Data-oriented prediction of rocks' Mohr-Coulomb parameters

Mohr-Coulomb shear strength parameters, namely cohesion strength and internal friction angle, are the key determinants of intact rock strength. Triaxial tests are generally performed at different stress levels to determine these shear parameters. However, as an alternative to this expensive and time-consuming method, data-oriented computational approach is available to employ and provides a new technical means for geomaterial strength parameters prediction. In this paper, random forest, alternating model tree, and support vector machine techniques are utilized to predict Mohr-Coulomb shear strength parameters by using the published database of the uniaxial compressive strength, uniaxial tensile strength, and different stress conditions in which failure occurs. In this regard, 80% of data (176 samples) are used to train models, while 20% (45 samples) is for testing the developed models. For internal friction angle, coefficient of determination (R-2), mean absolute error (MAE), and root mean squared error (RMSE) are R-2 >0.90, MAE < 1, and RMSE <1.6 degrees in the training and testing phases for established models. Based on the results, R-2 >0.98, MAE < 1.1, and RMSE <1.5 MPa are obtained for developed models of cohesion strength prediction in the training and testing phases, which demonstrate the efficiency of proposed approaches in predicting shear strength parameters of sandstone rock. The sensitivity analyses indicate that confining stress has the most significant influence in increasing the prediction accuracy. This work provides a general and robust data-centric intelligent framework for predicting micro-/macro-parameters of geomaterial, which improves shear strength design in the field.

Automated summary

This paper presents a data-oriented computational approach for predicting Mohr-Coulomb shear strength parameters. By utilizing a published database and random forest, alternating model tree, and support vector machine techniques, the shear strength parameters of sandstone rock were successfully predicted. The results showed that confining stress had a significant influence on prediction accuracy.