Uncertainty Quantification Using Reduced-Order Models

Reduced-order models (ROMs) provide computationally efficient surrogates of high-fidelity thermal models (e.g. Thermal Desktop® models). An approach for creating these surrogates using efficient sampling and Gaussian process data fitting was developed and successfully applied to a broad range of spacecraft applications. This approach provides numerous benefits including computational speed; however, uncertainty must be carefully evaluated.

This investigation attempts to understand, quantify, and mitigate uncertainty using ROMs. An overview of sources of uncertainty will be presented. Further, computational efficiency and uncertainty will be evaluated as a function of surrogate construction. Finally, an overview of how Uncertainty Quantification can be realized using ROMs will be presented. Examples of how these methods can be used in practice will be provided.