Strategies for Improved Performance of Probabilistic Simulations

In recent years, probabilistic methods have becomeincreasingly important in engineering applications. They allowa quantification of the impact of the variability of componentson result values. In this thesis, existing probabilistic methodsare analyzed and new ones are introduced to improve theirperformance, especially in the context of the probabilisticanalyses of jet engine components.A major focus of the thesis is on the analysis of samplingmethods, especially with regard to the resulting surrogatemodel quality. For this purpose, Latinized Particle Sampling isintroduced as a new method in which the realizations of thesample are considered as charged particles. This new methodis then compared with existing sampling methods.Another focus is on sensitivity analysis with correlated inputvariables. Established methods such as the Sobol indices orShapley values cannot reliably identify input variables withoutfunctional influence in such cases. Therefore, the modifiedcoefficient of importance is introduced as a new sensitivitymeasure. Finally, the discussed methods are applied to theanalysis of compressor blades subject to manufacturingvariability and their advantage is demonstrated.