2026 Volume 47 Issue 2 Pages 78-87
Materials informatics (MI) is rapidly being adopted in industrial materials development. However, network polymers are difficult to uniquely describe their three-dimensional crosslinked structures, and the success or failure of MI application depends on the representation design of the material system. This article roughly classifies MI tasks for network polymers into (i) tasks where multiple components and conditions can be defined and (ii) tasks where a representative structure can be defined, and guidelines for selecting approaches suitable for practical applications are summarized. In the former, the importance of designing tabular datasets, focusing on formulation ratios, material types, and process conditions, is explained using the case of thermosetting resin composites. On the other hand, in the latter, three approaches, (1) descriptor-based modeling, (2) structure-based modeling using graph neural networks (GNNs), and (3) sequence-based modeling using Transformer are compared to describe their characteristics and applications. Furthermore, based on the MI approaches presented in this paper, points to consider in practical applications and future prospects of this field are described.
