A data-driven multiplicative fault diagnosis approach for automation processes $ Haiyang Hao a,n , Kai Zhang a , Steven X. Ding a , Zhiwen Chen a , Yaguo Lei b a Institute for Automatic Control and Complex Systems (AKS), University of Duisburg-Essen, Bismarckstrasse 81 BB, 47057 Duisburg, Germany b State Key Laboratory for Manufacturing Systems Engineering, Xi 0 an Jiaotong University, Xi 0 an 710049, China article info Article history: Received 25 July 2013 Received in revised form 25 October 2013 Accepted 15 December 2013 This paper was recommended for publica- tion by Dr. A.B. Rad. Keywords: Multiplicative fault diagnosis Process monitoring Multivariate statistics Key performance indicator Data-driven methods Large-scale systems abstract This paper presents a new data-driven method for diagnosing multiplicative key performance degrada- tion in automation processes. Different from the well-established additive fault diagnosis approaches, the proposed method aims at identifying those low-level components which increase the variability of process variables and cause performance degradation. Based on process data, features of multiplicative fault are extracted. To identify the root cause, the impact of fault on each process variable is evaluated in the sense of contribution to performance degradation. Then, a numerical example is used to illustrate the functionalities of the method and Monte-Carlo simulation is performed to demonstrate the effectiveness from the statistical viewpoint. Finally, to show the practical applicability, a case study on the Tennessee Eastman process is presented. & 2013 Published by Elsevier Ltd. on behalf of ISA. 1. Introduction The process industries have always played an important role in the global economy. Tightening competition is setting even higher demands for product quality and overall production efficiency. This has significantly increased the automation degree and com- plexity of industrial processes. Today 0 s production lines generally contain a great amount of control loops with numerous embedded components like sensors and actuators [1]. Key performance indicator (KPI) is becoming an important concept for large-scale complex industrial processes. It establishes a quantitative relation- ship between the performance of low-level technical loops/com- ponents and high-level production quality, production efficiency, energy and raw material consumption, etc. To keep high enterprise profit, KPI-based process monitoring and fault diagnosis tools play an important role. Associated with technological advances are the reliability and availability of various components on the low-level, which gradually reduces enterprise 0 s interest in those monitoring and diagnosis tools developed for total component failure, e.g., abrupt additive fault. Instead, there is an increasing interest in the KPI-based monitoring and diagnosis techniques. This is greatly motivated by the fact that the majority of technical loops/compo- nents in large-scale plants may frequently be subjected to perfor- mance degradation, e.g., multiplicative fault. In industrial applications, data-driven approaches are widely adopted for performance monitoring because of their simple forms and less requirements on design and engineering effort [2]. Effective monitoring approaches include basic multivariate statistics based approaches which are utilized on static processes with constant mean vector, e.g., principal component analysis (PCA) [3–6], independent component analysis (ICA) [7,8], partial least squares (PLS) [9,10], and subspace identification aided techniques which are suitable for dynamic processes [11–13]. To cope with process dynamics and nonlinearity, some variants of basic multivariate statistics based approaches have been reported [14–19]. A recent survey for them is given by [20]. Among these approaches, PLS is a very popular one for KPI-based performance monitoring and has received considerable attention both in research and practice. PLS was originally developed for linear regression purpose to handle possible collinearity among process variables, i.e., singular covariance matrix of normalized process variables which may decrease the performance of least squares estimation [21]. Considering its efficiency in dealing with large problems with big data due to the nature of one-dimensional recursive computations, PLS has been successfully applied to monitor industrial processes. According to the correlation to KPI, PLS divides the process variable space into two oblique subspaces, Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/isatrans ISA Transactions 0019-0578/$ - see front matter & 2013 Published by Elsevier Ltd. on behalf of ISA. http://dx.doi.org/10.1016/j.isatra.2013.12.022 ☆ A brief version of the paper was presented at IEEE ICCA, June 14, 2013. This paper was recommended for publication in revised form by Dr. A.B. Rad. n Corresponding author. Tel.: þ49 2033792930. E-mail address: haiyang.hao@uni-due.de (H. Hao). Please cite this article as: Hao H, et al. A data-driven multiplicative fault diagnosis approach for automation processes. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2013.12.022i ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎