Jochem De Schutter
KU Leuven
Monday, July 06, 2015, 10:15 - 11:15
Room 01-210, Georges-Köhler Allee 102, Freiburg 79110, Germany
The design of the divertor in a nuclear fusion reactor goes beyond geometrical or magnetic optimization for optimal heat spreading or pumping efficiency. The optimal design should be robust with respect to uncertainties in the underlying physical model en should allow for a large operational window.
Therefore in this thesis a method is elaborated to compute the uncertainty interval of the optimal design in an efficient way. This interval is computed with gradient-based optimization techniques that rely on the ‘adjoint’-method for gradient computation. This way the cost of the uncertainty quantification method is quasi independent of the number of uncertain input parameters.
Because the robustness of the optimal divertor is dependent on the uncertainty of the parameters that serve as an input for its design, this thesis also elaborates a method for calibrating or estimating uncertain model parameters. This boils down to a non-linear regression, that computes the ‘best-fit’ model parameter values, that make sure that the simulated plasma state matches as good as possible with experimental data. Also here gradient-based optimization techniques based on the adjoint-method are applied.
Both methods have been elaborated in detail en succesfully executed in a testcase on a slightly simplified plasma edge model, that is representative for the physics contained in complete plasma edge codes.