(614g) Data-Driven Subsystem Configuration and Distributed Process Monitoring

Modern industrial processes that comprise interconnected operating units that are coupled through material, energy and information flows have been widely applied to achieve higher production efficiency and product quality as well as increase business profitability. Process monitoring, of which the objective is to determine whether the real-time process operation is in an abnormal condition, plays a critical role in avoiding the occurrence of plant-wide failures and safety incidents, such that the safe and economical operation of modern industrial processes can be ensured [1,2]. It is worth mentioning that these modern industrial processes are typically characterized by large scales and complex structures, which leads to difficulties in applying centralized process monitoring approaches to these processes. Therefore, it is necessary to exploit a new framework to develop new monitoring schemes that are capable of appropriately and efficiently handling medium- to large-scale modern industrial processes [2,3,4,5].

Data-based process monitoring within a multi-block/distributed framework is promising for modern industrial processes. This is because: 1) The development and implementation of data-driven algorithms do not need a first-principles model but only require process operation data. 2) By using a multi-block/distributed architecture within which the process variables are divided into groups of smaller sizes, limitations of conventional centralized monitoring induced by the large scale of a process and a large number of process variables can be bypassed, local process information can be more appropriately assessed, and the overall monitoring performance can be improved.

To develop a multi-block/distributed monitoring scheme, typically two key steps need to be conducted. The first step is to decompose all the process variables into smaller subsystems, the second step is to design local monitors based on the configured subsystems of variables after process decomposition. While in the existing literature, there are some results proven to be effective with their advantages. However, the optimality of the subsystems structure is not taken into account. Also, it is difficult to use most of the existing methods to handle increasingly complex processes.

Based on these observations, we aim at proposing a two-layer distributed monitoring method. In this work, we tackle both steps involved in the development of a distributed monitoring mechanism. In terms of subsystem configuration, the strength of correlation between each two process variables is assessed based on mutual information [6]. Taking advantage of the correlation strength information, a community detection [7,8] based subsystem configuration method is proposed to partition all the process variables into smaller subsystems. Then, a two-layer distributed monitoring scheme consisting of local monitors for the configured subsystems is developed for plant-wide fault detection. The first layer treats intra-subsystem correlations, while the second layer addresses correlations between different subsystems. Real-time information exchange among local monitors is enabled, and in this way, the local monitors are able to coordinate their decisions on whether faults have taken place in the process. A numerical example and a wastewater treatment process are introduced to illustrate the effectiveness and applicability of the proposed approach.

References

[1] Q. Jiang, X. Yan, B. Huang. Review and perspectives of data-driven distributed monitoring for industrial plant-wide processes. Industrial & Engineering Chemistry Research, 58(29):12899-12912, 2019.

[2] S. Khatib, P. Daoutidis. Optimal feature selection for distributed data-driven process monitoring. Industrial & Engineering Chemistry Research, 59:2307-2317, 2019.

[3] S. Khatib, P. Daoutidis, A. Almansoori. System decomposition for distributed multivariate statistical process monitoring by performance driven agglomerative clustering. Industrial & Engineering Chemistry Research, 57(24):8283-8298, 2018.

[4] H. Shahnazari, P. Mhaskar. Distributed fault diagnosis for networked nonlinear uncertain systems. Computers & Chemical Engineering, 115:22-33, 2018.

[5] Q. Jiang and B. Huang. Distributed monitoring for large-scale processes based on multivariate statistical analysis and Bayesian method. Journal of Process Control, 46:75{83, 2016.

[6] T. M. Cover. Elements of Information Theory. John Wiley & Sons, 1999.

[7] M. E. J. Newman and M. Girvan. Finding and evaluating community structure in networks. Physical review E, 69(2):026113, 2004.

[8] V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and E. Lefebvre. Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 2008(10):P10008, 2008.