Nuclear in-core fuel reload design: The trajectory of a sequence of projects

Abstract

Between 1986 and 2000, I have been involved in a sequence of projects, FUELCON and FUELGEN, applying artificial intelligence techniques to an economically important problem in nuclear engineering: how to design refuellings, i.e., the replacement of spent fuel and rearrangement of other fuel assemblies, inside the core of a nuclear reactor, so that power generation be efficient, and that the cycle be usefully long. Design used to consist of a fuel manager manually reshuffling fuel from given categories inside a grid representing a symmetrical slice of a reactor core section. This process was automated by the FUELCON expert system, by means of a ruleset. This was a breakthrough in incore fuel management. Then, after probing into the possibility of further automation in a hybrid architecture combining rulesets and neural computing, a different direction was taken: in FUELGEN, evolutionary computing was applied, and for a while, the results reported were at the forefront. Since then, the application to refuellings of other techniques from artificial intelligence has been reported. One derives the impression that the industry is now relatively satisfied with efficiency levels, so we no longer see the pioneering spirit of the 1990s, but nevertheless, there appear several relevant papers for novel situations or contexts, or with novel techniques or configurations of techniques, so the problem is still not considered solved.