Health Diagnosis Technology Research On A Certain Aeroplane Stabilizer

With the rapid development of Chinese aviation industry, health diagnostic technology of aircraft has become more and more important to ensure the flight safety. The stabilizer is an important component of the aircraft system to ensure the aircraft fly with various attitudes, whose health status plays an important part on the safety of the entire aircraft system. In this paper, the health monitoring and diagnostics research of stabilizer of a certain type of aircraft is discussed.Health diagnostic process has three parts: the collecting of diagnosis information, the state feature information extraction and pattern recognition.The state feature extraction and pattern recognition are the most important .The health status feature extraction of aircraft stabilizer is related to the reliability and accuracy of the health diagnosis, which is the most important in the health diagnosis of stabilizer. According to the non-linearity and non-stationary of the aircraft AE (acoustic emission) signals in the fatigue test, the methods of wavelet transform and EMD (Empirical Mode Decomposition, EMD) decomposition were respectively used in this paper to extract feature. By using these two methods, we can extract the different eigenvector of the absolute maximum, singular value, standard deviation of the wavelet coefficients, IMF energy, HHT energy IMF singular value of EMD, and then achieve the purposes of extracting the eigenvector of the health status.Health diagnosis is essentially pattern recognition, and a good identification method guarantees the diagnostic accuracy. In this paper, spectral analysis, support vector machines, probabilistic neural network and Elman neural networks were used to recognize and match the pattern. And the diagnosis result is satisfaying.In the end, the diagnosis result of different eigenvector and diagnosis methods were systematically analyzed and summrized.After analyzing the diagnosis result of the stabilizer, the following conclusions were obtained:(1) During the process of feature extraction from plane stabilizer, the correct rate of EMD eigenvector is higher than the wavelet eigenvector, and the IMF energy eigenvector gets the highest correct rate.(2) In the pattern recognition process, because the SVM has the unique advantages in a small sample classification, it is the best choice when difficult to obtain fault samples.Because of PNN'fast training and high precision, PNN is the best choice when easy to obtain diagnosis samples.