Application of robotic systems for inspecting underground utilities and tunnels with automated processing of defectoscopy data and construction of damage maps

Abstract

The article is devoted to the integration of robotic platforms and algorithms for automated data processing in diagnosing underground utilities and tunnels, followed by the formation of georeferenced damage maps. It examines the limitations of traditional visual inspections under conditions of high humidity, dustiness, confined spaces, and lack of satellite positioning, and justifies the need for a multisensor approach due to the diversity of materials and defects (cracks, cavities, corrosion, infiltration, destruction of the protective layer, profile deformations). Based on the systematization of modern solutions and analysis of inspection robot characteristics, the diagnostic capabilities of 4K video recording, laser profilometry, ultrasonic testing, and ground-penetrating radar are compared; the pronounced complementarity of the channels and the advantages of their integration are demonstrated, ensuring consistently high detection performance across all categories (around 90–98%) despite significant variability in individual sensor metrics. The software architecture for processing large data volumes (terabyte scale) is described, including defect segmentation using deep learning methods, as well as a two-level computation scheme: fast onboard real-time detection and more accurate server-side post-processing, selected based on a combination of quality, speed, and resource requirement metrics. For cartography, localization and trajectory construction based on SLAM using lidar, inertial measurements, and odometry are implemented; estimates of positional error, which increases in sections with complex geometry, wet surfaces, and partial flooding, are provided. An automated quantitative assessment of defect parameters with validation against control measurements is presented, demonstrating small deviations and high correlation. Spatial analysis of the maps reveals clustering of damages in zones of joints, junctions, and subsidence, while statistical relationships with age and laying conditions enable the use of a regression model for prioritizing inspections and repairs; the requirements for standardizing robotic defectoscopy results for practical implementation in urban infrastructure management are discussed separately.