Co-optimization of area ratio and material fraction in segmented N-type thermoelectric legs using XGBoost–ANN surrogates

Spacecraft operate in extreme environments and generate abundant waste heat from power, thermal-control and propulsion subsystems. Thermoelectric generators directly convert this waste heat into electricity, delivering solid-state, reliable, maintenance-free onboard power. However, the energy conversion efficiency and long-term stability of conventional thermoelectric technologies remain constrained under large temperature gradients. Here, a segmented thermoelectric generator is proposed with an optimized n-type leg architecture. By jointly tailoring the cross-sectional geometry and material distribution, the design strengthens carrier transport and boosts both conversion efficiency and output power. Comprehensive multiphysics simulations conducted on the COMSOL platform systematically evaluate the effects of cross-sectional area ratios and material fractions on device performance. The optimized segmented architecture achieves a 9.91% increase in output power and a 8.69% enhancement in thermoelectric conversion efficiency compared with conventional designs. Furthermore, eXtreme Gradient Boosting (XGBoost) and artificial neural network (ANN) models are developed to predict thermoelectric performance. XGBoost effectively captures the nonlinear relationships between input parameters and output metrics, yielding prediction errors as low as 0.42% and 0.51%, while ANN demonstrates strong performance in multi-output modeling, with coefficients of determination (R 2 ) of 0.9577 and 0.9603, respectively. These results show that combining segmented-structure optimization with surrogate modeling can markedly enhance thermoelectric energy conversion. This integrated strategy offers a practical route to high-efficiency onboard energy management in advanced spacecraft systems.