You are on page 1of 21

Provesende Bakery

A case study

Total Quality Management

Daniel Filipe up201505890@fe.up.pt


Valerie Geelen up201901590@fe.up.pt
José Ribeiro up201503561@fe.up.pt
Kasper Camp up201902097@fe.up.pt
Naomi Platenburg up201901179@fe.up.pt

Industrial Engineering and Management


22nd of December 2019
Provesende Bakery: a case study

Abstract
The following paper analyzes the bread production process in a traditional bakery located in
Provesende (Douro Valley, Portugal). The process’s dependent variable is weight, and the variables
that impact it (the independent ones) are the flour type, the oven, and the baker. After characterizing
the dependent variable, it was found that there was a disparity in bread weights, especially in
different flour types.

Next, a characterization of the independent variables was performed, which allowed us to


understand which ones had a significant impact (vital X’s) on the weight of the bread. This was done
recurring to the analysis of box plots and by using parametric tests to confirm the significance of
that impact. After an individual analysis of the variables, interactions between the independent
variables that had a significant impact on the dependent one were studied. Moreover, the statistical
control of the process was verified as well as the capacity and performance of the process through
control charts and the statistic Sixpack Analysis tool.

Finally, it was concluded that the flour type variable is the only independent variable that
significantly influenced the weight and had a great impact on its variability. It should be noted that
despite the division made, both flour types are performing differently. In general, flour type A
produces breads which are closest to the desired weight. Therefore, it is recommended to use only
one type of flour for all bakers, preferably flour type A.

ii
Provesende Bakery: a case study

Table of Contents

1. Introduction .............................................................................................................................................................. 1
2. Methodology............................................................................................................................................................. 1
3. Process Study ........................................................................................................................................................... 2
4. Data Analysis ............................................................................................................................................................ 2
4.1. Variable analysis ............................................................................................................................................ 2
4.2. Dependent variable characterization – Weight ................................................................................... 3
4.3. Independent variable characterization ................................................................................................... 4
4.3.1. Identification of vital X’s .................................................................................................................... 4
4.3.2. Statistical Inference ............................................................................................................................. 4
4.4. Joint Analysis ................................................................................................................................................... 5
5. Statistical Control ................................................................................................................................................... 7
6. Study of the capability and performance of the process ........................................................................ 10
7. Conclusions ............................................................................................................................................................ 12
8. Appendices ............................................................................................................................................................. 13

iii
Provesende Bakery: a case study

List of Figures
Figure 1 Fishbone diagram of the production of the bread ............................................................................... 2
Figure 2 Summary report for Weight ......................................................................................................................... 3
Figure 3 Average weight per day and flour type ................................................................................................... 4
Figure 4 Boxplot of the weight per flour type and driver ................................................................................... 6
Figure 5 Tukey Intervals for flour type ..................................................................................................................... 6
Figure 6 Xbar-S chart of Weight (20 samples)........................................................................................................ 8
Figure 7 Xbar-S Chart of Weight – flour type A..................................................................................................... 9
Figure 8 Xbar-S Chart of Weight – flour type B..................................................................................................... 9
Figure 9 Sixpack Report for Weight for flour type A ......................................................................................... 10
Figure 10 Sixpack Report for flour type B............................................................................................................. 11

List of Appendices
Appendix 1 Xbar-S Chart of Weight .................................................................................................... 13
Appendix 2 Sixpack Report for Weight ............................................................................................... 13
Appendix 3 Boxplot for Weight per baker ........................................................................................... 14
Appendix 4 Boxplot of Weight per Oven............................................................................................. 14
Appendix 5 Boxplot of Weight per flour type ..................................................................................... 15
Appendix 6 Tukey Intervals for Oven .................................................................................................. 15
Appendix 7 Tukey Intervals for Baker ................................................................................................. 16
Appendix 8 Tukey Intervals for Flour Type ......................................................................................... 16
Appendix 9 Boxplot of Weight for Flour Type and Oven .................................................................... 17
Appendix 10 Boxplot of Weight for Baker and Oven .......................................................................... 17

iv
Provesende Bakery: a case study

1. Introduction
For every organization, it is important to have a continuous improvement of production processes
and services to maintain and improve customer satisfaction, and to increase their competitive
advantage. It is therefore of crucial importance to maintain good quality management.

This case considers a bakery located in Provesende. The owner of the bakery is not sure if the
production process is stable, and what supplier ensures the best results. Since the loaves of bread
have a fixed price, it is important to maintain the weight of the loaves constant, to ensure that the
production costs are also constant. Therefore, the goal of this research is to understand what the
critical variables are associated with the final stage of the baking process. Additionally, it will be
assessed if the process is in statistical control.

First, the description of the methodology of the processes that are executed in this research is
provided. Next, the process within the bakery is thoroughly evaluated by means of a cause-and-
effect diagram. After this, the provided data is analyzed to see if the process is in statistical control.
Finally, a conclusion will be drawn and recommendations will be provided to improve the quality of
the processes.

2. Methodology
The realization of the study regarding bread production in a traditional bakery in firewood ovens was
based on a Six Sigma methodology, more specifically the Define-Measure-Analyze-Improve-Control
(DMAIC) method. Since the data that is required for the analysis of the problem was provided through
Excel file with all process variables, the problem definition and data collection phase (Define and
Measure respectively) are not considered in the report. Hence, the project presented will only address
the data analysis (Analyze) and process improvement (Improve) stages.

The first step in the analysis of the problem in question was the analysis of all variables. For the
analysis of variance, each one of these variables is characterized according to its qualitative and
quantitative nature and on their dependence on other existing variables. Next, the dependent
variable (weight) was characterized, with the purpose of determining the mean value, standard
deviation and the existence of outliers. In addition, a test for normality of the dependent variable
(Anderson Darling) is performed, to determine if the data is normally distributed over time.

Subsequently, independent variables that significantly influence dependent variables are identified
(vital X’s) by analyzing box plots and using parametric tests to confirm the significance. After
individual analysis of the variables, interactions between the independent variables that make
statistically vary the dependent variable are verified.

Next, the statistical control of the process was verified as well as the capacity and performance of
the process through control charts and the statistic Sixpack Analysis tool.

Finally, after a critical analysis of the results obtained, conclusions were drawn relative to the
variables that define the process.

1
Provesende Bakery: a case study

3. Process Study
In this section, the processes within the bakery are analyzed. First, we used the DMAIC-methodology
to understand the process and to see if we can find the causes for the variability of the weight of the
loaves. The use of this methodology provided useful insights into the possible root causes of the
issue that the baker faces. These causes are categorized into six categories, namely people, process,
environment, machines, measurement and materials, and are presented in the following fish-bone
diagram.

Figure 1 Fishbone diagram of the production of the bread

4. Data Analysis
a. Variable analysis
To evaluate the process and its capacity, the following variables were classified which allows a
correct interpretation throughout the analysis to be performed. The variables in this process,
which were recorded over 400 observations for 10 days are the weight (g) of the loaves, the baker
who baked the loaves, the type of flour used and in what oven.

The bakery has four bakers, Ana, Leo, João and Peter, who bake the bread and put it in the oven
into which are considered as a qualitative variable, nominal independent. The ovens, which are
used to bake bread (2 ovens working exclusively for baking breads) is also a qualitative variable,
nominal independent. The two types of flour used in the baking process is a qualitative variable,
ordinal independent. The weight of the bread is a dependent discrete quantitative variable,
where the limit tolerance is 300g ± 10g. The weight variable is classified as a continuous variable,
as the data provides the weight of the bread in decimal values.

2
Provesende Bakery: a case study

These variables do not present any kind of correlation and cannot be grouped together.

b. Dependent variable characterization – Weight

The dependent variable considered in this process was the Weight of the bread. The mean weight
of the bread is 302.16g and its standard deviation is 5.61g.

In order to test the normality of the observations, one can use both skewness and kurtosis. If
none of these values is close to 0, it is likely that the variable is not normally distributed. The
skewness value is 0.22 which allows us to conclude that the distribution is symmetric. The value
of kurtosis (0.02) is also close to zero, meaning that this distribution is the same as the normal
distribution - mesokurtic (medium peak). With the purpose of verifying if the dependent variable
follows a normal distribution, an Anderson Darling test was computed since it is more sensitive
than the K-S test. The p-value obtained is slightly higher than 0,005 (p = 0,0057), thus meaning
that it is possible to approximate this distribution to a normal distribution.

The last stage of the analysis was to assess how the weight varied over time, according to flour
type. The scatterplot represented below (figure 3) shows that loaves baked with flour type B are
always heavier than the ones baked with flour type A, leading to the conclusion that the type of
flour used may have an impact on the output variable and should be later studied.

Figure 2 Summary report for Weight

3
Provesende Bakery: a case study

Figure 3 Average weight per day and flour type

c. Independent variable characterization


i. Identification of vital X’s

The first step in the process is to determine which independent variables have an effect on
the response variables: this is called identifying the vital X’s. To make a good analysis of this,
three boxplots are constructed, one for each variable. These boxplots can be found in
appendix 3 to 5. The first boxplot shows that, in general, João and Peter make heavier breads
than Ana and Leo. Furthermore, João and Peter are in general further away from the desired
value of 300, however, their average values are still within the limit of +/ -10 grams. In the
second boxplot, it can be seen that there is no clear difference in weight between the breads
made with the primary oven compared to the ones made with the secondary oven. The third
boxplot indicates that when flour B is used, the weight of the bread is in general higher. It
can also be seen that when flour A is used for breads, the weight is closer to the desired
value of 300 than it is for flour B, however flour B is on average still within the limits of +/-
10 grams.

ii. Statistical Inference

It is necessary to verify whether the differences in the groups of variables are statistical
identifiable. Therefore, an ANOVA and two-sample t-test with (5% level) are performed. For
the ANOVA, variance equality is assumed, whereas for the two-sample t-test, the variance
equality is confirmed by the Levene method. The ANOVA test is used to check if the mean is
the same for all groups. It is done by finding the mean for each group and the overall mean.
Then, the within-group variation, which is the total deviation from each member compared
to the group mean, is calculated. Also, the between-group variation, which is the deviation
from each group mean from the overall mean is calculated.
4
Provesende Bakery: a case study

This is done using Minitab, and is used to test the following hypotheses:

H0: the expected values are the same for all groups

H1: the expected values are not the same for all groups

For the ANOVA analysis, the hypotheses are tested. For the Flour and Oven, it is not relevant
to perform an ANOVA, since they both have only two groups. Therefore, for these two
variables, only the t-test is performed. For the variable Baker, it is necessary to perform an
ANOVA since it has 4 groups. The results show that the p-value for the Baker is 0.000, which
is smaller than the significance level of 0.05. This means that the null hypothesis is rejected,
implying that the expected values are not the same for all groups.

The two-sample t-test is used to check if there is a significant difference between the means
of two groups. The difference between t-test and ANOVA is that the t-test only checks the
difference between two groups, and ANOVA can do more than two groups.

The hypothesis for the t-test are as follows:

H0: the expected values are all the same

H1: the expected values are not all the same.

The t-test will be performed for the variables Flour and Oven. For Flour, the p-value is 0.000,
which is smaller than the significance level of 0.05, meaning that the null-hypothesis is
rejected. For Oven, on the other hand, the p-value is 0.699, indicating that it fails to reject
the null hypothesis.

Now that it is known that for Baker and for Flour the null-hypotheses are rejected, a Tukey’s
range test is performed to see which of the groups are different. The results of this analysis
can be found in appendix 6 until 8. When an interval contains zero, the groups are not
significantly different. For the variable Oven, the t-test is confirmed, since the interval
contains zero. When taken a look at the bakers it can be found out that there is no difference
between Leo and Ana, and no difference between Peter and João. For the use of flour, a clear
difference is noticed between A and B.

d. Joint Analysis
After the individual analysis of the variables oven, baker and flour, the last two were found to
have a significant impact on the weight of the bread. Therefore, it is wise to do a joint analysis
of the variables.

Looking at the boxplots for both situations with focus on the flour variable it can be found that
Anna and Leo only use Flour A, while João and Peter are only using Flour B (figure 4). Therefore,
no differences in using Flour can be found for the same bakers. This indicates that the differences

5
Provesende Bakery: a case study

between bakers found in the previous section (João and Peter bake heavier bread than Anna and
Leo) might not be solely attributed to the bakers, but most likely due to the flour type they use.

Figure 4 Boxplot of the weight per flour type and driver

Furthermore, it seems that the use of the different Flours in different Ovens does not cause
significant differences in bread weight. In order to be sure about the differences an ANOVA test
was performed. It was found that, for both Ovens, there was a statistically significant difference
between using flour A and B. Once again, this corroborates the hypothesis that has been building
up, which is that flour type influences bread weight. The Tukey intervals are presented in the
figure below, while the boxplot is shown in appendix 9. The Tukey intervals clearly show the
impact of flour in the response variable.

Figure 5 Tukey Intervals for flour type

6
Provesende Bakery: a case study

When doing the test for the different Bakers using different Ovens it could be found that for
Primary Oven there is a significant difference between the weight of the bread of João and Leo.
However, this result is not useful since they do not use the same kind of Flour and this difference
might arise from that. For this reason, this type of cross-analysis (Oven-Baker) does not give new
insights towards reaching a conclusion about the flour’s impact on the bread weights. However,
it does seem to confirm that the ovens are not a significant cause for variation in the bread
weights. The boxplot for the Oven-Baker combinations is presented in appendix 11.

In conclusion, even though at first it might seem that bakers have an impact on the bread weight,
after the cross-analysis it is shown that this happens due to the usage of different flour types by
different bakers. For example, if João were to use flour type A instead of B, the results might have
not yielded a significant difference between him and Anna. If all bakers were to use both types
of flour, better conclusions could be drawn regarding the impact of the baker in the response
variable.

5. Statistical Control

Control charts are one of the fundamental tools of Quality Management, allowing the responsible
for the process to easily identify significant deviations from the “normal” functioning. If some
observation falls out of the control limits, the responsible can immediately try to identify the root
cause of the problem, which can be due to an assignable cause (type II error) or just a type I error (a
false negative). In the former case, the responsible should then act to solve the problem and bring
the process back to statistical control.

In this case, a variable control chart should be drawn, controlling for the bread weight across the
time series. Since the variable under investigation is the type of flour used, the control chart should
take this into account. For each day there’s a sample with 20 observations (N=20) of each type of
flour (A and B) per day, for a total of 20 samples across the time-series (K=20), hence, the choice of
the Xbar-S chart to analyze the process, with no standard given, i.e. unknown population parameters.
Figure 6 shows the control chart designed.

7
Provesende Bakery: a case study

Figure 6 Xbar-S chart of Weight (20 samples)

To understand the charts and assess if the process is under statistical control, firstly one should
analyze the S chart. The Xbar chart control limits are only valid if there are no samples out of control
according to the S chart. The S chart is measuring the within-sample variability, which should be
only caused by random causes. In this case, it is possible to conclude that the Xbar control limits are
valid. Moving to the observations in this control chart, it’s possible to see some observations outside
the control limits. This was to be expected since it was previously identified that flour type does
have an impact on the bread weight.

The next step should be to separate the data by flour type, presenting the Xbar-S control charts for
each of the flour types separately. Appendix 1 shows the control chart for flour type A. As seen in
appendix 1, in sample 6 the process was out of control. This sample was removed from the
calculations (assuming that the reason behind this abnormal value was an assignable cause of
variation) and the new control limits were computed. The new control chart is shown in Figure 7 (for
flour type A), while figure 8 shows the control chart for flour type B, which was under control in the
first place.

8
Provesende Bakery: a case study

Figure 7 Xbar-S Chart of Weight – flour type A

Figure 8 Xbar-S Chart of Weight – flour type B

Now that both processes are under control, it’s interesting to observe the differences between the
control limits of the Xbar charts. For flour A, the control limit (average of the samples’ averages) is
300,026, while for flour B is 304,320. Note that this value would be out of the control limits
computed for A. To assess how do these different flour types behave when compared to the target
value (300g per unit), a study of the process capability follows.

9
Provesende Bakery: a case study

6. Study of the capability and performance of the process

By now it has been established that flour type has an effect on the dependent variable (weight). To
assess the process capability, the data were then divided by flour type, just like it was done previously
for the control charts. For each dataset, the group ran a six-pack capability analysis. Figure 9 presents
the capability analysis for samples of flour A.

Figure 9 Sixpack Report for Weight for flour type A

Firstly, it’s important to make sure that the weights follow a Normal distribution, which is confirmed
by the Anderson-Darling test for normality (p-value=0,367). Looking at the capability histogram,
which compares the sample distribution to the specification limits, it’s visible that the process is not
capable, which is then confirmed by the Cp (short-term capability), which is 0,72, and by the Pp
(long-term capability), which is 0,73. These are values well below the benchmark 1,3 process
capability. It’s also worth noting that in this case, Cpk is equal to Cp (and Ppk is equal to Pp) since
the weights are centered in 300g, which is the target for this process. The difference between Cp
and Pp is also small, since the variability within groups is very similar to the variability between
groups. On the other hand, when all the data is analyzed together (Appendix X2) the variability
between groups is considerably larger than the variability within groups, since there are samples
from both flour types.

Moving on to the second dataset, concerning flour B, the capability six-pack analysis is presented in
figure 10. Once again, the normality of the response variable is assured. However, just like suspected,
the weights are off-target, as seen in the capability histogram and confirmed by the difference
between Cp (0,59) and Cpk (0,33) and between Pp (0,58) and Ppk (0,33). For flour B, not only are the

10
Provesende Bakery: a case study

values more off-target, which explains the lower Cpk and Ppk, but also the variability is greater, which

explains the lower Cp and Pp.

Figure 10 Sixpack Report for flour type B

After this analysis, one can conclude that a change in flour type has serious impacts on the process
capability and in the quality level of the bread. Thus, it’s important to assess if it’s possible to detect
this change, i. e. to assess the power of the test. Using the control limits of the process under control
for flour A and the parameters for flour B (mean and overall standard deviation), it’s possible to
calculate the power of the test, which is 70,3%. Consequently, ARL (Average Run Length) until this
change is detected is 2. Although it might seem a high power of the test, it seems reasonable since
the average weight of type B flour breads is 304,320g, which is above the Upper Control Limit for
flour type A weights (around 303g).

11
Provesende Bakery: a case study

7. Conclusions
After having conducted the individual analyses, it could be concluded the different types of ovens
had no impact on the variability in quality and constancy of the bread. It seemed that both Flour type
and Baker did have an influence on the variance in the weight of the bread. However, from the cross-
analysis it could be seen that the difference in Bakers is only there because the bakers use different
flour types; João and Peter both use flour type B and produce heavier breads, whereas Ana and Leo
both use flour type A and have in general lighter breads. From the capability test it could also be
seen that the average weight of the bread using Flour type B was much higher than the average
weight of Flour type A. Therefore, the overall conclusion of the research is that the only variable that
has a great impact on the variability in the weight of the bread is the flour type.

This is very valuable to know for the bakers, since it is an issue that can rather easily be solved. It is
recommended to use only one type of flour for all bakers, preferably flour type A since the breads
that are made with this flour type are closest to the desired 300g of bread.

12
Provesende Bakery: a case study

8. Appendices

Appendix 1

Appendix 1 Xbar-S Chart of Weight

Appendix 2

Appendix 2 Sixpack Report for Weight

13
Provesende Bakery: a case study

Appendix 3

Appendix 3 Boxplot for Weight per baker

Appendix 4

Appendix 4 Boxplot of Weight per Oven

14
Provesende Bakery: a case study

Appendix 5

Appendix 5 Boxplot of Weight per flour type

Appendix 6

Appendix 6 Tukey Intervals for Oven

15
Provesende Bakery: a case study

Appendix 7

Appendix 7 Tukey Intervals for Baker

Appendix 8

Appendix 8 Tukey Intervals for Flour Type

16
Provesende Bakery: a case study

Appendix 9

Appendix 9 Boxplot of Weight for Flour Type and Oven

Appendix 10

Appendix 10 Boxplot of Weight for Baker and Oven

17

You might also like