Fault diagnosis of discretely controlled continuous systems by means of discrete-event models

Discretely controlled continuous systems consist of continuous plants whose operation mode is switched by a feedback controller. Fault diagnosis has to use the measured switching sequence and the measured continuous movement to detect and identify faults. In order to get the diagnostic algorithm with the least possible complexity, the kind of measurement information and the granularity of the model have to be chosen in accordance with the faults to be detected. The paper presents five diagnostic methods in a uniform way, which differ with respect to the model and the measurement information used. From the hybrid model of discretely controlled continuous systems, four more abstract representations are derived, which have the form of embedded maps, semi-Markov processes, timed automata and nondeterministic automata. The validity of the diagnostic result is ensured by the claim that the models should be complete and, hence, consistent with all the input-output sequences of the discretely controlled system in the appropriate fault case. In this way a hierarchy of models and of diagnostic results is obtained. The methods are illustrated by an example.