Deep Learning Approaches to Predict Load -Carrying Capacity of Steel -Concrete Composite Beams

Accurately predicting the load-carrying capacity of steel-concrete composite beams is critical for ensuring structural safety and optimizing material use in modern construction. Traditional code-based methods often rely on simplified assumptions that may not fully capture the nonlinear behavior and complex interactions between steel and concrete components. This study presents a deep learning-based predictive framework for estimating the ultimate load capacity of composite beams using key geometric and material parameters as inputs. A dataset was constructed from a combination of experimental results and validated finite element simulations covering a wide range of beam configurations. Several deep learning architectures—including fully connected neural networks (FCNN), convolutional neural networks (CNN), and long short-term memory networks (LSTM)—were trained and evaluated. The models were benchmarked against conventional machine learning techniques such as support vector regression and random forests. Results indicate that the FCNN model outperformed others in terms of mean absolute error and R² score, demonstrating high predictive accuracy and robustness. Furthermore, feature importance analysis revealed that concrete compressive strength and steel section type are among the most influential parameters. The proposed deep learning approach offers a reliable and scalable tool for structural engineers seeking data-driven design support for composite systems.

کلیدواژه ها:

Steel-concrete composite beams ، Load-carrying capacity ، Deep learning ، Structural health prediction ، Neural networks ، Finite element simulation ، Feature importance ، Data-driven modeling