Managed Complexity in An Agentbased Vent Fan Control System Based on Dynamic Re-configuration

Transcription

1 Managed Complexity in An Agentbased Vent Fan Control System Based on Dynamic Re-configuration Fred M. Discenzo Rockwell Automation, USA Francisco P. Maturana Rockwell Automation, USA Dukki Chung Rockwell Automation, USA 1. Introduction Developments in advanced control techniques are occurring in parallel with advances in sensors, algorithms, and architectures that support next-generation condition-based maintenance systems. The emergence of Multi-agent Systems in the Distributed Artificial Intelligence arena is providing important new capabilities that can significantly improve automation system performance, survivability, adaptability, and scalability. The new capabilities provided by multi-agent systems has shifted control system research into a very challenging and complex domain. A multi-agent system approach enables us to encapsulate the fundamental behavior of intelligent devices as autonomous components. These components exhibit primitive attitudes to act on behalf of equipment or complex processes. Using this approach, we have implemented a set of cooperating systems that manage the operation of an axial vent fan.

2 We have implemented a laboratory vent fan system (Figure 1) that operates autonomously as a fan agent in the context of a chilled water system comprised of other agents such valve, load, and pump agents. This paper presents the foundation technologies that are essential to realizing an adaptive, reconfigurable automation system. The vent fan system serves to validate the agent methodology to manage the inherent complexity of highly distributed systems while responding dynamically to changes in operating requirements, degraded or failed components through prognostics, control alteration, and dynamic re-configuration. Figure 1. Vent Fan Demonstration System The concepts above and new engineering developments have helped achieve new and important capabilities for integrating CBM technologies including diagnostics and prognostics with predictive and compensating control techniques. Integrated prognostics and control systems provide unique opportunities for optimizing system operation such as maximizing revenue generated for capital assets, maximizing component lifetime, insuring machinery survival or mission success, or minimizing total life-cycle costs. 2. Intelligent Machines 2.1. Machine Intelligence Beginning with the famous Turing Paper about 50 years ago there has been an ongoing effort to understand cognition and intelligence in order to make machines more useful [Charniak 1985] [Nilsson 1980]. With a goal of enhancing the capability of machines, Artificial Intelligence (AI) includes techniques such as expert systems, fuzzy logic, artificial neural networks, and related model-based and model-free techniques [Charniak 1985]. Many of the automation successes reported apply biologically inspired adaptive and knowledge-based techniques to solve well-targeted, specific automation problems such as adaptive control, defect classification, and job scheduling to name a few [Zurada 1994]. The capabilities which may be provided by intelligent machines may be categorized based on the degree of embedded knowledge with the most capable systems employing real-time goal adjustment, cooperation, pre-emption, and dynamic re-configuration [Discenzo(a) 2000][Discenzo(b) 2000]. These capabilities may be effectively integrated in an agent-based system employing intelligent machines in a distributed automation system. This architecture built on a foundation of a society of locally intelligent cooperating machines provides a effective framework for the efficient and robust automation of complex systems.

3 2.2. Multi-Agent Systems Our approach is to encapsulate the fundamental behavior of intelligent devices as autonomous components. The components employ models of primitive device behavior and enable agents to act on behalf of physical devices or complex processes. The approach of establishing application-specific agent behavior in a reusable and scalable manner finds counterparts in other research activities such as Multi-agent Systems (MAS), Autonomous Agents, Flexible Manufacturing, and Virtual Enterprise [Shen 2001][Vasko 2000][Zhang 1999]. The evolution of web-based systems, which pushes Internet communication to accommodate agent-like services, also coincides with industrial automation requirements. Language and protocols can also serve as the basis to explore information exchange and resource discovery for agents (e.g., XML, SOAP, and UDDI [XML&DTD Specification][Xerces XML Parser][Universal Description UDDI White Paper, 2000][UDDI 2000][Simple Object Access Protocol, 2000]). In this paper, we focus on the requirements for the controller level. The Foundation for Intelligent Physical Agents (FIPA) [FIPA][FIPA-OS] provides welldefined and widely accepted standards for multi-agent systems development that coincide with several of the premises established in this work. FIPA implementations accelerate the development of multi-agent systems. Previous results demonstrated for autonomous control have been extended to incorporate agents that do planning, communication, diagnostics, and control [Vasko 2000][Maturana 2000][Maturana 2002]. Our focus is on the aggregation of autonomous behaviors and coordinated coalitions of smart resources. The approach we have taken is to establish a general architecture to deploy information agents for resource discovery in a highly distributed system. Resource capability may be established using an open system architecture that permits utilizing existing or future prognostic algorithms (Figure 2). OSA/CBM TRANSDUCER DATA ACQUISITION SIGNAL PROCESSING CONDITION MONITOR HEALTH ASSESMENT DECISION SUPPORT PROGNOSTICS PRESENTATION 3. Chilled Water Demonstration System Diagnostics Evaluate Configuration Equipment Model Module Hardware equipment Execution Execution Control Module 3.1. Vent Fan Hardware Configuration Fans and related air handling applications are often critical applications that occur in a wide range of automation, commercial, and military systems. Agent Diagnostic Module Inter-Agent communication Planner Module Figure 2. OSA-CBM in an Agent Framework

4 We have defined and implemented a vent fan system employing an integrated diagnostic / prognostic / controller system in an agentbased representation. This system has been implemented as one component in a hardware-inthe-loop simulation of a chilled water system. The vent fan system shown in Figure 3 is operated with a variable frequency drive (VFD) and programmable logic controller (PLC). During operation the system dynamically adjusts air flow based on state changes or new requirements and in collaboration with other agents operating in related parts of a chiller system. Figure 3. Vent Fan System Schematic The system in Figure 3 includes a 3-phase, 2 hp, 230 volt, a-c induction motor coupled to a fan. The motor is an Intelligent Motor that includes embedded current, voltage, temperature, and vibration sensors and a processor to enable the motor to continually assess its own health. The motor and fan are mounted inside a 20 diameter steel tube as an integrated axial vent fan and the VFD motor controller is mounted on top of the structure. This hardware system is operated as an intelligent sub-system in a chilled water system Chilled Water System Configuration Agent types and capabilities The chilled water system is simulated on a PLC and other than the fan system and a water pump system, all other components such as valves, pipes, and loads are simulated in a programmable logic controller. A model of the chilled water system is shown in Figure 4. The CWS is comprised of a community of agents that control the physical equipment. In this system, the Chillers (Chiller 1 and 2) are chilled water plants, each containing a chiller unit, one or two pumps, an expansion tank, regulation valves, and flow and pressure Cold valve Chiller1 Hot valve Load1 Load2 Load3 Load4 Chiller2 Figure 4. Chilled Water System Model

5 sensors. The Loads (1, 2, 3, and 4) are associated with a heat generator and correspond to some vital or non-vital operation. In addition, the Loads have an internal water circulation system that also provides heatexchanging services. In this manner, the heat is evacuated from the Load area using convection heat exchange. Cold water is transmitted from the chillers into the loads through the main supply water pipes. Hot water is transmitted from the loads back to the chillers through the main return water pipes. The main water circulation system has valves to control water flow and to isolate pipe ruptures. The function of Chiller1 is to provide water cooling. Inside the Chiller1 module there is a data repository and a connection to the physical vent fan. The vent fan increases the heat exchange capability of the chiller. The intelligent component contained in Chiller1 is a hybrid system comprised of simulation modules and the physical device. The vent fan system was connected to the simulation PLC using a wireless (IEEE b) network. Chiller2 provides water cooling capacity in simulation form only. The main purpose of the agent system is to react to the physical system (simulation) changes to regulate the temperature of the load components. There are several considerations in this simple model: The agents need to be goal-oriented in specifying cold water requirements. This permits agents to avoid equipment damage while optimizing cold water usage. The agents need to cooperate to resolve priorities since cold water may not be 100% available to all units. The agents need to adapt their set points locally to respond to limited availability of cold water. The agents should be aware of dangerous situations occurring in the heating process to enable actions based on urgency and context. The agents should have a set of self-regulating algorithms to do low-level automatic control Chilled Water System Operation The vent fan system with VFD may be operated in several modes one of which is under the control of an autonomous agent. In this mode, system operation changes dynamically and without regard to the set point speed or flow specified on the VFD. The communication layer of the vent fan system handles the cooperation with the agent-based system and enables the vent fan system to be interrogated or controlled by agents. When the agent-based system starts up, the axial vent fan system is registered as a resource that can provide cooling capability. If an agent decides to use the vent fan system, it can start the fan and command to provide airflow. The low-level controller implemented in the DSP maintains air flow at a prescribed level based on system need or operating objectives. Fan speed is automatically increased or decreased to maintain flow and protect the system (e.g. selfdiagnostics / prognostics) based on input from the vent fan agent. Later, the agent can check whether the fan system is providing the airflow it is asked to provide. The agent can also use the additional diagnostic information the fan system provides, such as system blockage. This

6 diagnostic information is published using OSA/CBM compliant XML pages. These XML pages are served by the web server which is running on the vent fan system Dynamic Virtual Clusters We propose the study and implementation of a self-emerging organization based on the dynamic gathering of system capabilities. We conceive the system capabilities as fine granular and distributed throughout the system. The dynamic gathering of capabilities is a mechanism to achieve organizational reconfiguration. The mechanism uses dynamic clustering of agents representative of system capabilities for short periods. User and system tasks trigger the dynamic clustering. The task complexity determines the size and configuration of the clusters. Gathering multiple clusters forms a cooperation domain. An important characteristic of this mechanism is the capability of the system to aggregate resources as needed into the emergent organizations. This organizational feature provides an architecture that is robust and survivable. Several different mechanisms may be used to prescribe and manage the formation of the clusters and to coordinate communication among the components. 4. Dynamic Re-configuration 4.1. Open Systems Architecture for Condition-Based Maintenance The utilization of open, industry standards for asset registry provides important capabilities for integrating operating information across a manufacturing plant and even across facilities. Recent developments have resulted in an Open Systems Architecture for Condition-Based Maintenance that provides a framework for the real-time integration of machinery health and prognostic information with decision support activities (Figure 2). This framework spans the range from sensor input to decision support ( This architecture specification is open to the public and may be implemented in a DCOM, CORBA, or HTTP/XML environment Integrated State Assessment and Control There are significant operational and econimic benefits possible by closely coupling machinery health (e.g. diagnostics) and anticipated health (e.g. prognostics) information with real-time automatic control. Given a current operating state for both the machinery and the process we can drive the system to achieve a prescribed operating state at a particular time in the future. This future operating state may an improved state than would occur if we did not alter the control based on machinery health information. Furthermore, the future state we achieve is chosen to be optimal in some sense such as machinery operating cost, machinery lifetime, or mean time before failure. The prescribed operating state of a particular machine may be sub-optimal however, as part of an overall system, the system-wide operating state may be optimal with regard to energy cost, revenue generation, or asset utilization. Dynamic re-configurable agents are an effective framework for realizing the benefits of integrated prognostics and control. 4.3 Agent Organization

7 The chilled water simulation system includes fluid mechanics equations to represent the water circulation in the pipe network for both the cold and hot transmission lines. Each simulated component generates input and output (I/O) data for the equipment. The simulation data is stored in the controllers inside data-tag symbols. The agents use the data tags as their interface with the physical equipment, The agents run in the controllers and utilize data-tags to manage the operating system and agent behavior using middleware [Maturana 2002]. Ladder logic algorithms synchronize the control events initiated by the simulation such as cooling load requirements, for response by the suitable suite of agents. Given a specific cooling demand on the chillers, the Chiller1 Agent may require an increase in the speed of the vent fan to increase the water cooling rate. This action corresponds to a first level reconfiguration. There are other levels of reconfiguration possible in this system. For example, the agents have been programmed to react to broken pipes events. 4.4 Dynamically Created Clusters In order to form the emergent organization, the agents register their respective capabilities in the organizational knowledge agents known as Directory Facilitators. To create decision-making organizations, agents need the ability to find suitable DF 1 DF 3 agents for particular tasks. DF 2 There are several Agent1 Agent2 Agent5 Agent6 architectures to accomplish this search. Examples of such Agent3 Agent4 approaches can be found in Figure 5. Distributed Structure of Directory Facilitators the Blackboard, Matchmaker, Brokering System, Mediator, Federated, and Acquaintance Models Architectures [Shen 2001]. Consistent with the FIPA standard, we use a Matchmaker-based architecture which is based on Directory Facilitators (DF). On each capability request, the DF agent provides a list of agents that coincide with the requested capability. For this model, it is required that each capability provider register its capabilities with its local DF agent. Later, when a request for services is initiated, the DF agent acts as a passive agent because it only provides information services. The DF agent organization may hierarchial and agent capability registration may be made in a breadth-first or depth-first manner. Alternatively, to avoid propogation, information may only be propagated locally and new capabilitity requests will be processed by local DF agents who carry out meta-level communications to discover needed capabilities. DF agents then provide location and service capability information of remote agents to the initial requesting agent. Relevant information may be stored locally to enhance the mount of organization learning (Figure 5). 5. Opportunities & Challenges Automation systems are being applied to more complex and safety-critical systems and there is a commensurate need for increased safety and reliable operation. The software content and complexity of automation systems continues to increase rapidly while software problems represent a leading cause of production breakdowns [Salimen 1992].

8 The new paradigm of Multi-agent / Holonic systems can provide significant benefits such as scalability, reliability, and survivability for complex critical systems. The broadscale deployment of intelligent devices, such as an intelligent vent fan system, utilizing this paradigm serves to reinforce these benefits and demonstrate an extremely flexible, adaptive, and robust system. Some of the challenges that remain include the need for a consistent framework and information model that will encompass integrating disparate agents, operating constraints, mission planning, dynamic optimization criteria, adaptive learning, and selforganizing behavior. The technological developments cited above combined with intelligent devices implemented in an agent-based / Holonic framework promise to provide unprecedented capabilities for the automation of a broad class of complex systems. A society of autonomous, cooperating agents are well-suited to address dynamic, complex systems of the type that were previously only the domain of biological systems. References Charniak, E., & McDermott, D, 1985, Introduction to Artificial Intelligence, Addison-Wesley Discenzo(a), F. M., Unsworth, P. J., Loparo, K. A., Marcy, H. O., Intelligent Motor Provides Enhanced Diagnostics and Control for Next Generation Manufacturing Systems, IEE Computing and Control Engineering Journal, Special Issue on Intelligent Sensors, Summer 2000 Discenzo(b), F. M., Marik, V., Maturana, F., Loparo, K. A., Intelligent Devices Enable Enhanced Modeling and Control of Complex Real-Time Systems, International Conference on Complex Systems ICCS, Nashua NH., May 2000 FIPA, The Foundation for Intelligent Physical Agents (FIPA): FIPA - OS (Open Source), Emorphia: Jeff Bradshaw (Ed.), Software Agents, MIT Press, Cambridge, 1997 Maturana F., Balasubramanian S., and Vasko D.,: An Autonomous Cooperative Systems for Material handling Applications. ECAI 2000, Berlin, Germany, 2000 Maturana F., Staron R., Tichy P., and Slechta P.: Autonomous Agent Architecture for Industrial Distributed Control. 56th Meeting of the Society for Machinery Failure Prevention Technology, Section 1A, Virginia Beach, April 15-19, 2002 Nilsson, N., 1980, Principles of Artificial Intelligence, Morgan Kaufmann (Los Altos) Salimen, V, Verho, A, 1992, Multidisciplinary Problems in Mechatronics and Some Solutions, Computers Electr. Engng., Volume 18, Number 1, pp. 1-9 Shen W., Norrie D., and Barthès J.P.: Multi-Agent Systems for Concurrent Intelligent Design and Manufacturing. Taylor & Francis, London, 2001 Simple Object Access Protocol (SOAP) 1.1, W3C Note 08 May 2000: UDDI Data Structure Reference v1.0, UDDI Open Draft Specification 30 September 2000: Universal Description, Discovery and Integration, UDDI Technical White Paper, September 6, 2000: Vasko D., Maturana F., Bowles A., and Vandenberg S.: Autonomous Cooperative Systems Factory Control. PRIMA 2000, Australia, 2000 Xerces XML Parser, part of Apache XML Project: XML & DTD specification, World Wide Web Consortium: Zhang, X., Norrie, D.: Agentic Control at the Production and Controller Levels. IMS 99, Sept , 1999, pp , Leuven, Belgium 1999 Zurada, J. M., & Marks II, R. J., & Robinson, C. J., 1994, Computational Intelligence: Imitating Life, IEEE Press, ISBN , 1994, pp. 5-12