Expert maintenance systems in telecommunications

An admissible heuristic algorithm for generating a network troubleshooting expert system that minimizes the expected troubleshooting cost and learns better troubleshooting techniques during its operation is presented. The algorithm requires three kinds of input: network topology, component failure rate, and relative costs of tests. It uses a modification of Huffman code that is proven to be an admissible heuristic on the A* search. Learning is shown to be facilitated by recomputing the optimal troubleshooting strategy when any of the inputs has changed. The algorithm is implemented in Prolog