Research On Signal Processing And Recognition Method For Pipeline Ultrasonic Guided Wave Detection

Pipeline transportation plays a very important role in the national defense and economy industry.Cracks,holes,and other defects pipelines can easily lead to major safety accidents and seriously threaten people's life and property.So using a fast and accurate non-destructive inspection technology is beneficial to reduce pipeline safety accidents.Compared with traditional nondestructive testing technology,ultrasonic guided wave testing technology has unique advantages.Studying the echo signal processing,feature extraction and pattern recognition in ultrasonic guided wave detection technology is beneficial to distinguish among pipeline defects rapidly and accurately,it is of great significance to protect the safe of pipelines.In this paper,the echo signal processing algorithm,pipeline feature extraction and pattern recognition algorithm are researched.Combined with the research to build a pipeline ultrasonic guided wave detection experimental system,the experimental system is used to distinguish between the pipeline features and defects.(1)Study the effect of different window lengths of short-time Fourier transform on signal time-frequency analysis and use short-time Fourier transform to deal with the echo signal.Design an optimized wavelet threshold noise reduction algorithm based on artificial fish swarm algorithm,select the appropriate wavelet basis function and decomposition layers,the artificial fish swarm algorithm is used to find the optimal threshold to optimize the wavelet threshold noise reduction and use this method to deal with the echo signal.Design a method based on wavelet threshold-EMD to deal with the echo signal.Analyze the processing effect of the three methods on the echo signal and the recognition accuracy of the pipeline features and defect positions.Study the influence of different echo signal processing methods on the feature parameters and choose appropriate signal processing methods to deal with the echo signals.The four common features and defects of pipeline welds,end faces,cracks and holes are taken as research objects,extract the feature parameters of each type of pipeline including kurtosis coefficient,skewness coefficient,discretecoefficient,entropy value,wavelet coefficient energy value(8),and wavelet coefficient variance value(8)and draw feature parameter curves.Analyze the difference and regularity of different types of pipeline feature parameter curves.(2)Combine the feature parameters to build a BP neural network,analyze the effects of different transfer functions between the neurons in each layer on the pipeline feature and defect recognition rate,optimize the BP neural network.Combine the feature parameters to build a LVQ neural network,analyze the influence of different number of neurons on pipeline feature and defect recognition rate,optimize LVQ neural network.A genetic algorithm combined with neural network is proposed to distinguish among the pipeline features and defects.Using genetic algorithm to screen feature parameters and reduce the correlation between feature parameters,the screened feature parameters are combined with the optimized BP and LVQ neural networks to identify the pipeline features and defects respectively.The recognition results show that the recognition rate of genetic algorithm combined with neural network method is significantly higher than that of BP or LVQ neural network alone,and the recognition rate is higher than 80%.(3)Based on previous research results,design the hardware system and software platform of ultrasonic guided wave detection,establish sample database of different types of pipeline features parameters.Build a pipeline ultrasonic guided wave inspection experimental system and use the system to conduct experimental research on artificially prefabricated features and defective pipelines.The experimental results show that,when the excitation frequency is 95 kHz,the experimental system has the highest recognition rate of pipeline features and defects.The positions recognition error of the pipeline features and defect is less than 10 cm,and the types recognition rate of the pipeline features and defect is higher than 90%,It is verified the pipeline ultrasonic guided wave signal processing algorithm,feature extraction and pattern recognition algorithm can well distinguish between the pipeline features and defects.