Bayrak B., Aydın A. C.
Journal of Brilliant Engineering (BEN), cilt.6, sa.3, ss.1-8, 2025 (Hakemli Dergi)
Özet
This study presents a comprehensive neural network approach for the structural design of reinforced concrete (RC) shear walls. A feed-forward back-propagation neural network (ANN) model was developed to predict the horizontal load capacity and maximum vertical load of continuous shear walls based on geometric and material parameters. The database, compiled from existing experimental studies and design recommendations, was divided into training, validation, and testing subsets in a 70-15-15 ratio. The Levenberg–Marquardt optimization algorithm was adopted to improve convergence efficiency and minimize mean-square error. The optimal architecture, consisting of two hidden layers with tan-sigmoid transfer functions and a linear output layer, demonstrated robust learning performance and generalization capability. The trained ANN model achieved up to 91% prediction accuracy when compared with design outcomes of real residential and commercial structures. Results indicated that the proposed model effectively captures nonlinear relationships between key variables such as wall aspect ratio, axial load ratio, and reinforcement spacing, yielding predictions that align closely with experimental results and code-based design equations. The study confirms that neural-network-based modeling provides an efficient and reliable computational framework for shear wall design, significantly reducing manual computation time while maintaining accuracy. Future research should focus on expanding the dataset and integrating adaptive learning strategies to further enhance model generalization for complex structural conditions.