Segmental material design in thermoelectric devices to boost heat-to-electricity performance

Abstract

In this study, a seven-layer segmented thermoelectric generator (STEG) is presented with four pairs of TE materials imposed on each semiconductor leg. Firstly, the performance of the STEG with every possible combination (16384 material sequences) of the leg materials for either leg is analyzed when the other leg is equipped with selected material. Machine learning is further employed to reveal the relationship of the electric power and the energy conversion efficiency with the leg material sequences via the ensemble-based regression model whose hyper-parameters are adjusted by the Bayesian optimization. The criteria of maximum output power and maximum energy conversion efficiency are employed to identify the optimal material sequences for each semiconductor leg, thus the overall optimal material sequences. Comparing to the traditional TEG, when operating between 600 K and 293 K, the electric power and energy conversion efficiency of STEG is 184.1 mW and 16.7%, which presents enhancement of 86.14% and 31.19%, respectively. Counterintuitively, the optimal sequence does not arrange in strict conformity with ZT value which has never been reported in the previous studies. The underling mechanisms are revealed via analyzing the temperature distribution, electric resistance, ZT value and voltage of each layers. The results in this study contributes to the rational design and fabrication of satisfied STEGs.