Prediction of compressive strength from microfabrics properties of banded amphibolite rocks using artificial neural networks and multivariate regression techniques

Abstract

In complex inherent characteristics of certain rocks, especially anisotropic rock, it may be difficult to measure the uniaxial compressive strength (UCS). However, the use of empirical relationships to estimate UCS of rocks can be more practical and economical. In this study, the prediction capability of the Artificial Neural Networks (ANNs) and multivariate regression methods has been carried out to predict UCS from microfabrics properties of banded amphibolite rocks. Based on statistical analysis, microfabrics parameters including grain Size, shape factor and quartz content that adequately affect the values of UCS have been adopted in this study. The ANNs model was performed using the same input variables as multivariate regression model. To assess the models performances, some performance indices such as correlation coefficient (R), variance account for (VAF) and root mean square error (RMSE) were calculated and compared for the two models. The results show that even though the developed two models are reliable to predict the UCS, the study clearly indicates the superiority of the ANN model based on the model performance indices. This approach can be easily extended to the modeling of rock strength and deformation parameters in the absence of adequate geological information or abundant data.