Research On Recognition Method Of Weld Defects In Natural Gas Pipeline

Natural gas pipelines are the preferred method for long-distance and large-scale transportation of natural gas.As natural gas has been widely used in industry,transportation,and residents’ lives,the safety of natural gas pipelines has also received extensive attention.In recent years,most of the safety accidents of natural gas pipelines are caused by defects in the pipeline welds.Therefore,efficient and accurate defect identification technology is an important guarantee for the safe and stable operation of natural gas pipelines.This thesis takes the weld defects of natural gas pipelines as the research object,and proposes a Parallel Convolutional Neural Network(P-CNN)based on multi-scale feature compression excitation to identify the weld defects of natural gas pipelines.Firstly,the necessity of the research on the identification of weld defects in natural gas pipelines is introduced.Then,from three aspects of defect characteristics,traditional machine learning,and deep learning,the research status of domestic and foreign gas pipeline weld defect recognition is reviewed.Finally,the core structure and core structure of Convolutional Neural Networks(CNN)are described.In the process of defect recognition,global feature extraction often only pays attention to the integrity of image features,and the obtained feature data information is singular and lacks diversity.A global feature of weld defects based on multi-scale modular convolutional neural network is proposed.method of extraction.Adding a multi-scale module to the traditional convolutional neural network not only retains the integrity of the global features to a certain extent,but also ensures that the extracted feature information is diverse.In order to improve the details of the local features in the defect recognition process,a local feature extraction method of weld defects based on Local Binary Patterns(LBP)and channel-weighted convolutional neural network is proposed.Combining the advantages of the local binary mode and the compressed excitation module of space plus channel,it highlights the detailed information with important calibration and enhances the extraction quality of feature images.In the defect recognition process,the traditional convolutional neural network can not take into account the integrity of the global feature information and the detail of the local feature information at the same time,so a parallel convolutional neural network-based weld defect recognition method is proposed.The weld defect image is divided into two branches:grayscale image and local binary mode.The two convolutional networks of different structures are input in parallel from the global feature and the local feature.The multi-scale space and the channel compression excitation module are used to input Multi-scale feature extraction and detailed feature calibration are performed on the image data to improve the classification accuracy.A parallel convolutional neural network model based on the above method is constructed,and the weld image data collected on site is input into the model for defect recognition.Through a comparative study with traditional methods,the effectiveness and superiority of the method proposed in this thesis are verified,and a higher defect recognition rate is obtained.