New considerations for empirical estimation of tensile strength of rocks

In this paper, a combined laboratory and modeling study was conducted to develop a database for predicting tensile strength of rocks. Six hundred eighty six rock samples from 24 different sites throughout eastern Turkey were collected and tested for the development of this database and evaluation of models. A total of 512 samples were used for developing the models and the remaining 174 samples were used as control dataset. The material parameters selected in the development of the models include tensile strength (σ t), point load index (Is (50)), Schmidt rebound number (N) and unit weight (γ). A total of four models, two regression models, namely, simple linear regression and multiple regression, and two feed forward-type artificial neural network (ANN) models, namely, radial basis function network (RBFN) and multi-layer perceptron network (MLPN) are developed. A commercial software, Statistica 8.0, is used to develop these models. The strengths and weaknesses of the developed models were assessed by comparing the predicted σ t values with the experimental values with respect to the R 2 values. Overall, the MLPN model was found to be the best model for the present development and evaluation datasets. As a result of these analyses, an equation was suggested based on ANN model to estimate the tensile strength of rocks.