Synergistic Effects of HPMC-HEMC Blending on the Rheological Properties of Composite Cement Paste Based on...

Fresh composite cement pastes modified with HPMC-HEMC blends were investigated using an explainable machine learning framework to clarify the coupled effects of cellulose ether dosage and blending ratio on rheological performance. A dataset containing 225 mixtures and 11 input variables was established, with fluidity, dynamic yield stress, and plastic viscosity used as target outputs. HPMC and HEMC dosages ranged up to 0.394% and 0.393%, respectively, while the total cellulose ether dosage varied from 0.041% to 0.398%. The results showed that total cellulose ether dosage strongly reduced fluidity but increased yield stress and viscosity, with Pearson correlation coefficients of -0.618, 0.653, and 0.848, respectively. Among the machine learning models, SVR achieved the best fluidity prediction where R² = 0.918, XGBoost performed best for dynamic yield stress where R² = 0.896, and CatBoost gave the highest accuracy for plastic viscosity where R² = 0.941. SHAP and PDP analyses identified total cellulose ether dosage, W/B ratio, silica fume, and sand-to-binder ratio as dominant variables. The optimal blending window was located near a balanced HPMC-HEMC ratio of 0.50, where HPMC = 0.092% and HEMC = 0.092%.