MRS-kNN fault detection method for multirate sampling process based variable grouping threshold. (January 2020)

• Record Type: Journal Article
• Authors: Feng, Jian; Li, Keqin
• Is Part Of: Journal of process control. Volume 85(2020)
• Page Start: 149
• Page End: 158
• Publication Date: 2020-01
• Subjects: Multirate sampling process -- Fault detection -- Group modeling -- Variable threshold -- k-Nearest Neighbor; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2019.11.007
• Languages: English
• ISSNs: 0959-1524

Abstract

Highlights: The inconsistent and incomplete training sample set is divided into several consistent and complete sample groups according to the different sampled variables of samples. These groups include all samples in the original sample set, and no repeated samples between them. A variable threshold instead of a stable threshold is used to detect the online samples with different lengths. Every online sample is compared with the training samples which are belong to the same group as the online sample, and the threshold of this group is used to detect this online sample. The establishment of this model proposed by the MRS-kNN method reduces the need for prior knowledge and parameter estimations. This model can be applied to various sampling rates process fault detection, and applied not only to linear processes, but also to nonlinear processes. Abstract: For the multirate sampling process, some traditional multivariate statistical process monitoring methods cannot perform well because the lengths of all samples are not consistent. To handle this problem, a multirate sampling k -nearest neighbor fault detection method is proposed in this paper. The training sample set is divided into different groups according to the length of the sample to ensure that the sample length of each group is uniform. For all the groups, we can get a variable threshold corresponding to samples of different lengths. Also, this model can be developed into one that is suitable for fault detection of various sampling rate processes. Finally, the effectiveness of the proposed method is demonstrated by the simulation experiments on a numerical example and an industrial process.