Cheap diagnosis using structural modelling and fuzzy-logic-based detection

Practical fault diagnosis can be based on simple, yet efficient, analysis of redundant information about the state of a plant, and diagnostic algorithms can be made without detailed and expensive modelling efforts. This paper shows how it is possible, using structural analysis, to find redundancy relations for linear or non-linear dynamic behaviour, and combine this with fuzzy output observer design to provide an effective diagnostic approach. An adaptive neuro-fuzzy inference method is used. A fuzzy adaptive threshold is employed to cope with practical uncertainty. The methods are demonstrated using measurements on a ship propulsion system subject to simulated faults.