ISSI Institute of Control & Computation Engineering


In the context of fault detection and isolation, the research is concerned with the investigation of both analytical and soft computing methods. Classic mathematical tools used in traditional Fault Detection and Isolation (FDI) are very sensitive to modelling errors, parameter variations, noise and disturbances. Thus, the main research attention is focused on eliminating these undesirable phenomena. The emphasis is put on soft computing methods and their integration with analytical approaches. In particular, various neural networks, fuzzy logic and genetic programming-based approaches are considered for both static and dynamic non-linear systems. An important research direction is concerned with experimental design performed in order to minimise model uncertainty, i.e. the mismatch between the system and the model. In particular, such a strategy was effectively applied to neural networks resulting in more reliable and accurate neural models than those designed without it. Knowledge regarding (possibly small) model uncertainty is especially important from the point of view of model-based fault diagnosis. It makes it possible to design a fault detection and isolation scheme, which minimises the probability of false alarms as well as undetected faults resulting in severe economical losses. This implies that neural networks designed with the above-mentioned experimental strategy can serve as an effective tool for fault detection.
The proposed solutions were examined and tested using various benchmark problems, e.g. a laboratory two-tank system with a delay line, a computer simulator of a power plant station and real data from the sugar evaporator.
In order to increase the reliability of fault diagnosis, several approaches were developed to design robust fault detection observers for non-linear systems. In particular, a number of relatively simple design procedures, which guarantee the convergence of the resulting observer, were proposed for non-linear deterministic systems.
Fault isolation problems are also investigated using neural, fuzzy, neuro-fuzzy and analytical classifiers. Based on expert system techniques, the integration problem of quantitative and qualitative methods was investigated. The research activities in fuzzy logic applications are mainly oriented towards developing effective methods of fuzzy knowledge representations in expert systems, the adaptation of fuzzy norms for fault classifiers, developing fuzzy model structures as well as fuzzy logic techniques in knowledge data discovery.
Recent research activity is concerned with the integration of fault diagnosis and control within the unified Fault-Tolerant Control (FTC) framework. In particular, special attention is focused on fault identification with both analytical and soft computing techniques.