(187g) Plant-Wide Visualization for Situation Awareness Using Ising Model Based Clustering of Vanishing Correlations

Process monitoring has an important part in safety operation of industrial systems such as chemical plants, electric power plants and so on. With the increasing accident risk originated from aging of their equipment, the importance have become larger and larger. Process monitoring typically consists of four components; fault detection, fault identification, fault diagnosis and process recovery [1]. Among them, fault identification is the most critical task, which determines a faulty subsystem. Since it is an initial task for operators in their workflow of abnormal situation handling. If the operators misjudge a faulty subsystem, they need more time to recover from the fault. In the worst case, the fault causes an accident. In fault identification, data driven approaches, which are based on statistical analysis, have been often taken due to its high feasibility [2-8]. Against dedicate improvement on finding the most suspicious measured variable, visualization for fault identification have not been studied well.
 The most common visualization is a bar chart or a heat map of fault identification index which shows degree of abnormality for each measured variable [9]. This visualization helps operators to pick up several suspicious measured variables. However, the visualization is often noisy [9, 10]. As a results, operators may not be able to extract meaningful features from the visualization. Although it is indicated that comparison the indices and a process flow diagram gives an insight for better fault localization [1], the methodology is still an open problem.
As more descriptive visualization of abnormal situation, fault propagation path diagram have been provided [10, 11]. The fault propagation path diagram consists of nodes connected with directed edges. The nodes represent measured variables and the edges represent causal dependencies and each of the directions is oriented in the causality, namely from the cause to the effect, and the root of the nodes represents the most relevant variable. They are often displayed on a process flow diagram [11]. It may pinpoint the most relevant variable to the root cause, but extraction of precise causal dependencies is still challenging task [10]. If the diagram is incorrect, operators blindly believe the results or are confused due to its complexity. Therefore the path diagrams are risky visualization for operators, since the visualization is too descriptive.
 In abnormal situation, operators are required to make a decision fully utilizing their domain knowledge and the latest situation of the system. In that sense, they require more intuitive and comprehensive visualization for situation awareness of plant-wide abnormal situation.
 In this work, it is proposed that the plant-wide visualization mapping clustered suspicious measured variables onto a process flow diagram and accompanying an identical symbol for each cluster. Each of the cluster is associated with a time range of the beginning of a remarkable change in system status. Therefore, the visualization enable operators easily to interpret a development of abnormal situation along time based on their domain knowledge. The clustered suspicious measured variables result from clustering of vanishing correlations. A vanishing correlation can be defined as a status when a residual of the pair-wise model exceeds its pre-fixed threshold. Since suspicious measured variables may contain much uncertainties [12, 13], vanishing correlations are focused in this work. Fault identification with vanishing correlations has been proposed in [5, 14-16]. In the method, vanishing correlations are represented as a part of a graph structure which called an invariant network. An algorithm of this visualization trace back vanishing correlations at time under consideration, typically the latest time. On the basis of their temporal features and their pattern in an invariant network, the clusters are determined. After that, the clustered suspicious measured variables are estimated with computing their fault identification indices within each of the time ranges. Finally, the suspicious measured variables are displayed on a process flow diagram with identical symbols according to their cluster. In order to obtain accurate clusters, Ising model [17] based clustering is also proposed.
 The proposed visualization is validated through a case study using simulation data of a vinyl acetate monomer production plant [18, 19]. In this case study, three consecutive but individual multiple faults are supposed. The proposed visualization displays three time ranges on remarkable changes in system status, and suspicious measured variables within each of the clusters are located around the corresponding faulty component. The results represented the abnormal situation correctly and were more intuitive than its fault identification map and its fault propagation path diagram.
 As demonstrated in this work, the proposed visualization will enhance the capability of operators in abnormal situation handling and effectively work for complex systems which tend to make situation awareness difficult.

[1] K. Severson, P. Chaiwatanodom, and R. D. Braatz, Perspectives on process monitoring of industrial systems, IFAC-PapersOnLine, vol. 48, no. 21, pp. 931-939, 2015.
[2] J. V. Kresta, J. F. Macgregor, and T. E. Marlin, Multivariate statistical monitoring of process operating performance, The Canadian Journal of Chemical Engineering, vol. 69, no. 1, pp. 35-47, 1991.
[3] P. Nomikos and J. F. MacGregor, Multivariate spc charts for monitoring batch processes, Technometrics, vol. 37, no. 1, pp. 41-59, 1995.
[4] G. Jiang, H. Chen, and K. Yoshihira, Discovering likely invariants of distributed transaction systems for autonomic system management, Cluster Computing, vol. 9, no. 4, pp. 385-399, 2006.
[5] G. Jiang, H. Chen, and K. Yoshihira, Modeling and tracking of transaction flow dynamics for fault detection in complex systems, IEEE Transactions on Dependable and Secure Computing, vol. 3, no. 4, pp. 312-326, 2006.
[6] M. Kano and Y. Nakagawa, Data-based process monitoring, process control, and quality improvement: Recent developments and applications in steel industry, Computers and Chemical Engineering, vol. 32, no. 1-2, pp. 12-24, 2008.
[7] S. Yin, S. X. Ding, X. Xie, and H. Luo, A review on basic data-driven approaches for industrial process monitoring, IEEE Transactions on Industrial Electronics, vol. 61, no. 11, pp. 6414-6428, 2014.
[8] S. Yin, X. Li, H. Gao, and O. Kaynak, Data-based techniques focused on modern industry: An overview, IEEE Transactions on Industrial Electronics, vol. 62, no. 1, pp. 657-667, 2015.
[9] X. Zhu and R. D. Braatz, Two-dimensional contribution map for fault identification [focus on education], IEEE Control Systems, vol. 34, no. 5, pp. 72-77, Oct. 2014.
[10] G. Li, S. J. Qin, and T. Yuan, Data-driven root cause diagnosis of faults in process industries, Chemometrics and Intelligent Laboratory Systems, vol. 159, pp. 1-11, 2016.
[11] L. H. Chiang, B. Jiang, X. Zhu, D. Huang, and R. D. Braatz, Diagnosis of multiple and unknown faults using the causal map and multivariate statistics, Journal of Process Control, vol. 28, pp. 27-39, 2015.
[12] J. A. Westerhuis, S. P. Gurden, and A. K. Smilde, Generalized contribution plots in multivariate statistical process monitoring, Chemometrics and Intelligent Laboratory Systems, vol. 51, no. 1, pp. 95-114, 2000.
[13] P. V. den Kerkhof, J. Vanlaer, G. Gins, and J. F. V. Impe, Analysis of smearing-out in contribution plot based fault isolation for statistical process control, Chemical Engineering Science, vol. 104, pp. 285-293, 2013.
[14] A. B. Sharma, H. Chen, M. Ding, K. Yoshihira and G. Jiang, Fault detection and localization in distributed systems using invariant relationships, 2013 43rd Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN), pp. 1-8, 2013.
[15] Y. Ge, G. Jiang, M. Ding, and H. Xiong, Ranking Metric Anomaly in Invariant Networks, ACM Transactions on Knowledge Discovery from Data, vol. 8, no. 2, pp. 1-30, 2014.
[16] W. Cheng, K. Zhang, H. Chen, G. Jiang, Z. Chen, and W. Wang, Ranking causal anomalies via temporal and dynamical analysis on vanishing correlations, in Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 805-814.
[17] E. Ising, Beitrag zur Theorie des Ferromagnetismus, Zeitschrift fur Physik, vol. 31, no. 1, pp. 253-258, 1925.
[18] H. Seki, M. Ogawa, T. Itoh, S. Ootakara, H. Murata, Y. Hashimoto, and M. Kano, Plantwide control system design of the benchmark vinyl acetate monomer production plant, Computers and Chemical Engineering, vol. 34, no. 8, pp. 1282-1295, 2010.
[19] T. Yumoto, S. Ootakara, H. Seki, Y. Hashimoto, H. Murata, M. Kano, Y. Yamashita, Rigorous Dynamic Simulator for Control Study of the Large-scale Benchmark Chemical Plant, IFAC Proceedings Volumes, vol. 43, pp. 49-54, 2010.