Turbo-Generator Unit Vibration Fault Diagnosis Research Based on Space Optimization Algorithm of Bayesian Networks

Recursive conditioning, RC, is any-space algorithm for exact inference in Bayesian networks because any number of results may be cached. But given a limited amount of memory, which results should be cached in order to minimize the running time of the algorithm becomes a key question. Aiming at this problem, depth-first branch-and-bound method is proposed to search all the potential goal states, average running time under each state is computed, then some optimal time-space tradeoff curves were made. Through the curves, an optimal discrete cache scheme can be found, with this scheme, significant amounts of memory can be removed from the algorithm´s cache with only a minimal cost in time. Applying this algorithm in turbo-generator unit fault diagnosis, based on analysis of vibration fault of the turbine machinery, the fault sets, manifestation sets, relation table for fault and manifestations to turbine fault were given, Bayesian networks for fault diagnosis was made, on the basis of which, a fault in turbine was identified. The results show that with the space optimization algorithm of Bayesian networks, the fault can be accurately diagnosed, and at the same time, and the storage capacity is reduced. It is estimated that the optimized strategy may be further applied in fault diagnosis.