Machine learning for revealing the relationship between the process–structure–properties of polypropylene in-reactor alloys

Abstract

Polypropylene in-reactor alloys present a complex structure influenced by diverse polymerization process parameters, posing challenges for traditional analysis methods in establishing a quantitative relationship between process conditions, alloy structures and mechanical properties. To address this issue, a series of polypropylene/poly(ethylene-co-propylene) alloys with varied structures were synthesized by gas-phase polymerization. Machine learning methods were employed to develop regression models for predicting flexural strength (FS), impact strength (IS) and rubber phase content. The importance of structure and process condition descriptors was further analysed to reveal the process–structure–property relationship. The FS and IS prediction models utilizing Extreme Gradient Boosting (XGB) algorithms achieved impressive R² scores of 0.9846 and 0.9841, respectively. Notably, the significant contribution of the rubber phase content to FS and IS prediction was observed in the structure descriptors. Furthermore, process condition descriptors (flowrate and initial pressure) played crucial roles in rubber synthesis, thereby exerting a substantial impact on FS and IS. In light of the feature importance analysis, new experimental runs were designed to synthesize alloys with enhanced IS. The experimental results closely aligned with the model predictions (RMSE = 4.4751 for IS). This research provides a new approach to establish process–structure–property relationships for in-reactor alloys, providing a convenient method for designing experiments to attain desired material properties.