A predictive model for the peak shear strength of infilled soft rock joints developed with a multilayer perceptron

Abstract

Several analytical methodologies help estimate the shear strength of rock discontinuities whose main limitations are the difficulty to obtain all necessary parameters to satisfactorily represent the boundary conditions and influence of infill materials. The objective of this study is to present a predictive model of peak shear strength for soft rock discontinuities developed making use of an artificial neural network known as multilayer perceptron. The model's input variables are: normal stiffness; initial normal stress acting on the discontinuity; joint roughness coefficient (JRC); ratio t/a (fill thickness/asperity height); uniaxial compressive strength and the basic friction angle of the intact rock; and finally the internal friction angle of infill material. To do so, results from 115 direct shear tests, with different soft rock discontinuities conditions were used. The herein proposed ANN predictive model, with an architecture 7-20-1, have shown coefficient of correlation in training and validation of 99.8 % and 99 %, respectively. The results from the model satisfactorily fit the experimental data and were also able to represent the influence of the input variables on the peak shear strength of soft rock discontinuities for different infill and boundary conditions.