Rock Brittleness Prediction Using Geomechanical Properties of Hamekasi Limestone: Regression and Artificial Neural Networks Analysis

The cold climate is a favorable parameter for the development of tension cracks and decrease of rock brittleness. Therefore, this paper attempts to investigate the Hamekasi porous limestone in order to predict the brittleness indices during freeze-thaw cycles. The freeze–thaw test was executed for one cycle including 16 h of freezing, and 8 h of thawing. The geo mechanical properties and brittleness indices (B1, B2, B3) of limestones were measured across freeze-thaw cycles from cycle 0 (fresh rock) to cycle 40. Statistical analyses, including simple and multiple regressions, were applied to identify those geomechanical parameters that are most influenced by the progression of freeze-thaw cycles and more appropriate for the brittleness prediction. Based on simple regression, all geomechanical properties including tensile strength ( ), uniaxial compressive strength ( ), P-wave velocity (Vp), porosity (n), and quick absorption index (QAI) (except dry density ( )) demonstrated good correlations with brittleness index (B3). The integrated prediction of brittleness is put forward to develop some models by multiple regression (MR) and artificial neural network (ANN) with some statistic parameters (R, RMSE, VAF and ME), based on all geomechanical properties examined in this research. It is concluded that models based on n, Vp and exhibited high performance according to the obtained statistic parameters. In spite of the fact that Vp has good correlation coefficient (R) with freeze-thaw cycles, and B3 (R2= 0.74, and 0.55, respectively) in simple regression, it does not have a prominent effect on B3 in MR models. Also, parameters with low correlation coefficient in simple regression ( =0.15) cannot improve the model performance in ANN methods.