基于能量解耦的风力发电旋转机械故障趋势预示方法研究

The fault trend information of wind-power rotating machinery under complicated varying working conditions not only includes fault trend information but also includes non-fault information such as varying loads, noise and so on. The information has the feature of energy decoupling. So the traditional prediction methods based on energy varying information can not reflect fault trend accurately,meanwhile,prediction methods utilizing single fault symptom information can not forecast fault development trend. To solve these problems, new prediction methods of fault trend based on energy decoupling were studied. In the research, a manifold feature extracting method of three-dimensional axis orbit for trend predicting and energy decoupling was presented, which constructed three-dimensional axis orbit, and mapped it into two-dimensional graphic by using nonlinear manifold analysis method. The graphic characteristic parameters extracting from graphic were used to represent fault information about rotating machinery; An independent component analysis method for trend predicting and energy decoupling was presented, which could extract fault information from energy monitoring information of wind-power rotating machinery by nonlinear blind signal separation and suppress the disturbance of wind information, noise information and so on;In order to reflect influence degree on different fault information from temporal and spatial scale, the key technology of various types of forecast information weight fusion was studied, which could improve the accuracy and effectiveness of prediction information. By this research, new methods of fault diagnosis and prediction maintenance for wind-power rotating machinery in real field under complicated varying working conditions would be provided and the level of safe operation and scientific maintenance would be improved.

在复杂变工况的风力发电旋转机械故障趋势信息中，不仅包含故障发展趋势信息，也包含载荷变化、噪声等非故障信息，这些信息具有能量耦合特征，传统的基于能量变化的趋势预示方法不一定反映故障趋势，同时基于单一故障征兆的预示方法难以全面、准确预示故障发展趋势，为此研究基于能量解耦的故障趋势预示新方法。提出趋势预示能量解耦的三维轴心轨迹流形图特征提取方法，构建三维轴心轨迹，通过非线性流形映射获取二维流形图像，提取用于描述流形图像形状的图形特征量，实现旋转机械的图形形式故障信息获取；提出趋势预示能量解耦的非线性独立成分分析方法，通过非线性盲信号分离，提纯能量形式故障信息，有效抑制风况信息、噪声信息等干扰；提出对多类预示信息加权融合方法，综合反映各类信息在空间与时间上对预示结果的影响程度，提高故障趋势预示信息的准确性和有效性。本研究能够为变工况风力发电机械故障预报提供新方法，有利于提高安全运行及科学维护水平。

传统的基于能量变化的趋势预示方法不一定反映故障趋势，同时基于单一故障征兆的预示方法难以全面、准确预示故障发展趋势,为此针对风力发电机组故障趋势信息具有能量耦合特征，研究能量解耦的故障趋势预示新方法。.（1）揭示了典型机电系统稳定性劣化发展演变的机理及传动系统动力学特性，构建了风电机组行星轮平移-扭转动力学模型，为分析故障发生原因与发展趋势提供动力学理论基础。.（2）提出了早期故障发展趋势弱信息的信号处理方法，在低信噪比情况下，实现噪声干扰抑制及早期故障信息分解。包括：基于改进的奇异值分解µ-SVD和局部均值分解的状态劣化弱信息处理；基于独立成分分析FastICA算法的故障信号提纯；自相关阈值法与自适应LMS组合降噪方法；基于不确定性推理的多传感器分布式检测融合算法等。.（3）提出了长历程变负载运行状态发展趋势特征能量解耦及时变非线性故障特征提取方法。包括：传动系统1.5维谱振动特征、阶次重采样信号包络特征、希尔伯特-黄变换敏感特征、小波阈值熵tSNE流形降维特征、小波时频图像纹理与矩融合特征等故障特征提取方法，实验研究表明所提出方法可有效提取故障演变信息。.（4）构建了系统运行稳定性劣化趋势预测模型与系统。包括：隐马尔可夫模型HMM与支持向量机SVM串联混合模型以及蚁群聚类RBF神经网络算法，以实现故障分类识别；建立基于信息熵和预测有效度的时变最优组合预测模型、T-S模糊神经网络预测模型，以实现劣化趋势预测；提出基于“劣化演化矩阵”的系统运行稳定性劣化状态评价方法，以实现运行状态评估。.同时，构建了风电机组故障预警实验平台，研发了机组状态监测诊断预警仪器，进行新方法的实验研究、实践验证及现场应用。.项目研究成果应用于风电机组等新能源装备的安全监测与故障预警，有利于保障设备安全可靠运行及实现科学维护，提升设备运行及管理水平，减少故障引起的经济损失，促进绿色能源利用与发展。