Research On Particles Detection And Identification System For Aerospace Power

Satellite power is used for supplying power in the satellite, and its reliability directly affects the safety of the whole system. Up to now, there are tin,iron,copper and wire line particles, which is one of the reasons resulting in breakdown, because the level of manufacturing is limited. Hence, researching on detection of particles in the satellite power has important theoretical significance and practical value for improving reliability of the whole satellite system.At present, particles in the relays and other electronic components are detected by Particle Impact Noise Detection (PIND) method, but there is no relevant theory and method used for detecting particles in the satellite. In this paper, a new detection method is proposed based on PIND method and the shortage of PIND is reduced, meantime, material identification is increased. The overall plan is designed based on a turntable. The particle signal is obtained by the acoustic sensor, and then amplified and filtered, reaching the initial detection. At the mean time, Programming the host computer software is to realize sampling, displaying and storing. Wavelet transform and stochastic resonance are used to detect whether there are particles or not, and neural network is used to realize identifying the material though using MATLAB tools.To solve the problem of low accuracy of PIND test, this paper adopts two different methods for testing. For the small particle signal mixed in the system noise, using wavelet de-noising method can effectively remove the system noise, with getting particle signal, and then threshold method is proposed to detect particles. For the tiny particle signal completely submerged in the system noise, if wavelet de-noising disposes the signal, it will eliminate useful signal. This paper puts forward stochastic resonance method to detect tiny particle signal, which makes good use of system noise and does not eliminate the noise, so it can improve detection accuracy of tiny particle.If material information of particles is known, the particles are avoided in the process of design and manufacturing. The key points of material identification are feature extraction and selection of classification algorithm. After analyzing the character of the four kinds of material, wavelet packet transform is adopted to extract the characteristics of particles. At last, LVQ neural network and BP neural network realize classification recognition, with the accuracy 80% and 82%, respectively, fully meeting the needs of practical application.