A hybrid exploration approach for the prediction of geological changes ahead of mechanized tunnel excavation

The application of effective exploration techniques in mechanized tunneling is crucial in order to obtain a knowledge of the subsoil prior to drilling. We present a three-staged method, which seeks to image the excavation environment in detail, also reducing the computational demand of common exploration techniques. The algorithm combines two approaches: supervised machine learning and full waveform inversion. Firstly, the machine learning algorithm is applied on data sets of measured pore water pressures and ground settlements during tunnel propagation, making a primary prediction of geological changes ahead of the boring machine. Secondly, seismic measurements are acquired for full waveform inversion based on parameter identification. This method incorporates the primary predictions from the supervised machine learning in the form of a parametrization of the position, shape and material properties of the disturbance. Thirdly, the subsurface model gained out of the second stage is utilized as a starting model for a second full waveform inversion using the adjoint method, providing an even more detailed image of the subsurface. The exploration algorithm is tested on a synthetic shallow tunnel environment with an unknown obstacle ahead of the tunnel boring machine. It is shown that the algorithm finds the unknown obstacle with improved accuracy.

Automated summary

This paper proposes a three-staged exploration method that combines supervised machine learning and full waveform inversion to obtain a detailed image of the subsoil in mechanized tunneling. The algorithm is tested in a synthetic shallow tunnel environment and shows improved accuracy in detecting unknown obstacles.