Paper No. 093 Β§ Full paper on enclosed CD-ROM Joint roughness has an important influence on the behaviour of rock discontinuities. In practice, determining the roughness with a high degree of confidence has been somewhat elusive. In the past, several methods have been employed relying on empirical, visual, statistical and fractals, with varying degrees of success. A popular means to quantify roughness is to employ the Joint Roughness Coefficient (JRC) which is used as the standard reference in engineering rock mechanics.
In this work, a series of direct shear tests were undertaken on natural discontinuities and the JRC values determined based on back-analysis. Prior and after each test, the joint profiles were recorded using laser equipment. It was then possible to construct the three-dimensional profile for each natural joint surface. Artificial neural networks were consequently trained on the laser-determined profiles and the back calculated JRC values for each joint. The developed networks were successful in quantifying and predicting joint roughness. The results obtained were superior to those derived by conventional statistical means. It is proposed that this methodology is a potentially useful tool in quantifying the joint roughness of natural rock joints.