Research On Feature Extraction And Recognition Method Of Corrosion Acoustic Emission Signal

Storage tank is the main oil storage equipment in petrochemical enterprises,and its safety is directly related to the production interests of enterprises,the surrounding ecological environment and the safety of people’s life and property.The traditional tank inspection method usually requires emptying the liquid in the tank before sending professionals into the tank for inspection.This process requires the tank to stop production for several days,which increases the operating cost of the enterprise,and also has the risk of safety accidents such as flash explosion and suffocation caused by improper operation.Acoustic emission technology,as a new on-line nondestructive testing technology,can complete the testing task from the outside of the tank,avoi d the tank production,and has high detection accuracy and sensitivity.This technology can accurately obtain the defect location and damage degree of the storage tank,so it is of great significance in ensuring the safe and stable operation of the storage tank during its service life.Corrosion is one of the common factors affecting the safety of storage tank,and the corrosion occurring in the tank interior and bottom plate will bring greater security risks.To solve this problem,the corrosion of storage tanks is studied in this thesis.Acoustic emission technology is used to detect the signals generated by the corrosion process of steel plates,and different types of corrosion acoustic emission signals are classified and identified.The research content of this thesis mainly includes the following aspects:(1)The concept and principle of acoustic emission detection technology are discussed,and the causes of tank corrosion and the mechanism of acoustic emission signal generated in the corrosion process are analyzed.According to the uniform corrosion,local corrosion and stress corrosion three different types of corrosion,the corresponding acoustic emission signal acquisition experiments are designed,and through the analysis of the waveform and spectrum diagram of the three types of signals,the conclusion is that the AE signals of different corrosion types have different duration,and have their own characteristics in frequency distribution.(2)The time-frequency analysis method is adopted to process the corrosion acoustic emission signal,so as to obtain the time-frequency diagram reflecting the relationship between time and frequency.Corrosion AE signals are processed by short-time Fourier transform,wavelet transform,S transform and generalized S transform respectively.Through comparative analysis,it is concluded that the time-frequency graph obtained by generalized S transform has the highest resolution and can effectively distinguish three types of corrosion signals.By changing the generalized parameters,the variation of window function with frequency can be more consistent with the frequency distribution of corrosion AE signals.The sparrow search algorithm with strong global and local search ability is introduced as the optimization algorithm of generalized parameters.The experimental results show that the optimized generalized S-transform has the highest aggregation degree of time-frequency graph.(3)The five-fold cross-validation method was used to divide the time-frequency graph data sets obtained from various kinds of time-frequency transform,and input them into Alex Net,VGGNet,Goog Le Net and Res Net convolutional neural networks respectively.Through comparative analysis,it is concluded that the optimal generalized S-transform data set has the best training effect,which verifies that the transform can effectively improve the time-frequency characteristics of corroded AE signals.By improving the residual module of Res Net and adding SE module,SE-Resnext network with better overall performance is obtained.To further improve the performance of the model,the Leaky Re LU activation function and the Adam W optimizer are introduced.Multiple groups of comparative experiments show that the improved SE-Res Ne Xt network has faster training convergence speed,higher accuracy and stronger generalization ability,and the recognition accuracy of the test set is up to 96.80%,which can effectively distinguish three types of corrosion signals.