Prediction\nand Explanation of Properties in Multicomponent\nPolyurethane Elastomers: Integrating Molecular Dynamics and Machine\nLearning

Establishing\nquantitative connections among the chemical composition,\nmolecular structure, and macroscopic properties of multicomponent\npolyurethane elastomers remains a challenging task. Molecular dynamics\n(MD) has been extensively utilized in the study of various materials\nand serves as a crucial tool for exploring the relationship between\nstructure and properties. However, the intricate modeling process\nand lengthy computation times associated with the MD method complicate\nthe attainment of complex combinatorial results for the various components\nof polyurethane elastomers. Machine learning (ML) offers a solution\nby integrating and analyzing existing data, along with the capability\nto predict new outcomes. Consequently, we combine MD and ML methods\nto conduct a comprehensive investigation of multicomponent polyurethane\nelastomers. MD simulations indicate the presence of various types\nof hydrogen bonds within the elastic matrix of polyurethane, and the\nstrong hydrogen bonds formed in the hard segments significantly affect\nthe tensile properties of material. While the incorporation of long\nmolecular chains in the soft segments enhances the material’s\nflexibility, it simultaneously diminishes its tensile strength. Feature\nengineering techniques, including parametric representation and feature\nscreening of the MD model, were employed to create a data set suitable\nfor ML applications. The application of the interpretable ML method\nhas demonstrated that the number of hydrogen bonds in the hard segment\nis regulated by the hydrogen bond donor and acceptor, while the rotatable\nbonds in the soft segment are the primary characteristics contributing\nto the material’s flexibility and are also key factors that\nregulate the number of free hydrogen bonds. This integration of MD\nand ML methods not only enhances predictive capabilities for novel\npolyurethane elastomers but also facilitates quantitative analysis\nof how microstructural characteristics affect macroscopic properties.