As global environmental sustainability becomes increasingly emphasized, the development of eco-friendly materials, including solutions to the issue of marine plastics, is thriving. However, the material parameter space is vast, making efficient search a challenge. Time-domain nuclear magnetic resonance offers material property information through the complex T2 relaxation curves resulting from multiple mobilities. In this research, we used the Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence to evaluate the binding state of water (water affinity) in polymers synthesized with various monomer compositions, which were immersed in seawater. We also assessed the T2 relaxation property of the polymers using the magic sandwich echo, double quantum filter, and magic-and-polarization echo filter techniques. We separated the T2 relaxation curves of CPMG into free and bound water for polymers by employing semisupervized nonnegative matrix factorization. By employing the features of separated bound water and polymer properties, a polymer composition optimization method offered crucial factors to monomers through random forests, predicted the components of the polymer using generative topography mapping regression, and determined expected values using Bayesian optimization for polymer composition candidates with the desired high water affinity and high rigidity.
Keywords: Machine Learning; Polymer; T(2) relaxation; Time-domain NMR; Water affinity.
Copyright © 2023 The Author(s). Published by Elsevier Inc. All rights reserved.