Silica Aerogel Synthesis/Process–Property Predictions by Machine Learning

Abstract

Silica aerogels are mesoporous high surface area materials with extensive synthetic and processing conditions. To effectively synthesize aerogels, the impact of synthetic pathways on the resulting aerogel properties must be understood prior to experimental investigation. We develop an information architecture, the silica aerogel graph database (10 3 ), and a supervised machine learning neural network regression model to examine these relationships. The property graph database enables rapid queries and visualization of the impact of the synthesis and processing conditions on the final aerogel properties. The model maps from silica aerogel synthetic and processing conditions to predict the aerogel BET surface area with an average error of 109 ± 84 m 2 /g. Following a validation experiment, the model was shown to predict the aerogel surface area from new synthetic and processing conditions with an error of less than 5%. The experiment demonstrates the usefulness of the model in surface area prediction through the compatibility between computational and experimental results. Both in its current form and with further expansion, the developed graph database could reduce experimental dimensionality, time, and resources, enabling the successful synthesis of high surface area silica aerogels, which are advantageous for applications including thermal insulation, sorption media, and catalysis.