Вы находитесь на странице: 1из 4

Modelling of Biological Manufacturing Systems for Dynamic Reconfiguration

Kanji Ueda' (2), Jari Vaarioz, Kazuhiro Ohkura' 'Department of Mechanical Engineering, Kobe University, Kobe, Japan Department of Information and Computer Science, Nara Women's University, Nara, Japan Received on January 3,1997

Abstract The concept of Biological Manufacturing Systems (BMS) aims at dealing with dynamic changes in external and internal environments in product life cycle from planningto disposal, based on biologically-inspired ideas such as self-growth, self-organization, adaptation and evolution. This paper describes modelling of BMS at a floor level and focuses on system reconfiguration. Computer simulation using the principle of selforganization shows that the proposed EMS model indicates adaptive behavior to the changes in products demands due to external environment and malfunction of manufacturingcells as an internal environment, and it provides the possibility of dynamic reconfiguration of manufacturingsystems. Kevwords : Manufacturing Systems, Dynamic Environment, Genetic Information

1. I n t r o d u c t i o n
Today's manufacturing faces significant trends of cultural diversification, lifestyle individuality, activity globalization and environmental consideration. These trends can be summarized as growing complexity and dynamics in manufacturingenvironments. It is not easy to deal with the complexity and dynamics [16] existing ideas with topby down type control of centralized systems, so that many new ideas have been proposed for next generation manufacturing, such as Fractal Company [15], Holonic Manufacturing [7], Random Manufacturing [4] and Sustainable Industrial Production [ 11. BiologicalManufacturing Systems (BMS) was proposed [9, 1 1 1 as a next generation manufacturing system concept. BMS aims to deal with non-predeterministic changes in manufacturing environments based on biologicallyinspired ideas such as self-growth, self-organization, adaptation and evolution. It also covers whole product life-cycle from planningto disposal. EMS [lo-141 has been developed in connection with new fields of computer science such as Evolutionary Computation and Artificial Life [6],and it has also been involved in a couple of researches including Intelligent Manufacturing Systems Program [5]and CAM-Internationalproject. Beside these, BMS has been discussed in several articles [3, 8,1 1 6. This paper describes modelling of BMS at a floor level based on evolution and self-organization, and it focuses on dynamic reconfigurationof manufacturing systems to adapt to changes in products demands and malfunction of manufacturing equipment, as an external and internal environment, respectively.

showing functions such as self-recognition, self-growth, self-recovery and evolution. These functions of organisms are displayed by expressing two types of biological information, i.e. genetic information evolving through generation (DNA-type) and individually achieved information during one's lifetime (EN-type). Unification of biological information with individuals makes living systems complex but adaptive.

All system elements in BMS such as work materials, machine tools, transporters, robots, etc., are comparable to autonomous organisms, as shown in Figure 1. Products develop from raw materials expressing their own DNA-type information. Manufacturing equipment "raises" the product using mainly BN-type information. The product continues to learn BN-type information as knowledge through its lifetime. As a result the product is able to deal with malfunctions autonomously, to be easily recycled and disposed, and to evolve its design for the next generation products [51.

2. ConceDt of BMS
Biological organisms are capable of adapting themselves to environmental changes and sustain their own life by

M : Machine Tool R :Robot Ti : Testing Instrument Tr : Transporter To: Tool Fig. 1 Concept of BMS at floor level

Annals of the ClRP Vol. 46/1/1997

343

3. Modellina of BMS
In orderto realize BMS, there are at least two issues to be examined. One is to develop the embedding technology of biological information in artifacts, and the other is to understand the general behavior of BMS, that is a theoretical aspect of the concept. Modelling of BMS here is related to the latter. The modelling theory is closely related to methods in Artificial Life such as Genetic Algorithm, Genetic Programming, Evolutionary Strategy, L-system, Cellular Automata, Reinforcement Learning and Immune Networks rat her than symbolic approaches in Artificial Intelligence. In the following two concrete examples are given: one based on evolution and one based on self-organization. On both these examples the emphasis is on the dynamic and adaptive behavior to environmental changes. The raw materials or jobs entering the manufacturingsystem is not known beforehand. Neither the internal condition of the system is known at the time of input. Without these information the production planning in the traditional sense is not possible. The production requests are carried by the input (DNA-type information) and the manufacturing system must adapt in real-time to new requests. In other words, the input is looking for manufacturing equipment capable of processing its request, and the manufacturing equipment is looking for input they are capable of processing. This concept also enables automatic recovery of the system in the case of malfunctions of manufacturing equipment. The manufacturing equipment is called manufacturing cell here by mapping from biological cell [13]. The production request received by the system is a part of external environment, while the state of manufacturing cell is a part of internal environment.
3.1 Evolution-based Model

3.2 Self-oraanization-based Model


In this model the movement of work materials is created by using "attraction" fields [13, 14). By calculating the attraction forces in the virtual space we can create dynamic simulation where manufacturing cells "attracts" jobs being "sensitive" for the attractions. The DNA-type information includes here the product description determining what processes the product requires, i.e., what kind of attraction fields to be sensitive. DNA-type information could be given for the manufacturing cells to describe the above attraction fields and their dynamics. These could be furthermore evolved similarly to the evolutionary model. The BN-type information could include howthe transporter selects a particular machine, and how this selection can be reinforced by experience.
4. Simulation R e s u l t s 4.1 Simulations bv Evolutionarv Model

The simulation model employs two kinds of product, WORK1 and WORK2, six kinds of cells 1 to 6 of two different processing types, A and B, with three levels of processing speeds, high, medium and low, and a goal cell, as shown in Figure 2. The process sequence, i.e., DNAtype information for WORK1 is A -> B -> goal, and that for WORK2 is B -> A -> goal.
~~

manufacturing processing processing speed cell No. type high 1 A medium 2 A low 3 A high 4 B medium 5 B low 6 B --_7 goal Fig. 2 Specifications for manufacturing cells Case 1: Change of External Environment

There are two population in this model: work materials and manufacturing cells. According to DNA-type information, the work material searches around looking for a cell with a certain objective. When it finds the objective cell, it tells request for morphology change, and waits for cell's reply. The cell checks the material arrived within a certain time, and replies to the material's requests. After all the changes in morphology are completed, the material goes to a final cell (goal). In the case of manufacturing cells, theirpositions are the f subject to be evolved. In the case o raw materials, their parameters defining the moving direction toward manufacturing cells will be evolved. The mathematical description of the model has been developed in [12]. Evolution Strategies [2] are employed here for the group of cells and the two kinds of work materials. The fitness functions are as follows:

The production request is given as external environment, but it changes. The simulation consists of serial requests numbered R1 and R2. These requests changes every 100 generations in a sequence R1 ->Ft2->R1->R2.. . The first request R1 consists of WORKl and WORK2, both 6 units. The second request R2 consists of WORKl 18 units and WORK2 2 units. Figure 3 shows an example of simulation result, indicating the rearrangement of manufacturing cells and the path of works. WORKl and WORK2 start from the upper left and lower right comers of the floor field, respectively. Each follows its own genetic information and aims for the goal while searching cells. In the beginning, works show random searching manner. After some generations, all works are able to reach the goal. By the 99th generation an ordered structure is formed as shown in (b), where the path for WORKl emerged as cell 1 -> 5 or 6 -> 7, and that for WORK2 as cell 4 -> 2 or 3 -> 7. At each generation the system will get a little better, measured by the total processingtime. Just after the change of request (at each 100th generation) the system shows chaotic behavior, but already after a few generations the organization has been stabilized rather well. In a long run the organization will fluctuate between two configurations according to the type

f,. = C,,time + C,,request (1) fwi =C,, time+ Cw2grow+Cw3collision+ refused ( 2 ) C,,
where, C,, time, request, grow, collisionand refused are weight constants, processing time, degree of achievement, degree of growth, the number of collisions and the number of being refused by cell, respectively.

344

of request. It can be said that the BMS model enables to adapt to external changes through self-organization.

Case 2: Change of Internal Environment


Figure 4 shows reconfigurable behavior of the BMS model which adapts itself to the case of malfunction and recovery of manufacturingcells. After showing a random behavior in the initial state (a), the system emerges a stationary structure using all cells as can be seen in (b). Malfunction occurs on two cells of 1 and 4 with the highest processing speeds at the 50th generation, and it continues until the 100th generation. The system selforganizes by excluding the disabled cells during the malfunction as in (d). When the cells recover at the 100th generation, the system aims for including the cells because their processing speeds are high as shown in (e). It achieves high performance in terms of the total processing time, by employing all cells so that the path for WORK1 emerges as cell 2 or 3 -> 4 -> 7, and that for WORK2 as cell 5 or 6 -> 1 -> 7, as shown in the state of (f).
4.2 Simulations bv Self-Oraanizational Model

(a) random behavior

(b) self-organization adapting to R l

05
\ 4

(c) reconfiguration adapting to R1->R2

(d) reconfiguration adapting to R2->R1

The following example is taken from the real factory environment. The problem is drilling holes on printed circuit boards (PCBs). In this particular case, there is a high volume of several types of boards to be processed every day. Also there is a great variety of machines with different capabilities. The problem is how to distribute the boards to the drilling machines assuming that the number of transporters and available machines varies daily depending on whether they are working or not. The problem decomposes into job, machine, and transporterdescriptions.Each job is defined by the properties of boards it consists of. Similarly, each machine is defined by its processing capabilities. The matching of job requirements and machine capabilities is illustrated in Figure 5. Exact matches are given priority. By accuracy requirement/ By boadtable size caoabilitv class Board Machine Board Machine A300

Fig. 3 Self-reconfigurablebehavior adapting to changes of production request. (9: generation)

starting point of work 1 starting point of work 2 /

lq;
Cf

(a) random behavior

(9=49) (b) self-organization

:--600

*-

. priority matches

800

800

(c) malfunction of cells 1 and 4


I

(9=99) (d) replacing cells 1 and 4 with 2 and 5

- - - - *secondary maches :

Fig. 5 Matching problem of PCBs and drilling machines. Each drilling machine creates attraction fields according to its capabilities. For example, in this case the attraction fields are for the size of the table and the accuracy capability. On the other hand, each transporter becomes sensitive for a particular attraction field according to the properties of the job it's carrying. The overall shop floor is illustrated in Figure 6.

(e) replacing cells 3 and 6 with recovered cells

(f) reconfiguration including all cells

Fig. 4 Self-reconfigurablebehavior adapting to malfunction of cells (9: generation)

The attraction fields are designed for this particular case. The priority of matching is defined by using a larger attraction field than for secondary matches. The used dynamics is defined so that input attractions are initially on, but are turned off when an input is received. When the processing is almost finished, they are turned back on (in
345

Collects empty transporters 'EMPTY' field/


\

Working

Collects finished jobs

The concept of BMS will be extended from the floor level to the handling problem of recycling and disposal.

Leaves a job to be processed

Picks up a finished job

)READY' field

6 Acknowledqment .
This study has been supported in part by the grant of IMS Program, Grant-in-Aid of the Ministry of Education, Science and Culture of Japan, the grant of Nissan Foundation, and the grant of Research for the Future Program of the Japan Society for the Promotion of Science. 7. References

Dispatching 1 of jobs 'Queue of empty transporters Queue of jobs to be processed


\

Malfunction

'\
\\

,'
,

. c - - - - - - - - - - - - - - - - - - - - - - - - - \* -

calsabilitv

finlri

Fig. 6 Factory floor layout for drilling machines and transporters navigating to "attractive" machines. order to receive next input) as well as an "empty" transporter attraction (in order to get the output transported). The result of the above simulation is a dynamic schedule, that can adapt to external and internal disturbances. For example, we can create malfunctions for the cells or the transporters, and the schedule will adapt in real-time to these changes. The main merit of the method is not to find the optimal schedule, but a method to give a dynamic schedule in very dynamic environment. This example is now to be verified against currently used manual scheduling. Other test cases in real world application are now under work.
5. Conclusion

One of the main issues that we have tosolve toward next generation manufacturing is how we can deal with growing complexity and dynamics in manufacturing environment. In this study, BMS proposed as one of candidate for next generation manufacturing systems has been modeled with focusing on the capability of configuration in dynamic environments. The simulation using the proposed model clearly demonstrates the adaptive behavior to the changes of production demands due to external environment and the malfunction and recovery of manufacturing cells as internal environment in the manner of self-organization without external control. Also, the modelling strongly suggests that BMS is capable of real time scheduling, in particular, the proposed system is not necessary to reach the optimal, but adapt even in the case of very dynamic disturbance of manufacturing environment.

(1) Alting, L., Jorgensen, J., 1993, The Life-Cycle Concept as basis for Sustainable Industrial Production, Annals of the CIRP, Vol. 42/1: 163-167 (2) Back, T., Schwefel, H.P., 1993, An Over View of Evolutionary Algorithms for Parameter Optimization, Evolutionarv Comoutation, MIT press: 1-10. (3) Boer, C.R., Jovane, F., 1996, Towards a New Model of Sustainable Production: ManuFuturing, Annals of the CIRP, VOl. 491 : 41 5-420. (4) Iwata, K., Onosato, M., 1994, Random Manufacturing Systems: a New Concept of ManufacturingSystems for Production to Order, Annals of the CIRP, Vol. 4311: 379-383. (5) Kurihara, T., Bunce, P.. Jordan, J., 1996. Next Generation Manufacturing Systems in the IMS Program, Advances in Production Management Systems, IFIP: 17-22. (6) Langton, C.G., 1989, Artificial Life, Addison-Wesley Pub. Comp. Inc.. (7) Markus, A., Kis Vancza, t., Monosstori, L., 1996, A Market Approach to Holonic Manufacturing, Annals of the CIRP, Vol. 4511: 433-436. (8) Tharumarajah, A., Wells, A.J., Neems, L., 1996, Comparison of the Bionic, Fractal and Holonic Manufacturing System Concepts, Int. J. Computer Integrated Manufacturing , Vol. 913: 217-226. (9) Ueda, K., 1992, A Concept for Bionic Manufacturing Systems Based on DNA-type Information, Proc. of IFIP 8th Inter. PROLAMAT Conf., Tokyo: 853-863. (10) Ueda, K., 1993, A Genetic Approach toward Future Manufacturing Systems, Flexible Manufacturing Svstems: Past-Present -Future, edited by J. Peklenik, CIRP: 21 1-228. (11) Ueda, K., 1994, Bioloqical Manufacturina Svstems, Kogyochosakai Pub. Comp., Tokyo. (12) Ueda, K., Ohkura, K., 1995, A Biological Approach to Complexity in Manufacturing Systems, Proc. of 27th CIRP International Seminar on Manufacturing Systems, Ann Arbor: 69-78. (13) Vaario, J., Ueda, K., 1996, Self-organization in Manufacturing Systems, Proc. of Japan/USA Symposium on Flexible Automation, ASME, Boston, VOI. 2: 1481-1484. (14) Vaario, J., Ueda, K., 1996, Biological Concept of SelfOrganization in Flexible Automation Systems, Advances In ProductionManagement Systems, IFIP: 33-38. (15)Warnecke, H.J., 1993, The Fractal Comoany, Springer -Verlag, Berlin. (16) Wiendahl, H.P., Scholtissek,p., 1994, Management and Control of Complexity in Manufacturing, Annals of the CIRP, Vol. 4312: 533-540.

346

Вам также может понравиться