- Email: [email protected]
Applying multi-agent technique in multi-section flexible manufacturing system
Expert Systems with Applications 37 (2010) 7310–7318
Contents lists available at ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
Applying multi-agent technique in multi-section flexible manufacturing system Kai-Ying Chen *, Chun-Jay Chen Department of Industrial Engineering and Management, National Taipei University of Technology, No. 1, Chunghsiao E. Rd., Sec. 3, Taipei 106, Taiwan, ROC
a r t i c l e Keywords: Multi-section FMS Multi-agent Dispatching JADE
i n f o
a b s t r a c t In the highly competitive market, cooperative multi-agent transaction and negotiation mechanism have become an important research topic. This paper uses multi-agent technology to construct a multi-section flexible manufacturing system (FMS) model, and utilizes simulation to build a manufacturing environment based on JADE framework for multi-agent to combine with dispatching rules, such as shortest imminent processing time (SIPT), first come first serve (FCFS) earliest due date (EDD), and Buffer Sequence. This paper finds that using multi-agent technique for multi-section FMS model can enhance the production efficiency in practice. Meanwhile, in this study, multi-agent systems combined with dynamic dispatching can be used to identify the best dispatching rules combination for achieving largest throughput, and thus it can provide the reference for production scheduling in advance. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction With the increasingly diversified market demands for small batch production, customized manufacturing mode must be employed so that the production system, production line and procedures can be adjusted flexibly to manufacture various kinds of products. Flexible manufacturing system (FMS) is an automated production mode most suitable for the aforementioned requirements, and has been valued in the manufacturing industry. It incorporates advanced computer application systems, such as Material Requirement Planning (MRP), Group Technology (GT), Computer Aided Process Planning (CAPP) and Multi-Processing Planning, ranging from ordering and material processing to delivery under system monitoring and resource assignment. This process is planned comprehensively to shorten the processing time and saving the cost. With the recent advancement of information technology, Artificial Intelligence (AI) has been developed prosperously, so multiagent technology has attention of researchers. The concept of agent has been applied to FMS, and the resource allocation for the manufacturing system is analyzed and discussed through the negotiation, coordination and cooperation mechanism among agents. In Smith (1980) proposed a multi-agent system – Contract Net Protocol (CNP), for resolving the contest of resources in a cooperative manner. This is a commonly-used coordination mode whereby the agents can resolve the resource allocation through negotiation and coordination based on Contract Net Protocol. Although there have been many researches focus on applying multi-agent tech-
* Corresponding author. Tel.: +886 2 27712171x2337; fax: +886 2 27317168. E-mail address: [email protected] (K.-Y. Chen). 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.04.024
nique on FMS. However, to the best of our knowledge, there is no literature of applying multi-agent technique on multi-section FMS. This paper applies multi-agent technology to multi-section FMS controller design, in order to build an agent-based control mode to improve the availability of machines, shorten the manufacturing time and increase the capacity. The objectives of this paper are to build a multi-agent based FMS cell controller via software agent technology, and to find the best dispatching rules combination for achieving largest throughput in a real case of multi-section FMS. 2. Literature review 2.1. Flexible manufacturing system The concept of FMS, originated from London-based Molins, was proposed by its R&D engineer David Williamson in early 1960s, and named as System24 (24 h operation, of which including 16 automatic operation) in 1965 for patent claims. The purpose of FMS is to satisfy the customer requirements for diversified and sophisticated products, and meet the expectations of the enterprises such as: shortening the delivery period, improving the quality and reducing the cost. Hence, FMS was planned for adapting itself to the aforementioned requirements as well as the production systems with medium and small throughput (Groover, 2001; Maleki, 1991). According to Nagarjuna, Mahesh, and Rajagopal (2006) and Viswanadham and Narahari (1992), FMS is defined as an integrated computer-controlled configuration, including: CNC machines, auxiliary production equipments and automated material handling system, with the purpose of medium and small batch production of a variety of products. Thus, FMS can manufacture diversified parts in a cost-effective way while minimizing the temporary inventory.
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
FMS aims at improving the product quality, reducing the production cost and shortening the lead time (Deng & Yang, 1999). As a real-time production system, FMS incorporates several CNC machines, material handling system and warehousing system, of which the computer of higher hierarchy is used for automatic monitoring. The communication mode between the cell controllers within FMS and other machines (e.g. RGV, CNC machine, Loading/ Unloading station), proposed by Deng and Yang (1999), is illustrated in Fig. 1. As for the hardware infrastructure of FMS, every independent machine is provided with a PLC for controlling its process; then, individual PLCs on the machines are connected to a master PLC, which is then linked to main computer via RS232 or RJ-45 (networked). Thus, the main computer can monitor the real-time status of every machine by reading the buffer of master PLC. Similarly, the main computer can also send message or command to various PLCs for performing the subsequent actions, or send the parameters and NC programs of machines via network. Thus, the coordination and control of FMS are assigned by main computer, and there is no signal communication among the machines and equipments, ensuring more simple and rapid control and operation of FMS. 2.2. Agent technology With the rapid development of e-commerce today, IT and Internet services are incorporated into many enterprise operations. In such case, artificial intelligence is developing quickly, of which the concept of intelligent agent has become a key example of today’s software development. The agents along with agent technology are widely discussed and applied in various fields. Stuart and Norvig (1995) suggested that, agent is defined to provide any service for other individuals or substances after authorization. An agent should set its target and act intelligently in an actual environment (Stone & Veloso, 2000). Agent should receive the external or internal commands, and handle the messages or tasks received through pre-established knowledge mechanism (Moore, Reyns, Kumara, & Hummel, 1997). According to the definition by Ferber (1999) in a broad sense, the agent may be a real individual or substance, or a virtual entity. Wooldridge (1997) indicated that, in a multi-agent system, every agent should have the ability of independence, reaction, prediction and interaction. A multi-agent system was used to resolve the system contest in a coordinated and competitive manner (Ishida, 1994; Jennings, Sycara, & Wooldridge, 1998). An agent should have the capability of independence, target-oriented communication and coordination (Fox, Barbuceanu, & Teigen, 2000). Huang and Nof (2000) concluded the characteristics of an agent such as: independence, coordination, target-orientation and responsiveness. Given the fact of well-developed IT and Internet technologies, many messages must be shared across the sectors, factories and borders. Many difficulties are encountered if message transfer is not conducted in a defined format or standard. Therefore, multi-
Status
Command & Request
Command & Request
agent system must be developed in accordance with certain framework and specifications. FIPA (The Foundation for Intelligent Physical) is an international organization of formulating the standards and specifications of agents, with its purpose of enabling the agents to perform their duties everywhere in a certain standard. FIPA was established in 1996, and recognized as 17th standard organization by IEEE in 2005. According to FIPA, the environment is divided into six parts, which are Software, Agent, Agent Management System (AMS), Directory Facilitator (DF), Message Transport Service (MTS) and Agent Platform, as shown in Fig. 2 (Bellifemine, Poggi, & Rimassa, 2001). JADE (Java Agent Development Framework) was developed by Telecom Italia in accordance with the standards of FIPA. JADE provides a series of function libraries and classes, enabling the program developers to freely and conveniently develop agent system. The agent platform of JADE was developed using JAVA language of Sun Microsystems. JADE has excellent performance thanks to the capabilities of JAVA including: object orientation, multiple sequences, cross-platform and high transportability. JADE agents can act in the platform with JAVA environment, and the agent platform can exist flexibly in various environments based on the capability of JAVA. The coordination and negotiation process among agents refer to the behavior of integrating multi-agent system. There are available with many negotiation modes, of which the most influential one is Contract Net Protocol. As proposed by Smith (1980), this concept was derived from the out-sourcing projects’ bidding procedures in human enterprises, and the issues and conflicts were addressed through multi-agent coordinative negotiation mode as shown in Fig. 3. (1) the manager issues bidding notice to the contractors when a new task is generated; (2) the contractors in idle state will receive the bidding notices from different managers; (3) the contractors make response to the bidding notices, and the managers will receive multiple responsive bidding information; (4) the managers evaluate the bids and select the contractors; and (5) a coordination is established with the selected contractors. 2.3. Using agents in FMS With the growing scale of FMS and increasingly complicated production processes, many scholars made attempts to improve the performance of FMS by combining it with the agents. To adapt
Status
Co-ordination For Load/Unload
Command & Request
Status
Co-ordination For Load/Unload
RGV
Fig. 1. Relation chart of sub-systems within FMS (Deng & Yang, 1999).
7311
Fig. 2. Agent management reference model (Bellifemine et al., 2001).
7312
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
Fig. 3. Contract network framework (Huang, 2004).
itself to highly flexible manufacturing requirements and complex organization, FMS is developed by integrating many sub-systems and performing tasks automatically when the operating environment is compatible with the commands, notwithstanding the system must have a certain degree of independence and coordination capability. The system performance could be improved by introducing the agent concept into the manufacturing system (Ming, 1996). The agent could make intelligent response and handling of external changes and the received messages, and transfer the suitable parts to the corresponding machines through message transfer and coordination (Choi, Kim, & Yook, 2000). The research on combination of agents with FMS contributed to establish the agents into the real-time FMS and apply them to distributed decision-making and dynamic scheduling system. Many studies focused on applying agent technology in real-time monitoring of FMS, of which Ouelhadj, Hanachi, and Bouzouia (2000) intended to realize real-time notification in the case of failure of FMS. Thus, multi-agent distributed monitoring system was used to monitor different resources through information-sharing capability of agents. He designed an agent monitoring function for two types of agents: task manager agent (TMA) and resource monitoring agent. The task manager agent is mainly responsible for managing and coordinating the behaviors among agents, receiving the commands of operators and transmitting messages to the operators; the resource monitoring agent is responsible for managing the assigned resources. A communication among the resource monitoring agents is possible to help realize various objectives by making decisions through the knowledge library or decision rules. Kouiss, Pierreval, and Mebarki (1997) applied distributed agent system to establish the dynamic dispatching rules, help the agents select appropriate dispatching rule in regional dispatching and resolve the global dispatching of manufacturing system through the cooperation of the agents. By comparing different dispatching rules in the FMS, this could support dynamically the multi-agent structure for dispatching purpose. Every agent is responsible for controlling the service center of manufacturing system, and selecting the optimum solution depending on its own database and status. With the help of dispatching rules, the agent could address the local dynamic dispatching. The dispatching rules include: SIPT, SCR*allowable dwell time/remaining parts processing time, EDD, CEXSPT and CR/SPT.
Li, Geng, Yang, and Xu (2002) applied agent system to the route dispatching of AGV, calculated the throughput by building a simulated environment, and then compared with conventional dispatching rules. The results indicate that the agent system is superior to conventional dispatching rules. Yamanoto and Marui (2005) proposed the possibility of a future FMS. An independent FMS was regarded as a production mode in the future, helping to make decisions and judgment by preset targets as well as preparations by finding out and predicting the possible situations. Farahvash and Boucher (2004) built a multi-agent architecture for control of AGV system. The experiment showed that using a combination of routing criteria regularly outperformed the use of shortest distance path. Reaidy, Massotte, and Diep (2006) compared five different negotiation protocols in dynamic agent-based manufacturing systems. They proposed that the ‘‘co-opetition” protocol achieved the best results in terms of tardiness and waiting time. Leitao (2009) surveyed the literature in manufacturing control systems using distributed AI techniques. He also discussed the reasons for the weak adoption of these approaches by industry and pointed out the challenges and research opportunities for the future. Turgay (2009) used Petri nets and object-oriented design to develop the integrated agent-based FMS control system. However, the above literatures only address the problem of single section FMS. Multi-section FMS plays more and more important role in modern production system, such as motorcycle cylinder production line, wafer manufacturing in an IC fab and defect LCD panel repair line. This paper will use multi-agent technique in multi-section FMS to illustrate its benefits of improve manufacturing efficiency. 3. System framework and scenario design 3.1. System framework The layout of a multi-section FMS is illustrated in Fig. 4. This is a flexible transfer line for producing motorcycle cylinders, including: buffer 1–16 storing 16 raw materials, intermediate buffers storing separately 9 work in process (WIP) (Buffer 19–27 and Buffer 31– 39), and the last buffer (Buffer 44) storing the defective parts from the last machine. This system is provided with 9 CNC machines as well as 3 RGVs for transferring parts between the buffers and the manufacturing zones. The parts enter the manufacturing system from Buffer 1–16, and the finished ones are fed back via RGV3, RGV2, RGV1 from the last machine. The processing step of multi-section FMS is: load the pallets from buffer(1–16) to TC1, TC2; when the processing in the first zone is finished, RGV1 transfers WIP to buffer(19–27); after receiving the manufacturing request of buffer(19–27), RGV2 transfers the pallets to individual machines according to the manufacturing procedure of WIP; when the task in the second zone is finished, RGV2 transfers WIP to buffer(31–39); after receiving the manufacturing request of buffer(31–39), RGV3 transfers the pallets to WM, LT, HC, IS in sequence; after completion of processing, the defective products are transferred to buffer(44), and qualified ones fed back
Fig. 4. Layout of flexible transfer line.
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
7313
to buffer(31–39), and where applicable, fed back to buffer(1–16) via RGV2, RGV1. In order to establish multi-agent environment in the FMS, this study applied JADE to build up a multi-agent system networked to the FMS, and assign an independent agent for monitoring the manufacturing machines. Depending on different machine functions, the agents are categorized into: RGV agent, CNC agent and AS/RS agent, of which RGV agent is set as the main agent, and various agents exchange message and make coordination through JADE for the purpose of distributed decision-making, as shown in Fig. 5. RGV agent is responsible for controlling RGV, and issuing command of FROM-TO to RGV. When there is a command for handling, an inquiry is made to the intended agents (CNC agent, AS/RS agent) through contract network framework; the agents having received the inquiry may feedback the bid value of handling depending on their own status; if there is a need, RGV agent feeds back confirmation messages to the selected machine agent; if there is a need of handling from two or more machine agents, RGV agent will compare their bid value and select the machines of biggest benefit. This process was performed in proper sequence and divided into four phases: initial phase, sending messages, receiving& handling messages and stop, as shown in Fig. 6. 3.1.1. Initial phase The buffer of Socket in JAVA is checked continuously; when FMS sends messages to JADE using Winsock, it represents RGV starts running or finishes the preceding work, so agent is notified to start selection of next CNC. 3.1.2. Sending messages The agent prepares for sending messages, then establishes message templates, and finally sends out the messages. 3.1.3. Receiving& handling messages RGV agent starts to wait for the feedback from other agents, including: ACCEPT_PROPOSAL and REFUSE. The machine is listed into the catalogs of candidates when ACCEPT_PROPOSAL is received, or neglected when REFUSE is received. If the quantity of fed back messages is equivalent to that of sent messages, RGV agent will select the optimum handling machine from the candidates according to the performance index. 3.1.4. Stop If some machines are listed into the candidates, INFORM message is sent to the successful machine agent, and the network message of Dispatch Information will be transferred to notify FMS for
Fig. 5. Framework of agents.
Fig. 6. Flow chart of RGV agent.
handling, and finally returned to initial phase; if no machine is listed into the candidates (the received message of REFUSE is equal to that of sent messages), it is returned directly to initial phase. AS/RS agent is responsible for controlling the buffer station, and ensuring if the available buffers and parts are suitable for handling. When the message of RGV is received, the parts to be handled are selected depending on the bid value, and then the parts along with the bid value are fed back to RGV agent for reference. When the system is activated, the relevant messages from RGV agent are processed, and the behavioral model of processed parts is selected, allowing to make response when receiving the messages. The messages received by AS/RS agent are divided into CFP and INFORM; when CFP is received, it represents that RGV agent makes inquiry to AS/RS agent if handling is required; after receiving the message, AS/RS agent will make a judgment, and after confirming the message, calculate the performance index and feed back the message; when INFORM is received, it represents that RGV agent selects the machine on a priority, and prepares for handling the parts, with the flow process shown in Fig. 7. CNC agent is responsible for controlling CNC and calculating the bid value. After receiving the inquiry from RGV agent, CNC agent checks the current machine status and make a response; in the event of ongoing processing, or lack of parts on the machines or presence of parts on the destination machines, CNC agent rejects the demand of RGV agent, or feed back to RGV agent the received information along with the bid value after processing is finished. When the system is activated, the relevant messages from RGV agent are processed, and the behavioral model including performance index is selected, allowing to make response when receiving the messages. The messages received by CNC agent are divided into CFP and INFORM; when CFP is received, it represents that RGV agent makes inquiry to CNC agent if handling is required; after receiving the message, CNC agent makes a judgment, and after confirming the message, calculate the performance index and feed back ACCEPT_PROPOSAL; REFUSE is fed back if the parts are not handled; when INFORM is received, it represents that RGV agent selects CNC on a priority, and prepares for handling the parts, with the flow process shown in Fig. 8.
7314
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
Fig. 7. Flow process of AS/RS agent.
messages required for FMS, with the basic experimental environment shown in Fig. 9. In order to provide normal operation of FMS cell controller and multi-agent system, Socket packages are established in JAVA system, and Winsock packages established in FMS, enabling the work of both systems via network. There are three types of messages available for network transfer: Start, Dispatch Information and Server-ok. In the case of Start (see Fig. 10), it represents that the FMS (Cell Controller) begins with writing the working status of machines into the database, updating the processing status and notifying the agent platform (JADE) for calculation; next, the agent platform dispatch information of the calculated results to the simulated FMS; after the completion of the manufacturing system’s operation, Server-ok message will be fed back, allowing for next cycle of agent system. In the multi-agent system, RGV agent is mainly responsible for system operation in 4 steps. Step 0: the agent will check continuously the network messages transferred from the manufacturing system; in the case of Start, the agent starts to check the handling requirement in the datasheet, and enters into Step 1; in Step 1, RGV agent will send internal message to the machine agent (CNC, ASRS) according to the handling requirement, and enters into Step 2; in Step 2, internal response messages are received; the successful agents are selected when the quantity of response messages is equal to that of sent messages, then enters into Step 3; in Step 3, in the case of presence of successful agent, the messages are sent to this agent, and dispatched to Cell Controller; after receiving the message of Server-ok, it returns to Step 0.
3.2. Integration of multi-agent and FMS The experimental environment is divided into two parts: one is an FMS cell controller developed by Microsoft Visual Basic 6.0, which provides a real-time monitoring screen displaying the simulation process and result; the other one is a multi-agent system, responsible for controlling the status and process of manufacturing system based on JADE platform. In addition, information exchange is made possible by TCP/IP for normal coordinative operation of two systems, and Microsoft SQL Server 2005 is used to record the
Fig. 8. Flow process of CNC agent.
3.3. Description of empirical scenario There are three manufacturing scenarios in the system implementation (see Fig. 11); during input of raw materials, the materials are selectively processed in the First Section by checking if TC1/ TC2 is in idle state; in the Second Section, the products are transferred to the respective machines according to their part type; in the Third Section, processing is conducted through WM, LT, HC, IS. These scenarios are discussed in two parts: the one is simulated experiment of FMS with introduction of agent technology VS. without introduction of agent technology; the other one is simulated experiment of dynamic rules combination for various sections of manufacturing system with introduction of agent technology into FMS. The manufacturing process of three parts in the experiment is simulated in individual scenarios of FTL. Take Fig. 11 Scenario A, for example, the manufacturing process of Part 1 is: TC1/TC2 > MC1 -> WM -> LT -> HC -> IS; the manufacturing process of Part 2 is: TC1/TC2 -> MC2 -> WM -> LT -> HC -> IS; the manufacturing process of Part 3 is: TC1/TC2 -> MC3 -> WM -> LT -> HC -> IS. Scenario B, the manufacturing process of Part 1 is: TC1/TC2 -> MC1 > MC2 -> WM -> LT -> HC -> IS; the manufacturing process of Part 2 is: TC1/TC2 -> MC1 -> MC3 -> WM -> LT -> HC -> IS; the manu-
Fig. 9. A basic environment of this research.
7315
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
System Start Send Message -Start-
Check demand
Send internal message System Check Read the instruction Receive internal messages Selected objective Execution Task Send Message -Server ok-
Send Message
Assumption 1. The standby capacity of the system is infinite; when the processed parts are fed back to Buffer 1–16, new parts are immediately put into production. Assumption 2. All parts must be subject to the last IS (Inspection Station), and then fed back to Buffer 1–16 via RGVs. Assumption 3. All parts must be handled by WM, LT, HC and IS in the last section, and then fed back to Buffer 1–16 via RGVs. Assumption 4. The system’s manufacturing and processing time is 50 h.
-Dispatch Information-
3.4. Operation of FMS Fig. 10. Integrated system process.
This section aims at testing the operation of FMS; after setup the procedures, one can firstly see the main screen, which can be switched to simulated monitor by clicking Monitor Button without changing the default values; after selecting the dispatching rules such as FCFS, SIPT, Buffer Sequence and MultiAgentsFMS, press START button, or press STOP button to stop the operation of procedures. Fig. 12A is a screen in operation, which displays the current position of parts; if the parts are raw materials, it is displayed as yellow, if the parts are semi-finished products, it is displayed as green, and if the parts are finished products, it is displayed as red; the current automated RGV handling commands as well as the types, quality and quantity of finished parts are displayed below. The throughput, cycle time and utilization of machines can be shown in Fig. 12B–D, respectively. 3.5. Operation of multi-agent system
Fig. 11. Three manufacturing scenarios.
facturing process of Part 3 is: TC1/TC2 -> MC2 -> MC3 -> WM -> LT -> HC -> IS. Scenario C, the manufacturing process of Part 1 is: TC1/TC2 -> MC1 -> MC2 -> MC3 -> WM -> LT -> HC -> IS; the manufacturing process of Part 2 is: TC1/TC2 -> MC1 -> MC2 -> WM -> LT -> HC -> IS; the manufacturing process of Part 3 is: TC1/TC2 -> MC3 -> WM -> LT -> HC -> IS. After the user stops the system operation, the system will record the starting and stopping time, observe and analyze the utilization time of RGV and CNC, etc.
3.3.1. Agent VS. without agent This experiment simulated 50-h manufacturing process in three manufacturing scenarios (Fig. 11), wherein the experiments with/ without introduction of agents were compared, by using the dispatching rules: FCFS, SIPT and Buffer Sequence. Therefore, there are totally six possible combinations.
3.3.2. Dynamic dispatching rules with agent This experiment simulated another 50-h manufacturing process in ABC manufacturing scenarios (Fig. 11), wherein the experiments for various combinations of sections were compared using different combination of three dispatching rules: FCFS, SIPT and Buffer Sequence. Therefore, there are totally 27 possible combinations as shown in Table 1. The assumptions of FTL simulated experiments are listed below:
In this research, SNIFFER AGENT provided by JADE was employed to monitor the communication among agents on a multiagent platform. For example, Figs. 13–15 depicts the communication during operation of multi-agent system. It can be seen from
Table 1 Combination of dynamic rules. Combination
Section 1
Section 2
Section 3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
SIPT BUFF SIPT BUFF FCFS BUFF BUFF SIPT FCFS BUFF FCFS FCFS FCFS BUFF SIPT FCFS BUFF BUFF FCFS BUFF SIPT FCFS SIPT SIPT SIPT FCFS SIPT
BUFF SIPT SIPT BUFF BUFF SIPT FCFS BUFF FCFS BUFF SIPT SIPT BUFF SIPT SIPT FCFS BUFF FCFS FCFS FCFS FCFS BUFF FCFS BUFF FCFS SIPT SIPT
SIPT FCFS FCFS FCFS FCFS SIPT FCFS FCFS SIPT SIPT SIPT BUFF BUFF BUFF SIPT FCFS BUFF BUFF BUFF SIPT BUFF SIPT SIPT BUFF FCFS FCFS BUFF
7316
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
Fig. 14. Communications of multi-agent system (Section 2).
Fig. 12. Operating screen of manufacturing system.
No. 1–4 that, RGV agent sends inquiry message of CFP to ASRS1, ASRS2, TC1 and TC2, which feedback the messages separately after handling the messages. In No. 5–7, ASRS2, TC1 and TC2 feedback REFUSE to RGV agent after confirmation, but ASRS1 feeds back ACCEPT_PROPOSAL; in such case, RGV agent will decide ASRS1 as the successful agent according to the performance index attached in the feedback messages, and then send INFORM to it as shown in No. 9, the successful agent will make preparations after receiving INFORM. Prior to termination of the system, the agents make communication in such a way. Fig. 15. Communications of multi-agent system (Section 3).
4. Simulation results This experiment simulated 50-h manufacturing process in ABC scenarios, and discussed in two parts: simulated experiment of FMS with introduction of agent technology VS. without introduction of agent technology; simulated experiment of dynamic rules for various sections of manufacturing system with introduction of agent technology into FMS.
4.1. Agent VS. without agent This section analyzed the throughput of three parts by totalizing and comparing the throughput in various scenarios, and also observed the trend as listed in Figs. 16–18, respectively.
Firstly, as shown in Fig. 16, the agent system has the largest throughput. In addition, the trend of the bar chart indicates that, among the agent system, the difference of throughput for three rules differs little. In Fig. 17, the agent system has also the largest throughput, the trend of the bar chart indicates that, BUFF in the agent has the biggest throughput. In Fig. 18, the agent system has also the largest throughput, the trend of the bar chart indicates that, FCFS in the agent has the largest throughput. The experiment found that, the throughput can be increased efficiently with introduction of agent system, and the agent system has excellent performance of capacity in various scenarios. Thus, it is feasible to introduce multi-agent system into multi-section FMS. 4.2. Dynamic dispatching rules combination with agent The experimental results revealed that, it is possible to increase the capacity by introducing agent technology into FMS. Next, this
Fig. 13. Communications of multi-agent system (Section 1).
Fig. 16. Bar chart of throughput in scenario A.
7317
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318 Table 3 Ranking of throughput of single rule.
F–F–F S–S–S B–B–B
Fig. 17. Bar chart of throughput in scenario B.
Fig. 18. Bar chart of throughput in scenario C.
section analyzed 27 sets of dynamic rules by totalizing the throughput of three parts and comparing the throughput in various scenarios, and also observed the trend to evaluate the feasibility of dynamic rules. Table 2 lists the throughputs of top 3 and bottom 3 agents in various scenarios. As shown in Table 3, the throughput using single rule is not the optimal throughput as for the arrangement of dispatching rules at various sections, so it is necessary to combine the dynamic rules to optimize the system throughput. The experiment found that, the agent system combined with dynamic dispatching can increase efficiently the capacity, and various sections have excellent performance of capacity with dynamic dispatching in various scenarios. Thus, it is feasible to introduce the agent system combined with dynamic dispatching into Multi-Section FMS.
5. Conclusions As compared with single-section FMS, the multi-section FMS features more complex operation, more product categories and more flexible manufacturing process, making system control and
Table 2 Ranking of throughput of dynamic rules.
No. No. No. ... No. No. No.
1 2 3 25 26 27
Scenario A
Scenario B
Scenario C
S–B–S B–S–F S–S–F ... F–S–F S–F–F S–S–B
F–B–S B–B–B F–B–F ... S–F–S S–F–F F–F–B
B–S–F S–S–F S–F–F ... F–B–S F–B–B S–B–F
Scenario A
Scenario B
Scenario C
No. 16 No. 15 No. 17
No. 4 No. 13 No. 2
No. 4 No. 22 No. 20
dispatching more important to the performance of manufacturing system. With introduction of the concept of agent technology, the multi-section FMS combines the agent technology with the control modules of various machines, and establishes a perfect simulated manufacturing system based on the planning of manufacturing modes. The message transfer and coordination among agents is made possible through the communication language, or through a coordination mechanism based on the contract network in the event of conflict of assigned tasks. This can select the priority machines and deliver services for maximizing the overall benefits of the system and efficient allocation of system resources. Multi-section FMS (Cell Controller) exchanges information with multi-agent system (JADE) through network technology, but many manufacturing processes or system status have to be provided from database. The advantage of agent system is that it can prevent the decline of system stability, shorten the development time and increase the operating efficiency. Moreover, the agent technology can sharpen the enterprises’ competitive edge by introducing new technologies or new algorithms quickly and stably. Thus, establishing a perfect agent system and defining its coordination criterions can guarantee satisfactory operating capacity while the tasks assignment of manufacturing system is not limited to certain algorithm or mechanism. With the introduction of agent technology, the existing manufacturing system is only required to handle partially the hardware assignment commands, making it possible to shorten greatly the development time and reduce the maintenance burdens. Limited literatures pay attention to the control of multi-section FMS. This is the first paper to apply multi-agent technique in multi-section FMS. This study integrated the multi-agent system with multi-section FMS based on the established agent knowledge base. It is suggested that the dispatching rules should be applied to the optimal combination in various sections, and independent agents should take charge of judging the scenarios of dynamic rules for increasing the system efficiency. Acknowledgment The authors would like to give thanks to the two anonymous reviewers for their valuable suggestions and comments. This work is partially supported by National Science Council, Taiwan under Grant NSC 97-2221-E-027-076. Appendix A. Appendices The following are abbreviated terms and definitions used in this paper: FCFS: first-come-first-served dispatching rule is implemented by simulated FMS. MAS-FCFS: FCFS is implemented by multi-agent system (MAS). SIPT: shortest imminent processing time dispatching rule is implemented by simulated FMS. MAS-SIPT: SIPT dispatching rule is implemented by multi-agent system (MAS). BUFF: buffer sequence dispatching rule is implemented by simulated FMS.
7318
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
MAS-BUFF: buffer sequence dispatching rule is implemented by multi-agent system (MAS). TC1, TC2: turning machine. MC1, MC2, MC3: milling machine. WM: washing machine. LT: leak testing machine. HC: horniness center. IS: inspection station. Utilization: utilization of machines: sum of processing time of machines / total actual operating time of system. Throughput: total output of various parts. MLT: manufacturing lead time, defined as the average processing time from the first to the last station.
References Bellifemine, F., Poggi, A., & Rimassa, G. (2001). Developing multi-agent systems with JADE. Intelligent agents. Springer-Verlag. Choi, K. H., Kim, S. C., & Yook, S. H. (2000). Multi-agent hybrid shop floor control system. International Journal of Production Research, 38(17), 4193–4203. Deng, Y., & Yang, C. R. (1999). Architecture-driven modeling of real-time concurrent systems with application in FMS. Journal of Systems and Software, 45(1), 61–78. Farahvash, P., & Boucher, T. O. (2004). A multi-agent architecture for control of AGV systems. Robotics and Computer-Integrated Manufacturing, 20(6), 473–483. Ferber, J. (1999). Multi-agent systems – An introduction to distributed artificial intelligence. Addison-Wesley. Fox, M. S., Barbuceanu, M., & Teigen, R. (2000). Agent-oriented supply-chain management. The International Journal of Flexible Manufacturing System, 12(2), 165–188. Groover, M. P. (2001). Automation, production systems, and computer-integrated manufacturing (2nd ed.). Prentice-Hall. Huang, C. Y., & Nof, S. Y. (2000). Formation of autonomous agent network for manufacturing systems. International Journal of Production Research, 38(3), 607–624. Huang, W. J. (2004). The development of communication protocols for production planning in distributed manufacturing systems, master thesis, Tunghai University, Taiwan.
Ishida, T. (1994). Parallel, distributed and multiagent production systems. SpringerVerlag. Jennings, N. R., Sycara, K., & Wooldridge, M. (1998). A roadmap of agent research and development. Autonomous Agents and Multi-agent Systems, 1(1), 7–38. Kouiss, K., Pierreval, H., & Mebarki, N. (1997). Using multi-agent architecture in FMS for dynamic scheduling. Journal of Intelligent Manufacturing, 8(1), 41–47. Leitao, P. (2009). Agent-based distributed manufacturing control: A state-of-art survey. Engineering Applications of Artificial Intelligence, 22, 979–991. Li, X., Geng, T., Yang, Y., & Xu, X. (2002). Multiagent AGVs’ dispatching system using multilevel decisions method. American Control Conference, 2, 1135–1136. Maleki, R. A. (1991). Flexible manufacturing systems (3rd ed.). Prentice Hall. Ming, L. (1996). Agent-oriented analysis methodology in intelligent manufacturing system. The 4th International Conference on Control, Automation, Robotics and Vision, 254–258. Moore, M. L., Reyns, R. B., Kumara, R. T., & Hummel, J. R. (1997). Distributed intelligent agents for logistic (DIAL). Computational Cybernetics and Simulation, 3, 2782–2787. Nagarjuna, N., Mahesh, O., & Rajagopal, K. (2006). A heuristic based on multi-stage programming approach for machine-loading problem in a flexible manufacturing system. Robotics and Computer Integrated Manufacturing, 22(4), 342–352. Ouelhadj, D., Hanachi, C., & Bouzouia, B. (2000). Multi-agent architecture for distributed monitoring in flexible manufacturing system. IEEE International Conference on Robotics and Automation, 3, 2416–2421. Reaidy, J., Massotte, P., & Diep, D. (2006). Comparison of negotiation protocols in dynamic agent-based manufacturing systems. International Journal of Production Economics, 99, 117–130. Smith, R. G. (1980). The contract net protocol: High-level communication and control in a distributed problem solver. IEEE Transactions on Computers, 29(12), 104–1113. Stone, P., & Veloso, M. (2000). Multiagent system: A survey from a machine learning perspective. Autonomous Robots, 8(3), 345–383. Stuart, R., & Norvig, P. (1995). Artificial intelligence: A modern approach. Prentice Hall. Turgay, S. (2009). Agent-based FMS control. Robotics and Computer-Integrated Manufacturing, 25, 470–480. Viswanadham, N., & Narahari, Y. (1992). Performance modeling of automated manufacturing systems. Prentice Hall. Wooldridge, M. (1997). Agent-based software engineering. IEE Proceedings Software Engineering, 144(11), 26–37. Yamanoto, H., & Marui, E. (2005). Intelligent communication between agents of autonomous decentralized FMS. IEEE International Symposium on Computational Intelligence in Robotics and Automation, 433–438.
Contents lists available at ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
Applying multi-agent technique in multi-section flexible manufacturing system Kai-Ying Chen *, Chun-Jay Chen Department of Industrial Engineering and Management, National Taipei University of Technology, No. 1, Chunghsiao E. Rd., Sec. 3, Taipei 106, Taiwan, ROC
a r t i c l e Keywords: Multi-section FMS Multi-agent Dispatching JADE
i n f o
a b s t r a c t In the highly competitive market, cooperative multi-agent transaction and negotiation mechanism have become an important research topic. This paper uses multi-agent technology to construct a multi-section flexible manufacturing system (FMS) model, and utilizes simulation to build a manufacturing environment based on JADE framework for multi-agent to combine with dispatching rules, such as shortest imminent processing time (SIPT), first come first serve (FCFS) earliest due date (EDD), and Buffer Sequence. This paper finds that using multi-agent technique for multi-section FMS model can enhance the production efficiency in practice. Meanwhile, in this study, multi-agent systems combined with dynamic dispatching can be used to identify the best dispatching rules combination for achieving largest throughput, and thus it can provide the reference for production scheduling in advance. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction With the increasingly diversified market demands for small batch production, customized manufacturing mode must be employed so that the production system, production line and procedures can be adjusted flexibly to manufacture various kinds of products. Flexible manufacturing system (FMS) is an automated production mode most suitable for the aforementioned requirements, and has been valued in the manufacturing industry. It incorporates advanced computer application systems, such as Material Requirement Planning (MRP), Group Technology (GT), Computer Aided Process Planning (CAPP) and Multi-Processing Planning, ranging from ordering and material processing to delivery under system monitoring and resource assignment. This process is planned comprehensively to shorten the processing time and saving the cost. With the recent advancement of information technology, Artificial Intelligence (AI) has been developed prosperously, so multiagent technology has attention of researchers. The concept of agent has been applied to FMS, and the resource allocation for the manufacturing system is analyzed and discussed through the negotiation, coordination and cooperation mechanism among agents. In Smith (1980) proposed a multi-agent system – Contract Net Protocol (CNP), for resolving the contest of resources in a cooperative manner. This is a commonly-used coordination mode whereby the agents can resolve the resource allocation through negotiation and coordination based on Contract Net Protocol. Although there have been many researches focus on applying multi-agent tech-
* Corresponding author. Tel.: +886 2 27712171x2337; fax: +886 2 27317168. E-mail address: [email protected] (K.-Y. Chen). 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.04.024
nique on FMS. However, to the best of our knowledge, there is no literature of applying multi-agent technique on multi-section FMS. This paper applies multi-agent technology to multi-section FMS controller design, in order to build an agent-based control mode to improve the availability of machines, shorten the manufacturing time and increase the capacity. The objectives of this paper are to build a multi-agent based FMS cell controller via software agent technology, and to find the best dispatching rules combination for achieving largest throughput in a real case of multi-section FMS. 2. Literature review 2.1. Flexible manufacturing system The concept of FMS, originated from London-based Molins, was proposed by its R&D engineer David Williamson in early 1960s, and named as System24 (24 h operation, of which including 16 automatic operation) in 1965 for patent claims. The purpose of FMS is to satisfy the customer requirements for diversified and sophisticated products, and meet the expectations of the enterprises such as: shortening the delivery period, improving the quality and reducing the cost. Hence, FMS was planned for adapting itself to the aforementioned requirements as well as the production systems with medium and small throughput (Groover, 2001; Maleki, 1991). According to Nagarjuna, Mahesh, and Rajagopal (2006) and Viswanadham and Narahari (1992), FMS is defined as an integrated computer-controlled configuration, including: CNC machines, auxiliary production equipments and automated material handling system, with the purpose of medium and small batch production of a variety of products. Thus, FMS can manufacture diversified parts in a cost-effective way while minimizing the temporary inventory.
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
FMS aims at improving the product quality, reducing the production cost and shortening the lead time (Deng & Yang, 1999). As a real-time production system, FMS incorporates several CNC machines, material handling system and warehousing system, of which the computer of higher hierarchy is used for automatic monitoring. The communication mode between the cell controllers within FMS and other machines (e.g. RGV, CNC machine, Loading/ Unloading station), proposed by Deng and Yang (1999), is illustrated in Fig. 1. As for the hardware infrastructure of FMS, every independent machine is provided with a PLC for controlling its process; then, individual PLCs on the machines are connected to a master PLC, which is then linked to main computer via RS232 or RJ-45 (networked). Thus, the main computer can monitor the real-time status of every machine by reading the buffer of master PLC. Similarly, the main computer can also send message or command to various PLCs for performing the subsequent actions, or send the parameters and NC programs of machines via network. Thus, the coordination and control of FMS are assigned by main computer, and there is no signal communication among the machines and equipments, ensuring more simple and rapid control and operation of FMS. 2.2. Agent technology With the rapid development of e-commerce today, IT and Internet services are incorporated into many enterprise operations. In such case, artificial intelligence is developing quickly, of which the concept of intelligent agent has become a key example of today’s software development. The agents along with agent technology are widely discussed and applied in various fields. Stuart and Norvig (1995) suggested that, agent is defined to provide any service for other individuals or substances after authorization. An agent should set its target and act intelligently in an actual environment (Stone & Veloso, 2000). Agent should receive the external or internal commands, and handle the messages or tasks received through pre-established knowledge mechanism (Moore, Reyns, Kumara, & Hummel, 1997). According to the definition by Ferber (1999) in a broad sense, the agent may be a real individual or substance, or a virtual entity. Wooldridge (1997) indicated that, in a multi-agent system, every agent should have the ability of independence, reaction, prediction and interaction. A multi-agent system was used to resolve the system contest in a coordinated and competitive manner (Ishida, 1994; Jennings, Sycara, & Wooldridge, 1998). An agent should have the capability of independence, target-oriented communication and coordination (Fox, Barbuceanu, & Teigen, 2000). Huang and Nof (2000) concluded the characteristics of an agent such as: independence, coordination, target-orientation and responsiveness. Given the fact of well-developed IT and Internet technologies, many messages must be shared across the sectors, factories and borders. Many difficulties are encountered if message transfer is not conducted in a defined format or standard. Therefore, multi-
Status
Command & Request
Command & Request
agent system must be developed in accordance with certain framework and specifications. FIPA (The Foundation for Intelligent Physical) is an international organization of formulating the standards and specifications of agents, with its purpose of enabling the agents to perform their duties everywhere in a certain standard. FIPA was established in 1996, and recognized as 17th standard organization by IEEE in 2005. According to FIPA, the environment is divided into six parts, which are Software, Agent, Agent Management System (AMS), Directory Facilitator (DF), Message Transport Service (MTS) and Agent Platform, as shown in Fig. 2 (Bellifemine, Poggi, & Rimassa, 2001). JADE (Java Agent Development Framework) was developed by Telecom Italia in accordance with the standards of FIPA. JADE provides a series of function libraries and classes, enabling the program developers to freely and conveniently develop agent system. The agent platform of JADE was developed using JAVA language of Sun Microsystems. JADE has excellent performance thanks to the capabilities of JAVA including: object orientation, multiple sequences, cross-platform and high transportability. JADE agents can act in the platform with JAVA environment, and the agent platform can exist flexibly in various environments based on the capability of JAVA. The coordination and negotiation process among agents refer to the behavior of integrating multi-agent system. There are available with many negotiation modes, of which the most influential one is Contract Net Protocol. As proposed by Smith (1980), this concept was derived from the out-sourcing projects’ bidding procedures in human enterprises, and the issues and conflicts were addressed through multi-agent coordinative negotiation mode as shown in Fig. 3. (1) the manager issues bidding notice to the contractors when a new task is generated; (2) the contractors in idle state will receive the bidding notices from different managers; (3) the contractors make response to the bidding notices, and the managers will receive multiple responsive bidding information; (4) the managers evaluate the bids and select the contractors; and (5) a coordination is established with the selected contractors. 2.3. Using agents in FMS With the growing scale of FMS and increasingly complicated production processes, many scholars made attempts to improve the performance of FMS by combining it with the agents. To adapt
Status
Co-ordination For Load/Unload
Command & Request
Status
Co-ordination For Load/Unload
RGV
Fig. 1. Relation chart of sub-systems within FMS (Deng & Yang, 1999).
7311
Fig. 2. Agent management reference model (Bellifemine et al., 2001).
7312
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
Fig. 3. Contract network framework (Huang, 2004).
itself to highly flexible manufacturing requirements and complex organization, FMS is developed by integrating many sub-systems and performing tasks automatically when the operating environment is compatible with the commands, notwithstanding the system must have a certain degree of independence and coordination capability. The system performance could be improved by introducing the agent concept into the manufacturing system (Ming, 1996). The agent could make intelligent response and handling of external changes and the received messages, and transfer the suitable parts to the corresponding machines through message transfer and coordination (Choi, Kim, & Yook, 2000). The research on combination of agents with FMS contributed to establish the agents into the real-time FMS and apply them to distributed decision-making and dynamic scheduling system. Many studies focused on applying agent technology in real-time monitoring of FMS, of which Ouelhadj, Hanachi, and Bouzouia (2000) intended to realize real-time notification in the case of failure of FMS. Thus, multi-agent distributed monitoring system was used to monitor different resources through information-sharing capability of agents. He designed an agent monitoring function for two types of agents: task manager agent (TMA) and resource monitoring agent. The task manager agent is mainly responsible for managing and coordinating the behaviors among agents, receiving the commands of operators and transmitting messages to the operators; the resource monitoring agent is responsible for managing the assigned resources. A communication among the resource monitoring agents is possible to help realize various objectives by making decisions through the knowledge library or decision rules. Kouiss, Pierreval, and Mebarki (1997) applied distributed agent system to establish the dynamic dispatching rules, help the agents select appropriate dispatching rule in regional dispatching and resolve the global dispatching of manufacturing system through the cooperation of the agents. By comparing different dispatching rules in the FMS, this could support dynamically the multi-agent structure for dispatching purpose. Every agent is responsible for controlling the service center of manufacturing system, and selecting the optimum solution depending on its own database and status. With the help of dispatching rules, the agent could address the local dynamic dispatching. The dispatching rules include: SIPT, SCR*allowable dwell time/remaining parts processing time, EDD, CEXSPT and CR/SPT.
Li, Geng, Yang, and Xu (2002) applied agent system to the route dispatching of AGV, calculated the throughput by building a simulated environment, and then compared with conventional dispatching rules. The results indicate that the agent system is superior to conventional dispatching rules. Yamanoto and Marui (2005) proposed the possibility of a future FMS. An independent FMS was regarded as a production mode in the future, helping to make decisions and judgment by preset targets as well as preparations by finding out and predicting the possible situations. Farahvash and Boucher (2004) built a multi-agent architecture for control of AGV system. The experiment showed that using a combination of routing criteria regularly outperformed the use of shortest distance path. Reaidy, Massotte, and Diep (2006) compared five different negotiation protocols in dynamic agent-based manufacturing systems. They proposed that the ‘‘co-opetition” protocol achieved the best results in terms of tardiness and waiting time. Leitao (2009) surveyed the literature in manufacturing control systems using distributed AI techniques. He also discussed the reasons for the weak adoption of these approaches by industry and pointed out the challenges and research opportunities for the future. Turgay (2009) used Petri nets and object-oriented design to develop the integrated agent-based FMS control system. However, the above literatures only address the problem of single section FMS. Multi-section FMS plays more and more important role in modern production system, such as motorcycle cylinder production line, wafer manufacturing in an IC fab and defect LCD panel repair line. This paper will use multi-agent technique in multi-section FMS to illustrate its benefits of improve manufacturing efficiency. 3. System framework and scenario design 3.1. System framework The layout of a multi-section FMS is illustrated in Fig. 4. This is a flexible transfer line for producing motorcycle cylinders, including: buffer 1–16 storing 16 raw materials, intermediate buffers storing separately 9 work in process (WIP) (Buffer 19–27 and Buffer 31– 39), and the last buffer (Buffer 44) storing the defective parts from the last machine. This system is provided with 9 CNC machines as well as 3 RGVs for transferring parts between the buffers and the manufacturing zones. The parts enter the manufacturing system from Buffer 1–16, and the finished ones are fed back via RGV3, RGV2, RGV1 from the last machine. The processing step of multi-section FMS is: load the pallets from buffer(1–16) to TC1, TC2; when the processing in the first zone is finished, RGV1 transfers WIP to buffer(19–27); after receiving the manufacturing request of buffer(19–27), RGV2 transfers the pallets to individual machines according to the manufacturing procedure of WIP; when the task in the second zone is finished, RGV2 transfers WIP to buffer(31–39); after receiving the manufacturing request of buffer(31–39), RGV3 transfers the pallets to WM, LT, HC, IS in sequence; after completion of processing, the defective products are transferred to buffer(44), and qualified ones fed back
Fig. 4. Layout of flexible transfer line.
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
7313
to buffer(31–39), and where applicable, fed back to buffer(1–16) via RGV2, RGV1. In order to establish multi-agent environment in the FMS, this study applied JADE to build up a multi-agent system networked to the FMS, and assign an independent agent for monitoring the manufacturing machines. Depending on different machine functions, the agents are categorized into: RGV agent, CNC agent and AS/RS agent, of which RGV agent is set as the main agent, and various agents exchange message and make coordination through JADE for the purpose of distributed decision-making, as shown in Fig. 5. RGV agent is responsible for controlling RGV, and issuing command of FROM-TO to RGV. When there is a command for handling, an inquiry is made to the intended agents (CNC agent, AS/RS agent) through contract network framework; the agents having received the inquiry may feedback the bid value of handling depending on their own status; if there is a need, RGV agent feeds back confirmation messages to the selected machine agent; if there is a need of handling from two or more machine agents, RGV agent will compare their bid value and select the machines of biggest benefit. This process was performed in proper sequence and divided into four phases: initial phase, sending messages, receiving& handling messages and stop, as shown in Fig. 6. 3.1.1. Initial phase The buffer of Socket in JAVA is checked continuously; when FMS sends messages to JADE using Winsock, it represents RGV starts running or finishes the preceding work, so agent is notified to start selection of next CNC. 3.1.2. Sending messages The agent prepares for sending messages, then establishes message templates, and finally sends out the messages. 3.1.3. Receiving& handling messages RGV agent starts to wait for the feedback from other agents, including: ACCEPT_PROPOSAL and REFUSE. The machine is listed into the catalogs of candidates when ACCEPT_PROPOSAL is received, or neglected when REFUSE is received. If the quantity of fed back messages is equivalent to that of sent messages, RGV agent will select the optimum handling machine from the candidates according to the performance index. 3.1.4. Stop If some machines are listed into the candidates, INFORM message is sent to the successful machine agent, and the network message of Dispatch Information will be transferred to notify FMS for
Fig. 5. Framework of agents.
Fig. 6. Flow chart of RGV agent.
handling, and finally returned to initial phase; if no machine is listed into the candidates (the received message of REFUSE is equal to that of sent messages), it is returned directly to initial phase. AS/RS agent is responsible for controlling the buffer station, and ensuring if the available buffers and parts are suitable for handling. When the message of RGV is received, the parts to be handled are selected depending on the bid value, and then the parts along with the bid value are fed back to RGV agent for reference. When the system is activated, the relevant messages from RGV agent are processed, and the behavioral model of processed parts is selected, allowing to make response when receiving the messages. The messages received by AS/RS agent are divided into CFP and INFORM; when CFP is received, it represents that RGV agent makes inquiry to AS/RS agent if handling is required; after receiving the message, AS/RS agent will make a judgment, and after confirming the message, calculate the performance index and feed back the message; when INFORM is received, it represents that RGV agent selects the machine on a priority, and prepares for handling the parts, with the flow process shown in Fig. 7. CNC agent is responsible for controlling CNC and calculating the bid value. After receiving the inquiry from RGV agent, CNC agent checks the current machine status and make a response; in the event of ongoing processing, or lack of parts on the machines or presence of parts on the destination machines, CNC agent rejects the demand of RGV agent, or feed back to RGV agent the received information along with the bid value after processing is finished. When the system is activated, the relevant messages from RGV agent are processed, and the behavioral model including performance index is selected, allowing to make response when receiving the messages. The messages received by CNC agent are divided into CFP and INFORM; when CFP is received, it represents that RGV agent makes inquiry to CNC agent if handling is required; after receiving the message, CNC agent makes a judgment, and after confirming the message, calculate the performance index and feed back ACCEPT_PROPOSAL; REFUSE is fed back if the parts are not handled; when INFORM is received, it represents that RGV agent selects CNC on a priority, and prepares for handling the parts, with the flow process shown in Fig. 8.
7314
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
Fig. 7. Flow process of AS/RS agent.
messages required for FMS, with the basic experimental environment shown in Fig. 9. In order to provide normal operation of FMS cell controller and multi-agent system, Socket packages are established in JAVA system, and Winsock packages established in FMS, enabling the work of both systems via network. There are three types of messages available for network transfer: Start, Dispatch Information and Server-ok. In the case of Start (see Fig. 10), it represents that the FMS (Cell Controller) begins with writing the working status of machines into the database, updating the processing status and notifying the agent platform (JADE) for calculation; next, the agent platform dispatch information of the calculated results to the simulated FMS; after the completion of the manufacturing system’s operation, Server-ok message will be fed back, allowing for next cycle of agent system. In the multi-agent system, RGV agent is mainly responsible for system operation in 4 steps. Step 0: the agent will check continuously the network messages transferred from the manufacturing system; in the case of Start, the agent starts to check the handling requirement in the datasheet, and enters into Step 1; in Step 1, RGV agent will send internal message to the machine agent (CNC, ASRS) according to the handling requirement, and enters into Step 2; in Step 2, internal response messages are received; the successful agents are selected when the quantity of response messages is equal to that of sent messages, then enters into Step 3; in Step 3, in the case of presence of successful agent, the messages are sent to this agent, and dispatched to Cell Controller; after receiving the message of Server-ok, it returns to Step 0.
3.2. Integration of multi-agent and FMS The experimental environment is divided into two parts: one is an FMS cell controller developed by Microsoft Visual Basic 6.0, which provides a real-time monitoring screen displaying the simulation process and result; the other one is a multi-agent system, responsible for controlling the status and process of manufacturing system based on JADE platform. In addition, information exchange is made possible by TCP/IP for normal coordinative operation of two systems, and Microsoft SQL Server 2005 is used to record the
Fig. 8. Flow process of CNC agent.
3.3. Description of empirical scenario There are three manufacturing scenarios in the system implementation (see Fig. 11); during input of raw materials, the materials are selectively processed in the First Section by checking if TC1/ TC2 is in idle state; in the Second Section, the products are transferred to the respective machines according to their part type; in the Third Section, processing is conducted through WM, LT, HC, IS. These scenarios are discussed in two parts: the one is simulated experiment of FMS with introduction of agent technology VS. without introduction of agent technology; the other one is simulated experiment of dynamic rules combination for various sections of manufacturing system with introduction of agent technology into FMS. The manufacturing process of three parts in the experiment is simulated in individual scenarios of FTL. Take Fig. 11 Scenario A, for example, the manufacturing process of Part 1 is: TC1/TC2 > MC1 -> WM -> LT -> HC -> IS; the manufacturing process of Part 2 is: TC1/TC2 -> MC2 -> WM -> LT -> HC -> IS; the manufacturing process of Part 3 is: TC1/TC2 -> MC3 -> WM -> LT -> HC -> IS. Scenario B, the manufacturing process of Part 1 is: TC1/TC2 -> MC1 > MC2 -> WM -> LT -> HC -> IS; the manufacturing process of Part 2 is: TC1/TC2 -> MC1 -> MC3 -> WM -> LT -> HC -> IS; the manu-
Fig. 9. A basic environment of this research.
7315
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
System Start Send Message -Start-
Check demand
Send internal message System Check Read the instruction Receive internal messages Selected objective Execution Task Send Message -Server ok-
Send Message
Assumption 1. The standby capacity of the system is infinite; when the processed parts are fed back to Buffer 1–16, new parts are immediately put into production. Assumption 2. All parts must be subject to the last IS (Inspection Station), and then fed back to Buffer 1–16 via RGVs. Assumption 3. All parts must be handled by WM, LT, HC and IS in the last section, and then fed back to Buffer 1–16 via RGVs. Assumption 4. The system’s manufacturing and processing time is 50 h.
-Dispatch Information-
3.4. Operation of FMS Fig. 10. Integrated system process.
This section aims at testing the operation of FMS; after setup the procedures, one can firstly see the main screen, which can be switched to simulated monitor by clicking Monitor Button without changing the default values; after selecting the dispatching rules such as FCFS, SIPT, Buffer Sequence and MultiAgentsFMS, press START button, or press STOP button to stop the operation of procedures. Fig. 12A is a screen in operation, which displays the current position of parts; if the parts are raw materials, it is displayed as yellow, if the parts are semi-finished products, it is displayed as green, and if the parts are finished products, it is displayed as red; the current automated RGV handling commands as well as the types, quality and quantity of finished parts are displayed below. The throughput, cycle time and utilization of machines can be shown in Fig. 12B–D, respectively. 3.5. Operation of multi-agent system
Fig. 11. Three manufacturing scenarios.
facturing process of Part 3 is: TC1/TC2 -> MC2 -> MC3 -> WM -> LT -> HC -> IS. Scenario C, the manufacturing process of Part 1 is: TC1/TC2 -> MC1 -> MC2 -> MC3 -> WM -> LT -> HC -> IS; the manufacturing process of Part 2 is: TC1/TC2 -> MC1 -> MC2 -> WM -> LT -> HC -> IS; the manufacturing process of Part 3 is: TC1/TC2 -> MC3 -> WM -> LT -> HC -> IS. After the user stops the system operation, the system will record the starting and stopping time, observe and analyze the utilization time of RGV and CNC, etc.
3.3.1. Agent VS. without agent This experiment simulated 50-h manufacturing process in three manufacturing scenarios (Fig. 11), wherein the experiments with/ without introduction of agents were compared, by using the dispatching rules: FCFS, SIPT and Buffer Sequence. Therefore, there are totally six possible combinations.
3.3.2. Dynamic dispatching rules with agent This experiment simulated another 50-h manufacturing process in ABC manufacturing scenarios (Fig. 11), wherein the experiments for various combinations of sections were compared using different combination of three dispatching rules: FCFS, SIPT and Buffer Sequence. Therefore, there are totally 27 possible combinations as shown in Table 1. The assumptions of FTL simulated experiments are listed below:
In this research, SNIFFER AGENT provided by JADE was employed to monitor the communication among agents on a multiagent platform. For example, Figs. 13–15 depicts the communication during operation of multi-agent system. It can be seen from
Table 1 Combination of dynamic rules. Combination
Section 1
Section 2
Section 3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
SIPT BUFF SIPT BUFF FCFS BUFF BUFF SIPT FCFS BUFF FCFS FCFS FCFS BUFF SIPT FCFS BUFF BUFF FCFS BUFF SIPT FCFS SIPT SIPT SIPT FCFS SIPT
BUFF SIPT SIPT BUFF BUFF SIPT FCFS BUFF FCFS BUFF SIPT SIPT BUFF SIPT SIPT FCFS BUFF FCFS FCFS FCFS FCFS BUFF FCFS BUFF FCFS SIPT SIPT
SIPT FCFS FCFS FCFS FCFS SIPT FCFS FCFS SIPT SIPT SIPT BUFF BUFF BUFF SIPT FCFS BUFF BUFF BUFF SIPT BUFF SIPT SIPT BUFF FCFS FCFS BUFF
7316
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
Fig. 14. Communications of multi-agent system (Section 2).
Fig. 12. Operating screen of manufacturing system.
No. 1–4 that, RGV agent sends inquiry message of CFP to ASRS1, ASRS2, TC1 and TC2, which feedback the messages separately after handling the messages. In No. 5–7, ASRS2, TC1 and TC2 feedback REFUSE to RGV agent after confirmation, but ASRS1 feeds back ACCEPT_PROPOSAL; in such case, RGV agent will decide ASRS1 as the successful agent according to the performance index attached in the feedback messages, and then send INFORM to it as shown in No. 9, the successful agent will make preparations after receiving INFORM. Prior to termination of the system, the agents make communication in such a way. Fig. 15. Communications of multi-agent system (Section 3).
4. Simulation results This experiment simulated 50-h manufacturing process in ABC scenarios, and discussed in two parts: simulated experiment of FMS with introduction of agent technology VS. without introduction of agent technology; simulated experiment of dynamic rules for various sections of manufacturing system with introduction of agent technology into FMS.
4.1. Agent VS. without agent This section analyzed the throughput of three parts by totalizing and comparing the throughput in various scenarios, and also observed the trend as listed in Figs. 16–18, respectively.
Firstly, as shown in Fig. 16, the agent system has the largest throughput. In addition, the trend of the bar chart indicates that, among the agent system, the difference of throughput for three rules differs little. In Fig. 17, the agent system has also the largest throughput, the trend of the bar chart indicates that, BUFF in the agent has the biggest throughput. In Fig. 18, the agent system has also the largest throughput, the trend of the bar chart indicates that, FCFS in the agent has the largest throughput. The experiment found that, the throughput can be increased efficiently with introduction of agent system, and the agent system has excellent performance of capacity in various scenarios. Thus, it is feasible to introduce multi-agent system into multi-section FMS. 4.2. Dynamic dispatching rules combination with agent The experimental results revealed that, it is possible to increase the capacity by introducing agent technology into FMS. Next, this
Fig. 13. Communications of multi-agent system (Section 1).
Fig. 16. Bar chart of throughput in scenario A.
7317
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318 Table 3 Ranking of throughput of single rule.
F–F–F S–S–S B–B–B
Fig. 17. Bar chart of throughput in scenario B.
Fig. 18. Bar chart of throughput in scenario C.
section analyzed 27 sets of dynamic rules by totalizing the throughput of three parts and comparing the throughput in various scenarios, and also observed the trend to evaluate the feasibility of dynamic rules. Table 2 lists the throughputs of top 3 and bottom 3 agents in various scenarios. As shown in Table 3, the throughput using single rule is not the optimal throughput as for the arrangement of dispatching rules at various sections, so it is necessary to combine the dynamic rules to optimize the system throughput. The experiment found that, the agent system combined with dynamic dispatching can increase efficiently the capacity, and various sections have excellent performance of capacity with dynamic dispatching in various scenarios. Thus, it is feasible to introduce the agent system combined with dynamic dispatching into Multi-Section FMS.
5. Conclusions As compared with single-section FMS, the multi-section FMS features more complex operation, more product categories and more flexible manufacturing process, making system control and
Table 2 Ranking of throughput of dynamic rules.
No. No. No. ... No. No. No.
1 2 3 25 26 27
Scenario A
Scenario B
Scenario C
S–B–S B–S–F S–S–F ... F–S–F S–F–F S–S–B
F–B–S B–B–B F–B–F ... S–F–S S–F–F F–F–B
B–S–F S–S–F S–F–F ... F–B–S F–B–B S–B–F
Scenario A
Scenario B
Scenario C
No. 16 No. 15 No. 17
No. 4 No. 13 No. 2
No. 4 No. 22 No. 20
dispatching more important to the performance of manufacturing system. With introduction of the concept of agent technology, the multi-section FMS combines the agent technology with the control modules of various machines, and establishes a perfect simulated manufacturing system based on the planning of manufacturing modes. The message transfer and coordination among agents is made possible through the communication language, or through a coordination mechanism based on the contract network in the event of conflict of assigned tasks. This can select the priority machines and deliver services for maximizing the overall benefits of the system and efficient allocation of system resources. Multi-section FMS (Cell Controller) exchanges information with multi-agent system (JADE) through network technology, but many manufacturing processes or system status have to be provided from database. The advantage of agent system is that it can prevent the decline of system stability, shorten the development time and increase the operating efficiency. Moreover, the agent technology can sharpen the enterprises’ competitive edge by introducing new technologies or new algorithms quickly and stably. Thus, establishing a perfect agent system and defining its coordination criterions can guarantee satisfactory operating capacity while the tasks assignment of manufacturing system is not limited to certain algorithm or mechanism. With the introduction of agent technology, the existing manufacturing system is only required to handle partially the hardware assignment commands, making it possible to shorten greatly the development time and reduce the maintenance burdens. Limited literatures pay attention to the control of multi-section FMS. This is the first paper to apply multi-agent technique in multi-section FMS. This study integrated the multi-agent system with multi-section FMS based on the established agent knowledge base. It is suggested that the dispatching rules should be applied to the optimal combination in various sections, and independent agents should take charge of judging the scenarios of dynamic rules for increasing the system efficiency. Acknowledgment The authors would like to give thanks to the two anonymous reviewers for their valuable suggestions and comments. This work is partially supported by National Science Council, Taiwan under Grant NSC 97-2221-E-027-076. Appendix A. Appendices The following are abbreviated terms and definitions used in this paper: FCFS: first-come-first-served dispatching rule is implemented by simulated FMS. MAS-FCFS: FCFS is implemented by multi-agent system (MAS). SIPT: shortest imminent processing time dispatching rule is implemented by simulated FMS. MAS-SIPT: SIPT dispatching rule is implemented by multi-agent system (MAS). BUFF: buffer sequence dispatching rule is implemented by simulated FMS.
7318
K.-Y. Chen, C.-J. Chen / Expert Systems with Applications 37 (2010) 7310–7318
MAS-BUFF: buffer sequence dispatching rule is implemented by multi-agent system (MAS). TC1, TC2: turning machine. MC1, MC2, MC3: milling machine. WM: washing machine. LT: leak testing machine. HC: horniness center. IS: inspection station. Utilization: utilization of machines: sum of processing time of machines / total actual operating time of system. Throughput: total output of various parts. MLT: manufacturing lead time, defined as the average processing time from the first to the last station.
References Bellifemine, F., Poggi, A., & Rimassa, G. (2001). Developing multi-agent systems with JADE. Intelligent agents. Springer-Verlag. Choi, K. H., Kim, S. C., & Yook, S. H. (2000). Multi-agent hybrid shop floor control system. International Journal of Production Research, 38(17), 4193–4203. Deng, Y., & Yang, C. R. (1999). Architecture-driven modeling of real-time concurrent systems with application in FMS. Journal of Systems and Software, 45(1), 61–78. Farahvash, P., & Boucher, T. O. (2004). A multi-agent architecture for control of AGV systems. Robotics and Computer-Integrated Manufacturing, 20(6), 473–483. Ferber, J. (1999). Multi-agent systems – An introduction to distributed artificial intelligence. Addison-Wesley. Fox, M. S., Barbuceanu, M., & Teigen, R. (2000). Agent-oriented supply-chain management. The International Journal of Flexible Manufacturing System, 12(2), 165–188. Groover, M. P. (2001). Automation, production systems, and computer-integrated manufacturing (2nd ed.). Prentice-Hall. Huang, C. Y., & Nof, S. Y. (2000). Formation of autonomous agent network for manufacturing systems. International Journal of Production Research, 38(3), 607–624. Huang, W. J. (2004). The development of communication protocols for production planning in distributed manufacturing systems, master thesis, Tunghai University, Taiwan.
Ishida, T. (1994). Parallel, distributed and multiagent production systems. SpringerVerlag. Jennings, N. R., Sycara, K., & Wooldridge, M. (1998). A roadmap of agent research and development. Autonomous Agents and Multi-agent Systems, 1(1), 7–38. Kouiss, K., Pierreval, H., & Mebarki, N. (1997). Using multi-agent architecture in FMS for dynamic scheduling. Journal of Intelligent Manufacturing, 8(1), 41–47. Leitao, P. (2009). Agent-based distributed manufacturing control: A state-of-art survey. Engineering Applications of Artificial Intelligence, 22, 979–991. Li, X., Geng, T., Yang, Y., & Xu, X. (2002). Multiagent AGVs’ dispatching system using multilevel decisions method. American Control Conference, 2, 1135–1136. Maleki, R. A. (1991). Flexible manufacturing systems (3rd ed.). Prentice Hall. Ming, L. (1996). Agent-oriented analysis methodology in intelligent manufacturing system. The 4th International Conference on Control, Automation, Robotics and Vision, 254–258. Moore, M. L., Reyns, R. B., Kumara, R. T., & Hummel, J. R. (1997). Distributed intelligent agents for logistic (DIAL). Computational Cybernetics and Simulation, 3, 2782–2787. Nagarjuna, N., Mahesh, O., & Rajagopal, K. (2006). A heuristic based on multi-stage programming approach for machine-loading problem in a flexible manufacturing system. Robotics and Computer Integrated Manufacturing, 22(4), 342–352. Ouelhadj, D., Hanachi, C., & Bouzouia, B. (2000). Multi-agent architecture for distributed monitoring in flexible manufacturing system. IEEE International Conference on Robotics and Automation, 3, 2416–2421. Reaidy, J., Massotte, P., & Diep, D. (2006). Comparison of negotiation protocols in dynamic agent-based manufacturing systems. International Journal of Production Economics, 99, 117–130. Smith, R. G. (1980). The contract net protocol: High-level communication and control in a distributed problem solver. IEEE Transactions on Computers, 29(12), 104–1113. Stone, P., & Veloso, M. (2000). Multiagent system: A survey from a machine learning perspective. Autonomous Robots, 8(3), 345–383. Stuart, R., & Norvig, P. (1995). Artificial intelligence: A modern approach. Prentice Hall. Turgay, S. (2009). Agent-based FMS control. Robotics and Computer-Integrated Manufacturing, 25, 470–480. Viswanadham, N., & Narahari, Y. (1992). Performance modeling of automated manufacturing systems. Prentice Hall. Wooldridge, M. (1997). Agent-based software engineering. IEE Proceedings Software Engineering, 144(11), 26–37. Yamanoto, H., & Marui, E. (2005). Intelligent communication between agents of autonomous decentralized FMS. IEEE International Symposium on Computational Intelligence in Robotics and Automation, 433–438.