Probabilistic analysis of tunnel loads using variance reduction

Abstract

The selection of input parameters for the numerical modelling of geotechnical structures is problematic due to their variability and uncertainty. Generally, parametric studies are required to evaluate the impact of input parameters on the modelling results. The utilisation of various statistical methods can bring significant benefits, such as probabilistic distributions of the modelling outputs and the probability of failure. This paper presents a variance reduction method known as ‘Latin hypercube sampling'. Special attention is paid to the Latin hypercube sampling mean, which represents a more efficient sampling scheme within the method's framework. The performance of the mean method is evaluated by comparison with the standard median method, and the necessary number of simulations is recommended. To take the statistical dependency of input variables into consideration the method mean is combined with the simulated annealing method. The effectiveness of Latin hypercube sampling methods and their practical application is demonstrated by static calculations for the Brusnice tunnel in the Czech Republic. The tunnel is a part of the Blanka tunnel complex on the Prague City Circle Road excavated by the new Austrian tunnelling method. The two-dimensional numerical model was prepared using the finite-element method. The modelling results were evaluated in terms of confidence intervals.

Numerical Modelling of the Soil Behaviour by Using Newly Developed Advanced Material Model

“…Unfortunately, one important fact is overlooked. Numerical analyses are only approximation of real behaviour and are highly dependent on the input parameters [9,10]. One area, which can affect the calculated results, is the correct choice of the material model [11][12][13].…”

Editorial

“…The sixth paper (Svoboda and Hilar, 2015) describes a method for understanding the probabilistic variance of soil properties with reference to the stresses acting on the Brusnice tunnel in the Czech Republic. The 'Latin hypercube sampling' (LHS) method is a statistical tool that can be used within an FE framework that is more computationally efficient than the Monte-Carlo method, reducing the number of simulations required to the order of tens or low hundreds.…”