F2008-08-107
Risk Estimation on the Basis of Fault-Identification and -Propagation
Abstract: An unexpected event or behavior ("symptom") of a system may lead to a serious state. Knowing the cause(s) of the symptom enables in general to predict (or at least to estimate) whether a serious state will occur and if so when this will be. In this context there are three main tasks to be performed: Firstly, fault-recognition and -identification, which means monitoring a system, recognizing an abnormal behavior and deducing the possible fault-space. Secondly, on the basis of the possible causes (explaining the symptoms), the potential effects (besides the observed symptoms) are to be calculated (propagated). Lastly, the probabilities of all the effects are to be calculated, taking into account the probability of the different causes in the calculated fault-space. The probabilities of the hazardous effects constitute the over-all risk estimated on the basis of the observed symptoms.
The tasks of fault-recognition and -identification are reflected in serious challenges in model-based-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.
(New and innovative aspects :)
In the approach to be presented here a (formal) petrinet-system model can automatically be enriched by "special" mappings that are appropriate to perform diagnoses. These mappings are called adjoint mappings and can be interpreted as dual mappings. With the help of these mappings it is possible to apply a fixpoint-algorithm to narrow down the uncertainties of the knowledge of the (global) system state and to identify the possible fault space. In the same model it is then possible to propagate the faults and in that way to specify the possible space of effects. In the last step the probabilities for the various faults (or fault-combinations) in the fault space and their effects are estimated. This estimation bases on the calculated quantity of the faults (e.g. the extend of a whole in the air path) and on expert knowledge.
After introducing the underlying general theory and method outlined above, a specific application from the automotive-area is presented.
Session: Accident Analysis II