基于机电混合数据驱动的风力发电机故障诊断与预测方法研究

As the number of wind turbines continues to grow and the high cost of maintenance, the need for online health monitoring and fault diagnosis and prognosis system becomes increasingly important. This project proposes a vibration and current-based data driven approaches for wind turbine fault diagnosis and prognosis. New time-frequency analysis methods, such as empirical wavelet transform and variational mode decomposition, will be used to analyze the signals to extract the fault-related features. It also aims to reveal the correlations between vibration and current signals. Synchronously sampling technique, which resamples the non-stationary signal such that the varying characteristic features of the wind turbine become constant values, is proposed to be used to diagnose wind turbine faults under the dynamic operational conditions. Since there’ve limited healthy and faulty data, a semi-supervised pattern recognition based approach is then proposed for wind turbine fault diagnosis. Considering the degradation of wind turbine is a dynamic and nonlinear process, a nonlinear degradation model will be built, and Bayesian filter based approach will be developed for updating the model parameters and predicting the remaining useful life of wind turbine. Finally, experimental studies will be carried out to demonstrate the effectiveness of the proposed vibration and current-based data driven approaches for fault diagnosis and prognosis of wind turbines operating in variable-speed conditions. This project is from the frontier of engineering, meets the demand of practical engineering, and has the significance to provide theoretical guidance and technique support to fault diagnosis and prognosis of wind turbines.

随着风力发电机装机容量的迅猛发展，维护问题突出，迫切需要研发有效的风力机健康监测和故障诊断与预测系统。本项目提出综合利用振动和电流信号，通过EWT和VMD等时频分析方法，提取风力机故障特征，揭示振动与电流信号之间的耦合特性。针对风力机复杂的运行条件，提出同步采样分析，将非稳态信号转换成稳态信号，进而实现变工况下的风力机故障诊断。考虑到获取风力机各种状况下的完整数据成本昂贵，通过对现有的有限数据的学习和建模，提出基于半监督模式识别技术的故障诊断方法。拟合分析风力机性能退化数据，构建非线性状态空间模型，提出基于贝叶斯滤波算法的故障预测方法，实现在线性能评估和剩余寿命预测。项目将通过搭建风力机故障仿真试验平台和设计性能退化实验，获取实验数据，验证基于机电混合数据驱动的风力机故障诊断与预测方法的有效性。本项目研究工作源于学科前沿和工程实际需求，对风力机的故障诊断与预测技术提供理论指导具有重要意义。

项目从故障诊断和故障预测两个方面，归纳风力发电机组的主要故障特点，梳理总结基于振动、电气信号和模式识别算法的故障诊断方法的研究现状，比较分析了各种方法的技术特点；参照风电机组中机械结构和电子系统性能退化的各自特点并结合SCADA数据，提出物理失效模型和数据驱动模型融合的故障预测方法。.故障诊断方面所开展的研究工作及重要结果如下：对装备信号进行S变换，结合时频信息构建二维随机变量的核密度函数，进而提取出与故障相关的冲击特征；基于装备多类故障状态信号，构建相应的标准密度函数库，采用频谱相似性比较方法，判别装备故障状态；提出一种基于高阶累积量分析的装备早期故障检测方法。.故障预测方面所开展的研究工作及重要结果如下：基于风电机组多类信息，通过融合分析，构建表征其健康状态的性能指标；利用异常阈值、矩形框截取等方法对装备不同健康阶段（正常工作阶段和退化阶段）进行了区分；拟合分析性能变化数据，构建装备的非线性退化模型，采用贝叶斯滤波（扩展/无迹卡尔曼滤波和粒子滤波）技术预测装备的剩余寿命；利用Copula函数分析不同性能指标间的相关特性，结合维纳随机过程实现装备的故障预测。.上述研究成果在理论方面可以为风电机组故障诊断与预测工作所涉及的故障发生、发展的演化机理，多类信号特征融合分析等方面的研究提供参考依据；在技术方面，可以为早期故障检测、故障诊断方法、剩余寿命预测等领域的研究提供重要的借鉴意义。