270 likes | 283 Views
Case Studies of MAS in Industrial Production: AARIA & The M. Fox Scheduling System. CS525M, Spring 2002 Presented by: Edwin P. Jacques. AARIA Introduction.
E N D
Case Studies of MAS in Industrial Production: AARIA & The M. Fox Scheduling System CS525M, Spring 2002 Presented by: Edwin P. Jacques
AARIA Introduction • Vision: Interconnect manufacturing processes implemented as agents to outperform current, centrally-controlled manufacturing systems. This is done by allowing quick reconfiguration to optimize profits and allow customization in a dynamic business climate. • AARIA • Autonomous Agents at Rock Island Arsenal • Agent research conducted at University of Cincinnati. • Rock Island Arsenal is a military production facility used to demonstrate results of the research.
AARIA: What Does It Do? • Dialog with customers and suppliers • Allocate resources to new jobs as they are introduced into system • Optimize schedules across resources • Recover from faults in the factory • Dispatch work according to a schedule • Report Results • Designed to be used in real life or simulation
AARIA: How is the system decomposed? • Persistent agents • Change little over time. Do not enter or leave the system without human involvement. • Resource broker agentsmanage capacity-constrained resources, e.g.– people, machines & facilities • Part broker agent manage material handling and inventory • Unit-process broker agents build products from parts • Transient agents • Rate of change comparable to part processing times. • For example, materials produced and sold by part broker go through states of birth, inquiry, commitment, availability, activity & death.
AARIA: How many agents are there? • Later stages of production are customers to earlier stages of production. • Stages can be added to simulate a supply chain of any depth. Each stage contains part, unit processor and resource brokers. • Final customer is viewed as just another customer in the supply chain.
AARIA: What is inside the agents? • AARIA is focused on agent organization, not implementation of specific agents. • Agents are viewed simply as problem-solvers. • The internal implementation is borrowed from agent infrastructure software, Cybele. e.g.– including discovery of trading partners & negotiation protocols.
AARIA: Agent Negotiations • Architecture calls for agent negotiation along axes of possible production (e.g.- price, quality, delivery time, product features & speed of answers) • Prototype only considers price and delivery time. • Sequence: • Customer orders product • Available Unit Process Brokers receive order and solicit bids from part brokers. • Continues up to end of the chain, and works its way back down as bids are returned. • If customer accepts bid, commitments are made in a similar manner.
AARIA: Scheduling • Problem: • Determining if agent can perform a task based on parameters (e.g.– time & cost) is not hard. • Determining optimal sequence in which to arrange tasks in all agents throughout system is NP-complete problem. • Solution: • When order presented, a possible schedule is developed for every possible delivery time. • When schedule is accepted, it is optimized as a background task.
AARIA: Maturity • Performance: • Inventory costs reduced as much as 47% • Lead times cut as much as 59% • Schedule reducible time down 93% • Overtime, inventory holding charges, etc. • Deployment: • Prototyped. Capable of simulation. • Being productized under brand name eFactory. • The Army might be using it, but that’s privileged information.
AARIA: References • Baker, Albert D, et al. Agents and the Internet: Infrastructure for Mass Customization. IEE Internet Computing. September, 1999. • Baker, Albert D, et al. Internet-based Manufacturing: A Perspective from the AARIA Project. Working Paper, Enterprise Action Group, Cincinnati, Ohio. August, 1999. • http://www.aaria.uc.edu/
FOX Introduction • Vision • Agents are a natural way of decomposing a supply chain. • Develop an architecture for agent-oriented supply-chain management. • FOX Scheduling System • Generic building shell that implements generic, reusable components for communicative-act based communication, conversational coordination & role-based organization modeling.
FOX: What Does It Do? • Provide mechanism for distributing supply chain activities to agents. • Provide coordination among agents. • Communicates plans/requests. • Communicates changes in plans (delays) • Handling responsiveness constraints (solution quality improves if more time given). • Allow information within agents available to be made available to other interested agents.
FOX: What are the agents? • System provides a shell that can be used to implement any distribution of supply-chain intelligence. • Decomposition is often at management (organizational) level. • Typical decomposition: • Order acquisition agent: negotiates with customers about orders. • Logistics agent: Coordinates plants, suppliers and distribution centers to achieve most on-time deliver and cost minimization. • Transportation agent: Assignment and scheduling of transportation resources. • Scheduling agent: Scheduling and rescheduling activities in the factory. • Resource agent: Manages inventory and purchases, with goal of minimizing cost and maximizing delivery. • Dispatching agent: Performs the order release and real-time floor control as directed by scheduling agent. Operates within degree of freedom allowed by scheduling agent.
FOX: Agent Shell Overview • Communication Service allows information exchange: • Domain-independent speech acts • Domain-dependent content • COOrdination Language (COOL), based on the conversation metaphor. • Express shared conventions • Capture coordination and social knowledge • Generic models of actions and behavior. • Not well specified. No detailed information available on components at this level.
FOX: Communication Language • Based on KQML (propose ;; communicative action :sender A :receiver B :language list :content (or (produce 200 widgets) (produce 400 widgets)) :conversation C1 :intent (explore fabrication possibility)).
FOX: Conversation Plans • Rule-based descriptions of how an agent acts and reacts in certain situations. • Defines a state machine that provides skeleton for conversation. (def-conversation-plan 'customer-conversation :content-language 'list :speech-act-language 'kqml :initial-state 'start :final-states '(rejected failed satisfied) :control 'interactive-choice-control-ka :rules '((start cc-1) (proposed cc-13 cc-2) (working cc-5 cc-4 cc-3) (counterp cc-9 cc-8 cc-7 cc-6) (asked cc-10) (accepted cc-12 cc-11))).
FOX: Conversation Rules • Describe actions to be performed when conversation is in a given state. • This rule says that when logistics is in the start state, and a proposal for an order is received, tell the customer it’s being worked on. (def-conversation-rule 'lep-1 :current-state 'start :received '(propose :sender customer :content(customer-order :has-line-item ?li)) :next-state 'order-received :transmit '(tell :sender ?agent :receiver customer :content '(working on it) :conversation ?convn) :do '(update-var ?conv '?order ?message)).
FOX: Control Architecture • Each agent operates in a loop where: • Events are sensed (such as messages from other agents) • Current situation is evaluated, • updating or creating beliefs • updating or creating conversations • Entry is selected from agenda • If new conversation, retrieve plan and instantiate • If existing conversation, apply conversation rules
FOX: Learning • When entering new state • Special rules dynamically estimate how well system does in respect to certain criteria • e.g.– time, quality, etc. • Rules look at underachieving criteria and compute new global criterion to correct it. • e.g.– give time greater weight in planning process
FOX: Implementation Details • COOL specification contains: • 12 conversation plans • 200 rules and utility functions • 2600 lines of COOL code to describe (concise!) • In a three-tier supply chain example, • overstock of raw inventory in case of breakdown reduced 26%. • 7500 lines of COOL code (plus 2000 for GUI) • Written by one author with no CS background in 3 months.
FOX: References • Fox, Mark S. Agent-Oriented Supply-Chain Management. The International Journal of Flexible manufacturing Systems. December, 2000. • http://www.eil.utoronto.ca/iscm/papers/index.html
Appendix A: When will agents be implemented? (Trudeau, 1996)