Mona Zebarjadi - Screening Materials with large Thermoelectric Power factors using Machine Learning

In the past, our group and others used experiments and computer simulations based on rigorous physics (called first-principles calculations) to search for good TE materials. However, these methods are time-consuming. Recently, the discovery process has been accelerated by incorporating artificial intelligence (AI), particularly machine learning (ML). In a recent project, we developed a database of binary metallic alloys and used different ML algorithms for training. Finally, we were able to predict the Seebeck coefficients as a function of temperature and composition. This helped us quickly find several new alloys that could be useful for TE devices.
Building on this experience, we are looking for 2 undergraduate students to lead new material screening projects. The idea is to build up material databases for  2 distinct classes of materials and screen potential materials with large TE power factors within these two classes. Each student will focus on a specific class of TE materials and use ML to optimize and screen candidates with high performance. 
Project 1: Metallic Alloys with high power factor for active cooling applications
The first class of materials is metallic alloys composed of 2-5 parent metals. The goal is to find metallic alloys, from binary to high-entropy ones, with large TE power factors that are low-cost and environmentally friendly (non-toxic elements). 
Project 2: Doped I-V-VI2 ternary TE materials with high figure of merit for power generation.
The second class of materials are  I–V–VI₂ ternary compounds such as AgSbSe2. There are several narrow gap semiconducting compounds with large TE power factor and low thermal conductivity within this class of materials, which are ideal for thermal to electrical energy conversion. Strategies such as doping and alloying can further improve the TE efficiency of this class of materials. In this project, a material database will be built using experimental data available in the literature. ML training and prediction will then be used to identify stable materials with optimum doping concentration to maximize the TE properties for energy conversion. 
In both cases, our final goal is to select a few materials, fabricate them, and verify the predictions experimentally. The undergraduate students can also participate in the experimental validation process.