You are on page 1of 6

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/268347733

Monte Carlo Simulation for Sparepart


Inventory

Article

CITATIONS READS

0 160

3 authors, including:

Nyoman Pujawan Niniet Indah Arvitrida


Institut Teknologi Sepuluh Nopember Institut Teknologi Sepuluh Nopember
41 PUBLICATIONS 573 CITATIONS 11 PUBLICATIONS 2 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Nyoman Pujawan on 18 August 2015.

The user has requested enhancement of the downloaded file.


The 11th Asia Pacific Industrial Engineering and Management Systems Conference
The 14th Asia Pacific Regional Meeting of International Foundation for Production Research
Melaka, 7 – 10 December 2010

Monte Carlo Simulation for Sparepart Inventory

I Nyoman Pujawan†1, Niniet Indah Arvitrida2, and Bathamas P. Asihanto


Department of Industrial Engineering
Sepuluh Nopember Institute of Technology
Surabaya, Indonesia
Email: pujawan@ie.its.ac.id1
niniet@ie.its.ac.id2

Abstract - Most inventory models are based on assumptions that demand follows a certain theoretical
distribution (such as Normal or Poisson distribution). For items with relatively fast moving, such an
assumption may hold true. However, for slow moving items such as spare parts, we rarely found that demand
follows such theoretical distribution. Consequently, using standard inventory control models to set inventory
parameters for spare parts would result in either low service level, high inventory level, or both. In this study
we use Monte Carlo simulation to determine inventory control parameters for spare parts. We employ a well
known periodic review base-stock inventory control models where the replenishment decisions are governed
by maximum and minimum inventory levels. The initial solutions were generated using standard inventory
control models. Monte Carlo Simulation is then used to evaluate the parameters against other neighboring
values in terms of inventory cost and service level. We apply the procedure for cabin spare parts in an aircraft
maintenance facility.

Keywords: inventory management, spare parts, simulation.

1. INTRODUCTION academics. One of the reasons why inventory management


for spare parts received only little attention is because the
Managing spare parts inventory has always been demand pattern for spare parts is slow moving, often
challenging for a number of (conflicting) reasons. First, intermittent or lumpy (Teunter et al, 2010). As a
demand for spare parts is often intermittent or lumpy consequence, they are difficult to forecast. Lumpy demand
making it difficult to forecast. Second, the required service is characterized by many periods with zero demand
level is high for most spare parts as unavailability would (Pujawan & Kingsman, 2003). The inventory control
result in costly consequences such as lengthy facility models that have been developed for fast moving items are
breakdown or high costs associated with urgent purchase. hardly ever appropriate for spare parts.
Third, the price for a spare part could be very high making There are a number of different stream of papers about
it less favorable to hold inventories. spare parts inventory. One of the streams is dealing with
For many companies, inventory investment for spare forecasting. Croston (1972) is among the early papers
parts is enormously large. An informal discussion with a proposing a method for forecasting demand of spare parts
gas exploration company in Indonesia reveals that the spare inventory. This model basically decomposes demand
parts inventory value in this company is about US$ 47 pattern into two components, i.e., the magnitude of positive
million. If the annual inventory holding cost is about 30% demand and time between demands. This method is further
then the spare parts cost this company about US$12 million discussed or developed by other authors such as Regattieri
per year. In addition, the turnover ratio of spare parts is et al (2005) and Syntetos & Boylan (2005).
usually very low (usually between 0.5 and 1 which means The other group of authors deal with optimum setting
that on average one spare part is consumed every one to of replenishment parameters such as reorder point and
two years. maximum inventory. The most popular method appears to
Despite that managing spare parts inventory is critical be the (s, S) or its periodic review version (R, s, S). This
to most companies, studies addressing this issue is fairly model works as follows. In a review period, when
limited, in contrast to inventory for raw materials or inventory falls to less than or equal to s then place an order
finished goods that have received much attention from to bring the inventory level back to S. The determination of

________________________________________
† : Corresponding Author
The 11th Asia Pacific Industrial Engineering and Management Systems Conference
The 14th Asia Pacific Regional Meeting of International Foundation for Production Research
Melaka, 7 – 10 December 2010

optimum s and S values has been discussed by many According to Regattieri et al. (2005), the demand pattern of
authors, but most of them have assumed that demand a material is characterized by two attributes, i.e., the
follows Normal or Poisson distribution. average demand internal (ADI) which is the average time
Due to its demand being intermittentt or lumpy, the between occurrence of two successive demands and the
demand distribution for spare parts hardly follows any coefficient of variation of the positive demand (CV). Large
theoretical distribution. Consequently, using standard values of both ADI and CV indicated high demand
inventory control models to set inventory parameters for uncertainty and hence, large inventory investment is
spare parts would result in either low service level, high required to achieve an acceptable service level. As shown
inventory level, or both. In this study we use Monte Carlo by figure 1, most items have CV larger than 50% and ADI
simulation to determine inventory control parameters for more than 1.5, meaning that the degree of lumpiness of
spare parts. By using simulation, we can capture various those items is high.
demand characteristics including those which do not follow
any theoretical distribution.

2. CASE STUDY

We do the case study in PT. GMF Aeroasia (will be


referred to as GMF in the rest of this paper), an aircraft
maintenance facility, a sister company of Garuda Indonesia.
The GMF manages the maintenance, repair, and overhaul
of aircraft from various airline companies. A critical
activity within the maintenance, repair, and overhaul is the
management of spare parts inventory. With the increasing
variety of aircrafts and the fact that each aircraft consists of Figure 1 – Demand pattern classification
a very large number of components, the management of
spare parts inventory is becoming so critical to achieve an
efficient operations as well as high service level.
In general, there are two major classification of spare 2.1 The Use of Analytical Models
parts related to an aircraft. The first is spare parts that are
required for safety flight such as break, wheel, and many Most inventory control models have been developed
others. If any of these components does not working well, based on an assumption that demand follows a certain
the aircraft can not or is not allowed to fly because it will theoretical distribution. In particular, majority of the
pose high safety risks. The second classification is related models have assumed that demand follows either normal or
to cabin materials. If any of the cabin materials is not Poisson distribution. In reality, when demand is lumpy,
functioning well, the aircraft is still able to fly, but it will such an assumption is hardly ever holds true. Consequently,
compensate the convenience of the passengers. the use of analytical models for lumpy demand often ends
In this study we will focus on seat materials for B737- up with inappropriate inventory parameters. In this study
800 as one class of cabin materials. In general the cabin we have used a periodic base-stock policy called (R, s, S)
materials can be classified into six groups, namely where R stands for interval between review periods, s is
components related to seat (such as hydro-lock, arm rest, in reorder point and S is maximum inventory position (Silver
arm table), in flight entertainment (such as seat electronic et al, 1998). The model reads as follows. During a review
box, video monitor, and harness), lavatory (such as portable period, if inventory position falls below s then we have to
water tank, toilet assembly), lighting, luggage bin and order to bring inventory position back to S. Inventory
galley (such as oven, coffee maker, and boiler). The current position is on hand inventory plus on order minus
situation in GMF shows that many of the cabin materials backorder.
are on what they refer to as “hot item list” or HIL, meaning Determination of optimum R, s, and S is not an easy
that those items are not available when needed. For B 737- task. All of these three parameters will affect both
800 alone, at the time this observation is made, there were inventory costs and service level. One of the analytical
45 items on HIL status. Among those on the HIL status, the methods often used for determining those parameters is
majority are for seat components. There are in total 94 called power approximation. However, this approximation
items that belong to seat components of B737-800. does not always provide good results, especially when
Like other spare parts in general, we found that the demand is lumpy. We have tried to use the power
demand pattern is lumpy for most of the seat materials. approximation to set inventory parameters for a couple of
The 11th Asia Pacific Industrial Engineering and Management Systems Conference
The 14th Asia Pacific Regional Meeting of International Foundation for Production Research
Melaka, 7 – 10 December 2010

spare parts. We then simulate the performance of those • Update on order quantity: O(t) = O(t-1)+Q(t)-R(t)
parameters to see if they can give good performance in • Determine order receipt schedule: R(t+L) = Q
terms of service level. One of our observations is that this Where I(t)is inventory at the end of period t, O(t) is on
approximation fails to achieve the intended service level for hand inventory, B(t) is backlog in period t, R(t) is schedule
almost all situations. As shown in figure 2, the actual receive in period t, Q(t) is order quantity in period t, and L
service level is always below the theoretical service level. lead time.
As an illustration, we will use one part called PN30-
802 (we have made an arbitrary part number, different from
the actual number used by the company). The baseline
parameters obtained from the power approximation is s = 5
and S = 6, with an intention to achieve a service level of
95%. We then do a parameter search by lowering and
increasing both s and S values around their baselines. The
result is shown in table 1.
Assuming that service level 95% is required, then we
have to exclude those alternatives with actual service
level below 95%. Figure 3 shows that the remaining
Figure 2 - The actual service level (vertical) is always candidates are those at the right side of the vertical dotted
lower than the intended service level (horizontal) line. Among the remaining candidates, we then have to
again exclude those that are not at the efficient frontier.
Efficient frontier can simply be defined as a curve
2.2 The Use of Monte Carlo Simulation connecting the most efficient candidates. The most efficient
candidates can be visually seen as those at the outer side of
We have used the Monte Carlo simulation to evaluate group. By looking at figure 3, there appears to be only 4
the performance of the parameters from power candidates are remaining. Now, the decision maker has to
approximation. As we believe that those parameters are make a choice among the 4 remaining candidates. Since the
rarely the best ones, we also evaluate other set of service level for each part should be dependent on part
parameters close to those recommended by the power criticality, choosing a solution with a service level much
approximation. Since the demand for parameters is mostly higher than the initial target would be inappropriate. Thus,
slow moving, the choice of s and S values is relatively as long as the service level target has been set carefully
limited, making the simulation quite effective for finding with consideration of item criticality then the simple rule of
the best set of parameters. thumb is to choose a candidate with lowest cost in the
The simulation procedure can be summarized as efficient frontier.
follows. First, based on historical data of 28 months, we
model the empirical demand distribution. As an illustration,
we will use item PN102-504 (we have made an arbitrary
part number, different from the actual number used by the
company). The historical data shows that 71% of months
has no demand for this item. About 3.5% has a demand of 1
or 2, 10.7% has a demand of 3 or 4 and the rest is 5 or 6. To
generate the demand, a uniform (0,1) random number is use.
If random number falls between 0 and 0.710 then demand
Efficient frontier
is set to be zero. If random number is between 0.710 and
0.745 then demand is either 1 or 2 with an equal probability,
and so on.
Let D(t) = demand in period t, obtained through the
above procedure. We then do the following steps:
• Update inventory: I(t) = I(t-1) + R(t) – D(t) –B(t)
• Update on-hand inventory: O(t) = max (0, I(t)) Figure 3 - Service level (horizontal) and cost (vertical) for
• Update backlog: B(t) = max (0, -I(t)) each candidate
• Calculate inventory position: IP(t) = I(t) + O(t)
• Determine ordering decision: If IP(t) < s, Q = S –
IP(t), otherwise Q = 0
The 11th Asia Pacific Industrial Engineering and Management Systems Conference
The 14th Asia Pacific Regional Meeting of International Foundation for Production Research
Melaka, 7 – 10 December 2010

demand: A case study. European Journal of Operational


3. DEVELOPMENT OF DECISION SUPPORT Research 205, pp. 313 – 324.
SYSTEM
Pujawan, I N. dan Kingsman, B. G. (2003). Properties
of lot sizing rules under lumpy demand. International
A spreadsheet-based decision support system (DSS)
Journal of Production Economics 88, pp. 295-307.
has been developed to assist decision maker in performing
the above procedure until the best candidate is obtained. We Regattieri, A., Gamberi, M., Gamberini, R., dan
have tested the DSS with the data of 20 seat components. Manzini, R. (2005). Managing lumpy demand for aircraft
The tests show that the DSS has worked well and practical spare parts. Journal of Air Transport Management 11, pp.
enough to use. 426 – 431.
Rustenburg, W. D., van Houtum, G. J., dan Zijm, W.
H. M. (2001). Spare parts management at complex
4. CONCLUDING REMARKS
technology-based organizations: An agenda for research.
International Journal of Production Economics 71, pp. 177
In this study, a decision support system has been
– 193.
developed to obtain the best replenishment policy for spare
parts. As the demand pattern for most spare parts typically Syntetos, A. A., dan Boylan, J. E. (2005). The
is lumpy, it is hardly ever true that it follows any theoretical accuracy of intermitten demand estimates. International
distribution such as Normal or Poisson. The analytical Journal of Forecasting 21, pp. 303 – 314.
models for inventory policy is normally assumed that the
Teunter, R. H., Syntetos, A. A., dan Babai, M. Z.
demand follows Normal, Poisson, or any other standard
(2010). Determining order-up-to levels under periodic
distribution. This has lead this study to use simulation to
review for compound binomial (intermittent) demand.
obtain the best inventory parameters. We use a periodic
European Journal of Operational Research 203, pp. 619 –
review (R, s, S) inventory model and the parameter s (the
624.
reorder point) and S (the maximum stock) is obtained
through simulation. The results suggest that the values of s
and S obtained by simulation are always better than those
from analytical models in terms of costs and service level. AUTHOR BIOGRAPHIES

Nyoman Pujawan is professor of supply chain engineering


REFERENCES at the Department of Industrial Engineering, Sepuluh
Nopember Institute of Technology (ITS), Surabaya -
Archibald, B. C., dan Silver, E. A. (1978). (s, S) Indonesia. He received his PhD from Lancaster University,
policies under continuous review and discrete compound UK. He has published papers in such journals as European
Poisson demand. Management Science 24, pp. 899 – 904. Journal of Operational Research, International Journal of
Production Economics, Production Planning and Control,
Babai, M. Z., Syntetos, A. A., dan Teunter, R. H. Business Process Management Journal, and many others.
(2010). On the empirical performance of (T, s, S) heuristics. He is the Editor-in-Chief of Operations and Supply Chain
European Journal of Operational Research 202, pp. 466 – Management: An International Journal. He can be reached
472. at< pujawan@ie.its.ac.id>.
Croston, J. D. (1972). Forecasting and stock control
Niniet Indah Arvitrida is a Lecturer at the Department of
for intermittent demand. Operational Research Quarterly Industrial Engineering, Sepuluh Nopember Institute of
23, pp. 289 – 303. Technology, Surabaya - Indonesia. She received her
Huiskonen, J. (2001). Maintenance spare parts bachelor and master degrees, both in Industrial Engineering,
logistics: Special characteristics and strategic choices. from ITS.
International Journal of Production Economics 71, pp. 125
– 133. Bathamas P. Asihanto graduated from the Department of
Industrial Engineering, Sepuluh Nopember Institute of
Nenes, G., Panagiotidou, S., dan Tagaras, G. (2010). Technology, Surabaya - Indonesia in 2010.
Inventory Management of multiple items with irregular
The 11th Asia Pacific Industrial Engineering and Management Systems Conference
The 14th Asia Pacific Regional Meeting of International Foundation for Production Research
Melaka, 7 – 10 December 2010

Table 1 – Simulation results for different parameters s and S for item PN30-802.

INPUT RESULTS
Experiment
Theoretical Actual Holding Order Shortage Total
No. S s
SL SL Cost Cost Cost Cost
1 8 92,50% 6,00 88,50% $138.342 $6.668 $68.376 $213.386
2 9 92,50% 6,00 91,74% $162.522 $6.337 $45.562 $214.421
3 10 92,50% 6,00 93,20% $178.029 $5.368 $37.686 $221.083
4 11 92,50% 6,00 94,47% $193.524 $4.634 $32.450 $230.608
5 12 92,50% 6,00 94,64% $204.378 $3.892 $32.780 $241.050
6 9 95,00% 7,00 91,97% $164.961 $6.668 $43.010 $214.639
7 10 95,00% 7,00 94,50% $190.116 $6.337 $27.346 $223.799
8 11 95,00% 7,00 95,68% $206.061 $5.368 $22.682 $234.111
9 12 95,00% 7,00 96,42% $221.937 $4.634 $20.240 $246.811
10 13 95,00% 7,00 96,02% $232.845 $3.892 $20.966 $257.703
11 10 97,50% 8,00 94,80% $192.624 $6.668 $25.300 $224.592
12 11 97,50% 8,00 96,82% $218.538 $6.337 $15.202 $240.077
13 12 97,50% 8,00 97,58% $234.840 $5.368 $13.156 $253.364
14 13 97,50% 8,00 97,53% $250.938 $4.634 $12.342 $267.914
15 14 97,50% 8,00 97,49% $261.726 $3.892 $12.188 $277.806
16 11 99,00% 9,00 97,07% $221.139 $6.668 $13.838 $241.645
17 12 99,00% 9,00 98,51% $247.659 $6.337 $8.184 $262.180
18 13 99,00% 9,00 98,35% $264.189 $5.368 $7.810 $277.367
19 14 99,00% 9,00 98,49% $280.272 $4.634 $6.886 $291.792
20 15 99,00% 9,00 98,53% $291.048 $3.892 $6.644 $301.584

View publication stats

You might also like