Research On Intelligent Diagnosis Method Of Oil And Gas Pipeline Defects Under Coating Layers

Oil and gas pipelines have the important task of transporting energy sources such as crude oil,refined oil and natural gas.Pipeline defects pose a serious threat to the safe operation of pipelines.To solve the difficulties in the practical application of coating pipeline defects diagnosis,this dissertation conducts in-depth research on signal noise reduction,feature extraction,feature selection and defect diagnosis.The main researches are as follows.1.Due to the problem that the traditional noise reduction method has poor noise reduction effect for non-stationary and noisy field detection signals,this dissertation proposes a DEMMD-MP noise reduction method based on Differential-based Empirical Mode Decomposition(DEMD)and Extremum Field Mean Mode Decomposition(EMMD).According to Monte Carlo simulation and Pareto optimal solution,the optimal IMF component combination is selected to reconstruct the noise reduction signal with the lowest noise content and the highest feature retention.Compared with Empirical Modes Decomposition(EMD)noise reduction methods,DEMMD-MP method resulted in a 20.9%increase in Signal to Noise Ratio(SNR)and a 53.1% decrease in Root Mean Squared Error(RMSE).The noise reduction effect is optimal.2.To address the limitations of traditional feature extraction methods in extracting signal feature information,this dissertation analyzes the oil and gas pipeline defects under coating layers by image visualization method.The Gray Level Cooccurrence Matrix(GLCM)and Gray Gradient Cooccurence Matrix(GGCM)methods were used to extract the texture features of gray image,and Fisher criterion and k-means clustering analysis were used to analyze the features.Continue Wavelet Transform(CWT)is used to convert the signal into a time-frequency scalogram,and a Convolutional Neural Network(CNN)model is established to identify pipeline defects.The experimental results show that the gray image texture feature and time-frequency image feature can realize the classification of different oil and gas pipeline defect signals,which can provide reliable input for subsequent defect recognition.3.A hybrid feature selection algorithm IGA-BSVM based on GA is proposed to address the problem that the unbalanced characteristics of the detection dataset makes traditional feature extraction methods difficult to apply.The selection strategy is optimized to ensure the global optimum,the fitness function is improved,and the characteristics conducive to defect recognition are extracted.The experimental results on unbalanced dataset show that this method can achieve higher classification accuracy with smaller feature subset dimensionality,a feature subset dimensionality reduction of 44.5%,and a recognition rate of 86.36% for pit defects and 87.72% for scratch defects.The highest defect recognition rate is achieved while reducing the dimension of feature subset.4.A semi-supervised neural network model based on SAE-AMSoftmax is constructed to make full use of the feature information contained in unlabelled samples and to improve diagnostic accuracy.Using SAE unsupervised training method and AMSoftmax supervised fine tuning method,the low-level and deep-level features extracted by SAE in different depths are integrated to comprehensively predict the defect types.Compared to SVM and DNN,the model improves the recognition rate of pit defects by 4.54% and 3.64%,and scratch defects by 4.55% and 1.82%,respectively,which proves the superiority of the model.5.Considering the influence of detection samples under different test scenes on defect diagnosis model,this dissertation proposes a weld defect group diagnosis method for oil and gas equipment based on lazy learning and CNN model.The test scene matrix is constructed,and the training samples of the diagnosis model are optimized.Compared with the traditional model,the model improves the recognition accuracy of defects by10.0 %,which has better defect diagnosis effect.The model is applied to the group diagnosis of weld defects in oil and gas equipment under coating layers to achieve endto-end intelligent diagnosis,and good application results are achieved.