Prediction of mechanical properties for a grafted polypropylene system via transfer learning with a small database

Abstract

The increasing demand for polymer dielectrics in different scenarios, such as electrical insulation cable, requires elevated electrical and mechanical properties. In order to understand the relationship between mechanical properties and the structure of grafted monomers in a grafted polypropylene (PP) system, a design framework based on the transfer-learning method has been proposed for designing grafted side chain structure in PP system. 22 samples of polypropylene-grafted polymer are synthesized. Through a well-trained multi-layer perceptron model for predicting glass transition temperature (Tg), mechanical properties of the grafted PP have been well predicted by the transfer-learning method. Meanwhile, the grafting group structures are divided into nine substructures. By assembling these substructures, the chemical space is established for prediction. Moreover, candidates with satisfied deep traps and flexible mechanical properties are screened out for reference. This work turns qualitative understanding into a quantitative analysis based on mathematical models and broadens the application of polymer materials design with a small dataset.