TY - EJOU AU - Xia, Pengcheng AU - Pan, Zhihong AU - Chen, Ruoyu AU - Wang, Linyuan TI - Study on Prediction of Grouting Material Curing Age Based on SAFT and Hyperparameter-Optimized XGBoost T2 - Structural Durability \& Health Monitoring PY - VL - IS - SN - 1930-2991 AB - The grouting sleeves in prefabricated structures critically depend on the strength development of the grout; however, existing non-destructive testing methods struggle to capture its time-dependent evolution. This study proposes a hybrid prediction framework that combines the Synthetic Aperture Focusing Technique (SAFT) with a hyperparameter-optimized XGBoost model. Ultrasonic signals were collected at five curing stages (0, 1, 3, 7, and 28 days), from which SAFT-derived features and the area ratios of six color regions were extracted as input variables, with the curing age serving as the model output. Following a correlation analysis with compressive strength, three optimization algorithms—Random Search (RS), Bayesian Optimization (BO), and Adaptive Particle Swarm Optimization (APSO)—were employed to tune the hyperparameters of XGBoost. Among these, the BO-XGBoost model achieved the lowest prediction error within the data range of this study (R2 = 0.9881, MAE = 0.2037), outperforming the other comparative models. Further validation using leave-one-specimen-out cross-validation (LOSO-CV) indicated that, under the current laboratory conditions, the model maintained good prediction consistency across the three available sleeve specimens. Feature importance and partial dependence plots were used to further verify the physical relevance of the SAFT-derived features and to elucidate their positive or negative effects on strength. This method provides a preliminary interpretable framework for the non-destructive evaluation of grout strength development in prefabricated structures. KW - Ultrasonic testing; synthetic aperture focusing technique; grouting material; Bayesian optimization; machine learning DO - 10.32604/sdhm.2026.080742