基于改进的多元离群检测方法的风机齿轮箱早期故障诊断

针对风电机组运行工况波动性以及机组早期故障特征不易提取的特点，提出一种基于改进的多元离群监测方法来实现风机齿轮箱故障的早期诊断。运用阶比重采样方法对原始振动信号进行预处理，并对处理结果进行量纲一因子分析；通过马氏距离建立风电齿轮箱的早期故障识别模型；利用多元线性回归改进多元离群检测算法进行实际数据的分析计算。结果表明，该方法较原始方法能够更早地察觉出风电齿轮箱早期故障。

Early Fault Diagnosis for Wind Turbine Gearbox Based on Improved Multivariate Outlier Detection

An improved method of multivariate outlier detection was used in early fault diagnosis for wind turbine gearboxes, which might extract the early fault features under fluctuation working conditions. First, the primitive vibration signals were preprocessed by order resampling, and the processed results were analyzed by dimensionless parameter analysis. Second, a model of early fault diagnosis was created based on Mahalanobis distance, which was used for turbine gearboxes. Finally, the actual data was analyzed by the method of multivariate outlier detection, which was improved by multiple linear regression. The results show that the new method may detect the gearbox faults earlier than original one.