Bayesian Networks for Fault Detection under Lack of Historical Data

In this paper we propose a Bayesian Network approach as a promissory data fusion technique for surveillance of sensors accuracy. We prove the usefulness of this method even in case when there is not enough feasible data to construct the model in traditional way. In presence of this data constrains we suggest an inversion of the causal relationship. This approach proves to be a possible solution to help the expert in the conditional probabilities assessment process. As a result a working model is constructed what would not be possible using traditional Bayesian Network approach.