Tunneling performance prediction using an integrated GIS and neural network

Abstract

This paper presents the use of artificial neural networks (ANNs) as an alternative tool for tunneling performance prediction required in routine tunnel design works. The proposed approach involves the development of ANNs using calibrated finite element models so that the trained ANNs can establish relationships between tunneling conditions and performance in terms of stability as well as impact on surrounding environment for a particular tunneling site. A novel feature of the proposed procedure is an ability to expedite the tunneling performance assessment process that otherwise requires significant time and effort, thus facilitating routine tunnel design works. The methodology was implemented using an urban high-speed railway tunnel design project to demonstrate its potential for use as a tunnel design tool. It is shown that ANNs, when generalized by the results of finite element analyses (FEAs), can make relevant predictions for routine design work with a comparable degree of accuracy of the FEA. This paper describes the concept and details of the proposed approach and its implementation to an urban tunnel design.