CAREER: SusChEM: Data Mining to Reduce the Risk in Discovering New Sustainable Thermoelectric Materials

University Of Utah, Salt Lake City UT

Investigators

Abstract

NON-TECHNICAL SUMMARY: Humanity faces a number of grand challenges in engineering in the 21st century ranging from making solar energy affordable, to inventing new tools for scientific discovery, to preventing nuclear terror and more. A common requirement to many of these challenges is the need to discover new materials but traditional materials discovery is slow, inefficient, and expensive. Clearly, a new tool is required to develop new materials faster and at a fraction of the cost. One possibility is to rely on big data to accelerate materials discovery. This project serves the national interest by using data mining tools to create a materials recommendation engine for new sustainable thermoelectric materials. This engine will provide recommendations for new materials based off of statistical probability of desired performance. Scientists will be able to use this tool to guide experimental efforts to explore totally new compounds that would be too risky to investigate otherwise. Since thermoelectrics are devices that can convert waste heat to electricity the potential for this project to benefit the United States is significant. Currently close to two thirds of energy is lost as waste heat and recovering even a small fraction of this with new thermoelectric materials would amount to enormous energy savings. The PI will also leverage this research opportunity to supplement his teaching and outreach efforts. Students will construct novel thermoelectric devices and use these devices to perform bilingual Spanish/English outreach to minority-majority high school and junior high students in Salt Lake City. TECHNICAL SUMMARY: Discovering new materials is slow, inefficient, and expensive. These factors make searching for novel new materials from chemical white space very high risk. Instead, most new developments occur incrementally in already known or established structure types, chemistries, and systems. However, the risk associated with exploring chemical white space for new compounds can be mitigated by utilizing the emergent field of materials informatics. In this proposal novel, sustainable thermoelectric compositions will be suggested using a materials recommendation engine for thermoelectrics. The engine uses composition only to make probabilistic estimates of performance rather than computationally expensive calculations which generally require knowledge of the crystal structure a priori. Avoiding crystal structure as an initial input means entirely new compounds can be discovered with this tool. The engine output is a probability of compositions lying within a desired performance range. Therefore, this project will combine these predictions with existing predictions of where compounds should form to experimentally explore novel thermoelectric materials. The training data set for algorithm and descriptor development will be improved by inclusion of performance of poor, mediocre, as well as good materials to overcome the bias in literature for high-performing materials. Experimental synthesis and characterization will be carried out on suggested compounds and for algorithm validation.