Research On Acoustic Emission Processing Algorithm Of Helicopter Composite Specimen

In the fatigue testing of helicopter component, it is very meaningful foraccurate assessment of the important helicopter component life and security tocarry out research of helicopter component acoustic emission detectiontechnology, to earlier detect the damage initiation, location and type, and toimprove and enhance the efficiency and accuracy of damage detection in thefatigue testing. The object of this paper is the composite specimens commonlyused in helicopter components. In this paper, depth study is carried outaccording to the issues of AE signal attenuation on the composite specimen,de-noising, location and AE source identification. The researched methods areevaluated using the AE signal by pencil lead break and pressure off experiment.It resolves some key technology of the helicopter component acoustic emissiondetection technology. What have been done about it can be summarized asfollows.The attenuation characteristics of AE signal across through the compositespecimens commonly used in helicopter components is studied. The attenuationcharacteristics data are gathered by pencil lead break experiment on carbon fibercomposites specimen and honeycomb specimen. Energy and amplitudeparameters are used to analyze the distance attenuation characteristics of AEsignal. To overcome the shortcomings of traditional wavelet packet includingenergy leakage, inflexible frequency band selection and different frequencyresolution on different levels. Harmonic wavelet packet is used to analyze theattenuation characteristics of AE signals on different frequency bands and thefeature band of AE signal is obtained. This provides a firmly basis for AE sensorarrangement on helicopter components.To resolve the de-noising issue of AE signals, and to overcome theshortcoming of Wavelet threholding method including influenced by signalcharacteristics and the wavelet basis function and only applied to hight signal tonoise ratio (SNR) case, de-noising approaches based on empirical modedecomposition and wavelet transform are proposed, including IMF-Waveletmethod, EMD-Wavelet method and Wavelet-EMD method. The adaptivedecomposition characteristics of EMD to signal and low SNR denoisingadvantages are used to compensate for the lackage of wavelet denoising. Thestandard noise signals and AE signals by pencil lead break are used to analyzethe de-noising performance. The simulation results indicate that whether in high or low SNR case, EMD-Wavelet and Wavelet-EMD method have stablede-noising performance for AE signals.To solve the AE source location issue, and to overcome the shortcoming ofneural network based intelligent location method including over learning, largesample and local optimum, a new approach for linear location of AE sourcebased on least squares support vector machine (LS-SVM) for regression and anew approach for planar location of AE source based on multi-output supportvector regression (M-SVR) are proposed. The advantages of global optimum,small sample prediction performance and better converge performance of SVMare used to compensate for the lackage of neural network.The results of pencillead break location experiment on specimen of carbon fiber materials indicatethat the proposed approach can implement linear and planar location of AEsource effectively, and has better performance on convergence rate and locationaccuracy than neural network locator.To solve the fatigue damage identification problem of helicopter movingcomponent, a new approach for AE source type identification based on harmonicwavelet packet (HWPT) feature extraction and hierarchy support vector machine(H-SVM) classifier is proposed. HWPT overcome the shortcomings oftraditional wavelet packet including energy leakage, inflexible frequency bandselection and different frequency resolution on different levels, which extractsthe AE source type feature more accurately. The structure of hierarchical supportvector machine (H-SVM) multi-classifier is designed using clustering method.The results of pressure off experiment on specimen of carbon fiber materialsindicate that the AE source type feature is extracted more effectively by HWPT,and it has better performance on computational efficiency and identificationaccuracy than WPT feature extraction. H-SVM has better performance aftermodel parameters optimizing which resolves the small sample problem in AEsource type identification.In order to optimize the model parameters of SVM in LS-SVM regressionlinear location and M-SVR planar location, the niche particle swarmoptimization algorithm is studied. In the LS-SVM regression linear locationmethod, k-fold cross validation is used to evaluate the performance of LS-SVMregression, the optimized parameters of LS-SVR locator is selected with leastlocation error. In the SVM classifier, the evaluation function based onidentification ratio and structure complexity is used to evaluate the performanceof SVM, the optimized parameters of SVM are selected with highestidentification ratio and simplest structure.