【作者】 崔国宁；

【导师】 杨理践；

【作者基本信息】 沈阳工业大学 ， 仪器科学与技术， 2022， 硕士

【摘要】 长输油气管道是石油和天然气能源的主要运输方式,该运输方式效率高、成本低,但是由于腐蚀、磨损及各种外力的损伤等原因,导致管道泄漏事故频发,给国家造成了经济损失,给人民带来安全隐患。管道漏磁内检测技术(Magnetic Flux Leakage)是长输油气管道的主要检测方法之一,通过管道漏磁内检测器检测出管道缺陷处的漏磁信号,从而分析识别出管道缺陷的损伤程度。目前,待检测的管道多,且管线长,导致需要识别的管道缺陷漏磁信号的数量较多,而目前漏磁信号的识别主要采用人工判读的方式,耗时耗力,容易造成误检、漏检,人工判读的方式显然已无法满足工程需求,因此亟需开展智能化缺陷识别方法提高缺陷的识别效率。深度学习能够对输入数据的特征学习,在分类、量化识别等领域均有广泛应用。本文以管道缺陷处的漏磁信号数据为研究对象,利用深度学习的基本框架对管道缺陷处的漏磁信号进行智能识别,识别出缺陷尺寸。针对长输油气管道缺陷尺寸的智能识别、有效评估管道损伤程度的问题,提出了一种基于深度学习的缺陷漏磁数据智能识别处理方法。该方法采用深层卷积神经网络,将管道缺陷处漏磁检测信号的结构化数据作为模型量化分析输入源,并对缺陷样本数据做归一化处理,可有效减小检测干扰的影响。卷积神经网络模型主要包含4个卷积层,4个池化层,3个全连接层和1个输出层,利用卷积核提取管道缺陷处的漏磁检测数据特征,改进输出层的输出方式,输出层的激活函数区别于用作分类识别的Softmax分类器,线性输出网络模型识别结果,实现对管道缺陷尺寸的智能识别。实验结果表明,提出的方法对管道缺陷长度和深度具有良好的量化能力,对缺陷长度的量化误差为2-7mm,对缺陷深度的量化误差为1-7mm,满足工程量化需求,同时该方法可快速对工程化数据进行批量识别,在管道漏磁内检测数据处理领域具有很好的应用前景。更多还原

【Abstract】 Long-distance oil and gas pipelines are the main mode of transportation of oil and natural gas energy.This mode of transportation has high efficiency and low cost.However,due to corrosion,wear and damage from various external forces,pipeline leakage accidents occur frequently,causing economic losses to the country,bringing security risks to the people.Pipeline Magnetic Flux Leakage(MFL)is one of the main detection methods for long-distance oil and gas pipelines.The MFL in-pipe detector detects the MFL signal at the pipeline defect,so as to analyze and identify the damage degree of the pipeline defect.At present,there are many pipelines to be inspected,and the pipelines are long,resulting in a large number of pipeline defect MFL signals that need to be identified.At present,the identification of MFL signals mainly adopts manual interpretation,which is time-consuming and labor-intensive and easy to cause misdetection and missing detection.Obviously,manual interpretation can no longer meet the engineering needs,so it is urgent to develop an intelligent defect identification method to improve the efficiency of defect identification.Deep learning can learn the characteristics of input data,and it is widely used in classification,quantitative recognition and other fields.This paper takes the MFL signal data at pipeline defects as the research object,and uses the basic framework of deep learning to intelligently identify the MFL signals at pipeline defects and identify the size of the defect.Aiming at the problem of intelligently identifying the defect size of long-distance oil and gas pipelines and effectively evaluating the degree of damage to the pipeline,a deep learning-based intelligent identification and processing method of defect magnetic flux leakage data is proposed.This method uses a deep convolutional neural network,takes the structured data of the magnetic flux leakage detection signal at the pipeline defect as the input source for quantitative analysis of the model,and normalizes the defect sample data,which can effectively reduce the influence of detection interference.The convolutional neural network model mainly includes 4 convolution layers,4pooling layers,3 fully connected layers and 1 output layer.The convolution kernel is used to extract the magnetic flux leakage detection data features at the defects of the pipeline,and the output of the output layer is improved.In this way,the activation function of the output layer is different from the Softmax classifier used for classification and recognition,and the recognition results of the network model are linearly output to realize intelligent recognition of the size of pipeline defects.The experimental results show that the proposed method has a good ability to quantify the length and depth of pipeline defects,the quantification error of defect length is 2-7mm,and the quantization error of defect depth is 1-7mm,which can meet the needs of engineering quantification.The rapid batch identification of engineering data has a good application prospect in the field of pipeline magnetic flux leakage detection data processing.更多还原