Uncertainty representation in practical decision support systems for the field service of large systems

The application of probabilistic reasoning in building a practical decision support system for servicing large physical systems is described. Certainty factors (CFs) with probabilistic semantics reduce both the representational and the computational complexities of probabilistic reasoning. When the modularity assumption is violated, however, their use results in counterintuitive beliefs. To overcome this problem, context-dependent CFs must be computed. Qualitative conditions that context-dependent CFs should satisfy are derived. These derivations assume a sub-synergistic influence of causes on effects, which is typical in physical systems. These qualitative conditions admit many solutions of context-dependent CFs; therefore, the choice of an exact solution is arbitrary. Experimental results show improvement in the quality of updated beliefs with respect to the modularity assumption