基于聚类分析的涡扇发动机的潜在故障检测

Potential failure detection of turbofan engine based on clustering analysis

针对多变量的过程统计监控问题,提出了一种基于聚类分析的潜在故障检测的方法。首先采用基于密度的减法聚类算法(SC)对数据进行聚类,然后结合基于分割的最大熵模糊聚类算法(MEFC)对数据再次聚类,利用聚类结果对数据进行状态划分并确定故障状态,最后实现潜在故障检测。通过对涡扇发动机数据集FD001进行实例验证,该方法能在故障发生的若干运行周期前检测到潜在故障。

For multivariate statistical process monitoring,the idea of potential failure detection based on clustering analysis is proposed. The data is clustered by means of density-based subtractive clustering and segmentation-based maximum entropy fuzzy clustering successively. Data based on clustering results is divided into several states and one state is thought of as faulty state,by which potential failure is detected. Through a case study performed on turbofan engines data FD001,the proposed method can detect potential faults in a number of operating cycles before failure occurs.