基于KPCA冗余检测的故障识别算法

Fault Identification Algorithm by Redundancy Supervision Based on KPCA Method

工业过程中,采用传统贡献图方法识别故障时,由于没有考虑传感器发生故障而导致测量数据异常的情况,可能得到错误的辨识结果,对此提出一种故障识别的新思路。该方法通过设置多个传感器监测某一变量的变化情况,建立起测量值与待测变量之间的冗余关系,提出将检测这种关系是否保持作为判定故障成因的依据,并利用KPCA方法将原始测量数据映射到高维空间进行分析,借助协方差矩阵的特征值大小实现对先前的冗余关系的量化和判定,在确定传感器是否发生故障后,参照贡献图定位故障源。结合风云某型气象卫星的姿态测量系统进行仿真验证,结果表明该方法有效克服了传统方法的缺陷,可以实现对故障的正确识别。

During the industry process, as the abnormal measurements detected by the faulty sensors are not taken into consideration, an erroneous conclusion could be obtained by the traditional contribution plot means. To solve these problems, a new method for fault identification is proposed. After establishing redundant relationship between the measurement and procedure variables by selecting multi-sensors to supervise one state of the process, this algorithm analyzes the structure of measure space via its eigen matrix through the KPCA method, and the cause of fault is therefore determined by the results whether the relationship are changed. Moreover, the source of fault is found through the contribution plots. Different fault modes are simulated on the basis of telemetric data of Feng-Yun satellite and comparing with the traditional method, the results demonstrate that the proposed algorithm can make a correct identification.