Research On Classification And Recognition Of Defect Magnetic Flux Leakage Signals

As an irreplaceable storage in petroleum and petrochemical fields,storage tanks play an important role in the development of our country's economy.Some defects will appear in some locations of the tank that results from various factors,which will seriously threaten the safe operation of the storage tank.The handling improperly will cause significant economic losses to the enterprise,resulting in serious environmental pollution and even casualties.Therefore,it is a very important work to detect the defects of the storage tank.In the defect detection method,magnetic flux leakage detection is an effective method in the detection of tank defects,and the signal recognition of classification in the later period is a very important link.This paper will make a detailed research on signal preprocessing,feature extraction,feature selection,classifier selection and design of optimization in the process of recognition of classification of defective magnetic flux leakage signals.The specific research work is as follows:1.consult the literature of the research field at home and abroad and summarize and analyze.As to the classification and recognition of defect magnetic flux leakage signals,a large number of domestic and foreign literatures are consulted,and the research progress and existing problems in the field are summarized.Based on some problems existing in the current research,the research ideas of this paper are put forward.2.The analysis and study of the basic theory.The basic theories related to preprocessing,feature extraction and selection,classifier selection and design of optimization are discussed in the recognition of classification of defective magnetic flux leakage signals.Genetic algorithm is applied to preprocessing and classifier selection and design of optimization.Therefore,the application of genetic algorithm on these two fields and the corresponding basic theories are also analyzed and studied.3.Research on signal preprocessing.Based on the wavelet packet analysis,the defect magnetic flux leakage signal is preprocessed.It includes the optimum selection of wavelet packet basis function,the determination of the optimum decomposition level of wavelet packet decomposition,the design and selection of the threshold criterion.The evaluation index of the three parts are based on the signal to noise ratio and the root mean square error.Finally,the coif2 wavelet packet basis function,the 4 layers of decomposition and the improved genetic algorithm threshold criterion are applied to the preprocessing of magnetic flux leakage signals.4.Study on selection and design of optimization of classifier.Selecting BP neural network as a classifier for recognition of classification of defective magnetic flux leakage signals.Then the weight threshold and the activation function are selected to be optimized by analyzing the influencing factors of the classification performance of the neural network model.Finally,the three improvement researches which are the weight threshold optimization based on the improved genetic algorithm,using the designed activation function as the hidden layer activation function,and using the purelin function as the output layer activation function can make the capability of classification of network optimal.And it is applied to the recognition of classification of the defect magnetic flux signal to make the recognition of classification realized.To sum up,this paper studies the pretreatment,feature extraction and selection and BP neural network classifier in the recognition of classification of defective magnetic flux leakage signals which have a certain reference and application value.