Recent Team Achievements - The Prediction of Tensile Strength Performance of Fiber-Reinforced Composites Based on a Knowledge Graph Constructed From Material Literature
Our team has published the paper “The Prediction of Tensile Strength Performance of Fibre-Reinforced Composites Based on a Knowledge Graph Constructed From Material Literature” in the international journal Polymer Composites (Impact Factor: 4.7, Q2 in CAS). The School of Computer Engineering and Science at Shanghai University is listed as the first affiliation. Qiaochuan Chen is the first author, Chen Zhao is the second author, and Yuexing Han and Na Song are the joint corresponding authors.
The prediction of tensile strength in fibre-reinforced polymer composites fundamentally relies on an accurate understanding of the relationship between “material components – forming processes – mechanical properties”. However, due to the lengthy experimental cycles and high costs associated with composites, obtaining sufficient data directly through experimentation is often challenging. Concurrently, the rapid accumulation of relevant literature has made literature mining and predictive modelling viable alternatives to partially replace experimental exploration.
To overcome the bottleneck of “data scarcity and fragmented knowledge”, our team proposed an integrated solution combining literature data extraction, knowledge graph construction, and machine learning prediction: First, we systematically organised and constructed the composite materials dataset ComMat, covering key elements such as materials, processes, testing, and performance; then utilised the joint extraction model PFPMHN to extract structured triples from literature, thereby constructing a domain-specific knowledge graph. Through reverse queries and feature screening, we identified key factors highly correlated with tensile strength.
Building upon this foundation, the selected features were employed to train a predictive model, achieving high accuracy in tensile strength forecasting. By integrating feature importance analysis, SHAP analysis, and OAT sensitivity analysis, we further identified and validated the critical variables influencing tensile strength, thereby providing actionable decision-making guidance for optimising composite material formulations and manufacturing processes.