In this work we present strategies for (optimal) measurement selection inmodel-based sequential diagnosis. In particular, assuming a set of leadingdiagnoses being given, we show how queries (sets of measurements) can becomputed and optimized along two dimensions: expected number of queries andcost per query. By means of a suitable decoupling of two optimizations and aclever search space reduction the computations are done without any inferenceengine calls. For the full search space, we give a method requiring only apolynomial number of inferences and guaranteeing query properties existingmethods cannot provide. Evaluation results using real-world problems indicatethat the new method computes (virtually) optimal queries instantlyindependently of the size and complexity of the considered diagnosis problems.
展开▼