Reaearch On Separation Technology Of Aero-engine Vibration Signals

The vibration sources and noise mixed with each other when the aero-engine operating, which has become a major challenge when fault detection and feature extraction. At present, modern signal processing methods can't separate the complicated multi-mixed vibration signals effectively without any prior knowledge. But the blind source separation(BSS)technology is still in the initial stage. Therefore, the systematical study on separating the mixed vibration signals in aero-engines has not only important theoretical significance, but also practical value.For the research of the vibration characteristics of aero-engine, a further study on the existing BSS algorithms and its theory was done, which will provide a new way for the condition monitoring and diagnosis of aero-engine. The main contents of this paper are summarized as follows:(1) Studied the basic theories and typical algorithms of BSS in details. The appropriate algorithms which suit for the mixed aero-engine vibration signals were found out from the existing algorithms.(2) A new BSS algorithm suits for separating the mixed vibration signals was proposed. For the shortcomings of second-order cumulant-based BSS algorithm and fourth-order cumulant-based BSS algorithm, a new method simultaneous diagonalizes the second- and fourth-order cumulant matrix was proposed. The method makes use of the signals time structure to reduce the computation and uses the fourth-order cumulant which is not sensitive to the white noise signal to reduce the impact of the noise signal. The experimental datas show that the algorithm has a stable separation performance at different signal to noise ratio(SNR).(3) A new BSS algorithm in strong noisy environment was proposed. For the disadvantages of existing BSS algorithm in strong noisy environment, using the time-delay auto-correlation de-nosing method to reduce the impact of noise to the separating results firstly, and then using the BSS algorithm separating the de-noised signals. The effectiveness of the method was verified though the simulation experiments and the separation of the real vibration data.(4) The problem of estimating the number of the sources in BSS was deeply studied. The source number estimation methods about noise-free and noise model were studied. The source number estimation method based on the power spectral density function was high lightly introduced. The feasibility of the method was verified under different conditions.(5) Two groups'aero-engine vibration signals were separated. Achieved some useful conclusions though analysis and computation, which was successful to confirm the validity of BSS applied to separate the mixed aero-engine vibration signals.(6) A BSS package using MATLAB and Visual C++ software was designed based on the theory of setting parameters. This package made the operation of BSS simulation and applications conveniently under a graphical interfaces control.