Condition Monitoring And Vibration Analysis Of Shaft Misalignment Using Discrete Wavelet Transform
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The condition monitoring (CM) of all kind of machines is inevitable task to keep healthy operating condition of entire plant. The condition based health monitoring involves faults viz. unbalance, misalignment, crack, looseness, insufficient lubrication etc. The Industry 4.0 insists to implement all process and CM activity based on ML and IoT. The literature survey reveals that, less work is contributed in ML based CM of shaft misalignment of rotary machines.
newline
newline
newlineIn this work, ML based condition monitoring of shaft misalignment is focused. The ANN-SVM based classification and prediction of shaft misalignment (CPSM) approach is proposed. In this method, kurtosis feature of normalized vibration signals are used for classification and measure of misalignment. The different mother wavelet viz. DB1, DB 2, SYM 2, BIOR
newline1.1, BIOR 1.3, COIF 2, COIF 3, RBIOR 1.3 and DMEY are compared at different level. The Shannon entropy based approach is used for selection of most suitable mother wavelet by comparing it to various levels of signal decomposition. The DB2 mother wavelet at first level decomposition is selected as most useful mother wavelet for feature selection. The DB2
newlineMother wavelet is used for selection of different features like Min, Max, RMS, En, Skewness and Kurtosis. The ReliefF algorithm is used to decide rank of selected features. The kurtosis feature is selected as first ranked feature. The classification and prediction accuracy of kurtosis alone was observed 19.8 % for misalignment fault. Therefore, to improve classification accuracy top eight ranked features are combined with kurtosis and tested. The combination of first and eighth i.e. kurtosis with Min_1X combination has shown 89.7 % classification accuracy for classification and prediction misalignment. The accuracy obtained in implementation of Fuzzy Logic
newline
newlineSystem (FLS) is 84%, Artificial Neural Network (ANN) is 94.17 % and average accuracy of combine artificial neural network and Support Vector Machine (SVM) is 97%.
newline
newline
newlineThe combined ANN-SVM based Classificati