The monitoring feedback analysis with measured data such as displacements, stresses and supporting forces is a common approach to appraise or predict the safety of a project in use or during construction. In this paper, the three-dimensional computer code for fast Lagrangian analysis of continua, FLAC3D, with the explicit finite difference method are used in the numerical simulation, while the artificial neural network (ANN) is exploited to back analyze the material parameters adopted in the final computation. The feedback analysis of the excavation of the underground caverns of Xiluodu hydropower station on Jinsha River in southwest China is conducted as an engineering application. The implementation process of this method is introduced in detail. The process can feed back and forecast displacements, stresses and supporting forces being interested in the plan and construction of underground power stations. It is shown by this paper that the ANN-based feedback analysis method is effective and feasible.
It is well known that the numerically computational results of rock engineering depend on the various parameters used in the idealized model. In the underground projects, e.g. underground hydropower plant, the analysis of the rock stability during the excavation is based on the measured data. The analysis accuracy is closely related to the selected rock properties. The monitoring feedback analysis with measured data such as displacements, stresses and so on is helpful to determine the values of these parameters (Feng 2000, Feng et al. 2007).
In the numerical simulation, the relation between the response of the underground hydropower plants and the adopted rock parameters is very complicated and implicit. As we know, the artificial neural network (ANN) is suitable to solve such kind of problems as finding the nonlinear mapping of inputs and outputs (Sonmez et al. 2006).
As for the numerical computation, there already exists a number of commercial software, among which the three-dimensional computer code for fast Lagrangian analysis of continua, FLAC3D, with the explicit finite difference method is usually the resort in the rock mechanics and rock engineering (Itasca 1997).
Recently, feedback analysis is used in underground projects more and more (Han et al. 1997, Li et al. 1998). The basic idea lies in: after the former excavation, we get the data of displacement or stress of surrounding rocks by measuring; calculate the excavation process using numerical model, adjust the material parameters until the calculation results are consist with the measured data; then calculate the next process, predict the variety of deformation and stress of rocks; forecast risk that may be met and giving the salvations.
Feedback analysis can reduce much more computational efforts, predict hidden risks. It contains two main stages, i.e. the forward calculation and the feedback analysis of parameters.
Thus this feedback analysis process based on ANN can consist of following steps: Firstly, build a FLAC3D model according to the geological data, excavation scheme and caverns' layout. Secondly, determine the rock-mass parameters which need to be back analyzed and arrange a series of numerical tests following orthogonal designation.