Reliability and Efficiency of Metamodel for Numerical Back Analysis of Tunnel Excavation

During tunnel construction, the ground properties, initially evaluated, are continuously assessed and verified through back analysis. This procedure generally requires many numerical analyses, so a metamodel based on artificial neural networks has been used to reduce number of analyses. More datasets can be create more reliable metamodels. However, there no established rules regarding optimum for metamodel. Metamodels predicting vertical displacement crown using five parameters (unit weight (?), uniaxial compressive strength (UCS), material constant mi, geological index (GSI), coefficient lateral pressure (K)), with 3, 4, 6, 8, 10 values per property, were created confirm reliability in this study. 6 8 each property showed 5% 1% mean absolute percent errors, respectively. These numbers properties would appropriate developing Among parameters, only results global sensitivity analyses GSI K higher than 0.9. According these results, it is that assessments sufficient