Generalized GMDH-Type Neural Network for Prediction of Concrete Compressive Strength via Core Testing

In this paper, generalized Group Method of Data Handling (GMDH)-type neural network has been successfully used for modeling concrete core testing including reinforcing bars based on various data obtained experimentally. Genetic Algorithm (GA) and Singular Value Decomposition (SVD) techniques are deployed for optimal design of such model. A set of input-output data for the training and testing the evolved models are employed in which core diameter, length-to-diameter ratio, number of reinforcing bars, distance of bar axis from nearer end of core as well as strength of cores, with or without reinforcing bars, are considered as inputs and standard cube strength of concrete is regarded as the output variables. The comparison of results obtained experimentally in this work with the proposed GMDH model depicted that this model has a great ability for prediction of the concrete compressive strength on the basis of core testing. Finally, sensitivity analysis was performed on the models obtained by GMDH-type neural network to study the influence of input parameters on model output. The sensitivity analysis reveals that the output variable (standard cube strength) is significantly changed by core strength and number of rebars in comparison with other input variables.