Numerical simulation, clustering, and prediction of multicomponent polymer precipitation

• DOI: 10.1017/dce.2020.14
• Publication Date: 2020-11-17T07:00:33Z
• Authors: Pavan Inguva; Lachlan R. Mason; Indranil Pan; Miselle Hengardi; Omar K. Matar
• Keywords: Autoencoder

Abstract Multicomponent polymer systems are of interest in organic photovoltaic and drug delivery applications, among others where diverse morphologies influence performance. An improved understanding morphology classification, driven by composition-informed prediction tools, will aid engineering practice. We use a modified Cahn–Hilliard model to simulate precipitation. Such physics-based models require high-performance computations that prevent rapid prototyping iteration settings. To reduce the required computational costs, we apply machine learning (ML) techniques for clustering consequent simulated polymer-blend images conjunction with simulations. Integrating ML simulations such manner reduces number needed map out blends as function input parameters also generates data set which can be used this end. explore dimensionality reduction, via principal component analysis autoencoder techniques, analyze resulting clusters. Supervised using Gaussian process classification was subsequently predict clusters according species molar fraction interaction parameter inputs. Manual pattern yielded best results, but were able ≥90% accuracy.