Applying finite different method FLAC and neural network to back-analysis the displacement of tunnel, the testing samples based on orthogonal test design and FLAC numerical simulation are used to constitute mapping of rock parameter and rock displacement, then the modulus of elasticity and lateral pressure coefficient of surrounding rock were obtained by the mapping from the measurement data. The accuracy of inverse parameters is checked by FLAC. The result shows that optimal solution of parameter can be reached from the displacement back analysis. The results can be feedback for the design of tunnel and also applied to the informative tunnel construction.
The complicated geotechnical condition makes some difficulties in determination of physical and mechanical parameters of rock mass. By the finite element and boundary element methods, the obtained mechanical parameters and initial geostress usually deviate from actual values. Therefore, back-analysis methods have been proposed in which the easiest, available and precise testing parameter is displacement in extremely complex rock and soil mass. The required parameters could be solved accurately by the back-analysis method based on the displacement monitoring (Zhu & Guo 2006, Lv & Jiang 1998). In this paper, the method of displacement back-analysis combining neural network with FLAC3D is studied, and a practical solution has been established. In practice, it enables the displacement back-analysis used for this kind of surrounding rock, and the mechanical parameters and initial geostress could be obtained.
On the issue of modeling, the closer to the objective reality in the employed mechanics model, the more number of parameters needed to be back-analyzed. However, the objects of study are rock and soil mass which are not homogeneous at all, it is a composite geologic body including joint, crack and faults.