Publications

Active learning for accurate settlement prediction using numerical simulations in mechanized tunneling

Saadallah, A. and Egorov, A. and Cao, B.-T. and Freitag, S. and Morik, K. and Meschke, G.

PROCEDIA CIRP
Volume: 81 Pages: 1052-1058
DOI: 10.1016/j.procir.2019.03.250
Published: 2019

Abstract
Finite Element simulation is a possible tool to investigate interactions between the Tunnel Boring Machine and the surrounding soil. Surface settlements can be predicted in real-time based on simulation results by machine learning surrogate models. However, to train such models, large amounts of computationally intensive simulations are required. To accomplish this step with minimal costs, we propose a hybrid active learning approach to select the minimal amount of simulations necessary to build an accurate model. During the tunnel construction, the real-time settlements prediction model will be used to analyze associated risks to ensure safe and sustainable constructions in urban areas. © 2019 The Authors. Published by Elsevier Ltd.

« back