Karlsruhe Institute of Technology, 2018 - 133 p.
The generation of multigroup cross-section libraries is a key point of multigroup transport calculations: a larger number of energy groups promotes the accuracy of the results, but hinders the time performance, which is a problem especially for 3D transient cases. Hence the need arises for multigroup deterministic transient calculations that, with a very limited amount of groups, can adequately represent the whole continuous energy space.
A new neutron cross-section collapsing tool has been implemented into the mechanistic codes SIMMER-III and SIMMER-IV, which introduces a cross-section condensation of the input multigroup libraries, which can then be provided with a finer structure than the one actually used by the transport solver. In this way the results are more accurate, as the nuclear data provided as input are closer to the original ones, but the computational time does not increase dramatically, as the transport solver will operate on a more limited number of energy groups.
A question, however, stays open: the determination of the energy discretization that best suits to the problem. This important issue, except for a few authors, has been considered mostly from the empirical point of view in the past, and required long tests to find out a reasonable energy structure, which is specific for the considered reactor and might be unsuitable for other systems.
This thesis proposes an automatic procedure, based on genetic algorithm optimization, aiming to choose the most appropriate energy structure for the considered system to collapse a fine multigroup library into a few-groups one. Such an innovative tool, used together with the cross-section condensation technique mentioned above, allows having specific libraries for each considered case, starting from a unique general library with fine energy discretization. The tests performed with different initial cross-section libraries and reactor systems show the strength of the technique in solving the energy meshing problem in many different conditions, returning structures which take into account the peculiar needs of each system. In addition, the analysis of the algorithm choices may reveal important information on neutron population physics, whose relevance during a manual cross-section library preparation can be easily underestimated. Tests show that the algorithm is able to find representative energy structures, providing accurate results on the multiplication factor, the reactivity feedback coefficients and the reaction rates. The results of each test are analysed, showing how different compositions, geometries and neutron spectra guide the algorithm choices and demonstrate the effectiveness of the method.