Abstract
KEYWORDS - Risk, Diagnosis, Formal Modelling, Petri-net, Fault
Not only in the automotive industry new requirements concerning aspects related to costs, emission and performance of systems are reflected in serious challenges in model-based and especially in model-based on-board diagnosis. A number of more or less formal methods have already been proposed. All these methods have in common, that models for different fault cases are built often independently from the system's model. In addition, combinations of faults are (at least generally) not taken into account.
The essential part of the stepwise approach to be presented here is a formal Petri-net model of the system to be examined. This model can automatically be enriched by special mappings that enable the model not only to predict the system states that are reached in the future. On the basis of observations (= observed symptoms) the potential fault-spaces (= spaces of causes) can be calculated within this model and reasonable (combinations of) faults can be identified. Therefore the modelling of (at best "semiformal" and) in general fault specific models is omitted in favour of an automatically generated formal (and global) diagnosis model.
This model can be simulated in forward as well as in backward direction and is used in step 1 and step 4 of the stepwise approach that will be presented here:
1. Fault identification: The possible fault space is calculated on the basis of observed symptoms and through backward simulation. All points in this space represent fault combinations that are in the position to explain the observed symptoms.
2. Fault estimation: The identified fault combinations are given occurrence probabilities.
3. Fault prediction: As faults are seldom constant in their size but change in time (e.g. the size of a leak may get larger), the changes of the identified faults in time are estimated.
4. Fault propagation and risk estimation: On the basis of these estimations, the expected consequences, i.e. the expected observations of symptoms are predicted. Through estimating the probabilities of the occurrences of faults in step 2 and their evolution in step 3, the probabilities of the expected observations of symptoms have been estimated implicitly, too.
The above mentioned special mappings - in fact these are "adjoint mappings", a well known concept in the theory of dual spaces - and their application within high-level Petri-net models as a basis to do diagnoses, are the essential but not domain specific concepts of the outlined approach.