- Email: [email protected]
Dynamic structuring of distributed manufacturing systems
Advanced Engineering Informatics 16 (2002) 127±133
www.elsevier.com/locate/aei
Dynamic structuring of distributed manufacturing systems Peter Butala*, Alojzij Sluga Department of Control and Manufacturing Systems, University of Ljubljana, Askerceva 6, SI-1000 Ljubljana, Slovenia Received 7 May 2001; accepted 1 March 2002
Abstract The paper addresses the problem of dynamic structuring of manufacturing systems. The approach presented in this paper is based on the decomposition of manufacturing objectives and the allocation of tasks to autonomous building blocks, i.e. work systems, in a dynamic environment. The allocation is based on a market mechanism that enables the self-structuring and optimization of a manufacturing system by evaluation and selection among competing work systems. The mechanism presented is based on logic relations and constraints. It enables the building of task-oriented manufacturing structures of work systems acting in series and/or in parallel. The approach is discussed in an example in the part fabrication domain. q 2002 Elsevier Science Ltd. All rights reserved. Keywords: Modeling; Work system; Fabrication; Multi-agent structure; Market mechanism; Virtual cluster
1. Introduction The rising complexity of products, production structures and processing procedures on one side and turbulent market excitations resulting in growing product variety, individualization and shortening time frames on the other are setting new frontiers for the manufacturing business. Despite research efforts and investments in the context of computer integrated manufacturing the existing manufacturing systems are still predominantly based on the obsolete Taylorian philosophy. Therefore they cannot adequately conform to these requirements because of their structural rigidity, deterministic approach to decision making in a stochastic environment, hierarchical allocation of competencies, and insuf®cient communication and exploitation of expertise. In order to face these challenges in manufacturing a shift of the existent manufacturing paradigm from deterministic into a new manufacturing prospect considering natural understanding and concern is needed. Several in¯uencing approaches are emerging. Fractal factory [1], bionic manufacturing systems [2], Holonic manufacturing systems [3] and distributed manufacturing systems [4] are some concepts that are making an appearance. Based on these concepts several advances to modeling a manufacturing system in terms of viable structures for more effective mastering of complex and dynamic behavior of the system * Corresponding author. Tel.: 14774-753; fax: 12518-567. E-mail address: [email protected] (P. Butala).
and its environment are being developed. Various contributions in this direction have been presented and intensively exercised recently [5±9]. In the paper an approach to dynamic structuring of manufacturing systems is presented. It is based on the market mechanism which induces a self-organizing principle and is implemented in the system entitled the adaptive distributed manufacturing system (ADMS). Section 2 brie¯y describes the context of distributed manufacturing systems and related works. Section 3 presents the concept of ADMS and the underlying principles. The case study in Section 4 illustrates the abilities of the system in the part fabrication domain. Particularities of ADMS and distinctions to related works are discussed in Section 5. 2. Issues of distributed manufacturing systems A market is the key segment of the environment which sets primary objectives for manufacturing systems. It demands innovative, customized products and quick responses. The character of the market is dynamic and stochastic. Traditional manufacturing systems cannot cope with market requirements due to: ² Time response to predominantly random excitations of the markets is unsatisfactory, too slow and in many instances too costly. ² Information is incomplete, inaccurate and unreliable.
1474-0346/02/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved. PII: S 1474-034 6(02)00007-1
128
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
Decision-making is based more or less on 'guessing', predominantly using the rule of thumb [10]. ² Organizational structures are predetermined and rigid. Traditional structuring of manufacturing systems is based on labor division and the optimization of performance is based on central planing and control [11]. Let us consider an example of a ¯exible manufacturing system (FMS). It is the most advanced structure of a traditional manufacturing system. It has built-in scenarios to anticipated contingencies. Its scope is ®xed and limited. Its structure is predetermined and cannot be modi®ed by adding or removing elements in a short period of time. Therefore it is a closed system. In order to overcome the earlier mentioned disadvantages of manufacturing systems, new, ¯exible and adaptable organizational structures capable of performing self-organization of the work process have to be established [10]. Several systems exist which are capable of adapting themselves to changes in the environment. Biological systems are an example of such systems that exhibit characteristics such as self-recognition, self-growth, self-recovery and evolution [2]. Another example of a self-organizing system is the social system called the market economy. It is an economic system controlled, regulated, and directed by markets alone; order in the production and distribution of goods is entrusted to this self-regulating mechanismÐthe `invisible hand' as noted by A. Smith two centuries ago. Both types of systems have inspired researchers to adopt their basic principles in the manufacturing world. Ueda [2] introduced biological manufacturing systems by mapping the analogies from the living world. In his latest work Ueda [7] expanded this concept into interactive manufacturing which recognizes the vital role of a human in manufacturing as proposed by Peklenik [10]. Various authors have adopted the micro-economic model of self-organization on the production planning and control scale. Some promising results have already been obtained [5,6,8,9,12]. The common observations can be summarized as follows: (1) a manufacturing system is a complex system, (2) the architecture is the essential issue, (3) the role of a human subject is dominant, (4) the information and communication technologies are the enabling technologies, and (5) the transition from highly data-driven to particular information, knowledge and learning driven organization is needed. Therefore new features of future manufacturing systems have to be developed. The most expected ones are: (1) an open multi-level architecture, (2) advanced communication capabilities, (3) decentralized decision making, (4) selfstructuring ability, and (5) rede®nition of the work systems in terms of autonomy, evolutionary adaptivity, re-con®gurability, co-operativeness, interactivity, task orientation within competence, ability of communication, coordination and co-operation, and learning capability. This research is focused on the problem of how a manu-
facturing system can be structured for a given objective (e.g. the realization of a product) and how the system can adapt its structure in the case of external and/or internal disturbances (e.g. machine break down). The key distinction between ADMS and other approaches [5,6,8,9] lies in the level of system decomposition. In other approaches building blocks are entities that model the factory's functions and/or physical entities (e.g. orders, parts, and resources). However, they do not represent a manufacturing structure and thus the reduction of complexity is limited. In ADMS the complexity is managed more ef®ciently as the building blocks are systemsÐelementary work systems (EWS). 3. Agent based approach to dynamic structuring in ADMS ADMS is based on the concept of distributed manufacturing systems. It is structured as a network of building blocks (work systems) acting as agents. Thus it represents an organic structure of interrelated building blocks which are acting in parallel and/or in series and are driven by cooperation and competition on various levels [4,13]. This concept synthesizes the agent structure as a decision network in order to optimize the overall system's behavior according to the current state of the environment. Structuring is thus an important issue for the achievement of the optimal performance of a manufacturing system. A single optimal structure for different manufacturing tasks cannot be de®ned. For each particular task the optimal structure has to be built up. A feasible approach to the structuring of the system is to structure the task ®rst. The complex task can be structured by the decomposition of the task into less complex tasks. Thus an incompletely speci®ed problem is transformed into a set of more speci®ed ones. Here it is assumed that complex tasks are decomposed into elementary tasks which can be executed by elementary work systems. The structuring process implies the incorporation of work systems for task execution. It is based on the market mechanism. The market mechanism is characterized by the operation of the self-adjusting forces of supply and demand. The motivation in the market mechanism is very basic: self-interest with respect to the economic gain. While self-interest may or may not be an universal part of human nature it is certainly a powerful motivating force. 3.1. Basic building blockÐthe work system In order to realize the concept one has to de®ne the basic building blocks with autonomous behavior and as having the competence and capabilities to perform a particular manufacturing operation. For the structuring of a manufacturing system a generic building block was introduced by Peklenik [13] in terms of the elementary work system. An EWS is capable of
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
129
Fig. 1. Elementary machining work system and its virtual counterpart.
performing single manufacturing tasks, e.g. process planning, machining. It consists of hardware elements necessary to implement a work process, work process identi®cation for process control and optimization, and a human operator as an autonomous subject for making decisions and synthesis. In ADMS a virtual work system (VWS) is introduced in order to delegate the EWS in a distributed environment. The VWS is an agent and it represents the EWS as its counterpart in the information space. Fig. 1 shows the composed structure of the EWS and VWS. Fig. 1 exhibits a generic structure of a computer numerical control (CNC) machining system as an example of an EWS. A process, e.g. turning, transforms inputs to outputs, that is a blank into a machined part, and is implemented by a process implementation device (PID). The PID is controlled by a CNC controller that transforms a reference, i.e. NCcode, and operator's commands into PID control signals. A human operator supervises inputs and outputs and manages the system. The VWS agent is a software entity having four basic functional elements: perceptor, evaluator, effector, and inference mechanism (lower part of Fig. 1) [14]. The
perceptor: (1) observes the state of the environment in the network in order to recognize information relevant to the agent, (2) forms inputs from this information, and (3) triggers the evaluation process. The inference mechanism controls the evaluation by reasoning. The reasoning process is based on embedded data and knowledge with the reference to a set of goals. Output is launched into the environment by the effector and thus affects the state of the environment. EWS and VWS are linked via an interface. The interface enables mapping of process data into the VWS data and knowledge base and interaction between VWS and operator. Autonomy, task-competence, alertness, reactiveness, ¯exibility, learning and communicability are the key attributes of a VWS. Thus the work system has acquired the functionality needed for operation in a distributed environment. 3.2. Task driven structuring of a manufacturing system in part manufacturing The work systems are interconnected via corresponding VWS agents into a network and thus constitute the ADMS.
130
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
The agents operate and communicate over the network and coordinate their actions to accomplish complex tasks by exploiting their competence, taking into account their own objectives. The coordination process implies task decomposition as a recursive task structuring process. The dynamic structuring process consists of bidding±negotiation± contracting phases. The process builds up task-oriented manufacturing structures. For example in the part fabrication domain the fabrication of a part is performed in a sequence of tasks that are executed by corresponding work systems. In principle within process planning there are several alternative sequences carried out for each part. The objective is to structure the manufacturing system in such a way that the fabrication is accomplished with the best-cost performance in a given time frame under the actual circumstances. The proposed mechanism of building task-oriented structures is carried out in the following procedure. Bidding phase. The objective of the bidding phase is to ®nd candidate work systems for a task. A mediator, i.e. an agent that controls the process of making coordinated solutions for a given task, starts the coordination process after perceiving a task from the environment. In the ®rst step it forms and releases a request for participation for a task deployment and execution. A request includes three kinds of information describing the task content, time frame, and commercial conditions. It has the following form:
Timing is a critical factor in dynamic structuring. Milestones that characterize the process are shown in Fig. 3. The request de®nes: (1) a task time frame in which the task is supposed to be executed, (2) request expiration time, and (3) bid validity. A bid is to be sent up to the request expiration time. It can agree with the requested task time frame, or it may offer an optional time frame. The bid validity milestone has a particular role. First it de®nes a time space for bilateral negotiations between the mediator and the bidding work system agent for a given task; second it has an impact on the coordination and determination of the manufacturing structure for the execution of the whole sequence of tasks. The ef®ciency of dynamic structuring mostly depends on a time lag between the task assignment milestone and the time of task execution. The shorter the time gap the more the assumption of a quasi-stable environment is realistic. Negotiation phase. The mediator negotiates with the bidding agents within the cluster. Here the mediator seeks for an overall optimum solution of a manufacturing structure for the realization of the whole sequence of tasks. The objective is a cost minimization at a harmonized lead-time. The participating work system agents try to optimize their local goals in terms of economic and machine utilization ef®ciency. The negotiation process iterates cyclically until the mediator states that the solution matches the requirements of the tasks within acceptable tolerances. Contracting phase. The mediator then assigns the tasks to the selected agents, which in turn con®rm acceptance of the
Request: {Head{Request_ID, Consigner_ID, {Adressee_ID}}, Request_speci®cation {Time_data{Time_frame_preiminary{Begin, End}, Request_expiration_time, (Bid_validity)}, (Commercial_data{Price, Award/Penalty}}, Task_speci®cation{Operation_Type, Input/Output_speci®cations{Dimensions, Material, Tolerance_Speci®cation}, Number_of_parts}, (Attachements{Documentation})}. All work system agents that have the functional capabilities and are willing to participate reply with a bid. The bid may be conformable to the request. It may include optional alternatives as well. The bid has the following form:
task. The selected agents then execute the assigned tasks according to the contracted references as shown in Fig. 2(b). During the negotiation phase particular attention is put on the coordination of agendas. The mediator takes care of the
Bid: {Head{Bid_ID, Request_ID, Consignee_ID, Adressee_ID}, Bid_speci®cation{Time_data{Time_frame_bidden{Begin, End}, Bid_validity}, Commercial_data{Price, Penalty}}, Task_speci®cation{Operation_Type, Input/Output_speci®cations{Dimensions, Material, Tolerance_Speci®cation}, Number_of_parts}, (Attachements{References})}. Thus contacts between demand and supply are established on the network resulting in a task-oriented dynamic cluster of cooperating and competing agents (Fig. 2(a)).
agenda for the sequence of tasks for a part. On the other side work system agents manage their own agendas which are composed of single tasks for different parts as shown in
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
131
Fig. 2. Dynamic structures of work systems: (a) dynamic structure; (b) executive structure.
Fig. 4. Basically the assigned tasks are ®xed on both sides. The mediator tries to relax time frame limits to attract more competitive work system agents. The work system agents try to increase their capacity utilization. Several considerations have to be taken into account, e.g. technological similarity of successive tasks, minimization of time gaps between tasks. In the case of a new good opportunity the work system agent may try to give away an already accepted task by launching it as a new request back into the network. On the basis of the proposed mechanism a dynamic manufacturing structure is built up for speci®c manufacturing objectives. Hence a non-permanent manufacturing system for the fabrication of a particular part is established. After the part has been manufactured the temporary form of the manufacturing system is decomposed back into autonomous work systems. The structuring process is initiated when a new task is perceived in the environment or when malfunctions and disturbances in the execution of already assigned tasks occur.
Fig. 3. Timing of a task coordination process.
3.3. ADMS implementation The inference mechanism of ADMS is implemented in Eclipse, a constraint logic programming language which is based on Prolog. Two types of modules were developed. The mediator module generates and optimizes solutions at the task level, while the VWS module performs decision-making processes on the work system level. Different optimization strategies can be applied according to the preset objectives. The state of the system and its environment are dynamic variables expressed as constraints. Thus the structuring process is inferred by the current state. The system's shell consists of conventions of the ADMS ontology expressed in the form of rules and serves for coordination control. Communication among agents is based on contract net protocol adopted from Ref. [15] and implemented in the distributed component object model (DCOM) technology.
Fig. 4. Agenda of cooperating agents.
132
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
Table 1 Requested tasks Task
Sketch
Table 3 Bid example PCS
S (m 2/pc)
L (m/pc)
Time frame Begin
End
K9
50
0,003
0,32
8:30
10:30
K15
40
0,107
4,39
10:00
13:00
K10
28
0,071
3,31
9:30
14:30
.. .
4. Case study The presented approach is demonstrated in an experiment in sheet metal manufacturing on industrial data. The requested set of tasks was composed of various sheet metal components with different shapes, dimensions and batch sizes. The selected material as well as sheet thickness was the same in order to provide an opportunity for the
Table 2 Request example
nesting of different components on the same sheet and thus for an optimization of material and time utilization. Examples of requested tasks (components to be cut) and corresponding basic data are shown in Table 1. Requests were coded as data structures composed of three parts: (1) heading data containing general information, (2) request speci®cation containing commercial and timing data and (3) task speci®cation data. An example of a request is shown in Table 2. The requests were validated by work systems. A bid was replied in the case that a particular work system was able to perform the requested task and was available in the requested time frame. The bid data structure is shown in Table 3. Examples of bids are shown in Table 4. Bids were evaluated by the mediator in terms of: (1) ful®llment of task objectives (in this case it was assumed that the bidding work system was capable of performing the task), (2) minimization of costs and (3) ful®llment of constraints (time frame limits, see Fig. 5). The best bids were selected and corresponding work system assigned as shown in Table 4.
5. Conclusions and discussion In the paper an approach to dynamic structuring of manufacturing systems is presented. It is based on the work system as the building block. The market mechanism Table 4 Bidden tasks Task
K9 K15 K10 .. .
Bidder
WS1 WS2 WS3 WS1 WS3 WS1 WS2 WS3
Price
13.029 13.920 12.314 12.343 11.728 10.018 10.315 9.695
Time frame
Assigned
BEGIN
END
9:30 10:30 12:00 10.45 13:00 11:45 10:30 14:00
10:45 11:30 13:00 11.45 14:00 13:00 12:30 15:00
u u u
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
133
Although tasks and time frames are de®ned, potential resources for task execution are not known in advance. So the point is to build dynamic structures for each order based on a self-organization principle. The resultant structure emerges from the actual state of the system and its environment expressed in constraints. However, there are two issues of particular interest: ®rst, how to balance the autonomy of an individual work system and the potential synergy rising from cooperation, and second, how to control the structuring process if the entire manufacturing system domain is taken into account. Nevertheless the presented approach indicates a viable frame for further investigation in structuring, optimization and control of distributed manufacturing systems.
Acknowledgements
Fig. 5. Constituting of agendas.
governs the dynamic structuring process. The approach is discussed in the part fabrication domain. In ADMS the virtual work system complements an elementary work system. It is an agent which represents the work system in a network. The structuring is task-driven and is based on bilateral, peer to peer communication and coordination between agents with particular reference to time harmonization. The presented structuring process clearly indicates its self-organizing nature driven by selfinterest and competition of autonomous EWSs. This concept combines some good properties of hierarchical and heterarchical systems in terms of robustness and stability against disturbances and dynamic ¯exibility with respect to changes in the environment. The ef®ciency of control of a complex manufacturing system largely depends on the level of its decomposition. In ADMS it is assumed that the complexity is minimized because the basic building block is the EWS. It consists of a process, a process implementation device and a human subject that is characterized by its competence in relation to the process. It is obvious that human competence cannot be decomposed such that its elements would be divided. That is why the complexity increases if the manufacturing system is decomposed to a greater extent, e.g. very basic elements, such as part description, tool elements, etc. In a traditional manufacturing system allocation of manufacturing tasks to resources is a typical scheduling problem. Scheduling is de®ned as the planning of n tasks over m resources to be processed at or during a particular time. The issue of planning and control in ADMS is not limited only to classical scheduling but ®rst of all it focuses on the appropriate structuring of the system.
This work was partially supported by the Slovene Ministry of Education, Science and Sport, under Grant No. J20778.
References [1] Warnecke HJ. Die Fraktale Fabrik, Revolution der Unternehmenskultur. Berlin: Springer, 1993. [2] Ueda K. A concept for bionic manufacturing systems based on DNAtype information. Proceedings of Eighth International Prolomat Conference. Tokyo; 1992. p. 853±64. [3] Yoshikawa H. Intelligent manufacturing systems program (IMS). Technical Cooperation That Transcends Cultural Differences. University of Tokyo; 1992. [4] Peklenik J. FMSÐa complex object of control. Proceedings of Eighth International Prolomat Conference. Tokyo; 1992. p. 1±25. [5] Gu P, Balasubramanian S, Norrie DH. Bidding-based process planning and scheduling in a multi-agent system. Comput Ind Engng 1997;32(2):477±96. [6] Markus A, Vancza TK, Monostori L. A market approach to Holonic manufacturing. Ann CIRP 1996;45(1):433±6. [7] Ueda K, Vaario J, Fujii N. Interactive manufacturing: human aspects for biological manufacturing systems. Ann CIRP 1998;47(1):389±92. [8] Sousa P, Ramos C. A dynamic scheduling Holon for manufacturing orders. Second World Congress on IMS, Budapest H; 1997. p. 542±7. [9] Parunak HV, Baker AD, Clark SJ. The AARIA agent architecture: from manufacturing requirements to agent-based system design. Workshop on Agent-Based Manufacturing. ICAA'98. Minneapolis, MN; 1998. p. 17. [10] Peklenik J. Structural and operational complexity of future manufacturing systems. CIRP J Manufact Syst 1998;27(1):5±11. [11] Westkaemper E. Manufacturing on demand in production networks. Ann CIRP 1997;47(1):329±34. [12] Sluga A, Butala P. Structuring of the information system for distributed manufacturing. CIRP J Manufact Syst 1998;27(2):177±82. [13] Peklenik J. Complexity in manufacturing systems. Manufact Syst 1995;24(1):17±25. [14] Sluga A, Butala P, Bervar G. A multi-agent approach to process planning and fabrication in a distributed manufacturing. Comput Ind Engng 1998;35(3±4):455±8. [15] Davis R, Smith RG. Negotiation as a metaphor for distributed problem solving. Artif Intell 1983;20(1):63±109.
www.elsevier.com/locate/aei
Dynamic structuring of distributed manufacturing systems Peter Butala*, Alojzij Sluga Department of Control and Manufacturing Systems, University of Ljubljana, Askerceva 6, SI-1000 Ljubljana, Slovenia Received 7 May 2001; accepted 1 March 2002
Abstract The paper addresses the problem of dynamic structuring of manufacturing systems. The approach presented in this paper is based on the decomposition of manufacturing objectives and the allocation of tasks to autonomous building blocks, i.e. work systems, in a dynamic environment. The allocation is based on a market mechanism that enables the self-structuring and optimization of a manufacturing system by evaluation and selection among competing work systems. The mechanism presented is based on logic relations and constraints. It enables the building of task-oriented manufacturing structures of work systems acting in series and/or in parallel. The approach is discussed in an example in the part fabrication domain. q 2002 Elsevier Science Ltd. All rights reserved. Keywords: Modeling; Work system; Fabrication; Multi-agent structure; Market mechanism; Virtual cluster
1. Introduction The rising complexity of products, production structures and processing procedures on one side and turbulent market excitations resulting in growing product variety, individualization and shortening time frames on the other are setting new frontiers for the manufacturing business. Despite research efforts and investments in the context of computer integrated manufacturing the existing manufacturing systems are still predominantly based on the obsolete Taylorian philosophy. Therefore they cannot adequately conform to these requirements because of their structural rigidity, deterministic approach to decision making in a stochastic environment, hierarchical allocation of competencies, and insuf®cient communication and exploitation of expertise. In order to face these challenges in manufacturing a shift of the existent manufacturing paradigm from deterministic into a new manufacturing prospect considering natural understanding and concern is needed. Several in¯uencing approaches are emerging. Fractal factory [1], bionic manufacturing systems [2], Holonic manufacturing systems [3] and distributed manufacturing systems [4] are some concepts that are making an appearance. Based on these concepts several advances to modeling a manufacturing system in terms of viable structures for more effective mastering of complex and dynamic behavior of the system * Corresponding author. Tel.: 14774-753; fax: 12518-567. E-mail address: [email protected] (P. Butala).
and its environment are being developed. Various contributions in this direction have been presented and intensively exercised recently [5±9]. In the paper an approach to dynamic structuring of manufacturing systems is presented. It is based on the market mechanism which induces a self-organizing principle and is implemented in the system entitled the adaptive distributed manufacturing system (ADMS). Section 2 brie¯y describes the context of distributed manufacturing systems and related works. Section 3 presents the concept of ADMS and the underlying principles. The case study in Section 4 illustrates the abilities of the system in the part fabrication domain. Particularities of ADMS and distinctions to related works are discussed in Section 5. 2. Issues of distributed manufacturing systems A market is the key segment of the environment which sets primary objectives for manufacturing systems. It demands innovative, customized products and quick responses. The character of the market is dynamic and stochastic. Traditional manufacturing systems cannot cope with market requirements due to: ² Time response to predominantly random excitations of the markets is unsatisfactory, too slow and in many instances too costly. ² Information is incomplete, inaccurate and unreliable.
1474-0346/02/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved. PII: S 1474-034 6(02)00007-1
128
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
Decision-making is based more or less on 'guessing', predominantly using the rule of thumb [10]. ² Organizational structures are predetermined and rigid. Traditional structuring of manufacturing systems is based on labor division and the optimization of performance is based on central planing and control [11]. Let us consider an example of a ¯exible manufacturing system (FMS). It is the most advanced structure of a traditional manufacturing system. It has built-in scenarios to anticipated contingencies. Its scope is ®xed and limited. Its structure is predetermined and cannot be modi®ed by adding or removing elements in a short period of time. Therefore it is a closed system. In order to overcome the earlier mentioned disadvantages of manufacturing systems, new, ¯exible and adaptable organizational structures capable of performing self-organization of the work process have to be established [10]. Several systems exist which are capable of adapting themselves to changes in the environment. Biological systems are an example of such systems that exhibit characteristics such as self-recognition, self-growth, self-recovery and evolution [2]. Another example of a self-organizing system is the social system called the market economy. It is an economic system controlled, regulated, and directed by markets alone; order in the production and distribution of goods is entrusted to this self-regulating mechanismÐthe `invisible hand' as noted by A. Smith two centuries ago. Both types of systems have inspired researchers to adopt their basic principles in the manufacturing world. Ueda [2] introduced biological manufacturing systems by mapping the analogies from the living world. In his latest work Ueda [7] expanded this concept into interactive manufacturing which recognizes the vital role of a human in manufacturing as proposed by Peklenik [10]. Various authors have adopted the micro-economic model of self-organization on the production planning and control scale. Some promising results have already been obtained [5,6,8,9,12]. The common observations can be summarized as follows: (1) a manufacturing system is a complex system, (2) the architecture is the essential issue, (3) the role of a human subject is dominant, (4) the information and communication technologies are the enabling technologies, and (5) the transition from highly data-driven to particular information, knowledge and learning driven organization is needed. Therefore new features of future manufacturing systems have to be developed. The most expected ones are: (1) an open multi-level architecture, (2) advanced communication capabilities, (3) decentralized decision making, (4) selfstructuring ability, and (5) rede®nition of the work systems in terms of autonomy, evolutionary adaptivity, re-con®gurability, co-operativeness, interactivity, task orientation within competence, ability of communication, coordination and co-operation, and learning capability. This research is focused on the problem of how a manu-
facturing system can be structured for a given objective (e.g. the realization of a product) and how the system can adapt its structure in the case of external and/or internal disturbances (e.g. machine break down). The key distinction between ADMS and other approaches [5,6,8,9] lies in the level of system decomposition. In other approaches building blocks are entities that model the factory's functions and/or physical entities (e.g. orders, parts, and resources). However, they do not represent a manufacturing structure and thus the reduction of complexity is limited. In ADMS the complexity is managed more ef®ciently as the building blocks are systemsÐelementary work systems (EWS). 3. Agent based approach to dynamic structuring in ADMS ADMS is based on the concept of distributed manufacturing systems. It is structured as a network of building blocks (work systems) acting as agents. Thus it represents an organic structure of interrelated building blocks which are acting in parallel and/or in series and are driven by cooperation and competition on various levels [4,13]. This concept synthesizes the agent structure as a decision network in order to optimize the overall system's behavior according to the current state of the environment. Structuring is thus an important issue for the achievement of the optimal performance of a manufacturing system. A single optimal structure for different manufacturing tasks cannot be de®ned. For each particular task the optimal structure has to be built up. A feasible approach to the structuring of the system is to structure the task ®rst. The complex task can be structured by the decomposition of the task into less complex tasks. Thus an incompletely speci®ed problem is transformed into a set of more speci®ed ones. Here it is assumed that complex tasks are decomposed into elementary tasks which can be executed by elementary work systems. The structuring process implies the incorporation of work systems for task execution. It is based on the market mechanism. The market mechanism is characterized by the operation of the self-adjusting forces of supply and demand. The motivation in the market mechanism is very basic: self-interest with respect to the economic gain. While self-interest may or may not be an universal part of human nature it is certainly a powerful motivating force. 3.1. Basic building blockÐthe work system In order to realize the concept one has to de®ne the basic building blocks with autonomous behavior and as having the competence and capabilities to perform a particular manufacturing operation. For the structuring of a manufacturing system a generic building block was introduced by Peklenik [13] in terms of the elementary work system. An EWS is capable of
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
129
Fig. 1. Elementary machining work system and its virtual counterpart.
performing single manufacturing tasks, e.g. process planning, machining. It consists of hardware elements necessary to implement a work process, work process identi®cation for process control and optimization, and a human operator as an autonomous subject for making decisions and synthesis. In ADMS a virtual work system (VWS) is introduced in order to delegate the EWS in a distributed environment. The VWS is an agent and it represents the EWS as its counterpart in the information space. Fig. 1 shows the composed structure of the EWS and VWS. Fig. 1 exhibits a generic structure of a computer numerical control (CNC) machining system as an example of an EWS. A process, e.g. turning, transforms inputs to outputs, that is a blank into a machined part, and is implemented by a process implementation device (PID). The PID is controlled by a CNC controller that transforms a reference, i.e. NCcode, and operator's commands into PID control signals. A human operator supervises inputs and outputs and manages the system. The VWS agent is a software entity having four basic functional elements: perceptor, evaluator, effector, and inference mechanism (lower part of Fig. 1) [14]. The
perceptor: (1) observes the state of the environment in the network in order to recognize information relevant to the agent, (2) forms inputs from this information, and (3) triggers the evaluation process. The inference mechanism controls the evaluation by reasoning. The reasoning process is based on embedded data and knowledge with the reference to a set of goals. Output is launched into the environment by the effector and thus affects the state of the environment. EWS and VWS are linked via an interface. The interface enables mapping of process data into the VWS data and knowledge base and interaction between VWS and operator. Autonomy, task-competence, alertness, reactiveness, ¯exibility, learning and communicability are the key attributes of a VWS. Thus the work system has acquired the functionality needed for operation in a distributed environment. 3.2. Task driven structuring of a manufacturing system in part manufacturing The work systems are interconnected via corresponding VWS agents into a network and thus constitute the ADMS.
130
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
The agents operate and communicate over the network and coordinate their actions to accomplish complex tasks by exploiting their competence, taking into account their own objectives. The coordination process implies task decomposition as a recursive task structuring process. The dynamic structuring process consists of bidding±negotiation± contracting phases. The process builds up task-oriented manufacturing structures. For example in the part fabrication domain the fabrication of a part is performed in a sequence of tasks that are executed by corresponding work systems. In principle within process planning there are several alternative sequences carried out for each part. The objective is to structure the manufacturing system in such a way that the fabrication is accomplished with the best-cost performance in a given time frame under the actual circumstances. The proposed mechanism of building task-oriented structures is carried out in the following procedure. Bidding phase. The objective of the bidding phase is to ®nd candidate work systems for a task. A mediator, i.e. an agent that controls the process of making coordinated solutions for a given task, starts the coordination process after perceiving a task from the environment. In the ®rst step it forms and releases a request for participation for a task deployment and execution. A request includes three kinds of information describing the task content, time frame, and commercial conditions. It has the following form:
Timing is a critical factor in dynamic structuring. Milestones that characterize the process are shown in Fig. 3. The request de®nes: (1) a task time frame in which the task is supposed to be executed, (2) request expiration time, and (3) bid validity. A bid is to be sent up to the request expiration time. It can agree with the requested task time frame, or it may offer an optional time frame. The bid validity milestone has a particular role. First it de®nes a time space for bilateral negotiations between the mediator and the bidding work system agent for a given task; second it has an impact on the coordination and determination of the manufacturing structure for the execution of the whole sequence of tasks. The ef®ciency of dynamic structuring mostly depends on a time lag between the task assignment milestone and the time of task execution. The shorter the time gap the more the assumption of a quasi-stable environment is realistic. Negotiation phase. The mediator negotiates with the bidding agents within the cluster. Here the mediator seeks for an overall optimum solution of a manufacturing structure for the realization of the whole sequence of tasks. The objective is a cost minimization at a harmonized lead-time. The participating work system agents try to optimize their local goals in terms of economic and machine utilization ef®ciency. The negotiation process iterates cyclically until the mediator states that the solution matches the requirements of the tasks within acceptable tolerances. Contracting phase. The mediator then assigns the tasks to the selected agents, which in turn con®rm acceptance of the
Request: {Head{Request_ID, Consigner_ID, {Adressee_ID}}, Request_speci®cation {Time_data{Time_frame_preiminary{Begin, End}, Request_expiration_time, (Bid_validity)}, (Commercial_data{Price, Award/Penalty}}, Task_speci®cation{Operation_Type, Input/Output_speci®cations{Dimensions, Material, Tolerance_Speci®cation}, Number_of_parts}, (Attachements{Documentation})}. All work system agents that have the functional capabilities and are willing to participate reply with a bid. The bid may be conformable to the request. It may include optional alternatives as well. The bid has the following form:
task. The selected agents then execute the assigned tasks according to the contracted references as shown in Fig. 2(b). During the negotiation phase particular attention is put on the coordination of agendas. The mediator takes care of the
Bid: {Head{Bid_ID, Request_ID, Consignee_ID, Adressee_ID}, Bid_speci®cation{Time_data{Time_frame_bidden{Begin, End}, Bid_validity}, Commercial_data{Price, Penalty}}, Task_speci®cation{Operation_Type, Input/Output_speci®cations{Dimensions, Material, Tolerance_Speci®cation}, Number_of_parts}, (Attachements{References})}. Thus contacts between demand and supply are established on the network resulting in a task-oriented dynamic cluster of cooperating and competing agents (Fig. 2(a)).
agenda for the sequence of tasks for a part. On the other side work system agents manage their own agendas which are composed of single tasks for different parts as shown in
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
131
Fig. 2. Dynamic structures of work systems: (a) dynamic structure; (b) executive structure.
Fig. 4. Basically the assigned tasks are ®xed on both sides. The mediator tries to relax time frame limits to attract more competitive work system agents. The work system agents try to increase their capacity utilization. Several considerations have to be taken into account, e.g. technological similarity of successive tasks, minimization of time gaps between tasks. In the case of a new good opportunity the work system agent may try to give away an already accepted task by launching it as a new request back into the network. On the basis of the proposed mechanism a dynamic manufacturing structure is built up for speci®c manufacturing objectives. Hence a non-permanent manufacturing system for the fabrication of a particular part is established. After the part has been manufactured the temporary form of the manufacturing system is decomposed back into autonomous work systems. The structuring process is initiated when a new task is perceived in the environment or when malfunctions and disturbances in the execution of already assigned tasks occur.
Fig. 3. Timing of a task coordination process.
3.3. ADMS implementation The inference mechanism of ADMS is implemented in Eclipse, a constraint logic programming language which is based on Prolog. Two types of modules were developed. The mediator module generates and optimizes solutions at the task level, while the VWS module performs decision-making processes on the work system level. Different optimization strategies can be applied according to the preset objectives. The state of the system and its environment are dynamic variables expressed as constraints. Thus the structuring process is inferred by the current state. The system's shell consists of conventions of the ADMS ontology expressed in the form of rules and serves for coordination control. Communication among agents is based on contract net protocol adopted from Ref. [15] and implemented in the distributed component object model (DCOM) technology.
Fig. 4. Agenda of cooperating agents.
132
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
Table 1 Requested tasks Task
Sketch
Table 3 Bid example PCS
S (m 2/pc)
L (m/pc)
Time frame Begin
End
K9
50
0,003
0,32
8:30
10:30
K15
40
0,107
4,39
10:00
13:00
K10
28
0,071
3,31
9:30
14:30
.. .
4. Case study The presented approach is demonstrated in an experiment in sheet metal manufacturing on industrial data. The requested set of tasks was composed of various sheet metal components with different shapes, dimensions and batch sizes. The selected material as well as sheet thickness was the same in order to provide an opportunity for the
Table 2 Request example
nesting of different components on the same sheet and thus for an optimization of material and time utilization. Examples of requested tasks (components to be cut) and corresponding basic data are shown in Table 1. Requests were coded as data structures composed of three parts: (1) heading data containing general information, (2) request speci®cation containing commercial and timing data and (3) task speci®cation data. An example of a request is shown in Table 2. The requests were validated by work systems. A bid was replied in the case that a particular work system was able to perform the requested task and was available in the requested time frame. The bid data structure is shown in Table 3. Examples of bids are shown in Table 4. Bids were evaluated by the mediator in terms of: (1) ful®llment of task objectives (in this case it was assumed that the bidding work system was capable of performing the task), (2) minimization of costs and (3) ful®llment of constraints (time frame limits, see Fig. 5). The best bids were selected and corresponding work system assigned as shown in Table 4.
5. Conclusions and discussion In the paper an approach to dynamic structuring of manufacturing systems is presented. It is based on the work system as the building block. The market mechanism Table 4 Bidden tasks Task
K9 K15 K10 .. .
Bidder
WS1 WS2 WS3 WS1 WS3 WS1 WS2 WS3
Price
13.029 13.920 12.314 12.343 11.728 10.018 10.315 9.695
Time frame
Assigned
BEGIN
END
9:30 10:30 12:00 10.45 13:00 11:45 10:30 14:00
10:45 11:30 13:00 11.45 14:00 13:00 12:30 15:00
u u u
P. Butala, A. Sluga / Advanced Engineering Informatics 16 (2002) 127±133
133
Although tasks and time frames are de®ned, potential resources for task execution are not known in advance. So the point is to build dynamic structures for each order based on a self-organization principle. The resultant structure emerges from the actual state of the system and its environment expressed in constraints. However, there are two issues of particular interest: ®rst, how to balance the autonomy of an individual work system and the potential synergy rising from cooperation, and second, how to control the structuring process if the entire manufacturing system domain is taken into account. Nevertheless the presented approach indicates a viable frame for further investigation in structuring, optimization and control of distributed manufacturing systems.
Acknowledgements
Fig. 5. Constituting of agendas.
governs the dynamic structuring process. The approach is discussed in the part fabrication domain. In ADMS the virtual work system complements an elementary work system. It is an agent which represents the work system in a network. The structuring is task-driven and is based on bilateral, peer to peer communication and coordination between agents with particular reference to time harmonization. The presented structuring process clearly indicates its self-organizing nature driven by selfinterest and competition of autonomous EWSs. This concept combines some good properties of hierarchical and heterarchical systems in terms of robustness and stability against disturbances and dynamic ¯exibility with respect to changes in the environment. The ef®ciency of control of a complex manufacturing system largely depends on the level of its decomposition. In ADMS it is assumed that the complexity is minimized because the basic building block is the EWS. It consists of a process, a process implementation device and a human subject that is characterized by its competence in relation to the process. It is obvious that human competence cannot be decomposed such that its elements would be divided. That is why the complexity increases if the manufacturing system is decomposed to a greater extent, e.g. very basic elements, such as part description, tool elements, etc. In a traditional manufacturing system allocation of manufacturing tasks to resources is a typical scheduling problem. Scheduling is de®ned as the planning of n tasks over m resources to be processed at or during a particular time. The issue of planning and control in ADMS is not limited only to classical scheduling but ®rst of all it focuses on the appropriate structuring of the system.
This work was partially supported by the Slovene Ministry of Education, Science and Sport, under Grant No. J20778.
References [1] Warnecke HJ. Die Fraktale Fabrik, Revolution der Unternehmenskultur. Berlin: Springer, 1993. [2] Ueda K. A concept for bionic manufacturing systems based on DNAtype information. Proceedings of Eighth International Prolomat Conference. Tokyo; 1992. p. 853±64. [3] Yoshikawa H. Intelligent manufacturing systems program (IMS). Technical Cooperation That Transcends Cultural Differences. University of Tokyo; 1992. [4] Peklenik J. FMSÐa complex object of control. Proceedings of Eighth International Prolomat Conference. Tokyo; 1992. p. 1±25. [5] Gu P, Balasubramanian S, Norrie DH. Bidding-based process planning and scheduling in a multi-agent system. Comput Ind Engng 1997;32(2):477±96. [6] Markus A, Vancza TK, Monostori L. A market approach to Holonic manufacturing. Ann CIRP 1996;45(1):433±6. [7] Ueda K, Vaario J, Fujii N. Interactive manufacturing: human aspects for biological manufacturing systems. Ann CIRP 1998;47(1):389±92. [8] Sousa P, Ramos C. A dynamic scheduling Holon for manufacturing orders. Second World Congress on IMS, Budapest H; 1997. p. 542±7. [9] Parunak HV, Baker AD, Clark SJ. The AARIA agent architecture: from manufacturing requirements to agent-based system design. Workshop on Agent-Based Manufacturing. ICAA'98. Minneapolis, MN; 1998. p. 17. [10] Peklenik J. Structural and operational complexity of future manufacturing systems. CIRP J Manufact Syst 1998;27(1):5±11. [11] Westkaemper E. Manufacturing on demand in production networks. Ann CIRP 1997;47(1):329±34. [12] Sluga A, Butala P. Structuring of the information system for distributed manufacturing. CIRP J Manufact Syst 1998;27(2):177±82. [13] Peklenik J. Complexity in manufacturing systems. Manufact Syst 1995;24(1):17±25. [14] Sluga A, Butala P, Bervar G. A multi-agent approach to process planning and fabrication in a distributed manufacturing. Comput Ind Engng 1998;35(3±4):455±8. [15] Davis R, Smith RG. Negotiation as a metaphor for distributed problem solving. Artif Intell 1983;20(1):63±109.