Prediction of load-bearing capacity of RC columns (CWA) using Artificial Neural Networks (ANN) trained on a hybrid experimental database HEXP

This study presents a comparative analysis of predicting the load-bearing capacity of Reinforced Concrete (RC) columns using contemporary design codes and alternative methodologies, specifically Artificial Neural Networks (ANN) and the Compressive Force Path (CFP) method. The ANN models were trained on a hybrid enriched experimental dataset (HEXP). Comparisons with current design codes, CFP, and ANN models reveal that the ANN predictions most accurately reflect the experimental results. The CFP method also provides estimates that closely match actual experimental outcomes. These comparative analyses identified and evaluated critical parameters width of columns in x-direction =b, width of columns in y-direction=d, Shear span ratio=av/d, Longitudinal steel ratio, Tensile strength of steel=fyl, Compressive strength of concrete=fc, Transverse steel ratio=pw, Axial Load=N, Flexural moment=M f , and Shear Capacity of member=V u , affecting RC column performance. The VANN model demonstrated superior stability and reliability with a Coefficient of Variation (CV) of 0.73, outperforming other models with higher CVs. The ANN model's predictions closely align with test data due to their derivation from experimental results. Furthermore, predictions from both ANN and CFP models were validated against ABAQUS simulations, with ANN predictions showing excellent agreement with ABAQUS outcomes. • The study evaluates load-bearing capacity predictions for RC columns using contemporary design codes, (ANN), and the (CFP) method. • ANN models trained on hybrid enriched experimental data (HEXP) demonstrated the highest accuracy, closely matching experimental results. • Key factors influencing the performance of RC columns were identified and evaluated, enhancing understanding of structural behavior. • Predictions from ANN and CFP models were validated using ABAQUS simulations, confirming the ANN model's excellent alignment with simulation results.