DIS 2006 15.6.2006 Helsinki University of Technology
Multi-Agent System Enhanced Supervision of Process Automation
Teppo Pirttioja1, Antti Pakonen2, Ilkka Seilonen3, Aarne Halme1, Kari Koskinen3
Helsinki University of Technology, Finland1Automation Technology Laboratory3Information and Computer Systems in Automation
2VTT Technical Research Centre of Finland
DIS 2006
IEEE 2006 Workshop on Distributed Intelligent SystemsJune 15 -16, Diplomat Hotel Prague, Czech Republic
DIS 2006 15.6.2006 Helsinki University of Technology
Agenda
Background and motivation Process automation and user needs Available technologies
Architecture Multi agent system based architecture Roles of agents Internals of an agent
Test scenario Conclusions and future work
DIS 2006 15.6.2006 Helsinki University of Technology
Background and Motivation:process automation Process Automation
Factories are running 24/7 and as effectively as possible General trend: Less personnel and bigger and more complex
factories, so there is more to supervise Ordinary IT solutions are used, because its cheap and it works
Results often data systems with mixed user interfaces Information is measured and gathered but not easily available
to users
User need easy and efficient access to all data Users need right information, in the right place, in right time,
and in right level of abstraction and easily! It’s not about tools and technology, it’s also the way people
are used to operate
DIS 2006 15.6.2006 Helsinki University of Technology
Background and Motivation: user needs within process automation
System within process automation is running mainly on it’s own User intervention mainly needed when there are problems in the system Users need to get information when critical changes happen Changes may result from input material variations, devices malfunctions, etc...
Various user profiles of supervision Process operators; physical process, measurements, ... Servicemen; device conditions for drives, valves, ... System developers; performance issues, control optimization, ... Business managers; orders, financial aspects, ...
What kind of concrete supervision tools are needed? Easy browsing and navigating trough all process related data Search engine to access data in separate and heterogeneous systems Find situations and data combinations that are interested by the user Easy setup for monitoring a combination of various measured variables
DIS 2006 15.6.2006 Helsinki University of Technology
Background and Motivation:What technologies might be useful?
Integration problem Semantic Web tools makes more and more information available
in machine understandable form May make the integration of separate systems easier There are no ontologies about factories available yet and there is no ready and standardized solution of how to
combine number of services to build up more complicated services (orchestration)
Supervision problem Automate routine supervision as much as possible Let people be just decision-makers for real problems that are not
possible to automate Proactive agents are argued to be suitable for active supervision
DIS 2006 15.6.2006 Helsinki University of Technology
Architecture: agents as information mediators in Semantic Automation
Proactive operation of agents
System integration with semantic web tools
DIS 2006 15.6.2006 Helsinki University of Technology
Architecture: agents operating in various roles
IA
PA
DF
CA
PA
PA
FA
AA
CA
DA
IA
CA CA
Continuous data analyzing
Various events from wrapper agents
CA = Client Agent, DA = Diary AgentIA = Information Agent, FA = Fault AgentAA = Alarm Agent, DF = Directory Facilitator
IAIA
Information processing
User Interface
DIS 2006 15.6.2006 Helsinki University of Technology
Architecture: internal structure of an agent
User configurable manager module Uses data processing modules for actual information processing Adapts to changing situations with context depended operation
(Belief-Desires-Intentions)
Data processing modules Communication with other agents Data I/O connections
Connecion directly to process automation measurements
Connection to other databases Data processing
Reasoning and logic Mathematic processing
Data storage modules Temporal storage of data
DIS 2006 15.6.2006 Helsinki University of Technology
User configurability supervision of process
How human users could define these tasks? Should be easy to desing, configure and maintain Framework, architecture etc. is needed
This solution should be as flexible as SQL servers are in data management area
But SQL servers as such can’t be used because Monitoring in automation is a task or a process, it’s not just QUERY ! Some information is ready to be access directly, but quite often we
need to wait for triggering events Data is stored in mixed formats (semantics)
What if we use the previously presented agent architecture? Let see how it could work in typical process related monitoring task
DIS 2006 15.6.2006 Helsinki University of Technology
Temporal monitoring scenarioProcess related problem explained
Problem: There are slowly drifting measurements in the processes Sensors that become dirty, etc.
Alarms provided by process automation system are not helping, because Limits are often set too wide to minimize false alarms Limits are set normally to check just one value at the time Control loops are compensating the error
User need a tool to monitor a combination of process quantities This kind of monitoring would be especially useful temporally e.g., startups and shutdowns of a process, or when there is a change in product
”for next 15 minutes make sure that pump is running and level in the tank is rising”
Solution proposal: fault may be found by gross checking process quantities User defines a set of constraints that agents are then set to monitor e.g., check that Measurement1 is always more than Measurement2
DIS 2006 15.6.2006 Helsinki University of Technology
Temporal monitoring scenarioOperation of agents
CA = Clint Agent Provides user interface and lets
user define constraints IA = Information Agent
Performs the monitoring task Divides the set of constraints
based on the initial values Delegates monitoring to peers Reports to the user when
needed PA1 = Process Agent 1
Knows Meas1 value PA2 = Process Agent 2
Knows Meas2 value DF = Directory Facilitator
Yellow page services Knows what agent is providing
what information
DIS 2006 15.6.2006 Helsinki University of Technology
Temporal monitoring scenarioActive monitoring of agents
Normally the situation is OK in the beginning so changes in measured values need to be monitored
Process agents monitor their own values and report when their partial constraints are broken
When partial constraints are broken the Information agent tests the set of constraints again
If constraints are ok the monitoring continues with updated values
If some constraint of the overall set of constraints is broken, then the user is informed
DIS 2006 15.6.2006 Helsinki University of Technology
Conclusions and Future Work Users within process automation need powerful tools to search, access,
and supervise information produced in all around the factory And this system should be easy to setup, configure, and maintain
Enabling technologies seems to be available Ontologies and semantic web tools “solves” interoperability issues Agents for proactive operation - monitoring on behalf of human user
Combining architecture - Semantic Automation
Future Work Testing on more complex monitoring tasks Test Scenarios motivated by real life problems and situated in real factories Use semantic web tools to provide user configurability
e.g., SPARQL (Query language for RDF/OWL from W3C) OWL for plant models, etc.
Internal operational principles are still much open
Consentrating more on the user viewpoint What services are really needed by the user? How user would like to configure and use various services?