Machine Learning Regression Guided Thermoelectric Materials Discovery – A Review

Abstract

Thermoelectric materials have increasingly been given attention recently due to their potential of being a solid-state solution in converting heat energy to electricity. Good performing thermoelectric materials are expected to have high electrical conductivity and low thermal conductivity which are usually positively correlated. This poses a challenge in finding suitable candidates. Designing thermoelectric materials often requires evaluating material properties in an iterative manner, which is experimentally and computationally expensive. Machine learning has been regarded as a promising tool to facilitate material design thanks to its fast inference time. In this paper, we summarize recent progress and present the entire workflow in machine learning applications to thermoelectric material discovery, with an emphasis on machine learning regression models and their evaluation.

Multiobjective Optimization and Machine Learning Algorithms for Forecasting the 3E Performance of a Concentrated Photovoltaic-Thermoelectric System

“…This means that it is crucial to determine the suitable property combinations needed to derive the optimum material. There seems to be a bottleneck in the discovery of more efficient thermoelectric materials [25,26] due to the reliance of the thermoelectric figure of merit, ZT, on the Seebeck coefficient (S), electrical (σ), and thermal (k) conductivities according to the expression, ZT = TS 2 σ/k. From this expression, there must be a reduction in the lattice thermal conductivity and an increase in the electrical conductivity for there to be an increase in the figure of merit.…”

Predicting thermoelectric transport properties from composition with attention-based deep learning

“…While DFT-based HTS is becoming more prevalent, there remains a large gap between the size of chemical space that is accessible with this approach, and the size of the space of all possible inorganic materials. To bridge that gap, and to further accelerate computational predictions of thermoelectric behaviour, techniques involving the use of machine learning (ML) have been gaining popularity in the search for new thermoelectric materials [47][48][49][50][51]. Data for these ML approaches can come from either theoretical calculations, or from physical experiments.…”

High Seebeck Coefficient Inorganic Ge15Ga10Te75 Core/Polymer Cladding Fibers for Respiration and Body Temperature Monitoring

“…The detection sensitivity of the thermal sensor can be assessed by the speed and magnitude of voltage generation from a minimal temperature change . Thus, TE material with a large Seebeck coefficient is preferred for thermal sensing that requires high sensitivity and accurate temperature-sensing ability. , Recent decades have witnessed great efforts in developing novel TE materials including organic and inorganic materials to realize highly sensitive thermal sensing. − The organic TE materials exhibit unique flexibility and ease of processing. This low Seebeck coefficient limits its potential application in high-performance thermal sensors. − Due to the relatively high carrier concentration, inorganic crystalline TE materials usually show a low Seebeck coefficient and high electrical conductivity, which is not suitable for thermal sensing. − In contrast, inorganic amorphous TE materials possess low thermal conductivity and high Seebeck coefficient in spite of a low ZT, which are promising candidates for thermal sensors with high sensitivity and temperature resolution. , The recently reported semiconducting chalcogenide glasses of Cu–As–Se–Te, As–Se–Sb–Cu, and Ge–Se–Sb–Ag systems present a high Seebeck coefficient of above 1000 μV/K, − while the low anticrystallization ability of these materials leads to uncontrolled crystallization during fiber drawing, which results in the decrease of the Seebeck coefficient .…”