Artificial Intelligence Guided Thermoelectric Materials Design and Discovery

Materials discovery from vast repositories of Earth's resources remains a significant impediment to revolutionary technological advancements. The labor-intensive and time-consuming nature of this process hampers the exploration of novel materials. While machine learning techniques have demonstrated their potential in expediting materials discovery, the challenge lies in obtaining effective material feature representations and achieving precise predictions of material properties. This research endeavors to establish an automated framework for material design and discovery, harnessing the power of data-driven AI models. Initially, we have developed a range of diverse material descriptors to enhance the representation and encoding of the distinctive characteristics of materials. This, in turn, leads to improved predictive performance across various molecular properties. As a baseline, we have focused on predicting thermoelectric (TE) properties of materials to demonstrate the framework's effectiveness. Remarkably, our proposed framework attains an accuracy rate exceeding 90% in forecasting TE properties. Moreover, our AI models have identified 6 promising p-type TE materials and 8 promising n-type TE materials. To validate our predictions, we have compared them with Density-functional Theory (DFT) calculations, and they align with the TE properties obtained from experimental results. By harnessing the potential of machine learning, deep learning, and data mining, this endeavor holds the potential to transform the paradigm of high-performance energy harvesting materials design, ushering in a new era of sustainable and efficient energy utilization.