Diagnostic and prognostic of intermittent faults (by use of machine learning).
| dc.contributor.advisor | Foote, Peter D. | |
| dc.contributor.advisor | Starr, Andrew | |
| dc.contributor.author | Sedighi, Tabassom | |
| dc.date.accessioned | 2023-09-20T14:35:05Z | |
| dc.date.available | 2023-09-20T14:35:05Z | |
| dc.date.issued | 2020-02 | |
| dc.description.abstract | This thesis investigates novel intermittent fault detection and prediction techniques for complex nonlinear systems. Aerospace and defence systems are becoming progressively more complex, with greater component numbers and increasingly complicated components and subcomponents. At the same time, faults and failures are becoming more challenging to detect and isolate, and the time that operators and maintenance technicians spend on faults is rising. Moreover, a serious problem has recently attracted a lot of attention in health diagnostics of these complex systems. Detecting intermittent faults that persist for very short durations and manifest themselves intermittently have become troublesome and sometimes impossible (also known as “no fault found”). In response to the above challenges, this thesis focuses on the development of a novel methodology to detect intermittent faults of these complex systems. It further investigates various probabilistic approaches to develop efficient fault diagnostic and prognostic methods. In the first stage of this thesis, a novel model (observer)-based intermittent fault detection filter is presented that relies on the creation of a mathematical model of a laboratory scale aircraft fuel system test rig to predict the output of the system at any given time. Comparison between this prediction of output and actual output reveals the presence of a fault. Later, the simulation results demonstrate that the performance of the model (observer)-based fault detection techniques decrease significantly as system complexity increases. In the second stage of this research, a probabilistic data-driven method known as a Bayesian network is presented. This is particularly useful for diverse problems of varying size and complexity, where uncertainties are inherent in the system. Bayesian networks that model sequences of variables are called dynamic Bayesian networks. To introduce the time variable in the framework of probabilistic models while dealing with both discrete and continuous variables in the fuel rig system, a hybrid dynamic Bayesian network is proposed. The presented results of data-driven fault detection show that the hybrid dynamic Bayesian network is more effective than the static Bayesian network or model (observer)- based methods for detecting intermittent faults. Furthermore, the second stage of the research uses all the information captured from the fault diagnostic techniques for intermittent fault prediction by using a probabilistic non-parametric Bayesian method called Gaussian process regression, which is an aid for decision-making using uncertain information. | en_UK |
| dc.description.coursename | PhD in Manufacturing | en_UK |
| dc.description.sponsorship | Engineering and Physical Sciences (EPSRC) | en_UK |
| dc.identifier.uri | https://dspace.lib.cranfield.ac.uk/handle/1826/20262 | |
| dc.language.iso | en | en_UK |
| dc.publisher | Cranfield University | en_UK |
| dc.publisher.department | SWEE | en_UK |
| dc.rights | © Cranfield University, 2020. All rights reserved. No part of this publication may be reproduced without the written permission of the copyright holder. | en_UK |
| dc.subject | No fault found | en_UK |
| dc.subject | intermittent fault | en_UK |
| dc.subject | model-based fault detection | en_UK |
| dc.subject | nonlinear systems | en_UK |
| dc.subject | nonlinear observer | en_UK |
| dc.subject | Bayesian methods | en_UK |
| dc.subject | hybrid dynamic Bayesian network | en_UK |
| dc.subject | Gaussian process regression | en_UK |
| dc.title | Diagnostic and prognostic of intermittent faults (by use of machine learning). | en_UK |
| dc.type | Thesis or dissertation | en_UK |
| dc.type.qualificationlevel | Doctoral | en_UK |
| dc.type.qualificationname | PhD | en_UK |