Predicting the Shear Strength of Rc Deep Beams with Wide Openings Using Fem and Machine Learning-Based Ni-Ti Sma Retrofitting

Abstract

The design of deep beams constructed with reinforced concrete (RC) having wide openings presents a significant engineering challenge due to the absence of specific provisions in existing design codes. This study explores the use of Ni-Ti shape memory alloys (SMA) for strengthening RC deep beams, aiming to mitigate brittle shear failure commonly encountered in conventional designs. Finite element modelling (FEM) and machine learning (ML) algorithms were employed to assess the performance of RC deep beams. FEM simulations investigated various SMA configurations, identifying the F_S7 layout as optimal, yielding a 10% increase in peak load capacity, enhanced ductile behaviour, improved energy absorption, and prevention of catastrophic structural failure. Additionally, six ML models were systematically evaluated using a dataset of 180 data points generated through Abaqus simulations to analyse the influence of key parameters such hole diameter, concrete compressive strength, and beam depth. Models’ performance was assessed using the coefficient of determination (R2) and other statistical metrics. To improve model’s interpretability, Shapley Additive Explanations (SHAP) and Partial Dependence Plots (PDPs) were utilized, providing insight into critical input-output relationships. Among the evaluated models, the Support Vector Machine (SVM) demonstrated superior performance, achieving R2 values of 0.957 for training and 0.956 for testing, establishing it as the most effective predictive model for RC deep beam performance. This study emphasizes a data-driven framework for enhancing deep beam behaviour and developing efficient, practical structural solutions.

Keywords: finite element modelling (FEM), Machine learning (ML), RC deep beams, AI in civil engineering