Optimizing interlaminar toughening of carbon-based filler/polymer nanocomposites by machine learning

Abstract

Currently, most designs for interlayer toughening of carbon-based filler/polymer nanocomposites are highly dependent on experimental iterative trial and error, and there is no rational design framework. This work uses machine learning to build a fast and accurate predictive model and assess the extent to which key features affect performance, giving researchers ideas for designing new materials and greatly improving efficiency. A training database is built by first collecting the features of the domain that affect the interlaminar performance. A stacking model fusion of the three machine learning models was then performed to construct a highly accurate fast prediction model. Besides, the importance of key features is evaluated during model training using the Random Forest Algorithm (RFA). Finally, by predicting the performance of materials and analyzing the importance of characteristics to guide material preparation, the development cycle is shortened and costs are reduced.