Page 220 - Numerical Analysis and Modelling in Geomechanics

Chapter 7
Parameter estimation using extended

Bayesian method in tunnelling
I.-M.Lee and D.-H.Kim

Abstract
This chapter addresses a parameter estimation technique to be applied in
tunnelling. The extended Bayesian method (EBM) is adopted, which can
systematically combine field measurements and prior information of
underground structures in order to obtain the best estimate of geotechnical
parameters. For the EBM, the relative importance of prior information over field
measurements is defined by the parameter, β, which is determined from the
sensitivity of geotechnical parameters and the uncertainty existing in the prior
information and the measurements. In the present study were various
geotechnical parameters were determined, including the elastic modulus (E), the
initial horizontal stress coefficient (K ), the cohesion (c) and the internal
0
frictional angle (ф). The validity of the feedback system proposed herein was
demonstrated through an elasto-plastic example problem. The proposed method
was applied to an actual tunnel site in Pusan, Korea and has shown to be highly
effective in actual field problems.

Introduction

In spite of dramatic developments in underground technology, there are still
many uncertainties that exist in the design and construction of underground
structures. This is mainly due to the discrepancy between laboratory and in situ
tests and limitations of site investigation techniques during the design stage. In
order to reduce these uncertainties, field instrumentation results obtained during
construction are compared with initially estimated ground properties. A feedback
system can be used to estimate optimum ground properties by minimizing the
difference between predicted and measured ground motions.
The ordinary least squares (OLS) method is widely used, because of its easy
application to non-linear geotechnical problems without complex mathematical
concepts. However, this method cannot consider prior information in the process
of parameter estimation. The Bayesian approach incorporates both prior
information and measurement data (Cividini, Maier and Nappi 1983:215). In the