Resistivity-enhanced multi-physics machine learning framework for dynamic stress prediction in high sensitive UHPC

Machine learning provides powerful capabilities for modeling complex nonlinear relationships in cement-based materials, particularly in predicting mechanical behavior. Traditional compressive stress prediction models for UHPC rely predominantly on strain or displacement inputs and suffer from limited accuracy. Critically, electrical resistivity, as an intrinsic material property, demonstrates strong correlations with stress-induced microstructural evolution. This study proposes a machine learning framework for dynamic compressive stress prediction in high sensitivity ultra-high performance concrete (HS-UHPC) integrated with carbon nanotubes. Three distinct machine learning algorithms, double-layer neural network (DLNN), boosting tree (BT), and squared exponential Gaussian process regression (SE-GPR) were employed to establish compressive stress prediction model of HS-UHPC. Results demonstrated that incorporating resistivity measurements alongside displacement data significantly enhanced predictive accuracy, increasing R2 by 0.06 compared to displacement-only models. The SE-GPR model achieved the highest performance (R2=0.85, RMSE=0.11) under dual-input conditions, exhibiting a 41.1% reduction in mean absolute error compared to displacement-only models. Integrating resistivity as a key input parameter fundamentally enhances model performance by directly capturing stress-driven microstructural changes, establishing a superior sensing paradigm for ML-based structural health monitoring.