In this research we want to improve and adapt different machine learning methods, which have so far mainly been applied in numerical simulations, now for both experimental and numerical investigation of complex aerodynamic problems and their optimization. Specifically, the required measurement conditions and procedures are investigated. Furthermore, the suitability of different data fusion methods for coupling of real, noisy and numerically obtained data is analyzed. In addition, methods and data structures for the use of metadata in databases are investigated and developed. Thereby, common methods are analyzed but also new methods are explored by modifications and couplings of existing concepts.
Across all parts of the project, machine learning methods are used in experiments and CFD simulations to optimize processes and bring the results from different sources to a higher usable level via data fusion. Experiments and simulations related to aerospace and automotive specific problems are conducted. Especially the data quality as well as the influence of systematic and random errors play a crucial role. This influence is quantified and subsequently for each used machine learning method a possibility to increase the robustness against such errors is explored. Furthermore, for data acquisition different sampling methods and strategies are investigated. In the area of optimization, different objective functions are chosen and the suitability of the machine learning methods to find optima are evaluated.