Research On Pipeline Defect Identification Method Based On Optimized Gaussian Mixture Model

Oil and gas are the lifeblood of China's national economy and occupy an important strategic position.With the development of the petroleum and petroch EMical industry,pipeline transportation has been widely used in the oil and gas industry and has great development prospects.However,due to the influence of various factors,pipelines will have defects,leaks,and even cause fires and explosions,which will lead to environmental pollution and casualties,causing national economic losses and waste of resources.Therefore,the inspection and evaluation of pipelines and the classification of pipeline defects are of great significance for predicting pipeline leakage and ensuring the safe operation of oil and gas pipelines.In this thesis,the PMS inspection system developed by the laboratory is used to detect the defects of the pipeline,and the pipeline defect detection experiment is designed to extract various types of pipeline defect signals and complete the preliminary processing of the signals.The optimized Gaussian mixture model is used to achieve the pipeline defects Signal identification.The main research work is as follows:(1)Two sets of pipeline defect detection experiments are designed.The first set of experiments detects pipeline defects under simulated open-air,underwater,mud and buried environments,and analyzes the collected signals.The second set of experiments manually produced 17 × 3 × 0.5mm,20 × 3 × 1mm groove defects,4.2mm diameter and5.0mm diameter circular defects on the pipeline,and collected and processed four types of defects and normal pipeline signals,Constitute a data set for identification work.(2)According to the characteristics of the data set and the advantages and disadvantages of the Gaussian mixture model and its valuation algorithm,the crossvalidation method is used to divide the training data and the test data to a ratio of 9: 1,and the grid search is used to select the relatively optimal model parameters.The cluster mapping method is used to solve the problem that the model lacks generalization ability to the data.The optimized Gaussian mixture model is used to identify the data.When the recognition category is five categories,the accuracy rate can reach 83.2%,which is 55.9%higher than the ordinary Gaussian mixture model;when the recognition category is two categories and three categories,the accuracy rate It can reach 95.8% and 85.8%respectively.(3)In view of the situation that some information in the signal does not conform to the Gaussian distribution,the model distillation method is adopted to set the weight of the combination of the optimized Gaussian mixture model,the support vector machine model,the Gaussian process classification model and the convolutional neural network model to 5: 2: 2: 2.The recognition type is set to five categories,and finally the overall accuracy rate reaches 96.6%.