Predicting the thermoelectric figure of merit in p-type BiSbTe-based alloys using artificial neural network modeling

Abstract

The polycrystalline Bi0.5Sb1.5Te3 compound has drawn promising thermoelectric material over the decades with a figure of merit (zT) close to unity and numerous experiments have been conducted through a trial-and-error approach to further improve their zT by tuning the chemical composition and processing approaches. Despite a lot of studies available on BiSbTe alloys, choosing the right combination of alloy system with high zT is still a challenging task. Herein, we produced a p-type Bi0.5Sb1.5Te3 alloy through casting followed by high-energy milling to enhance the zT via reducing thermal conductivity. The experimental results revealed that a peak zT of 0.95 was achieved at 300 K due to a substantial reduction in thermal conductivity via enhanced carrier/phonon scattering at refined grain boundaries. In addition to the current scenario, the machine learning (ML) assisted artificial neural network (ANN) modeling was performed to predict the zT value by applying the composition, electrical conductivity, Seebeck coefficient, thermal conductivity, and temperature as the input parameters. The predicted zT by ANN modeling shows excellent accuracy with experimental findings about an average absolute error of 9.34 %. Therefore, the developed ANN modeling shows a protocol for composition-based prediction of zT with reduced experimental costs, time consumption, and excellent accuracy in terms of the theoretical and experimental databases.