Interpretable and Uncertainty-Aware Machine Learning for Shear Strength Prediction of FRCM-Strengthened RC Beams

Abstract

Accurate prediction of the shear capacity of reinforced concrete (RC) beams strengthened with fabric-reinforced cementitious matrix (FRCM) systems remains challenging due to the complex interaction between geometry, internal reinforcement, and parameters related to the strengthening technique. This study proposes an interpretable and uncertainty-aware machine-learning framework for estimating the shear capacity of FRCM-strengthened beams. A database comprising 174 experimental beam tests collected from the literature was assembled and, to enrich the training space under limited experimental coverage, a hybrid tabular variational autoencoder (TVAE) framework was used to generate 6,000 synthetic samples from the training subset. The resulting experimental and augmented data sets were used to develop three machine-learning models: linear regression, support vector regression, and extreme gradient boosting. An existing analytical model was also evaluated for comparison. Among all approaches, the extreme gradient boosting model achieved the highest predictive accuracy, with R 2 = 0.911 on the testing set and R 2 = 0.949 for the complete data set, and also exhibited stable performance on the synthetic data set. Model interpretability was examined using shapley additive explanations (SHAP)-based explanations together with permutation-based importance analysis, which consistently identified effective depth as the most influential variable, followed by transverse-reinforcement ratio and shear span-to-depth ratio. To quantify feature-importance uncertainty, a fuzzy ensemble feature importance analysis was conducted. Effective depth exhibited the most stable importance pattern, whereas several FRCM-related parameters showed moderate importance with greater uncertainty. Introducing model-form uncertainty through a multimodel ensemble reduced the relative importance of several predictors. Contextual fuzzy rules further revealed distinct feature-state patterns associated with low, moderate, and high shear-capacity regimes. To improve practical applicability, the validated extreme gradient boosting model was further distilled into an explicit two-regime design-oriented formula with preliminary reliability calibration. Overall, integrating machine-learning prediction with fuzzy ensemble interpretability and TVAE-assisted design-oriented distillation enables accurate, transparent, and uncertainty-aware assessment of shear capacity in FRCM-strengthened concrete beams.