Skip to main content

Nonlinear models based on enhanced Kriging interpolation for prediction of rock joint shear strength

Authors: 

Mahdi Hasanipanah, Debiao Meng, Behrooz Keshtegar, Nguyen-Thoi Trung, Duc-Kien Thai

Source title: 
Neural Computing and Applications, 2020 (ISI)
Academic year of acceptance: 
2020-2021
Abstract: 

One of the most basic topics in rock mechanic is the shear strength criteria for rock joints. Thus, it is of high importance to accurately predict the shear strength of rock joints. In this study, the abilities for accuracy and agreement of Kriging model-based nonlinear interpolation strategy are investigated in terms of predicting the shear strength of rock joints. Totally 84 datasets were used to construct the Kriging models; the datasets were divided into two main parts: training and testing. The prepared database was applied to the training phase in the Kriging model; this way, several nonlinear basic functions were introduced to enhance the predictions of the Kriging model. The examined functions in this paper were linear, 2-order, 3-order, exponential, logarithmic, logistic, hyperbolic tangent, and hyperbolic sine. The sigmoid forms of the basic functions, including logistic and hyperbolic tangent, provide the superior predictions compared to other mathematical functions, while the 2-order and 3-order forms provide the worst performances than the linear, exponential, and logarithmic functions. According to the obtained results, the logistic-based model with coefficient of determination (R2) of 0.916 was found the optimal model that can be successfully applied to estimating the shear strength of rock joints.