Bayesian experimental design. Applications in nuclear fusion

This work introduces the idea of Bayesian experimental design to the field of high temperature plasma physics and gives examples for the optimisation of diagnostics as well as for the planning of future experiments. The approach is validated using data sets from the Wendelstein 7-AS experiment: The data was analysed and compared with the respective Expected Utility, which itself was calculated independently. A monotonous relationship between Expected Utility and the deviation χ2 of the measurement result was found. As a first application, a multi-channel interferometer at the Wendelstein 7-X stellarator is optimised using Bayesian experimental design. Three different density effects (measurement of the effects of Core Electron Root Confinement, estimation of high confinement regimes and the analysis of pellet injections) were used as design criteria. The results were analysed and compared, a final proposal for the sight line configuration is given and discussed. Bayesian experimental design can used as a tool for experimental planning, too. Here, the optimal experimental setting can be found to gain maximum information about a certain problem. For this, also data from previous experiments are incorporated, their respective usefulness can also be expressed by the Expected Utility. As an example, a data set from the stellarator database was examined with respect to the approach of power scaling laws. It was found out, that a good signal-to-noise ratio is preferred by Bayesian experimental design. Furthermore, a significant impact of the measurement error and the error statistics could be verified. The final design is determined by a combination of these criteria. For the interferometer, the optimal design is given for a beam line confiuration where a change of the parameters of interest has maximum impact on the expected data. In the case of experimental planning, the most informative experimental configurations for future measurements is found for settings which are not yet provided by the given database. (orig.)