Translated Abstract
With rapid development of modern industry and technology, wind turbine continues to be more precision, large-scaled, integrated, automatic and intelligent. The complex internal structure and the composite transmission movement of the wind turbine lead to inter-modulation and coupling of the dynamic signals, especially when the weak fault characteristics are easily covered by strong noises, causing fault identification even more difficult. In fault diagnosis research of wind turbines, it is important to not only realize the accurate identification of obvious faults in middle and late stages, but also study the potential danger, including early faults and weak faults. So as to timely identify failures when happening or developing. Changing ＂breakdown-maintenance＂ into ＂prior control＂ are preventive measures for operational reliability improvement, maintenance cost minimization, major accident prevention and accident risk reduction. Therefore, it is of great scientific significance and engineering value to study the effective processing of early fault diagnosis signal and the quantitative diagnosis method of fault degree.
Traditional signal processing methods usually extract fault features through suppressing or eliminating noise. However, the diagnostic capacity of weak fault signals is subjected to broken signal features because of loud noise or similar frequency of fault features and background noises. Contrary to making no use of noises, the noise utilization theory, which takes advantage of the signal noises to enhance signal features was proposed in the thesis. This method can not only avoid weakening signal feature with suppressing noise, but also promise the integrity of signal feature. The use of noise can detect the signals with lower signal-to-noise ratio (SNR), and opens up this new way forward enhancing detection of weak fault signals and has the great significance in theory and application.
The reasonable index for performance evaluation is a key to improve the adaptability of noise utilization theory and fault diagnosis. According to the mechanical fault features and characteristics of vibration signal, the collaborative SNR, covering factors of the tradition SNR, correlation coefficient, residual signal variance, zero point ratio, etc., as the index was constructed, which can realize the perfomance validation and parameters optimization of noise utilization theory. Numerical simulation shows that the collaborative SNR has better capability than other compared indexes. In view of the problem of lacking of the theory guidance that the experience and many experiments were needed for parameters and noise intensity adjusting in stochastic resonance (SR), the adaptive SR based on the collaborative SNR is proposed. The algorithm flow of the adaptive SR is given for mechanical fault diagnosis. Subsequently, the numerical simulation and simulated fault tests verify the effectiveness of the proposed method. In addition, this method is successful in bearing fault features in wind turbines.
Aiming at these problems that the result of single stochastic resonance detecting still includes the useful noise, and that the capability of using noise to enhance signal is limited, the second order stochastic resonance enhancement method based on the adaptive multiscale noise tuning of using Paul wavelet with the collaborative SNR as the objective function is proposed in the thesis. The ability of multi-resolution time-frequency analysis of wavelet, which can divide the input signals and noise into different frequency bands for realizing the control of intensity of signal and noise in different frequency bands, has been fully applied in the method. The property of the second order stochastic resonance, which can achieve the enhancement utilization of the noise energy in the signals, has been fully used. The proposed method has been validated by simulation tests and actual application in engineering.
As for the problem that the determination of noise amplitude and ensemble average number in ensemble empirical mode decomposition (EEMD) is short of effective theory guidance, adaptive EEMD based on noise utilization theory with the collaborative SNR index, in consideration of the time-frequency characteristic of signals and the principle of EEMD is introduced in the thesis. The proposed method revealed the mechanism of noise adding quantitatively, which can increase the detecting ability of EEMD for weak fault signals. The proposed method has been verified by experiment simulation and effective extraction of the fault feature of wind turbines coupling.
Nowadays, the condition monitoring systems for large-scale wind turbines usually adopt the traditional processing methods, which have high error report rate and low accuracy. In view of these problems, through the analysis of the operation characteristics of wind turbine and complex structure characteristics, condition monitoring and fault diagnosis software for large-scale wind turbines based on the B/S model was developed participantly. The software works well in the engineering practice for condition monitoring and fault diagnosis of wind turbines, especially for the weak fault signal. It is also a scientific proof for monitoring the operation of the wind turbine and making reasonable predictive maintenance plan.
Corresponding authors email
