(142i) Identifying Descriptors for Dielectric Breakdown Strength Using Genetic Programming
AIChE Annual Meeting
2016
2016 AIChE Annual Meeting
Computational Molecular Science and Engineering Forum
Data Mining and Machine Learning in Molecular Sciences I
Monday, November 14, 2016 - 2:24pm to 2:36pm
The development of phenomenological theories by identifying relevant descriptors for material properties enables rapid screening large numbers of materials and facilitates the design of new materials. One of the leading challenges for computational researchers is determining the best ways to analyze large material data sets to identify the best descriptors for a given property. In this presentation, we demonstrate the use of genetic programming to identify relevant descriptors of dielectric breakdown based on 82 representative materials. We identified band gap Eg and phonon cut-off frequency wmax as the two most correlated features, and new classes of phenomenological models featuring exponential functions of Eg*wmax were uncovered. The genetic programming model was found to outperform other models for descriptor identification, and we discuss some of the advantages of the genetic programming approach.