Abstract: Detecting problems in tunnel lining structures is often hampered by challenges such as complex equipment setup, low detection efficiency, and unrepresentative results. To address these issues, a novel tunnel-lining radar detection robot has been designed and developed. This robot features a four-rotor and four-drive configuration and is equipped with a 900 MHz wireless radar, and it can operate at speeds of up to 18 km/h. The robot traverses the circumferential lining surface using precisely calibrated mechanical indices and a custom gravity sensor to verify its path. The radar robot collects structural damage data, which is analyzed using a lightweight radar signal network model. This model automatically classifies common abnormalities, such as reinforcing bars, insufficient thickness, noncompactness, and cavities. The proposed method generates a signal time window that reflects the top and bottom boundaries of the detected cavities, thereby enabling precise positioning and depth measurements. Practical engineering testing and application are conducted in the Xiaolijian and Lushan tunnels, with subsequent on-site drilling verification. The application and verification results show that: (1) The radar robot can quickly and stably complete the automatic collection of structural data, significantly improving the accuracy compared with manual methods. The detection time for a 4 000 m² lining is shortened from 3 to 0.5 h. (2) The F₁ score and mAP@0.5 of the radar signal lightweight network classification module are 0.93 and 97.51%, respectively, outperforming traditional radar map models and requiring less computational effort. (3) The robot′ scavity depth predictions have an accuracy exceeding 80%, with the output position and depth information aligning closely with the field drilling results, validating the effectiveness of the radar robot for automatic damage detection.

Key words: tunnel lining, structural disease, radar robot, automatic collection, intelligent detection