Research And Application Of Rotating Machinery Fault Diagnosis Based On Estimation Of Underdetermined Blind Separation Mixed Martix

Rotating machinery is the key equipment in production.It is widely used in the fields of manufacturing,new energy and Aeronautics and Astronautics.Once a rotating machine fails,it will lead to production interruption and economic losses.Heavy mechanical accidents will happen and even casualties will happen.So the running state of the rotating machinery is monitored in real time so as to make a timely diagnosis of mechanical failure and take corresponding rescue measures to prevent the occurrence of safety accidents.The rolling bearing is an important part of the rotating machinery.When the rolling bearing fails,the vibration signal is nonstationary and often faces the condition of interference or multiple fault mixing.In this paper,taking the rolling bearing as the research object,in order to separate the mixed signal of the rolling bearing fault,the estimation method of the undetermined blind separation mixed matrix and its application in the fault diagnosis of the rolling bearing are studied.First,the development of blind source separation and the application of blind source separation in mechanical fault diagnosis are analyzed.The basic theory of blind source separation and underdetermined blind source separation is studied.The reasons for the effect of the blind separation of mixed signals are further analyzed,and the evaluation index of the blind source separation performance is given.Secondly,from the signal sparsity,the parameter solution method of signal purity is given,and the influence of signal sparsity and signal purity on the estimation of undetermined blind separation mixed matrix is studied,and the method of improving the purity of signal through single source detection is proposed.This method filters the single source points in the mixed signals,and compose the new signal samples,and then estimate the mixing matrix.The simulation experiment and the experiment of rolling bearing fault experiment verify that the single source point detection method can improve the purity of the signal obviously,thus improving the accuracy of the estimation of the mixed matrix.Finally,from the perspective of algorithm optimization,we study an underdetermined blind separation method based on the improved clustering by fast search and find ofdensity peaks(FSDPC).By calculating the density of each point in the mixed signal scatter plot,the point density is segmented by the Otsu to screen the target points,then the target points are normalized,and the center coordinates of the cluster are determined and converted into the mixed matrix.Finally,the L1 norm minimization method is used to separate the mixed signals.The experimental results show that the proposed method can estimate the mixing matrix under the condition of unknown source number and initial value of the cluster center.The underdetermined blind separation method based on the improved density peak clustering can effectively separate the multi fault signals of the bearing,and realize the fault recognition and diagnosis.