Identification Of Flow Blockage Levels In Centrifugal Pump By Machine Learning
Loading...
Date
2021-07-01
Authors
Ng, Woon Li
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
The current project focused on the flow blockage monitoring of centrifugal pump by vibration and acoustic analysis. The blockage of the pump inlet could result in cavitation or mechanical parts breakdown which would increase the maintenance cost. Machine learning can be used as a preventive measure to detect the blockage in the pump inlet at inception level. The purpose of this research is to develop an effective machine learning model for the classification of flow blockage levels in the centrifugal pump by using the statistically significant features from vibration and acoustic analysis. Raw data are collected by using Siemens LMS SCADAS Mobile data acquisition system. Furthermore, the statistical features are extracted and trained by multiple classifiers available in Classification Learner, MATLAB R2020b and validated by 5-fold cross-validation. RMS, variance and kurtosis of acoustic signals in time domain have strong predictive potential based on the Chi-Square test which ranked the predictive power of features for classification. Besides, Ensemble Bagged Tree and Support Vector Machine (SVM) with cubic kernel has achieved the highest accuracy, which are 95.8% and 94.4%, respectively. SVM model with cubic kernel is preferable as the training time taken is relatively lower than Ensemble Bagged Tree due to the ensemble algorithms are more complex. Hence, the SVM model with cubic kernel is exported as a MATLAB function to make predictions for the new data. Besides, the result of the Linear SVM model is validated by using the classification packages in Google Colaboratory. Several potential improvements are recommended in this report to increase the robustness and accuracy of the machine learning model.