Estimating the Hydraulic Conductivity of jointed rock mass using Genetic Programming

Abstract

Ground engineering through injection of cementitious grouts or polymer resins is an integral part of fractured rock mass stabilisation by improving its maximum load carrying capacity, stiffness, cohesiveness and reduction in permeability. To optimise product delivery and achieve a high penetrability of product into the rock mass, it is imperative to understand the permeability characteristics or hydraulic conductivity of the rock mass. The hydraulic conductivity of the rock mass also dictates the type of polyurethane (PU) or cement-based products to be injected. For this purpose, a Packer test has to be conducted to measure the hydraulic conductivity of a rock formation which is an expensive and time-consuming process. Alternatively, a number of empirical methods that use rock mass classifications and rock joint properties Rock Quality Designation (RQD), Q-system (Q), Geological Strength Index (GSI), Joint Spacing (JS), Joint Aperture (a) proposed by a number of researchers that are capable of predicting hydraulic conductivity (HC). In this research, a predictive model between rock mass properties and HC is proposed using new approach - genetic programming (GP). For this purpose, a database of rock mass parameters including RQD, Q, GSI, JS, Joint Aperture (a), Second Permeability Index (SPI) and Packer test results available in the literature is established. The database is split into randomly selected training and testing sets. To assess the fitting quality, the sum of the absolute difference is used, while maximum depth on trees is set to control the bloat of the model. The performance is assessed with four statistical criteria and three GP models using different input combinations are proposed. These models have been converted into simple mathematical equations to calculate HC based on collected input data. In summary, two out of three models have successfully predicted HC with high correlation to the actual HC (R2 of testing sets ≈ 0.92). Therefore, this study has shown the feasibility of applying GP models into future prediction of HC for the initial phase of rock grouting design.