Risk-Calibrated Sensor Fusion for Underground Utility Mapping: A Simulation-Based Decision-Support Framework for Excavation Safety

Abstract

A simulation study. No field data are used; all parameters are assumptions. Prepared for submission to Tunnelling and Underground Space Technology (Elsevier). A decision layer above ground-penetrating radar (250 and 600 MHz), an electromagnetic locator and a known anchor carries the probability that a utility lies on the tracked alignment along a corridor of 5 m segments with a Bayesian filter, and reports each segment as present, absent or unverified; unverified marks soil, utility and depth combinations whose expected detection probability is below 0.3, so that sensor silence is never reported as clearance. A seeded Monte Carlo study compares fifteen baselines and ablations, stress cases (lateral shifts, an uncertain anchor, sensor failure, correlated evidence), probability calibration and an excavation-clearance trial, and reports detection, calibration, location and depth accuracy and the share left unverified separately, with explicit denominators. Code and results: https://github.com/sandlerleon/subsurface-risk-calibrated-mapping, archived at 10.5281/zenodo.23306881.