|About this Abstract
||Materials Science & Technology 2019
||Ceramics and Glasses Simulations and Machine Learning
||Machine Learning Applied to Zeolite Synthesis Enabled by Automatic Literature Data Extraction
||Zach Jensen, Elsa Olivetti
|On-Site Speaker (Planned)
Zeolites are porous, aluminosilicate ceramics with many industrial and “green” applications. Despite their industrial relevance, the relationship between the synthesis and resulting zeolite structure remain poorly understood requiring expensive and time-consuming trial and error. To realize the full potential of zeolites, alternatives to trial and error synthesis need to be developed. We present an alternative, modeling the synthesis pathway with machine learning. We create natural language processing techniques to automatically extract synthesis information and trends from journal articles. We test this method by examining a data set of germanium zeolites. We create a regression model for a zeolite’s framework density as a function of synthesis parameters. This model has a cross-validation error of 0.98 T/1000 Å3, and the decision boundaries match many of the known synthesis heuristics. We propose that this methodology can be applied to many problems in zeolite synthesis and will enable synthesis of novel zeolite morphologies.