Tenacity Prediction of Polyester Materials Based on an Ensemble Soft Measurement Model

In the development process of polyester powder coating materials, it has great guiding significance for the development of new materials as the polyester material properties can be predicted through the given monomer structure and formula. In order to meet the requirements of rapid development of polyester materials, this paper establishes a prediction method based on an ensemble soft measurement model (ESM) for one key property--tenacity in prompting the acceleration of development cycle of polyester products. The processed data employed to train and test the model were collected from a local material company, and it covers over 100 kinds of polymerized resins. Firstly, the molecular expressions of monomers are converted to SMILES codes and further translated to ECFP fingerprints. Then, the principal component analysis (PCA) method is adopted to extract the elements with higher importance in each fingerprint. Finally, the ensemble model that consists of 4 machine learning models was established to make the multiclassification decision in tenacity prediction. The effectiveness of the prediction model is evaluated by comparing the predicted and the practical tensile test results. The results demonstrate that this approach achieves better performance in the material property prediction as well as generalization capability.