ABSTRACT
This paper reports on an investigation [Hut96] into using an on-line artificial neural network (ANN) to diagnose real time transients as part of an advisor to operators of a complex system, a nuclear reactor. An ANN is first developed to successfully diagnose six plant conditions using a selected input set of twelve reactor variables for three time steps. Once developed this network is embedded in a reactor simulator program to produce on-line diagnosis. The results show that not only can the system successfully diagnose transients quicker than a human operator but it can produce some unexpectedly accurate diagnosis of multiple faults.
