Diagnosis in general is the task to map symptoms to faults, it is thus denoted as symptom-fault classification in the general overview on the fault management page. In my work, I have been using a fuzzy logic based fault diagnosis approach, which is explained in the following.
Tha above figure nicely explains the individual steps which must be taken for a fuzzy-logic based fault diagnosis. The individual symptoms which are supplied at the left are first subject to fuzzification. This fuzzification is necessary since the models employed for the model-based feature extraction are typically tainted with an unavoidable uncertainty. The symptom is typically mapped to classes such as "normal", "increased", and "decreased".
A fault is typically characterized by more than one symptom, thus all symptoms which are characteristic for a certain fault are combined by a "Fuzzy-AND", which is realized by the minimum operator. This is the so-termed aggregation. Each symptom combination determines the possibility of a certain fault. The existence of a certain fault is indicated by a singleton whose value is between zero and one. Depending on the possibility as calculated by a certain rule, the correspoding singleton is activated. This is denoted as implication.
All individual faults are then combined (accumulation). The defuzzification is carried out by a simple maximum operation. Here, the fault with the highest possibility is assumed to be the fault most likely present in the system.