Defect identification for oil and gas pipeline safety based on autonomous deep learning network

The safety detection for oil and gas pipelines is more and more worthy of attention. It not only promotes the development of pipeline safety work, but also provides a guarantee for pipeline safety decision management. However, there are more and more safety problems in pipeline operation, causing immeasurable consequences. Therefore, the pipeline safety detection technology needs to be further improved. In this paper, a two-axis magnetic flux leakage detection device is used for safety detection of an oil and gas pipeline, and the detection results are analyzed and studied. 77 sets of detection data are collected through the detection device. Due to the harsh environment of the oil and gas station, the data is severely disturbed, so the data is filtered firstly. The filtered data can better reflect the safety status information of the pipeline. Secondly, In order to avoid the random error of single-axis data, a two-dimensional data fusion method is proposed. The fusion data improves the accuracy of recognition of pipeline failure features. Thirdly, autonomous deep learning recognition algorithm is used to classify and recognize pipeline failure features. The network in this algorithm includes convolutional layers, pooling layers and fully connected layers. Through multiple simulation calculations, the number of network layers has been optimized. Finally, experiments are carried out based on the data collected on-site. The experiment results show that the training accuracy is 99.19%, and the testing accuracy is 97.38%. In short, the entire pipeline safety inspection data processing algorithm reliably identifies the types of pipeline failure defects. And it will provide a basis for the safe construction of pipelines.