A Novel Cascaded Deep Learning Model for the Detection and Quantification of Defects in Pipelines via Magnetic Flux Leakage Signals

Yuksel V., Tetik Y. E., Basturk M. O., Recepoglu O., Gokce K., Cimen M. A.

IEEE Transactions on Instrumentation and Measurement, vol.72, 2023 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 72
  • Publication Date: 2023
  • Doi Number: 10.1109/tim.2023.3272377
  • Journal Name: IEEE Transactions on Instrumentation and Measurement
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Business Source Elite, Business Source Premier, Communication Abstracts, Compendex, Computer & Applied Sciences, INSPEC, Metadex, Civil Engineering Abstracts
  • Keywords: Convolutional neural networks (CNNs), deep learning, in-line-inspection (ILI), magnetic-flux-leakage (MFL), nondestructive testing (NDT), Swin-Tiny (Swin-T)
  • Istanbul Technical University Affiliated: Yes


In this article, we present a machine learning-based quantitative method for the interpretation of signals gathered from nondestructive testing (NDT) of steel pipelines via a semi-autonomous in-line-inspection (ILI) robot. The robot has a magnetic-flux-leakage (MFL) sensor that produces three axis data for each point of pipeline with specific intervals. Both the robot and the MFL sensor have been developed in-house. The signals collected via MFL sensor are converted into images to be used as an input for the proposed defect detection model. We propose a combination of a defect detection model based on Swin Transformer Backbone YOLOv5 (SwinYv5) object detection algorithm and a quantification model based on cross-residual convolutional neural network (CR-CNN). The detected defect locations are used to extract the region of interest (ROI) images of defects that are used as an input for the quantification model. In data collection phase, numerous tests have been conducted via a special test mechanism, and a custom data augmentation technique has been deployed in order to increase the amount and variety of training data. According to test results, the proposed method is capable of detecting defects with a precision of 98.9% and quantifying them with maximum errors of 1.30, 1.65, and 0.47 mm for length, width, and depth, respectively.