In the pursuit of sustainable energy solutions, thermoelectricity has emerged as a promising field. Thermoelectricity basically means the conversion of heat energy to electrical energy, and vice versa. This conversion occurs when a temperature difference exists across a thermoelectric material, generating electrical energy.
Thermoelectric materials, which are based on the concept of thermoelectricity, have applications in solid-state power generation, refrigeration, and industrial waste heat recovery. These materials are highly sought after because they are noise-free, environmentally friendly, have simple structures and distinctive properties.
With so many uses, thermoelectric materials have attracted the attention of scientists worldwide, and it is important to evaluate their efficiency and properties. The efficiency of thermoelectric materials is given by a dimensionless quantity called the figure of merit (ZT), where:
S is the Seebeck coefficient: it is a measure of the electric voltage generated in a material when a temperature difference is applied across it.
σ is the electrical conductivity: it is a measure of how well a material can carry an electrical current.
T is the absolute temperature: it represents the true physical measure of thermal energy in a system.
kl is the lattice thermal conductivity: it is the heat carried by atomic vibrations in a crystal.
ke is the electronic thermal conductivity: the heat transport due to moving charge carriers.
Thermoelectric materials exhibiting superior performance are generally characterized by a higher power factor (PF), represented as S2σ, coupled with lower thermal conductivity. Nevertheless, it is essential to acknowledge that the relationship between these parameters is intricately complex, and their modulation is not independent. Consequently, enhancing the ZT values of thermoelectric materials poses a significant challenge.
Half-Heusler (HH) alloys are a class of thermoelectric materials that have been gaining attention for their favourable mechanical properties, thermal stability, and use of relatively non-toxic elements. They are made from three different elements XYZ, where X is the most electropositive element, Y is a less electropositive transition metal, and Z is a p-block element (the p-block elements are – N, P, As, Sb, Bi). These materials also have a well-ordered crystal structure similar to the cubic MgAgAs-type.
To determine the properties of thermoelectric materials, such as half-Heusler alloys, scientists often use theoretical calculations, including density functional theory (DFT) and molecular dynamics (MD) simulations, to provide reliable predictions of thermoelectric properties, including how they conduct electricity and heat, and guide experiments to test their performance.
However, these methods become slower and require much more computing power when materials are complex, for example, when they have defects, extra elements, or mixed compositions. The high computational costs are particularly noticeable in high-throughput studies, where many materials are analyzed at the same time.
As the demand for energy-efficient technologies grows, the development of accurate predictive models becomes important. Machine learning (ML) techniques can be a viable alternative to DFT calculations, delivering results that are similarly accurate while significantly reducing computational time and cost.
In this study, the authors Dr. Vipin Kurian Elavunkel and Prof. Prahallad Padhan from the Department of Physics, Nanoscale Physics laboratory, Indian Institute of Technology (IIT) Madras, Chennai, India (Prof. Prahallad Padhan is also affiliated with the Functional Oxides Research Group, IIT Madras, Chennai, India), have successfully explored the electrical and thermal properties of half-Heusler alloys using a sophisticated machine learning model called stacking ensemble model, combining Random Forest and XGBoost algorithms.
This approach consistently outperformed individual models like DFT across all key thermoelectric parameters: kl, σ, S, and ZT, highlighting its robustness and enhanced predictive accuracy. The integration of physically meaningful composition-based features such as temperature, mean covalent radius, etc. consistently emerged as influential factors, revealing the intricate relationships influencing thermoelectric performance.
The validated predictive framework not only bridges experimental observations and theoretical insights, but also offers a reliable tool for screening and designing novel half-Heusler materials with optimized thermoelectric efficiency, thereby advancing the development of practical energy conversion technologies.
Prof. Manoj K. Harbola, Professor of Physics, and Professor In Charge, CERTEX, at the Indian Institute of Technology (IIT) Kanpur, Kanpur, India, acknowledged the importance of this work with the following appreciative comments: “I have gone through the paper by Dr Pradhan. It is a timely and relevant work that uses the latest computational techniques. It first trains a machine learning model on a large dataset and then tests it. The effectiveness of stacking-model predictions is checked against first-principles DFT calculations. This check highlights the limitations of DFT calculations for certain properties. This is very significant since a majority of electronic structure and properties calculations are carried out using DFT.
The paper is an excellent and contextual contribution to the literature on the thermal properties of materials.”
Article by Akshay Anantharaman
Click here for the original link to the paper
