风电机组齿轮箱早期故障预警方法研究

为实现风力发电机组等变工况机电设备的早期故障预警,研发了变工况齿轮箱状态监测系统。在基于该系统的多种变工况行星齿轮磨损实验研究基础上,提出了一种基于流形学习的早期故障预警方法。该方法首先研究采用完全总体经验模态分解与改进快速独立成分分析盲源分离技术,有利于对复杂振动信号的滤波与盲源分离;然后研究改进了局部线性嵌入流形学习方法,基于时域、频域信息融合提取了早期故障敏感特征;最后应用k-近邻分类器实现变工况齿轮箱早期故障预警。实验研究表明,该方法提高了早期故障预警准确率,能够应用于风电机组等变工况机电设备的安全保障及科学维护,具有广泛工程实用前景。

Study on the early fault warning method for wind turbine gearbox

In order to achieve the early fault warning of the electromechanical equipment such as wind turbine that works under variable condition, a condition monitoring system of variable conditional gearbox was developed. Based on the wear experiment study on a variety of variable conditional planetary gears in this system, a new early fault warning method based on manifold learning is proposed. Firstly, the method studies the blind source separation technique that adopts CEEMD (Complete Ensemble Empirical Mode Decomposition) and MFICA (Modified Fast Independent Component Analysis) methods, which is beneficial to the filtering and blind source separation of complex vibration signals. Then, LLE (Locally Linear Embedding) manifold learning method is studied and improved. The early fault sensitive features are extracted based on time and frequency domain information fusion. Finally, the k nearest neighbor classifier is applied to realize the early fault warning of the variable conditional gearbox. The experiment study results show that the proposed method improves the accuracy of early fault warning, can be widely applied in the safety guarantee and scientific maintenance of the electromechanical equipment in variable working conditions such as wind turbine, and has a wide range of engineering and practical prospects.