Development of Disulfide Epoxy Vitrimers via Transfer Learning: Bridging the Gap between Excellent Comprehensive Properties and Dynamic Cross-Linking Networks

Vitrimers are some cross-linked polymers with sustainability introduced by dynamic covalent bonds. However, vitrimers may exhibit more compromised thermal, mechanical, and electrical properties than traditional thermosets. Due to the structural diversity and complexity, developing vitrimers with both excellent comprehensive properties and eco-friendly characteristics remains challenging without quantitative structure–property relationships (QSPRs). Herein, we construct 2 data sets for disulfide vitrimers: one data set containing +100,000 entries of organic molecular properties and another data set with +1700 experimental entries for macroscopic properties of epoxy resins. A QSPR model (AdaQSPR) based on a 3D molecular representation learning framework and transfer learning is developed, which can efficiently and accurately predict the glass-transition temperature (T_g), initial thermal decomposition temperature (T_d5%), tensile strength (TS), relative permittivity (ε), dielectric loss (tanδ), and electrical breakdown strength (E_b) of epoxy resins. This model screens 216 vitrimers based on common disulfides and selects 5 promising candidates. Experimental tests verify their good comprehensive properties: e.g., for all selected vitrimers, T_g > 146 °C, E_b > 34.8 kV/mm; four selected vitrimers have TS ≥ 60 MPa. These vitrimers also possess repairability and degradability. This QSPR model can also be utilized for other novel epoxy resins to accelerate research and reduce R&D costs.