Explore Thermal and Mechanical Properties of Biobased Polyurethane Elastomers Through Machine Learning Models.

Abstract

Mechanical and thermal properties are the core to determine the application scenarios of biobased polyurethane elastomers (BPUEs). Here, six core properties were studied, including Young's modulus (YM), tensile strength (TS), elongation at break (EB), glass transition temperature (Tg), decomposition temperature at 5% weight loss (Td₅), and the energy dissipation factor (tanδ). We compiled a new dataset comprising more than 1500 samples with comprehensive information in composition, process, structure, and properties. Through domain-knowledge augmented feature engineering, a set of 26 features is sufficient to predict these core properties. Multi-target regression models for YM, TS, and EB delivered coefficients of determination (R²) better than 0.70 from validation and blind tests, and higher than 0.80 in the prediction of the remaining three properties, Tg, Td_5, and tanδ. Features to describe the chemical structure of polyurethane monomers and their formulation are dominant, and they contributed more than 70% of the explainability. Biomass feedstocks, molecular weights for polyols, hard segment contents etc., are important regulatable variables to prepare BPUEs with fitting-for-purpose products, and the stretching rate and the heating rate are also critical to harvest repeatable mechanical and thermal properties. This study provided data-driven insights for the rational design of BPUEs with desired mechanical and thermal properties.