Answers are grounded in this paper and cite their sources.
Localized surcharge identification above shallow buried structures depends on soil-arching state and the reliability of sparse monitoring channels. We examine an active trapdoor at H / B = 1.5 using three pre-displacement states and five observation sets combining surface settlements and support-reaction features. A 117-response plane-strain Mohr–Coulomb finite-element database was divided into training, development, independent numerical evaluation, and independent numerical confirmation sets. State-specific quadratic-trend Matérn-5/2 Gaussian-process surrogates were trained on the training set, while channel-wise discrepancy was calibrated on the development set and applied to the independent sets. Training-set leave-one-out cross-validation underestimated the out-of-sample discrepancy of the signed reaction moment. In the independent evaluation set, its normalized forward root-mean-square error was 8.56 and its mean contribution to the inverse objective at the true parameters was 62.7%. Development calibration reduced these values to 1.30 and 11.8%, respectively, eliminating the associated opposite-boundary solutions. With the calibrated scale, median combined errors across the five observation sets ranged from 0.302 to 0.359. The reaction-distribution scale improved width recovery in some states but provided no general combined-error advantage. In the confirmation set, median combined errors ranged from 0.493 to 0.511; the same feature reduced median width error from 47.35% to 35.14%, although 16 of 24 estimates still reached a parameter boundary. Observation design should therefore account jointly for arching state, parameter-specific information, and surrogate discrepancy. Width recovery remains limited in sparsely covered parts of the nominal load domain.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: