Premium
Fault detection method for non‐Gaussian processes based on non‐negative matrix factorization
Author(s) -
Li Xiangbao,
Yang Yupu,
Zhang Weidong
Publication year - 2012
Publication title -
asia‐pacific journal of chemical engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.348
H-Index - 35
eISSN - 1932-2143
pISSN - 1932-2135
DOI - 10.1002/apj.1669
Subject(s) - non negative matrix factorization , initialization , principal component analysis , kernel principal component analysis , fault detection and isolation , kernel (algebra) , dimension (graph theory) , constraint (computer aided design) , computer science , matrix decomposition , gaussian process , pattern recognition (psychology) , process (computing) , partial least squares regression , artificial intelligence , gaussian , mathematics , machine learning , kernel method , support vector machine , actuator , eigenvalues and eigenvectors , physics , geometry , quantum mechanics , combinatorics , pure mathematics , programming language , operating system
ABSTRACT In this work, a new fault detection method based on non‐negative matrix factorization (NMF) is presented for non‐Gaussian processes. NMF is a new dimension reduction technique that can preserve spatial relationships corresponding and retain the intrinsic structure of original data. The basic idea of our approach is to use NMF to extract the latent variables that drive a process and to combine them with process monitoring techniques. A modified alternating least squares algorithm with an order constraint and a fixed initialization is proposed for solving the NMF problem. In addition, kernel density estimation is adopted to calculate the confidence limits of defined statistical metrics for NMF‐based monitoring method. Afterwards, the proposed method is applied to the Tennessee Eastman process to evaluate the monitoring performance, comparing with principal component analysis and independent component analysis. The experiment results clearly illustrate the feasibility of the proposed method. © 2012 Curtin University of Technology and John Wiley & Sons, Ltd.