Development of a surrogate model for simplified neutronic calculations involved in the design stage of a thermonuclear fusion reactor

Abstract

Several system codes have been developed since the eighties, with different objectives and appropriate architecture and level of development, aiming to explore possible operating condition ranges of a fusion power reactor. In some “system codes” technology/engineering assumptions/models (e.g. thermodynamic efficiency of coolant cycle, neutron multiplication coefficient, Tritium Breeding Ratio, radial built) are treated as input data inserted by the user and integrated in a main module essentially describing “plasma physics” aspects. In a first stage, these values come from previous studies on equivalent reactor concepts. Subsequent and more complete analyses with detailed models allow to confirm/deny these values and modify, if needed, for a second run and most likely several run, up to convergence. Some other codes consist of different specific modules (calculation tools) each one treating a separate aspects (e.g. physics, engineering, costing...) and integrated together in a common multiphysics calculation platform. Appropriate modules consistently calculate needed values using simplified models or surrogate models that enable an acceleration of the convergence of these systems codes. A system code based on this approach, SYCOMORE, is under development at CEA. The characteristics of the plasma and of the various reactor subsystems are addressed by various codes/models which are linked together via an integrated tokamak modelling platform. This platform allows creating a system design workflow by chaining the execution of the various modules. In this framework, this document describes a methodology developed to build the neutronic module of SYCOMORE: a surrogate model, based on neural network giving main neutronic parameters characterizing a fusion reactor (tokamak): tritium breeding ration (TBR), multiplication factor, and nuclear heating as a function of the reactor main geometrical parameters (major radius, elongation…), of the radial built, Li enrichment, blanket and shield thickness, etc. (3) In order to obtain a reliable surrogate model, a consistent database is needed. Simplified 1D and 2D neutronic calculation carried out with APOLLO2 (deterministic) and and TRIPOLI-4 (Montecarlo) codes codes are therefore used to fill the database. The URANIE platform is used to build the surrogate model from neutronic results. The simplified 1D and 2D models are validated against more detailed 3D Monte-Carlo model conducted with TRIPOLI-4. This methodology is devoted to helium cooled lithium lead (HCLL) blanket, but it could be applied to any breeder blanket concept provided that appropriate validation could be carried out.