Search:
Computing and Library Services - delivering an inspiring information environment

Enhancement of Condition Monitoring Information from the Control Data of Electrical Motors Based on Machine Learning Techniques

Alabied, Samir (2020) Enhancement of Condition Monitoring Information from the Control Data of Electrical Motors Based on Machine Learning Techniques. Doctoral thesis, University of Huddersfield.

[img]
Preview
PDF - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (9MB) | Preview

Abstract

Centrifugal pumps are widely used in many manufacturing processes, including power plants, petrochemical industries, and water supplies. Failures in centrifugal pumps not only cause significant production interruptions but can be responsible for a large proportion of the maintenance budget. Early detection of such problems would provide timely information to take appropriate preventive actions.

Currently, the motor current signature analysis (MCSA) is regarded to be a promising cost-effective condition monitoring technique for centrifugal pumps. However, conventional data analysis methods such as statistical and spectra parameters often fail to detect damage under different operating conditions, which can be attributed to the present, limited understandings of the fluctuations in current signals arising from the many different possible faults. These include the fluctuations due to changes in operating pressure and flow rate, electromagnetic interference, control accuracy and the measured signals themselves. These combine to make it difficult for conventional data analyses methods such as Fourier based analysis to accurately capture the necessary information to achieve high-performance diagnostics.

Therefore, this study focuses on the improvement of data analysis through machine learning (ML) paradigms for promoting the performance of centrifugal pump monitoring. Within the paradigms, data characterisation methods such as empirical mode decomposition (EMD) and the intrinsic time-scale decomposition (ITD) reveal features based purely on the data, rather than finding pre-specified similarities to basic functions. With this data-driven approach, subtle changes are more likely to be captured and provide more effective and accurate fault detection and diagnosis.

This study reports the application of two of the above data-driven approaches, using MCSA for a centrifugal pump operated under normal and abnormal conditions to detect faults seeded into the pump. The research study has shown that the use of the ITD and EMD signatures combined with envelope spectra of the current signals proved to be competent in detecting the presence of the centrifugal pump fault conditions under different flow rates. The successful analysis was able to produce a more accurate analysis
of these abnormal conditions compared to conventional analytical methods. The effectiveness of these approaches is mainly due to the inclusion of high-frequency information, which is largely ignored by conventional MCSA.

Finally, a comprehensive diagnostic approach is suggested based on the support vector machine (SVM) as a diagnosing method for three seeded centrifugal pump defects (two bearing defects and compound defect outer race fault with impeller blockage) under different flow rates. It is confirmed that this novel data-driven paradigm is effective for pump diagnostics. The proposed method based on a combined ITD and SVM technique for extracting meaningful features and distinguishing between seeded faults is significantly more effective and accurate for fault detection and diagnosis when compared with the results obtained from other means, such as envelope, EMD and discrete wavelet transform (DWT) based features.

Item Type: Thesis (Doctoral)
Subjects: T Technology > T Technology (General)
T Technology > TK Electrical engineering. Electronics Nuclear engineering
Schools: School of Computing and Engineering
Depositing User: Christine Morelli
Date Deposited: 08 Dec 2020 14:08
Last Modified: 08 Dec 2020 14:15
URI: http://eprints.hud.ac.uk/id/eprint/35263

Downloads

Downloads per month over past year

Repository Staff Only: item control page

View Item View Item

University of Huddersfield, Queensgate, Huddersfield, HD1 3DH Copyright and Disclaimer All rights reserved ©