Abstract
Mixed raw materials are often used when thermosetting resins are synthesized. In such case, it is difficult to predict physical properties of resins by machine learning, owing to the deficits of dataset and/or to misclassification of raw materials. In order to solve this problems, we propose a method to calculate the feature of thermosetting resin based on raw material classification, and to predict physical properties of thermosetting resin by machine learning. First, the classificationｓ of raw materials were predicted by random forest using Extended Circular Fingerprint (ECFP) as an explanatory variable, representing the structural formula of raw materials. As a result, we could predict the classifications of raw materials, with high accuracy (F value = 0.96). Second, the resin physical properties were predicted by using ECFP. For the explanatory variables, we employed the features of the reaction intermediate, the curing condition of the resin, and the integration of ECFPs for each classification of raw materials. As a result, the elastic modulus could be predicted with the accuracy of R2 = 0.8. From these results, it was confirmed that the proposed method can predict the physical properties of the thermosetting resin synthesized by mixed raw materials.
