Performance Evaluation and Deep Learning-Based Prediction of CFRP-Strengthened RC Beams with Core-Cut Openings

The current study presents a data-driven framework to predict the structural performance of CFRP-strengthened reinforced concrete beams with various web openings using deep learning. Experimental data from beams with circular and elliptical openings under different CFRP wrapping configurations were used to train and evaluate machine learning models. The key structural parameters cracking load, initial/post-cracking stiffness, strain, and energy absorption served as input features, while ultimate load was the target variable. Four deep learning architectures ANN, CNN, RNN, and LSTM were implemented using TensorFlow/Keras and optimized using early stopping, dropout regularization, and uniform hyperparameters. Model performance was assessed using multiple statistical metrics including R², RMSE, MAE, VAF, NSE, and LMI. RNN and LSTM outperformed others, achieving R² values above 0.96 on the test set, with minimal residuals and stable loss convergence. Visualization tools such as regression plots, REC curves, ROC curves, and Taylor diagrams further validated predictive accuracy. The model’s interpretability was enhanced through Sensitivity and SHAP-based analysis, which identified ultimate load and initial load as the most influential predictors in determining structural behavior. The proposed approach offers a robust alternative to traditional analytical modeling by capturing nonlinear interdependencies and feature interactions within structural systems. The study demonstrated that deep learning, particularly recurrent architectures, can provide accurate and interpretable predictions of RC beam behavior, supporting efficient retrofitting decisions and structural safety assessments in real-world civil engineering applications.