A fuzzy surrogate modelling approach for real-time predictions in mechanised tunnelling. (2018)

• Record Type: Journal Article
• Authors: Cao, Ba Trung; Freitag, Steffen; Meschke, Günther
• Is Part Of: International journal of reliability and safety. Volume 12:Number 1/2(2018)
• Page Start: 187
• Page End: 217
• Publication Date: 2018
• Subjects: surrogate models -- proper orthogonal decomposition -- artificial neural networks -- uncertainty -- reliability analysis -- mechanised tunnelling -- real-time predictions -- fuzzy data; Reliability (Engineering) -- Periodicals; Industrial safety -- Periodicals; 620.00452
• Languages: English
• ISSNs: 1479-389X

Abstract

In mechanised tunnelling, it is important to perform reliability analyses with respect to the tunnel face collapse and the damage risks of the tunnel lining and existing structures on the ground surface due to the tunnelling induced settlements. The reliability assessment requires to deal with limited information describing the local geology and the soil parameters due to the availability of only a small number of borehole data. In this paper, it is focused on real-time reliability analyses in mechanised tunnelling considering different types of uncertain data, i.e. combining epistemic and aleatoric sources of uncertainty within polymorphic uncertainty models. The system output of interest in these analyses is time variant tunnelling induced surface settlement fields, which are computed by a finite element simulation model. However, for real-time predictions with uncertain data, efficient and reliable surrogate models are required. A new surrogate modelling strategy is developed to predict time variant high dimensional fuzzy settlement fields in real-time. The predicted results of the new surrogate model show similar accuracy compared to the results obtained by optimisation based fuzzy analyses. Meanwhile, the computation time is significantly reduced especially in case of high dimensional outputs and in combination with the p-box approach in the case of polymorphic uncertain data.