Verma, Ajeet Kumar
Reliability assessment of tunnels using machine learning algorithms - Vol.52(4), Aug - New York Springer 2022 - 780-798p.
In the present study, to examine the excavation-induced instability in the support system of rock tunnel, an analytical approach namely confinement convergence method (CCM) is employed with two different constitutive models, namely Mohr–coulomb criterion (MC) and Generalized Hoek–Brown criterion (GHB). Probabilistic analysis is performed on a circular tunnel with respect to two limiting functions: criteria regarding the radius of plastic zone and the tunnel convergence. Three surrogate models, namely collocation-based stochastic response surface method (CSRSM), multi-gene genetic programming (MGGP) and multivariate adaptive regression splines (MARS), are used in conjunction with the Monte-Carlo simulation (MCS) to calculate the probability of failure (Pf) of tunnel. The results of each methodology are compared with the traditional MCS, regarding the efficiency and the accuracy. Negative correlation between the shear strength parameters in the MC criterion is constructed using Gaussian copula. A detailed comparison is made among all three surrogates based on the Pf obtained using MCS. To incorporate the effect of the epistemic uncertainty, a Gaussian white noise having a specific variance level (p) is introduced and the noisy datasets are engaged by the MARS and CSRSM surrogate models for the probabilistic analysis.
Civil Engineering
Reliability assessment of tunnels using machine learning algorithms - Vol.52(4), Aug - New York Springer 2022 - 780-798p.
In the present study, to examine the excavation-induced instability in the support system of rock tunnel, an analytical approach namely confinement convergence method (CCM) is employed with two different constitutive models, namely Mohr–coulomb criterion (MC) and Generalized Hoek–Brown criterion (GHB). Probabilistic analysis is performed on a circular tunnel with respect to two limiting functions: criteria regarding the radius of plastic zone and the tunnel convergence. Three surrogate models, namely collocation-based stochastic response surface method (CSRSM), multi-gene genetic programming (MGGP) and multivariate adaptive regression splines (MARS), are used in conjunction with the Monte-Carlo simulation (MCS) to calculate the probability of failure (Pf) of tunnel. The results of each methodology are compared with the traditional MCS, regarding the efficiency and the accuracy. Negative correlation between the shear strength parameters in the MC criterion is constructed using Gaussian copula. A detailed comparison is made among all three surrogates based on the Pf obtained using MCS. To incorporate the effect of the epistemic uncertainty, a Gaussian white noise having a specific variance level (p) is introduced and the noisy datasets are engaged by the MARS and CSRSM surrogate models for the probabilistic analysis.
Civil Engineering