Probabilistic evaluation of ground-support interaction for deep rock excavation using artificial neural network and uniform design

An efficient approach is proposed in this paper for probabilistic ground-support interaction analysis of deep rock excavation using the artificial neural network (ANN) and uniform design. The deterministic model is based on the convergence–confinement method. The ANN model is employed as the response surface to fit the real limit state surface. The uniform design table is used to prepare the sampling points for training the ANN and for determining the parameters of the network via an iterative procedure. The probability of failure is estimated from the first-order and second-order reliability method (FORM/SORM) based on the generated ANN response surface and compared with Monte Carlo simulations and polynomial response surface method. The efficiency and the accuracy of the proposed approach are first illustrated with the case of a circular tunnel involving analytical solutions with respect to three performance functions. The results show that the support installation position and the parametric correlations have great influence on the probability of the three failure modes. Reliability analyses involving four-parameter beta distributions are also investigated. Finally, an example of a deep rock cavern excavation is presented to illustrate the feasibility of the proposed approach for practical applications where complex numerical procedures are needed to compute the performance function.

► ANN model is employed as the response surface to fit real limit state surface.
► Uniform design table is used to prepare training datasets for establishing ANN.
► Support installation position and parametric correlations significantly influence results.
► ANN method has advantage in accuracy and efficiency over polynomial RSM and MCS.