Académique Documents
Professionnel Documents
Culture Documents
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Interfaces.
http://www.jstor.org
INTERFACES
Vol.
11, No.
October
5,
Sciences
of Industrial
Engineering
Detroit,
Michigan
Andrea
Control
Data
Chelst
and Operations
Zundell
Corporation,
Research,
48202
State University,
Wayne
Tilles
20850
and
J. S. Pipis
Detroit
Edison,
2000
Second
Avenue,
Detroit,
Michigan
48226
Abstract.
Plant. Coal
Power
to explore
the impact on the system of unloader break
with adding a second unloader
associated
system. The
the relationship
between
the number
of trains, coal
satisfy
the
plant's
requirements.
factors are believed to have contributed to the rail shortfall. The major
contributor is believed to be the designed single-car unloader. Trains were frequently
attributed to unloader break
queued at the unloader system, which management
downs. These delays were costly for a number of reasons:
Several
trainload.
If the unloader
paper was
refereed,
as requested
by
12
INTERFACES
October
1981
significantly
is considering
To alleviate these problems, Detroit Edison
the addition of a
a simulation model of the
second multimillion
dollar unloader system. Although
entire rail coal movement
it was
system had already been built for other purposes,
decided to build an analytic model which focused on the unloader in order to isolate
its effect.
The specific parameters the model was to explore were: the impact of a second
unloader system on the coal throughput; L, the average number of trains in queue;
and W, the average time spent in the unloader system. In addition, the impact of
other changes were studied, including increasing the number of trains, reducing the
frequency of unloader breakdowns,
reducing the repair time, and changing the cycle
time between mine and power plant (puchasing coal from different mine fields). A
schematic
system
is displayed
in Figure
1.
Literature
Review
The nature of the unloader system suggested a queueing model subject to service
1958] viewed the
interruptions. The earliest work in this area [White and Christie,
service interruption as a high priority class of customer which preempts the service of
the primary customer, which in this case is a loaded train. Our problem involved a
finite source of customers (i.e., a limited number of trains), and the primary statistic
of concern is the average arrival rate of trains to the unloader system, which
is
to the coal throughput. This statistic is not addressed
in the priority
equivalent
literature. Because of the limited number of trains involved, we could
queueing
analyze this problem using only basic queueing theory at the level of Hillier and
text on Operations Research
Lieberman's
[1979].
INTERFACES
October 1981
13
system's
operation:
time
Unloading
time
Repair
Cycle time
trainload of coal
Failure
time
breakdowns
loader is operating
Though the exponential assumption was
breakdowns showed that the exponential
for the repair time (Figure 2) and failure
an exponential
since there
distribution,
when
the un
continuously.
motivated
1977 data on 15
by tractability,
distribution was a reasonable approximation
time. Cycle time we knew could not follow
was an obvious minimum
time for an un
a trainload of coal.
\
>i?
CO
as
LU
C3
\
\*^
EXP(-T/2.8)
N^l o
2.8
1977DATA
15
REPAIR
TIMEINDAYS
14
INTERFACESOctober 1981
Additional
assumptions
follow:
repair time).
for example, we are in state (0,2), two trains are in queue and
man, 1979]. When,
new trains arrive at a rate of (K? 2)X.2. Trains are unloaded at a rate of /?j for the
single unloader system (Figure 3); transitions occur from the top row to the second
row whenever
the unloader breaks down, which occurs at a rate of X2. Within
the
second
row,
are made
transitions
to a higher
state
with
the
arrival
of
a new
train
October 1981
INTERFACES
at a
15
Our discussion of the transition rates within row 1 and between rows 1 and 2
two minor
to Figure 4 of the two-unloader
system with
similarly
applies
service is
modifications.
there are two or more trains being unloaded,
Whenever
occur
a
at a
at
rate
downward
and
breakdowns
of
(i.e.,
transitions)
completed
2pl9
rate of 2X2 Row 2 ismodified
to allow for completion of unloading a train at a rate of
P! because coal can still be unloaded when only one unloader is broken. In addition,
we need to add state (1,0), since it is now possible to have one unloader broken and
no trains waiting
to be unloaded, which was not possible
in the previous model.
Lastly, we must now add another row of states to allow for two broken unloaders.
Breakdown
transitions from row 2 to 3 occur at a rate of \2 >while repair transitions
from row 3 to 2 occur at a rate of 2p2- Transitions within row 3 result from new train
can be represented by systems of difference
arrivals. In steady state these models
equations which equate the rate into and out of each state (available upon request
from the authors).
For the one-unloader model
involving K trains, the model results in 2^+1
simultaneous equations with 2K+ 1 unknowns. Any one of these equations is redun
dant and must be replaced by an equation that sets the sum of all of the state
probabilities equal to one. The two-unloader model involves 3K+ 2 equations and an
equal number of unknowns. These equations were rewritten so that the right-hand
side contained all zeroes except for the last equation, whose right-hand side was one.
A standard program to invert a matrix and which is available on any computer could
used to solve these equations. Because of the special structure of the
=
b) is contained in only
right-hand side, the entire solution of these equations (AX
one column of the inverse of the A matrix. We therefore wrote our own simple
program to solve this problem, which allowed us to perform basic sensitivity analysis
with little added cost.
have been
16
INTERFACES
October
1981
L= 1
=
j I Pu.
\
l
i=0
is idle is P0,o>
is broken is
K
Pu
I
The fraction of time the unloader
is busy
is
I=
of a queue of trains is
1 P0,o
The probability
The key
Poj
1
statistic
is the average
coal
(Po,i + Put)
which
throughput,
the average
equals
arrival
rate:
*_1
r
i
=
+
+
\t.
*1
Ln2- 1 i*"") <Po.n Pun) J KPOt0
This
formula
[Little,
1961] to determine
=
(9.5 hours/train) x (400 trains)/15
= 253 hours/breakdown
= 10.6
days/breakdown.
breakdowns
1977 data showed that the coal unloader system was out of service
Similarly,
total of 42.5 days. The mean repair time, 1//a2> was
= 42.5
1/jjl2
days/15 repairs
= 2.8
days/repair.
for a
The time for a train to complete a trip from the unloader to the mines and back
was 108 hours. When
this is coupled with a 9.5-hour average unloading
time, it
means each train can bring in a maximum of 74.6 trainloads per year. In Figure 5 we
have plotted
the number
October 1981
INTERFACES
of trainloads
throughput
17
the number
of
simultaneously.
COAL THROUGHPUT:
ONE-
VS TWO-UNLOADER
SYSTEMS.
600r
MAY
TWO UNLOADERS-BOTH
OPERATESIMULTANEOUSLY
TWO UNLOADERS-ONE
OPERATING
AT A TIME
ONE UNLOADER
TRAINS INSYSTEM
We can also look horizontally at the curves to determine how many additional
trains are required for the single-unloader
system to maintain a specified level of
an
To
maintain
annual
400
trainload
throughput, the single unloader
throughput.
one additional train
an
or
a
additional
of
two-thirds
train
requires
approximately
18
INTERFACES
October
1981
operating two-thirds of the time (these two things are not equivalent) when compared
to the two-unloader
system. As the throughput required increases to 475 or 520
the
trainloads,
single unloader requires one and one and a half additional trains,
the difference between the two systems grows as the demand
respectively. Clearly,
return from each train sets inmore quickly for the single
increases and diminishing
unloader
The
system.
eighth train only adds about 50 trainloads to the total
throughput for the single unloader, while it adds 60 trainloads to the total throughput
for the dual system. Remember,
each train is potentially capable of carrying 74.6
trainloads per year. These differences become especially significant if, for example,
after a coal strike, abnormally high amounts of coal are needed in a relatively short
time. The single unloader may need as many as five or six additional trains to bring in
the
same
amount
of
coal.
25 r
20
3
O
X
ONEUNLOADER
15
TWO UNL0ADERS-0NLY
ONEOPERATING
ATA TIME
?< 10
oc
L?
>
<
5h
TWO UNL0ADERS-B0TH
MAY
OPERATESIMULTANEOUSLY
6
7
5
NUMBEROF TRAINS IN SYSTEM
October 1981
INTERFACES
19
These
same
two
sets
of
performance
measures
can
be
better
perceived
by
com
bining them into a single figure. In Figure 7 we have graphed tons per day, number of
trains, and average waiting time. To unload 21.5 x 105 tons/day, the dual system
requires six trains and the average delay will be 0.8 hours. The single unloader
requires seven trains to match the throughput, and the average delay will be 17.6
hours.
ONE UNLOADER
(STRAINS)
24
HRSWAITING
inmodel
20
INTERFACES
October
1981
(Figure 8). The single unloader will feed through 470 trainloads with six trains, with
an 86.4 hour cycle time; two unloaders will feed through 445 trainloads with six
trains if the cycle time remains at 108 hours. In contrast, the impact on waiting time
of a 20% change is not as significant. The dual system will always have significantly
shorter waiting
periods
(Figure 9).
TIME: ONE-
VS TWO-UNLOADER
SYSTEMS.
600 r
TWO UNLOADERS
990
ONE UNLOADER
900
490
<
UJ
>
w 400
z
<
390 h
300
290
ONE UNLOADER
TWO UNLOADERS
200
_L
3
_L
6
J_
7
The number of trainloads was found not to be very sensitive to 20% changes in
the mean time between unloader failures or to the failure duration. As would be
expected, the single unloader was more affected by the changes. Delay in unloading
for the single-unloader
system was the parameter most affected by these changes. A
20% increase
INTERFACES
(decrease)
October
1981
in cycle
time reduced
(increased)
21
/^(+20%)
ONE UNLOADER
(-20%)
(+20%)
TWO UNLOADERS
checked against simulation runs and the results were close enough to use the
analytic model for planning purposes. In Table 1we present an illustrative compari
son for a six-train configuration.
22
October 1981
INTERFACES
One-Unloader
s.d.
Analytic Model
400 trainloads
18
= 469
System x
=
4
s.d.
Two-Unloader
450
The major advantages of the analytic models were the computer costs and run
time. A single run of the simulation model on an IBM 370/158 used an average of 2
and 50 seconds CPU time and cost $23.94.
minutes
(This included a $2.70 flat
one
for
To
obtain reliable estimates
handling rate.)
just
configuration
(e.g., a fixed
number of trains), at least 10 runs were typically made with different seed values and
at the above cost. In contrast, a single run of the analytic model could provide
for a range of train configurations.
The savings in computer costs and run time is more than just a financial factor.
It enables "what if" questions to be answered in a shorter period of time. This was
raised a question that could be answered with one run
exemplified when management
statistics
within
IMPLEMENTATION
The results of the study were presented to upper management
and are being used
to analyze whether or not to purchase the second unloader system. Perhaps of greater
to raise
is that the models allowed, and continue to enable, management
significance
and explore new questions:
What
What
time?
tional questions:
trains should be run to bring in X trainloads of coal?
trains can be run without the average delay exceeding
time?
How many
How many
waiting
a specific
In summary,
the analytic model,
which was originally
only to
developed
the addition of an unloader, provides ongoing information about changing
in order tomeet system constraints (such as coal throughput
the system configuration
and wait time). This enables management
and the people who run the system to tune
to
seasonal
the system
consider
and exceptional needs (such as recovery from a coal
use
miners'
Little's definition
the
strike). To
[Little, 1970] of implementation,
models are being used regularly to "update the intuition of decision makers."
evaluate
INTERFACES
October
1981
23
AC KNOWLEDGEMENT
I would
comments
whose
significantly
improved
F. S. and Lieberman,
G.
J.,
1979, Operations
3rd Edition,
Research,
Holden-Day,
Inc.,
San
Francisco.
J. D. C,
1961, "A Proof for the Queueing
3, pp. 383-387.
and Managers:
Little J. D. C,
1970, "Models
Science
Vol.
16, No. 8, pp. B466-B484.
Little,
Morris,
W.
White,
H. C.
Formula:
The
Concept
L = \W,"
Operations
of a Decision
Science
Preemptive
Research
Calculus,"
Vol.
9, No.
Management
Vol.
13, No.
12, pp. B707-B717.
or with Breakdowns,"
Priorities
24 INTERFACESOctober 1981
Detroit
2000SecondAvenue
,.
48226
Michigan
V":-^r\
v Detroit,
VJV!
(313)237-8000
November 8,
1979
RLHrjmg
attachment
INTERFACES
October
1981
25