Geotechnical Parameter Identification Beyond Force Reconstruction
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 657-660
DOI: 10.1609/aaaiss.v10i1.43470 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Honglai Peng und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Inferring geotechnical properties from operational measurements collected during lubricated microtunneling is inherently ill-posed. Measured jacking-force responses entangle face resistance, native pipe–soil friction, and latent lubrication effects, allowing different parameter combinations to produce similar observations. Resolving this ill-posedness is further constrained by parsimonious field data. Although Bayesian inverse methods can mitigate the data-parsimony challenge by incorporating scientific constraints and quantifying uncertainty, repeatedly recomputing the posterior for each new observation window imposes a computational-parsimony challenge. Thus, we propose a stage-wise physics-informed multi-head neural network (MHNN) that encodes the microtunneling force-transfer hierarchy in both the model architecture and training procedure to mitigate practical non-uniqueness and enable amortized inference of geotechnical properties. The joint inverse mapping is factorized into a physics-ordered sequence that estimates the empirical face-resistance factor, latent lubrication force, residual interface friction coefficient, residual friction angle, and total soil unit weight. Bounded residual heads further stabilize the weakly identifiable friction-angle and unit-weight estimates by constraining deviations from reference values. Physics-based consistency losses regularize the staged inference by enforcing agreement between the estimated quantities, decomposed force components, and reconstructed jacking-force responses. The effectiveness of the proposed framework is demonstrated using a domain-randomized, physics-consistent synthetic dataset and a documented microtunneling project, showing that the staged formulation improves geotechnical-property identification beyond force reconstruction alone.
Zitieren
APA
Peng, H., Borela, R., Staheli, K., & Frost, J. D. (2026). Geotechnical Parameter Identification Beyond Force Reconstruction. Proceedings of the AAAI Symposium Series, 10(1), 657-660. https://doi.org/10.1609/aaaiss.v10i1.43470
Vancouver
Peng H, Borela R, Staheli K, Frost JD. Geotechnical Parameter Identification Beyond Force Reconstruction. Proceedings of the AAAI Symposium Series. 2026;10(1):657-660. doi:10.1609/aaaiss.v10i1.43470
BibTeX
@article{Peng2026,
author = {Peng, Honglai and Borela, Rodrigo and Staheli, Kimberlie and Frost, J. David},
title = {Geotechnical Parameter Identification Beyond Force Reconstruction},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {657--660},
doi = {10.1609/aaaiss.v10i1.43470}
}
EUVIMED