Inventory Management of Returnable Bottles of Brewery

Inventory Management
of Returnable Bottles of Brewery

Master thesis - inventory management Sari Widi 2

Preface
This thesis is done in order to finalise my study of master Econometrics and Management
Science at the Erasmus University in Rotterdam. I have long lived under the impression that
every assignment, especially this thesis, was a fight against time. The months that I have
worked on this thesis have been a true journey. And as always with journeys, occasionally one
might get lost or need some reliable guidance on the way in such a manner that the destination
becomes clear and visible again. There are some people who I would like to thank.
The masters thesis that lies in front of you is the final product of my learning journey. This
thesis would probably never have been possible to complete without help of certain persons.
First of all I would like to express my gratitude to my supervisor, Prof. Dr. Ir. Rommert
Dekker, with whom I had information, knowledge and sources. I would never have been able
to write this thesis without the excellent support of my supervisor. I would like to thank to
Drs. Eelco van Asperen who put the time to share about simulation program, Dr. Erwin van
der Laan for sharing with me his knowledge and some papers, Cerag Pince MSc who put the
time to review my thesis.
Without support and warmth of my husband, my children and my family, this thesis probably
never have been possible to finish. Thank you for keeping me on the straight and narrow. Last
but not least I would like to thank some peoples who have helped me to finish this thesis.

Sari Widi
Augustus, 2009


Table of contents

Chapter

Page

Preface 2
1 Introduction 4

2 Background, Problem Formulation and Literature Review 6
2.1 Company and Process Description 6
2.2 Approach 7
2.3 Literature review 7

3 Data Description and Analysis 13
3.1 Data Description 13
3.2 Data Analysis 15
3.2.1 Descriptive statistics 15
3.2.2 Fitting probability distributions 17
3.2.3 Analysis for every periods and every products 25

4 A Basic Formulation 28
4.1 The Model 28
4.2 The Continuous-review (s,Q) inventory model 29

5 Analytical Reorder Point Calculations 31
5.1 Method 31
5.2 Calculation 33

6 The Simulation Model 37
6.1 Purpose 37
6.2 Set-up of simulation model 37
6.3 Implementation 39
6.4 Output & Analysis 39

7 Summary and Concluding Remarks 44

Appendix 1 : List of Tables 46

Appendix 2 : List of Figures 47

Appendix 3 : The Definition of all characteristics calculated 48

Appendix 4 : List of Outputs of Simulation 50
References 66


Chapter 1
Introduction
An important challenge arising in reverse logistics supply chains is the effective use of
returns so as to maximize the value of this resource. Recondition is the process by which used
products are recovered, processed, and used as new products. Product return has a major
influence on inventory management. Since cleaned returned products enter the inventory of
serviceable items, ordering strategies are affected both with respect to timing and quantity.

In a brewing manufacturing company, they use kegs, bottles and crates that can be
reused after inspection and cleaning. An adequate supply of empty kegs, bottles and crates
must be on hand to satisfy the demand. Part of this supply is a result of the returns of
previously issued ones. The time from issue to return of them however, is variable, without
any prior information about it. This make the inventory control of those items a difficult task.

In order to allow a continuous production with respect to inventory control, the
problems are as follows :
How to determine the number of kegs, bottles and crates needed to allow a continuous
production with respect to inventory control.
Accordingly, the objective in this thesis is to develop a method to assist in inventory
control of returnable items. Next to analyse data and apply the method in a case study.

To this end we model and determine the logistic process of returnable kegs, bottles
and crates of a large brewery manufacturing company in order to attain required service
levels with as low minimize average inventory levels as possible. We give advice on the
number of them needed to allow a continuous production with a certain probability.
Remanufacturing complicates inventory management by introducing return flow of used
products. The random lead time in this stage ( customer-use ) equals the duration of the
products stay with the customer, and the yield is the proportion of products that are
eventually returned. In production planning and inventory management decisions, the yield,
the lead time, and the on-hand inventory associated with a given stage are key pieces of
information.

In this thesis, we start with analyzing data on return cycles coming from a case study.
We apply the distribution fitting to describe the return. Next we use the results in an inventory
control model, and analyse it, both with analytical calculations and a simulation model that
are suggested by the supply chain structure of remanufacturable products. We calculate and
simulate with real data from the company.

Chapter 1 presents introduction. In chapter 2, we discuss the background, problem
formulation and literature review. Chapter 3 presents data description and analysis. We find a
good distributional model for the return lead time. Distribution fitting has obtained by using
SPSS vs. 15.00 program. Chapter 4 presents the models. Chapter 5 presents the analytical
calculation. Chapter 6 deals with the simulation models. The purpose of this chapter is to give
advice on the number of kegs / crates needed to allow a continuous production with a certain
Formatted: Indent: First line:
0 cm
1.25 cm
1.25 cm

probability. The simulation will be obtained by using arena program. Chapter 7 contains a
summary of results and some concluding remarks.


Chapter 2
Background, Problem Formulation and Literature Review
The problem originated from a beer brewery, we called the company with the
fictitious name, Morbeef. Below we briefly introduce and describe the company as well as
the circulation process of returnable beer items. Next we outline our approach and we provide
a literature review.

2.1. Company and Process Description
Morbeef is one of brewer manufacturing company that sell products in containers
(crates) that can be reused. An adequate supply chain of empty crates must be on hand to
satisfy the demand. We model the returnable crates of a large brewer manufacturing company.
Remanufacturing complicates inventory management by introducing return flow of used
products. The random lead time in this stage ( customer-use ) equals the duration of the
products stay with the customer, and the yield is the proportion of products that are
eventually returned.
Morbeef first entered Germany in 2001 with the acquisition of the Diebels Brewery,
followed by Beck & Co in 2002. In 2003, Morbeef acquired the Gilde, Hasseroder and Spaten
breweries. In 2006, Braunschweig, Zwickau and Stuttgart plants were sold. Volume of
product sold in 2006 is 10.2 million hectoliters. The market position of Morbeef Germany is
number 2 in the market, with market share 10.2%.
The beer supply chain in Morbeef is shown Figure 2.2 : (show the stocking points)
Figure 2.1. The beer Suply Chain in Morbeef
shop1 customer shop2
Manufacturing

The final products (beer ) are transported through the procurement Morbeef to various retails /
shops ( in figure 2.1 we called shop1 ). After a beer is purchased, the customer keeps the
bottles / crates for a certain amount of time and then take the crates to various retails / shops
(in figure 2.1 we called shop2 ) to be returned. The retails / shops are responsible for returning
the bottles / crates to the procurement Morbeef. The return time in this process is the duration
of the bottles / crates stay with the customer as well as the transportation time to the stocking
point.
Using that distribution process , there is some interrelated sets of decision typically
required on a daily basis. The decision is as follows :
1.25 cm
1.25 cm

The quantity of bottles or crates should be ordered at which point in time.

2.2. Approach

The case study is related to the model the returnable bottles / crates, we employ data
from the Supply Chain Planning Division of Morbeef in 2006 for Germany market. The
details of our model are rooted in the supply chain for brewer manufacturing company, in this
case is Morbeef. Our approach can be applied in a variety of manufacturing company.

To model the returnable bottles, crates, we construct the queueing network in to
model Morbeefs entire supply chain. A queueing network is a natural model to employ in the
remanufacturing setting because it captures not only the flow of materials through the
traditional stages of procurement, production, distribution, and sales, but also the dependence
of the return flow of used products on past sales, the return delay, and the return probability.
We find a good distributional model for return lead time. Distribution fitting has obtained by
using SPSS vs. 15.00 program and the analytical calculation. Finally we simulate the
probability distribution with the simulation models to give advice on the number of kegs /
crates needed to allow a continuous production with a certain probability. The simulation will
be obtained by using arena program. We compare the result between analytical calculation
with simulation.

2.3. Literature Review

In this thesis we conduct literature review to look at and study some theory and
methodology on inventory control with return flows. We used some Journals publications as
the source. The scope of this literature review are returnable items, inventory control, closed-
loop queueing network and supply chain management. We seek the source using
computerized methods, to identify a set of useful Journals, articles and books . Next we did
reference and citation search. Below we will discuss the most important papers or Journals,
which came out of the search.

In a paper by Dekker and de Brito (2001) literature on inventory control models with returns
can be distinguished into two streams : 1) typical repair models. 2) other models with
imperfect correlation between demand and returns. The review follows in two sections :
deterministic models and stochastic models.
A. Deterministic models.
Schrady (1967) considers a deterministic inventory model in which a certain
percentage of sold products come back, after a known period of time, to be repaired. Repaired
items are put in inventory to be eventually re-used. Since the demand and return processes are
assumed to be continuous deterministic flows, the dependency relationship between the
demand and return process is not explicitly modelled. Later, Richter (1994) and Teunter
(1998) extended this model with the option of product disposal. As regards the demand and
return process, Schradys assumptions remain the same in both extensions.

B. Stochastic models
The stochastic models can be divided into two groups that typically have different
assumptions : periodic review models and continuous review models.
1.25 cm
1.25 cm
1.25 cm
Deleted:

B1. Periodic review models.
This category of models typically focuses on proving the structure of the optimal
policy rather than finding optimal parameter values. Simpson (1978) provides for instance the
optimal policy structure for an inventory model with product return in which product demands
and returns can be stochastically dependent within the same period only. Demand and returns
are known through a joint probability function, which can differ from period to period.
Inderfurth (1997) extends the previous model with non-zero (re)manufacturing and
procurement lead times. All other assumptions equal the ones of Simpson. Inderfurth proves
that there is a simple optimal control policy structure as long as the lead time for
manufacturing and remanufacturing differ at most one period. Buchanan and Abad (1998)
consider a system with partial returns. Each period, a fixed fraction of products is lost while a
stochastic fraction is returned. The authors establish an optimal policy for the case that the
time until return is exponentially distributed.

Toktay et al.(2000) study ordering policies for a business case of single-use Kodaks
cameras. After using the camera, customers bring it to a photo laboratory where the film is
developed. The laboratories next returns the used cameras to Kodak (but sometimes they go to
the so-called jobbers). Kodak dismantles the used cameras and reuses the flash circuit broad
of every camera in the manufacturing of new ones. A closed queueing network model is
applied to decide on periodic ordering decisions. Custom demand is treated as a stationary
Poisson process from which a known percentage is returned. Time until return is modelled by
an infinite Customer and Lab server with general processing time. Another important feature
of this paper is the identification of the informations value according to different scenario.

Kiesmuller and van der Laan (2001) develop a periodic review inventory model where
product returns depend on the demand process. Both the demand and return streams follow a
Poisson distribution. All returns depend on previous demands through a constant time until
return, and two probabilities : the return probability (which underlying event is assumed to be
known upon the demand), and the probability that a returned item is in a sufficiently good
condition to be remanufactured. The authors compare this model with the situation of
independent demands and returns. The outcome supports that it is worth to use information
about the dependency structure between demands and returns.

B2. Continuous review models.
Heyman (1977) analysis different disposal policies for a single-item inventory system
with returns. He uses a model where demands and returns are independent compound renewal
processes and all lead times are zero. An explicit expression for the optimal disposal policy is
given when the processes are Poisson. Muckstadt and Isaac (1981) investigate too the control
of single-item inventory system with independent demands and returns following a Poisson
Process and derive some approximations.

Fleischmann and Dekker (1997) derive an optimal policy and optimal control
parameters for a basic inventory model with returns where demand and return are independent
Poisson processes.
Van der Laan et al.(1999) deal with policies in the context of two inventory facilities,
one of new products and the other of remanufactured items. The model considered is based on
unit demand and unit returns with independent Poisson processes.


Apart from the papers mentioned by Dekker and De Brito (2001), there are some
other papers relevant for this thesis.

Kelle and Silver (1989) describe forecasting methods the returns of reusable containers. The
time from issue to return of an individual container is a random variable with a distribution
that includes a finite probability of never being returned because of lost or damaged. In order
to properly establish the reorder point for purchasing new containers, it is necessary to
forecast the net demand ( demand minus returns ) during the purchasing lead time and the
variability of this net demand. The problem studied in this paper was motivated by
interactions (visits and consultations) with organizations selling products in returnable
containers. These include : liquid gases (in cylinders), beer (in kegs) and non-alcoholic
beverages (in returnable bottles and/or plastic cartons). Although detailed data on demands
and returns were not readily available from these sources, approximate parameter values were
obtained for use in simulation experiments. In this paper different forecasting methods,
dependent upon the available data, are developed to estimate the returns and net demand
during the lead time. A measure of forecast errors and the appropriate reorder point are also
estimated for each of the forecasting methods.
The paper describes four different forecasting methods. Method 1 is the simplest case.
It utilizes only the expected value and the variance of the demand during the lead time and the
the probability of each container eventually being returned. Method 2 uses more detailed
information, namely the actual issues during each previous period and the probability of
return in 1, 2,.n periods for any given container. Method 3 uses the same issue and return
probability data as Method 2, as well as the amount returned up to the present from each
previous issue. This additional information permits us to obtain the appropriate conditional
probabilities of return quantities during the lead time for the remaining outstanding portion of
each previous issue. Method 4 requires, besides the issue and return probability data of
method 2, only the total amounts returned in each of the recent periods without indentification
of when the associated containers were issued (i.e.. only aggregate return data).

The reorder point is calculated as calculated in the form
*
s ED k VD = + ensures an
optimal policy if the lead time net demand ED and its variance VD are estimated correctly. If
these estimates are incorrect, the reorder point will be set too high or too low. If the reorder
point is set too low, the increase in the expected shortage cost is higher than the decrease in
the expected holding cost. These properties are consequences of the choice of the appropriate
safety factor
*
k which minimizes the expected total of the two costs.

In a paper by Kroon and Vrijens (1994) Reverse logistics is an important issue.
Reverse logistics refers to the logistic management skills and activities involved in reducing,
managing and disposing of hazardous or non-hazardous waste from packaging and products.
Reverse logistics may be applied to several stages of the logistic chain. Both the materials
management part and the physical distribution part of the logistic chain are potential areas of
application. The application of reverse logistics in the area of physical distribution : the reuse
of secondary packaging material
Secondary packaging is packaging material used for packaging products during
transport from a sender to a recipient, either in retail or in industry, Stock J.R. (1992).
Traditionally, cardboard boxes are used as secondary packaging material. Since cardboxes can
be used only once, they are defined as one way packaging material. In contrast, returnable
packaging is a type of secondary packaging that can be used more than once in the same form.
Deleted:
Deleted:

Although returnable packaging maybe of different types, such as returnable containers,
irrespective of the actual type of the returnable packaging.

In 1993 the Frauenhofer Institut published the result of an ecological comparison of
one way packaging and returnable containers. On the basis of four criteria, it concluded that
returnable containers are less of a burden to the environment than one way packaging
material, provided each container is used a certain minimum number of times during its
lifetime. This minimum number is dependent on the type of container. The criteria taken into
consideration in this study were energy consumption, emission to the atmosphere, water
consumption and pollution, and solid waste.

The use of systems of returnable containers is being prompted by a growing concern
for the environment and by regulations from the government. For example, in 1991 the Dutch
government and industry signed the Packaging Covenant (1991) forcing industry to think of
new ways to deal with packaging material. In broad terms, the Packaging Covenant requires
that in the year 2000 the total amount of new packaging material in The Netherlands should
be reduced by 10 percent (relative to 1986), and that the total amount of packaging waste to
be dumped in the ecosystem should be reduced to zero.

Similarly, the German Packaging Order requires manufacturers to take responsibility
for their packaging waste. In order to comply with this, German manufacturers and retailers
created the non-profit organization Duals System Deutschland (DSD) to collect packaging
material for recycling. Participating companies pay a per-item free based on the amount of
packaging used and receive in return a green dot (grune Punkt) symbol that appears on their
one-way packaging material. The system is still suffering from a number of growing pains,
which, of course, works to the relative advantage of systems employing returnable containers.

A consequence of the use of returnable containers is that, after a container has been
used for carrying products from a sender to a recipient, the container has to be transported
from the recipient to the next sender, who need not be the same as the first one. In addition to
transporting the containers, the return logistic system also involves the cleaning and
maintenance of containers, as well as their storage and administration.

System with return logistics. In such a system the containers are owned by a central
agency. This agency is also responsible for the return of the containers after they have been
emptied by the recipient. Lutzebauer (1993) differentiates the following systems :
1. Transfer system : The sender always uses the same containers. The transfer system is
only concerned with the return of containers from the recipient to the sender.
2. Depot system : In this system the containers that are not in use are stored at containers
depots. From a container depot the sender is provided with the number of containers
he needs. After having been transported to the recipient, the empty containers are
collected and returned to a container depot.

System without return logistics. In this system the containers are also owned by a
central agency. The user of this system, the sender, rents the containers from the
agency. As soon as the sender no longer needs the containers, they are returned to the
agency. The sender is responsible for all activities involving the containers, such as
return logistics, cleaning, control, maintenance, and storage. By using this system, the

sender can decrease his fixed costs by renting warying numbers of containers as
required.

Whitt (1984) investigated the relationship between open and closed models for
networks of queues. In open models, jobs enter the network from outside, receive service at
one or more nodes, and eventually leave the network.. Thus, with a open model the total
external arrival rate or throughput is an independent variable (specified as part of the model
data), and the number of jobs in the system is a dependent variable (whose equilibrium
distribution is described in the model solution). On the other hand, in a closed models there is
a fixed population of jobs in the network. Hence, with a closed model the number of jobs in
the system is an independent variable (specified as part of the model data), and the throughput
(which may be defined, for example, as the departure rate from some designated node) is a
dependent variable ( to be calculated and described in the model solution). Since the
individual service rate is part of the model data, knowing the throughput is equivalent to
knowing the utilization, which is the expected proportion of the servers at the designated node
that are busy in equilibrium.

It might seem that open models would be more appropriate for most applications
because jobs do usually come from outside, flow through the system, and eventually depart.
However, closed models are often used instead. The representation of flow through the
system, i.e., the throughput, is easily handled in a closed model by assuming that a new job
enters the system to replace an old one whenever the old one has received all of its required
service. This can be represented in the closed model by a transition to a designated exit-entry
node. At this node arriving jobs complete. All of their required service, and departures are
new jobs. The rate of transitions through this node (which is both the arrival rate and the
departure rate) can be regarded as the throughput. If no such exit-entry node exists originally,
it is easy to add such a node. The modified if all jobs at this new exit-entry node have zero
service time.

Closed model are often applied because it seems natural to regard the number of jobs
in the system as the independent variable and the throughput as the dependent variable. The
number of jobs in the system is often subject to control; the queueing analysis is desired to
determine the associated throughputs and response times. For example, in production systems,
new jobs usually do not arrive at random; they are scheduled.

Similarly, in computing systems the total number of jobs in device queues tends to be
limited by resource constraints, so that it is natural to specify the number of jobs (the
multiprogramming level) as a decision variable and then calculate the associated throughput.
Also, in the time-sharing systems, so that the total number of jobs is not unbounded. Hence,
even though closed model are significantly more difficult to analyze because of the
normalization constant or partition function, there are good reasons for applying them.

A queueing network is a natural model to employ in the remanufacturing
setting because it captures not only the flow of materials through the traditional stages of
procurement, production, distribution, and sales, but also the dependence of the return flow
of used products on past sales, the return delay, and the return probability, Toktay et al
(2000).


Concluding, the review literature, in this thesis mainly we use literature by Dekker and
De Brito (2001) and Kelle and Silver (1989), especially when uncertainty is modelled, the
time from issue to return of kegs, bottles and crates is a random variable. In fact, to derive an
optimal policy and optimal control parameters for a basic inventory model with returns where
demand and return are independent Poisson processes.


Chapter 3
Data Description and Analysis

The data originated from a beer brewery, Morbeef. Below we briefly introduce and
describe the data. Next we analyze the data. The goal of this chapter is to find a good
distributional model for the return lead time. Once a good distributional model has been
determined, various percent points for the return lead time will be computed. Moreover, we
will be able to set-up in later chapters a model to calculate the number of bottles needed in the
factory.
3.1. Data Description
A data set ( refer to file Data Germany Cleaned 2006 ) was provided by Morbeef of the
Supply Chain Planning Division in 2006 for Germany market. The data corresponds to how
many days customer bring the bottles back to the shops ( return lead time ). Total data is
23.797 points, in excel format. The data includes 17 beer brands. The data came from 12
February 2006 until 13 Augustus 2006. We calculated data based on delivery time in 4
section.
- Data from 12 February 2006 until 17 February 2006, we assumed winter period.
- Data from 5 May 2006 until 20 may 2006, we assumed spring period.
- Data from 9 July 2006 until 21 July 2006, we assumed summer1 period
- Data from 13 Augustus 2006 until 25 Augustus, we assumed summer2 period.
Also, notice that data set has some shortcoming, for example, there is no information
on what happened at the retailer, no information of kegs lost, only information on the items
returned. In this case we can control the number of kegs lost from the serial number of the
kegs or delivery date number. Beside that the data set has some incorrect number ( maybe
because human error from the input), example Fill Code always have relation with day code
and year code and delivery date, and minimal have 3 digit number, 1 digit for day code, 1
digit for year code and the other digit for the code if they produce many products, so they
need more fill code. In table below we saw incorrect number in fill code, 0016.
Delivery
Date
Sort
Number
Product

Fill
Code
Customer
Code
Day
Code
Year
Code
Fill
Date
15-2-2006 23 Brand A GL 0,33 00166

308 1 6 1-1-2006
16-2-2006 23 Brand A GL 0,33 00166

410 1 6 1-1-2006
10-5-2006 23 Brand A GL 0,33 00166

413 1 6 1-1-2006
14-8-2006 8 Brand A Keg 50 l 0016

180 1 6 1-1-2006
15-8-2006 3 Brand A-AFP 0,33 00166

420 1 6 1-1-2006

15-8-2006 23 Brand A GL 0,33 00168

102 1 6 1-1-2006
15-8-2006 1 Brand A-Pils 0,33 00161

102 1 6 1-1-2006
15-8-2006 20 Brand A Gold 0,33 00161

351 1 6 1-1-2006
The data includes 14 columns. They are :
1. Column 1 : Delivery date, example : 12-JUL-2006
2. Column 2 : Sort number, example : 30, the sort number of Mini keg Home Draught
3. Column 3 : Product, example : Brand Mini keg Home Draught
Beer with type Draught and is served from mini keg. Draught beer is almost always un-
pasteurized and therefore is more fragile. It should be consumed after being "tapped", and is
generally truer to the flavours of the ingredients as pasteurization exposes the beer to heat and
changes the flavour profile. A keg has a concentrically located down tube and a valve that
allows beer in and gas out when filling and vice versa when beer is dispensed. Also kegs have
a simple concave bottom. This aspect of keg design meant that all the beer in the keg was
dispensed which therefore required that the beer be processed by filtration, fining or
centrifuging, or some combination of these, to prevent sediment formation. Lastly, kegs have
straight sides unlike the traditional barrel or cask shape. In order to get the beer out of a keg
and into a customers glass, it can be forced out with gas pressure, although if air or gas at low
pressure is admitted to the top of the keg it can also be dispensed using a traditional hand
pump at the bar. The mini keg is a 5-liter keg produced for retail sales. The example of keg is
showed below :

4. Column 4 : Fill code, example : 0016.
The codes have relation with column 6 ( days code), column 7 ( years code) and column 8
(Fill data). Fill code 0016 means the product is filled at the first day in 2006. The code in
column 6 should be 1 and the code in column 7 should be 6.

5. Column 5 : Customer code, example : 1
6. Column 6 : Day code, example : 1, the first day in that year.
7. Column 7 : Year code, example : 6, code for 2006.
8. Column 8 : Fill date, example : 27-JUN-2006
9. Column 9 : return lead time : 50 days
10. Column 10 : Correction number, example : 510 days. In this case we assumed 510
days is one year.
11. Column 11 : Delivery number, example : 132. The column has relation with
column 1. Delivery number 132 should has delivery date at 13 February 2006, because At 13
February 2006 have delivery number between 1 until 136.
12. Column 12 : return lead time in week, example : 50 days = 7 weeks

3.2 Data Analysis
3.2.1 Descriptive statistics
Graphical Output and Interpretation
The goal of this analysis is to determine a good distributional model for these data.
The first step is to generate a histogram to get an overall feel for the data. The histogram is
shown in figure 3.1. Figure 3.1 has been obtained by using the SPSS vs. 15.00 program.


Figure 3.1. Histogram The Return lead time
return lead time (days)
600,00 490,00 460,00 430,00 400,00 370,00 340,00 310,00 280,00 250,00 220,00 190,00 160,00 130,00 100,00 70,00 40,00 10,00
F
r
e
q
u
e
n
c
y
4.000
3.000
2.000
1.000
0
Histogram the return lead time

The histogram is shown in figure 3.1 is a skewed (non-symmetric) right distribution.
A "skewed right" distribution is one in which the tail is on the right side. For a skewed
distribution, however, there is no "centre" in the usual sense of the word. Be that as it may,
several "typical value" metrics are often used for skewed distributions. The first metric is the
mode of the distribution. Unfortunately, for severely-skewed distributions, the mode may be
at or near the left or right tail of the data and so it seems not to be a good representative of the
centre of the distribution. As a second choice, one could conceptually argue that the mean (the
point on the horizontal axis where the distribution would balance) would serve well as the
typical value. As a third choice, others may argue that the median (that value on the horizontal
axis which has exactly 50% of the data to the left (and also to the right) would serve as a good
typical value.
If the histogram indicates a right-skewed data set, the recommended next steps are to
(Anscombe, Francis (1973) :
1. Quantitatively summarize the data by computing and reporting the sample mean, the
sample median, and the sample mode.
2. Determine the best-fit distribution (skewed-right) from the
o Weibull distribution

o Gamma distribution
o Lognormal distribution
Next we have calculated the sample mean, sample media and the sample mode and some
number of statistics using SPSS vs 15.00. They are given in table 3.1

Tabel 3.1 Statistic Return lead time (all data)

Mean 73.6799 days
Median 50.0000 days
Mode 30.00 days
Std. Deviation 68.13137 days
Variance 4641.883 points
Skewness 2.836
Std. Error of Skewness .016
Kurtosis 11.627
Std. Error of Kurtosis .032
Range 640.00 days
Minimum 10.00 days
Maximum 650.00 days

The definition of all characteristics calculated are in the appendix.

The normal distribution has a skewness of zero. But in reality, data points are not
perfectly symmetric. So, an understanding of the skewness of the dataset indicates whether
deviations from the mean are going to be positive or negative.

A distribution with a significant positive skewness has a long right tail. A distribution
with a significant negative skewness has a long left tail. As a guideline, a skewness value
more than twice its standard error is taken to indicate a departure from symmetry. The
distribution has skewness value 2.836, the value indicates that the distribution has a long right
tail and a departure from symmetry. The coefficient of variation is 0.92470 (standard
deviation divided by the mean) equals almost one.

The distribution has positive kurtosis, 11.627 indicates that the observations cluster
more and have longer tails than those in the normal distribution. The standard error of kurtosis
is 0.32. The ratio of kurtosis to its standard error can be used as a test of normality (that is,
you can reject normality if the ratio is less than -2 or greater than +2).

3.2.2 Fitting probability distributions
Distribution fitting is the procedure of selecting a statistical distribution that best fits to
a data set generated by some random process. In other words, if we have some random data
available, and would like to know what particular distribution can be used to describe our
data.
Why should we use distribution in this thesis ? Random factors affect all areas of our
life, and businesses striving to succeed in today's highly competitive environment need a tool

to deal with risk and uncertainty involved. Using probability distributions is a scientific way
of dealing with uncertainty and making informed business decisions.
Why is it important to select the best fitting distribution? Probability distributions can
be viewed as a tool for dealing with uncertainty: you use distributions to perform specific
calculations, and apply the results to make well-grounded business decisions.
The use of a distribution allow us to describe the data in a compact way, through its
parameter. Moreover, it facilitates experiments by allowing drawing random numbers and it
facilitates drawing general inferences. However, if we use a wrong tool, we will get wrong
results. If we select and apply an inappropriate distribution (the one that doesn't fit to our data
well), our subsequent calculations will be incorrect, and that will certainly result in wrong
decisions.
In many industries, the use of incorrect models can have serious consequences such as
inability to complete tasks or projects in time leading to substantial time and money loss,
wrong engineering design resulting in damage of expensive equipment etc.
Distribution fitting allows us to develop valid models of random processes we deal
with, protecting us from potential time and money loss which can arise due to invalid model
selection, and enabling us to make better business decisions.
The next step is to try to fit various distributions to the data. To this end we will apply
probability plots.

Probability Plot
The probability plot (Chambers 1983) is a graphical technique for assessing whether
or not a data set follows a given distribution such as the normal or Weibull.
The data are plotted against a theoretical distribution in such a way that the points
should form approximately a straight line. Departures from this straight line indicate
departures from the specified distribution.
The correlation coefficient associated with the linear fit to the data in the probability
plot is a measure of the goodness of the fit. Estimates of the location and scale parameters of
the distribution are given by the intercept and slope. Probability plots can be generated for
several competing distributions to see which provides the best fit, and the probability plot
generating the highest correlation coefficient is the best choice since it generates the
straightest probability plot.
For distributions with shape parameters (not counting location and scale parameters),
the shape parameters must be known in order to generate the probability plot. For
distributions with a single shape parameter, the probability plot correlation coefficient (PPCC)
plot provides an excellent method for estimating the shape parameter.
There is a large number of distributions that would be distributional model candidates
for the data. However, we will restrict ourselves to consideration of the following
distributional models because these have proven to be useful in reliability studies.

1. Normal distribution
2. Gamma distribution
3. Weibull distribution
4. Lognormal distribution
There are two basic questions that need to be addressed.
1. Does a given distributional model provide an adequate fit to the data?
2. Of the candidate distributional models, is there one distribution that fits the data better
than the other candidate distributional models?
The use of probability plots provide answers to both of these questions.
If the distribution does not have a shape parameter, we simply generate a probability plot.
1. If we fit a straight line to the points on the probability plot, the intercept and slope of
that line provide estimates of the location and scale parameters, respectively.
2. The criteria for the "best fit" distribution is the one with the most linear probability
plot.
We analyzed the data using the approach described above for the following
distributional models. SPSS determined the parameters for a given distribution with
maximum likelihood.
Using SPSS vs 15.00, we generated a normal probability plot. The result is shown in
Figure 3.2.
Figure 3.2. Normal Probability Plot of Return lead time
Observed Cum Prob
1,0 0,8 0,6 0,4 0,2 0,0
E
x
p
e
c
t
e
d

C
u
m

P
r
o
b
1,0
0,8
0,6
0,4
0,2
0,0

Formatted: English (U.K.)

Using SPSS vs 15.00 we generated a gamma probability plot. The result is shown in
Figure 3.3.

Figure 3.3. Gamma Probability Plot of Return lead time

Observed Cum Prob
1,0 0,8 0,6 0,4 0,2 0,0
E
x
p
e
c
t
e
d

C
u
m

P
r
o
b
1,0
0,8
0,6
0,4
0,2
0,0

Using SPSS vs 15.00, we generated a Weibull probability plot. The result is shown in
Figure 3.4.
Figure 3.4. Weibull Probability Plot of Return lead time
Observed Cum Prob
1,0 0,8 0,6 0,4 0,2 0,0
E
x
p
e
c
t
e
d

C
u
m

P
r
o
b
1,0
0,8
0,6
0,4
0,2
0,0


Using SPSS vs 15.00, we generated a Lognormal probability plot. The result is shown
in Figure 3.5.
Figure 3.5. Lognormal Probability Plot of Return lead time

Observed Cum Prob
1,0 0,8 0,6 0,4 0,2 0,0
E
x
p
e
c
t
e
d

C
u
m

P
r
o
b
1,0
0,8
0,6
0,4
0,2
0,0

Using SPSS vs. 15.00, we determined the estimated distribution parameter. The result
is given in Table 3.2.

Table 3.2. Estimated Distribution Parameters

Distribution Scale Location Shape
Normal 73.6999 days 68.1314 days
Gamma .0160 days 1.1700 days
Weibull 77.9790 days 1.5180 days
Lognormal 54.1040 days .7780 days

1. Normal distribution - from the probability-plot above, the normal probability plot has
= 73.68 days and = 68.13 days.
2. Gamma distribution - the optimal value, in the sense of having the most linear
probability plot, of the shape parameter is 1.1700. At the optimal value of the shape
parameter, the estimate of the scale parameter is 0.0160.
3. Weibull distribution - the optimal value, in the sense of having the most linear
probability plot, of the shape parameter gamma is 1.5180. At the optimal value of the
shape parameter, the estimate of the scale parameter is 77.9790.

4. Lognormal distribution - the optimal value, in the sense of having the most linear
probability plot, of the shape parameter is 0.7780. At the optimal value of the shape
parameter, the estimate of the scale parameter is 54.1040.
We choose the 2-parameter Lognormal distribution as the most appropriate model
because it provides the best fit.
A variable X is lognormally distributed if Y = ln(x) is normally distributed with "ln"
denoting the natural logarithm. The general formula for the probability density function of the
lognormal distribution is

2 2
((ln( ) / )) /(2 ))
( )
( ) 2
x m
e
f x
x
| o
| o t

=
; , 0 x m | o > >
Where o is the shape parameter, | is the location parameter and m is the scale
parameter. The case where| = 0 and m = 1 is called the standard lognormal distribution. The
case where | equals zero is called the 2-parameter lognormal distribution.
The equation for the standard lognormal distribution is
2 2
((ln ) / 2 )
( )
2
x
e
f x
x
t
o t
= 0; 0 x o > >
Since the general form of probability functions can be expressed in terms of the
standard distribution, all subsequent formulas in this section are given for the standard form of
the function.
The next step in this analysis is to make a normality test for Y = ln(x) with normal
probability plot method ( chamber 1983). Using SPSS vs. 15.00, we determined the estimated
normal distribution parameter and generate probability plot. The result is given in table 3.3.
and figure 3.6.
Table 3.3. Estimated Normal Distribution Parameter

lnx
Location 3.9909 Normal Distribution
Scale .77815


Figure 3.6. Normal Probability Plot of ln x
Observed Cum Prob
1.0 0.8 0.6 0.4 0.2 0.0
E
x
p
e
c
t
e
d

C
u
m

P
r
o
b
1.0
0.8
0.6
0.4
0.2
0.0
Normal P-P Plot of lnx

The probability plot test indicate that Y=ln(x) is a straight line to the points on the
probability plot with mean 3.9909 and standard deviation 0.77815.
The next step, we try to analyse with an other test. The One-Sample Kolmogorov-
Smirnov Test (Chakravart, Laha, and Roy, 1967) procedure compares the observed
cumulative distribution function for a variable with a specified theoretical distribution, which
may be normal, uniform, or exponential. The Kolmogorov-Smirnov Z is computed from the
largest difference (in absolute value) between the observed and theoretical cumulative
distribution functions. This goodness-of-fit test tests whether the observations could
reasonably have come from the specified distribution.
The Kolmogorov-Smirnov test assumes that the parameters of the test distribution are
specified in advance. This procedure estimates the parameters from the sample. The sample
mean and sample standard deviation are the parameters for a normal distribution, the sample
minimum and maximum values define the range of the uniform distribution, the sample mean
is the parameter for the Poisson distribution, and the sample mean is the parameter for the
exponential distribution. The power of the test to detect departures from the hypothesized
distribution may be seriously diminished.

The Kolmogorov-Smirnov test is performed using the SPSS 15.00 program. We have
applied it to several distribution assumptions, like normal, uniform, and exponential. The
results are presented in table 3.4., 3.5., and 3.6.

Table 3.4. One-Sample Kolmogorov-Smirnov Test - Normal

lnx
N 23797
Mean 3.9909 Normal
Parameters(a,b) Std. Deviation .77815
Absolute .073
Positive .062
Most Extreme
Differences
Negative -.073
Kolmogorov-Smirnov Z 11.247

Table 3.5. One-Sample Kolmogorov-Smirnov Test - Uniform

lnx
N 23797
Minimum 2.30 Uniform
Parameters(a,b) Maximum 6.48
Absolute .263
Positive .263
Most Extreme
Differences
Negative -.119

Table 3.6. One-Sample Kolmogorov-Smirnov Test - Exponential

Lnx
N 23797
Exponential
parameter.(a,b)
Mean
3.9909
Absolute .481
Positive .223
Most Extreme
Differences
Negative
-.481

The Kolmogorov-Smirov Z value is lowest for the normal distribution (11.247)
compared to the value 40.588 for the uniform and 74.232 for the exponential distribution. The
results of Kolmogorov-Smirov test thus indicates that Y=LN(X) is normally distributed. The
test statistic for the normal distribution is noticeably higher than for exponential or uniform.
This provides additional confirmation that a variable X is lognormally distributed. (Massey, F.
J. Jr. (1951).

The Kolmogorov-Smirov test is a more powerful alternative to chi-square goodness-
of-fit tests when its assumptions are met. Whereas the chi-square test of goodness-of-fit tests
whether in general the observed distribution is not significantly different from the

hypothesized one, the K-S test tests whether this is so even for the most deviant values of the
criterion variable. Thus it is a more stringent test (Massey, F. J. Jr. (1951).

The Z value is the largest absolute difference between the cumulative observed
proportion and the cumulative proportion expected on the basis of the hypothesized
distribution. The computed Z is compared to a table of critical values of Z in the Kolmogorov-
Smirnov One-Sample Test, for a given sample size (cf. Massey, 1951). For samples > 35, the
critical value at the .05 level is approximately 1.36/SQRT(n), where n = sample size. If the
computed Z is less than the critical value, the researcher fails to reject the null hypothesis that
the distribution of the criterion variable is not different from the hypothesized (ex., normal)
distribution. In practice, computer programs like SPSS compute the probability of Z directly
without need to refer to such a table. SPSS prints the two-tailed significance level, testing the
probability that the observed distribution is not significantly deviant from the expected
distribution in either direction.
The next step, we calculated the mean return lead time and standard deviation of
return lead time for every periods.

3.2.3 Analysis for every periods and every products.

We want to know the relation between the mean return lead time with periods and
products. In this section we calculated the mean return lead time and standard deviation of
return lead time for every periods and products using SPSS vs. 15.00.
We calculated data based on delivery time in 4 sections, because the close of the data revealed
that a number of distinct periods can be distinguished. We want to know whether the return
data is the same in all periods.
- Data from 12 February 2006 until 17 February 2006, we assumed winter period.
- Data from 5 May 2006 until 20 may 2006, we assumed spring period.
- Data from 9 July 2006 until 21 July 2006, we assumed summer1 period
- Data from 13 Augustus 2006 until 25 Augustus, we assumed summer 2 period.
Table 3.7 presents the mean return lead time and standard deviation of return lead
time for every periods. Table 3.8 presents the mean return lead time and standard deviation of
return lead time for products.

Table 3.7. The mean return lead time and standard deviation of return lead time for every
period

Periods mean stdev
Winter 87.58 65.7732
Spring 85.03 75.0778
Summer1 70.58 64.6250
Summer2 75.92 66.0697


Table 3.8. The mean return lead time and standard deviation of return lead time for every
product

Product mean Stdev
Brand A CO 0,33 71.72 61.3538
Brand A GL 0,33 76.32 64.5263
Brand A Gold 0,33 75.31 72.1766
Brand A Gold 0,5 71.80 60.6601
Brand A Keg 15 l ITA 95.94 61.8154
Brand A Keg 30 l 78.32 57.0696
Brand A Keg 30 l ITA 92.60 61.0480
Brand A Keg 50 l 73.94 46.2403
Brand A L7 0,33 79.93 64.7511
Brand A-AFP 0,33 97.35 76.9932
Brand A-Pils 0,33 78.70 73.5243
Brand A-Pils 0,5 82.39 75.2185
Brand B LN braun 0,33 81.99 76.1907
Brand B NRW 0,5 93.83 92.3764
Brand B Steinie 0,33 75.96 63.0066
Brand C keg 30L 67.80 45.6866
Minikeg Home Draught 77.15 62.7769

Discussion and conclusions statistical analysis.
We can draw the following conclusions from the results listed above.
The best distribution for return lead time data is log normal distribution.
The average return lead time in July summer1 ( 70.58 ) and August summer2
(75.92 ) are better than in February - winter ( 87.58 ) and May - spring (85.03). They give
some implications :
1. The market is higher in the summer season ( July and August ) than in the winter
season ( February ) and spring season ( May ).
2. The market in the winter season and spring season maybe not so different but in the
winter ( February ) the customers keep longer the bottles in their house because they
shop less frequent.
Table 3.8 showed the average return lead time for products is between 67.80 days
(brand C keg 30L ) until 97.35 days ( brand A-AFP 0.33 ). If we compare with the result for
all products, refer to table 3.1, we got the average return lead time for all products is 73,68
days. Those results give some implications :

1. Brand C keg 30L is the fastest moving product in this case, the average
return lead time for this product is 67.80 days.
2. Brand A-AFP 0.33 is the slowest moving product in this case, the
average return lead time for this product is 97.35 days.
Concluding, the data analysis indicates that the return lead time distribution is not same in
every periods and for all products.


Chapter 4
A Basic Formulation
In the previous chapter we analysed the return lead time data. In this chapter we
introduce the model describing the circulation of the returnable items.
4.1. Descriptive Model
We can model the rotation of crates and kegs as follow.
Figure 4.1 The Model

Keep track of the inventory at Morbeef and of the number in the market (with the customer).
We assume, as is sometimes done in practice, that each demand has a fixed probability P of
an accompanying return of crates / kegs. The expected value ( )
L
E d and variance ( )
L
Var d of
the random lead time demand
L
d .
The expected return lead time, denoted by ER.
The variance of return lead time, denoted by VR .

The net demand is demand minus returns during the replenishment lead time. The expected
lead time net demand denoted by ED (Kelle and Silver, 1989).
In the calculation of the variance of lead time net demand, VD, we have to account for the
correlation between the random
L
d and the random return lead time. At Morbeef, we check
the inventory position and if it is below a level s, we order Q.
4.2. The Continuous-review (s,Q) inventory model
The assumption of this frequently used inventory model are as follows ( Tijms, 1994) :
1. Continuous review of inventory, that is the stock status is continuously monitored
and is updated each time a transaction occurs.
2. The individual demand transactions are small so that the inventory level can be
treated as a continuous variable.
3. A replenishment order of size Q is placed each time the inventory position drops to
the reorder point s.
4. The demands in disjoint time intervals can be treated as independent random
variables.
In practice it is often reasonable to model the lead-time demand by a normal
distribution. If the demand comes from a large number of independent sources, a justification
for use of the normal distribution is provided by the central limit theorem. We assume that
demand is normally distributed, with the following inputs :
D = average demand per period
D
o = standard deviation of demand per period
L = average lead time for replenishment
The ROP represents the available inventory to meet demand during the lead time L. A
stockout occurs if the demand during the lead time is larger than ROP. If demand across
periods is independent, demand during the lead time is normally distributed with the
following :
Mean demand during lead time,
L
D DL =
Standard deviation of demand during lead time,
L D
L o o =
Given the desire CSL, the required safety inventory ( ) ss are
1
( ) ( ) ,
S L L L
ss F CSL x NORMSINV CSL x ROP D ss o o
= = = +
Based on the continuous-review (s,Q) inventory model, Kelle and Silver in 1994
developed an (s,q) inventory system with return.

L
ROP D ss = +
*
s ED k VD = +
s indicates the ROP on the continuous-review (s,Q) inventory model. ED is indicates mean
demand during lead time on the continuous-review (s,Q) inventory model (
L
D ), and
*
k VD
indicates the safety stock, ss on the continuous-review (s,Q) inventory model. We can explain
more detail about that in chapter 5.


Chapter 5
Analytical Reorder Point Calculations

In the chapter 3 we analysed the return lead time data. Refer to the result from chapter
3, we determine in this chapter the reorder level. Based on the continuous-review (s,Q)
inventory model in chapter 4, Kelle and Silver in 1994 developed the (s,q) inventory system
with return. We use a continuous review policy has to account only for the uncertainty of
demand during the lead time. This is because the continuous monitoring of inventory allows
us to adjust the timing of the replenishment order, depending on the demand experienced. If
demand very high, inventory reaches the ROP quickly, leading to a quick replenishment
order. If demand is very slow, inventory drops slowly to the ROP, leading to a delayed
replenishment order. The available safety inventory thus must cover for the uncertainty
demand over this period. The objective in this chapter is to give advice on the number of keg,
crates needed to allow a continuous production with a certain probability.

5.1. Method

Which methods do we apply to determine the reorder level for purchasing new crates
based on average behaviour the return delay ? We will use and adapt the methods developed
by Kelle and Silver (1989). They called the method, Forecast based on average behaviour.
The method is used based on available information. Forecast based on average behaviour
utilizes only :
The expected value and the variance of the demand during the lead time, and
The probability of each kegs / crates being returned.

We assume, as is sometimes done in practice, that each demand has a fixed probability
P of an accompanying return of crates.
Let ( )
L
E d denote the expected value and ( )
L
Var d denote the variance of the random lead
time demand
L
d .
The expected return lead time, denoted by ER, can be expressed as
( )
L
ER PE d = (5.1.1)
The variance of return lead time, denoted by VR . The variance VR of return lead time has the
form :
2
( ) (1 ) ( )
L L
VR P Var d P P E d = + (5.1.2)
The net demand is demand minus returns during the replenishment lead time. The expected
lead time net demand denoted by ED , can be expressed as : (Kelle and Silver. 1989)

( ) (1 ) ( )
L L
ED E D ER P E D = = (5.1.3)
In the calculation of the variance of lead time net demand, VD, we have to account for the
correlation between the random
L
d and the random return lead time. Using the algebraic
expression below : Kelle and Silver (1989)
For a mixed binomial random variable b with random n and known p
2
( ) (1 ) ( ) (1 ) ( ) Var n b p Var n p p E n = +
( )
i
g P n i = =

For
1 1
( ) (1 )
i
k i k
i
k i
i
E nb ig k p p
k
= =
| |
=
|
\ .

2 2
1 1 1
1
(1 ) ( )
i
k i k
i i
i k i
ig k p p i g p pE n
k

= = =
| |
= = =
|
\ .

2
[ ( ) ( )] p Var n E n = +
Thus
2 2
( . ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Cov n b E nb E n E b pVar n pE n pE n pVar n = = + =
Further
2
( ) ( ) ( ) 2 ( . ) (1 ) ( ) (1 ) ( ) Var n b Var n Var b Cov n b p Var n p p E n = + = +
Thus returning to VD, the variance of lead time net demand,

2
(1 ) ( ) (1 ) ( )
L L
VD p Var d P P E d = + (5.1.4)
The reorder point s is expressed in the common way used in the inventory control literature
(e.g..Silver and Peterson (1985. Chapter 7) as :
*
s ED k VD = + (5.1.5)
Where
*
k is the appropriate safety factor (based on service considerations or on
minimizing expected total relevant costs). In practice it is often reasonable to model the lead

time demand by a normal distribution. Assume now that the lead time demand is normally
distributed with mean
L
and standard deviation
L
o . For detail, the safety factor
*
k is
calculated from the simple equation : Tijm (1994)
*
L L
s k o = +
In this case let the target stock out probability is 1% or service level percentage is
99%. We use Excel function NORMSINV to convert service level percentage to service
factor. We can find
*
k is 2.32.

5.2. Calculation

In this section we calculate the reorder level. Silver (1985) and Hadley (1963) use a
normal distribution approximation for demand during lead time in their inventory models. In
their case, where only demands are considered, demand during lead time is a positive random
variable. Since a normally distributed random variable takes on negative values, the expected
demand during lead time has got to be large so that the normal distribution can be reasonable
approximation for a positive random variable. When also returns are considered, net demand (
demands minus returns) during lead time takes on negative as well as positive values, so
approximating net demand during lead time by a normal distribution seems reasonable for all
values of expected net demand during lead time.

Refer to the data analysis in chapter 3, we found that the return lead time is normally
distributed with mean 74 days and variance 68 days. We assume :
- The target stock out probability is 1% or service level percentage is 99%. We
use Excel function NORMSINV to convert for a normal distribution the
service level percentage to a service factor
*
k . We find
*
k 2.32.
- the probability P that a crate is ever returned, 90%.
Using the expression 5.1.1, we can calculate the expected value of the random lead
time demand.
( )
L
ER PE d =
74
( ) 82
0.9
L
ER
E d
P
= = =
Using the expression 5.1.2, we can calculate the variance of the random lead time
demand
L
d .
2 2
( ) (1 ) ( ) 0.9 ( ) 0.9(1 0.9)82 68.1314
( ) 75.0017
L L L
L
VR P Var d P P E d Var d
Var d
= + = + =
=

Using the expression 5.1.3, we can calculate the expected lead time net demand ED
(1 ) ( ) (1 0.9)82 8.2
L
ED P E D = = =
Using the expression 5.1.4, we can calculate the variance of lead time net demand VD.

2 2
(1 ) ( ) (1 ) ( ) (1 0.9) 75.0017 0.9(1 0.9)82 8.1300
L L
VD p Var d P P E d = + = + =
Using the expression 5.1.5, we can calculate the reorder point s .
Reorder point =
*
8.2 2.32 8.1300 14.8 s ED k VD = + = + =
The results calculation for normal distribution are presented in table 5.1
ER : The expected return lead time
( )
L
E d : The expected value of the random lead time demand
L
d
VR : The variance of return lead time
( )
L
Var d : variance of the random lead time demand
L
d
ED : The expected lead time net demand
VD: the variance of lead time net demand

*
k VD : safety stock
s : reorder Point
Table 5.1 The result calculation
ER
( )
L
E d
VR
( )
L
Var d
ED VD
*
k VD s (crates)
74 82 68 75 8 8 6,6 14,8

We want to know the reorder point for every brand. In table 5.2 we calculated the reorder
point for every brand. Refer to the data in table 3.8 ( the mean return lead time and standard
deviation of return lead time for every products ) and using the expressions (5.1.1), (5.1.2),
(5.1.3) and (5.1.4), we calculate for every brand. The results calculation for every brand are
presented in table 5.2

Table 5.2 The results for every Brand

Brands ER
( )
L
E d
VR
( )
L
Var d
ED VD
*
k VD
s
(crates)
Brand A CO 0,33 72 80 61 67 8 8 6,50 14,47
Brand A GL 0,33 76 85 65 70 8 8 6,70 15,18
Brand A Gold 0,33 75 84 72 80 8 8 6,70 15,06
Brand A Gold 0,5 72 80 61 66 8 8 6,50 14,47
Brand A Keg 15 l ITA 96 107 62 64 11 10 7,42 18,08
Brand A Keg 30 l 78 87 57 61 9 8 6,74 15,44
Brand A Keg 30 l ITA 93 103 61 64 10 10 7,30 17,59
Brand A Keg 50 l 74 82 46 48 8 8 6,51 14,73

Brand A L7 0,33 80 89 65 70 9 9 6,84 15,72
Brand A-AFP 0,33 97 108 77 83 11 11 7,54 18,36
Brand A-Pils 0,33 79 87 74 81 9 9 6,84 15,58
Brand A-Pils 0,5 83 93 75 83 9 9 7,02 16,29
Brand B LN braun
0,33 82 91 76 84 9 9 6,97 16,08
Brand B NRW 0,5 94 104 92 102 10 10 7,48 17,91
Brand B Steinie 0,33 76 84 63 68 8 8 6,68 15,12
Brand C keg 30L 68 75 46 48 8 7 6,25 13,78
Minikeg Home
Draught 77 86 63 68 9 8 6,72 15,29

We can get service level as table 5.3 below.
Table 5.3. Service Level for every Brand

s ED VD
*
k
Service
Level
percentage
Brand A CO 0,33 14,47 8 8 2,29 0.9890
Brand A GL 0,33 15,18 8 8 2,54 0.9945
Brand A Gold 0,33 15,06 8 8 2,50 0.9938
Brand A Gold 0,5 14,47 8 8 2,29 0.9890
Brand A Keg 15 l ITA 18,08 11 10 2,24 0.9875
Brand A Keg 30 l 15,44 9 8 2,28 0.9887
Brand A Keg 30 l ITA 17,59 10 10 2,40 0.9918
Brand A Keg 50 l 14,73 8 8 2,38 0.9913
Brand A L7 0,33 15,72 9 9 2,24 0.9875
Brand A-AFP 0,33 18,36 11 11 2,22 0.9868
Brand A-Pils 0,33 15,58 9 9 2,19 0.9857
Brand A-Pils 0,5 16,29 9 9 2,43 0.9925
Brand B LN braun 0,33 16,08 9 9 2,36 0.9909
Brand B NRW 0,5 17,91 10 10 2,50 0.9938
Brand B Steinie 0,33 15,12 8 8 2,52 0.9941
Brand C keg 30L 13,78 8 7 2,18 0.9854
Minikeg Home Draught 15,29 9 8 2,22 0.9868
The calculations are repeated for every period for all brands, assuming a stationary demand
within a period. The results are presented in table 5.4
Table 5.4. The results calculation for every period

Periods ER
( )
L
E d
VR
( )
L
Var d
ED VD
*
k VD s (crates)
Winter 88 97 66 70 10 9 7,14 16,87
Spring 85 94 75 82 9

9 7,08 16,53
Summer1 71 78 65 71 8 8 6,47 14,31
Summer2 76 84 66 72 8 8 6,69 15,13


We can get service level as table 5.5 below.
Table 5.5. Service Level for every period

Periods s ED VD
*
k
Service
Level
percentage
Winter 16,87 10 9 2,29 0,9890
Spring 16,53 9 9 2,51 0,9940
Summer1 14,31 8 8 2,47 0,9932
Summer2 15,13 8 8 2,52 0,9941

In this chapter we have calculated the inventory reorder point for every brand and every
period. Refer to the result in table 5.2, the lowest value inventory reorder point is 13,78 crates
(Brand C keg 30L). The highest value inventory reorder point is 18,36 crates (Brand A-AFP
0,33). Refer to the result in table 5.3, the best service level is 99,45% and the lowest service
level is 98,54%.

Refer to the result in table 5.4, we can draw the following conclusions :
The reorder point in the summer1 period and summer2 period, 14,31 crates and 15,13 crates )
are better than in winter period, 16,87 crates and spring period, 16,53 crates ). Given the
continuous review policies, the purchasing department can order when the inventory drops to
the ROP. The results above give some implication, in the winter season the customer keep
longer the bottles in their house than in the summer season. Refer to the result in table 5.5, the
best service level is 99,41% (summer2) and the lowest service level is 0.98,90% (winter).


Chapter 6

The Simulation Model

Simulated random historical data were used to estimate the statistical measures of
relative performance, Kelton ( 2007 ). We simulated because we want to control the result
from the analytical calculation and assumption.

6.1. Purpose

One of the main objective of this simulation is to understand how we may control the
inventory into the front end and back end of the supply chain in order to minimize average
inventory levels and maintain service levels.
We have calculated necessary safety stock that the return time had a normal
distribution, although we found a lognormal distribution in chapter 3. In this simulation we
simulate the return time with lognormal distribution. We want to give advice on the number
of kegs, crates needed to allow a continuous production with a certain probability. Done at
each type of container (crates, keg).
For the purpose of this thesis, the inputs are the demand distribution, return delay and
loss percentage. The outputs are the required stock level and costs estimate.
6.2. Set-up of simulation model
In this thesis, the simulation is modelled using a very specific kind of simulation
known as discrete event simulation. In this type of simulation, individual entities in the
system are represented as unique work items, each with a appropriate set of attached
identifying characteristics. In discrete event simulation, everything is event driven, and each
event is treated individually. Because events are individualized, it is possible to have
enormous control over the way in which each event and the associated items flow through the
system. This control, in turn, makes it possible to create very accurate models.
Entities represent the objects moving through the system. In this simulation, the
entities are crates. We determine when did each individual crate went into inventory, when
was it made, when was it filled. Demand is according to a Poisson process with rate lambda.
Figure 4.1 shows the processes, which consists of production, return process and new
purchasing. Figure 6.1 contains the process flow diagram. Following the logic of the process
flow diagram, demand orders are generated. When demand orders are generated, they are
matched against the inventory. The level of inventory acts as a control on the rate at which
crates orders are released. If the level inventory exceeds the reorder point, they can send the
crates to the shops. But if the level inventory drops to the level reorder point, the process go to
order crates. The return time is modeled using distributions, because the data is too big to
simulate direct with the real data. Parameters for these distribution were based on historical
data. (Morrice and Valdez, 2005).

The inputs are demand and return time.
Demand : demand generated crates entities into the system based on a Poisson process with a
rate of the expected value of the random lead time demand, 82 crates. The Entities Per Arrival
is based on a Poisson process. The first batch of documents is generated at the level of safety
stock, 6.6 into the simulation run.
Return time : return time generate entities into the system based on an normal distribution
with a mean of the return lead time, 73 days and variance 68 and based on lognormal
distribution.
The Entities Per Arrival is based on a exponential distribution between 1 and 500 days refer to
the historical data, file Germany cleaned 2006 by Morbeef, the range of return days between 1
day until 500 days. The first batch of documents is generated at 1 day into the simulation run.
Control process : in this section they check inventory position. If the inventory position drop
below Reorder Point, then order one unit. Inventory level is number crates in stock. Inventory
position is inventory level plus number of outstanding orders.
Order crates : the order crates processing area is an automatic process while the crates in
storage is below the safety stock level. This is considered to be a value added process and the
time incurred will be added to the entitys Entity. VA time (Value Added) attribute. Value
added time Per Entity : The time each entity spent in any activity of a process, where the
allocation is specified as value added. The delay time is determined by an expression, which
distribution Refer to the historical data.
Shop 1 : the process is an automatic process where no resources are necessary. Considered to
be a value added process, the time incurred will be added to the entitys Entity.VATime
(Value Added) attribute. The delay type is constant.
Customer : the entity enters the Process module to undergo a shop 1. Considered to be a value
added process, the time incurred will be added to the entitys Entity.VATime (Value Added)
attribute. The delay type is normal.
Shop 2 : the entity enters the Process module to undergo a customer. Considered to be a value
added process, the time incurred will be added to the entitys Entity.VATime (Value Added)
attribute. The delay type is normal


Figure 6.1. The Process Flow Diagram of Simulation
demand choose order crates
True
Fal se
shop1 customer shop2 choose dispose 1
True
Fal se
Dispose 1
increase stock
Record 1
order crates
return time
0
0
0
0 0 0
0
0
0
0
0

6.3. Implementation
The next stage of this thesis involves conducting the simulation. The simulation for
this problem is developing by using the Arena vs. 10.00 program. Arena is an advanced
simulation system that provides an interactive environment for building, graphically
animating, verifying and analyzing simulation models. Arena combines the ease of use found
in high-level simulators with the flexibility of simulation language and even all the way down
to general-purpose procedural language ( Kelton, W(2007)). The simulation is developing
using basic process template.
A scenario is defined by a specific set of values for the parameters. Ten simulation
replications were made for each scenario in order to generate confidence intervals. Each
replicate is simulated for 2000 days after 10 days warm-up period.The warm-up period was
chosen by visual inspection using an approach similar to Welchs procedure (Law and Kelton
2000). By experimentation, we determined that a 2000 days simulation replication was
sufficient because statistics had stabilized indicating that we were approximating longrun
steady state results. Customers arrive with inter arrival times distributed as exponential, with
the first arrival occurring not at time zero but after one of these inter arrival times past zero.
The simulation results can be used as test for the analytical calculations. If the results
are almost the same as those from the analytical calculations, it means that the latter are
good.
6.4. Output & Analysis
We first report the main simulation results, and then perform sensitivity analysis with respect
to several key parameters. The running time of the simulation is 2000 days. The data cover a
period of 194 days.
The main simulation results for distributions are presented in table 6.1


Table 6.1 The simulated average order crates (95% confidence intervals) for distributions
Distribution ED VD
Half
Width of
value
added
time

s
Time spent in
the inventory
( days )

*
k

Service
Level
percentage
Normal
distribution

8,1889 8,1201 0,021

15 3.24

2.39

0.9916
Lognormal
distribution 8,1889 8,1201 0,020 16 2.25

2.74

0.9969

*
k is the appropriate safety factor (based on service considerations or on minimizing
expected total relevant costs). *. In this case let the target stock out probability is 1% or
service level percentage is 99%. We use Excel function NORMSINV to convert service level
percentage to service factor. We can find
*
k is 2.32. We determined s from the input data. We
applied an algorithm with start s = 1 and increase until we surpassed the service level. After
we got s we calculated the really
*
k based on the number of reorder level. Next we converted
the service factor to service level.
ED is the expected lead time net demand.
VD is the variance of lead time net demand.
s is the reorder level. The number of reorder level is output from the simulation.
The target stock out probability is 1% or which corresponds to a
*
k = 2.3. We got result
*
k =
2.39 and 2.74.
*
k is calculated from reorder level minus the expected lead time demand and then divided
with square root of the variance of lead time net demand.
95% Confidence intervals :
Value is returned in the Half Width category, this value may be interpreted by saying "in 95%
of repeated trials, the sample mean would be reported as within the interval sample mean
half width". The half width can be reduced by increasing the number of replications.

Refer to the result in chapter 3, using spss v15 program, we have calculated the estimated
distribution parameters. The results are presented in table 6.2.

Table 6.2 The Estimated distribution parameter normal distribution

Winter
( days)
Spring
( days)
Summer1
(days)
Summer2
(days)
Location 87.5800 85.0300 70.5800 75.9200 Normal Distribution
Scale 65.7733 75.0778 64.6251 66.0697


Refer to the result in chapter 3, using spss v15 program, we have been calculated the
estimated distribution parameters. The result is presented in table 6.3.

Table 6.3 The Estimated distribution parameter lognormal distribution

Winter
(days)
Spring
(days)
Summer1
(days)
Summer2
(days)
Scale 71.206 64.628 53.414 58.844 Lognormal Distribution
Shape .634 .720 .716 .687
Refer to data in the table 6.2 ( the estimated distribution parameter normal distribution) we
simulated using arena. The main simulation results for periods are presented in table below.
Table 6.4 The simulated average order crates (95% confidence intervals) for periods with
lognormal distribution
Periods ED VD s
Half
Width
of value
added
time
Time
spent in
the
inventory
(days)
*
k
Service
Level
percentage
Winter 9,7311 9,4619 19 0,034 3.24
3,01 0.9987
Spring 9,4478 9,3249 17 0,064 2.25
2.57 0.9949
Summer1
7,8422
7,7687 15 0,055
3.09 2.53 0.9943
Summer2
8,4356
8,3139 17 0,058
3.31 2.97 0.9985
Refer to data in the table 6.3 ( the estimated distribution parameter lognormal distribution) we
simulated using arena. The main simulation results for periods are presented in table below.
From the results in table 6.4 and table 6.5 we can see clearly why should we simulate for
periods (seasons). We got result
*
k = 2,74 and 2,39 from the simulation of all year. But with
simulation for periods we can see detail in which periods we need more number of reorder
level.
Table 6.5 The simulated average order crates (95% confidence intervals) for periods with
normal distribution
Periods ED VD s
Half Width of
value added
time
Time
spent in
the
inventory
(days)
*
k
Service
Level
percentage
Winter 9,7311 9,4619 18 0,0005 2.96
2.69 0.9964
Spring 9,4478 9,3249 17 0,0005 2.67
2.57 0.9949
Summer1
7,8422
7,7687 14 0,0009
2.21 2.21 0.9864
Summer
8,4356
8,3139 17 0,0005
2.42 2.97 0.9985
In table 6.6 we can compare the safety factor using simulation and analytical calculation.

Table 6.6 The Safety factor
Periods Analytical calculation
Simulation with
lognormal distribution
parameter
Simulation with normal
distribution parameter
February 2.36
3,01 2.69
May 2.47
2.57 2.57
J uly 2.21
2.53 2.21
August 2.28
2.97 2.97
We want to know the effect if the return time constant, instead of a distribution take a
constant value . The result of them are showed in table 6.7 and table 6.8 below. In table 6.7
the result simulation with the return time is constant, assume lead time is 1 month = 30 days.
Table 6.7 Simulation with the return time constant for all periods
Lead time/
return time ED VD s
Half
Width of
value
added
time
Time spent in
the inventory
*
k
Service
Level
Percentage
30 days 8,1889 8,1201 20 0,047 3.29
4.14 0.9999

In table 6.8 the result simulation with the return time is constant, assume lead time is 1 month
= 30 days for months.
Table 6.8 Simulation with the return time constant ( 30 days ) for every period
Periods ED VD s
Half
Width of
value
added
time
Time spent in
the inventory
(days)
*
k
Service
Level
percentage
Winter 9,7311 9,4619 23 0,042 3.29
4.31 0.9999
Spring 9,4478 9,3249 21 0,048
3.29 3.78 0.9999
Summer1 7,8422 7,7687 19 0,047
3.29 3.00 0.9987
Summer2 7,8422 8,3139 21 0,048
3.29 3.36 0.9909
Refer to the conclusion above, with the variation demand distribution, we can adapt the
inventory policy.
L
ROP D ss = +
*
s ED k VD = +

s indicates the ROP on the continuous-review (s,Q) inventory model. ED is indicates mean
demand during lead time on the continuous-review (s,Q) inventory model (
L
D ), and
*
k VD
indicates the safety stock, ss on the continuous-review (s,Q) inventory model.
Safety stock = ss =
*
k VD .
Table 6.9 is showed the safety stock for every periods with the variation demand distribution.
Table 6.9 The safety stock for every periods with the variation demand distribution
Periods ED
*
k

VD

Ss
( crates )
Winter 9,7311
4.31
9,4619
13
Spring 9,4478
3.78
9,3249
12
Summer1 7,8422
3.00
7,7687
8
Summer2 7,8422
3.36
8,3139
10
From the result in table 6.9 above we can see how many crate we need for the safety
stock. If demand high we need safety stock high too. The problem is if safety stock high, the
cost will be high too. But with a continuous review policy we can adjust the timing of the
replenishment order, so we can reduce the safety stock. If demand very high, inventory
reaches the ROP quickly, leading to a quick replenishment order. If demand is very slow,
inventory drops slowly to the ROP, leading to a delayed replenishment order. The available
safety inventory thus must cover for the uncertainty demand over this period.
Refer to the result in table 6.6, we can draw the following conclusions :
The safety factor from the simulation is better than the safety factor from the analytical
calculation. The result from the simulation is more accurate. Analytical results can be highly
precise and in most cases do not take very long to compute. Simulated results often take
longer to calculate and their accuracy depends on the number of simulation iterations
performed. Additionally, simulated results can vary slightly from run to run due to the
randomness of the analysis


Chapter 7

Summary and Concluding Remarks

The objective in this thesis is to develop a method to assist in inventory control of
returnable items. We analysed data and applied the method in a case study. To this end we
model and determine the logistic process of returnable kegs, bottles and crates of a large
brewery manufacturing company in order to attain required service levels with as low
minimize average inventory levels as possible.

In this thesis we used mainly literature from Dekker and De Brito (2001) and Kelle
and Silver (1989), especially when uncertainty is modelled. In fact, to derive an optimal
policy and optimal control parameters for a basic inventory model with returns where demand
and return are independent Poisson processes.
Fitting a probability distribution have been done. The best distribution for the return
lead time for the data given appears to be the log normal distribution.
The analytical calculation model have been done. the lowest value inventory reorder
point is 13,78 crates (Brand C keg 30L). The highest value inventory reorder point is 18,36
crates (Brand A-AFP 0,33. Refer to the result in table 5.3, the best service level is 99,45%
and the lowest service level is 98,54%.

The reorder point in the summer1 period and summer2 period, 14,31 crates and 15,13
crates are better than in winter period, 16,87 crates and spring period, 16,53 crates ). Given
the continuous review policies, the purchasing department can order when the inventory drops
to the ROP. The results above give some implication, in the winter season the customer keep
longer the bottles in their house than in the summer season. Refer to the result in table 5.5, the
best service level is 99,41% ( summer2) and the lowest service level is 98,90% (winter).

To assess the result of analytical calculation, a simulation has been done. Some
simulations have been done, both with the normal distribution and lognormal distribution. The
results of simulation with lognormal distribution parameter yield lower costs than the results
of simulation with normal distribution, which we can see from the appropriate safety factor
*
k (based on service considerations or on minimizing expected total relevant costs). We have
0.35 point less in case of the more with lognormal than normal distribution. Simulation with
the best fitting distribution is important, so we can get better result. Of course simulation with
the real data directly is better, but sometimes data is too much, and so we have problems to
simulate.
The safety stock for every period using the variation of the demand distribution has
been determined. . If demand high ( in winter 9.7 10 ) we need safety stock high too, 13
crates. The problem is if safety stock high, the cost will be high too. But with a continuous

review policy we can adjust the timing of the replenishment order, so we can reduce the safety
stock. If demand very high, inventory reaches the ROP quickly, leading to a quick
replenishment order. If demand is very slow, inventory drops slowly to the ROP, leading to a
delayed replenishment order. The available safety inventory thus must cover for the
uncertainty demand over this period.
The safety factor from the simulation is better than the safety factor from the
analytical calculation. The result from the simulation is more accurate. Analytical results can
be highly precise and in most cases do not take very long to compute. Simulated results often
take longer to calculate and their accuracy depends on the number of simulation iterations
performed. Additionally, simulated results can vary slightly from run to run due to the
randomness of the analysis


Appendix 1

List of Tables

Table 3.1 : Statistic Return lead time (all data)
Table 3.2 : Estimated Distribution Parameters
Table 3.3 : Estimated Normal Distribution Parameter
Table 3.4 : One-Sample Kolmogorov-Smirnov Test Normal
Table 3.5 : One-Sample Kolmogorov-Smirnov Test Uniform
Table 3.6 : One-Sample Kolmogorov-Smirnov Test Exponential
Table 3.7 : The mean return lead time and standard deviation of return lead time for
every period
Table 3.8 : The mean return lead time and standard deviation of return lead time for
every product
Table 5.1 : The result calculation
Table 5.2 : The result calculation for every brand
Table 5.3 : Service Level for every brand
Table 5.4 : The result calculation for every period
Table 5.5 : Service Level for every period
Table 6.1 : The simulated average order crates (95% confidence intervals) for
Distributions
Table 6.2 : The Estimated distribution parameter normal distribution
Table 6.3 : The Estimated distribution parameter Lognormal distribution
Table 6.4 : The simulated average order crates (95% confidence intervals) for periods
with lognormal distribution
Table 6.5 : The simulated average order crates (95% confidence intervals) for periods
with normal distribution
Table 6.6 : The Safety Factor
Table 6.7 : Simulation with the return time constant for all periods
Table 6.8 : Simulation with the return time constant (30 days) for every period
Table 6.9 : The safety stock for every periods with the variation demand distribution


Appendix 2

List of Figures

Figure 2.1 : The Beer Supply Chain in Morbeef
Figure 3.1 : Histogram The Return lead time
Figure 3.2 : Normal Probability Plot of Return lead time
Figure 3.3 : Gamma Probability Plot of Return lead time
Figure 3.4 : Weibull Probability Plot of Return lead time
Figure 3.5 : Lognormal Probability Plot of Return lead time
Figure 3.6 : Normal Probability Plot of ln x
Figure 4.1 : The Model
Figure 6.1 : The Process Flow Diagram of Simulation
Figure 6.2 : The Order Crates Levels


Appendix 3
The Definition of all characteristics calculated.

Skewness
A measure of the asymmetry of distribution. Skewness is written as
1
and defined as :
3
1 3
o
=
Where
3
is the third moment about the mean and o is the standard deviation. Equivalently,
skewness can be defined as the ratio of the third cumulant
3
k and the third power of the
square root of the second cumulant
2
k :
3
1 3/ 2
2
k
k
=
This is analogous to the definition of kurtosis, which is expressed as the fourth cumulant
divided by the fourth power of the square root of the second cumulant.

For a sample of n values the sample skewness is

'
3
3 1
1 3/ 2
3/ 2
2
1
( )
( ( )
n
i
i
n
i
i
n x x
m
g
m
x x
=
=
= =

Where
i
x is the
th
i value, x is the sample mean,
3
m is the sample third central moment, and
2
m is the sample variance.
Given samples from a population, the equation for the sample skewness
1
g above is a biased
estimator of the population skewness. The usual estimator of skewness is
3
1 1 3/ 2
2
( 1)
2
n n k
G g
k n
= =
Where
3
k is the unique symmetric unbiased estimator of the third cumulant and
2
k is the
symmetric unbiased estimator of the second cumulant. Unfortunately
1
G is, nevertheless,
generally biased. Its expected value can even have the opposite sign from the true skewness.

Skewness has benefits in many areas. Many simplistic models assume normal distribution i.e.
data is symmetric about the mean. The normal distribution has a skewness of zero. But in
reality, data points are not perfectly symmetric. So, an understanding of the skewness of the
dataset indicates whether deviations from the mean are going to be positive or negative.


A distribution with a significant positive skewness has a long right tail. A distribution with a
significant negative skewness has a long left tail. As a guideline, a skewness value more than
twice its standard error is taken to indicate a departure from symmetry. The distribution has
skewness value 2.836, the value indicates that the distribution has a long right tail and a
departure from symmetry.
The coefficient of variation is 0.92470 (standard deviation divided by the mean) equals
almost one. There is quite a variability in residence time.

Std. Error of skewness
The ratio of skewness to its standard error can be used as a test of normality (that is, you can
reject normality if the ratio is less than -2 or greater than +2). A large positive value for
skewness indicates a long right tail; an extreme negative value indicates a long left tail.

Kurtosis
A measure of the extent to which observations cluster around a central point. For a normal
distribution, the value of the kurtosis statistic is zero. Positive kurtosis indicates that the
observations cluster more and have longer tails than those in the normal distribution, and
negative kurtosis indicates that the observations cluster less and have shorter tails.

Std. Error of Kurtosis
The ratio of kurtosis to its standard error can be used as a test of normality (that is, you can
reject normality if the ratio is less than -2 or greater than +2). A large positive value for
kurtosis indicates that the tails of the distribution are longer than those of a normal
distribution; a negative value for kurtosis indicates shorter tails (becoming like those of a box-
shaped uniform distribution).

Range
The difference between the largest and smallest values of a numeric variable, the maximum
minus the minimum.


Appendix 4

List of Outputs of Simulation
Process Detail Summary normal distribution- all data
Process Detail Summary lognormal distribution- all data
Process Detail Summary season February-normal distribution
Process Detail Summary season May-normal distribution
Process Detail Summary season July-normal distribution
Process Detail Summary season August-normal distribution
Process Detail Summary season February-lognormal distribution
Process Detail Summary season May-lognormal distribution
Process Detail Summary season July-lognormal distribution
Process Detail Summary season August-lognormal distribution
Process Detail Summary with return time constant for 1 year
Process Detail Summary with return time constant for February
Process Detail Summary with return time constant for May
Process Detail Summary with return time constant for July
Process Detail Summary with return time constant for August


Normal Distribution for all periods
Process Detail Summary

Time per Entity

Total Time VA Time
crates 3,24 3,24

Crates

ED

8,1889

Number In Number Out
crates 15.427,00 15.378,00
Model Filename: C:\Documents and Settings\Administrator\Mijn
documenten\inbev\Modelnormaldistribution
Page 1 of 2

crates

Time per Entity

Average

Half
Width
Minimum

Maximum

Total Time Per
Entity
3,7498 0,021180488 0 14,0527

VA Time Per
Entity
3,7498 0,021180488 0 14,0527

Accumulated Time Value

Accum VA Time 125.408,23

Total Accum
Time
125.408,23
documenten\inbev\Modelnormaldistribution
Page 2 of 2


LogNormal Distribution for all periods

Time per Entity

Total Time VA Time
crates 2,25 2,25

Crates

ED

8,1889

crates 15.952,00 15.915,00
documenten\inbev\Modellognormaldistribution
Page 1 of 2

crates

Time per Entity

Average

Half
Width
Minimum

Maximum

Total Time Per
Entity
2,6239 0,020018721 0 15,0520

VA Time Per
Entity
2,6239 0,020018721 0 15,0520



Total Accum
Time
121.312,17
documenten\inbev\Modellognormaldistribution
Page 2 of 2


February normal Distribution

Time per Entity

Total Time VA Time
crates 3.75 3.75

Crates

ED

9,7311

crates 33.514,00 33.444,00
documenten\inbev\Modelseasonfeb
Page 1 of 2

crates

Time per Entity

Average

Half
Width
Minimum

Maximum

Total Time Per
Entity
3.7474 0,046489019 0 14.0527

VA Time Per
Entity
3.7517 0,034770794 0 13.9059



Total Accum
Time
125.408,23
documenten\inbev\Modelseasonfeb
Page 2 of 2


May normal Distribution

Time per Entity

Total Time VA Time
crates 3.76 3.76

Crates

ED

9,4478

crates 19.829,00 19.764,00
documenten\inbev\Modelseasonmay
Page 1 of 2

crates

Time per Entity

Average

Half
Width
Minimum

Maximum

Total Time Per
Entity
3.7596 0,060073662 0 14.9718

VA Time Per
Entity
3.7517 0,064790232 0 14.6754



Total Accum
Time
74.232,33
documenten\inbev\Modelseasonmay
Page 2 of 2


July normal Distribution

Time per Entity

Total Time VA Time
crates 3.09 3.09

Crates

ED

9,4478

crates 15.269,00 15.223,00
documenten\inbev\Modelseasonjuly
Page 1 of 2

crates

Time per Entity

Average

Half
Width
Minimum

Maximum

Total Time Per
Entity
3.0828 0,056869839 0 12.6704

VA Time Per
Entity
3.1025 0,055585836 0 12.3854



Total Accum
Time
47.090,54
documenten\inbev\Modelseasonjuly
Page 2 of 2


August normal Distribution

Time per Entity

Total Time VA Time
crates 3.31 3.31

Crates

ED

9,4478

crates 16.625,00 16.568,00
documenten\inbev\Modelseasonaugust
Page 1 of 2

crates

Time per Entity

Average

Half
Width
Minimum

Maximum

Total Time Per
Entity
3.3048 0,072335489 0 12.9227

VA Time Per
Entity
3.2845 0,058017551 0 13.1835



Total Accum
Time
40.806,40
documenten\inbev\Modelseasonaugust
Page 2 of 2


February lognormal Distribution

Time per Entity

Total Time VA Time
crates 2.96 2.96

Crates

ED

9,7311

crates 18.996,00 18.943,00
documenten\inbev\Modelseasonfeblognormal
Page 1 of 2

crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
2.9584 0,000513810 0 3.0511

VA Time Per
Entity
2.9587 0,000517906 0 3.0486



Total Accum
Time
56.044,25
documenten\inbev\Modelseasonfeblognormal
Page 2 of 2


May lognormal Distribution

Time per Entity

Total Time VA Time
crates 2.67 2.67

Crates

ED

9,4478

crates 18.996,00 18.943,00
documenten\inbev\Modelseasonmaylognormal
Page 1 of 2

crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
2.6668 0,000813501 2.5649 2.7670

VA Time Per
Entity
2.6671 0,000537651 2.5477 2.7753



Total Accum
Time
47.935,72
documenten\inbev\Modelseasonmaylognormal
Page 2 of 2


July lognormal Distribution

Time per Entity

Total Time VA Time
crates 2.21 2.21

Crates

ED

7,8422

crates 18.996,00 18.943,00
documenten\inbev\Modelseasonjulylognormal
Page 1 of 2

crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
2.2082 0,001020531 2.1165 2.3090

VA Time Per
Entity
2.2081 0,000924607 2.1069 2.3030



Total Accum
Time
18.499,77
documenten\inbev\Modelseasonjulylognormal
Page 2 of 2


August lognormal Distribution

Time per Entity

Total Time VA Time
crates 2.42 2.42

Crates

ED

8,4356

crates 12.878,00 12.873,00
documenten\inbev\Modelseasonaugustlognormal
Page 1 of 2

Crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
2.4164 0,000813765 2.3146 2.5097

VA Time Per
Entity
2.4167 0,000518773 2.3346 2.5073



Total Accum
Time
31.022,35
documenten\inbev\Modelseasonaugustlognormal
Page 2 of 2


1 year with return time constant

Time per Entity

Total Time VA Time
crates 3.29 3.29

Crates

ED

8,1889
crates 12.465,00 12.399,00
documenten\inbev\Model1yearreturntimeconstant
Page 1 of 2

Crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
3.2146 0,000813765 0 3.6171

VA Time Per
Entity
3.2146 0,000813765 0 3.6171



Total Accum
Time
42.133,21
documenten\inbev\Model1yearreturntimeconstant
Page 2 of 2


February with return time constant

Time per Entity

Total Time VA Time
crates 3.29 3.29

Crates

ED

9,7311
crates 12.477,00 12.349,00
documenten\inbev\Modelfebreturntimeconstant
Page 1 of 2

Crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
3.0013 0,0429823 0 3.5201

VA Time Per
Entity
3.0013 0,0429823 0 3.5201



Total Accum
Time
42.101,13
documenten\inbev\Modelfebreturntimeconstant
Page 2 of 2


May with return time constant

Time per Entity

Total Time VA Time
crates 3.29 3.29

Crates

ED

9,4478
crates 12.451,00 12.390,00
documenten\inbev\Modelmayreturntimeconstant
Page 1 of 2

Crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
3.3251 0,0487210 0 3.7641

VA Time Per
Entity
3.3251 0,0487210 0 3.7641



Total Accum
Time
42.357,10
documenten\inbev\Modelmayreturntimeconstant
Page 2 of 2


July with return time constant

Time per Entity

Total Time VA Time
crates 3.29 3.29

Crates

ED

7,8422
crates 12.330,00 12.281,00
documenten\inbev\Modeljulyreturntimeconstant
Page 1 of 2

Crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
3.6741 0,0478612 0 3.7780

VA Time Per
Entity
3.6741 0,0478612 0 3.7780



Total Accum
Time
42.167,31
documenten\inbev\Modeljulyreturntimeconstant
Page 2 of 2


August with return time constant

Time per Entity

Total Time VA Time
Crates 3.29 3.29

Crates

ED

7,8422
crates 12.430,00 12.371,00
documenten\inbev\Modelaugustreturntimeconstant
Page 1 of 2

Crates

Time per Entity

Average

Half
Width

Minimum

Maximum

Total Time Per
Entity
3.8971 0,0488321 0 3.9981

VA Time Per
Entity
3.8971 0,0488321 0 3.9981



Total Accum
Time
42.399,31
documenten\inbev\Modelaugustreturntimeconstant
Page 2 of 2


References
1. A. Gungor and S. M. Gupta, (1999), Issues in Environmentally Conscious
Manufacturing and Product Recovery: A Survey, Computers and Industrial
Engineering, vol. 36, no. 4, pp. 811-853.
2. Aitchison, J. and Brown, J.A.C. (1957), The Lognormal Distribution. Cambridge
University Press, Cambridge. http://en.wikipedia.org/wiki/Lognormal_distribution.
3. Anscombe, Francis (1973), Graphs in Statistical Analysis, The American Statistician.
4. Bose S.J. (2002), An Introduction to Queueing Systems, Kluwer/Plenum Publishers.
5. Buchanan D.J. and P.L. Abad (1988). Optimal Policy for a Periodic Review
Returnable Inventory System. HE Transactions 30:1049-1055.
6. Chakravarti, Laha, and Roy, 1967. Handbook of Methods of Applied Statistics,
Volume I.
7. Chambers, John, William Cleveland, Beat Kleiner, and Paul Tukey, (1983), Graphical
Methods for Data Analysis, Wadsworth.
8. Dekker R and De Brito M .2001. Modelling Product Returns in Inventory Control
Exploring the Validity of general Assumtions. Erasmus University, Rotterdam, The
Netherlands.
9. Dekker R and De Brito M .2001. Modelling Product Returns in Inventory Control
Exploring the Validity of general Assumptions. Erasmus University, Rotterdam, The
Netherlands.
10. Fleischmann M, Dekker R. 1997. Controlling Inventories with Stochastic Returns : a
Basic Models, Management Reports 43(13), Erasmus University Rotterdam, The
Netherlands.
11. Goh, T.N. and N.Varaprasad. 1986. A Statistical Methodology for the Analysis of the
life cycle of reusable Containers. HE Transactions. 18(1) 42-47, New Jersey.
12. Hadley, G., and T.M. Whitin. 1963. Analysis of Inventory System. Prentice-Hall.
13. Hauge, J and Paige, K. 2002. Learning SIMUL8 : The complete Guide, PlainVu
Publising, ISBN 09709384-1-1.
14. Heyman D.P. (1977). Optimal disposal policies for a single-item inventory system
with returns. Naval Research Logistics Quarterly 24:385-405
15. Massey, F. J. Jr. 1951. The Kolmogorov-Smirnov test of goodness of fit. Journal of
the American Statistical Association, Vol. 46. The table of critical values of D is found
on p. 70. n Wiley and Sons, pp. 392-394.
16. Kelle, P. and E.A.Silver. 1989. Forecasting the Returns of Reusable Containers.
Journal of Operations Management. 8. 17-35.
17. Kelton, W. David .2007. Simulation with Arena. Fourth Edition. McGraw-Hill
International Edition.
18. Kiesmuller G.P. and E. van der Laan (2001). An inventory model with dependent
product demands and returns, International Journal of Production Economics
72(1):73-78
19. Kroon, L and Vrijens G. 1994. Returnable Containers : an example of reverse
logistics, Erasmus University, Rotterdam, The Netherlands. International Journal of
Physical Distribution and Logistics Management. Vol 25 No. 2. pp. 56 58.
20. Lutzebauer, M.1993. System for Returnable Transport Packaging, Deutscher
Fachverlag Gmbh, Frankfrut .Morrice D, Valdez R. 2005 . Discrete Event Simulation
in Supply Chain Planning and Inventory Control at Freescale Semiconductor. The
university at Texas. US.

21. Muckstadt J.A. and M.H. Isaac (1981). An analysis of single item inventory systems
with returns, Naval Research Logistics Quarterly, 28:237-254
22. NIST/SEMATECH e-Handbook of Statistical Methods, http : // www.itl.nist.gov /
div898/handbook/.
23. 1991, Packaging Covenant (in Dutch: Verpakkings Convenant), Ministerie voor
Volkshuisvesting, Ruimtelijke Ordening en Milieu (VROM), Den Haag.
24. Schrady D.A. (1967). A Deterministic Inventory Model for Repairable Items. Naval
Research Logistics Quarterly 14:391-398
25. Silver, E.A. and R. Peterson. 1985. Decision System for Inventory Management and
Production Planning. John Wiley & Sons, 3
rd
edition.
26. Stock, J.R., Reverse Logistics. Council of Logistics Management. Oak Brook,
IL.1992.
27. Tijms H.C. 1994. Stochastic models, an algorithmic approach. John Wiley & Sons,
Chichester.
28. Toktay L.B., Wein L.M., Zenios S.A. 2000. Inventory Management of
Remanufacturable Products, Management Science 46(11):1412-1426.
29. Van der Laan E.A., M. Salomon, R. Dekker and L.N. van Wassenhove (1999).
Inventory control in hybrid systems with remanufacturing, Management Science
45(5):733-747
30. Whitt W, 1984, Open and Closed Models for Networks of Queues. AT&T Bell
Laboratories Tech. 63, 1911-1979

Inventory Management of Returnable Bottles of Brewery

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Inventory Management of Returnable Bottles of Brewery

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Inventory Management of Returnable Bottles of Brewery

Uploaded by

Copyright:

Available Formats

Inventory Management

of Returnable Bottles of Brewery

You might also like