Unsupervised Methods for Condition-Based Maintenance in Non-Stationary Operating Conditions
| dc.contributor.author | Fong, Stanley | |
| dc.date.accessioned | 2022-04-28T20:07:38Z | |
| dc.date.available | 2022-04-28T20:07:38Z | |
| dc.date.issued | 2022-04-28 | |
| dc.date.submitted | 2022-03-31 | |
| dc.description.abstract | Maintenance and operation of modern dynamic engineering systems requires the use of robust maintenance strategies that are reliable under uncertainty. One such strategy is condition-based maintenance (CBM), in which maintenance actions are determined based on the current health of the system. The CBM framework integrates fault detection and forecasting in the form of degradation modeling to provide real-time reliability, as well as valuable insight towards the future health of the system. Coupled with a modern information platform such as Internet-of-Things (IoT), CBM can deliver these critical functionalities at scale. The increasingly complex design and operation of engineering systems has introduced novel problems to CBM. Characteristics of these systems - such as the unavailability of historical data, or highly dynamic operating behaviour - has rendered many existing solutions infeasible. These problems have motivated the development of new and self-sufficient - or in other words - unsupervised CBM solutions. The issue, however, is that many of the necessary methods required by such frameworks have yet to be proposed within the literature. Key gaps pertaining to the lack of suitable unsupervised approaches for the pre-processing of non-stationary vibration signals, parameter estimation for fault detection, and degradation threshold estimation, need to be addressed in order to achieve an effective implementation. The main objective of this thesis is to propose set of three novel approaches to address each of the aforementioned knowledge gaps. A non-parametric pre-processing and spectral analysis approach, termed spectral mean shift clustering (S-MSC) - which applies mean shift clustering (MSC) to the short time Fourier transform (STFT) power spectrum for simultaneous de-noising and extraction of time-varying harmonic components - is proposed for the autonomous analysis of non-stationary vibration signals. A second pre-processing approach, termed Gaussian mixture model operating state decomposition (GMM-OSD) - which uses GMMs to cluster multi-modal vibration signals by their respective, unknown operating states - is proposed to address multi-modal non-stationarity. Applied in conjunction with S-MSC, these two approaches form a robust and unsupervised pre-processing framework tailored to the types of signals found in modern engineering systems. The final approach proposed in this thesis is a degradation detection and fault prediction framework, termed the Bayesian one class support vector machine (B-OCSVM), which tackles the key knowledge gaps pertaining to unsupervised parameter and degradation threshold estimation by re-framing the traditional fault detection and degradation modeling problem as a degradation detection and fault prediction problem. Validation of the three aforementioned approaches is performed across a wide range of machinery vibration data sets and applications, including data obtained from two full-scale field pilots located at Toronto Pearson International Airport. The first of which is located on the gearbox of the LINK Automated People Mover (APM) train at Toronto Pearson International Airport; and, the second which is located on a subset of passenger boarding tunnel pre-conditioned air units (PCA) in Terminal 1 of Pearson airport. Results from validation found that the proposed pre-processing approaches and combined pre-processing framework provides a robust and computationally efficient and robust methodology for the analysis of non-stationary vibration signals in unsupervised CBM. Validation of the B-OCSVM framework showed that the proposed parameter estimation approaches enables the earlier detection of the degradation process compared to existing approaches, and the proposed degradation threshold provides a reasonable estimate of the fault manifestation point. Holistically, the approaches proposed in thesis provide a crucial step forward towards the effective implementation of unsupervised CBM in complex, modern engineering systems. | en |
| dc.identifier.uri | http://hdl.handle.net/10012/18191 | |
| dc.language.iso | en | en |
| dc.pending | false | |
| dc.publisher | University of Waterloo | en |
| dc.subject | unsupervised learning | en |
| dc.subject | artificial intelligence | en |
| dc.subject | condition-based maintenance | en |
| dc.subject | condition monitoring | en |
| dc.subject | rotating machinery | en |
| dc.subject | signal processing | en |
| dc.subject | clustering | en |
| dc.subject | non-stationarity | en |
| dc.subject | degradation modeling | en |
| dc.subject | fault detection | en |
| dc.subject | machinery diagnostics | en |
| dc.title | Unsupervised Methods for Condition-Based Maintenance in Non-Stationary Operating Conditions | en |
| dc.type | Doctoral Thesis | en |
| uws-etd.degree | Doctor of Philosophy | en |
| uws-etd.degree.department | Civil and Environmental Engineering | en |
| uws-etd.degree.discipline | Civil Engineering | en |
| uws-etd.degree.grantor | University of Waterloo | en |
| uws-etd.embargo.terms | 0 | en |
| uws.comment.hidden | Hello, Thank you for reviewing my thesis submission. I have made the following corrections: 1) Copyright now appears on the title page 2) Updated the author's declaration accordingly 3) The publications in Appendix A are directly related to the thesis. I have added a line at the beginning of the Appendix to state this explicitly. If there are any other inconsistencies, please let me know. Thank you, Stan | en |
| uws.contributor.advisor | Narasimhan, Sriram | |
| uws.contributor.affiliation1 | Faculty of Engineering | en |
| uws.peerReviewStatus | Unreviewed | en |
| uws.published.city | Waterloo | en |
| uws.published.country | Canada | en |
| uws.published.province | Ontario | en |
| uws.scholarLevel | Graduate | en |
| uws.typeOfResource | Text | en |