Speaker
Description
Reliable tunnel magnet alignment demands a systematic precheck strategy to decide whether a candidate single-station setup is sufficient, requires additional key points, or should be adjusted before full observation. This work introduces a station precheck and decision-support prototype that integrates uncertainty-aware metrology with structured scene reasoning.
The uncertainty model combines transport residuals, single-station pose recovery, template-center priors, and empirical scene-level risk. These quantities are transformed into structured feature tables and action rules. A feature-level station precheck is then performed using standardized indicators of range, height, elevation span, and geometry conditioning. Partial-precheck simulations demonstrate that adding Level-2 key points significantly improves the identification of unfavorable station configurations compared with using basic station geometry alone. Accordingly, key-point selection is formulated as a planned geometric coverage problem rather than a random choice, establishing a clear selection strategy.
The framework also incorporates an LLM-based explanation module. Critically, the LLM is not used for numerical computation; it is constrained by verifiers and only interprets the structured precheck outputs and recommended actions, providing field-oriented explanations.
The prototype establishes a practical workflow from uncertainty-aware metrology results to measurement decisions. Future work will incorporate repeated transport datasets and tunnel field experiments to validate the framework under realistic measurement conditions.
