Date of final exam: 25/02/2000

E-mail: ssimani@ing.unife.it

Tutor: Prof.  S. Beghelli, Università di Ferrara

  ___________________________________________________________________________________________________________

Model-Based Fault Diagnosis in Dynamic Systems Using Identfication Techniques   ___________________________________________________________________________________________________________

Advisor:

Prof.  S. Beghelli, Università di Ferrara

Co-Advisor:

Prof.  R. Patton

Summary of the thesis

The control devices which are nowadays exploited to improve the overall performance of industrial processes involve both sophisticated digital system design techniques and complex hardware (input-output sensors, actuators, components and processing units). Such a complexity results in an increased probability of failure occurrence. As a direct consequence of this, control systems must include automatic supervision of the closed-loop operation to detect and isolate malfunctions as early as possible.
Since the early 1970's, the problem of fault detection and isolation in dynamic processes has received great attention and a wide variety of model-based approaches has been proposed and developed.
Model-based techniques have been widely recognized as powerful approaches for fault diagnosis and require a realistic mathematical model of the monitored system. An effective model-based fault diagnosis system should take into account noise and modeling uncertainties, always present in any real situation.
On the other hand, for the diagnosis of faults, mathematical models of the process under investigation are required, either in state space or input-output format.
A state space description of the system provides general and mathematically rigorous tools for system modeling and residual generation, which may be used in fault diagnosis of industrial systems, for both the deterministic and the stochastic case.
Residuals should then be processed to detect an actual fault condition, rejecting any false alarms caused by noise, spurious signals and modeling uncertainties.
This thesis aims to define a comprehensive methodology for actuator, component and sensor fault detection and isolation by using an output estimation approach. The method is based on residual processing schemes, which include a simple threshold detection, in deterministic case, as well as statistical analysis when data are affected by noise.
The final result consists of a fault detection and isolation strategy based on diagnostic methods to generate redundant residuals analytically. A number of strategies for the design of residual generators are then proposed.
The method proposed do not require any physical knowledge of the processes under observation since the mathematical description of the monitored system is obtained by means of a system identification scheme based on equation error and errors--in--variables models. The latter identification approach gives a reliable model of the plant under investigation, as well as the variances of the input--output noises affecting the data.
It is worthy to note how this work presents a novel point of view of the model-based fault diagnosis. The new aspect consists of exploiting linear system identification procedures in connection with the model-based residual generation problem.
The diagnostic tools presented in this thesis are well illustrated using practical application examples and the results show the effectiveness of the technique developed.
Finally, it is worthy to note how the main aspect of this work is the use of linear system identification and modelling methods, although the system considered are non--linear. This is considered important to avoid the complexities that would otherwise be inevitable when non--linear models are used. There is certainly an increasing interest in the use of non--linear methods (non--linear observers, extended Kalman filters, fuzzy-logic methods, etc). However, as the feature of system supervision is to monitor the operation and performance of the system with respect to an expected point of operation, linear system methods are still very valid. Deviations from expected behaviour can be used to minitor system performance changes as well as component malfunctions.
 

  _______________________________________