A comparative study for the concrete compressive strength estimation using neural network and neuro-fuzzy modelling approaches

In this paper, adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN) model have been successfully used for the evaluation of relationships between concrete compressive strength and ultrasonic pulse velocity (UPV) values using the experimental data obtained from many cores taken from different reinforced concrete structures having different ages and unknown ratios of concrete mixtures. A comparative study is made using the neural nets and neuro-fuzzy (NF) techniques. Statistic measures were used to evaluate the performance of the models. Comparing of the results, it is found that the proposed ANFIS architecture with Gaussian membership function is found to perform better than the multilayer feed-forward ANN learning by backpropagation algorithm. The final results show that especially the ANFIS modelling may constitute an efficient tool for prediction of the concrete compressive strength. Architectures of the ANFIS and neural network established in the current study perform sufficiently in the estimation of concrete compressive strength, and particularly ANFIS model estimates closely follow the desired values. Both ANFIS and ANN techniques can be used in conditions where too many structures are to be examined in a restricted time. The presented approaches enable to practically find concrete strengths in the existing reinforced concrete structures, whose records of concrete mixture ratios are not available or present. Thus, researchers can easily evaluate the compressive strength of concrete specimens using UPV and density values. These methods also contribute to a remarkable reduction in the computational time without any significant loss of accuracy. A comparison of the results clearly shows that particularly the NF approach can be used effectively to predict the compressive strength of concrete using UPV and density values. In addition, these model architectures can be used as a nondestructive procedure for health monitoring of structural elements.