Modeling uniaxial compressive strength of building stones using non-destructive test results as neural networks input parameters

Abstract

Uniaxial compressive strength value (UCS) is used as a critical input parameter in determining the engineering properties of natural building stones. The purpose of present study was to develop a model to determine the UCS of natural building stones via relatively simple and low-cost mechanical tests with the application of artificial neural networks. For this purpose uniaxial compressive strength, ultrasonic pulse velocities, Schmidt hammer hardness, and Shore hardness tests were performed on 37 different specimens of natural building stones collected from various natural stone processing plants in Turkey. The artificial neural networks (ANNs) approach was utilized for the development of the model that predicts the UCS.The major goal was to develop a model that makes the best prediction with the fewest number of input parameters. Therefore, analyses for verification of the models started with single input parameter and then combinations of two and three input parameters were used. For that purpose, two separate approaches were utilized with seven different sets of analyses in each method. The results of the ANNs models were compared with respect to the results of regression models. The criteria used to evaluate the predictive performances of the models were the coefficient of determination (R2), root mean square error (RMSE), and variance account for (VAF). The results show that the proposed ANNs method could be applied effectively for the prediction of UCS either from one of the input parameters or from their combinations i.e. ultrasonic pulse velocity, Schmidt hammer hardness and Shore hardness.