You are on page 1of 6

Multi-agent Modeling on Scheduling of Multi-variety and Multi-process Job-shop

1 2 3
Luo-ke HU , De-hai XIANG , Yong CHEN
IDepartment of Industrial Engineering, Zhejiang University of Technology, Hangzhou, China
huluoke@yahoo.com.cn

Abstract - A new method for modeling and analyzing characteristics of multi-agent and job-shop scheduling,
scheduling of multi-variety and multi-process job-shop this paper establishes a physical and mathematic
based on multi-agent was proposed to effectively solve the scheduling model of multi-variety and multi-process
dynamic and complex scheduling of multi-variety and job-shop and constructs target functions of the job-shop
multi-process job-shop problem. A physical and mathematic scheduling based on multi-agent, besides,
scheduling model of multi-variety and multi-process
comprehensively considers tendering and bidding
job-shop based on multi-agent was established. The
strategies and scheduling mechanisms to further optimize
analyzing scheduling system obtained the target function
based on six kinds of agents such as planning agents, the scheduling results. At the same time, a system was
searching agents, negotiating agents, deciding agents, designed to realize the analyzing scheduling of
logistics agents and database agents. A system was designed multi-variety and multi-process job-shop with Java
to realize the analyzing scheduling of multi-variety and programming language and this system was proved to be
multi-process job-shop with Java programming language. feasible when it was applied in the job-shop scheduling of
The scheduling of the job-shop in a certain national a certain national automobile liner production enterprise,
automobile liner production enterprise was taken as an results of which could provide powerful decision-making
example. Results proved the validity and reliability of this
references for the scheduling of multi-variety and
multi-agent model and scheduling system.
multi-process job-shop.
Keywords - multi-variety and multi-process, multi-agent
modeling, job-shop, scheduling
II. SCHEDULING MODELS OF MULTI-VARIATY
AND MULTI-PROCESS JOB-SHOP
I. INTRODUCTION
-A. The Physical Model Based on Multi-agent
Currently, the method of using primitive equipment
has become the mainstream of Zhejiang manufacturing The main features of multi-variety and multi-process
pattern, which largely promotes the high-speed includes the following aspects: the dynamic of processing
development of Zhejiang economy. In Zhejiang province, and transshipment batches; the diversity and dynamic of
there are more than 600 massive characteristic economic manufacturing resources; the possibility of great
zones producing more than a hundred million output difference of processing routes between different products
values annually. Numerous enterprises in those regions and it is non-linear. The scheduling of multi-variety and
adopt manufacturing methods of order-driven, multi-process job-shop could be defined as several tasks
comprehensive process, multi-functional equipment, of various products on some machines for processing and
multi-variety, multi-process, small and medium batch size. optimization of relevant target functions through
Compared with traditional discrete manufacture, the considering "how to schedule resources, such as
complexity of this typical multi-variety and multi-process machines and materials according to time". The operating
discrete production is caused by the following factors environment of multi-variety and multi-process
throughout the system: uncertain "relationship of supply workshops is filled with uncertainties and manufacturing
and demand", uncertain environmental factors and too tasks often vary dynamically, leading to the complexity
much random perturbation. And the quality of job-shop and difficulty of its scheduling. Multi-agent system is
scheduling plays an important role in the operating made up of several independent and coordinating agents,
efficiency of manufacturing systems and cost control [11. each of which has different solutions and according to the
The method of multi-agent system (MAS) takes predetermined agreement, communicates and cooperates
advantage of partial connection rules, functions, partial with others. It is quite suitable for complicated scheduling
detail models and good autonomous characteristics of system and it can realize independent and dynamic
agents to solve the token of individual intelligent scheduling. According to the characteristics of
behaviours and behaviours sprung up by the change from multi-agent and job-shop scheduling, we can get the
the small-scale character to the large-scale system quite physical scheduling model of multi-variety and
well in complex systems. Domestic and overseas multi-process job-shop based on multi-agent like the
researches suggest that the sheduling system of job-shop Fig.1, which includes the basic layer of scheduling and
based on multi-agent is a scheduling method, which suits the target layer of scheduling. The basic layer of
modem intelligent manufacturing systems and possesses scheduling includes six kinds of agents: planning agents,
wide prospects of applications[2-101• Combined with searching agents, negotiating agents, deciding agents,

978-1-4244-6484-5/10/$26.00 ©2010 IEEE

547
logistics agents and database agents. The target layer of it can affect itself and the environment, manipulate
scheduling includes the total processing time of environmental partial express and respond to the change
work-pieces, the total flow time of work- pieces, the total of environment, more importantly, each agent can
deferred time and the total number of deferred communicate and interact with others and work with each
[1. lli
work-pieces, etc . other to achieve the common target.
I) The planning agent: It can make corresponding
material selections and scheduling plans to ensure the
The basic la)er
workshop operate smoothly according to stock
information of existing raw materials and semi-finished of scheduling

products got from the searching agent and orders, and


lastly, forms task lists; besides, according to the stock
status analysis of finished products and product sales

[' � r
strategies got from the searching agent, it will formulate
the corresponding plans of stock distribution and allocate
finished products to different warehouses to be stored i ��
'-O---4' al otalnUmti ' " cessi�
based on different rules. At the same time, the planning
agent could send the relevant information of logistics
demand of scheduling plans to the searching agent.
sch�duling
:
The target:

i
la)'er of : � J__�
deferred
ti

[
of deferred
lI orkpieces

OtaI-nOlllng
weighted
. .
TOtalnOiling time o
J
of workpieces after

fl i
i
2) The searching agent: According to the order Total processing
time of workpiece� after :
information of planning agents, it can search out the
corresponding stock information of raw materials and
I
time of workpieces
L- lorkpieces . igb i
semi-finished products through the information database
and send the stock information to the planning agent.
3) The negotiating agent: It can receive the Fig.l. Physical scheduling model of multi-variety and
Multi-process job-shop basing on multi-agent
information of the searching agent and ensure each agent
communicate with others and avoid conflicts between
•B The Mathematical Model Based on Multi-agent
agents through the way that negotiation service centre
coordinates and manages other agents; besides, it is also
responsible for the information transmission between the There are mainly two factors contributing to the
basic layer of scheduling and the target layer of complexity of scheduling of multi-variety and
scheduling while it sends task information which has multi-process job-shop. Firstly, there are several
been negotiated and interacted to the deciding agent. processing routes for each work-piece while we need to
4) The deciding agent: According to material choose one route from them. Secondly, in one certain
requirements planning information and tasks from the processing route, this work-piece need to be processed by
negotiating agent and combined with information several serial working procedures and different machines
database and relevant knowledge and rules from the could be chosen in each working procedure, additionally,
when it is processed on different machines, the processing
scheduling rule library, it can decide scheduling rules and
get an initial scheduling rule solution and then send the time would be different, so a mathematical scheduling
result to the negotiating agent. model of multi-variety and multi-process job-shop based
on multi-agent could be established.
5) The logistics agent: The flow direction of
workshop logistics should be regarded as a key factor of First of all, relevant symbols are given as follows: J
stands for the aggregation of work-pieces in processing
scheduling because it is relatively complex to produce
environment and the number of work-pieces in J is n; i
multi-variety and multi-process products. The logistics
stands for the serial number of work-pieces, i=I,2 ...,n; M
agent provides the deciding agent with factors of job-shop
stands for the aggregation of machines in processing
scheduling about logistics, which need to be considered,
environment and the number of machines in M is m ' k
including large logistics from raw materials to finished
stands for the serial number of machines, k=1,2... ,m; P
products and small logistics in a working procedure itself.
6) The database agent: It is the basis of the initial stands for the aggregation of processing routes in
processing environment and the number of processing
system and preserves the result of operation. The database
agent is also responsible for collecting and sorting routes work-piece i could choose is PI and the serial
enterprise internal operation data and production-related number of processing routes of work-piece i is p, p=l,
operation data, such as the serial number of working
procedures, product information and processing
2, .... P;; JO stands for the aggregation of serial number
characteristics in the manufacturing process. At the same of all working procedures in the processing environment,
time, the database agent is responsible for sorting stock JO {I, 2, ... , q},
= the total number of working
information and collecting transferred information of raw
materials and semi-finished products and finished
procedures is q, jE JO means that the serial number
products in stock. of working procedures is j; Mallow (j) stands for the
Every agent in the model of Fig. I is independent and

548
aggregation of machines which could be used in working
L WiCi
OP(i,p)
of work-pieces required to be scheduled ;
procedure }, Mallow U) e M; means that
stands for the total flowing time of work-pieces after
the aggregation of working procedures work-piece i will
contain when work-piece i choose the processing route p,
weighted; L Wi 1; stands for the total processing time

OP(i,p) c JO;
{I
of work- pieces after weighted.

, w o rkpiece i adopts pro ces sin g r oute p;


xip =
0, o t hers.
(I) III. TENDERING AND BIDDING STRATEGIES AND
SCHEDULING MECHANISMS.
{L when vwrkpiece i adopts � sing route p,
y/ i ,p)= it includes vwrkingprocedure j; This system adopts tendering and bidding strategies
0, others. and determines the scheduling mechanism of
preference-selective styles to promote the gradual
(2) optimization of scheduling results' initial solutions. The

.
Zjk(l,p)=
1, when vwrkpiece i adopts processing routes p,
its vwrking procedure} chooses machine k;
main process of tendering and bidding strategies is as
follows: on the basis of the situation that five kinds of
agents on the basic layer can fully communicate and
interact with each other, the task agent (TA) is
0, others. constructed. At the same time, the planning agent
. 0) establishes the tendering document containing the
pt jk (i,p) stands for the time it takes to complete the
information of processing tasks, such as detailed
processing contents, processing batches, processing
working procedure} of work-piece i on machine k when accuracy, processing time, etc to send it to the resource
work-piece i choose the processing route p; st jk (i,p) agent (RA)-the sub-category of the database agent,
besides, the task agent registers to the management agent
stands for the time machine k start to process the working
(MA)-the sub-category of the strategy agent. After the
procedure} of work-piece i when work-piece i choose the
et jk (i,p)
resource agent receives the tendering document, firstly,
processing route p; stands for the end time the resource agent will measures itself whether it has the
machine k processes the working procedure } of ability to accomplish this task or not and checkup itself
work-piece i when work-piece i choose the processing whether it meets requirements of the tendering document
routep. based on the processing task information determined by
et jk (i,p) = st jk (i,p) +pt jk (i,p) ; (4) planning agents. If it meets requirements, a bidding
document will be made and the bidding coefficient should
eti stands for the time the processing of work-piece i be established and the bidding document is sent to the
management agent; after that, the management agent will

�>ip{I� { L }}
was completed.
judge the bidding coefficient of each resource agent and

et; = etjk(i'P)Zjk(i'P) '


establish corresponding scheduling strategies and then
( 5)
p=! ",OP(I,p ) kEMonow (j) send notification to the resource agent, which wins the
bids; lastly, processing will begin. While if the resource
The deciding targets of job-shop scheduling problems agent does not meet the requirements of tendering, it will
are as follows: select the processing routes of work-pieces quit bidding. If there is no resource agent to bid, it means
and the machines each working procedures need; that the work-shop does not have the ability to
schedule the order of each working procedure on accomplish this task and this task will be cancelled and
corresponding machines to make the deferred cost task orders will be withdrawn.
minimum; improve enterprises' productivity and get the
target function of scheduling of multi-variety and
multi-process job-shop based on multi-agent.

k=L Ui+ LCiWi+ �);Wi �min. (6)


In this function: Wi stands for the average loss rate of

work-piece i when it is deferred; L Ui stands for the

total number of deferred work-pieces; L Ci stands for


the total flowing time of work-pieces required to be
scheduled; L 1; stands for the total processing time

549
[MA' Li� [T:\' [RlIA' to the scheduling decision-making of logistics system of
I Ilt.ennin ion I multi-variety and multi-process job-shop. The scheduling
I
••
rules

irCf
o btaining inforntltion of the job-shop in a certain national automobile liner
P
It- -
,-, -
-
I nlonnatlOn or-+" ,
I of tasks
" �
Obtaining
-" " nI inromu. n production enterprise was taken as an example to get
b scheduling schemes of products in different categories.
schtduling beginn iRe. 'ection or unknow
�nalion of worki g
I infonnalion procedUI'f rules The information of the category of products, the status of
T. 'inrom"�"---i working centre, the amount of work-pieces, etc in this
�ng infonnalion of workshop have been known.
� resourres addl'fSs
Oelf'rmination of
� Oelemunalion
....-' of parts rules �('enlrerules Delfnnination of •A Scheduling Methods and Scheduling Targets.
-'------'--

�B
ment rules
--
--'- --
-'---infonnalion .." u,rrt' Il'qUl, rements �
Task tendering In!onnaiJon -'
I
Iktl'nninarion of This system researches five most frequently used
ble period
�a\ ilable inronnation � methods of job scheduling, including first come first

lJ 1,
X

_ _ a_lion
un_"'_ila_bl'_inronn
h�+-

"ailable
_----,;
.hai bj' inro!;r.;;-

:;::::§' Iec.ion orwork


\1-
�msselec n
.�
t�
J serve (FCFS), longest processing time (LPT), shortest
processing time (SPT), earliest due date (EDD) and
critical ratio (CR). The explanation of each method is as
follows:
I , I"fo,,",'io"of
successful disiribulion
� '::: ••ioilof
II

winning bids
"�Inlonnauon o�
l'fSources usage
1) FCFS-arrange processing according to the
sequence orders arrive;
Fig.2. Scheduling rules of the system 2) LPT- the remaining jobs which take the longest
processing time will be given the priority to be arranged;
The system implements and completes scheduling 3) SPT-the remaining jobs which take the shortest
activities through the way that each agent negotiates with processing time will be given the priority to be arranged;
others and selects rules automatically, besides, each agent 4) EDD-the one of the earliest due date will be
will adopt different negotiation process according to the processed first;
scheduling status and selects the rules based on the 5) CR-the job that has the smallest critical ratio
relevant knowledge and existing knowledge. Fig.2 will be given the highest priority of scheduling. It stands
illustrates scheduling mechanisms of the system. More for the radio of the remaining time to due and the
specifically, management agents or managers send the remaining processing time; it is an eclectic methods from
information about the beginning of scheduling to the task EDD and LPT.
management agent (TMA)-the sub-category of the In the scheduling process of multi-variety and
deciding agent after they determined relevant rules. The multi-process job-shop, the processing capacity of
task management agent negotiates with the resource workshops is the crucial factor to workshop resources and
management agent (RMA), which registers to the MA, production capacities, further more, processing time is the
about the distribution of resources according to the most important parameter, reflecting the workshop
information of starting working procedure tasks got from processing capacity. Therefore, processing time will be
the task agent and the resource requirements of working the only factor taken into account while the factors, such
procedures. In the process of negotiation, the RMA could as transition time and set-up time of machines could be
choose the specific RA according to its own preference neglected.
and the TMA will choose the suitable RMA as the final Seven scheduling targets of system researches with
selection of work centers according to its own preference five scheduling Cmax
methods: Tmax
L Ui ' LCi ' L7; , L WiC; and L w;7;.
and lastly, the result of distribution will be sent to the MA.
Therefore, it is quite concerted that the system adopts
tendering and bidding strategies through the way that the
scheduling mechanism reasonably arrange for its targets, •B Scheduling Results and Analysis
resources, etc towards several agents of different targets
in order to coordinate their own behaviours and achieve Set each basic parameter and build logistic
their own targets in maximum. relationship based on the practical status of the workshop.
Specifically, the value of delivery date should be set to
400 in order to be proportional with the previous inputted
IV. Scheduling examples data. Fig.3 is the parameter setting interface of work
centre and FigA is the parameter setting interface of work
A system was designed to realize the scheduling of piece and Fig.5 is the parameter setting interface of
multi-variety and multi-process job-shop with Java manufacture process and time.
programming language. After this system receives and
identifies specific information inputted, it will call some
rules of production scheduling to improve and adjust
workshop scheduling, which offers a powerful reference

550
�---�--x-" Scbeduful.gsuOllegieosoJ
tv, tC tTt tl,e tl,T,
Workc:ent.er ID ISDJ III III 11 11116 1916 11116 1916
Comment.s Beir sander..
811 III II 11I91 lIll lIlI! lIll

Ir-:-
#
2�
o£ machines '. I' = St.y1e III 112 U Illli 1I11 IlJll 1I11

dat.e (defau1t.) � III III IS 12111 lilt 12111 lilt


l
r;;:--
5et.up Me.t.rix
at.us (defau1t.)
I�I 116 I� II I� 2161 1!!4i 2161
Fig.7. Five scheduling results analysis of type 0102 product
OK Cance1 I He1p

Scl�ulillg!m';lleg:e+,
litt t., LV, E� E" EI,� EljT,
m l199 l199
Job ID ISDAOI
181 lIJO lIJO
CotHlent.s o..,ibltonllol "... lIi_I" .. L••,... "lblio"r .....I.J

� lt1 lIll lIJ)


jobs to ..dd Stylo
I II
� �I lt1 l6ll l6ll
;.1. •••• d .... Proces.inc Ti•• rs---
Due d.t.. � Rout.. IiI III II lill 31 lill II
W.icht rr- !ch t. Rout.. I Fig.

OK ClU\c.l Help
IX • CR
Fig.4. Parameters setting interface of work piece
� -----�--- � " IX • EDD

Workcenter
IX 10 FCFS

D
Ib"�3
IX LPT

I'll. 3 IX 0 SPT

I'll. 3 116
1'·]1' 3 132
OK Cancel I Load Route I Help
Fig.9 Netlike association graph of type 0101 product scheduling results
analysis
Fig.5. Parameters setting interface of manufacture process and time

Through scheduling methods, such as FCFS, LPT, rx D CR

rxlD
SPT, EDD and CR, multi-target netlike association graph
EDD
of scheduling target results analysis and scheduling
results analysis could be obtained. Fig.6 to Fig.8 are IX _ FeF3
scheduling results analysis of type 0101 product, type
0102 product, type 0103 product in five methods, JiC D LPT

respectively. Fig.9 to Fig.II are multi-target netlike rx 0 SPT


association graph of scheduling results analysis of type
0101 product, type 0102 product, type 0I 03 product,
res ectively.
lire � r.: �v, iC iT, E'� �,.r
Fig.10. Netlike association graph of type 0102 product scheduling
!OJ 10J 18 681' 814 Ill' IjO results analysis

)(1 1(4 66?l 1lO 66?l 1lO

III 118 � m � m

m Il 16 61U III 6lU III

IiiI lOl 101 lllO III lllO III

Fig.6. Five scheduling results analysis of type 0101 product

551
Ix I CR
national automobile liner production enterprise was taken
as an example, the results of which proved the validity of
IX _ EDD this system.

Ix D FCFS
REFERENCES
IX D LPT

Ix D
[1] CHEN Yong, WANG Xiao, LIN Fei-long. Research of
SPT
logistics scheduling of multi-variety and multi-technology
shop based on multi-agent [J]. Journal of Wuhan
University: Engineering Science, 2008, 41(Sup.): 104-107
(in Chinese).
[2] Y UAN Hong-bing, LI Xiao-ning, DONG Zhi-lin.
Fig.1 I. Netlike association graph of type 0103 product scheduling Application and research of agent technology in the
results analysis job-shop scheduling problem [1]. Computer Integrated
TABLE I Manufacturing Systems , 2001. 7(5):19-27 (in Chinese).
RESULT S OF JOB SCHEDULING BASED ON MULTI-AGENT OBJECT
[3] LIU Jin-kun, GONG Bao-jun, WANG Shu-qing. Research
on blast furnace distributed intelligent control system [J].
Rules of Type 0101 Type 0102 Type 0103
Journal of Zhejiang University: Engineering Science, 2001,
scheduling product product product
34(2):194-199 (in Chinese).
CR 7717 16449 3801
[4] ZHOU Xing-bin, LI Ping. Agent-based scheduling system
EDD 7229 14964 2950 of production optimization [1]. Computer Engineering,
FCFS 7313 14951 2952 2002, 28(3):196-198 (in Chinese).
[5] LI Zhang-wei, ZHANG Jian-ming, WANG Shu-qing.
LPT 7422 15202 3632
Research on network management systems via multi-agent
SPT 6134 15027 3134 technology [J]. Journal of Zhejiang University:
Engineering Science, 2004, 38(8):935-940 (in Chinese).
According to scheduling results analysis of Fig.6 to [6] BOCCALAT TE A, GOZZI A. A multi-agent system for
Fig.8, the target function result of scheduling of dynamic just-in-time manufacturing production
multi-variety and multi-process job-shop could be scheduling[C]//2004 IEEE International Conference on
Systems, Man and Cybernetics. [So 1.]: IEEE,
obtained. Thus, it can be seen that all these five
2004:5548-5553.
scheduling methods have their own advantages and
[7] AY DIN M E, FOGARTY T C. A simulated annealing
disadvantages in the scheduling of different types. So a algorithm for multi-agent systems: a job-shop scheduling
suitable scheduling method should be chosen based on application [J].Joumal of Intelligent Manufacturing, 2004,
the practical status. For instances, it is relatively 15(6): 805-814.
reasonable that type 0102 products adopt FCFS [8] CRAWFORD E, VELOSO M. Learning to select
scheduling rules and type 0101 products adopt SPT negotiation strategies in multi-Agent meeting scheduling
scheduling rules and type 0103 products adopt EDD [1]. Computer Science, 2005, 38(8):584-595.
scheduling rules. At the same time, the scheduling results [9] WANG Xue-hui, LI Shi-jie, ZHANG Yu-zhi. The
multi-agent technology used in the shop-scheduling
analysis of this scheduling system is proved to be suitable
system [J]. Journal of Hebei University of Technology,
for practical production status, so it could provide
2005, 34(2):105-109 (in Chinese).
logistics system scheduling of multi-variety and [10] LV Ci-xing, ZHU Yun-long, Y IN Chao-wan. Hybrid
multi-process job-shop with powerful decision-making negotiation mechanism for agent-based production
references. scheduling [1]. Computer Integrated Manufacturing
Systems, 2006, 12(12):2074-2081 (in Chinese).
V. CONCLUSION [11] CHEN Yong, LIN Fei-long. The multi-agent modeling of
manufacturing enterprise logistics and application [1].
Journal of Zhejiang University of Technology, 2006, 34(3):
Because of the relative complexity of scheduling
310-313,354.
problems of multi-variety and multi-process job-shop, it
should be the most feasible method to take advantage of
artificial intelligence technology to seek the optimal
solution at present. A physical and mathematic scheduling
model of multi-variety and multi-process job-shop based
on multi-agent was established and the target function of
job-shop scheduling was also obtained combined with the
characteristics of multi-agent technology and job-shop
scheduling. The scheduling results were further optimized
with the introduction of the tendering and bidding
strategies and scheduling mechanisms; besides, a system
was designed to realize the scheduling of multi-variety
and multi-process job-shop with Java programming
language. The scheduling of the job-shop in a certain

552

You might also like