Application of neural networks for the reliability design of a tunnel in karst rock mass
Meho Saša Kovačević (University of Zagreb)
M. Bacic (University of Zagreb)
K.G. Gavin (TU Delft - Geo-engineering)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
This paper offers a solution to overcome time-consuming numerical analysis for the evaluation of the impact of tunnel construction in a complex karst environment by implementing Monte Carlo Simulation (MCS) using a neural network (NN) tool. The rock mass is described using three parameters: Geological Strength Index, the uniaxial compression strength of the intact rock, and the Hoek–Brown parameter for the intact rock mi . By using their probabilistic distribution as an input, a developed neural network NetTUNN produces probabilistic distributions of tunnel crown displacement, rock bolt axial load, and shotcrete uniaxial compression stress. A full MCS is then applied on these NetTUNN outputs to determine the reliability index and probability of failure for the relevant limit states. To demonstrate the potential of NN in tunnel design, a case study of Tunnel Pecine in Croatia is used, where the NetTUNN-assisted MCS assessment served as a benchmark to evaluate approximate reliability assessment techniques. It was shown that the developed NN can be used as an accurate surrogate model for determination of probabilistic distributions of tunnel design parameters. Further, it was shown that approximate reliability assessment techniques generally overestimate the reliability index and underestimate the probability of failure when compared to the NetTUNN-assisted MCS.