Physics-informed theory modelling of rock engineering features from TBM operational data

Summary

Abstract

Accurate determination of rock engineering parameters is important for reliable prediction and optimization of tunnel boring machine (TBM) performance. In this study, TBM performance indicators and operational parameters are used to infer key geomechanical properties, including both intact rock and rock mass characteristics, across several lithological units (e.g. Paragneiss, Amphibolite, and calc- phyllite-schist with low to high grade of metamorphism) encountered along the Brenner Base Tunnel (BBT) alignment that connects the cities of Innsbruck (Austria) and Fortezza (Italy). The dataset, derived from both pre-construction investigations and construction-phase monitoring, is divided into geological zones and lithological units, with representative ranges of considering operational parameters and rock mechanical features defined for each zone. A framework based on physics-informed theory, engineering mechanics, and stacking ensemble learning model are developed to establish simulation between machine response, mechanical specifications, and ground conditions. The meta-learner stacking logic is implemented using the ridge cross-validation (RidgeCV) and the ensemble is built on three high-performance tree-based algorithms, including extra trees regressor (ET), hist gradient boosting regressor (HGB), and gradient boosting regressor (GB) based on the polynomial interaction technique. They were selected because they handle geological noise and non-linear relationships differently, therefore, providing a balanced prediction of rock mechanical features. The proposed modeling approach shows predictive accuracy, yielding reliable estimates of rock engineering properties and supporting its applicability to future double shield (DS)-TBM operations in metamorphic rock units.