This presentation introduces a novel deep learning framework for enhanced pipeline inspection using acoustic guided-wave signals in temperature-varying environments. Traditional guided-wave techniques for pipeline defect detection face significant challenges including environmental variations, sensitivity limitations, and complex signal interpretation that limit their practical effectiveness. This study addresses these limitations through an innovative dual-path one-dimensional convolutional autoencoder that simultaneously performs defect detection, localization, and temperature prediction.
The proposed framework utilizes multi-mode and broadband acoustic waves with an optimized sensor configuration that balances high accuracy with practical implementation simplicity. Experimental validation conducted on carbon steel pipes demonstrates exceptional performance, achieving precise defect localization with a mean absolute error of only 66 mm and temperature prediction accuracy within 0.2 °C mean absolute error. Comparative analysis reveals superior performance compared to traditional signal processing methods previously developed by the research team.
This work represents a significant advancement in nondestructive evaluation technology, demonstrating how deep learning integration can overcome fundamental limitations of conventional guided-wave signal processing approaches. The results highlight substantial potential for industrial implementation, particularly in oil and gas pipeline monitoring where early defect detection is critical for preventing costly maintenance and operational disruptions.