Improved Multi-Scale Entropy And Apply To Fault Feature Extraction And Diagnosis Of Rotating Machinery

Rotating machinery plays an irreplaceable role in the modern manufacture system.With the development of science and technology,due to every component of mechanical equipment are tightly coupled,chain effect will be happened if one part fails to work properly.As one of the most important parts in rotating machinery,it is necessary to diagnose the operating status of rolling bearing and hydraulic pump.Generally,the main steps of fault diagnosis are acquiring the vibration signals,fault feature extraction and working condition recognition.The second step is the core which was deeply studied in this paper.The fundamental principle and algorithm of Multi-Scale Entropy(MSE)was deeply studied.The effectiveness of the proposed method in revealing the signal complexity had been verified through simulated signals and experimental data.An improved algorithm was proposed to solve the stability of the sample entropy decreases and the sample entropy of different time sequence under the same condition.The effectiveness of the proposed method had been verified through simulated signals and experimental data.In this paper,a fault identification method based on MSE was proposed.Firstly,MSE was used to calculated the vibration signal.And the optimal scale factor was selected by ReliefF.Secondly,Partial Mean of Multi-Scale Entropy(PMMSE)was used to calculated the vibration signal.Then feature vectors of different signal types were extracted.Finally,feature vector sets of training samples were clustered by Kernel Fuzzy C-Means(KFCM),clustering centers of different signal types were obtained,the principle of the minimum Euclidean distance was adopted as the recognition method of feature vector sets of testing samples,and the effectiveness of the proposed method was verified through the bearing data published by Case Western Reserve University.Selecting hydraulic pump fault simulation test bench as the research object,the signal acquisition system was build based on LabVIEW.Using the proposed method,diagnosis of sliding shoes and loose slipper fault in different levels were achieved.In the same situation of diagnose data,the clustering results and fault recognition rates were compared between MSE and traditional time-frequency feature extraction method,the superiority of MSE was verified.