A new method for determining the hydraulic aperture of rough rock fractures using the support vector regression

Understanding fluid flow in rock fractures is relevant to many engineering applications, including underground nuclear waste disposition, oil and gas exploitation, dam and slope design, as well as contaminant transport. Rough rock fracture is usually idealized as two parallel plates separated by the hydraulic aperture, the determination of which is key to solving seepage problems in fractured rock using the cubic law. The mechanical aperture and fracture morphology are the main aspects affecting the hydraulic aperture. Here, an innovative hydraulic aperture prediction equation for two-dimensional fractures is established based on the mechanical aperture and fracture morphology. Six parameters were firstly selected to predict the hydraulic aperture, including three mechanical aperture parameters and three fracture morphology parameters. A training database was then established from numerical experiments of 300 fracture models, which were built based on Barton's 10 standard profiles and their combinations. The numerical experiments were conducted with FLUENT software. Based on the database, a hydraulic aperture prediction model was established using the support vector machine theory and the particle swarm optimization algorithm, which was transformed into an explicit equation. Finally, 20 natural fractures were used to validate the performance of this equation, with the comparisons of four existing empirical equations. The results show that this method worked well in predicting the hydraulic aperture. This method provides an alternative and effective way to directly determine the hydraulic aperture using the geometry information of rough rock fractures. •A training database for SVR (support vector regression) was built based on Barton's profiles and FLUENT software.•An SVR-based equation of hydraulic aperture was established by LIBSVM library and particle swarm optimization algorithm.•The proposed SVR-based empirical equation was shown superior to the existing empirical equations.