风电机组关键部件故障趋势预测方法研究

Wind turbine is the crucial equipment in renewable energy industry, whereas there lie some disadvantages such as high failure rates and short life-span in our domestic wind turbine in service. Based on the current operation status of bearing of main axis, gearbox and generator in wind turbine, and combining histroy data, structural characteristics and operation conditions, failure prediction for key components is an attractive technology to predict and determine the future failure status and realize precognition maintenance and high efficient operation for wind power industry. In this project, a multi-carrier-wave failure modulation model is built with the consideration of the characteristics of long drivetrain, multi-failure points and wide frequency band etc. Considering the effect of random wind load acting on wind turbine, the blind source deconvolution is adopted to decompose the vibration signal and independent components which are sensitive to fault and noise are obtained. Through constructing the characteristic index reflecting the fault level, the fault feature is separated from the non-failure energe in the vibration signal. Through hidden semi-Markov model, a failure prediction model is built based on data driven method of extracted fault feature to predict the failure trend of key components in wind turbine, and the effectiveness of the prediction model is verified through the comparison analysis and site tracking. There is momentous significance in theory and practice to develop the high efficient maintenance system for wind power industry and reduce the cost of operation and maintenance for wind turbine.

风电机组是可再生能源产业中的重要装备，而我国在役机组存在故障率高、寿命短等问题。关键部件故障趋势预测是以风电机组的主轴轴承、齿轮箱和发电机的运行状态为基础，结合历史数据、结构特性和运行工况，对机组未来可能出现的故障进行预测和判断，实现风电产业的预知维护和高效运营。项目针对风电机组传动链长、故障点多、频带覆盖宽等特点，研究其关键部件多载波故障调制模型；考虑风载荷等复杂激励对风电机组的影响，提出基于盲源解卷的方法进行振动信号分解，以获取故障敏感独立成分和噪声，构建反映故障变化程度的特征指标，实现故障特征与随机非故障能量的分离；以隐半Markov模型为理论基础，建立基于故障特征数据驱动的故障预测模型，预测关键部件故障状态发展趋势，通过对比分析与现场跟踪的方法验证预测模型的准确性。项目的研究对于形成风电产业合理高效的维修体制，降低风电机组的运营维护成本具有重要的理论与现实意义。

风力发电是改善我国能源产业结构、减少空气污染的重要型式。然而，由于设计与生产能力较弱、企业运行维护经验不足，我国在役风电机组存在故障率高、发电效率低下等问题。风电机组具有如下特点：1）传动链零部件众多、结构复杂；2）承受频繁的变速风载荷和刹车冲击；3）运行于环境恶劣的高空和边缘地区，维修困难。.项目分析风电机组不同类型传动链的结构特点，研究外部载荷激励下全尺寸风电齿轮箱的振动响应，发现一级行星加两级平行轴的传动结构比二级行星加一级平行轴传动结构的故障率高，原因在于行星轮部分由于结构对称且转速较低，载荷分配均匀，发生故障的概率较小。在平行定轴齿轮副中，小齿轮是其中的薄弱环节；研究先进的信号处理方法用以提取多故障耦合的、强噪声背景下的故障特征，发现齿轮箱高速轴的不对中容易导致轴承故障，进而引发高速级齿轮副故障。项目设计的循环比率系数可以发现强电磁振动下发电机轴承故障特征；研究变速、变载荷运行工况下的故障量化指标，为故障趋势发展和寿命预测奠定基础；提出利用关键部件自身早期发展规律训练神经网络模型并预测剩余寿命的方法，获得较为精确的预测结果。.项目的研究内容涵盖机械动力学、信号处理、机器学习与健康管理领域，能够解决复杂服役环境下的故障监测与寿命预测的问题，具有较高的学术价值，研究结果切合我国目前风电机组规模庞大、故障率高企不下的实际需求，对于节约风电行业的运行维护成本，提高风能的转换水平具有重要的工程应用前景。