Abstract: A multi-variable fault detection method based on reconstruction contribution analysis is proposed in view of the characteristics of multiple variables, large amount of collected data and varied data during the operation of complex systems. The improved reconstruction method can eliminate the shortcomings of the traditional SPE contribution graph method, such as insensitivity to fault data and insufficient diagnostic ability, and can conduct fault location when multi-variable faults occur simultaneously after the establishment of PCA model. The experimental simulation of the wind turbine system shows that this method can achieve accurate diagnosis when the faults of multiple variables occur simultaneously, no matter whether there are minor faults with gradual changes or abrupt changes.

Key words: Multivariable fault, fault diagnosis, principal component analysis, reconstruction method

摘要： 针对复杂系统运行过程中，变量多、采集数据量大、数据变化多样的特点，提出一种基于重构贡献分析的多故障变量检测方法。该方法在建立PCA模型之后，通过改进后的重构方法，可以消除传统SPE贡献图方法对故障数据不敏感和诊断能力不足的缺点，并且在多变量同时发生故障时进行故障定位。通过对风力发电机系统的实验仿真表明，该方法在对多个变量同时发生故障时，无论是存在渐变的微小故障还是突变故障，均能实现准确的诊断。

关键词: 多变量故障, 故障诊断, 主元分析, 重构法