Soft-computing based structural back analysis of material model parameters for soil in the context of tunneling according to the new Austrian tunneling method

This paper deals with determination of unknown parameters for a material model for soil used in Finite Element (FE) analyses of the structural behavior of tunnels. A gradient-free optimization method based on artificial intelligence is employed to find optimal parameters such that numerical results agree with available measurements as well as possible. A neural network is trained to provide an approximation to the FE simulations. Consequently, input parameters for FE simulations that have already been performed and the results from these simulations are the basis for training of the network. Using the trained network, a genetic algorithm searches for an optimal parameter set. The proposed parameter identification method (PIM) showed a satisfactory rate of convergence for a parameter identification problem involving FE simulations with one element only, Spira et al. (2001); Pichler and Mang (2001). It will be shown that this PIM is also well suited for the solution of real-life problems such as the investigation of the structural behavior of tunnels, involving time consuming FE simulations. For this purpose unknown soil parameters of a Drucker-Prager constitutive model for the soil are determined by means of back analysis in the context of a 2D simulation of a tunnel advance according to the principles of the New Austrian Tunneling Method (NATM).