Soft computing modeling of RC beams without web reinforcement

In this thesis, the availability of soft computing (SC) techniques (Neural Networks (NN), genetic programming (GP) and Neuro-Fuzzy (NF) for the prediction and formulation of shear strength of reinforced concrete (RC) beams without web reinforcement was investigated. Previous soft computing applications on shear strength of RC beams without web reinforcement have been surveyed firstly. It has been found that neural networks and genetic programming has been applied to modeling of shear strength of RC beams. Therefore the scope of the thesis is focused on neuro-fuzzy modeling which has not been studied so far. Literature survey on previous experimental studies has also been carried out regarding shear strength of RC beams without web reinforcement and a wide range of experimental database (664 tests) has been gathered from literature from 56 separate studies. The proposed neuro-fuzzy model is based on this wide range of experimental database. Various types of membership functions (MF) such as Gaussian, Gaussian combination, Generalized bell-shaped, Triangular-shaped and Trapezoidal-shaped membership functions are evaluated for varying number of membership functions to obtain the optimum NF model. The accuracy of the proposed NF model is compared with accuracies of current design codes and existing shear strength equations and found to be more accurate.