Premium
Fault detection based on weighted difference principal component analysis
Author(s) -
Guo Jinyu,
Wang Xin,
Li Yuan,
Wang Guozhu
Publication year - 2017
Publication title -
journal of chemometrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.47
H-Index - 92
eISSN - 1099-128X
pISSN - 0886-9383
DOI - 10.1002/cem.2926
Subject(s) - principal component analysis , kernel principal component analysis , fault detection and isolation , outlier , pattern recognition (psychology) , computer science , nonlinear system , kernel (algebra) , artificial intelligence , anomaly detection , independent component analysis , k nearest neighbors algorithm , data mining , mathematics , kernel method , support vector machine , physics , quantum mechanics , combinatorics , actuator
Abstract Recently, multivariate statistical methods, such as principal component analysis (PCA), have drawn increasing attention for fault detection applications in industrial processes. However, industrial processes typically have complex multimodal and nonlinear characteristics. In these situations, the traditional PCA method performs poorly due to its assumption that the process data are linear and unimodal. To improve fault detection performance in nonlinear and multimode industrial processes, this paper proposes a new fault detection method based on weighted difference principal component analysis (WDPCA). Weighted difference principal component analysis first eliminates the multimodal and nonlinear characteristics of the original data by using the weighted difference method. Then, PCA is applied to the preprocessed data, neglecting the influences of multimodality and nonlinearity. Two numerical examples and an industrial application in a semiconductor manufacturing process are used to verify the effectiveness of WDPCA. The simulation results demonstrate that WDPCA shows better fault detection performance than the PCA, kernel principal component analysis (KPCA), independent component analysis (ICA), k ‐nearest neighbor rule (kNN), and local outlier factor (LOF) methods.