Research On Early Fault Diagnosis Technology And System Of Large Wind Turbine Gearbox

China's wind power industry has developed rapidly in recent years, and the installed capacity increase year by year. Because the larger wind power generator works in wild field and bad working conditions continuously, many early mechanical failures are difficult to be timely detected and treated, and develop into serious trouble with long running, and even lead to major accidents, which will affect the economic benefits badly of wind power enterprise. In multiple key components of large wind power generator, speed increasing gear box is the fault prone parts. And when serious fault appears, it is very difficult to repair it and the cost is very high. Therefore, the early fault diagnosis for the speed increasing gearbox of large wind power generator will be studied, so the potential failure will be discovered to do predictive maintenance, and it is very important for enterprise to reduce the cost of maintenance and improve the economic benefits.The main study work focus on the speed increasing gearbox of main type of large wind power generators which are double feeding and have variable pitch and speed. With the method of theoretical analysis and experiment verification, the construction and analysis of nonlinear dynamics fault model, the signal denoising method, the early fault signal feature extraction and diagnosis, the early failure prejudging of gearbox by detecting metal grain in lubricating oil, and the construction of online monitoring and diagnosis are studied in this paper, the main studies include:(1) The study of nonlinear dynamics model of the speed increasing gearbox of large power generators. Based on the theory of nonlinear dynamics, the nonlinear dynamics model of planetary gear transmission is established with considering of the time-varying meshing stiffness. The time curve, frequency and phase diagrams of every model part are got with normal and fault state of gear meshing and bearing supporting under different speed and load conditions. The calculation results show that the rotating frequency of input shaft has modulation influence on inherent feature signal of system under normal and fault state of gear meshing and bearing supporting, and it causes every order main frequency appears side frequency band in response frequency of system. Under the condition of bearing meshing fault state, the 2th or 4th times frequency occupy the main energy, and under the condition of bearing supporting fault state, the frequencies caused by system supporting stiffness lead to phenomenon of left shift of frequencies and the frequency modulation phenomenon is obvious. The study results can provide analysis data for signal feature extraction and part of evaluation standards for fault diagnosis.(2) The study of signal denosing method of self-correlation coefficient threshold spectrum and improved two order adaptive NLMS. Usually, the large wind turbines work at poor environment conditions, so the sampled vibration signals include complex noise component and low frequencies caused by tower random vibration. To eliminate these two types noises, the self-correlation coefficient threshold spectrum signal denoising method is provided to eliminate the random noises, and based on this denoising method, grouped method and auto threshold calculation method are further provided, and the improved two order adaptive NLMS denoising method is provided to eliminate the low frequency signal component in vibration of gearbox modulated by the tower vibration. By analysis and verification of the simulated and really measured signal, it shows that this two denoising method have better preprocessing effect to eliminate the random noise and the low frequency noise modulated by the tower in vibration signal of gearbox.(3) The study of self-correlation power spectrum feature extraction method with Hilbert transform demodulation on the order resampling signal of gearbox vibration. Firstly, the method of shaft rotating angle fitted by cubic equation is studied to resample the signal by equal angle, and its better effectiveness is verified by simulated signal and really measured signal. Secondly, the Hilbert transform demodulation method for the resampled signal by equal angle is studied. Finally, the square calculation demodulation, power calculation demodulation and Hilbert transform demodulation are compared and analysed with the simulated signal and really measured signal, the results show that the autocorrelation power spectrum feature extraction method of the signal resampled by computed order and demodulated by Hilbert demodulation is better for feature extraction than the other two methods, and it can extract feature of signal more accurately.(4) The study of fault isolation method based on wave packet threshold entropy t-SNE manifold learning. The t-SNE fault identification method based on the wavelet packet decomposition is studied in time-domain and frequency-domain. According to practical measured data, the manifold structure processed by the t-SNE reducing dimension is distinct and prominent in characteristics, which can be better for the fault status identification of equipment. And the experimental results show that the proposed method has better effectiveness in faults isolation compared with other high dimensional data construction methods and manifold learning methods.(5) The study of early gearbox fault diagnosis method based on online detection of metal abrasive particle in all flowing lubricating oil. The corresponding grinding particle detection sensor and detection instrument system are designed and developed emphatically, and the metal grain recognition method based on the local maximum and minimum values is proposed, and the experiments of particle detection were studied in lubricating oil. The results show that the design of metal grain detection system can detect the minimum of 150 micron metal grain, which have got remarkable measurement effect. The instrument system can prejudge early non-obviousness faults of gearboxes of large power generators by detecting the size and number of metal grain in lubricant oil.(6) The study of early condition monitoring and diagnosis system of remote wind turbines transmission system. The embedded data acquisition system based on Ethernet was designed, and the remote data transmission protocol based on TCP/IP was formulated, and the software of the long-distance monitoring as well as early fault diagnosis system based on the Microsoft.Net was developed by the mode of B/S (Browser/Server), which realize the early remote fault diagnosis of gearboxes of large power generators.