Classification Recognition of Long Distance Oil and Gas Pipeline Girth Welds MFL Signal Diagrams Based on VGG16 Network

Abstract

Abstract In recent years, with the development of long-distance oil and gas pipelines, the number of accidents, most of which caused serious consequences, is increasing gradually. According to statistics, pipeline cracking caused by girth weld problems accounts for a large proportion of all. Girth weld defects of long-distance oil and gas pipelines, such as cracks, incomplete fusion and incomplete penetration, seriously threaten the safety of pipelines. Because the internal magnetic flux leakage detection (MFL) signal is complex at the girth weld, it is difficult to judge the defect type through conventional data analysis. In this paper, MFL signal screenshots of girth welds, which are automatically captured from MFL client software by the Python program, are the training sample set of VGG16. This training sample set which includes 1396 samples labeled with radiographic test results after girth welds excavation, is divided into three categories: qualified samples, unqualified samples with circular/strip defects, and unqualified sample with crack/incomplete fusion/incomplete penetration defects. The deep learning network after training is used to predict the classification of 156 test samples, and the prediction accuracy is 78%. This method can effectively judge whether there are hazardous girth weld defects types, which are significant for safety management of long-distance oil and gas pipeline girth welds.