Application Of Cyclic Spectrum In Vibration Test Of Wind Power Gearbox

Rotating machinery is the most widely used. Due to the impact of its structure and operational characteristics, the fault signal collected shows the characteristics of non-stationary, so the traditional stationary signal analysis methods cannot effectively extract fault characteristics. Its statistic parameters such as the mean and the correlation function change periodically or multiperiodically with time, so they meet the cyclostationarity (one kind of special non-stationary signals). In this paper, the noise sources from wind power gearbox in test-bed have been analyzed. The cyclic auto-correlation and the cyclic spectral density function are used to noise diagnosis.The experimental data analysis shows that cyclic spectral analysis helps to extract the gearbox failure, and the analysis also shows the effect of the cyclic autocorrelation and cyclic spectral density function in the coexistence of multiple faults in rotating machinery.This paper firstly develops a more comprehensive narrative for cyclostationary theory in gear fault diagnosis’s application status, progress and development trend. After a brief description of the concept of cyclic spectrum, the paper calculates the cyclic auto-correlation function and cyclic spectrum density function, of the simulated signals, and finds out its spectrum distribution,, so as to lay a good foundation for the next step of the engineering application. In addition, the paper studies second-order cyclic spectrum demodulation and noise reduction performance, and verifies the cyclic auto-correlation function of the FM and AM by comparing with the Hilbert demodulation method. The introduction of wind power gearbox in test-bed’s structural characteristics and the location of the sensor distribution, calculation of the characteristic frequency of the rotating machinery have been done. Meantime, spectral analysis, fast Fourier transform, power spectrum, cepstrum, Hilbert demodulation analysis, Wavelet packet decomposition are applied in the actual fault signal, and the conclusion is that the form of a single fault characteristic frequency is less than convincing. In addition, it is proved that the failure caused by not being enough stiffness of the shaft will bring out the FM information in the signal. Finally the irrationality of this test-bed is further verified by checking the high speed shaft stiffness. And we discuss the feasibility of the program, and solve the problem of this test-bed.At the end of this paper, a wind turbine gearbox alarm which is based on the cyclic auto-correlation function of wavelet analysis-probabilistic neural network method, has been provided. This method overcomes the traditional shortcomings fixed on alarm line and introduces the cyclic auto-correlation function to determine the fault location more accurately. Then the method determines the alarm threshold according to the historical data, with the changing of the alarm threshold an adaptive alarm line appears. Thus, it makes the needs of practical engineering more satisfied.