Reconfigurable robot and deep learning-based system for sewer pipeline inspection

The inspection of underground sewer pipelines presents significant challenges due to harsh environmental conditions and structural complexities. Conventional methods such as manual evaluations and CCTV-based assessments are often inefficient and lack adaptability to varied pipeline structures, while existing inspection robots suffer from limited manoeuvrability, power efficiency, and defect detection accuracy, especially in unstructured environments with sediment and water. To address these issues, this study proposes a reconfigurable pipeline robot with an adaptive locomotion mechanism designed for enhanced navigation and stability in complex sewer settings. The robot features a compact structure, underwater sealing, and a high-load transmission system, supported by a dynamic posture adjustment mechanism using yaw and pitch control to improve mobility across inclined and sediment-laden surfaces. Complementing the hardware, an enhanced deep learning model based on the IYOLOv8 architecture is introduced, incorporating RepGFPN and DynamicHead modules to improve multi-scale feature extraction and detection precision. Experimental results on a real-world sewer defect dataset show that the proposed system achieves a mean Average Precision (mAP) of 90.9%, with a detection speed of 61.5 FPS and robustness across various noisy conditions, outperforming the original YOLOv8 and other state-of-the-art models. These results confirm the system's effectiveness and reliability in sewer pipeline defect detection, providing a scalable solution for intelligent underground infrastructure monitoring.