Optimization of a segmented thermoelectric generator with various doping amounts using central composite design, multi-objective genetic algorithm, and artificial neural network

This study optimizes a segmented thermoelectric generator (STEG) under a 400 K temperature difference. Hot-side materials consider different doping amounts of indium (In). STEGs with different leg lengths and cross-section areas are explored for the first part of the study. It shows that the output power of the STEG with a leg length of 3 mm and a cross-section area of 4 mm × 4 mm is 84.28 % higher than that with a leg length of 6 mm and a cross-sectional area of 2 mm × 2 mm, but the conversion efficiency becomes 28.77 % lower. There have been no studies on segmented thermoelectric generators (STEGs) doped with different amounts of p-type and n-type thermoelectric materials, especially to analyze their performance through numerical predictions. The second part uses a multi-objective genetic algorithm (MOGA) for optimization analysis. The results show that the STEG using undoped p-type and 3 % n-type doping produces the best output power (1.337 W) and the highest conversion efficiency (18.71 %). Compared with the non-optimized STEG, the output power and efficiency of the optimized STEG are increased by 14.53 % and 32.49 %, respectively. Central composite design (CCD) is used for artificial neural network (ANN) model architecture, and ANN is used for STEG prediction and optimization. The calculation time of ANN is 1820 times less than MOGA, and the error is about 3–8% smaller, although the optimized value is 5–8% smaller.