B. Pichler:
"Effects of Monitoring-Data Errors of Tunnels on the Identification of Material Model Parameters of Soil";
Talk: 13th Inter-Institute-Seminar for Young Researchers, Vienna, Austria; 10-26-2001 - 10-28-2001.
The most popular method for analyzing the mechanical behavior of materials and structures is the FEM. Reliable analyses by means of this method require both an adequate mathematical formulation describing the characteristic physical properties of the material and a proper choice of the material parameters involved in this formulation. Commonly, some of the parameters cannot be determined directly from experimental results. Therefore, they need to be determined indirectly by means of Finite Element (FE) simulations and an identification procedure.Determination of unknown parameters for a material model for soil used in FE analyses of the structural behavior of tunnels is a basic requirement for progonoses of their structural behavior. Tunnel monitoring-data, e.g., displacements of measurment points of both the soil surface and the tunnel shell, can be used to identify unknown material model parameters of the soil. A gradient-free optimization method based on artificial intelligence [1] can be employed to solve this inverse problem. Consequently, it is used to find optimal parameters such that numerical results agree with available measurements as well as possible. A backpropagation artificial neural network is trained to provide an approximation to the underlying 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. The trained network provides a map of FE input-parameters onto selected values of respective numerical results. Using this network, a genetic algorithm and an extended version of the backpropagation algorithm yield a prognosis for an optimal parameter set [1].
This parameter identification method (PIM) was used to determine unknown soil parameters of a {Drucker}-{Prager} constitutive model for soil used in the context of an academic example [2]. A two-dimensional simulation of a tunnel advance according to the principles of the New Austrian Tunneling Method served as a numerical example of a complex geotechnical problem involving time-consuming FE simulations. The structure consists of two soil layers, sand and gravel, and the shotcrete tunnel shell. Results of a FE simulation of the structural behavior based on a reasonably chosen parameter set served as synthetic field data. In order to obatin an inverse problem, values for Young's Modulus, the angles of internal friction, and the characteristic time describing the evolution of viscoplastic strains were treated as unknown. The back analysis of these parameters was resting on the synthetic filed data, i.e., displacements of discrete points of both the soil surface and the tunnel shell, obtained from the aforementioned FE simulation. The PIM showed a satisfactory rate of convergence. The identification of five unknown parameters within a wide search range required 17 FE simulations. In real life problems, field data are afflicted with measurement errors. Although ranges of such errors can be estimated by experienced engineers, their exact values are unknown. This paper presents a study focusing on the effects of measurement errors on the identification of material model parameters of soil. Consequently, the aforementioned synthetic field data get affected with errors in a systematic manner. These error afflicted displacement data are the basis for back analyzing soil model parameters. Aftereffects on both (i) the discrepancies of identified parameters from their true values and (ii) the reliability of prognoses of the structural behavior of the tunnel resting on the obtained parameters are studied. It is shown that after adjusting the identification concept proposed in [2] satisfactory results can be obtained.
Keywords: parameter identification, inverse problem, NATM tunneling, measurement errors, prognosis of structural behavior