Plasma Diagnostic Signal Analysis: A Bayesian Based Genetic Algorithm Approach

Plasma physics is rich in phenomena, occurrences and applications: many instabilities exist due to the relatively long-range Coloumb forces that mediate constituent particle dynamics, most of the visible universe is in the plasma state, and plasmas have been used from lighting to computer chip manufacturing to (attempted) fusion power generation. Within plasma fusion research, Thomson scattering is a commonly used diagnostic. It allows the temperature and density of the plasma electrons to be measured without distorting the plasma. However, the scattering cross-section is small. Thomson scattering signals can be difficult to detect against the background emission of the plasma. In this thesis, the Thomson scattering diagnostic data from the COMPASS-D experiment is analysed. Several aspects of the diagnostic are presented along with detailed explanation of the inference procedure for determining the plasma's electron temperature. This temperature analysis was achieved by utilising a Bayesian inference model that allowed prior information about likely values to be systematically included. This prior information was found to remove the degeneracy present due to the low signal-to-noise ratio of the data. A genetic algorithm (GA) library, called ELGAR, was developed and used to solve the minimisation problem resulting from the Bayesian inference. The GA proved to be a reliable method of solving such problems. ELGAR was also used to investigate certain characteristics of the GA such as optimal choice of key parameters. These were found to be in disagreement with theoretical results but the difference was explained by the different mode of operation of ELGAR. The Thomson scattering analysis was extended to include two-temperature considerations. The set of observations consistent with an n-temperature distribution function was found to be bounded by a curve. Some data from the COMPASS-D experiment lay outside this boundary, but was bounded by a similar curve. This suggested that some systematic error had occurred. Some explanations of possible causes of this bias were suggested. Constraints were found for interpretations of any observations. These indicated that, for some observations, a set of temperatures is unavailable (as either to hotter or colder component) for a distribution function which is consistent with that observation. For certain observations, the least-squares temperature estimate is contained within the set of impossible temperatures. This indicates that the presence of a hotter species of electrons can bias the observations towards higher temperatures. The thesis concludes with a summary and a discussion of possible future work