A Mult-Agent Based Approach to Dynamc Schedulng of Machnes and Automated Guded Vehcles (AGV) n Manufacturng Systems by Consderng AGV Breakdowns Abstract Vaht Kaplanoğlu Unversty of Gazantep kaplanoglu@gantep.edu.tr Cenk Şah Cukurova Unversty sahn.cenk@gmal.com Adl Baykasoğlu Dokuz Eylul Unversty adl.baykasoglu@deu.edu.tr Rızvan Erol Cukurova Unversty rerol@cu.edu.tr Alper Eknc Unversty of Gazantep alperekncae@gmal.com Melek Demrtaş Cukurova Unversty demrtasm@cu.edu.tr In a compettve busness envronment, producng goods on tme plays a very mportant role. In addton to regular control complextes n manufacturng envronments, some unforeseen techncal problems may affect the effcency of producton. The breakdown of automated guded vehcles (AGV) durng manufacturng s one of these problems. Ths problem generally requres an nstantaneous soluton whle the system s operatng. However, tradtonal producton control systems and algorthms handle ths knd of problem centrally and usually are not able to provde effectve solutons promptly. Ths paper proposes a multagent based schedulng approach for AGVs and machnes wthn a manufacturng system where the AGV breakdowns are consdered. After mplementaton, ths approach s desgned to work under a real-tme manufacturng envronment and feasble schedules are supposed to emerge from negotaton/bddng mechansms between agents. Introducton Producng goods on tme plays a very mportant role n manufacturng control and plannng. Producton plans and schedules are generally nterrupted wth unexpected events around or
wthn the system. These problems may affect the effcency of producton plannng or they may collapse all the plans of operatons. The breakdown of automated guded vehcles (AGV) n flexble manufacturng systems s one of those problems. AGV systems are ndustral transportaton systems used n varous ndustral contexts: contaner termnals, part transportaton n heavy ndustry, and manufacturng systems [1-3]. They have consderable functonalty n manufacturng systems and contaner termnals. They may be the source of unexpected events wthn a manufacturng or logstcs system. The operatonal decsons of AGVs especally attracted researchers to desgn and mplement cost-effectve operatng decsons. However, the complexty of the problem has led the researchers to use dstrbuted methods other than central optmzaton approaches. Multagent-based systems, a newly maturng area of dstrbuted artfcal ntellgence, provde effectve mechansms for the management of such dynamc operatons n manufacturng envronments. As s expected from a farly young area of research, there s not yet an unversal consensus on the defnton of an agent [4]. However, the Wooldrdge and Jennngs defnton s ncreasngly adopted n ths feld: An agent s a computer system that s stuated n some envronment, and that s capable of autonomous acton n ths envronment n order to meet ts desgn objectves [5]. An agent s a component that can exhbt reasonng behavor under both proactve (goal-drected) and reactve (event-drven) stmul. When an agent s nstantated, t wll wat untl t s gven a goal to acheve or experences an event that requres a response [6]. Some of the authors of ths paper have prevously addressed a mult-agent based smultaneous AGV and machne schedulng approxmaton and tested t on test-bed problems [7]. Mult-agent based approxmaton has proven ts success n dynamc and volatle busness envronments. However, AGV breakdown occurrences are not consdered n ther prevous studes. The AGVs are assumed as beng operatonal wthout breakng down throughout the entre manufacturng process. In ths paper, the breakdowns of AGVs are consdered to extend the scope of the prevous studes. The ntenton of ths study s to get closer to real manufacturng envronments. Ths paper s organzed as follows: a bref lterature survey about the studes that consder AGV breakdown, the mult-agent based desgn of the proposed problem, the algorthms that the agent types uses under AGV breakdown condton and a summary wth future research possbltes. Lterature Revew The studes about AGV control have a wde scope n the lterature and range from traffc control on the AGV paths to AGV deadlock preventon [8, 9]. The applcaton areas range from manufacturng floors to contaner termnals [7, 10]. The soluton approxmaton for AGV control s also encompasses a wde research doman, from nteger programmng to meta-heurstcs and from Petr-net to mult-agent systems [7, 8, 10-13]. However, the
lterature revew n ths paper focuses on the AGV breakdown durng the real-tme manufacturng operatons. To the authors knowledge, there are few studes n ths area. AGV falures are neglected n most lterature on automated transportaton systems. Accordng to Mark Ebben (2001), when an AGV break downs, t may stop any other AGVs. There are two optons when the AGV breaks down: t can be fxed on the system or removed from the system to the repar secton. Choce depends on repar tme [14]. Taghabon and Tanchoco (1995) noted that routng flexblty allows a quck recovery to breakdowns and other dsruptve events, but ther study does not examne falures. Accordng to ther study, falures can be neglected n AGV systems when the AGV workload s low and falures can be resolved quckly [15]. Another study about AGV control that consders dsturbances s by Badr et al. (2010). They presented fve steps to clarfy the dsturbance handlng durng dynamc schedulng: dsturbance detecton, dsturbance analyss, acton selectng, acton announcement, and schedule repar [16]. Merdan et al. (2013) proposed an approxmaton for conveyor and machne falures n workflow schedulng by usng a mult-agent system. They tested dspatchng rules n combnaton wth the all re-routng re-schedulng polces under machne and conveyor falures. They then ranked the rules based on ther performance results at the smulaton [17]. Desgn of AGV Resource Agent durng Breakdown In ths study, AGV breakdown stuaton s modeled under a mult-agent based system approach. The proposed s desgned by usng the Prometheus methodology that defnes a detaled process for specfyng, desgnng, mplementng, and testng/debuggng agentorented software systems. Ths methodology s developed for specfyng and desgnng agent-orented software systems, and t s general purpose n that t s not ted to any specfc software platform. Prometheus dstngushes tself from other methods by supportng the development of ntellgent agents, provdng start-to-end support, evolvng out of practcal ndustral and pedagogcal experence, beng used n both ndustry and academa, and, above all, beng detaled and complete [4]. Fgure 1 presents the phases of Prometheus desgn methodology.
Fgure 1. The phases of Prometheus methodology [4] System Specfcaton The agent types are decded and desgned through ths desgn methodology stages. The followng are the agent types n the proposed system. Machne Resource Agent, Machne Scheduler Agents, AGV Resource Agent, AGV Scheduler Agents, and Operaton Agent In the system specfcaton stage of Prometheus, negotatons between agent types, system goals, agent roles n the system, and scenaros are dentfed. Fgure 2 shows the system specfcaton stage of Prometheus methodology. There are four man roles n the system: AGV management, machne management, system management, and negotaton management.
Fgure 2. System roles n PDT Ths study focuses on AGV management role n system specfcaton stage. The crcles n Fgure 2 show the goals of system elements. One of the goals of the AGV management role of the proposed system s AGV Schedulng after AGV Breakdown (see Fgure 2). The subgoal s also desgned n the system specfcaton stage. Three subgoals of the AGV Schedulng after AGV Breakdown goal are gven n Fgure 3: 1. AGV that s loaded and has a task n ts blackboard 2. AGV that s free and has a task n ts blackboard 3, AGV that s loaded and has no task n ts blackboard Fgure 3. Subgoals of AGV Schedulng after AGV Breakdown goal
Archtectural Desgn The negotaton protocols between agent types are desgned n ths stage of Prometheus methodology. A system overvew dagram s gven n Fgure 4. The AGV scheduler agent negotates wth operaton agents n order to fnd real-tme operaton transportaton and processng schedule. Fgure 4. System overvew Fgure 4 also shows an example negotaton protocol between operaton agents and scheduler agents. When an operaton agent enters to proposed mult-agent based system then t calls for proposals for the machne and scheduler agent that are avalable n the system. When the order agent fnds a proper machne agent to be processed, t then calls for a proposal to a scheduler agent to be transported to the machne. Detaled Desgn In detaled desgn stage, the capabltes of the scheduler agent type are defned by the breakdown condton. A resource agent would be n any followng states n a flexble manufacturng system: 1. Idle and ready 2. Transportaton of an operaton 3. Deadheadng trp (gong to take a job from machne)
Whle AGV resource agent s operatng, t can break down. The AGV resource agent has an attrbute of workng status of ether n workng condton or broken down. Status changes to broken down from n workng condton when t breaks down. In all three states, the resource agent updates ts status attrbute. The resource agent sends the breakdown nformaton to the scheduler agent after updatng ts attrbute. Fgure 5 shows the detaled desgn for the resource agent. Fgure 5: Detaled desgn of AGV resource agent Fgure 6 shows the negotaton protocol of resource agent and scheduler agents. Fgure 6. Negotaton protocol of AGV resource agent and AGV scheduler agent When the scheduler agent receves the breakdown message, t reasons n one of three ways by controllng the blackboard. Fgure 7 shows a detaled desgn for the scheduler agent.
Fgure 7. AGV scheduler agent wth detals When the scheduler agent takes the breakdown message from the resource agent, t sends the message to the operaton agents n ts blackboard, whch then start a new negotaton wth the scheduler agents n order to be transported. Fgure 8 shows the standard negotaton protocol between operaton and scheduler agent. Fgure 8. Negotaton of operaton agent and AGV scheduler agents
Algorthm for AGV Breakdown Condton Ths secton detals the scheduler agent s decson-makng. The operaton agent nforms the scheduler agents when the AGV broke down. The scheduler agent then assesses the coordnaton nformaton nsde the messages and performs a reward. Scheduler agents consder the proposal of machne operatons broken-down AGV accordng to the equaton 1. After the AGV breaks down and the blackboard resets, the current tme must be equal to the earlest pckup tme of operaton : t = EPT, so ELT = { t + t( CL, AGVBDP )}, =1 n (1) In the frst equaton, ELT denotes earlest loadng tme of operaton, CL s current locaton of AGV resource agent, AGVBDP s AGV s breakdown pont for operaton, t s current tme, t(.,.) s the requred tme between two locatons, and EPT s the earlest pckup tme of operaton. Scheduler agents evaluate the proposal accordng to followng equaton: t + t( CL, PCP ), ELT =, =1 n (2) t + max{ t( CL, PCP ),( EPT t) }, t EPT, t > EPT, In the second equaton, ELT denotes earlest loadng tme of operaton, CL s current locaton of AGV resource agent, PCP s pck up pont of operaton, t s current tme, t(.,.) s the requred tme between two locatons and EPT s the earlest pckup tme of operaton. Then, an operaton s selected from the AGV blackboard by usng the followng equaton: ELT s = mn{ ELT }, =1 n (3) The scheduler agent then proposes a tme to respectve operaton agents by addng ELTs to the related loaded trp tme as shown n equaton 4: PR=ELT s + t(pcp s, DP s ) (4) After the start of the negotatons, operaton agents call to all scheduler agents to submt a proposal. Ths plan frst checks whether an operaton has already been rewarded. If there s not a rewarded operaton, then t prepares an offer. When preparng a proposal, the scheduler agent fnds the operaton that has the mnmum ELT usng equatons 5 and 6, where EFT and NL denotes earlest free tme and the next locaton of the AGV resource agent, respectvely. EFT + t( NL, PCP ), EFT > EPT ELT = EFT + max{ t( NL, PCP ), ( EPT EFT )}, EFT EPT = 1 n (5) ELT s = mn {ELT }, =1 n (6)
If the operaton n the current negotaton matches the selected operaton n the scheduler agent s blackboard belef set, the scheduler agent proposes operatons by addng ELTs the related loaded trp tme, as n equaton 7: PR=ELT s + t(pcp s, DP s ) (7) Concluson and Future Research Resources that are used n flexble manufacturng systems pose unforeseen techncal problems n addton to regular control and mantenance complextes. The breakdown of AGVs durng real-tme manufacturng affects many related schedules of operatons and machnes. Ths problem generally requres an nstantaneous soluton whle the system s operatng. The proposed mult-agent based desgn s developed n order to solve these complextes durng the manufacturng process. The desgn uses the capabltes of multagent systems n order to solve real-tme schedulng complextes. Feasble and effectve schedules are supposed to be emerged from negotaton/bddng mechansms between agents. Future research drectons nclude Implementng the proposed desgn on a mult-agent programmng language. Fndng test-bed studes n order to compare the results of mult-agent systems wth other approxmatons. Developng mult-agent based smulaton models n order to test the effectveness of the proposed model Acknowledgment the present study s supported by The Scentfc and Technologcal Research Councl of Turkey (TUBITAK) (Grant number: 111M279). References [1] Le-Anh, T., & De Koster, M. B. M. (2006). A Revew of Desgn and Control of Automated Guded Vehcle Systems. European Journal of Operatonal Research, 171(1), 1-23. [2] Ln, L., Shnn, S. W., Gen, M., & Hwang, H. (2006). Network Model and Effectve Evolutonary Approach for AGV Dspatchng n Manufacturng System. Journal of Intellgent Manufacturng, 17(4), 465-77. [3] Vs, I. F. A. (2006). Survey of Research n the Desgn and Ccontrol of Automated Guded Vehcle Systems. European Journal of Operatonal Research, 170(3), 677-709. [4] Padgham, L., & Wnkoff, M. (2004). Developng Intellgent Agent Systems: A Practcal Gude. Melbourne, Australa: John Wley & Sons. [5] Wooldrdge, M., & Jennngs, N. R. (1995). Intellgent Agents: Theory and Practce. The Knowledge Engneerng Revew, 10(2), 115-52. [6] Hahn, C., Madrgal-Mora, C., & Fscher, K. (2009). A Platform-Independent Metamodel for Multagent Systems. Autonomous Agents and Mult-Agent Systems, 18(2), 239-66.
[7] Erol, R., Sahn, C., Baykasoglu, A., & Kaplanoglu, V. (2012). A Mult-Agent Based Approach to Dynamc Schedulng of Machnes and Automated Guded Vehcles n Manufacturng Systems. Appled Soft Computng, 12(6), 1720-32. [8] Herrero-Perez, D., & Martnez-Barbera, H. (2011). Decentralzed Traffc Control for Non-Holonomc Flexble Automated Guded Vehcles n Industral Envronments. Advanced Robotcs, 25(6-7), 739-63. [9] Km, K. H., Jeon, S. M., & Ryu, K. R. (2006). DeadlockPpreventon for Automated Guded Vehcles n Automated Contaner Termnals. OR Spectrum, 28(4), 659-79. [10] Gelareh S, Merzouk R, McGnley K, Murray R. (2013). Schedulng of Intellgent and Autonomous Vehcles under Parng/Unparng Collaboraton Strategy n Contaner Termnals. Transport Research Part C: Emergng Technologes, 33, 1-21. [11] Farahan, R. Z., Pourakbar, M., Mandoabch, E. (2007). Developng Exact and Tabu Search Algorthms for Smultaneously Determnng AGV Loop and P/D Statons n Sngle Loop Systems. Internatonal Journal of Producton Research, 45(22), 5199-222. [12] Wallace, A. (2001). Applcaton of AI to AGV Control-Agent Control of AGVs. Internatonal Journal of Producton Research, 39(4), 709-26. [13] Rajota, S., Shanker, K., & Batra, J. L. (1998). An Heurstc forconfgurng a Mxed Un/Bdrectonal Flow Path for an AGV System. Internatonal Journal of Producton Research, 36(7), 1779-99. [14] Ebben, M. J. R. (2001). Logstc Control n Automated Transportaton Networks. Enschede, The Netherlands: Unverstet Twente. [15] Taghabon-Dutta, F., & Tanchoco, J. (1995). Comparson of Dynamc Routng ttechnques for Automated Guded Vehcle System. Internatonal Journal of Producton Research, 33(10), 2653-69. [16] Badr, I., Schmtt, F., & Gohner, P. (2010) Integratng Transportaton Schedulng wth Producton Schedulng for FMS: An Agent-Based Approach. IEEE Internatonal Symposum on Industral Electroncs, 3539-44. [17] Merdan, M., Moser, T., Sunndyo, W., Bffl. S., & Vrba, P. (2013). Workflow Schedulng Usng Mult-Agent Systems n a Dynamcally Changng Envronment. Journal of Smulato, 7(3), 144-58. Bographes VAHIT KAPLANOGLU receved hs B.Sc. degree n Industral Engneerng from Unversty of Marmara, Department of Industral Engneerng n 2004. He receved hs M.Sc. and Ph.D. degrees n Industral Engneerng from Unversty of Gazantep, Department of Industral Engneerng n 2007 and 2011. He s studyng at the Unversty of Gazantep as an assstant professor. Hs research nterests are logstcs and supply chan management, mult-agent systems and dstrbuted artfcal ntellgence. CENK SAHIN receved hs B.Sc., M.Sc. and Ph.D. degrees n Industral Engneerng from Cukurova Unversty, Department of Industral Engneerng n 2001, 2004, and 2010 respectvely. He s studyng at the Cukurova Unversty as an assstant professor. Hs research nterests are producton schedulng, artfcal neural networks, logstcs and supply chan management and mult-agent systems.
ADIL BAYKASOGLU receved hs B.Sc., M.Sc. and Ph.D. degrees from Mechancal and Industral Engneerng areas n Turkey and UK. From 1993 to 1996, he was wth the Department of Mechancal Engneerng and Industral Engneerng at the Unversty of Gazantep, frst as a research assstant, later as an nstructor. He s presently a full professor n the Industral Engneerng Department at the Dokuz Eylul Unversty. He has publshed more than 260 academc papers, 3 books and edted several conference books on operatonal research, computatonal ntellgence, fuzzy logc, qualty, and manufacturng systems desgn. He s also edtor of Turksh Journal of Fuzzy Systems and servng on the board of several academc journals. RIZVAN EROL receved hs Ph.D. degree n Industral and Management Systems Engneerng from Arzona State Unversty n 1996. He s currently professor and charperson of Industral Engneerng at Cukurova Unversty n Turkey. Hs man research areas are appled operatons research, supply chan management, and health systems. ALPER EKINCI receved hs BSc degree n Industral Engneerng from the Unversty of Gazantep n 2012. He s currently studyng towards the Master of Scence Workflow Schedulng Usng Mult-Agent Systems n a Dynamcally Changng Envronment under the supervson of Vaht Kaplanoğlu. Hs research nterests nclude artfcal ntellgence and ts applcatons n mult-agent based machne schedulng. MELEK DEMIRTAS receved her B.Sc. and M.Sc. degrees n Industral Engneerng from Cukurova Unversty, Department of Industral Engneerng n 2010 and 2013 respectvely. She s currently studyng towards her Ph.D. under the supervson of Cenk Sahn. Her research nterests nclude mult-agent based machne schedulng and supply chan management.