Blind Source Separation Algorithm Based On Reference Signal And Its Application In Rolling Bearing Fault Diagnosis

Blind source separation is a method that uses the statistical independence between source signals to recover all source signals only by observing signals,also known as independent component analysis.The signal recovered by this method has uncertainty in amplitude and arrangement order.In engineering applications,some characteristics of the source signal,such as carrier frequency and other prior information,are known.Adding these prior information to independent component analysis to obtain the constrained independent component analysis method,which can solve the problem of uncertain separation results of independent component analysis.During the operation of rolling bearing,its fault vibration signal is often submerged in strong noise.In view of this phenomenon,it is necessary to explore new methods to extract fault features.Fault characteristic frequency is a common prior information,which can be used in constrained independent component analysis and in rolling bearing fault diagnosis.The main contents of this paper are as follows:(1)At present,there is a large error between the signal of interest extracted by constrained independent component analysis and the source signal.To solve this problem,an improved constrained independent component analysis algorithm is proposed.Firstly,the reciprocal of the proximity function containing a priori information is coupled into the objective function to obtain a new objective function,and then the Lagrange optimization algorithm is applied to the new objective function and combined with the Newton Like method to obtain the optimal separation matrix,so as to separate the interesting signal.The simulation results show that the improved algorithm is better than the original algorithm in signal-to-noise ratio and mean square error.(2)The above improved constrained independent component analysis algorithm needs to set a threshold parameter to distinguish the signal of interest from other signals.It is difficult to set the threshold parameter,which largely depends on the application experience of constrained independent component analysis.Therefore,it is proposed to alternately optimize the reciprocal of the objective function and the proximity function containing a priori information to derive the optimal separation matrix,so as to separate the signal of interest.Simulation results show that the improved algorithm not only ensures that the signal-to-noise ratio and mean square error are better than the original algorithm,but also avoids the problem of setting threshold parameters.(3)The improved constrained independent component analysis algorithm combined with minimum entropy deconvolution and local mean decomposition is applied to rolling bearing fault diagnosis.Firstly,the minimum entropy deconvolution is used to reduce the noise of the collected rolling bearing fault mixed signal.Then,the denoised signal is processed by local mean decomposition,and the appropriate components are selected to participate in signal reconstruction.Finally,the reconstructed signal and reference signal are input into the improved constrained independent component analysis to extract the fault signal.In the above process,the priori information of the theoretical fault characteristic frequency of the inner and outer ring is used as the reference signal.The experimental results show that the prominent fault characteristic frequency in the envelope spectrum is consistent with the calculated theoretical fault characteristic frequency,which shows that the algorithm proposed in this paper is effective.