Fault Detection Schemes for Dynamical Systems

Mohamadi, Lamine (2018) Fault Detection Schemes for Dynamical Systems. Doctoral thesis, Northumbria University.

Text (Doctoral thesis)
Mohamadi.Lamine_phd.pdf - Submitted Version

Download (1MB) | Preview


In the current digital age where automatic controlled systems are used in many fields such as industrial plants, means of transports and domestic electronics applications, the issue of safety and reliability has become of paramount importance. With that came the need to develop techniques to be implemented in control systems that would allow monitoring those systems and detect if some malfunctions or abnormalities are occurring. Fault detection (FD) emerged as one of the most widely used solutions to this issue and the so-called model based fault detection has received a lot of attention. In this approach, a model of the target system is involved to estimate the expected output of the system under healthy condition and then a fault can be detected by comparing the actual measured output to the estimated healthy output. By making use of the state estimation capability of observers, various observer-based fault detection schemes have been proposed to estimate the system output for the purpose of fault detection.

However, it is worth noting that, state observers have been designed for state estimation and have found use in being adapted for fault detection. In order to try to answer the increasing requirements on systems performances, this thesis focuses on developing observers to be used specifically in FD schemes and for which a systematic way to be designed is proposed.

The first solution based on proportional integral (PI) observer, relies on integrating the systems output to construct an augmented model. This technique has a double effect on achieving better fault detection performances. Indeed, disturbances effect is reduced by using a technique based on replacing part of the model information in the observer design. Besides, it allows having an additional degree of freedom when optimising the observer design to detect faults.

The second solution to the FD problem, a new type of observers, referred to as output observer is proposed for fault detection in both linear and nonlinear systems, where unnecessary state estimation in observer-based fault detection can be avoided. First, an input/output system representation, upon which the output observer design is based on, is introduced. Then, a new approach of output observer design, in which only the output variables are estimated, is developed. The convergence of the observer with respect to arbitrary initial conditions is proved and the fault detectability capabilities of the scheme are established. Another benefit of the proposed output observer design is the output injection feature, where the measured output is directly injected in the observer so as to linearise the estimation error dynamics. This feature is fundamental to the solution proposed in this thesis to deal with nonlinearities in the system's model so that the inclusion of those nonlinearities when tuning the observer is avoided. Furthermore, as time delays are ubiquitous in systems, and are one of the most important sources of estimation errors, a solution based on output injection is also proposed to set the condition that ensures the convergence of the observer while maintaining its output estimation performances.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: Observer based fault detection, Observer design, Output observer, Nonlinear systems, Time delay in systems
Subjects: H600 Electronic and Electrical Engineering
Department: Faculties > Engineering and Environment > Mathematics, Physics and Electrical Engineering
University Services > Graduate School > Doctor of Philosophy
Depositing User: Paul Burns
Date Deposited: 21 Jun 2019 16:16
Last Modified: 16 Sep 2022 08:30
URI: https://nrl.northumbria.ac.uk/id/eprint/39790

Actions (login required)

View Item View Item


Downloads per month over past year

View more statistics