Mechanical Property Prediction of Materials via a Literature Mining Approach

The prediction of material properties is critical for their design and application, with relevant process parameters frequently documented in scientific literatures. However, traditional methods of extracting information from literatures are often time-consuming and costly. To address the challenges, we propose a named entity recognition (NER) method based on heterogeneous graphs, referred to as CAGN, which enables the efficient extraction of key information from literatures. We conduct experiments on the Matscholor dataset and the results show that the accuracy of the method is 88.50%. Based on the results of the literature mining, this study identifies nine key features influencing the mechanical properties of composite materials through data consolidation and selection, and employs machine learning models to predict their flexural and tensile strengths. Experimental findings indicate that the XGBoost model outperforms others in predicting both properties, achieving R2 values of 0.87 and 0.83, respectively. The method in this paper provides valuable data support for applications in material science.