A predictive model for uniaxial compressive strength of carbonate rocks from Schmidt hardness

Uniaxial compressive strength (UCS) is considered to be one of the important parameters in rock engineering projects. In order to determine UCS, direct and indirect techniques are employed. In the direct approach, UCS is determined from the laboratory UCS test. In indirect techniques determine UCS based on the nondestructive test findings which can be easily and quickly performed and require relatively simple or no sample preparation. Indirect techniques are commonly preferred by rock and mining engineers because of their low cost and ease. This study presents the findings of an Artificial Neural Networks (ANN) based model for the prediction of UCS from Schmidt hardness. Schmidt hardness test (SHT) is a nondestructive test method which provides fairly good correlation about the strength of rocks. SHT can be easily and quickly conducted with a portable device known as Schmidt Hammer and it does not require any sample preparation. ANNs have been widely used in solving engineering problems and have emerged as powerful and versatile computational tools for organizing and correlating information in ways that have proved useful for solving certain types of problems too complex, too poorly understood, or too resource-intensive to tackle using more traditional numerical and statistical methods. For this reason, ANNs are used in this study to predict UCS of carbonate rocks from the Schmidt hardness rebound value (N R). A set of 37 test measurements obtained from 37 different carbonate rocks (marble, limestone, and travertine) are used to develop the ANN-based model. The results of the ANN model were also compared against the results of a regression model. The criteria used to evaluate the predictive performances of the models were the coefficient of determination (R 2), root mean square error (RMSE), and variance account for (VAF). The R 2, RMSE, VAF indices were calculated as 0.39, 46.51, 12.45 for the regression model and 0.96, 7.92, 95.84 for the ANN model, respectively. The results show that ANN-based model produces significantly better results than the regression model. It was concluded that the N R value is a useful indicator for the prediction of UCS from the ANN model developed in this study.