Comparative study of regression and evolutionary models for prediction of concrete compressive strength by means of concrete cores

Core testing is the most direct method to assess the in-situ concrete compressive strength in an existing structure, generally related to suspected construction malpractice or deficiency of concrete supply, to carry out the condition assessment of buildings before taking up repair and upgrading work. Although this test is quite simple to conduct, the results obtained may sometimes contain considerable errors because of the great variety of parameters involved. The general problems of core testing are well known. The factors including core diameter, length-to-diameter ratio (L/D), concrete age, aggregate characteristics, direction of coring and the moisture condition at the time of testing are known which affect the relationship between core strength and the corresponding standard cube or cylinder strength are fully reported by researchers. Another potential factor influencing the testing of cores is the presence of reinforcing bars within the core. The effects of the presence of steel bars on the strength of cores have been investigated by only a few researchers. Reinforcement bars passing through a core will increase the uncertainty of results and should be avoided wherever possible. Regression analysis and generalized GMDH network, whose structure is investigated using genetic algorithm and single-particle number optimization method for predicting the compressive strength of concrete using the results of coring tests with and without fittings. The form and ability of the multivariate linear regression models and the importance of regression coefficients based on the experimental data obtained for samples in two different processing conditions, in order to predict the cubic compressive strength of the concrete and using the input parameters including (1) the length to diameter ratio Core, (2) core diameter, (3) diameter, (4) number and (5) axial axial axis of the rebar in the core, (6) reinforcement of the rebar, and (7) core compressive strength as independent and input variables, as well as resistance Concrete pressure is evaluated as the response variable (or output of the models). This method is used for the GMDH neural network. The objective of the GMDH neural network method is to obtain a polynomial function that can be used to retrieve the output parameter by the input of the considered variables. The GMDH neural network can, after training, estimate the relationship between inputs and outputs in a polynomial, which depends on the accuracy of this polynomial on the data and structure of the network. The single-particle decomposition (SVD) method in the GMDH structure for the case where the number of equations is greater than that of unknowns, uses the least squares error method to solve such devices. The results showed that the models used have high ability to express the problem, since more than 95% of variations of response variables with fitted models in regression models and about 99% of changes in the response variable values ​​in the GMDH model can be expressed. But in a comparative position, GMDH model with a general structure optimized with Genetic Algorithm and SVD has shown the best performance, with this superiority becoming noticeable, considering that about 75% of the data is involved in training the neural model. Subsequently, nonlinear regression models show a certain advantage over linear models.