This paper presents the development of Artificial Neural Network (ANN) models for predicting the stress-strain response of intact and jointed rocks. The stress-strain behaviour of ten different types of intact rocks is predicted by specifying the uniaxial compressive strength of the intact rocks, confining pressure and axial strain as inputs. The stress-strain response of jointed rocks is predicted by specifying the intact rock properties, confining pressure, joint properties and axial strain as inputs. The database used in this paper for the ANN analysis consists of five types of jointed rocks namely Plaster of Paris, block jointed Gypsum Plaster, Jamrani Sandstone, Agra Sandstone and Granite tested in triaxial compression under a wide range of confining pressures at different joint orientations and joint frequencies. The results obtained from the ANN analysis are compared with the experimental measurements and the results obtained from the continuum approach. Results from the analyses showed that the neural network approach is effective in capturing the stress-strain behaviour of intact rocks and the complex stress-strain behaviour of jointed rocks.
Jointed rock mass is a natural material with a large variation in its properties (highly heterogeneous and anisotropic). The prediction and estimation of its properties involve many uncertainties, as the material behaviour is stress path dependent. Also, obtaining an undisturbed sample is very difficult in jointed rock mass. Particularly, developing a general model which incorporates many rock types and includes intact and jointed rocks considering all the parameters (confining pressure, strength properties and joint properties) which govern the engineering behaviour of these rocks in an attempt to predict their stress-strain response is next to impossible. No such model exists in the literature and all the available models are only highly empirical ones. Artificial Neural Networks have been found to be very efficient and intelligent in handling non-linear relationships among the parameters involved in numerical modelling. Unlike the standard computational methods, neural networks use a parallel approach analogous to the functioning of the human brain. Application of neural networks for problems in rock mechanics was explored by several researchers earlier (Feng et al., [1]; Meulenkamp and Grima, [2]; Sitharam et al., [3]; Sonmez et al., [4]; Arunakumari and Latha, [5]). In this paper, an attempt has been made to construct Artificial Neural Network (ANN) models for predicting the stress-strain response of intact and jointed rocks. The objective of this paper is to develop a generalized network architecture which can predict the stress-strain response of any type of jointed rocks given the intact rock properties, confining pressure under which the rock is tested and the joint properties. The present paper covers two parts. In the first part the stress-strain behaviour of ten different types of intact rocks is predicted by specifying the uniaxial compressive strength of the intact rocks, confining pressure and axial strain as inputs. In the second part, the stress-strain response of jointed rocks is predicted by specifying the intact rock properties, confining pressure, joint properties and axial strain as inputs.
