Design of a Multi-Agent System for Process Monitoring and Supervision

Abstract

New process monitoring and control strategies are developing every day together with process automation strategies to satisfy the needs of diverse industries. New automation systems are being developed with more capabilities for safety and reliability issues. Fault detection and diagnosis, and process monitoring and supervision are some of the new and promising growth areas in process control. With the help of the development of powerful computer systems, the extensive amount of process data from all over the plant can be put to use in an efficient manner by storing and manipulation. With this development, data-driven process monitoring approaches had the chance to emerge compared to model-based process monitoring approaches, where the quantitative model is known as a priori knowledge. Therefore, the objective of this research is to layout the basis for designing and implementing a multi-agent system for process monitoring and supervision. The agent-based programming approach adopted in our research provides a number of advantages, such as, flexibility, adaptation and ease of use. In its current status, the designed multi-agent system architecture has the three different functionalities ready for use for process monitoring and supervision. It allows: a) easy manipulation and preprocessing of plant data both for training and online application; b) detection of process faults; and c) diagnosis of the source of the fault. In addition, a number of alternative data driven techniques were implemented to perform monitoring and supervision tasks: Principal Component Analysis (PCA), Fisher Discriminant Analysis (FDA), and Self-Organizing Maps (SOM). The process system designed in this research project is generic in the sense that it can be used for multiple applications. The process monitoring system is successfully tested with Tennessee Eastman Process application. Fault detection rates and fault diagnosis rates are compared amongst PCA, FDA, and SOM for different faults using the proposed framework.