Knowledge-Based Prediction of Load-Carrying Capacity of RC Flat Slab through Neural Network and FEM

The present research work aims to compare the results for predicting the ultimate response of Reinforced Concrete (RC) members using Current Design Codes (CDCs), an alternative method based on the Compressive Force Path (CFP) method, and Artificial Neural Network (ANN). For this purpose, the database of 145 samples of RC Flat Slab with the simple supported condition under concentrated load is developed from the latest published work. All the cases studied were Square Concrete Slabs (SCS). The critical parameters used as input for the study were column dimension, c s , depth of the slab, d s , shear span ratio, , longitudinal percentage steel ratio, Ï ls , yield strength of longitudinal steel, f yls , the compressive strength of concrete, f cs , and ultimate load-carrying capacity, V us . Seven ANN models were trained using different combinations of input parameters and different points of hidden neurons with different activation functions. The results exhibited that SCS-4 was the most optimized ANN model, having the maximum value of R (89%) with the least values of MSE (0.62%) and MAE (6.2%). It did not only reduce the error but also predicted accurate results with the least quantity of input parameters. The predictions obtained from the studied models (i.e., CDCs, CFP, and ANN) exhibited that results obtained using the ANNs model correlated well with the experimental data. Furthermore, the FEM results for the selected cases show the closer result to the ANN predictions.