"Rapid technological advances have led to more and more complex industrial systems with significantly higher risk of failures. Therefore, in this dissertation, a model-based fault diagnosis and prognosis framework has been developed for fast and reliable detection of faults and prediction of failures in nonlinear systems. In the first paper, a unified model-based fault diagnosis scheme capable of detecting both additive system faults and multiplicative actuator faults, as well as approximating the fault dynamics, performing fault type determination and time-to-failure determination, is designed. Stability of the observer and online approximator is guaranteed via an adaptive update law. Since outliers can degrade the performance of fault diagnostics, the second paper introduces an online neural network (NN) based outlier identification and removal scheme which is then combined with a fault detection scheme to enhance its performance. Outliers are detected based on the estimation error and a novel tuning law prevents the NN weights from being affected by outliers. In the third paper, in contrast to papers I and II, fault diagnosis of large-scale interconnected systems is investigated. A decentralized fault prognosis scheme is developed for such systems by using a network of local fault detectors (LFD) where each LFD only requires the local measurements. The online approximators in each LFD learn the unknown interconnection functions and the fault dynamics. Derivation of robust detection thresholds and detectability conditions are also included. The fourth paper extends the decentralized fault detection from paper III and develops an accommodation scheme for nonlinear continuous-time systems. By using both detection and accommodation online approximators, the control inputs are adjusted in order to minimize the fault effects. Finally in the fifth paper, the model-based fault diagnosis of distributed parameter systems (DPS) with parabolic PDE representation in continuous-time is discussed where a PDE-based observer is designed to perform fault detection as well as estimating the unavailable system states. An adaptive online approximator is incorporated in the observer to identify unknown fault parameters. Adaptive update law guarantees the convergence of estimations and allows determination of remaining useful life"--Abstract, page iv.
Sarangapani, Jagannathan, 1965-
Balakrishnan, S. N.
Dagli, Cihan H., 1949-
Zawodniok, Maciej Jan, 1975-
Electrical and Computer Engineering
Ph. D. in Electrical Engineering
National Science Foundation (U.S.)
Missouri University of Science and Technology
Journal article titles appearing in thesis/dissertation
- A Unified Model-Based Fault Diagnosis Scheme for Nonlinear Discrete-Time Systems with Additive and Multiplicative Faults
- An Online Outlier Identification and Removal Scheme for Improving Fault Detection Performance
- Decentralized Fault Diagnosis and Prognosis Scheme for Interconnected Nonlinear Discrete-Time Systems
- A Decentralized Fault Detection and Accommodation Scheme for Interconnected Nonlinear Continuous-time Systems
- Fault Diagnosis of a Class of Distributed Parameter Systems Modeled by Parabolic Partial Differential Equations
xi, 151 pages
© 2013 Hasan Ferdowsi, All rights reserved.
Dissertation - Open Access
Library of Congress Subject Headings
Electric fault location -- Mathematical models
Adaptive control systems
Integrated circuits -- Fault tolerance
Detectors -- Reliability
System failures (Engineering)
Electronic OCLC #
Ferdowsi, Hasan, "Model based fault diagnosis and prognosis of nonlinear systems" (2013). Doctoral Dissertations. 1829.