Research On The Application Of VMD And DBN In Pipeline Defect Diagnosis

With the continuous development of China's economy,it is increasingly prominent that the importance of oil and gas in the development of national economy.At present,Pipeline transportation is the most important mode of oil and gas transportation.However,with the increase of the service life of the pipeline and the influence of many factors,the risk of leakage also increases gradually,so it is very important to detect the defects of the pipe regularly to master the integrity of the pipeline,so as to ensure the long-term safe and stable operation of the pipeline.In this thesis,detection method of pulsed magnetic eddy current is used to detect pipeline defects.The methods of noise reduction,feature extraction and classification of testing signals are studied emphatically.A set of signal processing and classification method suitable for the detection signals of pipeline pulse magnetic eddy current is proposed,and a set of software is compiled for the detection signal processing based on the method.The main contents of this thesis are as follows:(1)Propose SSE-VMD noise reduction algorithm based on the signal characteristics of pulsed eddy current detection.Firstly,the algorithm decomposes the signal though variational mode decomposition under different K-values,then reconstructs the signal,and uses the maximum value of the square error difference between the two adjacent IMF components to determine the best K-value for variational mode decomposition of the signal.Finally uses the average value of the correlation coefficient between each IMF component and the original signal as the threshold to screen the effective IMF component to reconstruct the original signal to complete the signal noise reduction.It is applied in the simulation and measured approximate noiseless signal with 5d B Gaussian white noise,the SNR of the signal was increased to28.5802 d B and 11.4074 d B.The test results show that the accuracy in classification increases from 90.20% to 92.40% after de-noising by SSE-VMD algorithm.(2)Establish a 900×41 pipeline pulse magnetic eddy current detection database and propose DBN-RELM feature extraction and classification algorithm to be used in pipeline pulse magnetic eddy current detection signal.The algorithm uses a regularized extreme learning machine to replace the reverse fine tuning and classify in the deep belief networks to reduce the model training time.Experiments show that,compared with the traditional deep belief networks algorithm.The algorithm can shorten the training time by about 45% on the basis of slightly reducing the accuracy in the classification.The algorithm's average classification accuracy of the pipeline pulse magnetic eddy current detection signal after noise reduction is 92.40%,which can meet the engineering needs.(3)Using the signal processing method proposed in this thesis,a set of visual signal processing software is designed.The software can reduce noise,extract feature and classify the detected signal.