Prediction of strength and crack coalescence mode of rock-like specimens with non-persistent joints under compression using machine learning algorithms

Sarfarazi, Vahab; Matinpoor, Fariborz; Mohammadi Bolbanabad, Shadman; Rezaei, Mohammad

Prediction of strength and crack coalescence mode of rock-like specimens with non-persistent joints under compression using machine learning algorithms

Document Type : Original Article

Authors

vahab sarfarazi ¹

fariborz matinpoor ²

Shadman Mohammadi Bolbanabad ³

Mohammad Rezaei ⁴

¹ mining engineering department

² tehran university

³ Department of Mining Engineering, Faculty of Engineering, Tarbiat Modares University, Tehran

⁴ Department of mining engineering, Faculty of engineering, University of Kurdistan, Sanandaj, Iran

Abstract

Jointed rock's strength characteristics and crack coalescence mode are considerably affected by its mechanical properties and pre-existing joint arrangement. In this paper, predictive models based on random forest (RF) and support vector machin (SVM) algorithms have been developed to predict strength (S) and crack coalesence mode (CCM) of jointed rock-like samples with non-persistent joints under uniaxial and biaxial compression tests. Mechanical properties of specimens (σc, σt, υ, E, C and tanφ), confining pressure (σn), number of joints (N), joint angle with respect to the horizon (β) and jointing coefficient (JC) were used as input parameters, while, S and CCM were assigned as output parameters. The performance of optimal RF and SVM models was evaluated based on the statistical criteria of coefficient of detemination (R2), root mean square error (RMSE), mean absolute error (MAE) and overall accuracy (OA). Also, multiple linear regression (MLR) model was used to predict S, as well as to evaluate and compare with optimal RF and SVM models. According to obtained results, it was concluded that both algorithms have superior efficiency and outperform the MLR model. Results of sensitivity analysis reveled that parameters C, σc, E and σt have the greatest effect on the strength of the samples, respectively, where JC parameter has the least effect on the strength of the specimens.

Keywords

non-persistent joint

Crack coalescence mode

SVM

Subjects