| In this paper, putting emphasis on the pretreatment of test data, faults detective and diagnose, the faults detection and diagnosis technique for Liquid-propellant Rocket Engine (LRE) sensors is studied, by using the methods such as median filter, wavelet analysis, liner principal component analysis (PCA) and nonlinear PCA based on auto-associative neural network (AANN).Firstly, the history and the status of the sensors'fault detective and diagnosis, yet the methods in this domain at home and abroad, are introduced.Secondly, to remove the noise from the data which for the LRE sensors'faults detective and diagnosis, a data pretreatment method based on median filter and wavelet analysis is proposed. The median filter is used to remove the impulse noise in the data, and the wavelet analysis to remove other random noise. The test results confirmed that this method is better at remove noise.Thirdly, the fault detection and diagnosis method based on the liner principal components analysis and auto-associative neural network which called nonlinear PCA is discussed. The liner PCA is used to the sensors between which there exists obvious correlation, while the nonlinear PCA is used to the sensors between which there exists no obvious correlation. And the fault data reconstruction method is studied yet. The results verified by simulated faults of sensors show that the methods can not only detect and diagnose faults of sensors effectively, and the data of faulty sensors can also be reconstructed accurately by the proposed methods.Finally, a application software which is used to simulate the LRE sensors'faults detective and diagnosis is built, which can realize the functions such as removing noise, faults simulation, faults detective and diagnosis.The research results in thesis have some important reference to the engineering application system for the LRE sensors'faults detective and diagnosis, and improving the dependability and the veracity of the fault detective and diagnose.
|
PDF Full Text Request