Transfer of Fault-Finding between Simulated Chemical Plants

Abstract

This study investigated the effect of training with either a qualitative model of plant functioning or a set of diagnostic heuristics on the transfer of fault-finding. Two simulated chemical plants were used. After training both conditions improve in accuracy of diagnosis of previously encountered faults, though the qualitative model condition is superior at transfer to novel faults. However, neither training condition exhibited positive transfer of training, in terms of accuracy, when tested on a second plant involving a recycle loop. Different reasons may explain this lack of transfer for each condition. Significantly, qualitative differences were found between the training conditions in the order in which plant variables were examined during fault-finding. These qualitative differences persisted in transfer to the second plant but only for faults whose symptoms were not affected by the recycle loop. Such qualitative measures offer a means of inferring the cognitive processes inculcated by different training programs.