Improving performance

of the Star Bottle

production line
A case study at Heineken

Author: Daan van Leer

Date: December 2, 2014

i
“The future belongs to those who prepare for it today.”
- Malcolm X

ii
HEINEKEN NEDERLAND B.V. University of Twente
Burgemeester Smeetsweg 1 Drienerlolaan 5
2382 PH Zoeterwoude 7522 NB Enschede
The Netherlands The Netherlands
+31 (0) 71 545 6111 +31(0) 53 489 9111
www.heineken.nl www.utwente.nl

Master Thesis Project

Study Industrial Engineering and Management
Section Production & Logistics Management
Date December 2, 2014

Daan van Leer

Student number s0217026
Email address daanvanleer@gmail.com

Graduation Committee
University of Twente
First supervisor Dr. P.C. Schuur

University of Twente
Second supervisor Dr. ir. M.R.K. Mes

HEINEKEN NEDERLAND B.V.

Supervisor ir. J. Bron

iii
Management Summary
HEINEKEN needs to stay ahead on the competitive beer market and therefore it constantly needs to
improve its performance. This report focuses on the production line of the Star Bottles, introduced at
HEINEKEN in 2013. Production line 11 produces Star Bottles and differs from other production lines at
HEINEKEN because it has multiple speeds on the filling machine. This line is known as a self regulated
production line, where speed levels of the machines are regulated by sensors on the production line.
Overall, the performance of line 11 is below target, so improvement is necessary. This leads to the
following research question:

How to improve line performance on the regulated production line (line 11) at HEINEKEN
Zoeterwoude?

First, to determine the focus of this research we performed a process analysis and data analysis. The focus
is on the pasteurizer and the labelers (CPLs). The pasteurizer is the bottleneck machine and most
inefficiencies occur on this part of the production line. These problems are formulated and displayed
below:

- Problem 1: Blockage on pasteurizer due to inefficient positioning of sensors. A blockage on the

pasteurizer occurs when the CPL111 (labeling machine) fails and CPL112 does not start
production, simply because sensors are not triggered when they should. A minute loss on the
bottleneck machine is a minute loss on the output of the total line. This problem is also known as
‘inefficient line regulation’.
- Problem 2: Unequal production balance between the labelers. CPL111 produces 57% of the total
output and CPL112 produces 43%. This is a problem because the maintenance schedule does not
fit and extra CILT-activities (Cleaning, Inspection, Lubrication and Tightening) by operators
need to be performed.
- Problem 3: Labeler 112 has an extremely high starvation percentage (64% of total time).

Problem 2:
Unequal production balance
to CPLs (labelers)

CPL111
Problem 1: 10
Inefficient line
CPL112
regulation

Problem 3:
High starvation
percentage on
CPL112

Based on the process and data analysis it is known that problem 1 is caused by inefficient line regulation.
The second problem is a lay out problem, but can also be solved with a more efficient line regulation.
Problem 3 is a mistake in the software. This problem is communicated to the software department and

iv
will be solved in January 2015. In order to improve the line regulation, a conceptual model is designed
that is the base for a simulation model. We use this simulation model to perform experiments of
alternative solutions to problem 1 and 2. In these experiments, the position of the sensors and the number
of speed levels of the machines are changed.

To find an alternative solution, we performed twelve different experiments. The results of the experiments
were ranged by two indicators, output quantity and production balance. Looking at both indicators, the
results of the experiments show that the current situation can be improved. In the table below, the current
situation is compared with the alternative solution.

Situation Output Production Difference

In bottles balance on CPLs
Average CPL111 CPL112
Current 441,313 57% 43% 14%
Alternative 453,103 53% 47% 6%
Difference 11,790 4% 4% 8%

The new situation increased the output with 11790 bottles and improved the production balance with 8%.
This improvement results in a yearly saving of €X (confidential).
We can conclude that the line performance at the regulated production line (line 11) at HEINEKEN
Zoeterwoude can be improved by:

 Adjusting the position of sensors and the amount of speed levels. These adjustments have a
positive influence on the output and production balance of the line. Thus, ‘efficient line
regulation’ improves the line performance.
 Reducing the amount of speed levels of CPL111 from four to three, where CPL112 remains the
same with three speed levels.
 Changing all three sensor positions regarding the speed levels of CPL111, and by changing only
one position of CPL112.
The alternative situation is implemented at the Star Bottle production line. The first results in real life
support our findings and conclusions from our simulation model. The first results show that the
production balance is improved (52%/48%) and that the throughput is increased. Nevertheless, we should
analyze our modification in real life more frequently to ensure that our modification is an improvement in
any situation on the Star Bottle production line.

In addition, we provide the following recommendations:

 Focus more on conveyors/lines. On all packaging lines the focus is on the machines, but the focus
should be on conveyors between the machines. The implementation of the modification is
relatively small, but the results are relatively large.
 Determine the functioning of all sensors on the production line. In order to improve the efficiency
between machines, it is necessary to have a clear understanding of the function of the sensors.
 Improve data registration at MES (information system). The data registration at MES should be
improved, especially for the Star Bottle production line. Due to the regulated production line,
MES is not capable to measure all parameters (e.g., speed levels).
 Hire an extra PA-/PI engineer. When inefficiencies are noted by employees, they have to fill in a
label. Different aspects on these labels are possible, and might vary from safety issues till

v
 machines issues. When such an aspect consists of technical issues, these arrive on the desk of a
PA-/PI engineer. Some filled in labels are on stack for six months. This slow response by
management discourages the operators to help improve the line performance.
 Improving the administration of inventory management of small objects. The exchange of small
objects (e.g., Teflon cylinders, glue sprayers) and their location is not registered by the
maintenance department.
 Visualization of inefficiencies for operators. Operators should be aware of all possible states and
errors of the production line.
The optimization of the line regulation of line 11 is now only performed between the pasteurizer and
CPLs. Therefore we suggest further research to improve the whole line regulation of line 11.

vi
Preface
This report presents the research I conducted in order to improve the line performance at the packaging
department of HEINEKEN Nederland B.V. This research is performed in order to graduate at the
University of Twente. Managing this research project was a real challenge and opportunity to develop my
personal and educational skills.

Conducting and finishing this report would have never been possible without the help of others. Therefore
I am using this opportunity to thank everybody directly and indirectly involved in the realization of this
research.

First, I would like to thank Jojanneke Bron for realizing my graduation internship at the packaging
department at the HEINEKEN Zoeterwoude brewery, for all the effort, enthusiasm, and contributing to
make it a fantastic experience. Furthermore I would like to thank Dennis van Strater, Toine van den Berg,
Ed van Dorp, Ernst Hageman and Peter Zandvliet for their effort, support and insights. During my 6
months internship, I believe to have developed myself at a professional and personal level due to their
constructive feedback and guidance.

A second word of thanks goes to Peter Schuur and Martijn Mes, for their valuable feedback and sparring
sessions. Their extensive input was indispensable in defining and executing this research, writing this
report and finishing the (simulation) model.

Finally, special thanks go to my parents, brother, sister and girlfriend for their support and inspiration,
which eventually is the key to my success.

Daan van Leer.

vii
Abbreviations
AGV Automated Guided Vehicle
BDA Break Down Analysis
CILT Clean Inspect Lubricate Tighten
CPL Cold glue Plastic Label (Machine)
CS&L Customer Service and Logistics
EBI Empty Bottle Inspector
FTE Full Time Equivalent
HL Hecto Liter
HNL Heineken Nederland
HNS Heineken Nederland Supply
IS Information System (see MES)
KPI Key Performance Indicator
MER Mean Efficiency Rate
MES Manufacturing Execution System
MRP Material Requirements Planning
MTBA Mean Time Between Assist
MTBF Mean Time Between Failures
MTTR Mean Time To Repair
MU Movable Unit
OPI Operational Performance Indicator
PDCA Plan, Do, Check, Act
PLC Programmable Logic Controller
QFD Quality Function Deployment
SAT Site Acceptance Test
SB Star Bottle
TEI Time Efficiency Improvement
ToC Theory of Constraints
TPM Total Productive Management
TQM Total Quality Management

viii
Table of contents
Management Summary ................................................................................................................................ iv
Preface ........................................................................................................................................................ vii
Abbreviations ............................................................................................................................................. viii
Table of contents .......................................................................................................................................... ix
1. Introduction ........................................................................................................................................... 1
1.1 Introduction HEINEKEN .............................................................................................................. 1
HEINEKEN Zoeterwoude .................................................................................................................... 2
1.2 Context Description ...................................................................................................................... 3
KHS....................................................................................................................................................... 4
1.3 Problem Statement ........................................................................................................................ 4
1.4 Research setup .............................................................................................................................. 5
1.5 Research scope .............................................................................................................................. 6
1.6 Research methods ......................................................................................................................... 6
1.7 Research deliverables.................................................................................................................... 8
2. Process Analysis ................................................................................................................................... 9
2.1 Packaging line 11 .......................................................................................................................... 9
2.1.1 Machinery ................................................................................................................................. 9
2.1.2 Conveyor/buffer strategy and sensors ..................................................................................... 12
2.1.3 Different states of a machine .................................................................................................. 14
MES .................................................................................................................................................... 14
2.2 Line regulation ............................................................................................................................ 15
2.2.1 Speed control .......................................................................................................................... 15
2.2.2 Speed levels ............................................................................................................................ 17
2.2.3 V-graph ................................................................................................................................... 18
2.3 Speed loss.................................................................................................................................... 20
Technology ......................................................................................................................................... 21
2.4 Measuring Line Performance ...................................................................................................... 21
3. Data Analysis ...................................................................................................................................... 24
3.1 Line Performance ........................................................................................................................ 24
Visibility/Transparency....................................................................................................................... 29
Measuring Speed Loss - Excel Tool .................................................................................................. 29
3.2 Problem design – layout.............................................................................................................. 30

ix
 3.3 Summary of data analysis ........................................................................................................... 34
Summary of analysis (Section 3.1) ..................................................................................................... 34
Summary of problem design (Section 3.20) ....................................................................................... 34
4. Literature Review................................................................................................................................ 35
4.1 Continuous improvement strategies ............................................................................................ 35
Lean management ............................................................................................................................... 35
Six Sigma ............................................................................................................................................ 35
Theory of Constraint (ToC) ................................................................................................................ 35
4.2 Total Productive Maintenance (TPM) ........................................................................................ 36
TPM philosophy.................................................................................................................................. 36
TPM pillars ......................................................................................................................................... 36
4.3 Performance measurement .......................................................................................................... 37
Six big losses....................................................................................................................................... 38
Operational Performance Indicator (OPI ............................................................................................ 38
CILT.................................................................................................................................................... 39
4.4 Related Research ......................................................................................................................... 39
Conveyor Theory ................................................................................................................................ 40
Conveyor systems in simulation ......................................................................................................... 40
Choice of method ................................................................................................................................ 41
Simulation type ................................................................................................................................... 41
Conclusion .............................................................................................................................................. 42
5. Solution design.................................................................................................................................... 43
5.1 Conceptual model ....................................................................................................................... 43
Model overview – Movement of Star Bottles ..................................................................................... 43
Model overview – Regulation ............................................................................................................. 46
Components of simulation model ....................................................................................................... 47
Assumptions........................................................................................................................................ 48
Conclusions ......................................................................................................................................... 48
5.2 Simulation model ........................................................................................................................ 49
Description of the simulation model ................................................................................................... 50
5.3 Experimental setup...................................................................................................................... 51
Input data ............................................................................................................................................ 51
Warm-up period .................................................................................................................................. 53

x
 Number of replications........................................................................................................................ 54
5.4 Verification & validation ............................................................................................................ 55
5.5 Experimental design.................................................................................................................... 56
5.6 Conclusion .................................................................................................................................. 59
6. Experimental results............................................................................................................................ 60
6.1 Performance measures ................................................................................................................ 60
6.2 Simulation results........................................................................................................................ 61
Current situation.................................................................................................................................. 61
All experiments ................................................................................................................................... 61
Correlation .......................................................................................................................................... 63
6.3 Risk analysis ............................................................................................................................... 65
6.4 Conclusions ................................................................................................................................. 69
7. Implementation ................................................................................................................................... 72
7.1 Implementation Procedure .......................................................................................................... 72
7.2 First results after implementation – 8hr shift .............................................................................. 72
7.3 Savings ........................................................................................................................................ 73
8. Conclusion and Recommendations ..................................................................................................... 76
8.1 Conclusions ................................................................................................................................. 76
8.2 Recommendations ....................................................................................................................... 77
8.3 Further research .......................................................................................................................... 78
References ................................................................................................................................................... 79
Appendices.............................................................................................................................................. 82

xi
 1. Introduction
In the framework of completing my Master thesis, I performed research at HEINEKEN BV to improve
the production line of the Star Bottle, introduced in 2013. This report describes the master thesis for the
study Industrial Engineering and Management at the University of Twente. In this research, we analyze
the production line with corresponding machines in order to improve the current situation. Section 1.1
contains general information and background information about HEINEKEN, where we describe in
Section 1.2 the context of the research. In Section 1.3 we define the problem statement followed by the
research question and sub-questions in Section1.4. In Section 1.5 we define the research focus followed
by the research methods in Section 1.6. We end this chapter in Section 1.7 with a list of the main
deliverables of this report.

1.1 Introduction HEINEKEN

HEINEKEN, a Dutch brewing company, is established in 1864 by the HEINEKEN family and is world’s
most international brewer. It has 165 breweries and is active in 71 countries in the world. With around
85,000 employees, HEINEKEN manages one of the world’s leading portfolios of beer brands.
HEINEKEN in the Netherlands owns three breweries, one
location for bottling of soft drinks and nine sales regions.
Production takes place in breweries at Zoeterwoude, Den
Bosch and Wijlre. The largest brewery of HEINEKEN is
located at Zoeterwoude, which is also the location of the
research. Figure 1.1 shows the brewery of HEINEKEN
Zoeterwoude. The beer production at Zoeterwoude is 10
million hl in 2013, from which 60% is dedicated to export.
The destination of the exported beer is especially the
Americas and Asia Pacific. The distribution network of
HEINEKEN BV is so efficient that distributing bottles F IGURE 1.1:HEINEKEN ZOETERWOUDE
from Zoeterwoude to America is more profitable than
brewing in America. In Figure 1.2 we show in what way HEINEKEN is present in the world. Where
HEINEKEN OpCo stands for “Operations Company” and means that in this country HEINEKEN has one
or more breweries.

F IGURE 1.2: GLOBAL PRESENCE OF HEINEKEN

1
In total HEINEKEN brews and sells more than 250 brands including international, regional, local and
specialty beers and ciders. Heineken® is the flagship brand and other brands that are part of the portfolio
are Amstel, Desperados, Tiger, Foster’s, Sol, Wieckse Witte but also ciders like Strongbow and Jillz.
HEINEKEN has three brands that are positioned in the top ten of the world’s leading brands, which are
shown in Figure 1.3. Note that Heineken® is
number one with 27.4 million hectoliter.

F IGURE 1.3: W ORLD' S LEADING BRANDS

( IN MILLIONS OF HECTOLITERS)

In addition HEINEKEN BV (all brands

together) ranks second in the top of the global
market share, with a percentage of 9.1%.
ABInBev (Belgium) and SABMiller (South- F IGURE 1.4: INDUSTRY CONSOLIDATION
Africa) are respectively number one and three. In
Figure 1.4 show these numbers. Nevertheless, based on volume HEINEKEN ranks third, after
respectively ABInBev and SABMiller. ABInBev’s portfolio consists of brands as Budweiser, Stella
Artois, Jupiler, Hertog-Jan etcetera. Whereas SABMiller has brands like Miller, Grolsch and Pilsner
Urquell.

Recently SABMiller proposed a takeover offer towards HEINEKEN, but HEINEKEN wanted to preserve
the firm as “an independent company”. Besides there has been speculation within the brewing industry,
for months, that SABMiller has been targeted by the world’s number one brewer ABInBev. This means
that there is a frequent activity around the top of the breweries.

HEINEKEN Zoeterwoude
HEINEKEN Zoeterwoude is divided into two divisions, HEINEKEN Netherlands (HNL) and
HEINEKEN Netherlands Supply (HNS). In this research we only focus on HNS which is shown in the
organizational structure of HNS in Figure 1.6. The chart narrows on the area of interest for this research,
which considers line 11. A rayon consists of (two or) three production lines which differ from bottles to
kegs.

2
 HNS

Secretary

Total
Brewing Technology & Technical Safety Health
Packaging Productive
/Filtration Quality Services Environment
Management

Rayon 1 Rayon 2 Rayon 3 Rayon 4 Rayon 5

Line 9

Line 11

Line 12 F IGURE 1.5:

F IGURE 1.6: ORGANIZATIONAL CHART
HNS HEINEKEN
S TAR BOTTLE
In 2013, HEINEKEN introduced the new Star-Bottle (SB) exhibited in Figure 1.5. In 0.3L
order to produce these new bottles, a new production line has been developed, which is
line 11. So this research focuses on line 11 with the Start Bottles. In Section 2.1 we describe this line in
detail.

1.2 Context Description

In today’s highly competed beer market, HEINEKEN needs to stay ahead of its competitors. More beer
brands will enter the market and as HEINEKEN Nederland Supply (HNS) states: “customer demand is
changing, volume is decreasing, fixed costs as well as variable costs are increasing, and customers expect
the same service and quality” (HNS visie 2015, 2011). Therefore, HEINEKEN is striving for continuous
improvement of their performance in order to stay ahead of the competition. The main goal is to gain
higher line performance, higher productivity, and eventually a lower cost price, while maintaining the
quality. HEINEKEN has decided to perform this continuous improvement using Total Productive
Management (TPM). TPM is an equipment management philosophy, focused on maximizing
performance and the ultimate goal is to reach zero losses (Nakajima 1988). TPM is preferred above TQM
and Six Sigma because of its strong focus on equipment and maintenance. Since the continuous
improvement philosophy, TPM has a strong focus on maintenance and is useful in organizations that have
a high level of equipment automation (Rolfsen, Langeland 2012). TPM is a philosophy to continuously
manage, optimize and improve a supply chain by eliminating all losses, involving all employees of the
organization (Chan, Lau et al. 2005, Ahuja, Khamba 2008). By systematically eliminating losses, TPM
improves the performance of a production system (Nakajima 1988, Hartmann 1992, Chan, Lau et al.
2005).

In order to know what performance is improved, the performance measure should be clear.
Currently, in most businesses, every performance is measured by various kinds of performance indicators
(PIs). Also departments in a company have their own PIs. Consider for example a car manufacturer:
where the sales department measures its performance on cars sold and number of customers satisfied and

3
the production department measures its performance by cars produced and cars rejected by lack of
quality. In literature it is a highly debated topic. According to Neely (2002), the definition of performance
measurement is: “The process of quantifying the performance of actions”. Measuring the performance is
important in order to be able to perform improvement activities based upon these measures and to keep
track of previous results (De Ron, Rooda 2006). In addition, only aspects, that have been measured, are
actively improved by the stakeholders (Ridgway 1956, De Ron, Rooda 2006). Therefore it is important
for businesses to identify the correct performance measurement and corresponding PIs for each process.
With an incorrect performance measure, the problem will not be measured correctly and therefore it is
unclear whether the problem is solved or not.

KHS
KHS, the German supplier of the Star Bottle production line, also matches the thoughts of TPM, where
reducing losses are a point of interest. The philosophy of KHS is to reduce losses by avoiding
start/stop situations of a machine. The losses that will be reduced are equipment failure, idling, minor
stoppage and reduced speed. KHS implemented line 11 in 2013. Thus the machinery is relatively new
and KHS expects less failure compared to other lines. Nevertheless there are some differences with older
production lines due to new insights in technology. KHS has a new philosophy to increase line balance by
introducing several speed levels in the machines, particularly on the filler. Previously, fillers on the
production lines at HEINEKEN have just one production speed, the machine produces or does not (resp.
at 100% or 0% of capacity). This stationary process is known as a non-regulated production line.
Nowadays the machines on line 11 have different speed levels (e.g., 0/25%/75%/100% of capacity),
which is a dynamic process. When a starvation/blockage impends, sensors on the conveyors will send a
message to the machine that it must change to, e.g., 75% of the capacity in order to prevent downtime.
This is in line with the philosophy of KHS, who believes that a continuous flow of products will reduce
failures. This dynamic process is known as a regulated production line. HEINEKEN adopted this
philosophy and line 11 is therefore a regulated production line.

1.3 Problem Statement

According to Nakajima’s (1991) findings mentioned in the previous section, HEINEKEN should reduce
their losses in order to improve their performance. First we have to clarify the definition of ‘losses’.
HEINEKEN uses an information system to get insight into their losses. A tool to determine the machine
states is the “DNA strand”, which is exhibited in Figure 2.8 . For a non-regulated production line this tool
works well. A green bar means that the machine is producing (100% of its capacity). Nevertheless, for a
regulated production line it is more challenging because on the green bar the production speed is not
visible ( e.g., if the machine is producing for 50% or 75% of its capacity). This means that the production
losses of line 11 are not measured completely. It is complex to reduce the losses if they are hardly visible.
There exists a mismatch between the philosophy of KHS and the measurement of losses at HEINEKEN.

It is hard to improve the line performance of line 11 because of the regulation and because losses are not
visible. Nevertheless, line performance will not improve when only the losses are made transparent. An
action or modification has to be made. Making these losses transparent is just a step in the process in
order to improve the line performance. When losses are more transparent, the machine efficiency and
relations could be determined and only then improvements can be made.

4
To summarize this section, we state that the main objective is how to improve the line performance and
how can losses made transparent.

1.4 Research setup

Based on Section 1.3, we formulate the following research question to reach the problem statement:
How to improve line performance at the regulated production line (line 11) at
HEINEKEN Zoeterwoude?
To answer the main research question, we formulate five sub-questions that give a deeper understanding
of the research. Each sub-question contains a brief description of what will be discussed in this sub-
question.

1. How is production line 11 currently organized?

a. What kind of machinery is located?
b. How is the line regulation organized?
c. What KPIs are currently used?
d. How is performance currently measured?
First, the layout of line 11 is explained in order to delineate the problem situation. We need a clear
understanding of how production line 11 is functioning. We need to know what the production processes
are and how the line regulation works. Furthermore, the KPIs need to be clear in order to measure
performance. Also the functioning of the different machines and conveyors are described.

2. What is the current performance of line 11?

a. What losses can be identified?
b. Which processes are bottleneck processes?
The second question focuses on the current performance and on what kind of losses HEINEKEN has to
deal with. A lot of losses are measured in the information system of HEINEKEN, but not every loss is
visible. Therefore all the losses are clarified in order to compare them and to determine the focus for
performance improvement. In addition, we give a clear definition of losses and what kind of performance
measurement must be used. Moreover a tool for determining the speed losses on bottleneck processes is
introduced.

3. What alternative approaches are described in the literature for the improvement of line
performance? What is the best alternative approach to use at HEINEKEN?
We conduct a literature study to increase the understanding of different approaches for improving line
performance. Different methods are compared so the best method is applied to the problem. We search
scientific articles in the field of production line improvement.

4. How can the alternative approach, described in sub-question three, be implemented in order to
improve the line performance?
In sub-question three, we compared different methods in order to improve the line performance. In this
question we propose suitable improvements/interventions for HEINEKEN. To display the impact of these
improvements, we develop a conceptual model. In order to validate and verify the authenticity of this
conceptual model we will use simulation. A simulation model mimics the reality. Different scenarios can

5
be experimented and the best alternative can be compared with the current situation. If this improves the
current situation, then implementation can be considered.

5. What are the results of the identified improvements, and what are the recommendations for
HEINEKEN?
In sub-question five we analyze whether the results from sub-question four will have a positive impact on
the current situation. Based on these results we describe the recommendations for HEINEKEN regarding
implementations as for further research.

1.5 Research scope

We base the research scope on the argumentation of the management of HEINEKEN, which stated that
the focus needs to be set only on the machines that have the biggest influence on the line performance.
The management argued that the area between the bottle washer and packer works inefficient. This
argumentation will be further explained in Chapter 3, using data analysis.

Furthermore, we decided not to include the breakdowns of the machines because these are mainly
technical issues that can be solved by operators or electricians. Moreover it is the wish of HEINEKEN to
improve the line with current breakdowns, because teams with operators/electricians will solve the
technical problems. Besides, mechanical/technical improvements are not in line with the study
background in which this research is conducted.

Also this research has a couple of restrictions that should be taken into account. Large investments
(>10,000 euro) cannot be done. Likewise, the layout of the conveyors at the production line cannot be
changed. Furthermore the quality and safety standards developed by HEINEKEN should be satisfied.

Below we give an enumeration of characteristics that can be changed, which is useful for developing the
conceptual model and simulating this model. The characteristics are:

 Production speed of machines.

 Number of speed levels of machines.
 Moment of switching to a certain machine speed. This depends on the location of the sensor and
the programming code related to the sensor. A more detailed explanation can be found in Section
2.2. Note that the location of the sensor is fixed, because modifications are restricted to the
current layout.
 The combination of sensors that will lead to a certain action (e.g., production of a machine).
Often one sensor is coupled to multiple actions.
These characteristics will be explained in Chapter 0.

1.6 Research methods

To answer the research questions, we will use several research methods. In order to answer the first and
second sub-question, we use the knowledge at HEINEKEN. To describe the current layout and machinery
it is useful to observe the line itself, with the use of empirical data. Furthermore some interviews will be
done with supervisors, experts and operators to gather information about the production line. The nature

6
of the interviews will also depend on the nature of the information that is to be gathered. Experts at
HEINEKEN and from the academic community will be approached accordingly to the required
information. Also the information from information systems and real life data is used to describe how
production line 11 is currently organized. The information system MES, used by HEINEKEN, stores all
relevant line and production data. This will be the main source in order to answer sub-question two.
Besides, we apply a data analysis in order to determine the bottleneck machine/process. Also a tool for
determining the speed losses will be introduced.

The research method for sub-question three, is a literature review. To perform all literature reviews,
academic publications, books, reports, internet, databases and proceedings of conferences will be
analyzed. These sources will also be used to perform desk research.

In sub-question four we use conceptual model building to design the solution. Furthermore, we use
verification and validation methods to assess the data that is used. This validation and verification will be
done, using simulation. It is appropriate to use simulation in order to mimic real life situations. A
simulation model is a simplified model of reality and is used to test out different production rules (Wein
& Chevalier, 1992). We will create several experiments in order to search for a best alternative.

Then in sub-question five the conceptual model will be developed by expertise and literature. In order to
assess the solution, we run several experiments. When the model is finished and potential losses are
identified, some improvements can be recommended. These recommendations will be discussed with
experts at HEINEKEN in order to create reliability during the implementation period.

To gather data and to perform an analysis a multifunctional team, consisting of experts and programmers,
is composed. The diverse knowledge in this team make the analysis more efficient and more valuable.

In Figure 1.7 we show a summary that depicts the research methods by making a difference between
academic literature & knowledge and current practices & knowledge at HEINEKEN .
Sub question 1

Interviews, observations and

participatory research at
HEINEKEN

Sub question 2

Analysis current approach

through desk research.

Sub question 3

Literature review and Current practices

Academical literature academic knowledge of
process improvement and knowledge
and knowledge
at HEINEKEN
Sub question 4

Description of conceptual
model and simulation

Sub question 5

Formulation of improvement
and recommendations

F IGURE 1.7: GRAPHICAL REPRESENTATION OF RESEARCH STRUCTURE

7
1.7 Research deliverables
- Process analysis.
- Data analysis.
- Insight into the operations of the regulated production line.
- Tool to visualize speed losses.
- Simulation model for production line 11, which can be used by the department of ‘maintenance’ and
‘packaging’ (when a product license is purchased).
- A guide that explains the concept behind the simulation model.
- An implementation plan that elaborates on the steps necessary to make the model operational.
- Recommendations for further research.

8
 2. Process Analysis
In this chapter, we provide insight into the processes of packaging line 11. In Section 2.1 we explain the
functioning of the different machines. Thereby we describe the functions of conveyors/buffers. The
conveyors and machines are related with the sensors on the production line. All these separate
components together are part of the line regulation and will be explained in Section 2.2. In Section 2.3 we
narrow down on the definition of speed loss and how these losses occur. Also the impact on the
technology is given. Followed by Section 2.4, where the performance measurement is highlighted. The
measurement of performance is a preface for Chapter 3, where we determine the line performance.

2.1 Packaging line 11

Line 11 have been developed for the HEINEKEN Star Bottle (SB) which was introduced in 2013. Line 11
differs from other lines on several aspects. An aspect is that the SBs consists of returnable bottles, which
means that they are recovered from the domestic market. Other lines (except Amstel, line 12) are one way
bottles, these bottles are disposed by customers after consuming. The functioning of line 11 depends on
the quality of the returned material. In this section the different machines with corresponding conveyors
are explained. Furthermore the functioning of the buffers is described and we take a closer look at the
structure of the production line.

2.1.1 Machinery

Line 11 consists of several machines. A brief description of the function of each machine is given below,
in sequence from start to end. Thus the production process for the SB’s starts at the depalletizer and ends
at the foil taper. In Figure 2.1, all machines are displayed in a schematic overview. In this
figure the green square represent the “wet area”. The wet area consists of machines from the bottle washer
till the CPLs (labeler). The management has argued that the most important machines of this research will
be the filler, the pasteurizer and the CPLs. This will be proved in Chapter 3. Therefore only these
machines are visualized.

Depalletizer:
The depalletizer removes the crates (returned from the domestic market) from the pallets, layer by layer,
and drops it on the conveyor to the depacker.
De-packing machine (Decrater):
The depacker picks the empty bottles out of the crates. The bottles move to the bottle washer and the
crates to the crate washer.
Bottle washer:
The bottle washer cleans the bottles. When the bottles are cleaned they move to the empty bottle
inspection.
Crate washer:
The crate washer cleans the crates.
Empty Bottle Inspector (EBI):
At this stage, the bottles are inspected for being empty. Several pictures are made to ensure the bottle is
clean according to predetermined standards. If a bottle does not meet quality standards it will be removed
from the line. The bottles that pass the EBI will move to the filling machine.

9
 CONFIDENTIAL

F IGURE 2.1: S CHEMATIC O VERVIEW P ACKAGING LINE 11

10
 Fillers (or Filling machines):
The fillers put the beer into the empty bottles and
closes the bottles with a crown. A filler machine is
shown in Figure 2.2.
Full Bottle Inspector (FBI):
The bottles are inspected again and are removed
from the line if quality is not met. If the bottles
passes the inspection they will move to the
pasteurizer.

F IGURE 2.2: F ILLER

Pasteurizer:
In the pasteurizer the bottles are heated to deactivate all
microorganisms and enzymes that can influence the quality of
the beer, and to increase the shelf life. The cycle time of the
pasteurizer is the largest of the whole line, with an average of
43.2 minutes. After the bottles are pasteurized, they will move
to the labelers. The pasteurizer is shown in Figure 2.3.

F IGURE 2.3: P ASTEURIZER

CPLs/Labelers:
The labelers stick three labels (front, back and the neck of the bottle). CPL stands for Cold glue Plastic
Label. Again the bottles are inspected and, if necessary, removed from the line. The quality checks at this
stage are strict, with a single
deviation, the bottle will be
removed. Perfectly labeled
bottles move to the packer. A
CPL machine with a detailed
view on the labels is shown in

Figure 2.4.
F IGURE 2.4:
CPL/L ABELER

Packer (Crater):
The packer puts full and labeled bottles

11
into a clean crate.
Cratecover:
The cratecover will put cardboard sheet on the upper side of the crate, covering the bottles (mostly with
attractive marketing promotion). After this, the crates move to the palletizer.
Sorter:
Before the crates move to the palletizer, this machine spins (some of) the crates to optimize the way there
are stacked on a pallet.
Palletizer:
The palletizer puts the crates on a pallet, layer by layer.
Sticker:
The sticker puts a foil and a label on the pallet. This label will be scanned and linked to an order in the
information system. The system contains specific data of each pallet, such as the date of production and
destination of delivery. When a batch needs to be retrieved from the market for some reason, HEINEKEN
can easily detect the specific batch. At the end the pallet is ready to enter the market.

All these machines are connected with conveyors, what will be explained in Section 2.1.2. In Figure 2.5
we show the layout of line 11 where all the machines are exhibited. At the right side we show the
depalletizer and sticker, these are the first and last machines. Therefore in front of the depalletizer there
are pallets returned from the market, and after the sticker there are standing finished goods. This means
that the department of Customer Service & Logistics (CS&L) with their Automated Guided Vehicles
(AGVs) are located at the right side of this picture. AGVs are the vehicles that transport the pallets to and
from the production line. This department is also responsible for the warehousing of the pallets.

F IGURE 2.5: LAYOUT LINE 11

2.1.2 Conveyor/buffer strategy and sensors

Conveyors are used to transport the SBs from one machine to another. The conveyors have different sizes
in width as well as in length. A conveyor can also be used a as buffer. A buffer is provided in order to
cope with unexpected failures of the installation (machines), which may cause interruptions of the

12
production process (Van der Duyn Schouten, Vanneste 1995). This is in line
with the current situation at HEINEKEN where some conveyors are used as
buffers. The speed (level) of a conveyor is predetermined and programmed into
the IS. Mostly the conveyors have different speed levels in order to comply
with the needs of the next conveyor/machine. The timing of switching speed
levels is dependent of the occupation of the buffer (number of bottles on the
conveyor). This can be measured by the use of sensors. In Figure 2.6 we show
a picture of a regular sensor at line 11. On each conveyor one (or sometimes
more) sensor is (are) located. The sensor is the metal ‘arm’ at the left side of
the picture. These sensors are triggered with the presence of the SBs, the bottles
will push the metal arm towards the left fence. Mostly the sensors are located in F IGURE 2.6: S ENSOR
such a way that bottles will not directly trigger these sensors when arriving at
the buffer. This happens when bottles stagnate and enumerate, due to the fact that machines further in the
line are already stopped producing or when the machine is in failure as shown in Figure 2.7. The green
bar at situation 1, is comparable with the two belts at the right side of Figure 2.6. The white bar, in Figure
2.7, is similar with the left belt in Figure 2.6. Mostly sensors are only triggered (yellow sensors) when the
buffers are full till the corresponding sensor. Thus, when the succeeding machine is not producing, the
bottles before this machine will enumerate, spread out and hit the sensor. When the upper bar in Figure
2.7 is full with bottles (green bar) the bottles are enumerated in front of the machine. There are two
different kinds of sensors present on line 11: switches and photocells. A switch must be triggered
physically with a bottle. A photocell beams a laser to a reflector and is triggered when the beam is
interrupted by a SB.
1. Ideal situation: Machine produces
Pasteurizer S1 S2 S3 CPL
Production Production

2. First sensor is triggered due to machine failure Triggered

Pasteurizer S1 S2 S3
CPL
Production Failure

3. Third sensor is triggered due to machine failure

Pasteurizer S1 S2 S3 CPL
Blockage Failure

Buffer is occupied

F IGURE 2.7: BEHAVIOR OF SBS ON SENSORS , WHEN BLOCKAGE ON PASTEURIZER OCCURS .

Figure 2.7 shows an example of how the sensors are located. In reality there are more sensors and
conveyors placed between the pasteurizer and CPL. Furthermore the figure gives only a situation of
blockage where the buffer is completely filled (situation 3). It is also possible that the buffer is completely

13
empty and that the CPL has a starvation (no sensor is triggered) instead of a blockage as shown in Figure
2.7.

2.1.3 Different states of a machine

Since a machine is not producing all the time, there are several states that indicate the condition of the
machine. A machine can be in different states, which are formulated below:

Producing: The machine is producing products. This could be with different speed levels.
Planned production stop: The machine is not producing due to planned maintenance.
Starvation: The machine is not producing due to a lack at the infeed. Mostly caused by failures of
preceding machines.
Blockage: The machine is not producing due to a backup at discharge. Mostly caused by failures of
succeeding machines.
Short failure: The machine has an internal or external failure with a duration less than 5 minutes.
Long failure: The machine has an internal or external failure with a duration longer than 5 minutes.
Unknown: The cause of the machine downtime is not registered. This state will be neglected in this
research because the unknown time is nil. If this time arises it will be often a downtime due to a test.

A machine is either producing, or not producing for one of these seven reasons. Besides the distinction in
short and long failure, there is a difference between an internal and an external failure. Internal failures
are failures caused by the machine itself. External failures are failures caused from external factors, e.g.
another department or bad quality of material.

In order to measure performance only on those aspects that are relevant for production line 11, some
aspects will be neglected. The aspects that we neglect are the unknown state, the planned production stop
and the external failures. The unknown state does not consist of valuable information and mostly arises
when a test takes place. The planned production stop is necessary and is known beforehand, this aspect
will not influence the performance of the production line. The external failures are not taken into account.
These failures are caused by other departments but do have an influence on the performance of line 11.
External failures could also arise by bad quality of material. Nevertheless, this research focuses on the
machine/conveyor efficiency and not on material quality.

MES
In order to register all the different machine states and create visibility among the machine conditions,
HEINEKEN uses the information system called MES. A print screen of the machine status of a 8-hour
work shift is given in Figure 2.8.

14
F IGURE 2.8: MES DNA STRAND - 8 HOUR WORK SCHEDULE

2.2 Line regulation

At this moment we described the machinery/conveyors and the different machine states. In order to create
a continuous flow among these machines/conveyors, the packaging line uses line regulation. Line
regulation is the overall term for speed changes and speed levels on different machines. First the term
speed control is explained followed by determining at what moment the machines will produce and at
what speed. At the end of this section we introduce the theory of the V-graph.

2.2.1 Speed control

In order to control the machine’s speed level, HEINEKEN makes use of sensors located on the conveyors.
The location and/or combination of the sensors play an important role by determining the efficiency of a
machine. Considering Figure 2.9, one can see a simplification of sensors located on a conveyor around
the CPL. Here s1, s2 and s3 are sensors placed before the CPL and s4, s5 and s6 are placed after the CPL.

S1 S2 S3 S4 S5 S6

CPL

F IGURE 2.9: S IMPLIFICATION OF SENSORS AROUND CPL

Each sensor manages a conveyor belt, which means that in the illustration there are also three conveyors
before and three after the CPL (illustrated as separated blocks). The sensors before the CPL will regulate
the starvation of the CPL. They measure the input quantity of the CPL, thus if there are enough bottles

15
available that should be labeled. On the other side of the CPL, the sensor measures the output and
therefore if there is no blockage.

Figure 2.10 shows three situations that indicate three different states. These three states are exhibited: 1.
Production, 2. Starvation and 3. Blockage.

S1 S2 S3 S4 S5 S6

1. CPL

S1 S2 S3 S4 S5 S6

2. CPL

S1 S2 S3 S4 S5 S6

3. CPL

F IGURE 2.10: S ENSOR POSITIONS IN 1. PRODUCTION 2. STARVATION 3. BLOCKAGE

In an ideal situation there are always bottles available at the in feed and there is no backup at discharge.
In the first situation this perfect scenario is given, all input sensors (s1,s2,s3) are triggered by bottles and
the output sensors are free (no bottles at the conveyor). In situation two the CPL has no bottles so sensors
s1,s2 and s3 and not triggered, hence the CPL will not produce. Also in situation three the machine is
down because sensors s4, s5 and s6 are triggered by bottles which create a blockage. The colors of this
image are equivalent to those of MES. Green is production, yellow is starvation and blue is blockage.

Line regulation endeavors a continuous flow on the production line, which is favorable according to the
philosophy of KHS. Multiple sensors accompany multiple conveyor belts. Besides, multiple sensors
should be available to create different speed levels. Some sensors are directly linked to the speed level of
the machine. The location (first, second or third) of the sensor has a correlation with the speed level of the
machine.

The start/stop of a machine can also be regulated by just one sensor before and one after the machine.
When the sensor before the machine is triggered and the one after is not, the machine should produce
(otherwise it should not). The disadvantage of this approach is the fluctuating activity of a machine,
because there is just one sensor. This leads to more start/stop situations compared to a situation where
more sensors are located. In Section 1.1, KHS stated that reducing start/stop situations will reduce losses.

16
We can conclude that start/stop situations should be avoided, what will be achieved when a continuous
flow is created. To create a continuous flow, several sensors and speed levels are necessary on the
machines at the production line.

2.2.2 Speed levels

As mentioned in the previous section, speed levels are correlated with the location of the sensor. We
continue with the situation from Figure 2.10. The situation is explained after the machine is repaired, see
Figure 2.11. This figure shows three sensors, and we assume that every sensor has a relationship with the
CPLs. This means that every sensor triggers a speed level at the CPL. In the coming figures we assume
that the CPLs have three speed levels: full (or max), nominal and low (or half). It is not usual that a
machine has more speed levels. For example, the pasteurizer has only one speed level.

Again consider Figure 2.11. The conveyors are occupied which mean that s1, s2 and s3 are triggered.
Now, the machine needs to reduce the buffer as quickly as possible to prevent a blockage of the
pasteurizer. Therefore, sensor s1 is linked to the CPL and sends the message to produce at full speed. In
order to control the continuous flow and prevent start/stop situations, the speed level of the CPL will
reduce when s1 is not triggered any more. The same holds for sensor s2 and s3. When even sensor s3 is
not triggered anymore the CPL will shut down. The reason for reducing this buffer, even when the
pasteurizer is producing, is the fact that the CPL has a higher production capacity per hour than the
pasteurizer, if the CPLs are producing on full or nominal speed. We will explain this in Section 2.2.3. The
maximum, nominal and low speed is different per machine and is therefore not exact. In Chapter 3 we
will mention the exact speed level(s) of the machines.

4. s1, s2 and s3 are triggered: CPL to max speed

Pasteurizer S1 S2 S3 CPL
Blockage Production

5. s2 and s3 are triggered: CPL to nominal speed

Pasteurizer S1 S2 S3 CPL
Production Production

6. s3 is triggered: CPL to half speed

Pasteurizer S1 S2 S3 CPL
Production Production

F IGURE 2.11: CORRELATION BUFFER & MACHINE SPEED

17
2.2.3 V-graph

The figures in the previous section only show a small selection of the whole production line. As explained
during the layout one can see that there are several machines. A main task to enhance the performance is
that buffers should be cleared in order to prevent blockages and should be filled in order to prevent
starvation. To fulfill this task, the machines should react on each other and speed levels should enhance
the continuous flow, with the help of sensors. To control the situation, HEINEKEN uses the theory of the
V-graph in order to establish the predetermined speed levels for all machines.

Härte (1997) stated that the V-graph is a theory based on the bottleneck machine, which contains the
bottom of the V. Härte (1997) stated that: “The machines on either side of the core machine have extra
capacity to restore the accumulation after a failure has occurred”. This overcapacity increases for
machines that are located at a larger distance from the core machine. Theoretically, the machine with the
lowest capacity, the core machine, on line 11 is the pasteurizer with a capacity of 80,000 bottles/hour.
This means that the capacity of the machines before and after this core machine should be higher.
Meaning that the de-palletizer and foil taper (as can been seen in the first and last machine in the line)
must have the highest capacity of the line. The V-graph is developed to cope with machine failures thus
when there is no machine failure, the graph will be flatten. The theory of the V-graph ensures that the
core machine has enough bottles as input to prevent the lack at infeed, and the machines after the core
machine will have a higher capacity in order to prevent backup at discharge.

A core machine can also be called the bottleneck machine, if it has in reality also the lowest capacity. The
situation can occur that the core machine is not the bottleneck machine. For example, if the filler has a
high failure rate and therefore produces less than the 80,000 bottles/hour of the pasteurizer. Then the filler
is the bottleneck machine and the pasteurizer is the core machine. So the core machine is theoretically the
machine with the lowest production capacity and the bottleneck machine is operationally the machine
with the lowest capacity.

Losses made by the bottleneck machine cannot be corrected by other machines. Thus a loss on the
bottleneck machine is a direct loss on total line performance.

In order to determine the bottleneck machine, Härte (1997) introduced the Mean Efficiency Rate (MER).
In Figure 2.12 we show an image where the machine capacity is compared with the Mean Efficiency Rate
(MER). The machine efficiency rate is calculated with the following formula:

The production time plus the internal failure time is the actual time that the machine could produce, so the
machine’s availability. This proves that the core machine is not the same as the bottleneck machine.
Every machine could be the bottleneck machine, dependent on the internal failure time. The machine with

18
the lowest MER is called the bottleneck machine. Note that in Figure 2.12 the bottleneck machines is the

filler.

F IGURE 2.12: V-GRAPH: MACHINE CAPACITIES , MER AND LINE EFFICIENCY (Harte 1997)
Regarding the theory of Härte (1997) the V-graph developed by the supplier of the line is given in Figure
2.13. Here 1 (=100%) is equal to the theoretically capacity of 80,000 bottles per hour of the pasteurizer.
This is also called the nominal speed of the production line. Thus the nominal speed is the speed of the
core machine. The pasteurizer is located at the bottom of the V-graph and represents the core machine.

1,8
1,6
1,4
1,2
1
0,8 default
0,6 minimum
0,4 maximum

0,2
0

F IGURE 2.13: V-GRAPH LINE 11 (BY KHS)

As shown in Figure 2.13, every machine has a higher default speed than the pasteurizer. For example, the
filler is calculated to have a higher capacity of 5% relative to the pasteurizer, and for the CPLs (labelers)

19
this is 15%. Therefore the capacity of
- The pasteurizer is 80,000 bottles/hr.
- The fillers is 1.05*80,000 = 84,000 bottles/hr. This is 42,000btls/hr per filler.
- The CPLs 1.15*80,000 = 92,000 bottles/hr. This is 46,000btls/hr per CPL.

The minimum and maximum speed levels shown in Figure 2.13 are related to the machine capacities of
KHS in general. This speed levels are neglected in our research.

A side note for line 11 is that not the pasteurizer but the filler determines the nominal speed. The speed of
the pasteurizer is hard to measure, because no counters are available on the pasteurizer. Besides the filler
has a higher failure rate than the pasteurizer. This can result in a situation where filler is the bottleneck
machine (if other machines work normal). HEINEKEN determined, from the practical point of view, to
measure the performance on the fillers.

2.3 Speed loss

Speed loss is one of the losses considered by Nakajima (1991). Speed loss is defined by the losses due to
reduced speed of a machine during operations. Speed loss arises from the fact that a machine has different
speed levels. There are two terms that will cover a machine’s speed usage. Machines produce
dichotomously or continuously. Dichotomous production means that a machine has only two speed levels,
not producing (0%) or producing (100%). Where continuous production can have different speed levels
between the 0 and 100%. To clarify, a machine that is down or up (0-100%), so without different speed
levels, does not have speed losses but has blockage, starvation, failures or planned downtime. A machine
with different speed levels can create speed losses when it produces on a lower speed than the nominal
speed. Important to know is that speed loss is a machine state (Section 2.1.3.), but is not mentioned before
because HEINEKEN’s information system is not capable of measuring this continuous state correctly.
Considering the V-graph, some machines have an overcapacity. This is called the maximum speed, which
is therefore higher than the nominal speed.

Example (maximum speed is not included):

- a machine runs for 3 minutes on 75% of the
nominal speed
- the nominal speed is 80,000 bottles per hour (on
line 11; the pasteurizer)

Then:
(100-75%) * 80,000 = 20,000 bottles per hour
for 3 minutes = 3/60 * 20,000
Speed loss = 1000 bottles

Speed loss is defined as the number of bottles produced under the nominal speed minus the number of
bottles produced above the nominal speed. The speed loss cannot be negative because the nominal speed
of the production line determines the output. An example is given below.

For HEINEKEN it is important to realize what amount of bottles they miss due to speed loss.
HEINEKEN compares the different production lines with each other to measure the performance and set

20
targets. When the speed losses of line 11 are not taken into account, an incorrect comparison is made and
targets can be misplaced.

Technology
Negative about introducing different speed levels is that the information systems of HEINEKEN,
including MES, do not match with the technological needs. This means that, looking at the DNA strand in
the IS as shown in Figure 2.14, it cannot be perceived whether a machine is producing continuously.

F IGURE 2.14: MES - DNA STRAND

The problem is that a machine can run for 10 minutes on 10.000 bottles/hour but can also run for 1 minute
on 110.000 bottles/hour and have a failure of 9 minutes. According to the DNA strand option 1 will be
preferable because the strand is all green (option 2 is almost fully red). Nevertheless, looking at
production quantity option 2 is better, because of a higher output. However, first the speed loss of
production line 11 should be made visible, then problems can be detected.

2.4 Measuring Line Performance

HEINEKEN uses the Operational Performance Indicator (OPI) as measure for its packaging
performance per production line. In Figure 2.15 one can find the calculation of OPI which is explained
below. OPI exists of three main components and its formula is given below:

Where the availability is explained in Section 2.2.3. This can also be given as

The performance is calculated with the following formula:

The production time is the time needed for producing the total number of products (good product + reject
and rework). The operating time is the production time + the speed losses and minor stoppages. Where
the minor stoppages are the failures of a machine less than 5 minutes.

The quality is defined as the fraction needed to produce the ‘good product’. This is the time needed to
produce the actual output divided by the production time ( = good product + reject & rework). The
formula is shown below:

21
F IGURE 2.15: CALCULATION OPI
HEINEKEN uses also the OPI to determine the performance of line 11. Nevertheless one can see in
Figure 2.15 that it also includes the external stops, changeover and planned downtime. As mentioned
earlier, these factors have no direct influence on the machine efficiency and the line regulations. For
example, considering department Customer Services & Logistics (CS&L). The situation occurs that an
Automated Guided Vehicle (AGV), a vehicle that is computer controlled and stores the pallets into the
warehouse, is in failure. Then this failure is for line 11 an external failure because the depalletizer (last
machine in line) is in blockage. It therefore decreases the OPI, even when all machines operate perfectly.
For that reason these external stops are neglected in this research. Even the changeover time is neglected
because line 11 produces just one product, so no changeovers take place. At last the planned downtime
(e.g., maintenance) will be neglected because it will not influence the machine efficiency during
production. It does not mean that planned downtime is not important, because with no maintenance the
production performance reduces excessive.

This research focuses on improving the line performance. For that reason the focus is on the performance
component. Therefore the minor stoppages and the speed losses are the main focus in order to improve
the current situation. These minor stoppages will influence the whole production line, which results in
different losses. These are shown in MES as the machine states other than the production state (described
in Section 2.1.3).

In order to focus on the performance measurement, the production time should be considered. The issue
arises that the production time says nothing about the output quantity because of the fact of different
speed levels. When the machine produces for 100% of the time at 10% of the capacity, the performance
will be 100%. Therefore the line performance should consider the output of the machines and compare
these numbers with the production time. Line performance is therefore measured as:

22
Where operating quantity is the number of product that could be produced in a certain period. Production
quantity is the number of ‘good products’ that are produced.

Summary
After reading this chapter we have a clear thought about the production line 11. We know what machines
are located on the production line. Furthermore the functioning of the sensors is explained and the
different machines states are described. The theory of the V-graph is given with the explanation of the
speed losses. We know that the technology of HEINEKEN does not fit with the speed levels which are
implemented by KHS. In the next chapter we discuss the different data of the production line and focus
on the losses.

23
 3. Data Analysis
An overall view of the layout and the philosophy of packaging line 11 was given in the previous chapter.
Based on these findings the current performance of the line will be determined. In Section 3.1 we show
the V-graph of the current situation. Further the problem scope will be narrowed down to create a
structural approach to detect the specific problems. In Section 3.2 we describe the problem in relation
with the layout. In Section 3.3 we will summarize both sections.

After this chapter the focus of the problem on line 11 should be clear in order to improve the line
performance. As been mentioned in the research scope, the focus will start on a few machines. According
to the management and operators opinions the scope should be between the bottle washer and the CPLs,
because problems arise in this area.

3.1 Line Performance

As the V-graph is an important philosophy at HEINEKEN, it is relevant to sketch the current situation
according to the theory of Härte (1997). In Figure 3.1 we show the V-graph. The blue line, with
corresponding rhombuses, represents the capacity of a machine when the line produces normally. It can
be stated that the pasteurizer is the core machine since it has the lowest capacity. Therefore the other
machines should have a higher production capacity. The red line with corresponding triangles represents
the MER, as explained in Section 2.2.2. The graph is expressed in an average amount of crates during an
8-hour production shift.
Availability x
minimum capacity =
# crates per shift (production time +
starvation + blockage)
35.000
x minimum capacity
30,000
30.000 28,000 27,667
28,000 26,667

25.000
25,891 25,699 25,031
24,695 23,892 Realized
20.000
production
quantity

15.000
Minimal
machine
10.000
capacity

5.000

-
Bottle EBI's Fillers Pasteurizer CPLs
washer

F IGURE 3.1: V- GRAPH CURRENT MACHINE PERFORMANCE (AVERAGE)

24
According to Figure 3.1 we can conclude that the pasteurizer is the machine with on average the lowest
realized production, and is therefore the bottleneck machine. This is consistent because the pasteurizer has
the lowest machine capacity. In order to improve the output quantity and therefore the line performance, it
is necessary to further improve the machine with the lowest realized production quantity (Härte, 1997). It
is not efficient to focus on the other machines, because improving these will not lead to a constant higher
output quantity of line 11. The bottleneck machine determines the output quantity of the production line
and therefore further analysis has to be done.

First the number of times that the pasteurizer is the bottleneck machine has to be determined. There is a
possibility that the average production quantity is influenced by outlying work shifts. The variation in
production quantity of the pasteurizer should be excluded to make the data valid and reliable.
In Figure 3.2 we show the number of shifts when a certain machine is the bottleneck. The left figure
shows the number of times a machine is the bottleneck at all shifts from 1 July ’14 till 31 August ’14. The
total amount of shifts is 100 whereof 30 shifts above 19,000 crates. The right figure shows the shifts that
are above the 19,000 crates in the same period. Shifts above 19,000 crates are defined by HEINEKEN’s
management as the performance target of production line 11 (line performance of
71%).

F IGURE 3.2: NUMBER OF TIMES THAT A MACHINE IS A BOTTLENECK – T OTAL SHIFTS: 100/ ABOVE 19,000 CRATES :
30

The following conclusions can be made from Figure 3.1:

 The bottle washer scores worst on machine efficiency. The biggest gap between machine capacity
and operational production is on the bottle washer.
 The fillers score second worst.
 Including the internal failures the pasteurizer is still the bottleneck.
The following conclusions can be made from Figure 3.2:

 Considering all shifts, the pasteurizer is 33% more often the bottleneck than the fillers and more
than 50% compared to the bottle washer.
 Fillers are less often the bottleneck machines when shifts are above 19,000 crates (5/30=16,67%)
compared with all shifts (29/100=29%).

25
Considering the conclusions we could focus on the bottle washer, the fillers, the pasteurizer or a
combination of these. According to Härte (1997) the focus should be on improving the bottleneck
machine. Therefore and because of the following reasons, we choose to focus on the pasteurizer:

 The efficiency of the bottle washer depends of the quality of the returnable materials from the
market. The biggest problems at the bottle washer arise from falling bottles. When the quality of
the returned material is bad, more bottles will collapse. When bottles collapse they will block the
entrance of the bottle washer, whereby the utilization of the bottle washer is below standard. This
problem will be solved by starting a ‘Break Down Analysis’, which is not part of this research.
 The inefficiency of the bottle washer arises because crates from the market are not filled
completely. The capacity of the depacker is based on crates with 24 bottles, which is often not the
case. This reduces the efficiency of the bottle washer. The management of HEINEKEN is busy
with solving this problem by introducing a new sorting machine.
 The bottle washer covers a relatively high position in the V-graph. This means that when the
quality of the material meets the expectations, the bottle washer will not be the problem.
 The fillers are considered to be the most important machines of the production line. HEINEKEN
has a close relationship with KHS in order to optimize the use of these fillers. The failures of the
fillers have the most influence on the efficiency of the fillers. Focusing on the failures of the
fillers will lead a mechanical approach. Besides, KHS is nowadays constantly looking for
improvements on these fillers. When we focus also on the fillers, our interests could struggle and
measurement can be become invalid.
 At the end, the pasteurizer is the bottleneck machine. When improvements are made on this
machine, it will directly positively influence the line performance.
According to Härte (1997) the focus should
be on the bottleneck machine and according
to Nakajima (1988) companies should
reduce their losses in order to improve the
line performance. Therefore we focus on the
losses of the pasteurizer. These losses are
shown in a pie chart in Figure 3.3.

Up to now we have considered the F IGURE 3.3: LOSSES P ASTEURIZER

bottleneck machine by determining the
machine efficiency. From here we can determine the efficiency between the machines, the conveyors and
corresponding sensors. According to the management of HEINEKEN, this part is forgotten most of the
time. The management looks only at the machine efficiency, because the efficiency of the conveyors and
sensors is not registered in their information system (e.g., MES). To determine the efficiency between
machines, MES can be considered and a special speed loss tool can support the correctness of MES. This
tool will be explained at the end of this section.

We focus on the pasteurizer and look further at the efficiency from the filler to the CPLs. Again consider
Figure 3.3. As one can see, the biggest losses on the pasteurizer are starvation and blockage. The
starvation and blockage of a machine occurs due the fact that other machines are in failure. Nonetheless
there could be other aspects that determine the length and reason of a starvation and blockage. For
example, if sensors are not aligned efficiently with the machines. For that reason an analysis is done to
determine the starvation and blockage of the pasteurizer. This can easily be done with the DNA strand
from MES. In MES, the relations of the machines are clearly visible. Furthermore, to validate the

26
information in MES, it should be compared with real-life situations at the production line. Different DNA
strands of different work shifts are analyzed and we show an example of a remarkable situation in Figure
3.4.

F IGURE 3.4: MES DNA STRAND ( EXPLANATION BLOCKAGE PASTEURIZER)

In Figure 3.4 two inefficiencies are shown. The first one, specified with the red circle, shows that the
packer is in failure. When the packer is in failure, both CPLs should create a blockage after the buffers
between CPLs and packer are filled. But as situated in the red circle, CPL112 has created a starvation
state. This means that the Programmable Logic Controller (PLC) measures the wrong indicator or MES
translates the wrong message. A PLC is “a digital computer used for automation of industrial
electromechanical processes” (Lin,
S., Huang, X. 1998). This means
that a PLC measures predetermined
variables and indicators on the
production line itself, this can vary
from the number of bottles
produced till the amount of
detergent in a bottle. This will be
translated by MES in order to make
it visible for operators and the
F IGURE 3.5: P ERCENTAGE OF LOSSES CPLS
management. The registration
fault can be present on both situations. Furthermore the red square presents a huge inefficiency. As stated
in before mentioned sections, a blockage/starvation on the pasteurizer results in a direct loss in output
quantity. In Figure 3.4 the red square shows such a direct loss. CPL111 is in failure, CPL112 is in
starvation and the pasteurizer is in blockage. When the pasteurizer is in blockage, a CPL may never be in
starvation. In this situation the production of CPL112 could prevent that the pasteurizer will create
blockage, and therefore prevent a loss in production quantity. This loss in production quantity has an
average of 5 minutes. This is based on an analysis of MES where all strands of the past two months are
considered. Some examples of those strands are shown in Appendix A.

Comparing these red shapes we can see a difference. The situation given in the red square is not a
measurement or translation error, such as the red circle. We can conclude this by looking at the state of
the packer. Consider the DNA strand of the packer (“Inpakker110” in Dutch) shown in Figure 3.4. Below
the red square the packer is in the starvation state. This means that the packer has no bottles at infeed. For
that reason it is valid to say that CPL 112 has to be in starvation mode too. This is different from the red
circle, where CPL 112 should create a blockage state because the packer is in failure. We can conclude
that the situation in the red square is not a translation error as shown in the red circle, but a loss in
production due to inefficiencies. The red square shows that the blockage of the pasteurizer arises from

27
inefficiencies on the CPLs. The losses of the CPLs are shown in Figure 3.5 to brighten these
inefficiencies. Three remarks regarding these losses/inefficiencies are explained below.

The first thing to notice is the large starvation percentage (64%) of CPL 112 and the small blockage
percentage (3%). This is in line with the situation we show in the red circle (Figure 3.4), this problem
arises from the fact of an error in data storage. From this point of view it is hard to define the losses on
CPL112 regarding the starvation and blockage. Expected is that some of the blockage time is measured as
a starvation time, in MES. Comparing the real-life data on the production line with the registered data in
MES, it can be stated that the regulation of CPL112 does not operate efficiently. This inefficiency is
described in Section 3.2. After noticing the incorrect regulation of CPL112 also the data registration of
CPL111 is checked by experts. Nevertheless, the registration of the data of CPL111 works correct when
the real-life data and the data in MES are compared. For that reason we consider that the blockage and
starvation percentages are valid.

Production balance
Production
Productie
CPL111
44% E111
56%
Productie
Production
CPL112
E112

F IGURE 3.6: P RODUCTION BALANCE CPLS

The second inefficiency is the difference in production quantity of the CPLs. The production quantity of
both machines is given in Figure 3.6. This difference is another argument for the inefficiency of the area
from the pasteurizer and the CPLs. The maintenance schedule of production line 11 is for parallel
machines exactly the same, so also for both CPLs. The activities that should be done for operators are
called CILT activities. CILT stands for Cleaning, Inspection, Lubrication and Tightening. These
activities are unequal because CPL111 produces more bottles than CPL112. For example, when we
consider parts that should be replaced after 1000 production hours. The production hours of the CPLs are
based upon the number of parts that are produced, measured in output quantity. This amount of parts is
divided by two to calculate the parts produced per CPL machine, because the maintenance department
assumes that both CPLs will produce the same amount. Thus the maintenance department creates their
maintenance schedule based on equal production. This means that in the current situation at the CPLs
problems arise. Parts on CPL111 are changed too late and parts on CPL112 are changed while they are in
good condition. In Appendix J we show the differences per CPL, registered by operators.

The difference in production quantity of Figure 3.6 emerges for different reasons. In order to prevent that
some aspects are overlooked and to determine the reasons for the unbalance we organized a brainstorm
session. This session is done in the presence of the operators which are familiar with the packaging
process and problems encountered with this process.

The third inefficiency is empirically observed at the production line. When CPL111 starts producing it
takes +/- 9 minutes before CPL112 starts with the production. This could overlap the two notes mentioned
before, but should definitely be taken into account.

28
Visibility/Transparency
In Section 1.3 we conclude that first the losses should be made visible for HEINEKEN. Especially the
speed losses are hard to define when we consider the MES DNA strand. For example, in Figure 3.4 we
see in the red square a large blockage time at the pasteurizer. It is hard to see what influence the blockage
time has on the fillers and CPLs. The reason is explained before, in Section 2.3 – Technology. The
conclusion was that the different speed levels on the fillers and CPLs can result in invisible losses in
MES. The fillers and CPLs can produce at a low speed level, where the pasteurizer has only the nominal
speed. Considering the red square in Figure 3.4, we can see that simultaneous with the blockage of the
pasteurizer, the filler is producing. Without different speed levels on the filler this situation cannot occur.
Nevertheless, the problem described in Section 3.1 holds, because the CPLs are in starvation state. When
the CPLs are producing, like the fillers, there is no visible problem. This means that the inefficiencies
between the pasteurizer and CPLs still exist.

Measuring Speed Loss - Excel Tool

In order to create insight into the non-transparency of MES, we developed a tool in order to measure the
speed losses. The goal of this tool is to visualize the speed losses on the fillers and CPLs. When these
speed losses are known, we can focus our research more specifically. This tool uses the bottle stop of the
CPL as measure. The bottle stop determines whether SBs are able to pass the CPL or not. This is
dependent on failures, material availability and operator presence. Nevertheless, the bottle stop is the
most reliable sensor on the CPL. The combination of the bottle stop with the speed level of the CPLs
determines the amount of bottles passed. This amount will be used in the data analysis. Figure 3.7 shows
the graph that is the result of the tool. In this graph we see at what level the CPLs have produced during a
work shift and during a production week. Thus, the tool measures the speed level of the fillers and CPLs
and visualizes these into a graph. This tool will be used as a KPI in order to measure the performance of
the line during this research. We describe the use of this model in more detail later on.

F IGURE 3.7: SPEED LEVEL CPLS - EXCEL TOOL

29
 3.2 Problem design – layout
The conclusion from previous sections is
that the current situation at the area
between the pasteurizer and the CPLs is
not efficiently organized. In order to
determine the causes of these
inefficiencies this area should be outlined
specifically. Therefore the layout of this
area is shown in Figure 3.8. In this figure,
the bottles move from the pasteurizer (left)
to the CPLs. Parts A, C and E represents
the lower deck of the pasteurizer and part
B, D, F and G represents the upper deck.
The green octagons show the location of
the sensors, where no distinction is made
of the nature of the sensor
(switch/photocell). Every sensor is linked
with its corresponding conveyor. The use
of the sensors on the conveyors and how 45%
these are triggered was explained in
Section 2.1.2. Triggering a sensor can lead
to several actions. For example, sensors
determine the speed level of the conveyor. 55%
Besides, some sensors also determine the F IGURE 3.8: LAYOUT CONVEYOR BETWEEN
speed of the pasteurizer and/or CPLs. P ASTEURIZER AND CPLS
These sensors determine when a machine
starts/stops producing and at what speed. We start with focusing on these sensors. The sensors that
determine the speed of the machines are interesting, because these sensors determine the production
quantity of the machine. Other sensors just have the function of determining the moment of changing of a
conveyor speed.

The sensors that trigger the CPLs are highlighted in three different situations. Situation 1, shown in
Figure 3.11, explains when CPL111 starts producing to low speed. The triggered sensors are colored blue
and the non-triggered sensors are colored green. As explained in Section 2.1.2, sensors are triggered when
SBs enumerate from the CPL to the concerning sensor, this means that a CPL or a succeeding machine is
in failure. Note that CPL112 has no low speed. Situation 2, shown in Figure 3.10, explains when the
machines start at nominal speed. Note that CPL112 directly moves to nominal speed. The last situation is
number 3 and is shown in Figure 3.9. This figure displays the situation that both CPLs move to high
speed. The number of speed levels is predetermined and the values are fixed. We can conclude from the
figures explained above, that CPL111 has three speed levels and CPL112 has only two.

30
 CPL112: NO
10 = triggered
low speed 10
= not triggered

CPL111 to
low speed

17

F IGURE 3.11: S ITUATION 1: CPL111 TO LOW SPEED

CPL 112 to 10 = triggered

nominal speed 10
= not triggered

10 12
CPL111 to 14

nominal
speed
17

F IGURE 3.10: S ITUATION 2: CPLS TO NOMINAL SPEED

10 = triggered
13 10
= not triggered

10 12
14
CPL 112 to
full speed

17
CPL 111 to
8 OR 9 full speed

F IGURE 3.9: S ITUATION 3: CPL S TO FULL SPEED

31
When the CPL or the succeeding machine that was in failure is repaired, the amount of products on the
line decreases. This means that the CPLs will now produce at a higher frequency than the pasteurizer. The
buffer/conveyors will slowly decrease in occupation as we explained in Section 2.2.2. This decrease will
influence the sensors and therefore the CPLs. This happens in the same way as the figures on the previous
page shows, but the other way around. This means that at sensor 10 in Figure 3.9, CPL111 will change
from full speed to nominal speed if sensor 13 is not triggered anymore. CPL112 will reduce its speed
from full to nominal if sensor 8 or 9 is not triggered anymore. CPL111 will change from nominal speed to
zero if sensor 14 is not triggered anymore. CPL112 will fall down if sensor 10 is not triggered for 30
seconds. This is the only difference compared to the enumeration of the bottles, explained before. This
delay of 30 seconds ensures the continuity of the material flow and the start/stop situations of CPL112.
Without this delay, CPL112 would stop if the sensor is not triggered and start again if the sensor is
triggered. Experience has shown that the material flow around sensor 10 is highly fluctuating. Therefore
the delay of 30 seconds ensures continuity.

Figure 3.12 shows a detailed version of the conveyor belt

part I. Location A represents the output from the lower
deck of the pasteurizer and B is the output from the upper
deck. Exit X move towards CPL111 and exit Y moves to
CPL112. We would expect that if a SB enters conveyor
belt H at location A it exits I at location X and when a
SB enters at location B it exits at location Y. In real-life
this will not be the case, if conveyor H is not completely
filled. For example, bottles arrive (entrance) at locations
A and B. Then the SBs all move to the red dot, because
bottles are transported to the end of conveyor belt H and
are ‘guided’ by the fence towards conveyor I. As a result,
the bottles will all exit at location X. Note that if bottles

reach part I at the blue dot, they will exit at location Y. F IGURE 3.12: BEHAVIOR OF SB AT
BENDED CONVEYOR BELT
If we zoom out from this picture and get back to the three
situations on the previous page we can conclude that, if conveyor H is not fully occupied, CPL111 (X)
gets more SBs than CPL112 (Y). This is one of the reasons that explain the production unbalance
between the CPL111 and CPL112. If both conveyor H and I are filled with SBs, this bend is not a
problem regarding the production balance. So, if just one deck of the pasteurizer is filled (entrance A or
B) the bottles will move to exit locations X and therefore towards CPL111. This whole situation is clearly
exhibited in Figure 3.13. In this figure the upper right corner shows from which angle this picture is
taken. The entrance of part H is at the right side of Figure 3.13 and the exit at part I is at the upper left
corner. Furthermore, sensor 10 is highlighted. The bottles at the right side of the ‘T-junction’ at part I will
move to CPL112 and the left side moves to CPL111.

32
 Sensor 10

I
I

F IGURE 3.13: P ROBLEM SITUATION - REAL LIFE

The problem that we explained in the previous paragraph ‘triggers’ another problem. Again consider
Figure 3.12. From previous pictures we know that the green octagon represents a sensor, which is
triggered when bottles ‘hit’ the sensor. As we know from the previous paragraph, it will be hard to trigger
sensor 10 because all bottles will move towards the red dot. Sensor 10 is only triggered when conveyor
belt I is completely filled with bottles. This is clearly visible at Figure 3.13, where we can see that the
bottle flow move along part H and not
towards the sensor. Besides it occurs that if
CPL111 fails, the material flow on conveyor 1
H also stops. When sensor 10 is not triggered, 1 1
CPL112 will stay down, while it is ready to
2 7
produce. This situation was already known
and showed in Figure 3.4, in the red square. 2

When the entrance at H (locations A and B) is

not completely filled, bottles move towards
the red dot. From there all the bottles will
move to the exit X. As we know from Figure F IGURE 3.14: CPLS IN STARVATION MODE - GREEN BAR:
3.10, CPL111 will start producing when FILLED WITH BOTTLES

sensors 14 & 17 are triggered. When CPL111

is producing it will prevent that conveyor belt I is completely filled, therefore that sensor 10 is triggered
and at the end that CPL112 starts producing. Even worse, CPL111 will start producing at full speed, even
when CPL112 is not started yet.

33
We can conclude from this section that the input towards CPL111 is higher than the input towards
CPL112 because of layout inefficiency, the transition from part H to part I. SBs will move to the outer of
the bended conveyor I. Therefore CPL 111 starts earlier with the production and also ‘stealing’ the bottles
in front of sensor 10, wherefore CPL112 does not start producing.

At the moment that the pasteurizer or a preceding machine is in failure, no bottles will enter part A & B.
As explained before CPL 111 & CPL112 will produce until the corresponding sensors are not triggered
anymore. In Figure 3.14 the green bars represent the conveyors filled with bottles.

3.3 Summary of data analysis

Summary of analysis (Section 3.1)
As determined in this section, the scope of this research will be on the area between the pasteurizer and
the CPLs. The pasteurizer creates blockage while the CPL 112 is in starvation. This means that the
regulation of the CPLs is not operating efficiently. The results from before mentioned analysis and the
results from the brainstorm sessions that contribute to losses of the pasteurizer are listed below:

 Blockage at CPL 112 is not registered correctly. The blockage time is displayed as
starvation time.
 Sensors are not efficiently regulated with the CPLs.
 The CPLs do have inefficient predetermined speed levels.
 The conveyors between the pasteurizer and CPLs are not reacting efficiently to the
sensors.
 The conveyors do have inefficient predetermined speed levels.

The first node is explained before and is a problem within the information system and has to be solved by
the PA/PI-engineering group within HEINEKEN. The other nodes will be discussed in Chapter 5.

Summary of problem design (Section 3.20)

Based on the layout of line 11 we can summarize the following points:

 If part H of the production line is not completely filled, bottles will mainly move to CPL111.
 CPL111 produces at full speed even when CPL112 is not started.
 CPL111 ‘steals’ bottles from CPL112 and therefore prevents that sensor 10 will be triggered and
prevents CPL112 to start producing.
 If CPL111 fails and sensor 10 is not triggered, than CPL112 will stay down and the pasteurizer
creates blockage. Note that this happens not all the time, but is dependent of the utilization of the
bottles.
From these summaries we define the next ‘keys issues’ that define our problem and which will contribute
to the line performance of line 11.

 How should the sensors be regulated, in order to create an equal production balance.
 What is/are the ideal speed levels of the CPLs (to minimize the failures).

34
 4. Literature Review
A lot of research is done in the field of improving line performance. A lot of improvements are based on
continuous improvement theories. In order to know how HEINEKEN continuously improves the
performance, a comparison is made between HEINEKEN’s Total Productive Management and alternative
improvement strategies that are made in literature. This is explained in Section 4.1. Based on the
comparison the strengths and weaknesses of HEINEKEN’s TPM are shown, which we explain in Section
4.2. Furthermore in Section 4.3 we will discuss the performance measurement according to the literature.
In Section 4.4 we consider some related research. At last, in Section 4.5, we describe what simulation is
and what types there exist. These sections will serve as the basis of our research.

4.1 Continuous improvement strategies

There are multiple improvement strategies and it hard to separate them from each other. There are some
close related programs such as Total Quality Management, Just in Time (Cua et al., 2001), Lean (Arlbørn
& Freytag, 2013), Theory of Constraints (Rahman, 1998), and Six Sigma (de Mast & Lokkerbol, 2012;
Schroeder, Linderman, Liedtke & Choo, 2008). These improvement strategies have grown to
comprehensive management strategies. Implementing them requires a changing a working culture
towards a continuous improvement culture, which can prove to be difficult and have an impact on
involved personnel (Farris et al., 2009). There are discussed four of these strategies in more detail, which
are: Lean management, Six Sigma, Theory of Constraint and Total Productive Maintenance. For every
strategy an analysis from literature is made. The Total Productive Maintenance in covered in Section 4.2.

Lean management
There is no commonly accepted definition of lean management, and therefore there are a number of views
on lean: “Ranging from a focus on waste elimination, utilizing operational tools and implementing
specific production-related principles, to identifying conditions that are linked to the product and/or the
service and the predictability of demand and its stability” (Arlbørn & Freytag, 2013). Nevertheless, the
basic principle of lean management is eliminating waste. Wastes are all activities that add no value to the
end product. The principle assumes that eliminating waste will increase the business performance. The
focus lies on the improvement of small improvements, where the overall flow time can be reduced, the
variation can be reduced and the quality will increase (Shah & Ward, 2003). However, critiques against
lean management involve a decrease in operator autonomy and multi-skilled labor qualities.

Six Sigma
Six Sigma tries to solve problems from a data driven point of view (Pepper & Spedding, 2010). Six sigma
focuses on process variation reduction. Projects are addressed from start to finish, and each project is
controlled by a certified project leader. A project is signed off only when the target financial savings are
verified (Bendell, 2006). Critique on Six Sigma is aimed on three main aspects. The first one is the lack of
taking into account the system interaction. The second one is that it is a cost driven approach instead of
focusing on the customers. The third aspect is that tools as innovation and creativity are neglected and
only the (statistical) data analysis is taken into account (Bendell, 2006).

Theory of Constraint (ToC)

The principle of the Theory of Constraint can be formulated into two statements according to Rahman
(1998):

- Every system must have at least one constraint (no constraints means unlimited profit).

35
 - The existence of constraints represents opportunities for improvements (positive constraints
determine the performance of a system).
Therefore these constraints form the focus of improving the production processes within a company. The
main focus lies on the throughput. The theory also involves the research to hidden bottlenecks. The
critique on the theory is aimed at the lack of involvement of operating employees. The ToC focuses on
the whole system and therefore, employees working at part of this process can contribute a very limited
way.

4.2 Total Productive Maintenance (TPM)

TPM is mostly known from Japanese car manufacturers like Toyota, and was introduced in the early
1970s. The section ‘TPM philosophy’ will discuss this concept in more detail. This philosophy consists of
several “pillars” that represents together the framework of TPM. The explanation of TPM is relevant
because HEINEKEN uses also TPM.

TPM philosophy
TPM is founded by Nakajima (1988) and is a continuous improvement philosophy. The definition in
literature states that Total Productive Maintenance is a methodology to continuously mange, optimize and
improve a supply chain by eliminating all losses, and involving all employees of the organization (Ahuja
& Khamba, 2008). The methodology aims to “increase the availability and effectiveness of existing
equipment in a given situation, through the effort of minimizing input and the investment in human
resources which results in better hardware utilization” (Chan, Lau et al. 2005, Ahuja & Khamba, 2008).
TPM is applied through the entire organization and involves directors, management, support and
operators. By training employees, a working culture can be created in which losses are not accepted and
processes are structurally improved. Especially the cooperation between maintenance and operations is
very important, since operators shift from pure operational tasks to a more all-round shop floor
management role (Ahuja, 2001). TPM is an aggressive maintenance strategy that focuses on actually
improving the functioning of the production equipment (Tsarouhas, 2007). TPM is especially used in
organizations with a high level of equipment automation (Rolfsen & Langeland, 2012).

Due to its focus on personnel, the fragile point of the methodology is the capability of the personnel
(Willmott & McCarthy 2001). The capability of the personnel determines in what way TPM can be
successful in an organization.

TPM pillars
According Nakajim (1988), TPM has eight different pillars. Within an organization these pillars together
form the framework for TPM (Rolfsen & Langeland, 2012). These pillars have their own direction
regarding losses. Ahuja & Khamba (2008) defined each pillar in relation with operational skills. These
combinations are shown in Table 4.1.

Pillar Operational skills

Autonomous maintenance (AM) Fostering operator ownership
Perform CILT, adjustment and readjustment of
production equipment
Focused improvement (FI) Systematic identification and elimination of
losses.

36
 Working out loss structure and loss mitigation
through structured why-why, failure mode and
effects analysis. Achieve improved system
efficiency. Improved OEE on production systems
Planned maintenance (PM) Planning efficient and effective PM, predictive
maintenance and time base maintenance systems
over equipment life cycle. Establishing PM check
sheets. Improving mean time before failure, mean
time to repair and mean time between assists.
Quality maintenance (QM) Achieving zero defects
Tracking and addressing equipment problems and
root causes
Setting 4M (machine/man/material/Method)
conditions
Training and Education (T&E) Imparting technological, quality control,
interpersonal skills
Multi-skilling of employees
Aligning employees to organizational goals
Periodic skill evaluation and updating
Safety, health and environment (SHE) Ensure safe working environment. Provide
appropriate work environment. Eliminate
incidents of injuries and accidents. Provide
standard operating procedures
TPM office Improve synergy between various business
functions
Remove procedural hassles
Focus on addressing cost-related issues
Apply 5S in office and working areas
Measurement of TPM performance
Development management (DM) Minimal problems and running in time on new
equipment
Utilize learning from existing systems to new
systems
Maintenance improvement initiatives, Early
equipment management
TABLE 4.1: TPM P ILLARS (AHUJA & KHAMBA, 2008)

4.3 Performance measurement

Neely et al. (1995) define performance measures (PMs) and metrics as the process of quantifying the
efficiency and effectiveness of an action. The term metric refers to the definition of the measure, how it
will be calculated, who will be carrying out the calculation, and from where the data will be obtained.
According to Fitzgerald et al. (1991), there are two basic types of PMs in any organization; those that
relate results (competitiveness and financial performance) and those that focus on the determinants of the
results (quality, flexibility, resource utilization and innovation).

According to Neely (1999), two features are necessary for a business performance measurement system;
performance measures and a supporting infrastructure. Although the existence of measures is often taken
as a given, there is no such agreement on the nature and design of those measures. A supporting
infrastructure can vary from very simplistic manual methods or recording data to sophisticated

37
information systems and supporting procedures which might include data acquisition, collation, sorting,
analysis, interpretation and dissemination (Neely, 1999).

Six big losses

Nakajima (1991) stated that a loss of an production facility is the difference between an OPI of 100% and
the actual OPI. By reducing the losses, the actual OPI increases. Nakajima (1991) categorizes losses into
“six big losses”. Nakajima (1991) categorizes these losses into “six big losses”: equipment failure, setup
and adjustment, idling and minor stoppage, reduced speed, defects in process and reduces yield. As one
can see in Figure 4.1, these losses are used to compute the OEE.

F IGURE 4.1: RELATION BETWEEN OEE AND SIX BIG LOSSES - (CHAN , 2005)
With OEE, an organization looks at the total time that is available, down time losses, speed losses and
defect losses. These three types of losses are translated into Availability, Performance and Quality.

Operational Performance Indicator (OPI)

Within the PIs there can be made a difference between performance indicators (PI) and key performance
indicators (KPIs), the last one indicates which actions are needed to dramatically increase performance
(Parmenter 2010).

To measure the performance, HEINEKEN uses a variant of Nakajima’s (1988) overall equipment
effectiveness (OEE) in TPM, as a KPI. This variant is the Overall Performance Indicator (OPI). The OPI
is measured over the performance of the fillers. The OPI is determined by the product of Availability,
Performance and Quality, like the OEE. As stated by Nakajima (1991), the OEE identifies (hidden) losses
related to any decrease in performance by evaluating each component. Eliminating these losses results in
a higher performance, where according to Nakajima (1991) zero losses will result in an OEE of 100%.

The Operational Performance Indicator (OPI) is calculated as follows:

OPI = Availability * Performance * Quality

38
Where these three indicators have their own equations which are stated below

These indicators are calculated in order to measure the line performance of a production line. As stated
above, these indicators are multiplied which means that the weight of these indicators are the same. The
quality measures the ratio of good products, which are the products that exit the production line in order
to enter the market. The performance measures the efficient production time of all operating time. The
– –
– – – –
This means that only the blockage and starvation times are the difference between operating time and
production time. These times are used in order to calculate the performance. The availability is the
operating time (described above) divided by the manned time. The manned time is the time that operators
are working on the production line, which is in total 9600 minutes per week.

CILT
An important part of TPM for production line 11 is the use of CILT-activities. CILT consists of Cleaning,
Inspection, Lubrication and Tightening. These activities play an important role in order to maintain the
machines. Every operator on the production line has its own responsibility. These activities should
prevent machine breakdowns and improve the line performance.

4.4 Related Research

According to Kegg (1990), in the 1970s, companies with transfer lines began to study the productivity of
their lines. Each found that the actual number of parts produced per year was about half of the theoretical
maximum. This result was widely discussed and published, but the causes of these production losses were
kept classified. This led to the conclusion that sensors were needed in order to measure inefficiencies on
different places on the production line. These sensors are called the Programmable Logic Controllers
(PLCs). It was the first major milestone in the use of electronics to extract information from sensors in
manufacturing. The PLCs became a reliable measure to collect data from the production line. The data
from the PLCs supports technicians to detect problems earlier and therefore the productivity increased
(Kegg, 1990). In the 80s the combination of PLCs and use of measurement systems allows to detect
trends on machine failures and other inefficiencies, therefore the PLCs play in important role in the
automation of production lines.

Another use of sensors on a production line was to cope with high flexibility and productivity (Mahalik,
Lee, 2001). Sensors do not only register information about machine breakdowns but also about starvation
and blockage at the production line. Sensors are linked with conveyors, but also with machines. Often the
PLCs are positioned on the conveyors and collect information of the (number of) products.

39
Conveyor Theory
Conveyor systems can most of the time be built from basic units as linear conveyor systems. “Conveyor
systems are typically installed as simple straight assembly lines and a number of workplaces are set on
each side of the conveyor for manual and/or automated operations.” (Yeung & Moore, 1996). For simple
configurations the design and implementation of conveyor system is relative easily. Furthermore, the
control programs are easily developed and executed by PLCs. Nevertheless, the demand for multi-product
mixes and flexibility can require more complex conveyor systems (Yeung & Moore, 1996). Conveyor
systems which support the multi-product mixes and variable product routing need high control
requirements.

Bastani (1988) analyzes in his paper a multiple homogeneous closed-loop conveyor system with discrete
and deterministic flow of material, taking into consideration the unit length of products. Where Kwo
(1958) established three fundamental principles that govern the satisfactory operation of conveyor
systems, also known as the Conveyor Theory:
- The speed of the conveyor must be within the permissible range (Speed Principle).
- The conveyor must have enough capacity (Capacity Principle).
- The number of items loaded onto the conveyor must equal the number of items unloaded (Uniformity
Principle).

Kwo’s work is expanded by Muth (1974) who has treated both continuous time and discrete time material
flow, multiple load and unload stations (input/output) and stochastic material flow. Additionally,
according to Belzer et al. (1978) several authors have applied waiting line analysis and simulation to the
problems included with the field of conveyor systems.

Conveyor systems in simulation

“Conveyor systems in simulation can be classified by the type of conveyor as well as the size of the load
moving on the conveyor” Banks(1998). Difference is made between a non-accumulating conveyor, where
a load stops when the entire conveyor stops and an accumulating conveyor. On accumulating conveyors,
loads continue moving and bench products from behind that are stopped. Banks (1998) considers different
load sizes as pallet conveyors, case conveyors and power-and-free conveyors. Power-and-free conveyors
have carriers that attach to the load being transported and are often seen in automotive paint applications
(Banks, 1998).

Note that single product conveyors are not discussed.

40
Choice of method
Two types of models are typically used to estimate performance measures: simulation models and
analytical models. The definition of simulation is: “the process of designing a model of a system and
conducting experiments with this model for the purpose either of understanding the behavior of the
system or of evaluating various strategies (within the limits imposed by a criterion or set of criteria) for
the operation of the system” (Shannon, 1975). Discrete-event simulation models mimic the real system by
constructing a list of events that occurs in the real life. At each event occurrence, such as a process
completion or a breakdown, new events are scheduled and added to the event list. The randomness in
times between two events (arrival or breakdowns) is captured by drawing random numbers from pre-
specified distributions. These distributions can be derived from data of the production system; both
empirical and fitted distributions can be used. These distributions can be translated into stochastic
variables. “A great benefit of simulation is the ability to include stochastic variables, for example the inter
arrival time of products and the breakdowns of machines” (Wein & Chevalier, 1992). A simulation
model is a simplified model of reality and is used to test out different production rules.

Analytical models try to capture the system in terms of sets of equations and then solve these equations.
In many cases, the solution of these equations is numerical. Since most systems in practice are too
complex to analyze exactly, heuristic methods need to be constructed to obtain approximate results.
According to Tino & Khan (2013) “Simulation techniques are often time consuming. Therefore,
analytical models are often used to generate solutions in a fraction of the time. These models however are
complicated and take effort to derive.”

An analytical analysis is basically a set of formulas, where simulation is a graphical tool for analysis.
Simulation enables us to analyze the impact of, e.g., breakdowns and inter arrival times. At the production
line of HEINEKEN these events should be considered to mimic real life situations. This would be too
hard to solve with an analytical model due to the dynamic production environment.

Simulation type
Law (2006) distinguishes several types of simulation models. First we determine which dimensions are
applicable to this research. There are three dimensions, which are:

 Dynamic or static simulation models. A dynamic model shows how a system evolves over time
while a static simulation model represents the system at a certain time.
 Stochastic vs. deterministic simulation models. A stochastic simulation model exists of random
input components while a deterministic model does not contain any probabilistic components.
 Discrete vs. continuous simulation models. In a discrete simulation model the state variable
changes at different points in time while a continuous model has continuous state changes.

Furthermore there is a distinction between terminating and non-terminating simulations. In terminating

simulation there is a natural event that specifies the end of the run. This can be for example, the end of a
shift or end of a day. Non-terminating simulations consider a steady state performance measure. The
performance depends on initial conditions, and after time t the simulation can turn into steady state
behavior. This is not always the case, because parameters might be changing over time which results in a
continued transient system behavior. Considering steady state parameters, the time it takes until the
system turns in a steady state has to be determined in order to measure performance.

41
Other subdivisions of simulations discussed by Law (2006) are:

 Monte Carlo simulation. This contains a static discrete simulation model and can be stochastic or
deterministic.
 Spreadsheet simulation. Spreadsheet simulation refers to the use of a spreadsheet as a platform
for representing simulation models and performing simulation experiments.
 Continuous simulation
 Discrete-event simulation
 Combined discrete-continuous simulation

The discrete-event simulation, models the operation of a system as a discrete sequence of state changes in
time.

Conclusion
TPM is used by HEINEKEN and the philosophy is based to “increase the availability and effectiveness of
existing equipment in a given situation, through the effort of minimizing input and the investment in
human resources which results in better hardware utilization”. The performance measure at HEINEKEN
is based on the six big losses of Nakajima (1988). The performance is measured by the Operational
Performance Indiactor (OPI). OPI = Availability * Performance * Quality. All these performance
measures have the same weight.

We use simulation in order to determine a solution for the problem statement. The discrete-event
simulation fits the best with our parameters and therefore we use this. This is explained in more detail in
Section 5.3. Furthermore we use accumulating simulation in order to model the Star Bottles.
Nevertheless, we considered the related research in order to simulate products on the conveyor system at
HEINEKEN. In the literature only single loads are discussed, as pallets, cases, automobiles etc. No
literature is written about how to simulate multiple products in a sequence on a conveyor system in
discrete-event simulation. In this paper we describe how to perform this.

42
 5. Solution design
In Section 5.1 we describe the conceptual model. In Section 5.2 we describe the simulation model and
how we transformed the conceptual model. In Section 5.3 we describe the experimental setup in order to
determine the input data, warm-up period and the number of replications. In Section 5.4 we explain the
verification and validation of the model. Furthermore we describe the experimental design in Section 5.5.
At the end a conclusion is made, based on the whole chapter.

5.1 Conceptual model

In this section we describe what the model is and how it is functions.

Model overview – Movement of Star Bottles

As described in the previous section, the model in a simulation study is a simplification of the reality.
This model overview is intended to explain this simplification compared to the real-life situation of the
SB production line.

The model is divided into several lines that depict the conveyors. The length and width of the lines are
developed on scale. These lines transport the SBs to the CPL machine.

This model focuses on the behavior of the bottles when changing from conveyor part. In real life the SBs
are positioned in multiple rows next to each other. This makes it different compared to an assembly line,
where products are positioned in a single line.
See Figure 5.1 ‘Real life’ for the difference
between the SB line and an assembly line. The
green circles are SBs and the red squares are
example products, which we need to compare
the situation. These are generally large
products and transferring in single rows. For
the simulation study this difference is
important, because SBs cannot be modeled in
multiple rows next to each other. To simulate
the movement of each different SB across the
line we choose to split the line in several
conveyors. The model conveyor SB production
line, separated in three conveyors, is shown in
Figure 5.1 ‘Model’.

Another aspect that needs attention is the

sensors positioned on the production line. The
specific working of a sensor was discussed in
Section 2.1.2. In real life sensors are triggered F IGURE 5.1: CONVEYOR BELT - DIFFERENCES
when bottles hit the sensor. This only occurs REAL LIFE AND SIMULATION
when a conveyor is occupied. In real life, the
sensor is placed vertically at one side of the conveyor as shown in Figure 5.1 ‘Real life’ with the red line.
In the model, a sensor could only be placed horizontally confiscating a total line, shown in Figure 5.1
‘Model’ again with the red line. This sensor is triggered every time when a single bottle passes. This

43
sensor is denoted by the horizontal red line. In order to prevent that a sensor is triggered by every bottle,
first the conveyors are divided into multiple parts. Now consider the upper left conveyor belt, with the red
line. Secondly, when a SB enters this conveyor a method is triggered which can determine how much SBs
there are currently on the other conveyor. The capacity is
known and therefore the occupancy can be determined.
When this conveyor in the model is occupied, the sensor is
triggered, just like in real life.

As described in the previous section, the model should

determine the production balance between the CPLs.
Therefore it is important how the distribution of bottles is
modeled. Consider Figure 5.2, where the movement of SBs
is shown in the conceptual model. This conveyor consists of
a bend, which can be compared with part I in Figure 3.10
and Figure 3.9. Section 3.2 also considers this bend, but then
in real life. The conclusion from this section was that SB
F IGURE 5.2: BEHAVIOR OF A SB IN A BEND
will move towards the outside of the bend.

Figure 5.2 consists of three line conveyors 1,2 and 3. These conveyors are separated into three parts, A, B
and C. In this figure we consider a red SB that moves towards the outside of the bend, which happens in
real life. Considering the conceptual model, we see that the SB moves from 2A  2B  1B  1C.
Therefore the conceptual model should take into account the distribution of the SBs between conveyors.
Furthermore the possible successors of a SB should be determined. For example, it is not possible for a
SB to move from 1A to 3B, if a bend ‘turns’ right.

The destination table determines the behavior of a SB. We make a distinction between straight lines and
bended lines. For the explanation of the destination table at input data, we only consider a bended line.
Bottles on a bended line have a tendency to move towards the outer of the bend. As mentioned before we
consider this as a deterministic process, which is modeled with priorities. Figure 5.2 explains the behavior
of a bottle. A SB actually wants to move towards the outside of the bend. These priorities are determined
in a destination table.

In order to determine possible successors we consider Figure

5.3. In the conceptual model there are four possibilities where
the SB can flow after triggering a method at the ‘end of the
line’. In situation A we can see the different possibilities,
shown with numbers. Consider the red SB (with number 4 in it).
This SB has four succeeding options; number 1, 2, 3 and 4. The
SB can move to three positions: 1, 2 or 3. Number 4 means that
the SB stays on the same position. This can only occur when 1,
2 and 3 are not available. Note that at the right side of number 3
is also space, but it is not a possible successor. It is not realistic
to move to this position, because the distance is too large. So,
determining the possible successors is the first step in the F IGURE 5.3: P OSSIBLE SUCCESSORS
conceptual model. FOR A MU

44
The second step is determining the occupancy of the first position. The first position is the one with the
orange circles (this means that in this example, the lines have three positions). In situation B we can see
that the middle conveyor is occupied, and that it is not a possible successor anymore. Determining the
occupancy of the first positions is the second step in the conceptual model. This is done by determining
the capacity of a line and counting the current amount of SBs on this line. If this is equal, the succeeding
conveyor is occupied.

The third step is prioritizing the possible successors. We consider two different scenarios, a straight line
and a bended line. In a straight line the SB will, when possible, move in a straight line. In Figure 5.3 this
means that the red SB (with #4) will move to number 2 with priority 1. When situation B occurs the SB
will move to position 1 or 2. These positions have the same priority in a straight line and therefore the
chance of moving to these positions is random. Considering a bended line the chances are deterministic.
As explained in Figure 5.2, the SBs will move to the outside of the bend. This means that considering the
SB in Figure 5.2, it will move eventually to position 1C with priority 1, to 2C with priority 2 and to 3C
with priority 3. These priorities are determined beforehand, and are input data to this conceptual model.
Prioritizing the possible successors is the third step in this conceptual model.

When there is a possible successor that has a free first position, the SB moves to this position. If there is
no possible successor and option 4 is chosen, the SB is placed on the blocking list. As the name reveals, it
is a list of all blocked SBs, which are ‘waiting’ to move to a possible successor.

The sensor that triggers the blocking list is placed on the succeeding
production line. Thus, when a first position becomes empty, a method
checks whether there are SBs on the blocking list. When there are SBs on
the blocking list, we must check if the SB in the list is allowed to fill the
first position, as described in Figure 5.3. When this is the case, pick the SB
which is ranked highest in the blocking list (longest waiting), and delete
this SB from the blocking list. This means that a SB from the blockings list
has preference above a part that triggers the end of a line, and wants to
move directly from a conveyor.

When there are SBs on the blocking list, but when a SB is not permitted to F IGURE 5.4: SBS IN
transfer to the first position, the neighbor lines should be considered. BLOCKING LIST

Consider Figure 5.4, the red SB (4) is located in the blocking list. A
succeeding line triggers a sensor (red line) and therefore a method of the conceptual model. However, the
orange (first) position is not a possible successor of the red SBs. In real life the conveyors will be
constantly flowing, and therefore the SB with number 3 eventually moves to the orange position. This is
also taken into account in the conceptual model. The model includes a method that compares the amount
of SB on the line of the neighbor, if no SBs are available on the blocking list. When the amount of SBs on
the neighbor line is more than 2, it takes the last SB of the line. In Figure 5.4 the succeeding line has only
one neighbor, and the amount of SB on the line next to the orange circles is more than 2, so it moves SB
#3 to the orange circle. Then it is possible for the red SB to move to position 3.

Summary

In order to summarize the previous steps, we created two flowcharts. In Figure 5.5 we describe the
flowchart of a moving a SB over the lines.

45
F IGURE 5.5: F LOWCHART CONCEPTUAL MODEL 1A - MOVING SB FORWARD
This blocking list is triggered by another part of the model, which is described in Figure 5.6.

F IGURE 5.6: F LOWCHART CONCEPTUAL MODEL 1B -

TAKE SB FROM BLOCKING LIST

Model overview – Regulation

To create a better understanding of the scale of conceptual
model part 1 we consider Figure 5.7. Here we zoom in into
part H, and consider Figure 5.4 again. Note that part H in the
conceptual model consists of eight lines, but in this example it
is simplified to four lines. As mentioned in Section 3.2, there is
a sensor (sensor 10) located at part H/I. This specific sensor
ensures that CPL112 will start producing when triggered, and
stops producing when it is not triggered anymore.

In the conceptual model, the conveyor line with the orange

circles is the one where the sensor is located. Due to the
separated conveyor lines, we can easily model that CPL112
should start producing by counting the amount of SBs on F IGURE 5.7: REGULATION IN CONCEPTUAL MODEL
the conveyor with the sensor. In this example there are

46
three possible positions on the conveyor. This means that when all three positions are occupied, the sensor
should be triggered. Therefore it is modeled that when the number of SBs on the conveyor with the sensor
is equal to three, the processing time should go to nominal speed. If all three positions are empty for 30
seconds, CPL112 will stop. This modeling is done for all relevant sensors explained in Section 3.2.

Furthermore, the conceptual model works with aggregated sizes of SBs. In real life every hour there are
entering about 70.000 SBs and staying in the system for several minutes. It costs a lot of processing time
in order to mimic the real life situation, and for that reason the conceptual model uses aggregated size of
1:100. Thus, 1 SB in the conceptual model represents 100 SBs in real life.

Components of simulation model

We use several components for the simulation model. These five main components are:
- Input data. This is data that will not be changed during the experiments. It is implemented once, and
will not be influenced.
- Stochastic variables. The values of these variables are subject to variations due to chance (e.g., a
machine failure).
- Experimental factors. These are controllable variables, set by the experimenter and can be different per
experiment.
- Output data. This data results from a run of an experiment. It is influenced by the stochastic variables
and experimental factors.
We use the input data with the stochastic variables and the experimental factors which results in the
output data.

Input data
The input data we use for our model is:

 Process times. The time it takes for a machine to perform an activity (e.g., filling or labeling).
 Aggregated size of Star Bottles. In the conceptual model the SBs are aggregated to a ratio 1:100
(in real life 1 bottle equals 100 bottles in the conceptual model).
 Speed per conveyor.
 Length per conveyor.
 Strokes per conveyor (this determines the width of a conveyor).
 Buffer capacity per line.
 Machine failures

Stochastic variables
The stochastic variables of the model are:

 Machine availability. In the conceptual model the machine availability includes the machines that
are not modeled (fillers & packer) but have an influence on the machines described in the
conceptual model. For example, the impact of starvation, blockage and failures of machine
outside our scope on the machines modeled. It also includes the breakdowns of the machines
modeled.
 Arrival times. Arrival times of the Star Bottles are dependent on the machines availabilities of the
pasteurizer and preceding machines. These machines are analyzed and this stochastic variable is
used in order to ‘create’ new bottles.

47
Experimental factors
The experimental factors of the model are:

 Speed level of the machine. This will be the processing time of a machine and the level of the
speed might differ per experiment.
 Number of speed levels. Some machines have multiple speed levels. The amount of speed levels
might differ per experiment.
 Position of sensors/ moment of switching speed of CPLs. Machines are dependent of the sensors
when they will change to a certain speed. This might differ per experiment.
 Position of buffers. The position of the buffer determines the effect on the pasteurizer.

Output data/ performance indicators

To measure output data we use five performance indicators. These indicators are:

 Production/ Output quantity. The total quantity that is produced on both CPLs.
 Production balance. The production balance between the CPLs.
 Time of machine failure. The failure time of the CPLs.
 Time of machine stopping of a machine. The stopping time of the CPLs is dependent on the
position of the sensors. When sensors are positioned inefficiently the stopping time will increase.
 Time of waiting of a machine. The waiting time of the machine is dependent on the output of the
pasteurizer. This is also the starvation time of the CPLs.
These five performance indicators determine the performance per experiment. The weights of the
performance measures are described in Section 6.1.

Assumptions
We made several assumptions because it is almost impossible to approximate a real life situation. We will
formulate the assumptions that we made below:

 No bottles will collapse on the production line. This means that we create no losses due to bad
quality of the material.
 Bottles react the same on an identical situation.
 Processing times of machines have fixed values.
 The average mean time to failure and mean time to repair of the last year is representative.
Therefore we can use this as input data for the model.
 The lines/conveyors will not fail/ have breakdowns. The machines breakdowns are included, but
the conveyor breakdowns are excluded. In real life these are negligible.
 No maintenance activities have to be done.
 Extra material is available and setup times are zero. For some machine extra material should be
present in order to fulfill the machines’ activity (e.g., the labeling machine needs labels).

Conclusions
The conceptual model is a simplification of reality and contains two parts, 1A and 1B. Part 1A consists of
three steps:

- First the possible successors should be determined

- Secondly the occupancy of the possible successors should be determined
- The third step is to prioritize the possible successors.

48
When all these steps are performed a MU moves towards the next positions, or it is moved to the blocking
list. Subsequently we consider part 1B which continues from the blocking list. This part:

- Check whether there are SB on the blocking list

- If this SB matches a possible position
- And if there is no SB on the blocking list, it checks neighbor lines.
Furthermore, this sections described the components needed for a simulation model and the assumptions
made during this conceptual model.

5.2 Simulation model

We choose to work with Siemens’ Tecnomatix Plant Simulation. Tecnomatix Plant Simulation is a
simulation tool to create digital models of logistical problems (e.g., production systems), in order to
examine the system characteristics and optimize performance (Siemens, Tecnomatix Plant Simulation,
2012). Furthermore, Tecnomatix Plant Simulation is used by the University of Twente in their simulation
coursed

The representation of the different flows should correspond with reality. Some assumptions are made,
which are described at the end of the previous section. We use the simulation in order to find optimal
locations for the sensors with ideal speed levels for the CPLs. Figure 5.8 shows a print screen of the
simulation model from Tecnomatix Plant Simulation.

F IGURE 5.8: P RINT SCREEN OF MAIN FRAME PLANT SIMULATION

49
 In this main layer we consider different frames and the flow line of bottles. A frame consists of data
and/or methods for a specific area. A detailed explanation of the flow line is given in Appendix A.
Besides the Layout of the Pasteurizer-CPLs, there are five different frames which are:
- Event Control - Counters
- Methods & Data - Performance Measurement
- Experimental Factors

Description of the simulation model

Here we explain the different frames that are used during the simulation study as mentioned before:

Layout Pasteurizer-CPLs
The layout consists of all conveyors that are positioned between the pasteurizer and the CPLs. Differences
between conveyors are the straight and bended lines, as explained in Section 5.2. The characteristics of
the conveyors are the same as reality. This means that the length, wide and speed of the conveyors are
similar. Furthermore the aspects discussed in Section 5.2 are considered.

Event Control
The event control regulates the simulation from initializing till resetting. The generator triggers a method
that ensures that different experiments and shifts will run.

Methods & Data

Methods are indicated with the large blue Ms. In these methods we program different codes in order to
run the model. The aspects discussed in model overview are written into these methods, such as the
behavior of SBs and the regulation of the CPLs.

Experimental factors
The experimental factors determine the different experiments that will be done. For example the different
locations of sensors can be implemented in the experimental factors.

Counters
Counters easily serve for determining the amount of SBs produced and the number of batches produced.
In this way it is clearly visible during a run of an experiment.

Performance measures
The frame with performance measures contains all table files. A table file is a table in Tecnomatix Plant
Simulation that collects data from the experiments. Performance is measured using the data from these
table files. In the table files the producing times, waiting times, failures is measured. Furthermore the
output of the process with the production balance is stored into the table files.

50
 5.3 Experimental setup
We described the conceptual model and the simulation model. Before we can run our model we have to
determine the input data, the warm-up period and the number of replications. This is covered in this
section.

Simulation type
Several types of simulation models are defined by Law (2006), as we explained in Section 4.1. In our
simulation model, the stochastic variables indicate that the system evolves over time and the state of the
system changes, which means that we should use random input components. Our model is a discrete
event simulation, because the state of the system only changes when an event occurs. Furthermore, the
processing times of the SBs are stochastic, because breakdowns on machines can occur. Taken the
different dimensions into account and the reason that we deal with a complex system, the conclusion is
that we use a discrete-event simulation.

Input data
There are three types that we have to define in our simulation model.

 Processing times. Time that a machine needs to produce a SB.

 Mean Time Between Failures (MTBF). This is the mean time it takes between machine
failures.
 Mean Time To Repair (MTTR). Time it takes for repairing a machine after it failed.

In order for the simulation model closely mimic to reality we try to define these three types with a
theoretical distribution. If this is not possible, the empirical distribution function will be used.
Considering a theoretical distribution function, several distributions in Tecnomatix Plant Simulation can
be used. These distributions are described in Appendix A. From this appendix we can conclude that we
will use both a Chi-squared test, as a Kolmogorov-Smirnov test to check if the data fits and theoretical
distribution.

Furthermore, the three types discussed need to be determined for the machines that are available in our
simulation model. These machines are:

 CPL111
 CPL112

Besides these functions and distributions we need a deterministic destination input table, to determine the
behavior of the SBs. As mentioned in Section 5.2, this is different for straight lines and bended lines.

In the following sub-sections the processing times, MTTF, MTTR and destination table are determined.

Processing times
The processing times of the Pasteurizer and CPLs are not registered at MES. This means that there is no
data available to determine a theoretical distribution. An empirical distribution is time consuming to
measure and validate. Therefore we determine the processing times on the set points of the machine
speed. As one can see in the Figure 5.8, the pasteurizer has two sources, one for the upper deck and one

51
for the lower deck. The lower deck is the left side of the lines from the pasteurizer towards part I and the
upper deck is the right side. As described in Section 3.2, the lower deck is always filled with SBs, due to
the failure mode of the fillers. Therefore, we can state that the source of the lower deck produces more
SBs compared to the upper deck. In order to determine the processing time of sources, we have done a
test on the pasteurizer. This test determines the output of the pasteurizer measured over both sources, and
was described in Section 3.2. The SBs were counted with a production counter. The partition of the deck
was as follows:
- Lower deck: 39138 bottles per hour.
- Upper deck: 36257 bottles per hour.

The difference between the lower and the upper deck is 7,4%. This means that the upper deck should
produce 7,4% less than the lower deck. The source at the upper deck has therefore a failure rate of 7,4%.
This should make our model more realistic. In order to create relevant results, we make this a constant
failure rate, so it would not influence our results that much. This means that the upper deck has a
availability of 92,6% and we entered a MTTR of 1 minute. Therefore 92,6% of the total time, the upper
deck has SBs at in feed.

The machine speeds of the CPLs are shown in Figure 5.9. Note that the speeds should be multiplied with
factor 10. The use of batches requires that the machine speeds have the current setting.

F IGURE 5.9: MACHINE SPEEDS /P ROCESSING TIMES

MTBF
To calculate the MTBF we use operating-dependent failures, this means that a machine can only have a
breakdown when it is in operation. To determine the mean time between failures we will look at the
production time between to machine failures, which means that we exclude starvation and blocking
periods. In the information system MES, which is explained in Chapter 2, the information is stored. From
the data in MES, we try to find a theoretical distribution. In Appendix B this approach is described.

Nevertheless we can conclude that for the machines there is no theoretical distribution that fits with the
data from the process. Therefore we use an empirical distribution which is also explained in Appendix B.

52
MTTR
In order to calculate the MTTR we first have to consider the duration length of a machine failure. In MES
there is a distinction between short (< 5 minutes) and long (>= 5 minutes) failures. The conclusion from a
discussion with the production manager and operators is that short failures can be considered as incidents.
These are for example fallen bottles and are not part of a pattern in the duration of the failure mode. For
that reason we only consider the long failures. Appendix C describes the result of the analysis for
determining a theoretical distribution of the MTTR in more detail. Regarding this appendix we can
conclude that of the MTTR there is a theoretical distribution that fits the data from the process, namely
the Weibull distribution. In Table 5.1, one can find the parameters of the Weibull distribution of both
CPLs.

Distribution Parameters

CPL111 Weibull α = 0.83029 β = 36.428

CPL112 Weibull α = 0.78302 β = 28.755

TABLE 5.1 – DISTRIBUTIONS WITH CORRESPONDING PARAMETERS - MTTR

Destination Table
In order to deliver these priorities as input to our simulation model, we developed a destination table. An
example of a destination table in Plant Simulation is shown is Figure 5.10.

F IGURE 5.10: DESTINATION TABLE PART I OF LAYOUT

This destination table shows the priorities from the first layer at part I to the second layer at part I.

Warm-up period
We use initial conditions for the start-up and therefore the steady state behavior is not directly reached.
The warm-up period is the period before the steady state is reached. The warm-up period should be
determined in order to neglect warm up ‘problems’. To enter the steady state in our system, the first SBs
should exit the system. This takes in our system 6 minutes for the CPL112 and 8 minutes for the CPL111,
which is negligible. Therefore we do not use a warm-up period.

53
Number of replications
We have to determine the number of replications that should be done within an experiment. According
Law (2006) we can use the replication-deletion method in order to determine the number of replications.
By means of the number of replications we guarantee that the 95% confidence interval has a width of at
most 5% of the mean. We use the following formula to compute the required precision: γ’ = . Where

γ’ is the required precision and has a value of: γ’ = = 0.04619. If the precision is not sufficient,
another replication is executed in order to decrease the confidence interval half width until the required
precision is achieved. In Table 5.2 the computation of the number of replications is shown. The required
precision is achieved in replication 5 where the width of the confidence interval is lower than the relative
error. This means that we will run 5 replications per experiment.

Replication Data Average Variance T-value Relative Confidence

error interval
width

1 461600 461600 0,047619

2 427580 444590 578680200 12,7062 0,047619 0,48613901
3 448250 445810 293805300 4,302653 0,047619 0,095511502
4 435920 443337,5 220323225 3,182446 0,047619 0,053275405
5 423330 439336 245302430 2,776445 0,047619 0,044264771
6 451950 441438,3 222760777 2,570582 0,047619 0,035481758
7 402980 435944,3 396925895 2,446912 0,047619 0,042266183
8 468830 440055 475405971 2,364624 0,047619 0,041423091
9 401590 435781,1 580375361 2,306004 0,047619 0,042493735
10 461570 438360 582395889 2,262157 0,047619 0,039382294
11 456620 440020 554467900 2,228139 0,047619 0,03595106
12 454830 441254,2 522339736 2,200985 0,047619 0,032908964
13 459530 442660 504504200 2,178813 0,047619 0,030662698
14 411750 440452,1 533941049 2,160369 0,047619 0,030290872
15 466360 442179,3 540550207 2,144787 0,047619 0,029117766
5.2: NUMBER OF REPLICATIONS

54
 5.4 Verification & validation
Verification and validation is an important part of our simulation study. A verified and validated model
means that we can use our model to run experiments, and assures that the model mimics a real life
situation. Below we discuss briefly the meaning and importance of verification and validation.

Verification
Verification is the process to ensure that the simulation model is correctly implemented with respect to
the conceptual model. Verification is practically seen as debugging the simulation model, which can be
done with several tools. The tracing function within simulation is very useful, and mostly used in
Tecnomatix Plant Simulation. There is a possibility to use ‘breakpoints’ which in order to go through a
method, step by step. In this way, different parameters can be controlled which contributes to the
verification of the model. This function helps to debug the code and fix programming mistakes.

Another verification tool within Tecnomatix Plant Simulation is the animation. By running experiments,
there is a possibility to have a look at the Movable Units. This animation makes us aware if MUs stuck on
a certain conveyor. When this is the case, we know that there is a bug in our model otherwise the MUs
will move to the next conveyor.

The last possibility to check the verification of our model is to compare the output of the model with the
input, which should be equal if no MUs remain in the system. To further verify the simulation model we
check how this aspect sustains regarding the output, when we change the input variables, e.g.,
distributions and processing times. We used all of these verification tools in order to check our simulation
model.

Validation
When the verification of the model is done, we Output = (crates Real life Simulation
continue with the validation. Validation of the * # of btls. in model - #
model measures the accuracy of the simulation crate) of SBs
model, with reality. There are several options
to measure validation. Sargent(2005) states CPL111 Output (18109 * 24 ) * 253100
that a possibility to measure the validity of the 0,55= 239038 btls.
model, is to determine the output quantity of btls.
the MUs. Furthermore the lead time is
important in order to validate our model. In CPL112 Output (18109 * 24 ) * 206990
order to check the lead time we consider the 0,45= 195577 btls.
output quantities over a time period of 8 hours. btls.
The output of our simulation model compared
with the output in real life is shown in Table 5.3. TABLE 5.3 - VALIDATION OF SIMULATION MODEL
Furthermore the production balance is checked
in order to validate the model. The production balance consists of the production quantity between the
CPLs. CPL111 produces in real life 55% of the total output and CPL112 produces 45%. As shown in
Table 5.3, the difference between real life and the simulation model is sufficiently small. Even note that
the production balance in our simulation model is 55,15% (CPL111) against 44,85% (CPL112).

55
In Chapter 7 we discuss the implementation of an alternative solution. The results after the
implementation confirm the validation of our model. The results of the implementation are discussed in
Section 7.2. Also the real life output before and after the implementation is compared, which supports the
validation in this section.

5.5 Experimental design

In Section 5.2 we mentioned that there are several experimental factors for the configuration of the
experiments. The main experimental factors are the location of the sensors and the number of speed levels
of the CPLs. When considering all sensors and a lot of number of speed levels, the number of experiments
will be too high which takes too much calculation time. Therefore, it is necessary to make a well-advised
decision about which factor should be taken into account. The rayon manager and installation
administrator are asked for advice to make that decision, after the improvement team proposed some
recommendations.

The first decision is that the moment of switching of the CPLs should be programmed on current
locations of the sensors. This means that it is not possible to choose random positions to switch to another
speed of the CPLs. The current positions of the sensors should be used. This makes a potential solution
less expensive because sensors do not have to be displaced.

Furthermore the different speed levels of the machines will be three or four. CPL112 has at the current
situation no low speed and therefore it has only three speed levels. CPL111 has four speed levels, which
are:
1. Down.
2. Low.
3. Nominal.
4. High.
In the past, multiple improvement teams have determined the ideal amount of speed levels. These differ
per machine, but labelers at other production lines at HEINEKEN vary at three or four levels. To reduce
the amount of experiments we agree with three or four levels. Thus at CPL112 we should add a position
of a sensor and at CPL111 we should skip one sensor in a certain experiment.

The improvement team made some recommendations to the management, in order to reduce the amount
of experiments. In these recommendations they have taken the problem definition and KPIs into account.
These were an optimal production balance, and a high output of the CPLs. The conclusions are some
specific possible positions of sensors. These positions will be taken into account in order to reduce the
amount of experiments. There are 17 sensors where the speed of CPLs can be regulated with, as described
in Section 3.2. The current layout of the CPLs concerning the change of speed is shown again in
Appendix F and Appendix G. Trying to combine and change all these sensors together with the different
speed levels will not be realistic for the experiment. Therefore we will use the sensors that are
recommended by the improvement team and approved by the management. These are the most relevant
scenarios for our simulation study.

Possible options for our simulation study are shown in Figure 5.11. The sensors colored green will be
neglected in our simulation study for several reasons. Sensors 1 till 7 are too close to the pasteurizer, and
are used to determine the speed of the pasteurizer. When we use these sensors for changing the speed of
the CPLs, the risk that the pasteurizer will create blockage will increase significantly. This has a reverse

56
result on the desired situation. Sensor 9 will not
be used because we use sensor 8. As described
in Section 5.4 the lower deck from the
pasteurizer is always filled. Skipping the use of
sensor 9 will decrease the amount of
experiments, without having any influence on
the outcome. Sensors 15 and 16 will be
neglected because these sensors serve for a
security. When these are not triggered the line
has an emergency shutdown. If not, the CPLs
will be damaged. When there are no SBs, the
labels stick in the machine. Sensor 11 is
neglected because, according the improvement
team, this has little value when also sensor 13,
14 and 17 are regulating CPL111. Sensor 11
regulates conveyor K, and therefore it is
positioned at that location. The colors in Figure
5.11 mean that these will change over the
experiments. Sensor 12 (yellow) and sensor
(17) are only considered that the CPLs have a
low speed or will not (on/off).

F IGURE 5.11: P OTENTIAL SENSORS FOR

CHANGING SPEED OF CPL S IN
EXPERIMENTS

Besides this figure one should consider Table

5.4. In this table the experimental changes are shown. These are beforehand discussed with the
management as mentioned earlier. One remark on this table is, that when a sensor of a higher speed is
triggered, the sensor of the lower speed is overruled. For example when in the current situation sensor 13
is triggered, so CPL111 changes to high speed , then sensor 14 is overruled until sensor 13 is not
triggered anymore.

As we can see in Table 5.4, we have 4 different factors which have two different speed levels. No low
speed means that the CPLs directly change to the nominal speed, so only three speed levels are available.
Thus, at the moment CPL112 has no low speed, and the alternative situations checks if is valuable to add
a low speed on the CPL112 on sensor 12. The colors are equal to those of Figure 5.11, so one can see
what is changing. This means that we have = 16 different experiments. This amount of experiments is
realistic. However, there is one combination that is not valid. It is not relevant to change from low speed
to high speed and then change to nominal speed. This situation would occur if sensor 10 and sensor 13
will be in an alternative solution. This combination will be excluded, and therefore we have only 12
different experiments. Table 5.5 shows the different experiments. Note that the first experiment is the
current situation.

57
 Current situation Alternative situation
Machine – ‘Change to’
Triggered Sensor Triggered Sensor

CPL112 – Low speed No low speed Sensor 12

CPL111 – Low speed Sensor 17 No low speed

CPL111 – Nominal speed Sensor 14 Sensor 10

CPL111 – High speed Sensor 13 Sensor 8

5.4: SPEED CHANGES DEPENDENT ON SENSORS - CURRENT AND ALTERNATIVE SITUATION

The speed changes from Table 5.4 are translated into the letters in Table 5.5. The letters correspond with
the conveyors given in Figure 5.11. These are explained below:

NOSPEED : No low speed

O4 : Sensor 17
M4 : Sensor 14
L111 : Sensor 13
I8 : Sensor 10
E51 : Sensor 8
J4 : Sensor 12

The columns of Table 5.5 consist of changing the speed of the machine indicated. First column therefore
means: changing low speed of CPL112.

CPL112 <> low CPL111 <> low CPL111 <> nominal CPL111 <> high
speed speed speed speed
Experiment 1 NOSPEED O4 M4 L111
Experiment 2 NOSPEED O4 M4 E51
Experiment 3 NOSPEED O4 I8 E51
Experiment 4 NOSPEED NOSPEED M4 L111
Experiment 5 NOSPEED NOSPEED M4 E51
Experiment 6 NOSPEED NOSPEED I8 E51
Experiment 7 J4 O4 M4 L111
Experiment 8 J4 O4 M4 E51
Experiment 9 J4 O4 I8 E51
Experiment 10 J4 NOSPEED M4 L111
Experiment 11 J4 NOSPEED M4 E51
Experiment 12 J4 NOSPEED I8 E51
TABLE 5.5: EXPERIMENTS

58
Running all these experiments takes s certain period. In order to calculate how long it takes to run all
experiments we determine the total run time. The total run time of all experiments is, 2.5 hours
(= ).

5.6 Conclusion
We started this chapter with the explanation of two possible methods to analyze our problem. Then we
described our conceptual model and explained the simplifications from reality. We used this conceptual
model to build our simulation model. We looked at the processing times, MTBF, MTTR and the
destination table. Furthermore we determined the warm-up period and the number of replications in order
to run our experiments. After designing this model, we verified and validated the simulation model. We
can conclude that our simulation model is valid. The experimental factors of the simulation model are the
positions of the sensors and the number of speed levels. This covers the last section of this chapter. We
described why we have chosen for the different experiments.

59
 6. Experimental results
In this chapter, we discuss the results that are retrieved from our simulation model. In Section 6.1, we first
introduce how we compute the scores of the experiments. We use different performance measures and
therefore we have to determine the weight. In Section 6.2 we describe the results of the experiments and
consider alternatives to improve the current situation. In Section 6.3 we make a risk analysis in order to
assess the new settings in real life. In Section 6.4 we will write a conclusion of the experiments.

6.1 Performance measures

First we will describe the different performance measures. In Section 5.2 the different performance
measures are mentioned, but one can find them below:

 Production/ Output quantity. The total quantity that is produced with both CPLs, which consists
of the output of our model.
 Production balance. The production balance between the CPLs.
 Time of machine failure. The failure time of the CPLs.
 Stopping time of a machine. The stopping time of the CPLs is dependent on the position of the
sensors. When sensors are positioned inefficiently, the stopping time will increase.
 Time of waiting of a machine. The waiting time of the machine is dependent on the output of the
pasteurizer. The waiting time increases when the upper deck of the pasteurizer has a higher
failure rate.
The stopping time together with the waiting time is the starvation time of a CPL. It means that the
efficiency of the positions of the sensor together with the failure rate of the upper deck of the pasteurizer,
determine the starvation time. In this research we measure the stopping and waiting time separate, but we
look at the total starvation time of the CPLs to determine the results. The blockage time is neglected
because this state is for the conveyors after the CPLs, which is not part of this research.

In consultation with the management, there is decided that the output quantity is by far the most important
performance measure. The output quantity of the production line at HEINEKEN is used in order to
measure the line performance. Nevertheless the production balance is also very important to reduce the
start/stop situations, as discussed in Section 3.1. Start/stop situations will increase the failure rate of a
machine, which is the philosophy of the supplier and is confirmed by the management of HEINEKEN.
The combination of start/stop situations and the failure rates is not included in our simulation model, but
should be included to determine the best alternative solution. Therefore is the production balance an
important measure to compare the alternatives. According the management at HEINEKEN these two
performance measures are far more important than the other three performance measures, mentioned in
the bullet points. We could use a graph to get a first impression of the results. Thereafter we could use
weights to include the last three bullet points. The combination of these tools will underpin our choice for
the best solution. First we discuss the results of the experiments, which is done in the next section.

60
 6.2 Simulation results
This section covers the simulation results from the experiments. As mentioned in Section 5.6 the first
experiment contains the current situation. Therefore we start with the results of the first experiment, then
we describe the rest of the experiments.

Current situation
The total average output of experiment 1 is 441313 bottles per shift, with a production balance of 57% on
CPL111 and 43% on CPL112. The other performance measures are shown below in Table 6.1.

CPL111 CPL112
Waiting 0,78% 38,08%
Stopping 28,99% 0,00%
Failure 2,22% 0,85%
6.1: RESULTS EXPERIMENT 1 CONTINUED
As mentioned in the previous section, the waiting time + the stopping time is the starvation time. We can
conclude that the starvation time of CPL111 (29,77%) is less than on the CPL112(38,08%).

All experiments
As mentioned in Section 6.1 we use the two most important performance measures, in order to get a first
impression of the best alternative. In Figure 6.1 we show all the experiments in a graph, with on the X-
axis the output quantity and on the Y-axis the production balance. The experiment which lies the closest
to the 50% (marked with the red line) and the closest to the 460,000 is the best option. Experiments
located above the red line have more SBs produced on the CPL111 than the CPL112, and for experiments
below the red line it is the reverse. Consider Figure 6.1. Option 1 is the current situation and option 6 is
the best alternative. Alternative 6 scores the best on both performance measures. The second best will be
10 or 12, depending on the weight of the performance measure as described in the previous section.

Results Experiments
70%
7
60% 4 5 10
Production Balance

50%
11 1 12 6

40%
9
30% 2
Experiments
8
20% 3
10%
0%
360000 380000 400000 420000 440000 460000
Output Quantity (# of bottles)

F IGURE 6.1: RESULTS OF EXPERIMENTS REGARDING OUTPUT AND BALANCE

61
In Table 6.2 all the results from experiments 1 till 12 are shown. The stopping time and waiting time are
combined as the starvation time. In Appendix H the whole table, including the stopping time and waiting
time is shown. Looking at the starvation time and the failure time we see that alternative 6 is the best
solution. The second best solution is alternative 10 or alternative 12. As mentioned earlier, the
management’s main target is to increase the output quantity and then option 10 would be preferable. Also
the starvation percentage of experiment 10 is lower which means that this performance measure is also
better in experiment 10.

As one can see in Figure 6.1: Results of experiments regarding output and balance, experiments 8 and 3
have a lower output quantity compared with the other experiments. When all buffers are completely filled
with star bottles, the source will stop producing. This means that when sensors are positioned inefficiently
and therefore the output reduces, a source block will occur. The source block is in line with the
Uniformity Principle (Kwo, 1958) explained in Section 4.4.This principle states that the input quantity
equals the output quantity.

Experiment Output Production Starvation Failure

(# of balance
bottles)

Average CPL111 CPL112 CPL111 CPL112 Total CPL111 CPL112

1 441313 57% 43% 29,77% 38,08% 67,85% 2,22% 0,85%

2 416625 29% 71% 67,77% 9,51% 77,28% 0,43% 1,33%

3 388495 19% 81% 69,40% 7,03% 76,42% 1,03% 0,24%

4 435440 58% 42% 1,72% 39,03% 40,75% 1,65% 0,54%

5 444508 57% 43% 0,82% 38,79% 39,61% 1,20% 0,18%

6 453103 53% 47% 0,01% 30,61% 30,62% 1,42% 0,04%

7 439100 62% 38% 24,65% 48,17% 72,82% 0,84% 0,13%

8 379278 23% 77% 76,67% 10,80% 87,47% 0,46% 1,36%

9 408198 31% 69% 66,44% 13,59% 80,03% 0,39% 1,06%

10 449990 58% 42% 2,90% 28,48% 31,38% 1,99% 1,03%

11 430915 57% 43% 0,78% 37,09% 37,86% 2,83% 0,86%

12 444338 54% 46% 0,08% 33,84% 33,92% 0,78% 0,80%

6.2: RESULTS EXPERIMENTS 1 TILL 12

62
Experiment Output Production Waiting Stopping
balance
Average CPL111 CPL112 CPL111 CPL112 CPL111 CPL112
1 441313 57% 43% 0,78 38,08 28,99 0,00
2 416625 29% 71% 0,05 9,51 67,71 0,00
3 388495 19% 81% 0,00 7,03 69,40 0,00
4 435440 58% 42% 1,72 39,03 0,00 0,00
5 444508 57% 43% 0,82 38,79 0,00 0,00
6 453103 53% 47% 0,01 30,61 0,00 0,00
7 439100 62% 38% 2,45 0,00 22,20 48,17
8 379278 23% 77% 0,03 1,39 76,64 9,41
9 408198 31% 69% 0,00 1,34 66,44 12,25
10 449990 58% 42% 2,90 8,77 0,00 19,71
11 430915 57% 43% 0,78 34,63 0,00 2,46
12 444338 54% 46% 0,08 33,33 0,00 0,51
6.3: EXPERIMENT 10 VS 12

Correlation
In Figure 6.2 we consider the results of the experiments again, but now we determine if there is a
correlation with the production balance and the output quantity. In first instance it seems that there is a
correlation between the performance measures. Nevertheless, there should be some correlation because
one CPL cannot produce more than 360,000 bottles (45,000btls/hr*8hr) bottles. Thus when the
production balance is out of proportion, the output quantity should be less than average.

Results of Experiments
70%
7
60% 4 5 10
Production Balance

11 1 12 6
50%
40%
9
30% 2 Experiments
8 Linear (Experiments)
20% 3
10%
0%
360000 380000 400000 420000 440000 460000
Output Quantity (# of bottles)

F IGURE 6.2: CORRELATION PRODUCTION BALANCE AND OUTPUT QUANTITY

63
When we consider all the experiments above the red line, which mean that the CPL111 produces more
than the CPL112, we see no clear correlation. Nevertheless we see that all the experiments close to the red
line have a higher output quantity. Overall this means that there is some correlation. Therefore we
conclude that, over all experiments, an equal production balance (50/50) increases the output quantity.
This means that we conclude that an equal production balance improves the output quantity, and therefore
the line performance.

In order to determine if there is a correlation between the starvation and output, we consider Figure 6.3.
100,00
90,00
Starvation percentage

80,00
70,00
60,00
50,00
40,00 Experiments
30,00 Linear (Experiments)
20,00
10,00
0,00
360000 380000 400000 420000 440000 460000
Output Quantity

F IGURE 6.3: CORRELATION STARVATION PERCENTAGE AND OUTPUT QUANTITY

In Figure 6.3, we weigh the starvation percentage with the output quantity. We can conclude that there is
a negative correlation between the two performance indicators. This means that when the starvation
percentage decreases, the output quantity increases. This is obvious because when a CPLs in starvation, it
cannot produce.

The next correlation that we check is the starvation percentage with the production balance. These
performance indicators are shown in Figure 6.4.

100,00
Starvation percentage

80,00

60,00

40,00
Experiments
20,00

0,00
0% 10% 20% 30% 40% 50% 60% 70%
Production balance

F IGURE 6.4: CORRELATION STARVATION PERCENTAGE AND PRODUCTION BALANCE

64
In this figure there is no obvious correlation between the starvation percentage and the production
balance. Nevertheless we see that the experiments with a production balance around the 50/50 (60/40)
have a lower starvation percentage. This conclusion fits the explanation in Section 3.2, where the bend
conveyor in part I is described. The conclusion from this section was that when a shift has a starvation
percentage above average, the CPL111 produced more bottles than CPL112. This was because the SBs
have the tendency to transfer to the outer of the bend. This matches with the results from our experiments.
Nevertheless, when considering the best experiments regarding the production balance, we see no
correlation with the starvation percentage.

Experiment CPL112 + low speed CPL111 - low speed CPL111 <> nominal speed CPL111 <> high speed
Current NOSPEED O4 M4 L11
6 NOSPEED NOSPEED I8 E51
10 J4 NOSPEED M4 L11
12 J4 NOSPEED I8 E51
6.4: SENSOR POSITIONS TOP 3 ALTERNATIVE SOLUTIONS
Remarkable on Table 6.4 is that experiment 10 is close to the current situation and experiment 6 and 12
are different in almost every setting. It is remarkable because this proves that the combination of sensors
is far more important the sensors itself. Furthermore we can conclude that the amount of speed levels at
CPL111 decreases at all the three alternative solutions. In experiment 10 and 12, the amount of speed
levels on the CPL112 increases to three.6.4: Sensor positions top 3 alternative solutions

6.3 Risk analysis

In our simulation model we assumed a constant input from the pasteurizer because these numbers are not
available at HEINEKEN. Nevertheless, it is important to have a critical view on the performance of the
pasteurizer when we want to implement our new experiment. When an experiment creates a higher risk
for the blockage of the pasteurizer for any reason, the experiment will not be implemented. Factors as
fallen bottles are not taken into account because these are very unpredictable, not measurable and hard to
include. The blockage of the pasteurizer should be measured and therefore we will compare the current
situation (experiment 1) with the best alternative solution (experiment 6).

When we implement experiment 6, this means that during a starvation on the CPLs, the buffer comes
closer to the pasteurizer. In Figure 6.5 the difference in buffer capacity between the two experiments is
shown. The left figure shows the current situation and the right figure shows the new alternative.

65
F IGURE 6.5: BUFFER P OSITION – CURRENT ( LEFT) AND NEW SITUATION ( RIGHT)
The first positive result from the change in buffer capacity is that the problem with the bend is solved. At
the current situation the problem arises that after a starvation all bottles move to CPL111 and assumed
was that this was the reason for a production imbalance. With the use of the buffer before the bifurcation
we solved this problem. Nevertheless, the change of buffer capacity also has negative effects. The buffer
capacity during a starvation in the new alternative is lower than the capacity in the current situation. This
means that the risk that the pasteurizer creates blockage increases when after a starvation period, a CPL
fails. Despite the fact that the blockage of the pasteurizer is included in our simulation model, we want to
exclude any contingencies.

Therefore we consider the capacity of the buffer in the new situation. In Figure 6.5 the difference in
buffer size is shown with the red part. The capacity of the red part is 2517 SBs. This means that in the
current setting, when the CPLs have starvation, CPL111 produces 2517 SBs more than CPL112. In
addition, in the current situation the CPL111 starts at high speed when CPL112 is still down. On average
this is 5 minutes, which means that another 3500 SBs are produced by CPL111 until CPL112 starts
producing. When we combine these SBs, we can conclude that by every starvation, CPL111 produces
6017(=2517+3500) SBs more than CPL112. The calculations of the buffer capacity are explained in more
detail in Appendix I.

66
F IGURE 6.6: BUFFER ENLARGEMENT - CURRENT VS NEW ALTERNATIVE
Considering the new alternative solution, both effects will be solved. In the new situation CPL111 and
CPL112 will start and end simultaneously on sensor 10. This means that the buffer stops before the
bifurcation. The negative result of the change in buffer capacity is, as mentioned before, that the risk
increases that the pasteurizer will create blockage. In order to consider this risk we consider Figure 6.7
The capacity from the pasteurizer till the new buffer location is 7576 SBs. Consider Table 6.5.

Speed per buffer Time to

pasteurizer fill with
SBs
Cap. Bottles minutes bottles minutes
Pasteurizer
Red part 100% 80000 60 2517 1.38
Figure 6.6
Blue part 100% 80000 60 7576 5.08
Figure 6.7
6.5: RISK ON PASTEURIZER IN MINUTES
As shown in this table, we see that it takes 1.38 minutes to fill the red part and 5.08 minutes to fill the
lines from the pasteurizer till the buffer (blue part).

67
In the current situation the buffer is located at 6.36minutes from the pasteurizer and the new situation the
buffer is 5.08minutes from the pasteurizer. This means that the risk on blockage at the pasteurizer is
increased with 1.28 minutes.

F IGURE 6.7: DETERMINING CAPACITY PASTEURIZER -BUFFER – BLUE PART

Nevertheless, there is a possibility to move the buffer closer towards the bifurcation. This is possible by
implementing a delay of 30 seconds on sensor 10. The 30 seconds is determined by measuring the time it
takes from sensor 10 to the bifurcation, which is done on the production line. This means that, according
to the calculations in Appendix I, the risk on blockage at the pasteurizer is reduced with 30 seconds. The
fact remains that the new solution still has a risk of 0.58 minutes more than the current situation.

Therefore we looked at the probability that a CPL is in failure for more than 5.38 (=5.08+0.30) minutes.
Table 6.6 shows the amount of long failures in the month August. The whole table is shown in Figure 8.4,
Appendix I.

Long Failures
(> 5 min)
CPL 112 46
CPL 111 36
total 82

2,928571429 per 24u

0,976190476 per shift
average less than 1x per shift
6.6: NUMBER OF LONG FAILURES CPLS
We can conclude that the risk of implementing the new solution is that on average, one time per shift the
pasteurizer creates a blockage of 0.58 seconds more than the current situation. In 0.58 seconds the
pasteurizer produces approximately 1300SBs (80,000/60 = 1333 per minute).

68
Comparing this amount with the amount of SBs that experiment 6 yields over the current situation it is
still the best solution to implement experiment 6, as one can see in Table 6.7. With an output of 452903
experiment 6 is still the best experiment. Even when we consider experiment 2, which does not change
the location of the buffer, experiment 6 is the best solution.

Rank Experiment Output Buffer Real Ouput

Average Average
Current: 1 441313 441313
1st 6 453103 1300 452903
2nd 10 449990 449990
3rd 12 444338 1300 443038
6.7: CONCLUSION RISK ANALYSIS
Therefore we finally conclude that experiment 6 should be implemented on the SB production line at
HEINEKEN.

6.4 Conclusions
The conclusion of all the experiments of is that the following experiments ranked 1st, 2nd en 3rd:

Rank Experiment Output Production

balance
Average CPL111 CPL112
Current: 1 441313 57% 43%
1st 6 453103 53% 47%
2nd 10 449990 58% 42%
3rd 12 444338 54% 46%
6.8: B EST THREE ALTERNATIVE SOLUTIONS
This means that the current regulation should be changed into the settings of experiment 6. Translating
the Table 6.4 into the different sensors will result in the following Table 6.9

Experiment CPL112 + low speed CPL111 - low speed CPL111 <> nominal speed CPL111 <> high speed
Current NOSPEED Sensor 17 Sensor 14 Sensor 13
6 NOSPEED NOSPEED Sensor 10 Sensor 8
10 Sensor 12 NOSPEED Sensor 14 Sensor 13
12 Sensor 12 NOSPEED Sensor 10 Sensor 8
6.9: SENSORS OF BEST ALTERNATIVES
The new regulation of sensors of experiment 6 is visualized in Figure 6.8 and Figure 6.9.
Furthermore we conclude that the amount of speed levels at CPL111 will reduce from three levels to two
levels. This means that the amount of speed levels of the CPLs is the same in the new situation.

69
 11

CPL 112 to
13
nominal speed

12
14

17

CPL111 to
10
nominal speed
10

10 = triggered
= not triggered

F IGURE 6.8: NEW SITUATION CPLS TO NOMINAL SPEED

70
 11

CPL 112 to
13
high speed

12
14

17

CPL111 to high
10 speed

10

10
= not triggered

F IGURE 6.9: NEW SITUATION CPLS TO HIGH SPEED

71
 7. Implementation
In this chapter, we discuss how the results from our experiments can be implemented at production line
11 at HEINEKEN. In Section 7.1 we describe the implementation procedure. In Section 7.2 we describe
some preliminary results of a first implementation attempt. At last we formulate the savings of the
implementation in Section 7.3.

7.1 Implementation Procedure

In order to implement experiment 6 we need one resource. This resource is an expert of PA-/PI
engineering, the software department of HEINEKEN. Every conveyor part (in the simulation model
shown as: A, B, .. , InlinerP) has its own regulation which is programmed in the software. Also the
different sensors with their actions are programmed. For example, if sensor 10 is triggered than the
processing speed of CPL111 & CPL111 move to 30,000 bottles per hour, is programmed in a code. This
programming langue differs from the language we use in our simulation program, and therefore we need
an expert to implement our solutions. These programming skills are out of scope for this research.

Nevertheless, the experts implemented our solution and therefore it is possible to check if the results have
a positive influence on the line regulation, as we expected beforehand. This is done in the next section.

7.2 First results after implementation – 8hr shift

After the PA-/PI engineer implemented our solution, the modification is tested for 8 hours (1 work shift).
Actually we should test the modification for several weeks, but this cannot be realized in this research.
Therefore we show the results after an 8hr work shift.

The first improvement is that the problem described in Section 3.2 is solved. The situation of this problem
is shown at the left in Figure 7.1 and was the result of a bad line regulation. After a failure on CPL111,
the sensor before the bending conveyor was not hit by SBs, but after our modification this was solved and
is shown at the right. We moved the buffer before the bifurcation as described in the previous chapter and
therefore we assured that the sensor is hit.

F IGURE 7.1: BEFORE ( LEFT) AND AFTER ( RIGHT) THE IMPLEMENTATION OF MODIFICATION
In Section 3.2 we stated that these inefficiencies take 5 minutes per shift, which are solved now. More
examples of the DNA strands before the implementation are shown in Appendix A.

The second improvement is on the CILT activities. As described in Section 3.1 the CILT activities on
CPL111 were higher than those of CPL112. CILT is described in Section 4.1 and consists of operator
tasks such as, Cleaning, Inspection, Lubrication and Tightening. The CILT activities are scheduled by the

72
maintenance department equally over both CPLs, which means that they both get the same attention
regarding maintenance. Due to the effect that before the implementation, CPL111 produces many more
products than the CPL112, also more CILT activities were needed on the CPL111 and less on CPL112.
Therefore extra activities need to be performed by operators on CPL111, which increases the costs. At
the moment we can conclude that the production balance is around 52% (CPL111) and 48% (CPL112).
Therefore we asked operators, if extra CILT activities occur during or after this work shift. This was not
the case and therefore we conclude that the extra CILT activities on CPL111 are reduced and the CILT
activities on CPL112 are still below the scheduled amount. Note, that in order to validate this conclusion,
we should run the test for several weeks longer. In Appendix J, some pictures are shown where operators
report the differences in CILT activities, thus extra activities on CPL111. After the implementation, an
operator does not have to do extra activities and therefore the lists in Appendix J have no extra
information, so it is hard to show the results on the CILT-forms (no text is added by the operator). If the
production balance approaches the 50/50 balance, then the CILT-activities fits with the maintenance
schedule, so costs are reduced. Considering these CILT-activities we can conclude that these will
decrease with 10 minutes per shift, according to the operator lists.

The output quantity is not improved as expected, but this depends on a lot of variables. Therefore a larger
testing period is needed.

7.3 Savings
To determine the savings we again consider the results between the current situation and experiment 6.
The savings are based on the results of the simulation study, because we believe that only 8-hours testing
are not enough to support our savings. Nevertheless, we can confirm that the results of our
implementation closely match with those of our simulation study as we consider Table 7.1, where REAL
test show the results in real life after our implementation.

Situation Output Production Difference

balance on CPLs
Average CPL111 CPL112
Current (simulation) 441313 57% 43% 14%
Alternative (simulation) 453103 53% 47% 6%
Difference (simulation) 11790 4% 4% 8%

Average(real life before 420193 57% 43% 14%

modification)
REAL test (real life after 447480 52% 48% 4%
modification)
Difference (reallife) 27287 5% 5% 10%
7.1: DIFFERENCE CURRENT , NEW ALTERNATIVE AND REAL LIFE
This table shows the differences between the current and alternative situations of both our simulation as
real life. We show that the modification has a positive effect on the output and production balance.

Furthermore, based on the 8-hour test, we support our statement mentioned before that our simulation
model is valid. Namely, also after the implementation the values are close and reliable. Besides, the
production balance moves towards the 50/50 which was a constraint for a validated model, explained in

73
Section 5.4. Nevertheless, in order to validate our modification further we should run the modification for
several weeks more. Now the 8-hour work shift has an output with 27287 SBs more than the current
situation. It is not realistic to compare one shift with an average over three quarters, therefore we base our
savings on the difference between the current situation in our simulation model with the alternative
situation, colored yellow. We believe this is a realistic comparison because the model is valid and all
values of the model are comparable with real life.

Furthermore, this table shows that the output per shift increases with an average of 11790 SBs and the
production difference between the CPLs is reduced from 14% to 6%, with a total of 8%.

Now we consider what these results yield for HEINEKEN. Unfortunately we cannot say that the output of
the production line increases with 11790 SBs, because HEINEKEN produces on order. This means that
we can state that an order is finished 11790 SBs earlier. Therefore we calculate with reducing costs of
employees, because employees that are paid to complete this order can do other activities instead. We
base the savings of the throughput on the salary of the employees. This means that the production of an
order is 8.42 minutes finished earlier, as calculated in Table 7.2. The savings per year are calculated in
Table 7.3, in the column ‘Higher throughput’.

Output Yield
Star Bottles 80,000 11,790 SB

Per minutes 60 8.42 Min

7.2: REVENUE PER SHIFT – HIGHER OUTPUT
Furthermore the results of the blockage of the pasteurizer described in Section 7.3 results in 5 minutes
less blockage per shifts. Remember that the pasteurizer is the bottleneck machine, and therefore this has a
direct positive influence on the production output. Therefore we made a calculation for the savings per
year, which is shown in Table 7.3 in the column ‘Line regulation’. The determination of the OPI
downtime per hour and operator costs per FTE (shown in the first column) is done by the financial
department and is shown in Appendix K. As mentioned in the previous section, the CILT activities also
contribute to the annual savings, and therefore these are also calculated in 7.3 in the column ‘Decrease
CILT’.

In Table 7.3 we define higher throughput, but as mentioned before this means that the production stops
earlier at the end of a production week, because HEINEKEN produces on order. A production week at the
SB production line consists of 5 production days instead of 7 production days, which means that extra
output plays no role in our savings. Extra output plays a role if a line is producing for 24 hours a day and
7 days per week.

In the Table 7.3 the savings per year are calculated. This means that the total savings per year are €X.

74
 Higher Line Decrease Total
throughput regulation CILT
Reduction of blockage
on pasteurizer per shift(8hr) in 8.42 5
minutes
Shifts per week 20 20
Finished earlier per week (minutes) 168.4 100
OPI downtime (per hour) € €
Real OPI downtime per hour € €
Production weeks per year
Less CILT-activities per shift 10
(minutes)
Shifts per week 20
Less CILT-activities per week (200/60 =)
(hours) 3,33
Hours per work week
Real reduction percentage
Costs of 1 fte per year €
€ € €
Savings per year
€
7.3: S AVINGS PER YEAR
In order to support the table above, we developed three formulas in order to determine the savings per
year:

Savings per year of higher throughput:

Savings per year line regulation: (same formula as higher throughput)

Savings per year of decrease in CILT-activities:

Total savings per year: Savings per year of higher throughput + line regulation + CILT decrease =
€X

75
 8. Conclusion and Recommendations
To finish our research, we describe the overall conclusion, recommendations and we propose ideas and
insights for further research. This advice is at the same time the answer to our research question which is
stated in Section 1.4, and is as follows:

“How to improve line performance at the regulated production line (line 11) at HEINEKEN
Zoeterwoude?”

We summarize the conclusion of the previous chapters and answer our research question in Section 8.1.
We describe the recommendations for the management of HEINEKEN in Section 8.2. In Section 8.3 we
propose ideas and insights for further research.

8.1 Conclusions
The mission of this research was to improve the performance of the SB production line. However, the
question was how to do this. Therefore we started with a process analysis in order to be sure that we
consider every part of the production line. After describing the process, we continued with a data analysis
to detect the bottleneck of the SB production line, the part that has the most negative influence on the line
performance. The result from this process and data analysis was that we focused on the area from the
pasteurizer to the CPLs. There are two main issues after these analyses, which are:
- The pasteurizer creates blockage due to an inefficient regulation of the CPLs. This results in an
incorrect downtime of the CPL112.
- The production balance between the CPLs was uneven (CPL111: 55% against CPL112: 45%). This
results in extra activities (CILT) of an operator, due to an incorrect maintenance schedule (which is based
on a 50/50 balance).
These main issues arise due to inefficiencies of the regulation at the production line. This regulation is
done by sensors and therefore we consider different solutions regarding this regulation.

In order to solve these two inefficiencies and therefore to improve the line performance, we developed a
conceptual model. Review of theory helped us to substantiate ideas and decisions regarding interventions
and possible improvements for this conceptual model. We translated this conceptual model into a
simulation model in order to test possible changes on the production line. We run 12 different
experiments, including the current situation, to determine the best solution. The best solution was
experiment 6, which states that three out of four sensors settings have to be changed and that the speed
level of CPL112 should be decreased from three to two levels. CPL111 & CPL112 are triggered on the
same sensor, which means that they will start and stop at the same time.

The efficiency of the regulation between the pasteurizer and CPLs increases because a production week
stops on average 168.4 minutes earlier, in the new situation. This is because the throughput of the
production line is increased, and therefore more products can be produced at the same time. In addition,
the two main issues described above are solved. The inefficiency of the blockage of the pasteurizer is
corrected, which results that the production week stops 100 minutes earlier. In total this is 268.4 minutes
of the 9600 minutes per week. Furthermore the CILT tasks over the operators are reduced, because the
production balance in the new situation is CPL111: 53% against CPL112: 47%. The reduction of these
CILT tasks results in 10 minutes less CILT-activities per shift. All these improvements lead to yearly
savings of €X.

76
So, in order to improve the line performance on a regulated production line, the new changes should be
implemented.

8.2 Recommendations
In addition to the recommendation to implement the new regulation settings, we found some other
inefficiencies or possible improvements during this research. Below, we provide an overview of our
recommendations:

 Focus more on conveyors/lines. On all packaging lines the focus is on the machines. Several
teams focus on improving machine efficiencies. Mostly the thoughts at HEINEKEN consists, that
the line performance is determined by all machine performances, which is understandable.
Nevertheless, the conveyors and buffers also play an important role in the line performance. The
conveyors between the machines can be seen as a machine itself, which is proven by this
research. The implementation of the outcome of this research is relative small, but the results are
relative large.
 Create an overview of the functioning of sensors on the production line. In order to improve the
efficiency between machines, it is necessary to have a clear understanding of the function of the
sensors. Then superficial inefficiencies can be solved directly and the rest can be completed with
further research. This is also very useful to visualize the operation of the production line.
 Improve data registration at MES. The data registration at MES should be improved, especially
for the SB production line. Due to the regulated production line, MES is not capable to measure
all parameters. When a machine is producing at full or half speed, MES registers only
‘producing’ in a green DNA strand. Some tools presented in Section 3.1 will help to give an
insight into the speed losses, but this should be done in MES. Furthermore, some PLCs measure
wrong data. Not all machine states are registered correctly. This makes the data unreliable. For
example, as described in Section 3.1. PLCs measure at CPL112 a starvation instead of a
blockage. Overall, this recommendation is important, because a lot of decisions are based on data
at MES.
 Hire extra PA-/PI engineer. When inefficiencies are noted by employees, they have to write a
label. Different aspects on these labels are possible, from safety issues till machines issues. When
such an aspect consists of technical issues these arrive on the desk of a PA-/PI engineer. Some
filled in labels are on stack for six months. This slow response discourages the operators to help
improving the line performance.
 Improving the administration of changing small objects. The exchange of small objects (e.g.,
Teflon cylinders, glue sprayer) and their location is not registered by the maintenance department.
Known is the amount of spare parts changed, but not the destiny of it. Therefore it is not possible
to determine the frequency and amount of small objects changed on parallel machines.
 Visualization of inefficiencies for operators. At the moment every machine has its own ‘light’ that
visualizes the machine state. Nevertheless, not everything is visualized. For example, when on the
bottle washer a couple of fallen bottles block the entrance, no light is shown. Sometimes these
fallen bottles cause a machine inefficiency of 11,5% (6 out of 52 empty pockets). Therefore an
operator should know if fallen bottles are present at the entrance of the bottle washer. This can be
done with another light for ‘fallen bottles at entrance’ in order to prevent machine inefficiencies.

77
 8.3 Further research
We finish our research with the proposition of ideas for further research. The ideas for further research are
given below:

 Optimizing all conveyors/lines between machines. The subject will be the same as this research,
but should consider every flow line between two consecutive machines. The part between the
pasteurizer and CPLs which is done in this research is just one part. Multiple improvement steps
can be made, when considering the whole production line.
 Improving the use of the crate buffer. At the moment operators choose when crates should be
stored into the crate buffer or when crates should be pulled from this buffer. This means that
when an operator is busy on another machine, he/she is not able to regulate the crate buffer.
Further research should indicate whether if it is profitable to make this automatic and how this
should be done.
 Reducing downtime of the packaging machine. The packaging machine has a high down time.
This causes a blockage on the CPLs and therefore on the pasteurizer. This is a direct loss.
 Visualization of inefficiencies for operators. In the previous section an example is given.
Nevertheless this should be researched for other machines.

78
References

AHUJA, I.P., 2011. Manufacturing excellence through total productive maintenance paradigm.
International Journal of Technology, Policy and Management.

AHUJA, I.P. and KHAMBA, J., 2008. Strategies and succes factors for overcoming challenges in TPM
implementation in Indian manufactruing industry. Journal of Quality in Maintenance Engineering, 14(2),
pp. 123-147.

ARLBJØRN, J.S., FREYTAG, P.V., 2013. Evidence of lean: a review of international peer-reviewed
journal articles. European Business Review, 25(2), pp. 174-205.

BAN KS, J., 1998. Handbook of Simulation. Library of Congress Cataloging in Publication data.
Copywright by John Wiley & Sons.

BASTANI, A.S., 1988. Analytical solution of closed-loop conveyor systems with discrete and
deterministic material flow. European Journal of Operational Research, 35, pp. 187-192.

BELZER, J., HOLZMAN, A.G., KENT, A., 1978. Encyclopedia of Computer Science and Technology.
Volume 10 - linear

BENDELL, T., 2006. A review and comparison of six sigma and the lean organizations. The TQM
Magazine, 18(3), pp. 255-262.

CHAN, F., LAU, H., IP, R., CHAN, H. and KONG, S., 2005. Implementation of total productive
maintenance: a case study. International Journal of Production Economics, 95(1), pp. 71-94.

CUA, K.O., McKONE, K.E., SCHROEDER, R.G., 2001. Relationships between implementation of
TQM, JIT, and TPM manufacturing performance. Journal of Operations Management, 19(6), pp. 675-
694.

DE MAST, J., LOKKERBOL, J., 2012. An analysis of the Six Sigma DMAIC method from the
perspective of problem solving. International Journal of Production Economics, 139(2), pp. 604-614.

DE RON, A., and ROODA, J., 2006. OEE and equipment effectiveness: OEE revisited. Semiconductor
Manufacturing, IEEE Transactions, 18(1), pp. 190-196.

FARRIS J.A., VAN AKEN, E.M. DOOLEN, T.L., WORLEY, J., 2009. Critical success factors for
human resource outcomes in Kaizen events: An empirical study. International Journey of Production
Economics, 117(1), pp. 42-65.

FITZGERALD, L., JOHNSTON, R., BRIGNALL, S., SILVESTRO, R. and VOSS, C., 1991.
Performance Measurement in Service Business. CIMA. London.

HÄRTE, F.L., 1997. Efficiency analysis of packaging lines. Delft University Press.

HARTMANN, E., 1992. Successfully installing TPM in a non-Japanese plant: Total productive
maintenance. TPM Press London.

79
HEINEKEN. 2011. BCS Data Definitions Version 5.5.0. Zoeterwoude: Heineken.

HNS visie 2015. 2011. HNS nieuws special (pp. 1-12). Zoeterwoude: Heineken Nederland Supply.

KEGG, R.L., 1990. The Development of Sensors for Manufacturing Automation. Manufacturing Control
and Monitoring. Computers in industry, pp. 123-129.

KWO, T.T., 1958. A theory of conveyors. Management Science, 5(1), pp. 51-71.

LAW, A. (2006). Simulation Modeling and Analysis. Europe: Mc-Graw-Hill Education.

LIN, S., HUANG, X. 2011. Advances in Computer Science, Enviroment, Ecoinformatics , and Educatoin.
International Conference, CSEE, Wuhan China, part 3, pp.552-556.

MAHALIK, N.G.P.C., LEE, S.K., 2001. A study on production line automation with LonWorks control
networks. Computer Standards & Interfaces pp. 21-27.

MERTON, R.C., 1972. An Analytic Derivation of the Efficient Portfolio Frontier. The Journal of
Financial and Quantitative Analysis, pp. 1851-1872.

Muth, E.J., 1974. Analysis of closed-loop conveyor systems; the distribution flow case. AIIE
Transactions 6(1), pp. 73-83.

NAKAJIMA, S., 1991. Continuous Improvement in Operations. Cambridge, MA: Productivity Press.

NAKAJIMA, S., 1988. Introduction to TPM: Total productive Maintenance. Cambridge, MA:
Productivity Press.

NEELY, A.D., GREGORY, M. and PLATSS, K., 1995. Performance measurement system design.
International Journal of Operations & Production Management, 19 (2), pp. 205-228.

NEELY, A.D., 2005. The performance measurement revolution: why now and what next?. International
Journal of Operations & Production Management, 19(2), 205-228.

NEELY, A.D., 2002. Business performance measurement: theory and practice. Cambridge University
Press.

PARMENTER, D., 2010. Key performance indicators (KPI): developing, implementing, and using
winning KPIs. John Wiley & Sons.

PEPPER, M.P.J., SPEDDING, T.A., 2010. The evolution of lean Six Sigma. International Journal of
Quality & Reliability Management, 27(2), pp. 138-155.

RAHMAN, S., 1998. Theory of Constraints: A review of the philosophy and its applications.
International Journal of Operations & Production Management, 18(4), pp. 336-355.

ROLFSEN, M. and LANGELAND, C., 2012. Successful maintenance practice through team autonomy.
Employee Relations, 34(3), pp. 306-321.

SARGENT, R.G., 2005. Verification and Validation of Simulation Models. Winter Simulation
Conference, pp. 130-143.

80
SCHROEDER, R.G., LINDERMAN, K., LIEDTKE, C., CHOO, A.S., 2008. Six Sigma: Definition and
underlying theory. Journal of Operations Management, 26(4), pp. 536-554.

SHAH, R., & WARD, P.T., 2003. Lean manufacturing: Context, practice bundies, and performance.
Journal of Operations Management, 21(2), pp. 129-149.

SHANNON, R., 1975. Systems simulation: the art and science. Upper Saddle River, New Jersey:
Prentice-Hall.

SIEMENS. 2012. Tecnomatix Plant Simulation.www.plm.automation.siemens.com .Plano: Siemens.

TINO, A., KHAN, G. 2013. High performance NoC synthesis using analytical modeling and simulation
with optimal power and minimal IC area. Journal of Systems Architecture. 59(3), pp. 1348-1363.

TSAROUHAS, P. 2007. Implementation of total productive maintenance in food industry: a case study.
Journal of Quality in Maintenance Engineering.

VAN DER DUYN SCHOUTEN, F.A., VANNESTE, S.G., 1995. Maintenance optimization of a
production system with buffer capacity. European Journal of Operational Research, 82(3), pp. 323-338.

WEIN, L. M., and CHEVALIER, P. B., 1992. A broader view of the job shop scheduling problem.
Management Science, pp. 1018-1025.

WILLMOTT, P. and MCCARTHY, D. , 2001. TPM a route to world-class performance. Oxford. Boston.
Butterworth-Heinemann.

YEUNG, W.H.R., MOORE, P.R., 1996. Object-oriented modeling and control of flexible conveyor
systems for automated assembly. Elsevier Science Ltd. Mechatronics, 6(7), pp.799-815.

81
Appendices
APPENDIX A: EXAMPLES OF MES DNA INEFFICIENCIES

82
APPENDIX B: F LOW LINE - SIMULATION MODEL
In Figure 8.1 the main frame of the simulation model is showed again.

F IGURE 8.1: MAIN F RAME P LANT S IMULATION

The flow line of bottles is separated in multiple conveyors, because every conveyor has its own
characteristics. Some characteristics are length, width, number of strokes and speed. We want to improve
the understanding of the flow line, and therefore we consider conveyor. We choose conveyor C, which
makes no difference because the regulation of all conveyors is the same. Only the destination tables and

83
the characteristic will differ. In Figure 8.2, conveyor C is shown in more detail.

F IGURE 8.2: CONVEYOR C IN DETAIL

As one can see C has five different strokes. In real life these can move independently. Not every conveyor
has the same amount of strokes. Therefore these are measured on the production line. The results are
explained in Calculation of capacity.

Calculation of capacity
In Table 8.1 one can the information per line in more detail. The ‘% of line used’ is the percentage that is
efficiently used on a line, because on every line a SB is forced to move to another line. This is done by a
fence that is positioned above the strokes. This means that the place behind those fences is not uses by the
SBs, and is therefore excluded from the capacity. The capacity included the % of line used is the efficient
capacity.

scale
1:100 in Simulation Reality 1 line =
Length in Eff
Length in cm m # of strokes capacity % of line used capacity
Name Line
A 2,6 2,60 5 343,2 99% 339,8
B 2,8 2,80 5 369,6 99% 365,9
C 11 11,00 5 1452 99% 1437,5
D 9 9,00 5 1188 99% 1176,1
E 12 12,00 5 1584 98% 1552,3
F 7 7,00 5 924 98% 905,5
G 9 9,00 5 1188 98% 1164,2

84
H 4,4 4,40 8 932,8 92% 858,2
I Angle 1,10 1,10 8 233,1062 95% 221,5
Straight 4,5 4,50 8 954 95% 906,3
5,6 5,60 1187,106 95% 1127,8
J Angle 0,55 4
Straight 4 4
4,5 4,55 477,7268 95% 453,8
Inliner N 4 4,00 12 1279,844 10% 128,0
K 4,25 4,25 4 446,25 98% 437,3
L Angle 0,55 4
Straight 6 4
6,55 6,55 687,7268 95% 653,3
M 8 8,00 4 840 98% 823,2
O 4 4,00 4 420 98% 411,6
Inliner P 4 4,00 12 1279,844 10% 128,0
SKUs
Total
15787 capacity: 13090
8.1: LINE INFORMATION
In Table 8.2 the translation from strokes to the capacity is shown.

Diameter SB 60 mm

Stroke
82.5 mm 84
Capacity (# of
Stroke bottles) per meter
# of strokes width stroke
1 24
2 51
3 78
4 105
5 132
6 159
7 186
8 212
9 239
10 266
11 293
12 320
13 347
14 374
8.2: CALCULATION OF AMOUNT OF BOTTLES

85
APPENDIX C - DISTRIBUTIONS DATA SET
In this data set, we try to find a theoretical distribution. As mentioned before we try to find distributions
for the processing time, the MTBF and the MTTR. Plant Simulation tests the data against twelve
theoretical distributions which are listed below:

 Beta distribution
 Binomial distribution
 Erlang distribution
 Gamma distribution
 Geometric distribution
 Lognormal distribution
 Normal distribution
 Negative exponential distribution
 Poisson distribution
 Triangle distribution
 Uniform distribution
 Weibull distribution
According to Law (2006), these twelve distributions are the most promising to fit with our process
and failure times. Furthermore there are three tests that can check if the data fits with a theoretical
distribution. Law (2006) describes three different tests which are:

 Chi-squared test (CS). This is used to determine if a sample comes from a data set with a
specific distribution. The test is applied to binned data, so the value of the test statistic
depends on how the data is binned. There is no optimal choice of number of bins (k), which
is determined by

 Kolmogorov-Smirnov test (KS). This is used to decide if a sample comes from a

hypothesized continuous distribution.
 Anderson-Darling procedure (AD) . This is a general test to compare the fit of an observed
cumulative distribution function to an expected cumulative distribution. This test gives
more weight to the tails than the Kolmogorov-Smirnov test.
In order to find a theoretical distribution function we will use a combination of the Chi-squared
test and the Kolmogorov-Smirnov test. The Anderson-Darling test focuses more on tails, which is
something that is not preferable in our model. We will use a level of significance of 5%, which is
common in statistics.

86
APPENDIX D – DETERMINING THEORETICAL DISTRIBUTION: MTBF
In order to rank the different distributions we used two different tests. The first one is the Kolmogorov-
Smirnov (KS) test and the second one is the Chi-Squared (CS) test. These tests are performed using
Excel. Number 1 means that it ranked first and therefore the best. Number 2 ranks the second best and so
on. In Table 8.3 one can find the different rankings on the two tests.

CPL111 CPL112

KS CS KS CS
Beta 4 - 1 -

Exponential 7 5 6 5

Gamma 2 2 2 3

Log normal 3 4 4 4

Normal 5 1 5 2

Uniform 6 - 7 -

Weibull 1 3 3 1

TABLE 8.3 – KS AND CS TEST FOR MTBF

For both CPLs the P- value is very close to zero or zero as one can see in Table 8.4. Considering a
significance level of 5%, we can conclude there is no distribution that fits to the data, based on the two
tables. Therefore we use empirical data in order to determine the MTBF for the simulation model.

This MTBF is stored in a table file and linked with the option ‘cEmp’ in Tecnomatix Plant Simulation.

KS P-value CS P-value

CPL 111 Weibull 0 0

Normal 0 0

CPL112 Beta 0 0

Weibull 0 0.038

TABLE 8.4 – P- VALUE: MTBF

87
APPENDIX E - DETERMINING THEORETICAL DISTRIBUTION : MTTR
Determining the fit of data from the process with a theoretical distribution of the MTTR is done the same
as described in Appendix B for the MTBF. The ranking of the distributions is also the same, which means
that number 1 is the best, number 2 the second best and so on. In Table 8.5 we show that both test score
the best on a Weibull distribution.

CPL111 CPL112

KS CS KS CS
Beta 5 4 5 5

Exponential 4 5 4 4

Gamma 2 2 2 2

Log normal 3 3 3 3

Normal 6 6 6 6

Uniform - 7 - 7

Weibull 1 1 1 1

TABLE 8.5 – KS AND CS FOR MTTR

KS P-value CS P-value

CPL 111 Weibull 0 0.052

CPL112 Weibull 0.103 0.173

TABLE 8.6 – P- VALUE: MTTR

Considering the P-value in

KS P-value CS P-value

CPL 111 Weibull 0 0.052

CPL112 Weibull 0.103 0.173

Table 8.6 and a significance level of 5% we can conclude that a Weibull distribution satisfies. For
CPL111 we see not fit the Kolmogorov-Smirnov test. The significance level of the Chi-squared test
satisfies narrowly (5.2% > 5%). The P-values of the KS test and CS test on the CPL112 are convincing
(10.3% and 17.3% respectively). This means that we use for the MTTR the Weibull distribution with the
following parameters, shown in Table 8.7.

88
 Distribution Parameters

CPL111 Weibull α = 0.83029 β = 36.428

CPL112 Weibull α = 0.78302 β = 28.755

TABLE 8.7 – F IT W EIBULL DISTRIBUTION WITH CORRESPONDING PARAMETERS

The Weibull distribution with the corresponding parameters is implemented in Tecnomatix Plant
Simulation. Furthermore one can see the interval of the duration of a failure, which contains the empirical
data of the MTBF. A screenshot of the simulation is shown in Figure 8.3.

F IGURE 8.3: P RINT SCREEN OF WEIBULL DISTRIBUTIONS OF MTBF AND MTTR

89
 APPENDIX F: CPL S TO NOMINAL SPEED

CPL 112 to
nominal
speedCPL 112
12
10 14
to nominal 12
10 14
speed

17
17

CPL 111 to
nominal
speedCPL 111
to nominal
speed

10 = triggered=
10 triggered
10 = not triggered=
10 not triggered

90
APPENDIX G: CPLS TO HIGH SPEED

CPL 112 to
13
full speed

10 12
14

17
8 OR 9

CPL 111 to
full speed

10 = triggered

10 = not triggered

91
APPENDIX H: RESULTS EXPERIMENTS 1 TILL 12

Experiment Output Production balance Waiting Stopping Starvation Failure

Average CPL111 CPL112 CPL111 CPL112 CPL111 CPL112 CPL111 CPL112 Total CPL111 CPL112
1 441313 57% 43% 0,78 38,08 28,99 0,00 29,77 38,08 67,85 2,22 0,85
2 416625 29% 71% 0,05 9,51 67,71 0,00 67,77 9,51 77,28 0,43 1,33
3 388495 19% 81% 0,00 7,03 69,40 0,00 69,40 7,03 76,42 1,03 0,24
4 435440 58% 42% 1,72 39,03 0,00 0,00 1,72 39,03 40,75 1,65 0,54
5 444508 57% 43% 0,82 38,79 0,00 0,00 0,82 38,79 39,61 1,20 0,18
6 453103 53% 47% 0,01 30,61 0,00 0,00 0,01 30,61 30,62 1,42 0,04
7 439100 62% 38% 2,45 0,00 22,20 48,17 24,65 48,17 72,82 0,84 0,13
8 379278 23% 77% 0,03 1,39 76,64 9,41 76,67 10,80 87,47 0,46 1,36
9 408198 31% 69% 0,00 1,34 66,44 12,25 66,44 13,59 80,03 0,39 1,06
10 449990 58% 42% 2,90 8,77 0,00 19,71 2,90 18,48 31,38 1,99 1,03
11 430915 57% 43% 0,78 34,63 0,00 2,46 0,78 37,09 37,86 2,83 0,86
12 444338 54% 46% 0,08 33,33 0,00 0,51 0,08 33,84 33,92 0,78 0,80

92
APPENDIX I: RISK ANALYSIS - BLOCKAGE ON PASTEURIZER

93
In Table 8.8 one can find the capacities per line.

Number of
Length strokes capacity % of stroke used Eff capacity
Name line m
A 2,60 5 343,2 99% 339,8
B 2,80 5 369,6 99% 365,9
C 11,00 5 1452 99% 1437,5
D 9,00 5 1188 99% 1176,1
E 12,00 5 1584 98% 1552,3
F 7,00 5 924 98% 905,5
G 9,00 5 1188 98% 1164,2
373.1 or
H 4,40 8 932,8 32% or 68% 559.7
I 5,60 1187,106 95% 1127,8
J 4,55 477,7268 95% 453,8
Inliner N 4,00 12 1279,844 10% 128,0
K 4,25 4 446,25 98% 437,3
L 6,55 687,7268 95% 653,3
M 8,00 4 840 98% 823,2
O 4,00 4 420 98% 411,6
Inliner P 4,00 12 1279,844 10% 128,0
Total eff
Capacity 14600 capacity: 13090 SBs
Total eff
capacity: 2517 SBs
Total eff
capacity: 7576 SBs
8.8: DETERMINING BUFFER CAPACITY

94
F IGURE 8.4: NUMBER OF FAILURES OF CPLS

95
APPENDIX J: EXTRA CILT ACTIVITIES ON CPL111

96
APPENDIX K: OPI DOWNTIME PER HOUR AND OPERATOR COSTS

97
98
99