A Multi-Output Deep Learning Framework for Automated Reinforced Concrete (RC) Beam Design and Detailing

This study presents a machine learning framework for the automated design of reinforced concrete (RC) beams in compliance with ACI 318-19 code provisions. Traditional RC beam design requires iterative calculations to satisfy strength and serviceability criteria across a wide range of geometric, material, and loading parameters. To enhance efficiency and consistency, a dataset comprising 10,000 synthetically generated beam configurations was developed by varying span length, concrete compressive strength, steel yield stress, and applied loading. Each configuration was generated using ACI 318-19 design equations to ensure code compliance and structural validity. A deep neural network (DNN) regression model was trained to learn the nonlinear mapping between input parameters and corresponding design outputs, including required tensile reinforcement, bar diameters, stirrup spacing, and ultimate moment capacity. Model performance was evaluated using a quantitative error matrix reporting mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R<sup>2</sup>) for each output variable. The model achieved MAE values ranging from 0.35 to 10 units, RMSE values from 0.45 to 12 units, and R<sup>2</sup> values between 0.95 and 0.99, demonstrating high predictive accuracy and strong agreement with ACI-based reference designs. These results confirm that the framework can automatically generate code-compliant RC beam designs with high fidelity to ACI 318-19 specifications. By providing consistent, rapid, and interpretable predictions, this approach establishes a foundation for AI-assisted structural engineering tools that reduce computational time, improve design accuracy, and support data-driven automation in structural design workflows.