AN INTELLIGENT PIPELINE DEFECT DETECTION SYSTEM USING AN IN-PIPE ROBOT

Abstract

This paper presents an integrated intelligent in-pipe inspection system combining the SIPIR mobile robotic platform, a sensor-computing unit, computer vision methods, and spatial mapping of the inner pipe surface. The system is intended for automated detection of corrosion, anomalies, and weld defects under limited illumination and restricted access. The robot is equipped with an endoscopic camera, an Intel RealSense L515 RGB-D camera with an infrared channel, an inertial measurement unit, and an onboard computing module. A hybrid processing pipeline combines image preprocessing, edge extraction and clustering of anomalous regions with lightweight neural models suitable for onboard inference. Experimental validation was performed on a test bench containing real pipe segments. A dataset of 4,212 images obtained from 53 pipes was formed. The prototype has a mass of approximately 4.8 kg, a traction force of approximately 93 N, and operates in pipes with diameters of 350–450 mm. The experiments confirmed weld traversal, onboard recognition and early anomaly detection, as well as three-dimensional mapping using RTAB-Map. The proposed approach transforms inspection data into an engineering-oriented representation that specifies what was detected, where it is located, and how maintenance should be prioritized.