Professional Documents
Culture Documents
Dispatching Multiple-Vehicle: Composite Systems
Dispatching Multiple-Vehicle: Composite Systems
121
Tanchoco 1984, Russel and Tanchoco 1984, Egbelu queue at any time, and (7) shop arrival time (select
1987, Yim and Linn 1993). However, most of the the workcenter with the material that has been in the
common rules proposed in previous research do not shop the longest.) They compared the performance of
provide information to break the tie when two or the rules in a batch manufacturing system and found
more jobs receive the same priority. For example, that in a busy shop the vehicle-initiated rules have
First-Come-First-Serve (FCFS) is a common rule for more significant effects on the system performance
AGV to select a job. In a busy manufacturing envi- than the workcenter-initiated rules.
ronment, jobs may arrive at various workcenters at Newton (1985) applied the vehicle-look-for-work
the same time. A tie-breaking rule is then needed to (VLFW) logic in dispatching AGV where the nearest
further prioritize the jobs. In general, the tie-breaking station rule was combined with the first-come-first-
rule can be another simple dispatching rule. But served concept so that demands waiting an inordi-
whether a tie-breaking rule can significantly affect nate amount of time are serviced within reasonable
the performance of the AGV system has not been time. Hodgson et al. (1987) developed modified
fully explored in the published research. VLFW rules for a job shop environment. The selec-
This research expands the previous work by tion of a target station for service in their rules is
considering and evaluating three practical yet simple dependent on existing as well as potential demands
composite rules which combine the primary dis- from the stations. Lee (1994) evaluated several simple
patching rules with tie-breaking rules. Discrete-event vehicle-initiated rules where an AGV can stay in the
simulation models were developed to implement the same workstation to minimize the travel distance, or
requiring service to select a vehicle from a set of tion of the selected load. In the pull- dispatching
vehicles available. Common workcenter-initiated rules, a workcenter with the highest need for load
rules include (1) random vehicle, (2) nearest vehicle, (job) replenishment is selected first. Then, a job is
(3) farthest vehicle, (4) longest idle vehicle, and (5) selected among a set of jobs which can move to the
least utilized vehicle. A vehicle-initiated rules allows selected workcenter. By pairing the load and process
an idle vehicle to select a material
handling task from selection, they evaluated numerous push and pull
one of the workcenters or warehouse. Common rules dispatching rules. The results showed that there was
under this category are (1) random workcenter, (2) no significant difference between the push and pull
shortest travel time or distance, (3) longest travel based dispatching rules.
time or distance, (4) maximum outgoing queue size, Sabuncuoglu and Hommertzheim (1992, 1993)
(5) minimum remaining outgoing queue space, (6) studied the performance of both job scheduling rules
modified first come first served where each and vehicle-initiated AGV dispatching rules in a FMS
workcenter is allowed to have only one request in the model. The Largest Queue Size (LQS) rule, where the
122
AGV was dispatched to the machine with the largest oversimplified as they do not provide information to
queue, produced better mean flow-time and due date break the tie when two or more jobs receive the same
performance than the FCFS rule for all the machine priority, or designed for systems with specific
scheduling rules tested. In Tang et al. (1993), they requirements. The assembly system configuration
considered both vehicle-initiated and workcenter- and the composite rules which combine the concepts
initiated AGV dispatching along with the job sched- of the dispatching rules discussed in current literature
uling rules. Their results show that the various with tie-breaking rules are presented in the next
dispatching rules have different degrees of influence section.
on the performance criteria. No single rule can
dominate other rules in all performance criteria.
An issue that could affect the performance of the 3. SYSTEM CONFIGURATION AND AGV
dispatching rule is the type of AGV track layout and DISPATCHING
control mechanism. To reduce the complexity of The hypothetical assembly system studied in this
traffic control, Bartholdi and Platzman (1989) proposed
paper consists of four major assembly lines or
an idea of decentralization of control over AGVs workcenters and a warehouse. Each of the four
using single vehicle loops. In their control system, workcenters performs different types of assembly
each AGV can be programmed independently of the tasks and is capable of producing specific groups of
others so that vehicles can be added or removed from
products. Raw materials enter the system at the
the system. It offers flexibility in the design stage of warehouse entrance where they are bundled and
AGVS since the total number of vehicles needed in a stored by type. They are then transported to the
system can be changed. Bozer and Srinivasan (1991) corresponding workcenters based on the part type by
proposed a tandem configuration for AGVS, which is AGVs for assembly. Upon completion of the assem-
aimed at simplifying an AGV control system. In the
bly tasks, the products are transported back to the
tandem system the stations or workcenters are warehouse. Since assembly systems usually involve a
partitioned into non-overlapping single-vehicle large number of processing operations, the products
closed loops with additional pick-up/delivery points from each of the workcenters can also be considered
provided as an interface between adjacent loops. This as subassemblies requiring further processing in
approach provides the simplicity for traffic control. other parts of a plant.
Choi et al. (1994) reported in a simulation study that
the traditional &dquo;multiple-vehicle&dquo; layout could 3.1 AGV Track Layout
produce a shorter job flow time and AGV waiting The assembly system with the AGV track layout is
time in the stage area, while the tandem layout might
shown in Figure 1. This layout was first studied in
perform better based on the number of jobs completed Maxwell and Muckstadt (1982). Several subsequent
and vehicle blocking time.
studies related to AGV systems have also used this
Mahadevan and Narendran (1990) analyzed
several issues related to the design and operational layout. It is a unidirectional and multiple-vehicle
control of AGV systems. Their result indicates that system. Each of the four major assembly lines has a
the single-vehicle loop configuration performs better pair of drop-off station (D) and pick-up station (P) for
material handling purpose. The possible routes of
than the multi-vehicle systems. One disadvantage of
the single-vehicle loops is their inflexibility in tolerat- AGVs among the workcenters and the warehouse
can be thought as directed links of a network. The
ing vehicle breakdowns which will present a prob- nodes in the network are labeled from 1 to 14 in the
lem to normal operations. Sharp and Liu (1990)
examined a design issue with respect to adding figure including the incoming dock (node 1) and
shortcuts and off-line spurs to fixed-path, closed-loop outgoing dock (node 10) in the warehouse, drop-off
stations (nodes 2, 4, 6 and 8), pick-up stations (nodes
AGV systems. They showed that the savings in travel
time and waiting time due to congestion was signifi- 3, 5, 7, 9), and junctions (nodes 11 through 14) on the .
123
Figure 1. The AGV Track Layout (in feet).
picked up from the pick up station at the end of the must be larger than the length of any AGV used in
assembly line. The warehouse can be divided into the system, and the length of each junction can be
three areas. The area near the incoming dock (node 1) considered as a regular zone. The right to use a zone
is for the storage of raw materials. The finished or a junction is based on the First-Come-First-Serve
products are stored in the area near the outgoing (FCFS) concept. With the unidirectional track design
dock (node 10). In the center of the warehouse is the in the studied system, there are four junctions as
staging area for the AGVs, where the load and shown in Figure 1.
unload operations are carried out. The idle AGVs can
also be parked in the staging area. 3.2 Composite AGV Dispatching Rules
Since multiple vehicles are allowed in the system,
Vehicle-initiated Dispatching Rules
traffic control is needed. To avoid collisions, the AGV
control system often has a zone control capability On completion of the unloading operation at a
that allows only one AGV to access the junction or a drop-off station, a vehicle-initiated dispatching rule
section of the track at a time. In this study, the track must be used to send the idle AGV to one of the pick-
layout can be viewed as a network consisting of up stations or the warehouse to perform other
many junctions or nodes and links. Each link is material handling activities. Four vehicle-initiated
divided into a sequence of non-overlapping zones so dispatching rules were evaluated in this study. The
that one vehicle is allowed in one zone at a time, first rule, Stay in Same Station (SS), was used as a
preventing collisions between AGVs. The zone length benchmark for comparison purpose. The other three
124
rules are composite rules. The composite rules are the list, it also goes to the nearest pick-up station
proposed for study because they are based on the and wait for the next available assignment.
concepts of shortest distance and maximum outgoing IF look-for-work list is not empty, GO TO the
queue size (high queue), which have shown superior station with the highest queue;
or competitive performance over the other rules in
IF two or more stations have the highest queue,
previous research (Egbelu and Tanchoco, 1984;
Sabuncuoglu and Hommertzheim, 1992 and 1993). GO TO the nearest station;
Also, the logic of the selected rules can be easily IF look-for-work list is empty, GO TO the pick-
programmed in the control system and easily under- up station on the same line.
stood by the shop floor personnel. They are described
below: If we assume that the assembly jobs are released to
the workcenters based on the due date, then the SS
1. Stay in Same Station (SS): When an AGV completes rule can be considered as a demand-driven or pull-
a drop-off assignment at an assembly line, it goes
based rule because the idle vehicle always stays at
to the pick-up station of the same assembly line, the same line to &dquo;pull&dquo; the finished product as soon
expecting to carry some finished goods back to the as it becomes available. The NS/HA and HQ/NS
warehouse. The AGV may need to wait at the pick- rules are push-based rules since the idle vehicle
up station if finished goods are not available, but selects a load of materials to move from an output
the level of work-in-process (WIP) on the shop
queue that has the highest priority. The NS/SS rule is
floor and the AGV travel distance could be reduced. based on a mix of pull and push concept.
125
and looks for the next task. When an AGV approaches Therefore, each experiment was replicated three
a junction, a FCFS control scheme is used to avoid times. The total number of simulation experiments
possible collisions. performed is 4 (rules) x 12 (experiments) x 3 (replica-
Three factors which might affect the performance tions) which is 144.
of the rules were identified, including (1) mean time
between arrivals (TBA), (2) arrival distribution (AD), 5. OUTPUT ANALYSIS
and (3) ratio of AGV travel time to the assembly time The performance measures collected from the
(RT). On the average an assembly job may arrive to simulation are: (1) the number of jobs completed per
the warehouse at every 4 minutes or every 8 minutes. shift or throughput, (2) average flowtime per job, and
This inter-arrival time can be uniformly distributed (3) average inventory level in the system, or work-in-
with a possible 50% variation above and below the process (WIP). The throughput measures the produc-
mean, or exponentially distributed. Finally, the ratio tivity, while the flowtime and WIP show the effort in
of travel time to assembly time can be 0.25, 0.5 or 1 in reducing on-line inventory. Statistical tests on
the simulation. The minimum &dquo;round trip&dquo; travel normality and homoscedasticity were performed to
distance from node 1 to node 10 is 340 feet, connect- validate the assumptions needed on the data sets
ing nodes 1,14,8,9,13, and 10. The travel time is 1.7 representing performance measures of the rules. The
minutes if an AGV is operated at 200 feet/minute. test results show that the data sets satisfy both
When the assembly time is 6.8 minutes, then the RT normality and homoscedasticity at 0.05 level of
ratio is 1.7/6.8 or 0.25. If the AGV speed is reduced to significance, so further statistical analysis can be
100 feet/minute, then the ratio would be 3.4/6.8 or carried out.
0.5. The ratio would increase to 1 if the AGV speed is To measure the performance difference of heuristic
further reduced by half. The problem data and the rules and to identify factors differentiating the
experiment factors and their levels tested are summa- system performance, the percent differences is an
rized in Table 1. appropriate measure. Given two heuristic rules HI
Given a simulation model, the factors were tested and H2, the percent difference in throughput (TP),
at 2 by 2 by 3 levels resulting a 2x2x3 factorial design %TP(H1/H2), flowtime (FT), ‘%FT(H1/H2), and
with 12 experiments. A number of trial runs were work-in-process (W), o,{,W(H1 /H2) are defined as
performed to validate the model and to determine a follows:
proper simulation warm-up period. First, it was
observed that several statistics in the simulation
started to show a smaller variation after about 200
minutes. Then, there was very little variation among
the replications. With this in mind, each simulation
in our experiment was run for 480 minutes after a
warm-up period of 240 minutes, simulating an 8-hour
shift. The method of common random numbers
(Kleijnen, 1987) was used to reduce the variance in
the simulation output. Statistical test using ANOVA
has shown that the replication effect did not signifi-
cantly affect the statistics collected in the simulation.
126
The first set of tests performed were the paired-t The results from statistical tests indicated that the
tests to determine whether the performance measures NS-SS and HQ-NS performed equally well in
by rules are significantly different. Paired percent throughput and WIP, while NS-SS outperformed
differences were used instead of paired differences HQ-NS on flowtime. It is not surprising to see HQ-
since percent differences are more likely to be homo- NS perform well in throughput, because both the HQ
scedastic across the sample space than simple differ- and NS concepts have been shown to yield competitive
ences. Under the null hypothesis, the percent differ- results in the literature. But it is interesting to see that
ence between a pair of heuristics is
equal to zero. If NS-SS, a rule which minimizes the travel distance, is
the null hypothesis is rejected at 0.05 level of signifi- competitive with the other rules. The NS-HA rule
cance, it is concluded that the-difference is signifi- ranks third in throughput, but it has the best
cant. The average values and the t tests results from flowtime and WIP. Overall the NS-SS can be consid-
the simulation are summarized in Table 2 by rules. ered as the best rule when both the throughput and
The rank given in Table 2 is based on the result of flowtime are considered.
paired-t test at 0.05 level of significance, where two One observation on our assembly system is that a
rules would receive the same rank if they are not lower part release rate (i.e. releasing parts from
significantly different. The percent difference and the warehouse to assembly lines) can cause the average
level of significance from the t tests are shown in part flowtime to decrease, because the part waiting
Table 3. time and queue size are reduced. However, the
127
Table 4. ANOVA Results -
Factors Affecting the Performance Differences
The ANOVA results suggest that further statistical travel time isrelatively short as compared to the
analyses are needed to gain a better understanding of assembly time. Under the operation condition of a
the relationship between the magnitude of the low RT, the Stay-in-same-station (SS) concept would
performance differences and the significant factors. not work very well. Sending the AGVs around to
To accomplish this, experiment data were grouped perform services to other workstations seems to be
by factor settings and the mean values from different a more efficient operating strategy. The difference in
data groups were compared by Tukey’s tests. Table 5 throughput between SS and other rules ranges from
shows the mean values of groups of data by factor 56.2% to 57.6% when RT is 0.25. This percent difference
settings which are significantly different as detected reduces to 2.5% to 5.1% when the RT is 1. Notice that
by the Tukey’s tests. In most cases a lower mean time RT does not significantly affect the performance
between arrivals (TBA) or higher arrival rate could differences in composite rules. The NS-SS rule, which
enlarge the performance differences between two results in the best throughput, is the one which has
heuristic rules. For example, when the TBA is 8 the smallest travelling distance among the three
minutes, the percent difference between NS-SS and composite rules. z
SS is 9.3%. The difference increases to 51.7% when
TBA is 4 minutes. An exception is observed when
comparing two composite rules: NS-SS and NS-HA. 6. CONCLUSIONS
We learned from t-test that NS-SS outperformed NS- Four vehicle-initiated AGV dispatching rules were
HA rules. Table 5 shows that the difference in tested in this study. Three of the four rules are based
throughput between these two rules is higher (3.9%) on combinations of simple rules proposed in previ-
when the arrival rate is at the lower level. Mixed ous research, while the SS rule is used as a bench-
results were also found in arrival distribution (AD). mark for comparison purpose. The result revealed
In general the percent difference between two rules that the NS-SS and HQ-NS performed equally well in
increases when the arrival distribution is completely throughput and WIP, while NS-SS outperformed
random (exponential distribution), but this is not HQ-NS on flowtime. The performance difference
always the case. between the SS rule and the composite rules was
Finally, consistent results were obtained from the significantly affected by the job inter-arrival time
analysis of factor RT. A low RT indicates that AGV (TBA), the ratio of AGV travel time to assembly time
128
Table 5. Tukey’s Mean Tests on Performance Difference
*
Difference not significant at 0.05 Level
129
Kleijnen, J.P.C. (1987), Statistical Tools for Simulation JIM LEE is Associate Professor of
Practitioners, Dekker, New York. Engineering Management at the
Lee, J. (1994). "Dispatching AGVs in an Apparel Manufac- University of Southwestern Louisiana.
turing System," Proceedings of 5th Annual Academic He received a BS degree in Industrial
Apparel Research Conference, Lafayette, Louisiana. 10.1-10.8. Engineering from Tunghai University,
Mahadevan, B. and T.T. Narendran (1990), "Design of an and MS and PhD degrees in Industrial
Automated Guided Vehicle-based Material Handling and Management Engineering from the
System for a Flexible Manufacturing System," Interna- University of Iowa. His research areass
tional Journal of Production Research, 28(9), 1611-1622. include simulation, production and
Maxwell, W. L. and J. A. Muckstadt (1982), "Design of quality management, decision support
Automatic Guided Vehicle Systems", IIE Transactions, systems, and computer-integrated manufacturing. He has
14, 2, 114-124. performed and managed research projects in various areas
of computer aided modeling of engineering problems. His
Newton, D. (1985) "Simulation Models Calculate How industrial experiences include being a manager of a
Many Automated Guided Vehicles Are Needed,"
Industrial Engineering, February, 17, 68-73. consulting group in Taiwan, providing services to local
manufacturing companies in building production planning
Pegden, C. D., R.E. Shannon and R.P. Sadowski (1990), and quality control systems, and as a production planner
Introduction to Simulation Using SIMAN, McGraw-Hill. with a leading U. S. electronics manufacturer.
Raju, K. R. and O. V. K. Chetty (1993), "Design and
Evaluation of Automated Guided Vehicle Systems for
Flexible Manufacturing Systems: An Extended Timed
Petri Net Based Approach," International Journal of
Production Research, 31(5), 1069-1096.
Russel, R. S. and J. M. A. Tanchoco (1984), "An Evaluation
of Vehicle Dispatching Rules and Their Effect on Shop
Performance," Material Flow, 1, 271-280.
Sabuncuoglu, I. and D. L. Hommertzheim (1992), "Experi-
mental Investigation of FMS Machine and AGV
Scheduling Rules Against the Mean Flow-time Criterion,"
International Journal of Production Research, 30, 1617-1635.
Sabuncuoglu, I. and D. L. Hommertzheim (1993), "Experi-
mental Investigation of FMS Due-Date Scheduling
Problem: Evaluation of Machine and AGV Scheduling
Rules," International Journal of Flexible Manufacturing
, 5, 301-324.
System
Sharp, G. P. and F. F. Liu (1990), "An Analytical Method
for Configuring Fixed-path, Closed-loop Material
Handling System," International Journal of Production
Research, 28, 4, 757-783.
Tang, L. L., Y. Yih, and C-Y. Liu (1993), "A Study on
Decision Rules of a Scheduling Model in an FMS,"
Computers In Industry, 22, 1-13.
Yim, D. S. and R. J. Linn (1993), "Push and Pull Rules for
Dispatching Automated Guided Vehicles in a Flexible
Manufacturing System," International Journal of Production
Research, 31, 1, 43-57.
130