Lamb wave-based MVDR imaging and CNN classification of defects in pipelines

Significant progress has been made in ultrasonic guided wave (UGW) technology for pipe signal processing and defect imaging recently. However, developing a defect localization and imaging algorithm that requires fewer parameters, offers a wide imaging range, and achieves high positioning accuracy remains a considerable challenge. Traditional direction-of-arrival (DOA) estimation methods primarily focus on the single-angle estimation with low resolution, failing to satisfy the spatial localization requirements for pipeline defects. Therefore, a high-resolution spatial spectrum estimation algorithm is introduced to realize the two-dimensional DOA estimation. By distributing multiple sensors in a specific geometric configuration to form an array, this method employs the array signal processing technology to accurately obtain the DOA of spatial signals. A uniform circular array (UCA) is employed in the present study to receive signals from pipe defects, and high-precision localization and imaging of defects are achieved based on the two-dimensional minimum variance distortionless response (MVDR) beamforming algorithm, with a relative positioning error of less than 1%. An image classification method based on the convolutional neural network (CNN) is further developed to distinguish the defect types. By constructing a novel CNN model to extract defect features and perform classification, this model achieves a prediction accuracy of 97.50%, which effectively distinguishes between defect types.