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

All kinds of pressure pipeline and high pressure boiler can be seen everywhere today when industrial was developed quickly. It may lead to leak when the pipe or wall material was damaged due to reasons such as corrosion or wear in use process. Acoustic Emission (AE) detection technology is a testing technique using of local materials rapid release of energy generated by the transient excitation of elastic wave which occupies an important position in the nondestructive testing. Through analyze the detected acoustic emission signal processing, then combined the leakage signal detection theory and detection methods, to realize the classification of the leakage signals. It has an important theoretical value and practical significance to maintain the safe running of the pipe, boiler, etc, to avoid the waste of resources. Therefore, it is imminent to research a kind of method applicable to industrial leak detection.Three types of feature extraction and recognition method is proposed mainly in the paper. Acoustic emission data was collected in the laboratory environment. It has collected three kinds of simulated leakage acoustic emission signals:rap, sand paper and broken lead. All experimental data used in the paper come from the signal acquisition.Considering the acoustic emission signal is similar to speech signal, so it can use framing method to process acoustic emission signal. Then using the Autoregressive Model (AR) to extract the signal characteristic value of per frame. Put the signal characteristic values together and constitute the feature vectors. The method can overcome the defects of traditional AR Model parameter extraction method in application of long time series. Finally, using hidden Markov model (HMM) to train and test the feature vectors.Because the empirical mode decomposition (EMD) can do the stationary processing to signals, first take EMD to non-stationary acoustic emission signal and get the smooth IMF component. Then take AR model for each order IMF component, and feature vector consist of AR parameters. Finally, using support vector machine (SVM) to realize classification recognition. Introduced the BP (Back Propagation) neural network recognition method for the sake of comparison analysis at the same time. The experimental results show:the SVM recognition accuracy is higher than the BP neural network’s.Mathematical morphology (MM) signal processing method has been widely used in various fields. Based on this, introduced the knowledge of Mathematical Morphology, this paper proposes a new method to identify the acoustic emission signal. First respectively take multi-scale decomposition Mathematical Morphology to three types leakage acoustic emission signals. Then calculate the spectral energy in different scales, and calculate the spectral entropy. Then calculate the proportion of energy spectrum entropy of every dimension, constitute the feature vectors; Finally use the SVM to train and test feature vectors. Results show that the proposed feature extraction method’s recognition accuracy is the highest. The experiment proved that introduced the knowledge of mathematical morphology and entropy into acoustic emission leak detection is feasible.