Analysis of Principal Component Analysis-Based Reconstruction Method for Fault Diagnosis
详细信息 查看全文
- 作者：Zhe Zhou ; Chunjie Yang ; Chenglin Wen ; Jie Zhang
- 刊名：Industrial & Engineering Chemistry Research
- 出版年：2016
- 出版时间：July 13, 2016
- 年：2016
- 卷：55
- 期：27
- 页码：7402-7410
- 全文大小：605K
- 年卷期：0
- ISSN：1520-5045
文摘
The reconstruction method, defined in the frame of principal component analysis, is a popular and representative method of fault diagnosis. It has been successfully applied in several industrial processes. Although the sufficient condition of the unique fault subspace identification has been provided in the literature, it cannot be used in practice because the true fault subspace and magnitude are unknown. Thus, this condition cannot be used to confirm whether an identified subspace is the true fault subspace once there are more than one subspace determined as the possible candidates. This paper shows that the true fault subspace is hard to identify when a fault occurs and causes a significant change in the correlation of the variables while the T2 index is still in control. Through theoretical analysis and some illustrative examples, it is shown that the reconstruction method cannot always uniquely identify the true fault subspace in these cases. This indicates that care must be taken when using the reconstruction method. Some possible solutions are also provided to further improve the reconstruction method.