Professional Documents
Culture Documents
Systems, and
Manufacturing
Abstract
Process planning and scheduling are important manufacturing planning activities
which deal with resource utilization and time span of manufacturing operations.
The process plans and the schedules generated in the planning phase shall be
modified in the execution phase due to the disturbances in the manufacturing
systems. This paper deals with a multi-agent architecture of an integrated and
dynamic system for process planning and scheduling for multi jobs. A negotiation
protocol is discussed, in this paper, to generate the process plans and the schedules
of the manufacturing resources and the individual jobs, dynamically and
incrementally, based on the alternative manufacturing processes. The alternative
manufacturing processes are presented by the process plan networks discussed in
the previous paper, and the suitable process plans and schedules are searched and
generated to cope with both the dynamic status and the disturbances of the
manufacturing systems. We initiatively combine the heuristic search algorithms of
the process plan networks with the negotiation protocols, in order to generate
suitable process plans and schedules in the dynamic manufacturing environment. A
simulation software has been developed to carry out case studies, aimed at
verifying the performance of the proposed multi-agent architecture.
Key words: Multi Agent System, Integrated Process Planning and Scheduling, Real
Time Scheduling, Flexible Manufacturing System, Petri Nets.
1. Introduction
Process planning and scheduling are important manufacturing planning activities which
deal with resource utilization and time span of manufacturing operations. In order to cope
with competitiveness and globalization of todays business environment, supply chains
become more complex, and manufacturing processes have become more advance, however
products have to be manufactured in higher varieties and smaller batches. It is essential to
establish effective and efficient process plans and production schedules to cope with the
highly dynamic manufacturing requirements. Some automobile manufacturers are gradually
adopting their production ways to support the diversity of the customer needs and increase
the changing speed for the developing the new products (1).
The process planning and scheduling tasks are very complicated and time consuming, if
it is applied to the dynamically changing FMSs (Flexible Manufacturing Systems). To
facilitate the dynamic and adaptable manufacturing activities in the FMS, a new
methodology with appropriate architecture is needed to handle the process planning tasks,
719
in order to cope with dynamic changes of the products and the manufacturing processes in
the FMSs.
The multi-agent approaches are becoming increasingly important for the FMSs to
increase the productivity and the profitability through the greater shop floor agility and
flexibility (2). The multi-agent systems provide us a systematic way to deal with the dynamic
changes of the products and the manufacturing processes in the FMSs.
This research applies a multi-agent architecture to a real time integrated process
planning and scheduling system in the FMSs, which generate suitable process plans and
schedules based on the status of the FMSs. The objective of the paper is to develop a
multi-agent architecture for the process planning and scheduling of multi jobs that rapidly
adopt the process plans and schedules to cope with the job changes and the unexpected
disturbances in the FMSs. A process plan network and a heuristic search algorithm are
proposed to generate suitable process plans and schedules real-timely.
2. Literature review
In recent years, Multi-Agent Systems (MAS) have been widely applied in
manufacturing applications because of its flexibility, re-configurability, and scalability (3-5).
Gu et al. (3) proposed a multi-agent system where process routes and schedules of a part are
accomplished through the contract net bids. The system addresses some practical issues for
merging the CAPP with shopfloor scheduling such as the parts feature representation and
operation specification. The task allocation and process alternative selection are achieved
through the hierarchical bidding processes between machine agents and shop floor manager,
between upper level machine agents and lower level machine agents, and between machine
agents and tool agents, etc. A cascading auction protocol (6) was proposed as a framework
for integrating process planning and heterarchical shop floor control. The integration of the
real-time online process planning (alleviating the selection of routing alternatives, resource
alternatives, detail process planning, etc.), and shop floor control (resource allocation,
scheduling, transportation, tooling, and fixturing) is accomplished progressively through a
recursive auction process carried out in parallel among part management agent and multiple
resource management agents.
IDCPPS (7) is an integrated, distributed and cooperative process planning system. The
process-planning tasks are separated into three levels, namely, initial planning,
decision-making, and detail planning. The results of these three steps are general process
plans, a ranked list of near-optimal alternative plans and the final detailed linear process
plans, respectively. The integration with scheduling is considered at each stage with process
planning. Wang and Shen (8) presented a new Distributed Process Planning (DPP)
methodology by integrating machining feature-based planning, function block-based
control, and agent-based distributed decision making. It proposes to use two-level
decision-making supervisory planning and operation planning. The supervisory planning
focuses on product data analysis, machine selection, and machining sequence planning, and
the operation planning considers the detailed working steps of the machining operations
inside each process plan and is accomplished by intelligent NC controllers. Lim and Zhang
(9)
have developed an agent-based integrated dynamic process planning and scheduling
system to increase the responsiveness of the manufacturing systems. This system does not
only effectively integrate the dynamic process planning and scheduling, but also optimizes
the machine utilization and provides a platform to assess the reconfiguration of the
manufacturing systems.
Sugimura et al. (10) proposed a basic architecture for integrated process planning and
scheduling from the view points of the distributed decision making.
720
Machining Center
Preparation Stations
Turning Center
Figure 1. Configuration of the target manufacturing system
A systematic approach was proposed to select suitable combination of the process plans and
to generate suitable production schedules for all the jobs by applying the genetic algorithm
and the dispatching rules. Wong et al. (11) proposed an agent-based approach for the dynamic
integration of the process planning and scheduling functions. In consideration of the
alternative processes and alternative machines for the production of the parts, the actual
selection of the schedule and the allocation of the manufacturing resources is achieved
through the negotiation among the part and machine agents which represent the parts and
the manufacturing resources, respectively.
The multi-agent architecture has been widely applied to the process planning and the
scheduling, as discussed in (12). The methods proposed in the literatures deal mainly with the
process planning and scheduling tasks in the static environment in which the jobs
specifications and the manufacturing system status are stable. However, it is now required
to develop an integrated process planning and scheduling systems applicable to the dynamic
environment in which some unforeseen disturbances may occur. The objective of the
research is to propose a multi-agent based integrated system for process planning and
scheduling in the dynamic environment, in order to cope with the jobs specification changes
and the unforeseen disruptions, such as the malfunction of the machine tools. The following
issues are discussed in the paper.
(1) Target flexible manufacturing systems.
(2) Multi-agent system for dynamic process planning.
(3) Flexibility and robustness of the proposed architecture.
(4) Synchronization
(5) Simulation software and experimental results.
721
722
723
724
Size
( Character)
1
5
5
2
2
2
2
2
2
5
5
Production
Engineering
Agent
Job
Agents
Description
Message Status
Valid time window for message
Message unique ID
Message type
Current position of requester
Sender ID
Receiver ID
Machining process ID
Machining feature IDs
Manufacturing time of next operation
Estimation of completion time
Job
Order
Agents
Machine
Tool
Agents
Machining
Process
Agent
Step 1: Initialization
Step 2: Request for available machining processes
Remaining Machining Features
Available Machining Processes
Send Requests
Send Proposals
725
MF1
MF3
MF2
MF1
MF2
N0
RMF = {MF1,MF2,MF3}
AMF= {MF1,MF3}
N2
mp3=(mt1, fi2,ct2) -> MF3
RMF = {MF1,MF2}
AMF= {MF1}
MT = 1397
ECT = 2227
N3
mp4=(mt2, fi2,ct1) -> MF1
RMF = {MF2,MF3}
AMF= {MF2,MF3}
MT = 685
ECT = 2525
N4
mp6=(mt2, fi1,ct2) -> MF3
RMF = {MF1,MF2}
AMF= {MF1}
MT = 1312
ECT = 2440
First Level
N5
mp2=(mt1, fi1,ct2) -> MF2
RMF = {MF3}
AMF= {MF3}
MT = 435
N6
mp3=(mt1, fi2,ct2) -> MF3
RMF = {MF2}
AMF= {MF2}
MT = 897
N9
mp3=(mt1, fi2,ct2) -> MF3
RMF = {}
AMF = {}
MT = 897
N10
mp6=(mt2, fi1,ct2) -> MF3
RMF = {}
AMF = {}
MT = 1531
N7
mp5=(mt2, fi2,ct2) -> MF2
RMF = {MF3}
AMF= {MF3}
MT = 943
End
N8
mp6=(mt2, fi1,ct2) -> MF3
RMF = {MF2}
AMF= {MF2}
MT = 1341
Second Level
Third Level
726
representing the machining sequences of the machining features based on the precedence
constraints. The information of the generated nodes is sent to the job order agents for the
negotiations.
Step 3: Request generation by job order agents
The job order agents create requests for the machining process execution for the
individual nodes of the process plan networks, which are the groups of the machining
features that can be generated by same machine tools. The generated requests are sent to the
request boards of the corresponding machine tool agents. The content of request includes
the machining features and selected machine tool, cutting tool and fixture. As you can see in
Fig. 6, there are four nodes N 1 to N 4 in the first level of the process plan network. For
each of them, the requests shown in the Table 3 are generated according to the message
format in Table 1, and sent to the related machine tools MT1 and MT2.
Step 4: Proposal preparation by machine tool agents
The machine tool agents read all the requests from the request boards every RTIP
(Reading Time Interval Period). The machine tool agents analyze the request messages, and
generate appropriate proposals to all the requests. In this paper, we consider a heuristic
algorithm to estimate minimum completion time for generating appropriate proposal for
each request by the machine tool agents, as shown in the followings.
Minimum completion time
The machine tool agents need to estimate the completion time of the remaining
machining features of the job agents. A procedure is developed and given to the job
agents to estimate the minimal completion time of the remaining machining features,
based on the process plan networks shown in Fig. 6. When a machine tool agent
requires a job agent to estimate the minimum completion time, the job agent starts the
procedures from the start node which is specified by the machine tool agent, and repeat
to generate and to select suitable successive nodes with the minimum machining time.
When all the machining features are included in the process plan networks, the job
agent find both the machining sequences of the machining features and the estimated
minimal completion time. Consider a case where we are at node N i of the process
plan network and we are going to estimate the manufacturing time from the node N i
to the end node. The algorithm for calculating the estimated minimum completion time
from node N i to the end node is summarized in the followings.
Initialization:
Set RMF and AMF. The RMF is the set of the remaining machining features. The
AMF is the set of the available machining features that do not have any preceding
machining features and could be done firstly considering the precedence
constraints among the nodes of the process plan networks.( AMF RMF ).
Put the node N i in ECTS set. The ECTS is the set of the nodes in the path from
node the N i to the end of the process plan network, which has the minimum
manufacturing time.
727
Z1V01000N00001T01C00S01R01P01F01
Z1V01080N00005T02C00S01R01P01F01X738E2070
Z1V01000N00002T01C00S01R01P03F03
Z1V01080N00006T02C00S01R01P03F03X1397E2227
Z1V01000N00003T01C00S01R02P04F01
Z1V01080N00007T02C00S02R01P04F01X685E2525
Z1V01000N00004T01C00S01R02P06F03
Z1V01080N00008T02C00S02R01P06F03X1312E2440
(1) Generate a set of successor nodes SN = { N j j = 1,2,... SN } of the node N i for all
feasible machining processes mpr = (mti , fx f , ctt ), r = 1,..., R , ( R = total number of
available machining processes) by applying the following algorithm.
Cluster all features of the AMF set of the node N i that could be machined with
the machining process mp r ,
Generate a new node N j representing a set of machining features which can be
machined by the machining process mp r and put it in the SN set. The links to
the nodes N j , which are successor nodes, are stored in the node N i for further
processing,
Estimate the manufacturing time for node N j that includes the time of the
machining, the transportation and the re-fixturing processes,
Update the RMF and AMF sets for the node N j ,
(2) Select a successor node N k from the SN set which has the minimum machining time
for the next step of extension, and move it to the ECTS set.
(3) If RMF set of N k is not empty consider node N k as node N i and go to (1).
(4) If RMF set of N k is empty, it means that we are in the end of the process plan network.
The sum of the manufacturing time for the nodes in ECTS set is the estimation of the
minimum completion time from node N i to the end.
Let us consider a case where we are going to calculate the estimation of minimum
completion time for node N1 at the process plan network shown in Fig 6. We start with
node N1 , and there are four successor nodes N 5 , N 6 , N 7 , N 8 from the node N1 as
shown in Fig. 6. We select the node N 5 which has the minimum manufacturing time, and
we put it in the ECTS set. We expand the node N 5 at the next stage of the algorithm and
there are two successor nodes N 9 , N 10 . The node N 9 is selected and which has the
minimum manufacturing time, we put it in the ECTS set. As you can see in Fig. 6, for the
node N 9 the RMF set is empty and the algorithm stops. It is because that there are no
remaining machining features in the node N 9 . The sum of the manufacturing time for the
nodes in ECTS set is the estimation of the completion time from node N1 until end.
Following this, the job agent returns the estimated completion time to the machine tool
agent. As you can see in the Fig. 6, the estimation of completion time for all nodes N1 ,
N 2 , N 3 , N 4 are calculated and these values are returned to the machine tool agent. This
procedure can estimate the completion time of all the remaining machining features,
however it requires the additional communications between the machine tool agents and the
job agents.
The machine tool agents generate proposals for each request as shown in Table 4, based
on the minimal completion time of the remaining machining features and send them back to
the related job order agents.
728
729
730
MF5,MF9,MF11
MF20
MF14 MF17
MF14
MF11,MF25,MF30
MF10
MF18
MF26
MF27
MF24
MF1
MF9
MF18
MF22
MF28
MF16
MF21
MF19
MF13
MF8
MF2
MF29
MF15
MF3,MF8,MF10
MF12
MF15
MF23
MF31
MF2
(a)
MF14
MF1,MF2
MF7,MF11
MF1
MF22
MF17,MF23MF32
(b)
MF14
MF6,MF10
MF5,MF9
MF2MF6MF21
MF13
MF7MF10MF20
MF13
MF4
MF8,MF12
MF8
MF15
MF5
MF9
MF15
MF18
MF16,MF20
MF3
MF12
MF3,MF4
MF17
(c)
MF19
MF6MF19MF22
MF1MF17MF23
MF18
MF11
(d)
8. Conclusion
A multi-agent based integrated process planning and scheduling system was proposed
for the integrated process planning and scheduling systems for the FMSs. The following
remarks are concluded.
731
Table 5. Generated process plans and for jobs for case studies
Job
Job
No
Type
Conditions
Process Plans
Flow Time
[(MT14,FI1,CT135)(MF6,MF14,MF17)] [(MT14,FI1,CT135)(MF10)]
Without
unforeseen
changes
[(MT15,FI2,CT149)(MF15)][(MT15,FI2,CT145)(MF3)]
[(MT15,FI2,CT145)(MF4)]
35924
[(MT14,FI1,CT133)(MF1,MF5,MF7,MF8,MF13,MF16,MF18,MF19)]
[(MT14,FI1,CT135)(MF20)] [(MT14,FI1,CT133)(MF2,MF9,MF11,MF12)]
[(MT15,FI2,CT149)(MF15)] [(MT15,FI2,CT145)(MF3)]
(c)
Malfunction of
[(MT15,FI2,CT145)(MF4)][(MT12,FI1,CT117)(MF13,MF14,MF16,MF18,MF19)]
machine tool
[(MT12,FI1,CT117)(MF20)][(MT14,FI1,CT133)(MF1,MF5,MF6,MF7,MF8,MF1
30676
7)] [(MT14,FI1,CT135)(MF10)][(MT14,FI1,CT133)(MF2,MF9,MF11,MF12)]
[(MT15,FI2,CT149)(MF15)] [(MT15,FI2,CT149)MF21)]
Job specification
changes
[(MT15,FI2,CT145)(MF3)] [(MT15,FI2,CT145)(MF4)]
[(MT12,FI1,CT117)(MF13,MF14,MF16,MF18,MF19)][(MT12,FI1,CT117)(MF20)]
34346
[(MT14,FI1,CT133)(MF1,MF5,MF6,MF7,MF8,MF17)]
[(MT14,FI1,CT135)(MF10)][(MT14,FI1,CT133)(MF2,MF9,MF11,MF12)]
[(MT9,FI1,CT81)(MF9,MF10,MF11,MF12,MF13,MF20,MF23,MF24,MF27,MF29)]
Without
unforeseen
changes
[(MT9,FI1,CT81)(MF16,MF30,MF32)]
[(MT14,FI1,CT135)(MF14,MF15,MF18,MF19,MF22,MF26,MF28,MF31)]
16900
[(MT14,FI1,CT133)(MF1,MF2,MF3,MF4,MF5,MF6,MF7,MF8)]
[(MT17,FI1,CT163)(MF21)] [(MT17,FI2,CT169)(MF17,MF25)]
[(MT9,FI1,CT81)(MF9,MF10,MF11,MF12,MF13,MF20,MF23,MF24,MF27,MF29)]
(b)
Malfunction of
machine tool
[(MT9,FI1,CT81)(MF16,MF30,MF32)] [(MT17,FI1,CT163)(MF21)]
[(MT14,FI1,CT135)(MF14,MF15,MF18,MF19,MF22,MF26,MF28,MF31)]
17792
[(MT14,FI1,CT133)(MF1,MF2,MF3,MF4,MF5,MF6,MF7,MF8)]
[(MT17,FI2,CT169)(MF17,MF25)]
[(MT9,FI1,CT81)(MF9,MF10,MF11,MF12,MF13,MF20,MF23,MF24,MF27,MF29)]
Job specification
changes
[(MT3,FI1,CT27)(MF16,MF30,MF32)] [(MT17,FI1,CT163)(MF21)]
[(MT17,FI2,CT169)(MF17,MF25)]
20780
[(MT14,FI1,CT135)(MF14,MF15,MF18,MF19,MF22,MF26,MF28,MF31)]
[(MT14,FI1,CT133)(MF1,MF2,MF3,MF4,MF5,MF6,MF7,MF8)]
Without
unforeseen
changes
[(MT3,FI1,CT27)(MF1,MF2,MF3,MF4,MF5,MF6,MF7,MF12,MF13,MF15,MF21,MF2
3)] [(MT3,FI1,CT27)(MF8,MF9)]
[(MT14,FI1,CT133)(MF14,MF16,MF17,MF18,MF19,MF20,MF22,MF24)]
8683
[(MT17,FI1,CT165)(MF11)][(MT17,FI1,CT163)(MF10)]
[(MT3,FI1,CT27)(MF1,MF2,MF3,MF4,MF5,MF6,MF7,MF12,MF13,MF15,MF21,MF2
(a)
Malfunction of
machine tool
3)][(MT3,FI1,CT27)(MF8,MF9)][(MT17,FI1,CT165)(MF11)]
[(MT17,FI1,CT163)(MF10)][(MT12,FI1,CT117)(MF14,MF16,MF17,MF18,MF19,
25500
MF20,MF22,MF24)]
[(MT3,FI1,CT27)(MF1,MF2,MF3,MF4,MF5,MF6,MF7,MF12,MF13,MF15,MF21,MF2
Job specification 3)][(MT3,FI1,CT27)(MF8,MF9)][(MT3,FI1,CT23)(MF25)][(MT17,FI1,CT165
changes
)(MF11)][(MT17,FI1,CT163)(MF10)][(MT14,FI1,CT133)(MF14,MF16,MF17,
17922
MF18,MF19,MF20,MF22,MF24)]
Note: [(MT, FI, CT)|(MF,)] indicate [(Machine tool number, Fixture number,
Cutting tool number)|(Machining feature number,..)]
(1) A multi-agent system consisting of five basic agents and a negotiation protocol among
the agents were proposed to carry out the various tasks in the process planning and the
scheduling. The individual agents have the capability for the distributed
decision-making and the communications with the other agents.
(2) A systematic procedure was proposed to generate suitable process plans of the jobs and
suitable schedules of the machine tools. The proposed method is able to solve the
process planning and scheduling problems concurrently and dynamically, with use of
the search algorithms of the process plan networks.
732
733
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
734