Research On Simulation And Classification Of Pipeline Damage Defects Based On Magnetic Flux Leakage Detection

As the pipeline is about to face or has entered the extended service state,the defects on the surface of the pipeline are becoming more and more complicated,and the possibility of pipeline leakage is increasing.Thus,the type of pipeline defects and the situation of development of defects become an effective criterion for evaluating whether the pipeline can continue to serve.At present,the MFL is one of the most mature methods for the detection of underground pipeline defects.The magnetic flux leakage detection signal has a good correspondence with the defect size.Moreover,the magnetic flux leakage detection signal is used to classify the pipeline defects.And the identification can provide a good evaluation standard for pipelines that are about to face or have entered the extended service pipeline,and it can also provide a reliable basis for the prediction of the remaining life of the pipeline.Therefore,how to use the magnetic flux leakage signal to classify and identify pipeline defects has become the focus of research.First,This paper summarizes the defects of underground pipelines and quantifies the defects of different types of pipelines.Then,based on magnetic flux leakage detection method,Maxwell is used to establish the magnetic flux leakage simulation model,and 1000 sets of simulation experiments are carried out to obtain sample data.The magnetic flux leakage signal of the defect is extracted and the signal characteristics are extracted.Finally,the three kinds of machine learning algorithms,SVM,random forest and GBDT,are used to classify and identify the feature quantity of the defect signal.And,comparing the feasibility and recognition accuracy of the three methods,in the existing data range,the accuracy of the classification identification of GBDT algorithm has reached 100%.It is an ideal classification learning algorithm for pipeline defect detection.It lays a good foundation for the classification and identification of pipeline defects by using machine learning theory.