Citation

Wan Y, Wang Y, Yang Y, Liu C, Dai Y. Harbin Gongcheng Daxue Xuebao 2021; 42(9): 1321-1329.

Copyright

(Copyright © 2021, Gai da xue)

DOI

10.11990/jheu.202005049

PMID

unavailable

Abstract

Ensuring the normal operation of pipelines and preventing pipeline defects before they occur are important tasks in oilfields. Metal magnetic memory detection technology may be especially helpful in these endeavors. In this paper, a fluxgate magnetometer is first used to collect pipeline magnetic flux leakage signals. Then, various characteristic quantities, including magnetic induction peak, maximum value, minimum value, average value, energy, area, maximum gradient, average gradient, and wavelet packet energy, are calculated, and the most representative properties are selected. Finally, three methods, namely, the collaborative representation classification method, the traditional support vector machine method, and the improved support vector machine method, are employed to establish several pipeline defect classification models. The defect recognition rate of the optimal model could reach 99.513 0%, with the model training and recognition time being only 3.55 s.

RESULTS show that the optimal model can effectively identify pipeline corrosion and defects, as well as bend and weld stress concentration defects. This study may be applied to actual oilfield pipeline classification and provides a reliable reference for future research on defect classification. © 2021, Editorial Department of Journal of HEU.

Keywords: Pipeline transportation

Language: zh

Keywords

Defects; Pipeline corrosion; Pipelines; Magnetic leakage; Magnetic storage; Magnetometers; Support vector machines; Welds