Classification of damage types in liquid-filled buried pipes based on deep learning

Abstract

In long-distance pipelines, this type of local damage can lead to different forms of damage. Ultrasound (UT)-guided wave technology can detect channel damage at a distance and reduce the workforce and material resources. Deep learning has the advantages of high efficiency and accuracy for pipeline damage classification and identification. This study proposes a classification method that combines UT-guided waves with deep residual neural networks. First, the time-series data of the defect echoes are encoded into different types of images using the glare angular field matrix. Then, the features of the generated images are extracted using ResNet. Finally, it is put into Faster-RCNN for training, validation and defect type recognition. Finite element models containing cracked, square and circular defects were built to verify the effectiveness of the method. The network models were trained for classification, testing and validation using pipes with broken defects. Finite element analysis results show that the network model classifies cracked, square and circular defects with different damage levels with accuracy, recall and F1-score indices close to 90%, and the experimental results show that the network model has an identification accuracy of approximately 90%. Furthermore, the results show that the Faster-RCNN-ResNet model is more accurate in identifying complex pipeline defect types than machine learning and other deep learning methods. The model shows good feasibility and effectiveness in classifying the damage types of long-transmission buried pipelines.