Research On Acoustic Emission Evaluation Technology Based On Feature Selection

Acoustic emission detection technology has high sensitivity to the internal active defects of equipment.With the expansion of the application field of technology,the demand of extracting more valuable information from waveform and parameter data becomes more and more extensive.Meanwhile,the complexity of noise increases,so does the difficulty of signal processing.The development of applied mathematics and statistics provides reference for the development of acoustic emission signal processing technology.Convex optimization can avoid initializing the algorithm and falling into local extremum,and it performs well in the selection of retrieval step size.In this paper,convex optimization theory is applied to acoustic emission signal processing technology.Combining optimization algorithm,statistical theory and traditional acoustic emission characteristic analysis method,the following work is carried out for characteristic parameters used in acoustic emission evaluation:1.According to the theoretical knowledge of statistics,the maximum dependent feature selection algorithm based on mutual information is optimized and converted into the maximum correlation minimum redundancy feature selection algorithm.According to the characteristics of convex sets,the optimal incremental search algorithm is designed to implement the mRMR algorithm.2.Based on the simulation test platform of valve internal leakage,the simulation study of valve internal leakage of different types and sizes was carried out.Fourier analysis and wavelet packet decomposition were used to process the experimental data,and finally the experimental signal frequency domain characteristics and filtered experimental data of different sizes and types of valves were obtained.Combining with the current signal waveform parameters in electricity,the waveform parameters of electricity are extended to acoustic signal processing and combined with the traditional acoustic emission characteristic parameters to form the candidate feature set of the feature selection algorithm.3.In this paper,weighted mRMR algorithm is used to select evaluation features from candidate feature sets,and support vector machine is used to carry out clustering analysis on experimental data.According to different number of evaluation accuracy of feature conditions,the optimal combination of evaluation feature sets for clustering analysis is finally determined.4?The mRMR evaluation feature set,ReliefF evaluation feature set and AE evaluation feature set were used for data clustering respectively,and the clustering accuracies werecompared.The results show that the accuracy of mRMR feature clustering is up to 70%,which is 10.2% higher than that of the other evaluation features.