基于单通道盲分离算法的大型风电机组早期机械故障诊断

发布时间：2018-05-18 17:46

本文选题：风电机组 + 盲分离　；参考：《沈阳工业大学》2013年博士论文

【摘要】：由于大型风力发电机组工作的条件比较恶劣，而且运行时通常不是长时间稳定地处于一种载荷工况，而是随着风、电网、温度等条件的变化而不断的进行调整，因此机组的传动链所传递的载荷是不断变化的，这就会对传动链上的各个零部件提出一定的可靠性要求：一是零部件质量的可靠性，二是当零部件出现早期损伤时，能够及时的发现，以便作到故障的早发现早处理。对于第一点与设计和制造有关，本文不作讨论，这里只讨论第二种情况。当风电机组出现故障时，故障零部件通常会产生具有一定特征的振动信号。但是在故障初期，这种故障特征并不明显。同时由于在风机运行时，许多零部件都会发出振动和噪音，振动传感器拾取信息时难免会受到强信号和噪声信号的影响,例如润滑和散热系统的运作、偏航和变桨机构的动作、电气系统的运行和发电机的励磁振动等，这些强信号和噪音之间还会互相干扰形成复杂的背景噪音，使早期故障特征振动信号湮没于背景噪音，提取真实准确的信息比较困难。同时，盲源分离算法对噪声很敏感，当利用该算法直接对混叠信号进行分离时，会造成很大的误差或得出错误的结论。因此，对采集到的振动信号进行盲分离前的强信号的去除以及降噪，对提高信噪比就显得尤为重要。本文采用自相关方法和EEMD方法对采集的信号进行降噪；采用扩展多虚拟通道FastIca技术进行强信号分离；针对诊断领域中的单通道信号难以应用盲源分离方法的难点，采用EEMD-FastIca技术，可以满足盲源分离（BSS）算法的多入多出（MIMO）条件，实现信号的盲分离。这种方法的优点是既不必先知道源信号的数量，，也不必先了解信号的产生和传递的参数，就能实现采集的信号中的各数据得盲分离；该方法可以提取风电机组传动链中的早期信号特征，提高了诊断的效率和准确性。为了验证方法的有效性，本文先通过仿真实验模拟出风电机组机械系统中的典型振动信号，并用上述方法分别进行分析测试，以确定其可以有效地分离信号。机组的增速箱分离出特征振动信号；通过对佳木斯风电场采集的数据进行分析，诊断和分析一台1.5MW级风力发电机组的轴流风机散热器的早期故障信号，验证该方法在风电机组振动信号处理的有效性和基于EEMD-FastIca算法和强信号去除的虚拟通道盲分离方法及其扩展算法适用风电机组信号的处理、预测机械系统的早期故障。
[Abstract]:Because the working conditions of large-scale wind turbines are relatively bad, and the operation time is usually not in a load condition for a long time and stable, but with the changes of wind, power grid, temperature and other conditions, they are constantly adjusted. Therefore, the load transmitted by the transmission chain of the unit is constantly changing, which will put forward certain reliability requirements for each component on the transmission chain: first, the reliability of the parts quality; second, when the parts are damaged early, Be able to find out in time, in order to make the fault early detection and early processing. The first point is related to design and manufacture, which is not discussed in this paper. Only the second case is discussed here. When wind turbine failure occurs, fault components usually produce vibration signals with certain characteristics. However, at the beginning of the fault, the characteristics of the fault are not obvious. At the same time, since many parts and components will emit vibration and noise when the fan is running, the vibration sensor will inevitably be affected by strong signals and noise signals when picking up information, such as the operation of lubrication and heat dissipation systems, the action of yaw and propeller mechanism. With the operation of electrical system and excitation vibration of generator, these strong signals will interfere with each other to form complex background noise, which makes the vibration signal of early fault feature obliterate in the background noise, so it is difficult to extract the true and accurate information. At the same time, the blind source separation algorithm is sensitive to noise, when it is used to separate the aliasing signal directly, it will cause great error or get the wrong conclusion. Therefore, the removal of strong signals and noise reduction before blind separation of the collected vibration signals is particularly important to improve the signal-to-noise ratio (SNR). In this paper, the autocorrelation method and EEMD method are used to reduce the noise of the collected signal; the extended multi-virtual channel FastIca technique is used for strong signal separation; and the EEMD-FastIca technology is adopted to solve the difficulty of blind source separation method in the single channel signal in the diagnosis field. Blind source separation (BSS) algorithm can satisfy the multi-input and multi-output MIMO-conditions and realize blind signal separation. The advantage of this method is that it can achieve blind separation of each data in the collected signal without first knowing the number of the source signal and the parameters of the signal generation and transmission. This method can extract the early signal features from the transmission chain of wind turbine, and improve the efficiency and accuracy of diagnosis. In order to verify the effectiveness of the method, the typical vibration signals in the wind turbine mechanical system are simulated by simulation experiments, and the above methods are used to analyze and test respectively to determine that the signals can be separated effectively. By analyzing the data collected from Jiamusi wind farm, the early fault signals of the axial fan radiator of a 1.5MW wind turbine are diagnosed and analyzed. The validity of this method in wind turbine vibration signal processing and the blind separation method of virtual channel based on EEMD-FastIca algorithm and strong signal removal and its extended algorithm are proved to be applicable to wind turbine signal processing and early fault prediction of mechanical system.
【学位授予单位】：沈阳工业大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TM315;TH165.3

【参考文献】