Key Technology Research On Avionics Health Management Based On Multi-wavelet And Support Vector Machine

With (Prognostics and Health Management, PHM) technology is becoming animportant part in the design and the usage of a new generation of aircraft, buildingavionics equipment failure prognostics and health management system is becoming thedevelopment direction of the signal process. As a fault diagnosis and managementmethods, PHM technology can monitor the important components and systems ofavionics to comprehensive health condition, fault prediction, and analysis and judgmentprocessing. As for monitoring aircraft critical components in working condition,assessing its state of degradation trends and predicting its remaining service of life, it issignificance to ensure its safe operation.Failure prediction capability of avionics system is a remarkable feature of the healthmanagement system. In order to improve the accuracy of failure prediction in avionicshealth management I proposed a prediction method based on multi-wavelet and supportvector machine. Multi-wavelet compared to single wavelet has better advantages ofextracting data band features, and coupled with strong generalization ability of themethod of least squares support vector machine has the advantage of overcoming thecurse of dimensionality and local extremum problem, and combine both, first withmulti-wavelet make data preprocessed, it will make data noise removed, thendecompose the denoised data with multi-wavelet into sub-sequences of a number ofdifferent bands, by analyzing the characteristics of each sub-sequence of digital,confirm the kernel function to construct support vector machine, and then predict thedata for each band respectively, and finally get the predictive value of each bandwavelet reconstructed to obtain final predicted value. The GHM multi-wavelet withprocessed SNR in the case of the input signal to noise ratio for12.90db can reach23.78db, mean square error of0.2365. With the high reliable thumb force controllingsensitive switch assembly data, get5step prediction with GHM multi-wavelet methodand least squares support vector machine, the prediction mean square error is only3.7307verified by simulation.