Hardness Prediction of WC-Co Cemented Carbide Based on Machine Learning Model

Abstract

The hardness of cemented carbides is a fundamental property that plays a significant role in their design, preparation, and application evaluation. This study aims to identify the critical factors affecting the hardness of WC-Co cemented carbides and develop a high-throughput predictive model for hardness. A dataset consisting raw material composition, sintering parameters and characterization results of cemented carbides was constructed in which the hardness of cemented carbide was set as the target variable. By analyzing the pearson correlation coefficient, shapley additive explanations(SHAP) results, WC grain size and Co content were determined to be the key characteristics influencing the hardness of cemented carbide. Subsequently, machine learning models such as support vector regression (SVR), polynomial regression (PR), gradient boosting decision tree (GBDT), and random forest (RF) were optimized to construct prediction models for hardness. Evaluations using 10-fold cross-validation demonstrated that the GBDT algorithm model exhibits the highest accuracy and strong generalization ability, making it most suitable for predicting and analyzing the hardness of cemented carbides. Based on predictions from GBDT algorithm model, PR algorithm model was established to achieve high-precision interpretable prediction of the hardness of cemented carbides. As a result, a quantitative relationship between hardness and Co content and WC grain size were obtained, which show that reducing grain size and Co content is the key to obtain high hardness of cemented carbide. This research provides a data-driven method for accurately and efficiently predicting cemented carbide properties, offering valuable insights for the design and development of high-performance cemented carbide materials.