The Uniaxial Compressive Strength (UCS) of intact rocks is an essential index of strength in rock engineering. Laboratory based direct compressive strength estimation may be problematic, as obtaining fresh samples is not always feasible. Thus, the aim of indirect methods index test such as point load index test, and empirical correlations with UCS of indexes like the Brazilian indirect tensile strength test, serve as an alternative for many geotechnical engineering projects. The aim of this paper is to propose a relationship between UCS and indirect tests or indexes for some sedimentary and igneous rocks in KwaZulu-Natal using the technology of artificial intelligence. These tests include the point load index (Is (50)) test and Brazilian Tensile Strength (σt, test. Block samples were collected in KwaZulu Natal, among these include sedimentary rocks (sandstones, siltstone, tillite) and igneous rocks (granitoids and dolerite). A back propagation artificial neural network was developed and trained in order to predict UCS. The input parameters were unit weight γ, (Is (50)), (σt), and lithology. The lithology was introduced in the neural network as a qualitative input parameter, in order to indirectly incorporate in the model the mineralogical content. Training results returned, R value of 0.99% for the training set, and R = 0.92% for the test set, which is conveying to the conclusion that the approach is valid and could be used, as an alternative indirect approach to UCS estimation.
Geotechnical design requires the estimation of intact rock properties. Strength of intact rock is a key parameter required in rock mass classification systems (RMR) , Q-system , and Hoek - Brown criterion . Typically, the strength and the modulus of elasticity of intact rock can be determined by the unconfined compression test. This test is standardized by the International society for Rock Mechanics (ISRM 2007) . Direct determination of these properties in the laboratory are complicated and often time consuming [5, 6]. Hence, the indirect estimation of UCS using rock index tests is of interest. The aim of this paper is the assessment of a relationship between UCS and indirect tests or indexes used to estimate the value of UCS based on data from sedimentary and igneous rocks in KwaZulu-Natal. These tests include the point load index (Is (50)) test and the Brazilian Tensile Strength test. An artificial neural network was developed to predict a reliable UCS value. A back propagation ANN was developed and trained in order to predict UCS value based on blocks sample data from the (29) sedimentary and (14) igneous rocks. Among these 43 rock blocks samples were cored for the UCS, test, 129 for the point load test and 258 for the Brazilian test. The input parameters were unit weight γ, (Is (50)), (σt), and lithology. The lithologies that are abundant in the KwaZulu-Natal province and were available for this study, are granitoid rocks, dolerite, sandstone, and tillite.