Numerical Simulation Using Nonlinear Finite Elements and Machine Learning for the Evaluation of Failure Probability of Reinforced Concrete Beams

Resumo

Reinforced concrete beams are fundamental structural elements designed to resist bending moments and shear forces resulting from gravitational loads. Their performance is critical for the structural integrity and serviceability of buildings and infrastructure. Accurate prediction of their load-bearing capacity contributes to safer and more economical structural designs.In recent years, artificial intelligence-based methods, particularly machine learning techniques, have gained increasing relevance in structural engineering due to their ability to model complex nonlinear behavior with high accuracy and reduced computational effort. Among these techniques, genetic programming has proven effective in discovering symbolic relationships and generating predictive equations for structural performance parameters.In this study, a numerical simulation of a reinforced concrete beam is carried out using the Abaqus software with nonlinear finite elements. The beam’s load-bearing capacity is evaluated for different combinations of parameters, including compressive strength, yield stress, reinforcement area, and depth. Based on the simulation results, a data cloud is created to train a genetic programming model capable of deriving an equation that estimates the beam’s ultimate strength without the need for further simulations.The predictive equation generated through genetic programming is employed to estimate the probability of failure of the beam using structural reliability methods, allowing for the probabilistic evaluation of safety in reinforced concrete beam design.