Interactive GPSS-PC Program Generator for

Automated Material Handling Systems

J. Ashayeri and L. F. Gelders
Katholieke Universiteit Leuven, Celestijnenlaan 300A,
B-3030Leuven-Heverlee, Belgium

Software aids for simulation are very important to practitioners of simulation.

The widespread availability of inexpensive computing power now allows
computer assistance in each stage of" simulation activities such as input data
analysis, modelling, programming, output analysis and so on. The purpose of
this paper is to describe an interactive microcomputer GPSS simulation
program generator for automated material handling systems. The program is
written in Pascal and consists of several modules to capture data, build the
model, and generate the corresponding GPSS simulation program for auto-
mated guided vehicle systems as well as surge systems. The application of" the
program to a real life project is then used to highlight practical advantages of
this approach. The paper also presents current research (e.g. inclusion of an
automated storage and retrieval system) and the degree to which such software
can be extended.

1. I N T R O D U C T I O N

Historically, simulation models have been built using general purpose simulation languages
(GPSS, GASP, SLAM) which require expertise on the part of the modellcr. In recent years,
there has been a trend in the simulation market toward software that, in effect, generate
(produce) simulation models for specific classes of problems. Examples of this type are
simulation generator programs developed for the design of flexible manufacturing systemsl~].
automated warehousesPl, automated guided vehicle systemsl3-41and many others.
A simulation program generator is an interactive software toot that translates the logic of a
model into the code of a simulation language for execution. The model is described through a
set of enquiries.
This facility gives the user the benefit of communicating with a computer in simple language,
on the one hand (especially valuable for those who are not familiar with simulation modelling
and programming practices), and, on the other hand, it provides tile user with the power of a
high-level simulation language in which models will be written. With such special purpose
software aids, a particular class of problems can easily be modelled, but the modeller is
essentially constrained to stay within that class.
In this paper an interactive Pascal program which generates the GPSS-PC models for auto-
mated material handling systems is discussed. The program to be presented in this paper can be
a useful tool for designing an efficient and economical new advanced material handling system
as well as effectively operating or expanding an installed one. The paper is organised as
follows: in Section 2 the basic structure of the considered material handling system is identified
and then the role of simulation in designing and operating such systems is clarified. The
characteristics of the simulation program generator such as the assumptions, modelling logic,
decision rules, required input and output are described in Section 3. To illustrate how the

The International Journal of Advanced Manufacturing Technology, 2 (4), 63-77, 1987 ~) IFS (Publications) Ltd,
0268-3768

63
The International Journal of Advanced Manufacturing Technology

program can be applied to a complex real life problem, a design project is studied in Section 4.
The undergoing and future extensions of the program and conclusions are given in Section 5.

2. S T R U C T U R E OF A U T O M A T E D MATERIAL HANDLING SYSTEMS

At present, an AMHS generally refers to a combination of a set of hardware systems (e.g.

automated storage/retrieval units, automated guided vehicles, conveyors, accumulators) and
a software system (such as management software, operation software and control software)
under computer control. Note here that even in advanced automated handling systems,
operators will still play important roles with regard to activities such as picking, loading,
supervision and maintenance of handling units. Therefore, operators are taken into account as
one of the sub-systems of the AMHS dealt within this study.
The scope of the material handling considered in this study will be basically confined to time
integration of automated guided vehicle systems (AGVS) and a shipping surge system. The
hardware components of an AGVS and its operation software have already been discussed in
literature (see e.g. 13' 5.6. 71). Here, a specific class of operation software has been considered
which will be described in the next section, but time hardware components are the ones
commonly used in practice, i.e. the vehicles, guidepaths, number of accumulators (buffers)
used as loading/unloading stations, battery charging and so on. Thus, the AGVS will not be
detailed and the surge system will be explained further.
A shipping surge system consists of a number of parallel accumulators placed at the edge of a
loading dock to ship large quantities of palletised goods. These accumulators are filled at one
side with finished product pallet loads and emptied at the other side by a truck/train loading
operator. Normally a surge is linked to warehouse and production lines by means of AGVS.
The surge concept as considered in this study is rather new. The purpose of using surge is to
uncouple the inner warehouse operations (storage/retrieval of pallets) from loading operations
on one hand, and, on timeother hand, increase the number of shipments from production lines to
loading docks (direct shipment), i.e. to reduce double material handling (from production to
warehouse, warehouse to loading dock).
A surge system can constitute a good solution to diverse problems of shipping, especially in
cases where the supply rate of pallets from the warehouse or production lines is lower than the
loading potential of the loading operators at the dock. It can also be a valuable help when the
arrival sequence of trucks and the sequence of orders to be loaded are unknown. These random
arrivals yield sudden peaks of activity followed by periods of idleness which have, of course, a
negative influence on the performance of equipment and loading operators. Therefore, three
main functions can be distinguished for a surge:
O To maximise direct shipment
O To smooth activity for fork lift truck (FLTs) and AGVS
O To ensure continuous loading activity
Surge systems can be classified into two categories depending on the type of allocation:
brand dedicated lanes and truck dedicated lanes.
Brand Dedicated Lanes
One lane contains pallets of one brand only. The operator loads a truck by picking pallets from
different lanes to complete an order. When a plant produces numerous types of product, the
high investment cost of a pallet position (about $1920, which includes the price of the conveyor
and the floor space), prohibits the building of as many lanes as there are different brands
produced. Therefore, one has to carefully allocate the brands to the available lanes. One or
several lanes may be non-dedicated for products which cannot occupy a separate lane in the
surge.
Truck Dedicated Lanes
One or several lanes of the surge are exclusively reserved for pallets to be loaded on a single
truck. Therefore, it is possible to find in any lane a complete mix of the brands produced and

64
 Interactive GPSS-PCProgram Generator forAutomated Material Handling Systems

shipped in a plant. Once again, due to the high cost of the surge, it is impossible to have as many
lanes as the number of trucks visiting the dock in a day or even in a shift.
Thus, the surge is a quite simple concept but its application in combination with an auto-
mated guided vehicle system can bring important benefits.
The optimal size of the surge will be defined as the one corresponding to the best compromise
between, on the one hand, the cost of equipment and the cost of floor space and, on the other
hand, the savings realised through the increased percentage of direct shipments and the reduc-
tion of the truck loading time. Unfortunately the exact evaluation of these savings cannot be
done analytically because of the large number of parameters involved and their dynamic
behaviour.
The same is true for an AGV system. Analytically, one can compute the minimum number of
AGVs required but many questions can not be answered and design parameters determined
due to the dynamic behaviour of such systems. These are related to routing layout, direction of
movements, the location of bypasses along the layout, number of idle stations, operating rules
for dispatching AGVs, number of positions on output buffers and so on.
It is obvious by now that the design and operating of such automated handling systems poses
quite a complicated problem and analytical models are generally not suitable for such complex
systems. A more accurate approach is simulation which takes into account the dynamic
behaviour of the system (blocking congestion, etc.). What follows the study will be limited to
the modelling of an AGV system and surges with brand dedicated lanes for the above
mentioned reasons. The simulation model can be used to evaluate the system performance and
verify the influence of different parameters on the decision criteria (number of AGVs, buffer
length, dispatching rules, allocation of brands to lanes, loading r u l e s . . . ) . This enables the
design engineer to assess the behaviour of the system without the need for experimentation
with a real system.

3. M O D E L L I N G PROCESS
3. ! Choice of Programming
The simulation program generator is implemented in Pascal and utilises GPSS blocks. There
are four reasons to use a generating program to write the GPSS simulation program:
O To relieve the design engineer of the need to develop and program the simulation model in
GPSS.
O The ease by which the simulation program can be modified (i.e. to allow the design engineer
to concentrate on alternative designs).
9 To provide an interactive feature to input and output data.
9 To make it easier to include some 'intelligence' in a generating program, instead of using
high level simulation language like GPSS.
Pascal was chosen for programming the simulation program generator because of its
dynamic data structure which allows the storage of data in a functional way. Pascal also
enables one to write the program interactively. This choice has been made by other authors
(see e.g. Is.9, L0,1[, 12]).
The choice of GPSS is basically related to the nature of the system under study. There are
many elements involved in an automated handling system representing the state of system
which is changing in a discrete way. GPSS is a discrete simulation language which can be used
for process modelling.

3.2 Concept of tile Model

To overcome the rigidity of GPSS and to provide the flexibility required, the model is first
divided into two modules (AGVS module and surge module) Then each module is split up in
different parts. Modules are described as follows:

65
The International Journal of Advanced Manufacturing Technology

3.2.1 AGVS Module

This module consists of following blocks:
0 Layout block
0 Demandgenerator block
0 Decision block (dispatching and routing)
These blocks and their relationships are now briefly explained:

0 Layout block
The layout determines the design and the structure of an AGVS. A layout is built up of two
basic elements: nodes and tracks between these nodes. This block contains the physical
location and characteristics of the nodes which interact with AGVs and describes the possible
linking routes and their distances. The Pascal program first builds a database of the physical
layout through a set of enquiries. Then it transforms the physical layout in an equivalent GPSS
program. To create a layout, information is required on tracks and nodes. Eight different nodes
were implemented in the layout block. They are:
Split
- -

- Merge
-

- - Cross
- Pick-up
-

- - Drop-off
- - Pick-up/Drop-off
- - Multiple drop-off
- Control station
-

Having defined the nodes and tracks, one can also define areas in which only one AGV can
travel at a time. These areas are particularly useful to stimulate situations in practice where
congested areas are restricted to one AGV (to avoid collision). Reserved areas can also be used
to handle three-way splits and bi-directional tracks.
The layout block also requires knowledge about the AGVs (i.e. the number of AGVs in a
system, the length of an AGV, the travelling speeds of a full and empty AGV). To determine the
capacity of each track the distances between nodes are divided by the size of an AGV. Note
that these figures can be modified by the user.
This part of the program generator also has the ability to check the layout for possible mis-
takes (loose ends, false connections) and prompts the user to correct them.

0 Demandgenerator block
A demand is a physical unit which must be transported by an AGV. In the simulation package,
it indicates to the AGVS what and how much is to be transported. Two types of demand can be
distinguished in the system. The first type can be expressed by a fixed from/to table and the
second type are demands coming from the surge system. The interarrival times and variation
on interarrivals are defined by two from/to tables.

0 Decision logic block

The brain of an AGV system consists of a dispatching and a routing logic: it controls which load
will be handled and in which sequence, and what route the AGVs will follow. Therefore the
decision logic was extensively worked out in this study. Two types of decision logic can be
distinguished in literature, according to two types of AGV systems:
(i) AGV systems with fixed demands: the best way to operate such a static system is to
implement a fixed routing scheme for the AGVs. This results in a generalised dispatching rule.
This routing scheme can be determined by a global optimisation algorithm. Kusiak1131
recognises this as a multiple Travelling Salesman problem. Maxwell and Muckstadt 1~41
proposed a useful heuristic dispatching rule and Kusiak1131also describes, in a recent article,

66
 Interactive GPSS-PCProgram Generatorfor Automated Material Handling Systems

the state-of-the-art for this type of dispatching rule. For the routing logic all of the preceding
articles assume that the shortest path is followed.
(ii) A G V systems with dynamic demands: here the dispatching decisions will be taken when-
ever required, and it depends on the current state of the system. Newton [41 proposed three
different dynamic dispatching rules; Vehicle Looks For Work (VLFW), F~rst In First Out
(FIFO) and Hybrid Strategy. In the latter, a load becomes a critical demand after waiting for a
certain time. Newtonl41 also suggests that routing should not only be a function of distance
(shortest path) but also a function of occupation of tracks. Ashayeri et al. TMalso propose a kind
of hybrid strategy, Egbelu and Tanchoco ljSI list some possible real-time heuristic dispatching
rules. The reader is referred to the literature for further details.
In reality, dispatching rules are rather simple because communication between AGVS and
the cental control computer is a delicate and very expensive part of the AGVS. For the same
reason in most applications, the routing logic is rigid and is known in advance (the shortest path
with/or without some deviations).
According to the foregoing discussion, the shortest path algorithm in the generator program
was used to determine the shortest routes in the layout. However, the user has the choice of
changing any route. To enable the user to try out several dispatching rules, dynamic and fixed
destination assignments were considered. Dispatching decisions can then be taken by looking
at a number of objectives at the same time. They are related to utilisation of AGVs, the number
of direct shipments and the status of pick-up buffers. This part distinguishes the package from
any other similar packages written to simulate an AGV system.
To implement such a dispatching logic, some characteristic continuous and dynamic
function must be used for every pick-up station to represent the priority of derriand at a station.
A priority function is defined as follows:

ei = Ci X~i + C2 X2, + C3 X3i

= Priority for pick-up station i.

C1, C2, C3, = Weighting coefficient for objectives 1,2 and 3.

Xli, X2i, X3,= Measures of objectives 1,2 and 3.

Ci = 1 Xu, Xzi, Xaie[O, 11

Objective 1: avoid stopping production activities. These activities will be stopped whenever
a buffer (pick-up station) is full. Thus X1 is increasing as occupation of buffer
increases. This relation can be defined as a linear or quadratic function.
Intuitively the relationship between Xi and buffer occupation is likely to be
quadratic.
Objective 2: avoid double material handling. This objective eventually increases the direct
shipments. X2 is 1 if the destination of a load is surge, otherwise X2 is zero.
Objective 3: maximise the performance of the AGVS. To do so, empty travel times should be
minimised. This is done by giving higher priority to stations closer to the AGV
looking for a load (i.e. for the closest X3i = 1 and for the farthest station A~i = 0).

The user can specify a critical priority level for each station. Empty AGVs looking for a
demand scan the stations which have a priority larger than the critical level and will be dis-
patched to the station with the highest priority value. If one uses a dynamic destination assign-
ment, the destination of AGVs can be changed on split nodes. If no load is waiting for trans-
portation, the empty AGVs will be sent to the closest idle station or continue their way to the
next split node depending on the choice of the user.
67
The International Journal of Advanced Manufacturing Technology

3.2.2. Surge Module

The surge module is built up by collecting the following information:
0 Generalphysicaldefinition
0 Truck arrivals pattern
0 Demand pattern
0 Production pattern
0 Loading rules

@ Generalphysicaldefinition
To build a surge simulation model, first the physical characteristics of the system have to be
defined. These are the capacity of the surge (number of lanes and number of positions per lane),
the number of loading operators, the loading time, truck capacity, allocations of brands to the
lanes, percentage of order picking pallets per truck (mix of brands not available in the surge),
and duration of operations and delays.

0 Truck arrivalspattern
To generate the arrival times of trucks, it is necessary to define the arrival distribution of trucks
over a day or a shift or in a shorter interval (e.g. hours).

9 Demandpattern
A demand for the pallets to be loaded on the truck is generated after arrival of a truck. The
loading content of a truck is divided into different orders: each order consists of a set of many
different brands. The probability distribution (observed in reality or estimated) is required to
generate the number and size of orders per truck, the number and size of the lots in each order
and the brands of each lot.

Q) Production pattern
The production pattern in time has to be defined to decide whether the required pallets at the
surge can be brought from production lanes or from the warehouse.
To evaluate the performance of the surge under different loading rules, a set of loading rules
is provided in the program from which the user can select one. These rules deal with:
- - the choice of assigning a number of operators to one or to more non/dedicated lanes
- - the choice of the next lot to load
(i) random loading
(ii) loading the lot (brand) with largest quantity
(iii) loading according to a predefined priority
(iv) loading according to the size of lot
(v) loading sequentially a pallet of each lot
I reservation choice
depending on the foregoing options the user has the option of reservation, that is to reserve a
number of pallets in the surge for a truck after its arrival.
- - waiting
this option forces the operator to wait for pallets that are not yet available to complete a lot.
The GPSS-PC generator program containing the logic presented above was implemented in
Pascal on IBM-PC XT. The linked version of the generator program is about 12,000 lines long
and occupies 350k bytes on a diskette. However, for running the program only 64k is required
as the program uses the overlay procedure of TURBOPASCAL.

4. CASE STUDY

In order to verify and validate the program, an existing consumer product manufacturing
system was simulated. The objectives of the study were: to test the accuracy of the simulation
modcl, to analyse the sensitivity of the system performance to changes in the parameters and to
68
 Interactive GPSS-PCProgram Generator for Automated Material Handling Systems

find out how performance can be improved by applying certain loading or dispatching rules~
The existing layout of the material handling system for finished goods is shown in Figure 1
(29 nodes represent the whole layout). Nodes 1,2, 4 and 5 are unit load formers (ULFs) at the
end of the production lines. At these points pallets are taken from pick-up buffers by AGVs to
warehouse 1 or warehouse 2 when there is no demand at the surge. Pallets which are not
produced at production lines and demands for produced brands when exceeding a certain level
(four pallets) are retrieved from the warehouses (pick-up node 16 and pick-up/drop-off node
19).
Node 22 represents a surge of 16 lanes with seven places per lane. Two idle stations exist in
the system (track 23 to 24 and track 26 to 27). Free AGVs stay on either of these tracks and scan
the pick-up points for a new demand. Two reserved areas in which only one AGV can travel at a
time are shown in the layout by dotted lines.
Before simulating the whole system, a static calculation (see [141was carried out to de'termine
the minimum required number of AGVs for the given throughput (see Table 1). The computa-
tion showed that at least seven AGVs were required to transport all demands. To verify the
dynamic behaviour of the system and check the validity of the AGV's module, three runs were
carried out (case 1,2 and 3 ill Table 2). These runs are different in only the number of AGVs in
the system. The real dispatching rule (FIFO) without any priority was considered for such
cases.
The simulation results showed that seven AGVs cannot handle all the jobs and some of the
pick-up buffers were always full. Eight AGVs were capable of transporting all demands but the
average number of free AGVs in the idle stations was rather low in comparison with the data
collected in reality. By increasing the number of AGVs to nine (equal to the existing number),
the system performancc (occupation rate of pick-ups, average number of free AGVs in the
ideal station) given by the simulation became almost identical to the information observed in
practice.
Having checked the validity of the AGV simulation module, several runs were performed to
verify the effects on system performance by changes in the weighting coefficients of the
priority function for dispatching the type of destination assignment, number of AGVs, and
specd of AGVs (cases 3 to 22 in Table 2). These runs are grouped in different sets based on the
number of AGVs in the system (nine, eight and seven AGVs).

Table 1. Demand table

From To 11 19 20

1 10/15 10/10 20/9

2 9/20 9/20 18/10

4 5/36 5/36 10/l 8

5 5/36 5/36 10/l 8

16 46/4

19 46/4

Demand per hour/Variation on interarrival time in seconds

69
The International Journal of Advanced Manufacturing Technology

I j

P4

=.

a~

i =ol_. ~ is e

I
y i~ E
__J

U~

J~

/ ---op
o=
E
r
I

?0
 Interactive GPSS-PCProgram Generatorfor Automated Material Handling Systems

r~ ~.=
m
i

a
- ~+!
.~.

@
e.+

.~.
~4

[-

' ~ ~ ~ ~ ~ ~

~ 2 22 222 ISt 222 Sll " 22~

~ Z

?1
 The International Journal of Advanced Manufacturing Technology

In the simulations with nine AGVs (cases 4 and 5), there was on average more than one A G V
at the idle stations. This showed that the hybrid dispatching logic out-performs the real dis-
patching rule. Notice that the fixed destination rule was implemented during these runs. In the
second set of runs, eight A G V s were considered in the system (cases 6 to 21 ). In all these cases,
the AGVs handled the required throughput with less than one A G V at the idle station. In runs 6
to 8, the fixed destination rule and in runs 9 to 11 the dynamic destination rules were con-
sidered. Application of the dynamic destination rules resulted in a better utilisation of AGVs
(see the average number of AGVs at idle stations for these runs in column 10, Table 2). This
illustrates the advantage of a more intelligent dispatching logic.
Cases 12 to 17 are similar to the cases 6 to 11 except for the speed of AGVs. Here, different
speeds were considered for a loaded and an empty AGV. As a result of this change, the
performance of AGVs was increased in runs 12 to 17 (see again Column 10, Table 2). Notice
that the impact of changing the fixed destination rule to the dynamic destination rule was more
important than the impact of increasing the speed of empty AGVs. This can be seen by looking
at the maximum average buffer utilisation and the average waiting time of loads in the buffers
(Columns 7 and 9, Table 2).
Another important aspect which was checked in runs 6 to 21 was the choice of the weighting
coefficients of the hybrid priority dispatching function, The choice 0.7/0/0.3 (for objective 1 to
3 respectively) usually gave the worst results since the buffer occupation has a larger weighting
than the A G V performance (i.e. AGVs were despatched to stations without enough attention
to the travelling distance). The choice 0.3/0/0.7 resulted in a better performance for AGVs but
increased the average buffer occupation. For this layout, the choice 0.5/0/0.5 proved to be
good and yielded a better overall performance.
Decreasing the number of AGVs to seven (case 22) resulted in a critical situation with very
high buffer and AGVs utilisation. Nevertheless, the required throughput was handled during
the simulation period. Alternative layouts could lower the utilisation but it was not possible to
change the layout in practice. The general conclusion is that with more intelligent rules the
system can operate with eight AGVs without any problem (blocking, congestions and so on).
In order to verify the validity of the surge module, a typical working day was simulated.
During this 7, 3 and 2 lanes were dedicated to brands 9, 2 and 3 respectively. Four lanes were
non-dedicated for a number of different brands brought from the warehouse. A truck visiting
the plant had a capacity of 30 pallets from which four were order picking pallets. The number of
operators for loading was six.
Other information required to generate simulation models are loading time, preparation
time in the warehouse, distribution function for the number of orders per truck, lot size
function, brand function, production pattern (e.g. see Figures 2 to 5 and Table 3), and arrival
distribution of trucks, the latter being considered to be different for each hour. The production
pattern and brand function were also different for each two hour period.
This configuration was linked with the AGVS module using nine AGVs and 0.5/0.2/0.3 as
weighting coefficients for the priority despatching function. Notice 0.2 gives a low priority
(weight) to direct shipments which was evident in the real system. The complete configuration
was simulated for 16 hours (two shifts of eight hours). The simulation results (see Column 1,
Table 4) were practically the same as those observed in practice. This showed that the surge
program is accurate and reliable.
To evaluate the reaction of the system, a number of important parameters were modified and
the simulation was carried out for each of them (see cases 2-11, Table 4). In case 2 a higher
weighting was considered for the direct shipments (using 0.3/0.4/0.3 as weighting coefficients).
As a result of this change the average percentage of direct shipments increased. By decreasing
the block effect (decreasing number of lanes blocked by one A G V ) in case 3, only the average
waiting time for a pallet (time to receive a pallet at the surge from the moment requested) was
decreased.
72
 Interactive GPSS-PC Program Generator for Automated Material Handling Systems

0.9

0.8

0.7

0.6

0.5

0.4

0,3

0.2

0.1

0
1 2 3 4 5

Figure 2. Cumulative distribution function for number of orders/truck.

1 \\

O.e

0.8

0.7

0.6

0.5

0.4

0,5

0.2

0.1

0
6 7 8 lO 15 30

Figure 3, Cumulative distribution function for order size.

73
 The InternationalJournal of Advanced Manufacturing Technology

1 m ,.--
m

0.9
F- \ \ \ \\ \

0.8 .-, \ ~ \,, \

\\ \ \ \

0.7 \ \ \
\ \ \
0,6 \ 9 "-. \ \ \ \ \ \ \ \ \., \
\\ \\'q ,,, .,," \ \ \\ ",,, \ ~
0.5 \ '-, ',.,, \ \ "-.,, \ \ "\ \ \ \\ \ 9
\ ~\,,, \ -., .q \\ \ "-\
0.4
\ \ \ \ \ \ '\ \\ \
\ \ \ \ \ \ \ \\
0,3
\ \ \ \ \ \ \ \ \
0+2 \ \ \ \ \ \ \ \ \ \\ \

\ \ \ \ \ \ \
0+1 \ \ \ \ \ \ \ \ -., \ \

\ \ \ \ \ \ \ \ \:\ \ \. "\ \ \
0 I I 1 I I I I I I I I 1 I I ! I
2 ,._'3 4- ,5 6 7 8 9 10 11 12 1,3 14 1,5 16 17 18 19

Figure 4. Cumulative distribution function for lot size.

0.9

0.8

0.7

0.6

0.5

0.4

0,3

0.2

0.1

0
1 2 ..3 4 5 6 7 8 9

Figure 5. Cumulative distribution function for brand number.

74
 Interactive GPSS-PCProgram Generator for A utomated Material Handling Systems

Table 3. Production pattern for the first period

ULF No. Max demand Production Produced

(A G V loads) rate~hour brand

1 4 40 1

2 4 36 9

3 4 20 2

In cases 4 and 5 the number of brands having a dedicated lane increased from 3 to 4 and 7
respectively. For example, in case 4 the fifth lane was allocated to brand 1 which was pre-
viously brought to non-dedicated lanes from the warehouse. As a consequence, brand I might
be brought directly from the production unit. The results of these runs showed that a different
allocation of lanes to the brands can increase the overall performance of the system. In
comparison with the first run, in case 4, the average direct shipment percentage and the
average number of AGVs at idle stations were increased by 34% and 40% respectively.
Case 6 simulates the system with 20 lanes while case 7 considers 12 lanes in the system.
Increasing the number of lanes had a positive effect on the truck loading and waiting times and
the average percentage of direct shipment. However, it requires additional investment. On the
other hand, decreasing the number of lanes increases the truck waiting time.
In cases 8 and 9, the number of positions per lane was increased to 10 and 13 respectively.
These changes did not improve the overall performance of the system. Eight operators instead
of six were allocated for case 8. This only lowered the average operator utilisation. In the last
case, a different loading rule was introduced (see Table 4) but the results did not show much
improvement in comparison with the first run.
By looking at the results of case 1 to 11 it can be concluded that a careful allocation of the
lanes was the most important variable. In general, it was shown that the package is a powerful
tool for sensitivity analysis as well as for solving the design problem.

5. C O N C L U S I O N S

Simulation is an important tool to assist in the design of automated material handling systems
and to improve the operation of existing systems. Microcomputers have encouraged the wide
use of simulation rather than using analytical alternatives, within more complicated models
which encompass greater volumes of detail. The relative expense of simulation is now largely a
function of the modelling process rather than computer-time. Thus, software tools are
required for reducing the time to define a simulation modelling process.
This paper presents a simulation program generator for automated handling systems con-
sisting of an AGVS and surge system. Application of this program which was implemented in
Pascal and utilises GPSS-PC, greatly shortens the design process, helps in producing an
accurate and workable design, and enables control software to be tested. For instance, the
total simulation time for the linked version was about three hours on IBM-PC (modification
time of input data, generation time of approximately 800 GPSS statements, reading time of the
GPSS program, execution time, and saving time of the results). Adding the preparation time
for the input data one would need halfa day to simulate such a complicated case. However, the
75
The International Journal of Advanced Manufacturing Technology

Table 4. Results of the linked simulation runs

~ Cmfi~ztim ~ ~ 1 P ~ i 5 5 7 6 9 I0 II

klber of L~eS 16 16 16 16 16 20 12 16 16 16 J6
limber of Pallets I Lime 7 I 7 7 7 ? 7 lO I~ ~ 7

L~ 1 9/0.135' 910.~81 910.419 9leA7 9/0A~2 9/0.368 $'I0.Y+-] 910.17 $'10.~ 910.~1 9/e.~2
2 $'/0.579 910A73 9101M~ 910.4W 9]0,578 9/0.6 9/0.4~ 9/91677 9[0.713 9/0.53~ 910.~
] 9/0.619 9/9.537 91~.~7 $'/0.614 9/9.672 9J0.675 9/0.528 $'/6.721 9/018~9 9/0+619 9[6.~84
4 9/0.714 $'/0.5% 9[0.671 $'10,~2 $'/D.T.L9 910.791 9/0.6L] 9/0.874 910.9~9 910.65~ 916.&98
s $'/0+721 9 / 0 . ~ +19.722 [[6.487 9 ~ 0 . ~ 9/0.~4 2[e 298 $'/0.972 910.$'~ 9/01115 9JO.]~
~le . 7~ 9/01~ 919.767 2/0+~ 5/0.552 $'[0.~ 2/9.318 $'16.921 910.~',6 $'10.257 $'to.Tas
7 $'/9+81~ 9/6.75.5 $'10.94s 2/0.534 @6+$57 9/I go V$'.~4 9/0.917 9/i.00 9/0+/7 9/0.70z
Srzeds ie 1 ~ / a 71e+11 2/0.394 210+~33 2/0.724 1 / 0 . ~ 2/o.lm 319 as4 2/0.17~ 2/0.515 2/g+~$' 210A49
O~c~ti~e of Lme $' 7/6+515 2[9.546' 2/6.177 ~10.66~ 210.@5 7/6.~ z/01171 7/0.7s~ 7/0.791 2/oA] 2/0.s
19 2/O.~a~ 2/0A~D3 2/0.64~ ~10.858 2/0.599 2/9.781 X/O.l?2 2/9.817 2/9.94 2/0.639 7/0.65'4
11 3/0J~i5 ]/01535 ~ / 0 . ~ ~/0.879 270.72 1/0.~% X/0 319 3/0.d~ ~/0.758 3/9.616 ]/0.658
12 5f0.8-~ 3/0.778 316.~6 ~(10.d~ ]/0.5~k~ ~/I).~l~l X191~9 110.913 310+~4 ]/O,~+~l 3/o.a~s
13 X/o.468 x/o_326 ]/o.q4 x/o.~9 +/o_869 5/0.~9 x/e.+9 x/e.+7~ x / o . ~ z/o.+m
14 x/o.7,~2 x/o.2~ x/o.+12 X/o.214 +/o.e~ 5/e_717 x/o.~75 i/o.2~1 x/o.+~4 x/o.~5
i5 Z/0 . ~ l i d . ~ x [ O1] x/o . ] % 4/6.761 6/0.~2 I/0.2~ IleAl XIO.~7 1/e.~5
t6 x/o.2~ X/6.206 Ue.2ea 110.159 I/BA~0 6/0+717 1[0397 Z10.059 x/O,3~ X/0.281
17 lto+t$'l

]9 710.7~5
20 9/0,918

No. of ~,~at~rs 6 6 6 6 6 6 6 6 6 8 6
Loid/~g Dale II~ ~ I~ ]~ ~ HI~ t~ MiOID I~ ~ PLk~V~
J~ 1 ~ a ~ s Ukilizitioi 0.892 8.903 0.893 0.897 O_g~ 0.177 0.8~ o.Bgl 0.695 6.776 0.897

tile. of ha:ks e,ived 81 84 84 82 8~ 81 82 83 84 81 I~

~1 O~ lr [ ~ l~'ded 79 73 83 76 ~O 91 73 81 79 79 73

nero m i t i N tizo (k~r) lag 2.2~ 1.~8 1.94 1.~ O.~8 2.26 117~ 2.91 D+&? 7.B2
$td. NO~. 1.15 1.85 1.12 1.71 1.21 0.6 116 1.29 1.41 0+~ ).54

lieu im~di~ tipa(Im~) 1.17 1.29 1.14 1.77 1.17 I 1.3 1.17 1.21 1.25 1.24
Std. I ~ . 0.27 0.34 0.26 9.47 0.76 0.11 0.~ 0.29 O.~ 6.32 9.2~

~i~t~ Tota| I~. ~f PalleLs

~. of direc~ skip.
sr~d ! of d i r e r ~lip.
k+e. miti~ t i e (mi~.) 14.~ 17.5 |5.15
Total Ik d pall~t~ 478 ~18 ~ 418 394 ~4~ ~64 ~ ~8 ~ 152
NO. ~f diF~t ~bip+ Z60 2~ ~ 272 270 294 ~ 283 275 290
Item I .f d i r e r S~ip. 60 66 ~ 11 70 79 77 ~ 61 ~ 64
- #re+ Niti W 1i~ (ziL) 17 17.8 16.~ fiX3 17.23 17.47 20.41 1s 16.11 16+98 16.3~
Tetal No. of Pallets 218 218 252 2~0 ~ 24~ 229 2~0 2t8 2~ 2'2'2
No. of direct skle.
BrZL~d I of direct ship.
i~. miti@ t i ~ (sin.) 9.7 16-2 9.43 11.~6 ~.O 16.~ 919 II.16 10.~ 10.16 9.23
Total Mo_ o~ Pallets 114 1B8
No. of dire~t ~Jp.
Ir~l % of direct ~ip.
AVe. paitiW t i ~ (*is.] 9.6 9.85
~ t a [ No. ~f Pallets
~ . of direct ~i).
Irmd I of direct skip.
9.77 9.$'
Total ~ . of ~llels 25~ 272
]~, of direr ~ip.
~ra~d Z 0~" direr s~ip.
A~. ~ziti~ tim Wi.+) 9.3 lfl.2
Iota| ~e. of pallets 126
kiD. of direct d~ita.
|r~e~ 7 I ef ~rect ~llipl
*~. ~ t i ~ lime (,!L} lO.~
1ot~] ~ . o[ pallets M
~ . of die~t s~ip.
k~ 9 %~f dirett ~Jp.
A~. ~iti~ tipa (sin.) 10.2~
Total I~. of pallets ?16 $76 921 772 ~6 700 875 926 ~ ~4 M8
MO. of dire~t skip. 471 461 616 484 +74 492 4~ t78 4~ 426 @P
sr~ma 9 Z of dirL~t ~kip, 51 5?. $l 62 65 70 $2 51 53 46 53
, kve. I~itin~ t i m {me.) ]1.$ 16.4 11.3 12.~ 11.18 11.57 12.85 ]2.15 1i.85 i1.85 11.~
I~-d~icited Total He. of Pallets 770 634 811 ~ P26 698 7~ 762 812 704
|~ls k~e+ pa[tie9 t i m (sin.} It.g ~.45 15.17 12.85 IB.~ 16.~ 12.1B ll.l[ 19.7[ i~58

Ave. ~if~k s~ipl~t ~ r r ~ t ~ 32 34 3! 43 38 3~ ~4 ~I $2 E~ ~I

Ave. liB. D[ ~V's in idle station 1,34 1.4 1.~ 1.89 1.45 1.31 2.g2 1.39 1.47 1.25 1 67

L o a d J n g Rules :
NRND = NO p r i o r i t y for non-dedicated b r a n d ~ , Random S e q u e n c e o f J o t s , No r e s e r v a t i o n ,
Wait for p a l t e t ~ o f d e d i c a t e d brand§
PLA(4ND : Priority roy g o n - d e d i c a t e d bFands, L a r g e s t a v a i ] a b e quantity loaded fizst:,
No re_~ervation, Wait For p a l l e t 5 of d e d i c a t e d bPand.

76
 Interactive GPSS-PCProgram Generator for Automated Material Handling Systems

authors' experience show that for similar cases with the same complexity, a straightforward
GPSS programming would require four man weeks.
The possible extension of this program are: (a) development of an automated storage
retrieval system (AS/RS) simulation module, (b) consideration of surges with truck dedicated
lanes, (c) introduction of new decision rules for AGVS and surge system, (d) development of
software to support input, output data analysis and, (e) graphical animation. Currently efforts
are being made to extend the package to areas (a), (b) and (c). In the AS/RS simulation module
different storage and retrieval policies have been considered. This module can simulate
systems with cranes dedicated to single aisle or multiple aisle architectures. []

REFERENCES

1. J. E. Lenz, MAST, 'A simulation as advanced as the FMS it studies', Proceeding of the 1st
International Conference on Simulation Manufacturing, 5-7 March 1985, Stratford-
Upon-Avon, UK, pp. 311-320.
2. J. Ashayeri, I,. F. Gelders and P. M. Van Looy, ' A simulation package for automated
warehouses', Material Flow, 1983, I (3), pp. 189-198.
3. J. Ashayeri, L. F. Gelders and P. M. Van Looy, 'Micro-computer simulation in design of
automated guided vehicle systems', Material Flow, 1985, 2 (2), pp. 37-48.
4. D. Newton, 'Simulation model calculates how many AGVs are needed', Journal of
Industrial Engineering, February 1985, pp. 68-78.
5. G. Lofgren. 'Automatic guided vehicles perform as production line system', Journal of
Indgrtrial Engineering, November 1981, pp. 16-20.
6. J. C. Quinlan, "The great AGVs race', Material Handling Engineering, June 1980, pp. 56-
64.
7. R. E. Smith, 'Robotic vehicles wilt perform tasks ranging from product retrieval to sub-
assembly work in factory of future', Journal oflndustrial Engineering, September 1983,
pp. 60-2.
8. D. W. Balmer and R. J. Raul, 'CASM - t h e right environment for simulation', Journalof
the Operational Research Society, 1986, 37 (5), pp. 443-452.
9. D. P. Christy and H. J. Watson, 'The application of simulation: survey of industry
practice, Interfaces, 1983, 13 (5), pp. 47-52.
10. J. G. Crookes, D, W. Balmcr and Chew Sew Tee, ' A three-phase simulation system
written in Pascal', Journal of the Operation Research Society, 1986, 37 (6), pp. 603-618.
11. L. P. Jennergren, 'Simulation in microcomputers revisited', Journal of the Operational
Research Society, 1983, 34 ( 11 ), pp. 1053-1056.
12. R. O'Keefe and R. Davies, 'A microcomputer system for simulation modelling', Euro-
pean Journal of Operational Research, 1986, 2, pp. 23-29.
13. A. Kusiak, 'Material handling in flexible manufacturing system', Material Flow, 1985,
2 (2), pp. 79-95.
14. W. Maxwell and J. Muckstadt, "Design of automatic guided vehicle systems, lIE Trans-
actions, 1982, 14 (2), pp. 114-124.
15. P. Egbelu and J. Tanchoco, 'Characterisation of automatic guided vehicle dispatching
rules', International Journal of Production Research, 1981,19 (5), pp. 359-374.

77