Wow! It’s hot! Even when it rains, the heat in Chennai doesn’t let up. But guess what? Heat might not always be a bad thing. In fact, in the field of thermoelectrics, scientists study the conversion of heat into electricity, and vice versa, using special materials known as thermoelectric materials.
Thermoelectric technology is a method of converting heat into electricity or utilising electricity to generate heating and cooling without the need for moving parts. It utilises special materials that generate electric power when a temperature difference exists between their two sides. This can help recover wasted heat in cars, factories, and power plants, or provide reliable power sources for space missions. Thermoelectric devices are also utilised in wearable electronics to efficiently manage temperature. Because they are solid-state, they are durable, silent, and require little maintenance, making them useful for many applications where traditional heating, cooling, or power systems are less practical.
The performance of thermoelectric materials depends on several key factors. The most important is the figure of merit (ZT), a number that shows how efficient the material is at converting heat into electricity. Other important factors include:
- Thermal conductivity (κ): it is a measure of how well a material can conduct/transfer heat.
- Electrical conductivity (σ): it is a measure of how well a material can carry an electrical current.
- Seebeck coefficient (S): it is a measure of how much electric voltage is generated in a material when there is a temperature difference across it.
- Absolute temperature (T): it represents the true physical measure of thermal energy in a system.
Researchers in the field of thermoelectrics are always seeking ways to increase the figure of merit. They also keep developing new thermoelectric materials such as skutterudites, Half-Heusler alloys, clathrates, layered chalcogenides, etc.
Researchers aim to develop thermoelectric materials that are flexible or have nanostructured features, making them suitable for use in wearable devices and small electronics.
In this study, the authors Mr. Vipin K E 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, Indian Institute of Technology (IIT) Madras, Chennai, India), have focused on skutterudites.
Skutterudites are a class of crystalline compounds, known for their cage-like structure and excellent thermoelectric properties. They consist of a transition metal like cobalt (Co), Nickel (Ni), or iron (Fe), along with a pnictogen element like phosphorous (P), arsenic (As), or antimony (Sb).
Skutterudites have been extensively studied using density functional theory (DFT) to investigate their electronic structure, yielding valuable insights into their conduction mechanisms and thermoelectric properties. By providing an atomic-level understanding of these materials, DFT plays a crucial role in guiding the design of skutterudites with enhanced thermoelectric performance.
However, artificial intelligence (AI)-powered machine learning (ML) has revolutionized the exploration and optimization of thermoelectric materials. ML offers several advantages, including speed, scalability, generalization, and the ability to handle large, complex datasets. ML is particularly valuable for:
- High-throughput material screening,
- Large-scale predictions, and
- Rapid predictions for complex systems where traditional DFT is computationally prohibitive.
XGBoost, a powerful machine learning technique, was utilised in this study to predict the thermal conductivity, electrical conductivity, Seebeck coefficient, and figure of merit for skutterudite materials.
By analyzing the results using Shapley values, a technique that explains which factors most affect predictions, the study identified temperature, atomic composition, and structural features as the most important influences on performance.
The model predicted that a compound called Nd(CoP3)4 [Nd – neodymium, Co – cobalt, P – phosphorous], a skutterudite material could achieve a high ZT value of about 1.06 at 800 K, making it a potentially efficient thermoelectric material.
To test the efficiency of the machine learning method, the thermoelectric properties of Nd(CoP3)4, were studied using DFT. It was found that the XGBoost ML technique was in good agreement with the DFT calculations, particularly at higher temperatures. Unlike traditional computation methods that require extensive computational resources and time, the XGBoost-based approach enables rapid screening and property prediction of new skutterudite compositions.
Using the XGBoost machine learning model, the study predicted that several skutterudite compounds could exhibit high thermoelectric efficiency, highlighting these materials—particularly Nd(CoP3)4—as promising candidates for further experimental investigation and optimization to advance next-generation energy conversion technologies.
By harnessing the unified power of machine learning and materials science, researchers can streamline the development of novel thermoelectric materials, enabling innovative solutions for energy conversion and environmental sustainability.
Prof. Sashi Satpathy, from the Condensed Matter Theory Group, at the University of Missouri, Missouri, United States, gave his analysis and appreciation of the work done by the authors with the following comments: “There has been a long-standing search for advanced materials with superior thermoelectric performance. Achieving this goal is challenging because the key requirements, viz., high electrical conductivity and low thermal conductivity, are inherently conflicting. Conventional theoretical approaches such as density functional theory (DFT) are limited to idealized systems, while experimental exploration by trial and error is both difficult and time-consuming. This study establishes a powerful machine-learning (ML) framework for predicting thermoelectric behavior in skutterudite materials, offering a data-driven alternative to traditional discovery methods. The ML model accurately predicts key transport properties (κ, σ, S, and ZT) and, through Shapley value analysis, reveals why certain compositional and structural features enhance the performance. The identification of Nd(CoP3)4 as a new skutterudite compound with a predicted high figure of merit (ZT) demonstrates the method’s potential for uncovering promising thermoelectric candidates. This is a significant work because it not only establishes a powerful ML framework for predicting thermoelectric behavior, but also because the idea can be extended to other areas of materials optimization.”
Article by Akshay Anantharaman
Click here for the original link to the paper
