Design and Experimental Validation of Pipeline Defect Detection in Low-Illumination Environments Based on Bionic...

Area of Science:

• Robotics and Automation
• Computer Vision
• Non-Destructive Testing

Background:

• Internal pipeline defect detection is challenging due to poor lighting and complex geometries.
• Existing methods struggle with reliable perception and collision-free navigation in narrow, curved pipelines.

Purpose of the Study:

• To develop an event-aware framework (ES-YOLO) for enhanced pipeline defect detection.
• To create a robotic system for collision-free inspection using a hyper-redundant manipulator and event camera.
• To validate the system's effectiveness in low-light, narrow pipeline environments.

Main Methods:

• Proposed the ES-YOLO framework to convert RGB data into an event dataset (N-neudet) for training.
• Developed a pipeline defect inspection system integrating ES-YOLO with a hyper-redundant manipulator controlled by NMPC and SCA.
• Conducted comparative experiments on steel and acrylic pipelines under varying illumination.

Main Results:

• The event-based ES-YOLO model significantly outperformed RGB models in low-light defect recognition.
• The developed system achieved 84% detection accuracy for steel pipeline defects in a 2 Lux environment.
• Demonstrated collision-free inspection capabilities within pipelines using the hyper-redundant manipulator.

Conclusions:

• The event-aware ES-YOLO framework and robotic system offer a robust solution for internal pipeline defect inspection, especially in challenging low-light conditions.
• The proposed approach significantly improves defect detection rates and enables safe, autonomous navigation within pipelines.
• This technology holds promise for enhancing the safety and efficiency of infrastructure maintenance.