Diagnosis when there are temporal dependencies and uncertainty using a cost bounded possibilistic ATMS

An incremental approach for generating multiple fault explanations when the system behaviour model is incomplete or contains temporal dependencies has been developed using a cost bounded ATMS as an underlying implementation mechanism. An extension of the basic cost bounded ATMS suitable for cases when the incompleteness is modelled using possibilistic logic and the temporal dependencies are propagation delays in the system is described. The possibilistic cost bounded ATMS is integrated into a diagnostic system where there is uncertainty, precedence relationships and temporal delays along different causal paths. Electrical paths which are almost instantaneous, for example, can cause identical symptoms to thermal paths which may have considerable time lags. Uncertain and temporal information are used to discriminate hypotheses. Some experimental timings of the basic algorithm are described. An advantage in using a CBATMS is that explanations, which can include multiple fault explanations, can be generated incrementally. Initially only explanations below a chosen cost bound are generated and further explanations are generated, if required, by increasing the cost bound. In many diagnostic applications the structure of the model can be such that the cost of explanations for the symptoms increase monotonically as explanations are generated in the (CB)ATMS network. Even if the condition for monotonicity is not true, in can be possible to restructure the model to ensure that it is. The CBATMS uses monotonic increase in costs to control the blocking and unblocking of propagation