Documente Academic
Documente Profesional
Documente Cultură
15
C H A P T E R
Simulation Modeling
227
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 228
The following random numbers have been generated: SOLUTIONS TO DISCUSSION QUESTIONS
60, 87, 46, 63 (set 1) AND PROBLEMS
52, 78, 13, 06, 99, 98, 80, 09, 67, 89, 45 (set 2)
15-1. Advantages of simulation: (1) relatively straightforward;
Use set 1 of these to generate lead times and use set 2 to simulate (2) can solve large, complex problems; (3) allows what if ques-
daily demand. Simulate 2 ordering periods with this and determine tions; (4) does not interfere with real-world systems; (5) allows
how often the company runs out of stock before the shipment arrives. study of interactive variables; (6) allows time compression; (7) al-
SOLUTION: lows inclusion of real-world complications. Disadvantages;
(1) cost; (2) no optimal solutions; (3) managers must generate con-
Sales RN Lead Time RN ditions to test; (4) each model is unique.
7 0130 3 0120 15-2. a. Inventory ordering policy: May require simulation if
8 3170 4 2180 lead time and daily demand are not constant. Also useful if data do
9 7100 5 8100
not follow traditional probability distribution.
b. Ship docking in port to unload: If arrivals and unload-
First order: RN 60 so lead time 4 days.
ings do not follow Poisson/exponential distributions common to
Demand day 1 8 (RN 52) queuing problems, or if other queuing model assumptions are vio-
day 2 9 (RN 78) lated (for example, FIFO not observed).
day 3 7 (RN 13) c. Bank teller service windows: If arrivals or service
day 4 7 (RN 06) times do not follow standard distributions, or if several waiting
Total demand during lead time 31. Since the reorder point is 32, lines exist, may be easier to use simulation.
there is no stockout. d. U.S. economy: Because mathematical equations and
Alternative Example 15.3: The time between arrivals at a relationships are too complex to solve mathematically and because
drive-through window of a fast-food restaurant follows the distrib- an optimal solution may not exist.
ution given below. The service time distribution is also given in 15-3. Problems with conditions of certainty can be solved more
that table in the right column. Use the random numbers provided easily by other QA techniques. Problems that require quick an-
to simulate the activity of the rst ve arrivals. Assume that the swers that cannot wait for a simulation model to be built are a sec-
window opens at 11:00 A.M. and the rst arrival is after this, based ond category.
on the rst interarrival time generated. 15-4. Major steps are: (1) dene problem, (2) introduce impor-
tant variables, (3) construct model, specify values to test, (4) con-
Time
Between Service duct simulation, (5) examine results, (6) select best plan.
Arrivals Probability Time Probability 15-5. Monte Carlo steps: (1) set up probability distribution(s),
1 0.2 1 0.3 (2) set up cumulative probabilities, (3) establish random number
2 0.3 2 0.5 intervals, (4) generate random numbers, (5) simulate trials.
3 0.3 3 0.2 15-6. Random numbers can be generated by: (1) computer pro-
4 0.2 grams such as Excel, (2) spinning a dial on a uniform wheel,
(3) pulling numbers from an urn, (4) using a random number table,
Random numbers for arrivals: 14, 74, 27, 03 and (5) creating an algorithm such as the midsquare method.
Random numbers for service times: 88, 32, 36, 24
15-7. Validation is the process of comparing a model to the real
What time does the fourth customer leave the system? system that it represents. Verication is the process of determining
SOLUTION: that a model is internally consistent and follows the logic of the
conceptual model.
Time
15-8. A next event time increment model should be used when
Between Service
it is necessary to record information each time the system status
Arrivals Prob. RN Time Prob. RN
changes. For example, if we wish to determine the average time a
1 0.2 0120 1 0.3 0130 customer waits in line, it is necessary to know precisely what time
2 0.3 2150 2 0.5 3180 that person enters the line and the time that person leaves the line.
3 0.3 5180 3 0.2 8100
This cannot be accomplished with a xed time increment model.
4 0.2 8100
15-9. The results would very likely change, and perhaps signif-
First arrival (RN 14) at 11:01. Service time 3 (RN 88). icantly, if a longer period was simulated. The 10-day simulation is
Leaves at 11:04. valid only to illustrate the features of the system. It would not be
safe to forecast based on that short a span.
Second arrival (RN 74) at 11:04 (3 minutes after rst).
15-10. A computer is necessary for three reasons: (1) it can do
Service time 2 (RN 32). Leaves at 11:06.
time periods or trials in a matter of seconds or minutes, (2) it can
Third arrival (RN 27) at 11:06. Service time 2 (RN 36). quickly examine and allow change in the complex interrelation-
Leaves at 11:08. ships being studied, and (3) it can internally (through a subroutine
Fourth arrival (RN 03) at 11:07. Must wait 1 minute for service or function statement) generate random numbers by the thousands
to start. Service time 1 minute (RN 24). Leaves at 11:09. or millions.
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 229
15-11. Operational gaming is a simulation involving competing 15-15. a, b. Lundbergs car wash:
players. Systems simulation tests the operating environment of a
large system such as a corporation, government, or hospital. Random
Number of Cumulative Number
15-12. Simulation may very well increase in use for several reasons:
Cars Probability Probability Interval
(1) computers are in all types and sizes of businesses; (2) simulation
languages may be rened and made easier for noncomputer man- 3 or less 0 0.00
agers to use, especially with the advent of spreadsheet approaches; 4 0.10 0.10 0110
(3) the mass of graduates educated in QA entering the corporate 5 0.15 0.25 1125
6 0.25 0.50 2650
world is growing, decreasing resistance to sophisticated techniques;
7 0.30 0.80 5180
and (4) complex problems will not become fewer in nature. 8 0.20 1.00 8100
15-13. C is a common, popular language. Learning a special- 9 or more 0.00 1.00
ized simulation language can be time consuming and difcult. For 1.00
regular simulation, regular languages may sufce. They may take a
few extra seconds of computer CPU time (GPSS and SIMSCRIPT c.
are very efcient), but computer time is often a xed cost.
Random Simulated
15-14.
Hour Number Arrivals
Random
1 52 7
Number of Number
2 37 6
Failures Interval
3 82 8
0 0106 4 69 7
1 0719 5 98 8
2 2044 6 96 8
3 4572 7 33 6
4 7392 8 50 6
5 9399 9 88 8
6 00 10 90 8
11 50 6
Number of A.C. 12 27 6
Simulated Random Compressors Simulated 13 45 6
Period Number to Fail This Year 14 81 8
15 66 107
1 50 3 105
2 28 2
3 68 3 105
4 36 2 Average number arrivals per hour 7.00 cars
15
5 90 4
6 62 3 15-16. Using the probability distribution developed in Problem
7 27 2 15-15, the expected value is E(X) 4(0.10) 5(0.15) 6(0.25)
8 50 3 7(0.30) 8(0.20) 6.35. The average number of arrivals in
9 18 1 Problem 15-15 was 7. If another simulation were performed, the
10 36 2 average number would not be 7 but would be some other number.
11 61 3 If enough simulations were performed, the average number com-
12 21 2 puted in the sample would approach the expected value computed
13 46 3
with the formula.
14 01 0
15 14 1 15-17. Higgins plumbing:
16 81 4
17 87 4 Random
18 72 3 Number
19 80 4 Heater Sales Probability Intervals
20 46 3 3 0.02 0102
4 0.09 0311
No, its not common to nd three or more years in a row with two 5 0.10 1221
or less compressor failures. 6 0.15 2236
7 0.25 3761
8 0.12 6273
9 0.12 7485
10 0.10 8695
11 0.05 9600
1.00
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 230
a. 15-18. a.
Random Simulated
New
Week Number Sales
Random Number
1 10 4 Unloading Rate Interval
2 24 6
1 0103
3 03 4
2 0415
4 32 6
3 1655
5 23 6
4 5683
6 59 7
5 8495
7 95 10
6 9600
8 34 6
9 34 6
10 51 7
11 08 4 6
b. Average number delayed 0.40
12 48 7 15
13 66 8 31
14 97 11 Average number of arrivals 2.07
15
15 03 4
16 96 11 31
Average number unloaded 2.07
17 46 7 15
18 74 9
The short span simulated (15 days) introduces volatility in the
19 77 9
20 44 137
daily arrival rate (from 2.73 arrivals/day in Table 15.12 to only
139 2.07 in the above simulation). This, coupled with a speedier un-
loading rate, produces a much lower average delay rate (from
1.33/day down to only 0.40/day).
With a supply of 8 heaters, Higgins will stock out 5 times during
the 20-week period (in weeks 7, 14, 16, 18, and 19).
Interval Demand RN
X P(X) Cum. Prob. of RN Day RN (100s) Probability Interval Weather Day RN Weather
23 0.15 0.15 115 1 7 23 0.8 180 Good 1 88 Bad
24 0.22 0.37 1637 2 60 25 0.2 8100 Bad 2 2 Good
25 0.24 0.61 3861 3 77 26 3 28 Good
26 0.21 0.82 6282 4 49 25 4 49 Good
27 0.18 1.00 8300 5 76 26 5 36 Good
6 95 27 6 87 Bad
7 51 25 7 21 Good
8 16 24 8 95 Bad
9 14 23 9 50 Good
10 85 27 10 24 Good
c.
Start
Yes
Yes
Yes
End
Cumulative RN
Probability Probability Interval Demand
0.20 0.20 120 0
0.40 0.60 2160 1
0.20 0.80 6180 2
0.15 0.95 8195 3
0.05 1.00 9600 4
Cumulative RN Lead
Probability Probability Interval Time
0.15 0.15 115 1
0.35 0.50 1650 2
0.50 1.00 5100 3
Random numbers will differ from student to student. Ours were 15-24. Flow diagram for Port of New Orleans simulation:
selected from the right-hand column of Table 15.5.
Daily order cost ($10)(0.2 order/day) Begin Day
of Simulation
$2.00
Daily holding cost ($.50/unit/year)(4.8 units/day)
$2.40 Select Random
Number to Generate
Daily stockout cost ($8/lost sale)(0.6 lost sales/day) Overnight Arrivals
$4.80
Total daily cost $9.20 Add New Arrivals
This cost is greater than the Q 10, ROP 5 policy. However, to Number Delayed from
the short period simulated does not really permit a valid analysis Previous Day to Find
Total to Be Unloaded
and comparison.
Select Random
Number to Generate
Daily Unloadings
Is Total to
Be Unloaded No
> Todays Number Delayed = 0
Unloading
Rate?
Yes
Have
No Enough
Days Been
Simulated?
Yes
Compute Appropriate
Averages and Costs
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 235
Time
Time Repair-
Between Person Repair
Break- Break- Time of is Free Time Time No. Hours
down Random down Break- to Begin Random Required Repair Machine
Number Number (Hours) down This Repair Number (Hours) Ends Down
1
1 69 22
02:30 02:30 37 1 03:30 1
2 84 3 05:30 05:30 77 1 06:30 1
1
3 12 12
07:00 07:00 13 12 07:30 12
4 94 3 10:00 10:00 10 12 10:30 12
5 51 2 12:00 12:00 02 12 12:30 12
6 36 2 14:00 14:00 18 12 14:30 12
1
7 17 1
2
15:30 15:30 31 1 16:30 1
8 02 12 16:00 16:30 19 12 17:00 1
1
9 15 1 2
17:30 17:30 32 1 18:30 1
1 1
10 29 2 19:30 19:30 85 12
21:00 12
1
11 16 12
21:00 21:00 31 1 22:00 1
1 1
12 52 2 23:00 23:00 94 12
00:30 12
1 1
13 56 2 01:00 01:00 81 12
02:30 12
14 43 2 03:00 03:00 43 1 04:00 1
1
15 26 12
04:30 04:30 31 1 05:30 11
1
Total 14
2
10 repairs at 10 repairs at
$50/per hour $50/per hour
(1 hour per (2 hours per Random
repair) repair) Arrival Cumulative Number
$80 $500 $320 $1,000 Distribution Probability Probability Interval
$580 $1,320 20 minutes early 0.20 0.20 0120
Cost/hour $580/380 hours Cost/hour $1,320/1,210 10 minutes early 0.10 0.30 2130
hours On time 0.40 0.70 3170
$1.53 per hour $1.09 per hour 10 minutes late 0.25 0.95 7195
20 minutes late 0.05 1.00 9600
b. Analytical approach to Brennan Aircraft problem:
Expected number of hours between failures if one pen
Random
replaced Exam Time Cumulative Number
10(0.05) 20(0.15) 30(0.15) Distribution Probability Probability Interval
40(0.20) 50(0.20) 60(0.15) 20% faster than expected 0.15 0.15 0115
70(0.10) In expected time 0.50 0.65 1665
20% later than expected 0.25 0.90 6690
42 40% later than expected 0.10 1.00 9100
Random
Order Cumulative Number
(Sq. Ft.) Probability Probability Interval
8,000 0.45 0.45 0145
11,000 0.55 1.00 4600
Demand distribution:
Pelnor should add more storage capacity if the current policy will
be maintained. However, students should recognize that a buildup
will continue innitely. The expected supply is 0.45(8,000)
0.55(11,000) 9,650 sq ft. The expected demand 8,750 sq ft.
Over the long run, the on-hand supply will grow and grow to an
innite level.
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 238
15-29. a. Random number intervals must be set for each from- 15-30.
to combination:
Time Random
From-To Random Number Between Number
Combination Interval Arrivals Probability Interval
Random
Person Number From To
1 52 Initial exam Operating room
37 Operating room Observation
82 Observation Out processing
2 69 Initial exam Observation
98 Observation Out processing
3 96 Initial exam Out processing
4 33 Initial exam X-ray
50 X-ray Observation
88 Observation Out processing
5 90 Initial exam Out processing
6 50 Initial exam Operating room
27 Operating room Observation
45 Observation Out processing
7 81 Initial exam Out processing
8 66 Initial exam Observation
74 Observation Out processing
9 30 Initial exam X-ray
59 X-ray Observation
67 Observation Out processing
10 60 Initial exam Operating room
60 Operating room Observation
80 Observation Out processing
Daily
Random Sales Gross Random Ingred. 2 Ingred. 2 Ingred.
Day Number Price Sales Number Cost/Unit Cost Total 1 Cost Profit
1 52 $350 10,500 37 $40 $1,440 $1,250 $7,810
2 06 300 9,000 66 40 1,440 1,250 6,310
3 50 350 10,500 91 45 1,620 1,250 7,630
4 88 400 12,000 35 40 1,440 1,250 9,310
5 53 350 10,500 32 40 1,440 1,250 7,810
6 30 350 10,500 00 45 1,620 1,250 7,630
7 10 300 9,000 84 45 1,620 1,250 6,130
8 47 350 10,500 57 40 1,440 1,250 7,810
9 99 400 12,000 07 35 1,260 1,250 9,490
Random number intervals for sales price: Random number intervals for cost 2:
0120 $300 0110 $35
2170 $350 1170 $40
7100 $400 7100 $45
d. Expected prot from simulation $7,770/day
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 241
*Random numbers taken from column 18 of Table 15.5, reading top to bottom, then from column 17, reading bottom to top.
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 242
*Random numbers taken from column 18 of Table 15.5, reading top to bottom, then from column 17, reading bottom to top.
Maruggis balance never drops below $400 and he should be able A spreadsheet model for the scenario should be constructed
to balance his account. consisting of the following items:
The above table is, in effect, the actual model of the scenario. The trial. Here, for example, they are the average and maximum of
headings are the key to expressing the relationships between the time in the system (that is, time on hold time being served),
various model elements as relatively simple (that is, intuitive) for- time on hold, time server idle, and server utilization. (All times
mulas. The relationships need only be expressed once, for cus- are in minutes.)
tomer/caller 1, are then copied down for as many callers as are To obtain steady-state results, the trial above has to be repeated
required in the simulation window (here 2 hours). Relevant mea- a sufcient number of times, where each time the measures of
sures of performance have to be chosen and evaluated for each performance are stored in a summary table seen on the next page.
Summary of Trials
TIME TIME TIME TIME TIME TIME
IN IN SERVER PERCENT IN IN SERVER PERCENT
TRIAL SYSTEM QUEUE IDLE UTILIZATION TRIAL SYSTEM QUEUE IDLE UTILIZATION
Avg.: 8.9 5.7 0.4 88.4 Max.: 18.0 14.5 4.3 98.1
0 7.6 4.0 0.43 89.5 0 14.0 10.0 4.0
1 8.6 5.1 0.23 93.9 1 17.0 15.0 5.0
2 3.8 1.2 1.00 72.4 2 10.0 7.0 5.0
3 17.3 13.8 0.25 93.4 3 30.0 27.0 3.0
4 4.8 1.8 0.65 82.3 4 11.0 6.0 6.0
5 6.0 2.8 0.80 80.0 5 14.0 9.0 4.0
6 7.8 4.1 0.25 93.6 6 13.0 11.0 5.0
7 4.7 1.9 0.55 83.5 7 9.0 7.0 5.0
A straightforward way of determining when the various mea- Importance of having relevant and meaningful measures of per-
sures of performance have stabilized is graphically (see Figures 1 formance (for example, average wait times may be low but if
and 2), and these long-term averages can be used to make recom- their standard deviation is large, the maximum times in line
mendations regarding hiring a second reservation agent. may be unacceptable.
For part (b) essentially the approach above is repeated with Human and soft issues: Is utilization of, for example, 95% for
the new arrival distribution. a human server appropriate or indeed desirable?
It is possible to automate the procedure above by using What are the assumptions in the model [for example, callers do
macros, although by no means are they necessary or required. not hang up, changeover times are negligible (or included in
(Students with more advanced spreadsheet experience often elect service times), there are no calls in the system at time zero].
to do this spontaneously.) What are the implications/restrictions?
Discussion: A number of items can be addressed in the dis- The model and simulation can easily be extended to become
cussion of the case: more realistic by incorporating the various assumptions and simulat-
Reliability of the recommendations, vis--vis the number of tri- ing callers hanging up, multiple servers, and so on, depending on
als, structure of the model, and so on. the level of the class and the students familiarity with spreadsheets.
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 245
14
12
Minutes
10
8
6
4
2
0 10 20 30 40 50
Number of Trials
Figure 2. Time in System and Time in Line: Summary of Trials
Conclusion: The application described above, although not changing the arrival interval and service time distributions. Fi-
exceedingly complex, is nevertheless considered to be more realistic nally, a not insignicant benet is the popularity among students
than could normally be attempted without resorting to a simulation of this approach, where they can dramatically demonstrate for
model. Even the least quantitatively oriented students persevere themselves the consequences of various what if scenarios by
with implementation and experimentation of these models and amending a few parameters.
based on formal feedback (evaluations) and informal feedback
(verbal comments) view the experience as highly benecial. Two SOLUTION TO STATEWIDE DEVELOPMENT
reasons are given for this: First, the use, abuse, and relevance of
modeling in decision making is demonstrated dramatically in
CORPORATION CASE
We will assume that the time of concern is from 8:00 a.m. until
problems more realistic and complex than the typical compara-
8:00 p.m. each weekend day. The results will vary based on the
tively trivial or textbook examples, which require hand or calcula-
random numbers used. However, the following tables can be used
tor computation. Second, the time and energy that students spend
to assign random numbers to the events. The times are expressed
in acquiring the appropriate spreadsheet skill levels are perceived
in hours.
by them as acquiring an extremely useful skill, since an in-depth
familiarity with a spreadsheet package is now a necessary require-
ment for a business major.
The application described above has been used successfully Time between Cumulative Random number
calls (Hrs.) Probability probability interval
in the classroom to demonstrate the principles and usefulness of
simulation. Clearly, many other (nonqueuing) situations lend 0.5 0.15 0.15 115
themselves to be modeled in the same manner. The case can read- 1.0 0.30 0.45 1645
ily be updated/renewed from semester to semester (1) by changing 1.5 0.30 0.75 4675
2.0 0.25 1.00 7600
Alabama Airlines to a bank, movie theater, and so on, and (2) by
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 246
The time for a repair that requires a trip to the supply house
will be 2 hours and 30 minutes 30 minutes to nd the problem,
1 hour to make the trip, and 1 hour to install the part.
Using an Excel spreadsheet, the simulation can be performed
as shown.
*Because a trip to the supply house was necessary for this service call, the time for the repair is 2.5 hours. The random number for the repair is not used for
this call.
**Only 5 calls are shown. For the entire simulation, 150 calls were simulated. This was repeated several times.
Table 1
Trucks Trucks
Freight Required Required
per Day 75% 3 Trips/Day 25% 3 Trips/Day
Mo. RN Freight (30 Days) Noncontainerized 60 Tons/Trip Containerized 53 Tons/Trip
1 63 171,000 5,700 4,275 24 1,425 9
2 88 197,000 6,567 4,925 27 1,642 10
3 55 165,000 5,500 4,125 23 1,375 9
4 69 176,000 5,867 4,400 24 1,467 9
5 13 124,000 4,133 3,100 17 1,033 7
6 17 131,000 4,367 3,275 18 1,092 7
7 36 150,000 5,000 3,750 21 1,250 8
8 81 186,000 6,200 4,650 26 1,550 10
9 84 190,000 6,333 4,750 26 1,583 10
10 63 172,000 5,733 4,300 24 1,433 9
11 70 177,000 5,900 4,425 25 1,475 9
12 06 110,000 3,667 2,750 15 917 6
It is noted that a truck can carry two 20-foot containers, so Biales Waste Disposal, GMBH.:
that the total cargo is 40 tons. Thus average cargo hauled by con- Costs in German Marks (DM):
tainerized freight is 53 tons, or 0.6 60 0.2 45 0.2 40. Shipment: DM 900 per load
Daily cargo hauled is 3 53 159 tons/day. Loading/unloading DM 120 per load
Table 1 shows simulated monthly freight shipments. Cargo is Overhead (DM 41,000/25) DM 1,640 per load
then categorized into containerized and noncontainerized ship- DM 2,660 per load
ments. The daily truck requirements for each month is projected.
As seen from the simulated years operation, the daily truck
requirement for noncontainerized cargo ranges from 15 to 27. For
containerized cargo it ranges from 6 to 10 trucks. The maximum
number of trucks required to handle the load in any single period
is 37. It should be noted that the utilization factor is 0.96. Hence
the number of trucks should be adjusted upward accordingly.
A discussion of obtaining a simulated Freight from a Ran-
dom Number should be highlighted; for example, why does the
random number 63 result in a freight of 171,000?
Number Revenue Repair times are simulated by the use of cumulative normal
Truckload Random of Random per Total distributions for routine and accelerated repair procedures. It is
Simulation Number Barrels Number Barrel Revenue suggested that the student begins the simulation process by using
only the routine repair procedure. Calciner availability should be-
1 52 38 06 DM50 DM1,990
come apparent as the simulation process is executed.
2 37 38 63 60 2,280 Table 1, on the next page, is a one year (366 days of failures)
3 82 43 57 60 2,580 simulation of calciner failures and repairs using both normal and
4 69 43 02 50 2,150
accelerated procedures. (For practicality in scale interpretation,
two-digit random numbers that would yield greater than 60 days
5 98 48 94 80 3,840 between failures were not accepted. This corresponds to a failure
6 96 48 52 60 2,880 of 0.5 or less.) Cumulative calciner downtime for normal repairs
7 33 38 69 70 2,660 was 355.5 days, and for accelerated repairs was simulated to be
181 days.
8 50 38 33 60 2,280
Lost production for normal repairs:
9 88 43 32 60 2,580
355.5 calciner days 100 tons/calciner day 35,550 tons
10 90 48 30 60 2,880
Lost production for accelerated repairs:
11 50 38 48 60 2,280
181 calciner days 100 tons/calciner day 18,100 tons
12 27 33 88 70 2,310
Additional output for accelerated repairs:
13 45 38 14 50 1,900 35,550 tons 18,100 tons 17,450 tons
14 81 43 02 50 2,150
Prot and overhead contribution:
15 66 43 83 70 3,010 17,450 tons $12/ton $209,400
16 74 43 05 50 2,150 For calciner failures 116 or one years simulation, the fol-
17 30 38 34 60 2,280 lowing are rental costs for crane. (Note: there is a $12,000 mini-
18 59 43 55 60 2,580 mum charge per rental.)
19 67 43 09 50 2,150
Dryer Failure Repair Days Cost
20 60 43 77 70 3,010
1 13 $19,500
21 60 43 08 50 2,150 2 9.5 14,250
3* 7 10,500*
22 80 43 45 60 2,580
4* 11 16,500
23 53 43 84 70 3,010 5* 12.5 18,750
24 69 43 84 70 3,010 6* 8 12,000
7 9.5 14,250
25 37 38 77 70 2,660 8 11 16,500
DM 63,260 9 9.5 14,250
10 13 19,500
11 10.5 15,750
Average income per load DM2,544.40 12* 13.5 20,250
13 7.5 11,250*
Loss per load DM2,660 2,554.40 DM105.60 14 8.5 12,750
Loss per year 25 DM105.60 DM2,640 15 10.5 15,750
16 10.5 $215,750
The conclusion, based on just one short simulation, is that money
Total $247,500
will be lost by continuing service to Italy.
*As seen in the simulation, the crane was rented continuously for repairs
Buffalo Alkali and Plastics 36 and 1213. These represent only one rental and the minimum charge
The solution to the case is based on Monte Carlo simulation of of $12,000 applies once each for these two cases.
shell failures and repairs by normal as well as accelerated proce-
dures. Simulation models are the cumulative probability distribu- Hence, rental costs for 16 failures $247,500; prot and over-
tions displayed in the tables on the next page. head contribution amounts to $209,400 for additional output.
Simulation of shell failures is achieved through the use of the Since costs exceed contribution, the accelerated procedure should
cumulative Poisson distribution that is determined by an average not be adopted.
failure rate of 1.5 per month. Failure frequency is converted to
time between failures and is plotted against cumulative proba-
bility. The arrival of failures is simulated by selecting two-digit
random numbers which are projected on the cumulative probabil-
ity axis. Corresponding time between failures are then deter-
mined by these randomly selected probabilities.
REVISED
M15_REND6289_10_IM_C15.QXD 5/15/08 10:38 PM Page 249
*37 days is calculated as follows: 3.5 days (elapse before repairs can begin) plus 14 days (repair) plus 19.5 (previous downtime).
**22.5 days is calculated as follows: 9.5 days (repair time) plus 13 days (previous downtime).