Uncertainty Quantification for Motor Digital Twin Models

Uncertainty quantification (UQ) is becoming more and more necessary for the adequate simulation of real-world applications based on non-deterministic processes. Therefore, UQ finds a central application in the digital twin (DT) field, which is an emerging technology for comprehensively modeling physical phenomena. In this work, the analysis of a thermal model as part of a DT - cloning an asynchronous motor - is conducted by using variance-based sensitivity analysis to improve the DT’s prediction performance and therefore the efficiency of the asynchronous drive. The sensitivity information is extracted from a polynomial chaos expansion (PCE) surrogate of the thermal model. Therefore, the thesis also focuses on the adaptive construction of PCE bases that exploit possible anisotropies in the model input space to provide the best possible surrogate models for the sensitivity analysis. In this context, this work also contributes with two extensions of an already applied algorithm for adaptive basis construction for PCE surrogates. In a comparison of different basis construction approaches, it was shown that the adaptive approach outperforms basis construction approaches that do not consider anisotropies in the input space.