Presentation Title

Applying Machine Learning To Materials Science: Classification and prediction of perovskites

Faculty Mentor

Sam Behseta, Allison Fry-Petit

Start Date

23-11-2019 8:00 AM

End Date

23-11-2019 8:45 AM

Location

153

Session

poster 1

Type of Presentation

Poster

Subject Area

engineering_computer_science

Abstract

Minerals, otherwise known as perovskites, are compounds that are used in a wide variety of technologies. The rational design of perovskites has the potential to improve the performance of technologies such as MRI machines, however, their discovery to date has often been limited to a slow synthesis and testing process mostly based on trial and error. In this work, we have employed a series of machine learning techniques to better understand the structurally driving forces in perovskites towards the rational design. Our method uses an unsupervised learning technique from the set of 883 perovskite compounds with 72 predictors, to categorize known materials by structure type. The classification of data is a primary step towards the identification of features that can potentially contribute to building a predictive model for perovskite design. We applied Principal Component Analysis (PCA) on the original space of 72 variables. PCA reveals natural clustering of certain perovskites in the direction of the most significant loadings or eigenvectors of the sample covariance matrix. Specifically, results show that predictors that separate structure type are very sensitive to anion type. This result has the potential to elucidate the currently unknown structural driving forces that can favor perovskite formation with differing anions.

This document is currently not available here.

Share

COinS
 
Nov 23rd, 8:00 AM Nov 23rd, 8:45 AM

Applying Machine Learning To Materials Science: Classification and prediction of perovskites

153

Minerals, otherwise known as perovskites, are compounds that are used in a wide variety of technologies. The rational design of perovskites has the potential to improve the performance of technologies such as MRI machines, however, their discovery to date has often been limited to a slow synthesis and testing process mostly based on trial and error. In this work, we have employed a series of machine learning techniques to better understand the structurally driving forces in perovskites towards the rational design. Our method uses an unsupervised learning technique from the set of 883 perovskite compounds with 72 predictors, to categorize known materials by structure type. The classification of data is a primary step towards the identification of features that can potentially contribute to building a predictive model for perovskite design. We applied Principal Component Analysis (PCA) on the original space of 72 variables. PCA reveals natural clustering of certain perovskites in the direction of the most significant loadings or eigenvectors of the sample covariance matrix. Specifically, results show that predictors that separate structure type are very sensitive to anion type. This result has the potential to elucidate the currently unknown structural driving forces that can favor perovskite formation with differing anions.