Identifiability: construction error as apparent soil stiffness in back-analysis

Abstract

Geotechnical back-analysis usually estimates ground parameters while treating the construction process as known. Because ground and construction jointly generate the stress path, an error in the assumed construction state can be absorbed by the calibrated soil8
parameters without degrading the fit. This letter writes the inverse problem as y = ℳ(ℱ(θ, c) + δ) + ε, separating ground parameters θ from the mechanically effective construction state c, and distinguishes confounding, when c is estimated jointly with θ, from information loss, when an uncertain recorded value is fixed. Both are quantified with a verified semi-analytical shield-tunnelling model applied to the paired TULIP sections. With section specific grout pressure records, one site investigation stiffness reproduces both measured settlement peaks; replacing the records by a common nominal pressure creates an apparent stiffness contrast of about 40% with equally good fits. Because operational grout records do not directly measure the effective boundary pressure, settlement data cannot decide between these interpretations, and additional settlement points do not help. Lining deformation, which responds oppositely to grout pressure, would restore most of the lost information. Goodness of fit alone is therefore not evidence that calibrated soil parameters are correct.