Optimization of focusing neutronic devices using artificial intelligence techniques

Abstract

The successful use is reported of a particle-swarm optimization algorithm to design a focusing, multi-channel neutron guide for the measurement of millimetre- and sub-millimetre-sized samples. For a 5 Å incident neutron wavelength on an IN5-type instrument, this results in a ninefold gain in the peak neutron count rate, and around an eightfold average gain in the count rate over the crucial 3–6 Å wavelength range, averaged over a 2 × 2 mm sample. A particle swarm method and a genetic algorithm were compared for simple neutron flux maximization, and the particle swarm was found to be faster for these kinds of problems. The focusing device was then designed by coupling the particle swarm algorithm to a full Monte Carlo neutron ray-tracing system. This realizes the `holy grail' of autonomous, self-optimizing virtual neutron devices based on life processes. The end result is superior to the manual (human) design of a focusing guide, and the design can be entirely re-optimized within a few days if the design requirements for a specific instrument should change.