The authors describe a diagnostic procedure aimed at identifying the most likely causes of malfunction of a composite device and at generating an optimal plan of action for repair, that is, a sequence of observations and repairs of individual components that allows one to minimize the expected costs (an optimal troubleshooting plan).
The proposed approach consists of a series of approximations to an exact method for a simple case, which allow one to identify an optimal sequence of observation and repair actions in a time proportional to the number of components in the device, without explicitly constructing and rolling back a decision tree, as is usual in developing an optimal solution to the general troubleshooting problem. These results are obtained by using a Bayesian network to compute the probabilities that components have failed.
The work is interesting, well described, and framed in the research context. The situations in which the procedure can be applied are clearly identified and of general interest. The formal description of the procedure is completed by the illustration of the results obtained in a series of experiments with real-world troubleshooting problems (printing problems, car start-up problems, copier feeder systems, and gas turbines). A comparison with the performance of other kinds of troubleshooting planners is also given. The method used to measure performance is illustrated in detail.
The paper is well written and well organized. More references to recent literature would have improved the paper.