Procedia CIRP 7 ( 2013 ) 437 – 442
2212-8271 © 2013 The Authors. Published by Elsevier B.V.
Selection and peer-review under responsibility of Professor Pedro Filipe do Carmo Cunha doi: 10.1016/j.procir.2013.06.012
Forty Sixth CIRP Conference on Manufacturing Systems 2013
Agent based manufacturing simulation
for efficient assembly operations
Yasuhiro Sudo
a*, Michiko Matsuda
aaKanagawa Institute of Technology, 1030 Shimo-Ogino, Atsugi , Kanagawa, 243-0292 Japan
* Corresponding author. Tel.: +81-46-291-3194, fax: +81-46-242-8490, E-mail address: [email protected]
Abstract
This study experiments with the manufacturing efficiency by layout change of a factory by means of agent-based autonomous production scheduling, using the virtual factory on a multi-agent simulation system. As infrastructure software for agent based simulation, the artisoc(c) is used. In this virtual factory, three types of agents are equipped. Users can alter a configuration such as input new jobs, adjusting a machine setting, etc, with monitoring conditions of agents. As a result, by adjustment of the agent's behavior with shop floor detail, the assembly schedule becomes more effective. The experiment is carried out to show that local negotiations contribute to global optimization when resources in the factory are effectively distributed and shared. In this paper, the effectiveness of job-list cleanup method is shown. In addition, the scheduling influence is simulated by the communication range of agents. A part agent chooses a machine, by the length of a job list and the conveyance cost. But the communication cost between agents increases with the size of the communication range. From experimental results, when extending the communication range simply, the conclusion is reached that optimization did not necessarily result in progress.
© 2013 The Authors. Published by Elsevier B.V.
Selection and/or peer-review under responsibility of Professor Pedro Filipe do Carmo Cunha Keywords: Virtual factory; Dynamic scheduling; Multi agents; Concurrent engineering;
1. Introduction
Recently a manufacturing system is shifting into mass production, a high-mix very low volume production and flexible order-made production to respond to customer needs. With this conversion production planning becomes more complicated, and researches on production scheduling have taken a technological turnaround.
Because the solution space is too large, the mathematical programming based on job-shop scheduling is powerless to obtain the optimal solution. To build a practical production plan adopting suboptimal solutions is essential, using a parallel processing and problem reductions. With such background, autonomous manufacturing systems using multi-agents have been proposed [1-2]. In such autonomous manufacturing system, a production plan is generated autonomously and dynamically, using communication and negotiation
between agents that correspond to factory components. As a result, the system has flexibility and continuous activeness. Even when an emergent trouble occurred, the necessity for a fresh start had been reduced. Each agent's own action is determined by reference to simple rules and local negotiation, the assembly operation progresses autonomously [3-4]. This agent based system adjusts the production schedule dynamically using only local negotiation when conditions have to be changed. The most important feature of this architecture is, there is no manager that controls the factory as a representative.
Previously, authors proposed a method that is using agents for decentralized autonomous control. This autonomous assembly type system consists of the following: assembly machine agents, product agents and parts agents [5-8]. In addition, it has also discussed the possibility of implementation as a multi-agent system [9]. In this paper, the experimental results that the effectiveness of changing in behavior and parameters of the agent are shown, using the virtual factory built on a © 2013 The Authors. Published by Elsevier B.V.
Selection and peer-review under responsibility of Professor Pedro Filipe do Carmo Cunha
Open access under CC BY-NC-ND license.
multi-agent simulator artisoc(c) [10]. The first experiment is about the effectiveness of the job-list cleanup method. Next is the changing size of the communication range of agents. As a result, it was found that both parameters had some influence on production planning.
2. Agent based autonomous assembly system
In the autonomous manufacturing system, a production plan is generated autonomously and dynamically, using communication and negotiation between agents that correspond to factory components. 2.1. Structure of the autonomous assembly system
The structure of a traditional manufacturing system is a device oriented structure. There are several attempts to construct such kind of autonomous decentralized manufacturing system. One example is the holonic manufacturing system [11-13]. The elements of the system act autonomously. As a result, they organize the system cooperatively. Usually some manager functionality is installed as an agent or blackboard. If the manufacturing system doesn’t need the manager function that manages the entire system by controlling each agent, it becomes a more flexible system. This means that the system should be constructed as an event driven type system. As a solution for the above mentioned requirements, a configuration of a work-piece agent and a machine tool agent for an autonomous machining system has been introduced [4-5]. In such manufacturing systems, a manufacturing activity unit such as a machine tool, assembly machine, robot, AGV (Automatic Guided Vehicle), and manufacturing cell has autonomous functionalities that are configured as agents. In these cases, the system structure was a device driven system structure (Fig. 1).
Moreover about the implementability of such a software agent have been also discussed. A part is just a part, it is impossible to install an agent into parts. To install an agent’s capability into pallets (trays) which are used in transporting parts is proposed [9]. Parts are transferred to an assembly machine by AGV or a kind of conveyor belt. In this case, the container is used in general if packaging is considered. This pallet is reusable to install agents; it contributes to the realization of the agent based autonomous assembly system. 2.2. Characteristics of agent based system
The system consists of a distributed structure, it has flexibility and is active continuously. Even when an emergent trouble occurs, the necessity for a fresh start had been reduced. Each agent's own action is determined
Shop Floor
New Product Order Product’s Specification
Control Yard
...
Product Agents
Assembly Machine Agents
Work’s Specification Work’s Specification Work’s Specification Work’s Specification Work’s Specification Work’s Specification Assembly Request Assembly Request Assembly Request Assembly Request Assembly Request •Product model •Production plan •Arrow diagram Shop Floor
New Product Order Product’s Specification
Control Yard
...
Product Agents
Assembly Machine Agents
Work’s Specification Work’s Specification Work’s Specification Work’s Specification Work’s Specification Work’s Specification Assembly Request Assembly Request Assembly Request Assembly Request Assembly Request •Product model •Production plan •Arrow diagram
Fig. 1. Outline of an agent based assembly system
by reference to simple rules and local negotiation. Then, the assembly operation progresses autonomously. Accordingly, transformation of scheduling results from agent's behavior and factory parameter had been further explored [14-16]. In such an agent based system, since there are too many indefinite elements, in order to predict the behaviour of a factory line, the computer simulation is indispensable. In recent years, concurrent engineering with a digital virtual factory has attracted attention. These simulation systems are aimed at cost saving, shortening development time, improved quality etc. This means that imperfections, troubles and problems can be found by using pre-manufacturing computer simulation.
3. Agents in the autonomous assembly system
The established autonomous assembly system is structured of three kinds of agents [5-9]. The required functional capabilities of each agent are as follows: 3.1. The product agent
The product agent generates the assembly work process from the product model. Based on this work process, product agents put parts agents onto the shop floor. After that, product agents watch delays of operations, and check a change of the deadline and quantity. When there is a need for a rescheduling plan, the product agent notifies parts agents of the related information. If a new
machine problem arises, the trouble information is sent to parts agents, they will then re-select another assembly machine. If each parts agent does work greedily, it is not an efficient assembly production line. Each parts agent must know its own priority for the assembly job. The priority is derived from an arrow-diagram obtained by the assembly process model. The product agent’s capabilities are as follows:
Generation of the assembly process plan Generation of the parts agent
Communicating ability with parts agents
Management of the arrow diagram using feedback of the working results
3.2. The parts agent
Parts agents correspond to each assembly work. They have the assembly process model handed by the product agent, and select the assembly machine based on the estimated result. The scheduling optimization process is generated by negotiations between parts agents, asynchronous distributed in real-time. When a priority change of a deadline is notified by the product agent, the parts agent redoes the selection of the assembly machine according to the contents. Parts agents must have abilities as follows:
Communication capability with other agents Computation for the operations completion Request the estimate to an assembly machine Request of conveyance
3.3. The assembly machine agent
The assembly machine agent has its own machine model that is described by specifications and capabilities of the assembly machine. The machine agent manages its own operation schedule using this machine model. Moreover, the machine agent checks itself and notifies the assembly machine’s condition to other agent. Machine agents must have abilities as follows:
The management function of a work list Correspondence to work estimated request
(Calculation of the completion time of work) The notice function of work schedule delay Arrangements of attachment parts and pickup
4. Construction of the virtual factory
The above mentioned agent based manufacturing system does not have advantages over other manufacturing systems with respect to every factor. However, the proposed system has superior performance under a particular set of conditions. The virtual production plant built on a computer is called a virtual factory [17-18]. It can be used for probing the problem
Parts Agent #1 Machine Agent a1 Parts Agent #2 Parts Agent #3
Assembly Simulation System
Simulation Manager Simulation Monitor Machine Agent a2 Machine Agent b1 Virtual Factory Virtual Factory Machine Data Production Plan Data Parts Agent Generator Shop Floor Configurator Product Data XML XML Operator Parts Agent #1 Machine Agent a1 Parts Agent #2 Parts Agent #3
Assembly Simulation System
Simulation Manager Simulation Monitor Machine Agent a2 Machine Agent b1 Virtual Factory Virtual Factory Machine Data Production Plan Data Parts Agent Generator Shop Floor Configurator Product Data XML XML Operator
Fig. 2. The structure of the assembly simulation system [14]
of the manufacturing line in operation, or attaining an increase in efficiency when newly designing a factory. Moreover, it is possible to perform a simulation when parts of the plant stop working due to an accident or power failure.
On the other hand, in the autonomous manufacturing system that is using a software agent, cooperation with computer systems is needed. Moreover a lot of agents are autonomous-decentralized and actively act, thus, a prior simulation is indispensable. As a consequence, development of the virtual factory as a verification system [14-16] has also been performed simultaneously with constructing an autonomous manufacturing system.
Fig. 2 shows the structure of the assembly simulation system. This simulation system is structured on the virtual assembly factory. The operator can set up the initial shop floor configuration through the user interface. The simulation manager generates parts agents from product models, and generates machine agents from machine models. Product models and machine models are input as XML description files. The simulation monitor shows progress status and condition of assembly processes from the product view and assembly machine view. This assembly simulation system is prototyped using the multi-agent simulator called artisoc(c) [10] as a development environment. Fig. 3 shows simulation displays. In the virtual assembly factory, each type of agent is implemented with the following key features are shown in Fig. 3:
(A) Preview window: the agent's position can be dynamically checked by movement of icons.
(B) Machine agent viewer: Supervise the situation of each assembly machine (job lists, developed power, mechanical condition, etc.)
(C) Work-advance graph: checking the completion rate of products
Fig. 3. Prototyping of an assembly simulation system
(E) Control Panel (production request etc.)
(F) The work log of assembly machines (with a file output function)
The input of a programming level is needed for performing a detailed setup and control. It is possible to copy the position and conveyance course of an operating machine with the layout of a real factory. And the user is able to change and adjust a parameter in real time and visually. In recent years, the design of a dynamic factory or production control which used such a system is attracting attention, it is called concurrent engineering.
5. Improvement of productivity
In the agent based assembly system model presented, after a part agent selects the assembly machine, it does not negotiate for a change of the order to other parts agents. Here, it is excluded in case that product agent announces a change in the time for delivery. The following method is newly proposed for improving productivity.
5.1. The job list clean-up method
A machine agent locks the order of the job List when a real job is imminent. Then the parts agent starts preparation of assemblies, such as supply of sub-assembled parts. A machine agent transposes the sequence of operations on a small scale only within the case where the influence is sufficiently small at the time
Job List
...
Approach (Order Locked)
1.
2. 3.
4.
Assembly Machine 1. Checking work type2. Searching for a similar job 3. Notification of the delay 4. Interruption
5. Preparation of assembled parts 5.
Job List
...
Approach (Order Locked)
1.
2. 3.
4.
Assembly Machine 1. Checking work type2. Searching for a similar job 3. Notification of the delay 4. Interruption
5. Preparation of assembled parts 5.
Fig. 4. The job list clean-up method
Shop Floor Human Worker
Screwing Machine Bonding Machine
Bonding
Machine Human Worker
Screwing Machine
Shop Floor Human Worker
Screwing Machine Bonding Machine
Shop Floor Human Worker
Screwing Machine Bonding Machine
Bonding
Machine Human Worker
Screwing Machine
Fig. 5. The layout of assembly machines and human workers (1)
for delivery, just before approaching a limit. In other words, a machine agent tries to reduce the time for initial set-up and tool change by continuously processing similar type jobs. In this regard, a check is not carried out to the tail end of a list, it is limited to near the approach line (Fig. 4).
5.2. Verification of the job-list cleanup method
The effectiveness of the job-list cleanup method is shown using a virtual factory demonstration. Fig. 5 shows the layout and number of assembly machines. Table 1. shows parameters of manufactured products from the simulations. Two types of products were assembled. Straight type mobile phones which have two assembly sequences and the flip type mobile phones which have twelve sequences. On the shop floor, there is one bonding machine and one screwing machine and 6 human workers. Human workers can do both of the operations, but it requires time for changing the tools used.
The bar graph in Fig. 6 shows the relations between the number of steps to the completion of the work and
(B)
(A)
(F)
(A)
(E)
(D)
(B)
(A)
(F)
(C)
(E)
(D)
Table 1. Data of assembled products used in simulations Category of Products Straight type phone Flip-type phone
Number of Parts 5 19
Assembly Sequences 1 bonding and 1 screwing
9 bonding and 3 screwing
Amount of Assembled Products [lot]
Steps Transpose Times
Amount of Assembled Products [lot]
Steps Transpose Times
Fig. 6. The relations between amount of assembled products and number of transpose incidences
the amount of assembled products
.
The latter number is proportional to the length of the waiting job-list for a machine. The line chart in Fig. 6 shows the frequency of transpose incidences. The clean-up method does not run when assembly machines (including human-workers) have few jobs. On the other hand, when the work list of a machine agent becomes long it may be associated with a high probability of an existence of the same kind of work in the job-list, consequently the number of times of calling the clean-up method increases. As a result, the length of time required to finish decreases up to 4% compared to the case without the clean-up method. This means reduced time for the exchange of tools when human-workers take up the next work. This method is an effective thing especially when employing a general-purpose machine in which various types of assembly processing could be utilized.6. Influence of changing communication range
Usually, the same configuration of machines is applied to various products in small-lot production. Therefore the efficiency depends on placements of resources. A part agent chooses a machine, by the length of a job list and the conveyance cost to get the parts there. But the communication cost between agents increases with the size of the communication range (Fig. 7). Therefore cost reduction will be also possible. The desired extent of the communication range can also be tested.
Shop Floor
Parts Agent Parts Agent
Smaller Scope
Larger Scope Shop Floor
Parts Agent Parts Agent
Smaller Scope Larger Scope
Fig. 7. The size of the agent’s communication range
Bonding Machine Human Worker Screwing Machine Bonding Machine Human Worker Screwing Machine
Fig. 8. The layout of assembly machines and human workers (2)
20 lots of Straight type phones and 30 lots of flip type phones are put onto the assembly line. The parameters of productions are the same as in Table 1. Each of the assembly machines are laid out as indicated in fig. 8. Simulation results are shown in Fig. 9, where the viewing ranges of parts agents were set at 20%, 30%, 40%, 50%, 60%, 70% and 80% of the diagonal diameter of the shop floor. The vertical axis means the number of steps to the completion of the work, and total distance of parts agents moved. When the communication range is narrow, the dispersion efficiency of work is bad because requests for work may concentrate on a nearby machine. On the other hand, if the communication range is extended, the length of movement tends to increase since a vacant machine located at a further distance may be chosen. With the factory composition of this experiment, while parts agents are acting in about 40% to 50% of the range, the operation step has been relatively small. From these results, it is necessary to provide a suitable parameter according to the content to optimize the situation.
Steps Distance
Ratio of Agent’s Communication Ranges to Floor Size [%]
Steps Distance
Ratio of Agent’s Communication Ranges to Floor Size [%]
Fig. 9. The simulation results on changing agent’s viewing scopes 7. Conclusions
The first experimental result showed the effectiveness of the job list clean-up method. The next one showed the relationship between agents’ viewing scopes and conveyance costs. From the second experimental result, when simply extending the communication range, the conclusion that optimization did not necessarily progress was obtained. In the case of such very large scale manufacturing, the autonomous decentralized assembly system may have advantages. One item for future work is designing an agent's algorithm according to various purposes, such as not only shortening manufacturing time but also reducing energy consumption.
Acknowledgements
The authors are grateful to Dr. Udo Graefe, retired from the National Research Council of Canada for his helpful assistance with the writing of this paper in English.
References
[1] Monostori, L., Váncza, J., Kumara, S., 2006, “Agent-based systems for manufacturing”, CIRP Annals-Manufacturing Technology, vol. 55, no. 2, p. 697.
[2] Fujii, N., Kobayashi, M., Makita, T., Hatono, I. and Ueda, K., 2004,”Integration of Facility Planning and Layout Planning Using Self-Organization in Semiconductor Manufacturing”, Proceedings of the 37th CIRP-ISMS, p. 175.
[3] Matsuda, M., Ishikawa, Y. and Utsumi, S., 2006,”Configuration of Machine Tool Agents for Flexible Manufacturing”, Proceedings of the 39th International Conference on Manufacturing Systems, p. 351.
[4] Matsuda, M., and Sakao, N., 2008,”Configuration of An Autonomous Decentralized Digital Factory Using Product and Machine Agents”, Innovation in Manufacturing Networks, IFIP vol.266, p. 215.
[5] Sakao, N., Sudo, Y. and Matsuda, M., 2008,”Product and Machine Agents for an Autonomous Assembly Production System”, Proceedings of Joint 4th International Conference on Soft Computing and Intelligent Systems and 9th International Symposium on advanced Intelligent Systems, p. 1271.
[6] Sakao, N., Matsuda M. and Sudo, Y., 2009,”Assembly planning for an autonomous decentralized manufacturing system led by a product part agent”, Proceedings of 42nd CIRP International Seminar on Manufacturing Systems (CD-ROM).
[7] Sudo, Y., Sakao, N. and Matsuda, M., 2010,”An Agent Behavior Technique in an Autonomous Decentralized Manufacturing System”, Journal of Advanced Mechanical Design Systems and Manufacturing Vol. 4, No. 3, p. 673.
[8] Matsuda, M., Sakao, N., Sudo, Y. and Kashiwase, K., 2010,”Flexible and Autonomous Production Planning Directed by Product Agents”, Proceedings of the 43rd CIRP International Conference on Manufacturing Systems, p. 876.
[9] Sudo, Y., Kashiwase, K. and Matsuda, M., 2011, “The Implementability of Agent Based Autonomous Decentralized Assembly System”, Proceedings of International Symposium on Scheduling 2011, p. 101.
[10] Kozo Keikaku Engineering Inc., Artisoc User Manual English Edition, http://mas.kke.co.jp/modules/tinyd0/index.php?id=9 [11] Brussel, H. V., Wyns, J., Valckenaers, P., Bongaerts, L. and Peeters,
P., 1998,”Reference architecture for holonic manufacturing systems: Prosa,” Computers in Industry, vol. 37, no. 1, p. 255. [12] McFarlane, D.C. and Bussman, S., 2000,”Developments in Holonic
Production Planning and Control”, Production Planning and Control, vol. 11, no. 6, p. 522.
[13] Sugimura, N., Shrestha, R. and Inoue, J., 2003, ”Integrated process planning and scheduling in holonic manufacturing systems -Optimization based on shop time and ma-chining cost”, Proceedings of the 2003 IEEE Interna-tional symposium on Assembly and task planning (ISATP2003), p. 36.
[14] Matsuda, M., Kashiwase, K. and Sudo, Y., 2011, “Configuration Of A Digital Factory For Autonomous Virtual Manufacturing”, Proceedings of 21st International Conference on Production Research, -Innovation in Product and Production (CD-ROM) [15] Matsuda, M., Kashiwase, K. and Sudo, Y., 2012,”Agent oriented
construction of a digital factory for validation of a production scenario”, Procedia CIRP (Special Issue on) 45th Conference on Manufacturing Systems, Elsevier, vol.3, p.115.
[16] Sudo, Y., Kasiwase, K. and Matsuda, M., 2012,”Verification of scheduling efficiency of an autonomous assembly system using the multi-agent manufacturing simulator”, Proceedings of the ASME 2012 International Symposium on Flexible Automation. [17] Bley, H., Franke, C., 2004, Integration of product design and
assembly planning in the digital Factory, CIRP Annals - Manufacturing Technology, Vol. 53, Issue 1, pp. 25-30. [18] Butterfield, J., Crosby, S., Curran, R., Price, M., Armstrong, C. G.
and Raghunathan, S., 2007,”Optimization of Aircraft Fuselage Assembly Process Using Digital Manufacturing”, Journal of Computing and Information Science in Engineering, Vol. 7, No. 3, p. 269.
