sistemas multi-agentes no contexto da indústria 4eol/ssiim/1718/seminars/l08 (pl...10/11/2017 2...
TRANSCRIPT
10/11/2017
1
Sistemas Multi-agentesno contexto da Indústria 4.0
Paulo Leitã[email protected]
http://www.ipb.pt/~pleitao
Seminário de Sistemas Inteligentes, Interacção e Multimedia
Porto, 9 de novembro de 2017
Paulo Leitão, 2
Agenda
Indústria 4.0
Sistemas multi-agente (MAS) como uma solução para desenvolver sistemas ciber-físicos (CPS)
Discussão de aplicações de MAS na indústria e em projetos R&D
Análise das barreiras para uma maior adoção de MAS pela indústria e desafios futuros
10/11/2017
2
Paulo Leitão, 3
Agenda
Indústria 4.0
Paulo Leitão, 4
Evolution of complexity
Spirit of St. Louis,National Air and Space Museum, Smithsonian Institution
Airbus A380
10/11/2017
3
Paulo Leitão, 5
Complexity in engineering problems
Processes plants
Logistics
Smart grids
Taxis fleet
Manufacturing plants
Airport management
Paulo Leitão, 6
New demands in manufacturing
1
Markets are imposing strong changing conditions
Customization of products in high flexible production
Reduction of time to reconfigure
(usually weeks and months)
Plug and produce
Time on market / Time to market
Tesla’s robotic factory in Fremont, California
10/11/2017
4
Paulo Leitão, 7
Moving to decentralized structures
Traditionally: centralized and monolithic structures
Production processes
Sensors / actuators
Control
SCADA
MES
ERPANS/ISA
95
Challenge: decentralize and distribute functions
Paulo Leitão, 8
Industrie 4.0
• Digitization of traditional factories, aiming:
– Modernization of industry (intelligent factories)
– More productive, efficient, flexible, adaptive reliable
globally more competitive!
• Based on:
– CPS systems
– Using emergent technologies, particularly IoT and IoS
10/11/2017
5
Paulo Leitão, 9
Key Design Principles
• Decentralization
• Connectivity and interoperability
• Service orientation
• Optimized and real-time decision-making
• Modularity
• Virtualization (modelling and simulation)
• Human-machine integration
Paulo Leitão, 10
Disruptive technologies enablingdigitization
10/11/2017
6
Paulo Leitão, 11
These concepts around the world …
Paulo Leitão, 12
Fourth industrial revolution
1st: mechanization of production using water
and steam power
2nd: mass production with the help of electric power
3rd: use of electronics and IT to further
automate production
4th: ICT, IoT, CPS, Self-*
1780 - 1830
1900 - 1930
1955 - 1970
2013 -
10/11/2017
7
Paulo Leitão, 13
Why is seen as an revolution?
• Challenges that force a disruptive technological change!
• Examples:
– Big data analysis
o Huge amount of data to be collected and processed in real-time
– Additive manufacturing
o More complex parts, faster and with more quality
– Collaborative robots
o Already available for small/medium sizes but what happen for robots with higher speed and payload? (example 5 m/s and 1ton)
Paulo Leitão, 14
Industrial Internet
• GE, Intel, IBM, Cisco, US Government, …
• Integration of complex physical machinery with networked sensors and software
• Broader than industrial production, e.g., also smart electrical grids, transportation
• Several topics to process data from machines
– E.g., machine learning, big data, Internet of things, M2M communication
• Analyze collected data (often in real-time) and dynamically adjust operations
10/11/2017
8
Paulo Leitão, 15
Waves of innovation and change
Source: Peter C. Evans and Marco Annunziata, «Industrial Internet: Pushing the Boundaries of Minds and Machines”, General Electric, November 26, 2012
Paulo Leitão, 16
Cyber-physical systems (CPS)
• Concept introduced in 2006
• Featuring:– Tight combination of
computational and physical elements
– Distributed organized in a network
– Interaction each other to achieve a common goal
• Intelligent, dynamic and self-* large-scale systems, covering inter- and intra-enterprise integration
10/11/2017
9
Paulo Leitão, 17
CPS vs ES and CPS vs IoT
• CPS vs embedded systems (ES)
– ES: computational elements hosted in stand-alone devices
– CPS: network of interacting computational and physical devices
• CPS vs IoT
– IoT : focus is in the interconnection of cooperative objects
– CPS: also considers the computational decisional components to provide intelligence, responsiveness and adaptation
Paulo Leitão, 18
CPS in practice
Cyber-physical components
DB Intelligent, self-contained and plugable SW
modules
Intelligent, self-contained and plugable HW
modules
global self-adaptation and
self-optimization
self-learning
10/11/2017
10
Paulo Leitão, 19
Application domains
Smart buildings
Smart electrical grids
Smart transportation and mobility
Smart manufacturingSmart healthcare
Defense
Paulo Leitão, 20
Challenges
• Way to realize Industrie 4.0 and Industrial Internet
• “CPS are now a national priority for federal R&D”President’s Council of Advisors on Science and
Technology (PCAST) in a 2007 report
• Novelty is not in new technology but in the way to combine existing technologies!
• Heterogeneous HW and SW technologies
• Interconnection with legacy systems and HW devices
• Such systems can be difficult and costly to design, test and maintain
10/11/2017
11
Paulo Leitão, 21
Agenda
Sistemas multi-agente (MAS) como uma solução para desenvolver sistemas ciber-físicos (CPS)
Paulo Leitão, 22
MAS for distributed intelligence
• Agents to introduce intelligence and adaptation
• Agents are not a restyling of objects!
inter-agent communication
decision
physical interface
An autonomous component that
represents physical or logical objects,
capable to act in order to achieve its
goals, and able to interact with other
agents to reach its objectives.
10/11/2017
12
Paulo Leitão, 23
MAS paradigm
• Rare applications consider agents in an isolated manner multi-agent systems!
Network of intelligent agents …,
capable to interact to reach their
individual goals when they haven’t
enough knowledge/skills
• “Intelligent” behavior emerges from the interaction among distributed agents
• Characteristics: modularity, flexibility, robustness, adaptability and re-configurability
Paulo Leitão, 24
MAS working in practice
visão local
Comportamento local
10/11/2017
13
Paulo Leitão, 25
What multi-agents can offer
Reusabilityold components can be re-used to develop new components or new systems
Distributed thinkinga complex problem can be divided into several small problems
Modularitybuilding the system by
pieces like using LEGO
Robustnesslosing one decision node doesn’t implies the system failure
Reconfigurabilitychanges can be performed on the fly
Smooth migrationfrom old technologies to new ones
Paulo Leitão, 26
Need of MAS in practice
• System comprising a sequence of modular conveyors
• Individual conveyor comprises:
- 1 motor
- 2 light sensors
How to implement the control system?
S0
S1
C1 C2
- C1 only stops when the part arrives to S1
- C2 starts when the part arrives to S0
• Transfer a part from an input to an output location
10/11/2017
14
Paulo Leitão, 27
Using the traditional solution
• Use a centralized logical control approach
• Programmed using IEC 61131-3 running in a PLC
• Advantages:
– Simple to program
– Industrial adopted
• Disadvantages:
– Lack in supporting scalability and re-configuration of the conveyor system
– Interdependencies between conveyors increases development effort and time!
Paulo Leitão, 28
Particularly, what happen if we need to …
• Switch the order of the conveyors?
We need an alternative design approach to support the easy reconfiguration on-the-fly!!!
A B C DBC
• Add a new conveyor?
10/11/2017
15
Paulo Leitão, 29
Using the MAS solution
• Create cyber-physical components
Source: J. Barbosa, P. Leitão, J. Teixeira (2017), “Empowering a Cyber-Physical System for a Modular Conveyor System with Self-organization”, Proceedings 7th Workshop on Service Orientation in Holonic and Multi-agent Manufacturing (SOHOMA'17).
Paulo Leitão, 30
Intelligent MAS solution
Each agent has a modular and similar logical control
Instantiate as many agents as modular conveyors
Agents interact among them to get sequence knowledge
Plugability and reconfiguration on-the-fly!
video
10/11/2017
16
Paulo Leitão, 31
Combining with IEC 61131-3
• MAS usually misses real-time constraints
• Preserve low-level control to ensure responsiveness
How to standardize the
interface?
Paulo Leitão, 32
IEEE P2660.1 Working Group
• Recommended practices to solve the interface problem
– integrating software agents with low-level real-time automation control devices
– in the context of cyber-physical systems
– allowing the reuse and transparency
• Contributions are welcome!
10/11/2017
17
Paulo Leitão, 33
Integration of agents and SOA
• SOA have potential to solve interoperability problems
• New design approach:
Agents: implement control functions
Services: encapsulate functionalities provided by agents
• Similarity with the LEGO concept
Paulo Leitão, 34
Agents distributed among different levels
Cloud level to make accurate and
optimized analysis
Edge and fog levels to make a fast (real-
time) analysis
10/11/2017
18
Paulo Leitão, 35
The role of edge vs cloud
• Processing as close to source as possible
• Real-time monitoring and decision-making
• Fast response to condition change
• Data pre-processing and filtering
Cloud layer Edge layer
• Central data aggregation
• Long-term data processing
• Historical analysis of data
• Optimization and prediction
Paulo Leitão, 36
Agenda
Discussão de aplicações de MAS na indústria e em projetos R&D
10/11/2017
19
Paulo Leitão, 37
MAS applications in industrial domain
1995 20152005 20102000
EXPLANTECH
FABMAS
CambridgePacking Cell
ABAS
AgentSteel
Air cateringscheduling
NovaFlex
SemanticMAST
ADACORFMS AgentFly
ChilledWaterSystem
Axion-Holding scheduling
Ciudad Real
Airport
Source: P. Leitão, V. Marik, P. Vrba (2013), “Past, Present, and Future of Industrial Agent Applications”, IEEE Transactions on Industrial Informatics, vol. 9, n. 4, pp. 2360-2372 (DOI: 10.1109/TII.2012.2222034)..
Paulo Leitão, 38
Objectives
Partners:
• Integration of quality control and process control in real-time
• Use of a distributed MAS infrastructure to support feedback control loops
10/11/2017
20
Paulo Leitão, 39
• Preserving existing low-level control
• Distributed collection of data in real-time
• Correlation of data in real-time using data analysis
Functionalities
Paulo Leitão, 40
• MAS solution developed using JADE
• Installed in the Whirlpool’s factory plant producing washing machines
• 11 QCAs and 6 RAs were running in PCs distributed along the production line
• In stable production flow, approximately 400 PAs are running simultaneously
Deployment numbers
10/11/2017
21
Paulo Leitão, 41
Examples of intelligence” for data analysis
grace-demo
• Early detection of non-conformities
Customization of functional tests according to the on-line gathered production data
Customization of the controller’s parameters of each machine considering the production data
Early detection of products that never reach desired quality
Dynamic adaptation of process parameters considering data gathered from quality control
Paulo Leitão, 42
Main benefits
• Increase of production efficiency
• Reduction of non-conformities
• Increase of products’ quality
• Reduction of costs related to scraps
• Increase of products’ customization
Source: P. Leitão, N. Rodrigues, C. Turrin, A. Pagani (2015), “Multi-agent System Integrating Process and Quality Control in a Factory Producing Laundry Washing Machines”, IEEE Transactions on Industrial Informatics, vol. 11, n. 4, pp. 879 - 886 (DOI: 10.1109/TII.2015.2431232).
10/11/2017
22
Paulo Leitão, 43
• Mitigation strategies to respond faster to unexpected events in ramp-up production of complex and highly customized products
• ESB integrating heterogeneous SOA and knowledge-based MAS applications
Objectives
Paulo Leitão, 44
• Integration of strategic planning and operational scheduling systems
• Real-time data analysis, namely to detect deviations, tendencies, etc.
• “What-if” simulation to support strategic decisions
Functionalities
Source: P. Leitão, N. Rodrigues and J. Barbosa (2015), “What-if Game Simulation in Agent-based Strategic Production Planners”, Proceedings of the 20th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA'15), Luxembourg, 8-11 September, pp. 1-8 (DOI: 10.1109/ETFA.2015.7301438).
10/11/2017
23
Paulo Leitão, 45
Strategic Planner
Operational Scheduler
Scenario Designer
Manufacturing Incident Detection
Agent Solution
iESB (Intelligent Enterprise Service Bus)
Life-cycle Management
Data Transformation
Node Management
Sniffer
ARUM database
(tripe store)
Ontology service
Legacy data
sources SC
AD
A
ER
P
ME
S
GatewaysGateways
Gateways
Architecture
Source: C. Marín, L. Mönch, P. Leitão, P. Vrba, D. Kazanskaia, V. Chepegin, L. Liu, N. Mehandjiev (2013), “A Conceptual Architecture Based on Intelligent Services for Manufacturing Support Systems”, Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics (SMC’13), pp. 4749-4754 (DOI: 10.1109/SMC.2013.808).
Paulo Leitão, 46
• Main benefits:– Reduction of the ramp-up time
– Early detection of events and failures
– Improvement of the adaptation to the condition changes
– Improvement in strategic decision-making facing unpredictability
arum-demo
10/11/2017
24
Paulo Leitão, 47
Consortium
Paulo Leitão, 48
Agenda
Análise das barreiras para uma maior adoção de MAS pela indústria e desafios futuros
10/11/2017
25
Paulo Leitão, 49
Road-blockers
investment
distributed thinking interoperability
scalability standardization real-time constraints
integration with physical devices supporting
technologies and methodologies
Paulo Leitão, 50
Current trends
• Convincement of industry stakeholders of the benefits of this approach
• Compliance with industrial standards
• Technical issues (relevant in industrial systems): interoperability, resilience, scalability, security and privacy
• Integration of humans (in-the-loop and in-the-mesh)
• Smooth migration from existing running systems
10/11/2017
26
Paulo Leitão, 51
In a short …
• CPS assumes a key role to realize Industrie 4.0
• MAS is a suitable approach to implement the distributed intelligence in CPS
• MAS is being applied in industrial environments through several European R&D projects
• Important notes:
– Hide complexity, reveal functionality
– Think globally, act locally
• A long and hard path is needed to face several challenges and open questions
Paulo Leitão, 52
Some open questions for research
How to properly design SOMAS systems?
How hierarchies are dynamically formed, evolved and removed?
How optimization is achieved in decentralized systems?
How to control nervousness in self-organized systems?
How to ensure the emergence of desirable behaviours?
And so many others …