separation.A set of fault diagnosis for pumping unit based on proposed separation algorithms is designed and achieved.Experimental results and field test show the proposed algorithm has great separation performance,providing a new method for single channel blind source separation.