Hybrid surrogate modelling for mechanised tunnelling simulations with uncertain data

Abstract

For numerical reliability analyses of engineering structures, highly efficient computational models are required, in particular if numerical simulations of construction processes are to be performed in real time during the construction. Adopting process-oriented finite element simulations in mechanised tunnelling as a specific field of application, surrogate models are proposed to substitute inevitable expensive complex finite element simulations with uncertain parameters. The focus in this paper is laid on the uncertainty of geotechnical and tunnelling process parameters described by stochastic numbers, intervals and interval stochastic numbers. A new hybrid surrogate modelling concept is presented, which is based on combining artificial neural networks and the proper orthogonal decomposition method. Thereby, uncertain time variant surface settlements of several monitoring points are predicted by recurrent neural networks. Based on these predictions, the complete surface settlement field of each time step is computed using the gappy proper orthogonal decomposition approach. The hybrid surrogate model is applied for reliability analyses of a mechanised tunnelling process, where limit states at multiple surface positions are evaluated.