A Method for Generating and Evaluating Magnetic Flux Leakage Data for Large-diameter Pipeline Crack Defects Based on Numerical Simulation and Diffusion Algorithm
Considering that the sample size of magnetic flux leakage （MFL） data for large-diameter pipeline crack defects is insufficient to support the quantitative identification， a database generation method integrating numerical simulation of MFL field and diffusion algorithm was proposed， and it was verif...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Office of China Petroleum Machinery
2025-07-01
|
| Series: | Shiyou jixie |
| Subjects: | |
| Online Access: | http://www.syjxzz.com.cn/en/#/digest?ArticleID=4975 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Considering that the sample size of magnetic flux leakage （MFL） data for large-diameter pipeline crack defects is insufficient to support the quantitative identification， a database generation method integrating numerical simulation of MFL field and diffusion algorithm was proposed， and it was verified through full-scale pull test and three-dimensional MFL simulation modeling. The results show that the crack defect size has a clear impact on the MFL signal. For the axial component， the peak value is positively correlated with the defect depth， and the trough spacing is positively correlated with the defect length. For the radial component， which shows an antisymmetric distribution， the absolute peak value is positively correlated with the depth， and the spacing between the positive and negative peaks is positively correlated with the length. The accuracy of the three-dimensional MFL simulation model reaches the engineering requirement （average error <5%）， and 240 groups of crack defect simulation databases were constructed. The data generated from the diffusion model have the same distribution as the simulation data. After optimizing the model based on the pull test data with environmental noise， 75.4% of the generated data pass the KNN test （K=5）， and have the feature distribution highly similar to the real data. The research conclusions provide a basis of MFL data for the quantitative evaluation of pipeline crack defects. |
|---|---|
| ISSN: | 1001-4578 |