Artificial neural network surrogate modelling for real-time predictions and control of building damage during mechanised tunnelling. (November 2020)

• Record Type: Journal Article
• Authors: Cao, B.T.; Obel, M.; Freitag, S.; Mark, P.; Meschke, G.
• Is Part Of: Advances in engineering software. Volume 149(2020)
• Publication Date: 2020-11
• Subjects: Surrogate modelling -- Artificial neural networks -- Finite element method -- Mechanised tunnelling -- Real-time prediction -- Building damage analysis; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2020.102869
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Real-time prediction of building damage in mechanised tunnelling. Finite element simulation of mechanised tunnelling processes including soil-structure interaction. Combination of two artificial neural network based surrogate models. Computation time is reduced from ca. 12 h two one second. Steering support software for tunnel boring machines to control the building damage risk during construction. Abstract: Tunnelling induced surface settlements can cause damage in buildings located in the vicinity of the tunnel. Currently, surface settlements and associated building damage risks usually are estimated based on empirical equations, e.g. by assuming Gaussian curves for the settlement trough and by applying the Limit Tensile Strain Method or the tilt-based method to evaluate and categorise the expected building damage. In this paper, finite element simulations are used to predict the soil-structure interaction in mechanised tunnelling during the tunnel advancement. The time variant surface settlement field and the corresponding tunnelling induced strains in the facade of a building are computed by two independent finite element models. Coupling both models allows predicting the expected category of damage (cod) for the building, given the operational parameters of the tunnel drive. Based upon this coupled approach, a method is proposed in the paper, which provides optimised operational parameters (e.g. tail void grouting pressure and face support pressure) during the advancement of tunnel boring machines below vulnerable buildings, such that the risk of damage for existing buildings is minimised. For real-time applicability of this method two different types of Artificial Neural Networks in combination with the Proper Orthogonal Decomposition approach are generated as surrogate models of the finite element simulations. The surrogate models are finally linked and implemented into a user-friendly application, which can be used as an assistant tool to adjust the operational parameters of the tunnel boring machine at the construction site.