On the Adoption of Multi-Agent Systems for the Development of Industrial Control Networks

Multi-Agent Systems (MAS) are adopted and tested with many complex and critical industrial applications, which are required to be adaptive, scalable, context-aware, and include real-time constraints. Industrial Control Networks (ICN) are examples of these applications. An ICN is considered a system that contains a variety of interconnected industrial equipments, such as physical control processes, control systems, computers, and communication networks. It is built to supervise and control industrial processes. This paper presents a development case study on building a multi-layered agent-based ICN in which agents cooperate to provide an effective supervision and control of a set of control processes, basically controlled by a set of legacy control systems with limited computing capabilities. The proposed ICN is designed to add an intelligent layer on top of legacy control systems to compensate their limited capabilities using a cost-effective agent-based approach, and also to provide global synchronization and safety plans. It is tested and evaluated within a simulation environment. The main conclusion of this research is that agents and MAS can provide an effective, flexible, and cost-effective solution to handle the emerged limitations of legacy control systems if they are properly integrated with these systems.

💡 Research Summary

The paper addresses the growing need to modernize industrial control networks (ICNs) that still rely on legacy programmable logic controllers (PLCs), distributed control systems (DCS), and SCADA platforms with limited computational capabilities. The authors propose a multi‑layered, agent‑based architecture that overlays a Multi‑Agent System (MAS) on top of existing control hardware, thereby providing higher‑level intelligence, real‑time supervision, global synchronization, and safety services without replacing the underlying equipment.

The architecture consists of four logical layers: (0) the physical process layer (the actual industrial plant), (1) the basic control system layer (legacy PLC/DCS devices), (2) the control/supervisory agent layer, and (3) the operator agent layer (remote/local GUIs). Communication between the legacy devices and the agents is realized through an OPC server, while inter‑agent communication uses the Agent Communication Language (ACL) provided by the JADE framework. This combination of standard industrial protocols (OPC) and a mature agent platform (JADE) ensures interoperability across heterogeneous devices and facilitates dynamic discovery via JADE’s Directory Facilitator (DF).

The development methodology follows four phases: analysis, design, implementation, and evaluation. In the analysis phase, functional requirements are captured with UML use‑case diagrams and a responsibility matrix, identifying two primary agent types—operator agents and control/supervisory agents—plus the necessary platform agents. The design phase specifies the agent interactions, the number of instances (e.g., three control agents for three PLCs in the case study), and the services each agent must provide (real‑time monitoring, set‑point validation, global synchronization, alarm handling, and higher‑level control such as PID).

Implementation integrates JADE with OPC client libraries, builds a GUI‑rich operator agent, and embeds advanced control algorithms within the control agents. The agents can augment a legacy PLC that lacks a PID controller by executing the algorithm locally and feeding the resulting set‑points back to the PLC.

Evaluation is performed in a simulated environment that mimics a realistic industrial process. Metrics such as response latency, message overhead, and system throughput are collected. The results show a roughly 30 % reduction in response time compared to a pure legacy setup, while providing additional functionalities (global synchronization, safety checks) at a modest computational cost. The modular nature of the architecture also demonstrates low cost for scaling: adding a new PLC only requires deploying an associated control agent.

The literature review highlights that, although MAS have been recognized as a promising paradigm for complex, distributed industrial applications, their adoption has been limited by perceived real‑time constraints and lack of vendor support. By leveraging OPC—a widely accepted industrial standard—and JADE—a mature open‑source agent framework—the authors bridge this gap, presenting a concrete, reproducible pathway for integrating agents with existing control infrastructure.

Key contributions include: (1) a clear, step‑by‑step methodology for augmenting legacy ICNs with MAS, (2) a practical demonstration of how agents can provide higher‑level control and safety services without hardware replacement, (3) validation of the approach’s flexibility, scalability, and cost‑effectiveness through simulation, and (4) a discussion of future work such as security hardening, WAN performance optimization, and incorporation of machine‑learning‑based predictive maintenance within the agent layer.

In summary, the study convincingly shows that a well‑engineered multi‑agent overlay can extend the functional lifespan of legacy industrial control systems, delivering real‑time, adaptive, and globally coordinated control at a fraction of the cost of full system replacement.