Optimization Methods for Segmented Thermoelectric Generators

Abstract

Future National Aeronautics and Space Administration deep-space missions are seeking radioisotope propulsion systems (RPS) to have specific powers above 8 [W e /kg], while having thermal conversion efficiencies greater than 12%. The design and optimization of segmented thermoelectric unicouples used within RPS requires a multi-faceted approach to maximize device performance. The design space of a unicouple can span multiple dimensions, requiring immense computational resources to conduct parametric studies. These dimensions include, but are not limited to, the independent cross-sectional areas of the nand p-type legs, the total height of the unicouple, the length of the high-temperature nand p-type segments, the cold-side junction temperature and the load resistance applied to the couple, considering a fixed hot-side junction temperature, fixed per-couple heat input, and desired output voltage. To this end, computationallyinexpensive methods that optimize segmented unicouples are presented and compared. These methods include physics-based algorithms that dynamically reduce the design space when nonviable configurations are found, implementation of Golden Section Search (GSS) algorithm when uni-modal behavior is observed for a specific degree of freedom, and successive design space refinement. When using both GSS and successive design space refinement algorithms, an optimum geometry was found with 5,755 times fewer solver calls in comparison to the conventional parametric study without any loss of fidelity. This comparison indicates the proposed optimization methods are robust and accurate, while also drastically reducing the computation time to find the optimum unicouple configuration that maximizes system-level power output. These methods allow for exhaustive trade studies to be conducted of newly proposed heat sources, converter materials and designs, and heat exchange systems.