Fault diagnosis based on a qualitative bond graph model, with emphasis on fault localisation

Fault diagnosis of engineering systems requires finding the primary components causing the discrepancies between normal behaviour and observed abnormal behaviour of a system. An approach to diagnosis is based on an underlying model of a devices´s structure and behaviour. However, the availability of appropriate system description models that explicitly offer all the information required for diagnosis is the key to the success of this methodology. In this paper, a formal modelling scheme integrating qualitative reasoning with bond graphs is proposed to generate qualitative deep models to represent system structure and to predict system behaviour for diagnosing system failures. Moreover, a universal inference mechanism is developed to locate fault candidates an measurement selection via reasoning from the qualitative proposed system structure and observed abnormal behaviour.