Project Report of DOE Course - Tsinghua · 2017. 4. 26. · Design of experiment ... At first we...
Transcript of Project Report of DOE Course - Tsinghua · 2017. 4. 26. · Design of experiment ... At first we...
Project Report of DOE
Course -- Research on Factors Affecting the Stickiness of
Home-made Paste
Group Leader:曾夏明
Group Members:刘宝龙,赵妤婕,毛茅,闫博,权
宣亨
2010/6/20
Contents
1. Motivation ................................................................................................................................. 2
2. Literature Review ...................................................................................................................... 2
2.1 The history of home-made paste .................................................................................. 2
2.2 Principle of making paste: ............................................................................................ 3
2.3 Common methods of making paste .............................................................................. 3
2.4 Procedure of making paste ........................................................................................... 4
3. Factor Decomposition ............................................................................................................... 4
3.1 Assumption .................................................................................................................. 4
3.2 Variable definition ........................................................................................................ 4
3.2.1 Response variable ................................................................................................ 4
3.2.2 Factors interested ................................................................................................. 5
3.2.3 Noise factor .......................................................................................................... 5
3.2.4 Constant factor ..................................................................................................... 5
4. Design of experiment ................................................................................................................ 5
4.1 Sample size analysis ..................................................................................................... 6
4.2 Factor level definition .................................................................................................. 6
4.3 Implementation of the experiments .............................................................................. 7
4.4 The results of the experiment ....................................................................................... 9
5. Data analysis ........................................................................................................................... 10
5.1 Basic analysis ............................................................................................................. 10
5.1.1 Main Effects ....................................................................................................... 10
5.1.2 Interaction Plot ................................................................................................... 11
5.1.3 Half Normal Plot ................................................................................................ 11
5.1.4 Pareto Plot .......................................................................................................... 12
5.1.5 Half Normal Standardized Effects: .................................................................... 13
5.1.6 ANOVA ............................................................................................................. 13
5.1.7 Residual Plot ...................................................................................................... 14
5.1.8 Contour Plots of Result ...................................................................................... 15
5.1.9 Response Surface Plot ....................................................................................... 15
5.2 Further analysis .......................................................................................................... 16
5.3 Optimization design ................................................................................................... 17
6. Conclusion .............................................................................................................................. 18
7. Possible improvement for our experiment .............................................................................. 18
8. Reference ................................................................................................................................ 18
1. Motivation
Home-made paste is a Chinese tradition which could date back to ancient times.
The use of paste is to paste couplets and framed painting.
The reason that we choose home-made paste for the prototype of the Design of
Experiment project is that the theorem and process of making paste is relatively
simple and easy to handle. Moreover, the procedure of manufacturing is convenient to
control for the reason that there’re relatively fewer noise factors. As a result we form
the process of making home-made paste as a manufacturing process, and are trying to
explore some of the factors that affect the quality and feasibility of paste.
2. Literature Review
2.1 The history of home-made paste
Home-made paste is mostly used as the adhesion of framed painting throughout
Chinese history. The techniques for making paste have had received great
achievements in the ancient times. The most famous pieces are:
“Famous Paintings in History” (《历代名画记》) from Yanyuan Zhang in
Tang Dynasty
“Evaluation of Ancient Framed Painting” (《论鉴赏装裱古画》) in Song
Dynasty
“Record of Framing” (《装潢志》) from Jiazhou Chou in Ming Dynasty
It is said in “Zhuang Huang Zhi” that:
We know that the techniques of making paste do not only focus on the quality of
adhesion, but also on the maintenances of it. Moreover, the level of paste that is
pleasing to the eye is also an important criterion of home-made paste for framed
paintings.
“表于之糊,犹墨之于胶。墨以胶成,表以糊就。胶用善则灵液清虚, 糊用仁则卷舒温
适。调用之宜,妍媸攸赖。良之用糊如水,止在多刷,刷多则水泌透纸。凝结如抄成者,不全
恃糊力矣。”
2.2 Principle of making paste:
The material of making paste is flour. When flour is solved in water and formed
as glue, the original phase of paste is achieved. Of course, the addictive to paste varies
greatly corresponding to the purpose of paste. It is obvious that paste for framed
painting is more refined than that of daily use. In order for the feasibility of making,
we simplified the components of paste:
Starch
It is insoluble in water, alcohol and ether, but its cell membrane will rupture
when heated together with water, swelling into a transparent paste.
NaOH
It vaginally discharges the water molecules in starch granules, producing
expansion of the viscosity of starch as well as lowering the gelatinization
temperature.
Phenol
Pest control, anti-corrosion
Water
The environment for the starch to get paste.
2.3 Common methods of making paste
There’re many kinds of methods to make paste. The most commonly used four are:
Water washing Method
Use water in to dissolve the flour and form the glue phase of the paste.
And it is used in our experiment.
Steam Blowing Method
Blowing steam into the flour with water
Boiling in the Pot Method
Boling the paste of flour and water in the pot, the high temperature may lead to
the form of paste.
Agar-paste Method
Agar is a kind of expensive but exquisite material for making paste. The paste
made from Agar is relatively with better quality and appearance. However, the
cost of it is high, which is not suitable for common experiments.
2.4 Procedure of making paste
According to the knowledge of making paste and some chemical principles, we
generate our procedure of making paste.
Figure 1 the procedure of making paste
At first we dissolve NaOH into water and form the NaOH solution, than add flour
to the solution and at the same time pour hot water into it. After that, we add in phenol
and stir the mixture always in the same direction until the paste cools off.
Since we have already made our paste, the next procedure is to add weight and
measure the strength and viscosity of paste correspond to the magnitude of weight.
3. Factor Decomposition
3.1 Assumption
When considering the viscosity of paste, the four factors: temperature of water,
mess ratio of water and flour, value of NaOH, and Phenol are of the interests to
research. According to the literature we found, the addition of Phenol is used for
anti-corrosion, and perhaps has nothing to do with the viscosity.
3.2 Variable definition
3.2.1 Response variable
The viscosity (stickiness) of home-made paste
NaOH SolutionAdd flour to the solution
Add hot water
Add PhenolStir the paste
and let the paste cool off
Add weight and measure
the weight
We measure the viscosity of paste by measuring the strength of paste to bear
pulling force. Namely, we use some force to stretch the paste and see how much the
extreme forces are that the paste can bear. Specifically, we stick a pocket under to the
pasted papers. And then we add coins in the pocket to increase the weights that apply
to the paste. Accordingly, we can convert the force into the weight of coins that
correspond to the force.
3.2.2 Factors interested
Concentration of NaOH Solution
Water Temperature
Mass ratio of Water to flour
Whether or not to Phenol
3.2.3 Noise factor
Factors that we could not handle are the noise factors, and we try to eliminate their
effects on our experiment. The ways that we use generally are blocking and
randomization. The followings are the noise factors in our experiment as far as we
concerned.
Operator variation
Measurement of force
Way to paint the paste
3.2.4 Constant factor
Factors that we are not interested in and can be controlled are set to be constant.
This is a good way to simplify our experiment. The followings are the constant factors
in our experiment as far as we concerned.
Stirring time
The room temperature
Flour lot
Time after pasting
4. Design of experiment
After defining the factors, we continue to come up with a feasible and effective
experiment design. We thought that a full factorial design is good one based on
the following reasons:
a) there are in total four factors, not too much;
b) relationships among all factors are to be identified;
c) all main effects and all interactions are to be estimated;
We can measure the responses by using various combinations of factors and levels in
order to determine the large-effect factors and distinct interactions between factors.
4.1 Sample size analysis
Sample size analysis is important in the design of experiments. However, we can’t
find a way to determine sample size for a four factor experiment.
So intuitively we finally choose an unreplicated full factorial design for the
following reasons:
a) The experiment takes relative a long time;
b) Caustic materials are used, such as NaOH and Phenol, the more trials, the
more risk.
4.2 Factor level definition
In order to take a full factorial design, we need to set all the factors at two
levels—low and high. Considering it is an unreplicated experiment, we do not want
the noise to overwhelm the signal in data. Thus the distance between two levels of
factors should be appropriately increased. The problem comes that how to determine
it? Since there is no data to be referred to, we do it by ourselves. Some tentative
experiments are done for the purpose to find a relatively accepted distance. We try
several levels of concentration of NaOH, water temperature and mass ratio of water to
flour, and obvious difference of the pastes is obtained to increase the span between
two levels. The result is shown in the following diagram:
Factor\level Low High
NaOH 0.00001mol/L 1mol/L
Water_Temp 60°C 80°C
Water_to_flour 4:3 2:1
Phenol 0g 5.0g
Table 1 Low-high levels of factors
4.3 Implementation of the experiments
With the help of Minitab, we create a factorial design. The combination of factors
and orders, the run order are given by Minitab.
StdOrder RunOrder NaOH Water_Temp Water_to_Flour Phenol
15 1 1 1 1 -1
1 2 -1 -1 -1 -1
5 3 -1 1 -1 -1
7 4 -1 1 1 -1
13 5 1 1 -1 -1
12 6 1 -1 1 1
14 7 1 1 -1 1
8 8 -1 1 1 1
10 9 1 -1 -1 1
2 10 -1 -1 -1 1
4 11 -1 -1 1 1
6 12 -1 1 -1 1
16 13 1 1 1 1
9 14 1 -1 -1 -1
3 15 -1 -1 1 -1
11 16 1 -1 1 -1
Table 2 2^4 full factorial design given by Minitab
After the design is finished, some detailed methods of operation are then taken
into account.
1) Measurement methods. For good accuracy and experiment result, a set of
fine measurement methods are critical. Some of the factors can be well
measured by specific instruments, but some cannot. How to measure the
stickiness of pastes is a difficult problem which confuses us for long time.
The conclusion of discussion is using a cardboard which is attached to a
receptacle which can hold numbers of coins and give a controllable load to
the cardboard. We smear the paste on one side of the cardboard and stick it to
the table. After a several minutes the paste becomes dry and it’s time for
measuring the stickiness.
Factor Method
Stickiness: Force Needed to
make the cardboard fall
Coins Number and the
Weight of the Coins
Concentration of NaOH Solution PH Meter
Temperature of the Water Thermometer
The Ratio of Water to Flour Graduated Cylinder
The Weight of Flour Balance(30.0g)
Table 3 Measurement of factors
2) Preparation of experiment. According to the design plan of the experiment
execution, we prepare necessary materials, including Flour, NaOH, Phenol,
Hot and Cold Water, as well as experiment and measurement instruments:
Plastic cup
50ml Graduated cylinder
100°C Thermometer
Balance
pH meter
Pencil
Rope
Cardboard
Thermos bottle
3) Since we use plastic cups as the receptacle for making paste, we write the
combinations and levels of factors on the lateral of cup in order to
differentiate every paste and the result, as the following picture shown.
4) We perform the experiment Performance according to the experiment flow
chart shown in Figure1. And it is worth notice that the stir of paste should be
consistent in direction for a good result.
Figure 2: The proces of our experiment
4.4 The results of the experiment
After the construction of the experiment, we got the 16 different pastes and then
tested the stickiness of them, and get the following raw data.
The original data set is listed below.
StdOrder RunOrder CenterPt Blocks Phenol Water_to_Flour Water_Temp NaOH result
1 1 1 1 -1 -1 -1 -1 500.2
2 2 1 1 1 -1 -1 -1 636.5
3 3 1 1 -1 1 -1 -1 1023.1
4 4 1 1 1 1 -1 -1 628.9
5 5 1 1 -1 -1 1 -1 651.7
6 6 1 1 1 -1 1 -1 640.3
7 7 1 1 -1 1 1 -1 1090.3
8 8 1 1 1 1 1 -1 847.1
9 9 1 1 -1 -1 -1 1 530.7
10 10 1 1 1 -1 -1 1 741.6
11 11 1 1 -1 1 -1 1 1101.1
12 12 1 1 1 1 -1 1 812.4
13 13 1 1 -1 -1 1 1 736.5
14 14 1 1 1 -1 1 1 884.7
15 15 1 1 -1 1 1 1 1267.5
Figure 3: Results
16 16 1 1 1 1 1 1 1653.5
17 17 0 1 0 0 0 0 832.9
18 18 0 1 0 0 0 0 964.5
19 19 0 1 0 0 0 0 821.3
Table 4: Original Data Set
5. Data analysis
We want to achieve a model between the mass of coins (response variable) and the
4 factors to interpret the relationships of them and find the effects which influenced
the stickiness of paste.
5.1 Basic analysis
5.1.1 Main Effects
With the help of Minitab, we got the Main Effect Plot:
Figure 4 Main Effects Plot for the Mass of Coins
We can find that all the factors except Phenol have significant affects to the
response variable. And the high level of all the three factors will result in high
stickiness of paste. To get more information of the interaction effects, we had the
interaction plot for result.
5.1.2 Interaction Plot
Figure 5 Interaction Plot among the factors
There are no obvious interactions from the plot even though NaOH and
Water_Temp, NaOH and Phenol have weak interactions. To select key factors which
have great effects on stickiness of the paste, we formed the Half Normal Plot of the
effects.
5.1.3 Half Normal Plot
Figure 6 Half Normal Plot of the Effects
Only C (Water_to_Flour) is significant. But the other factors and interactions
don’t fit the normal probability plot well; maybe there are some other main effects or
interactions. Therefore we had the Pareto Plot to help us to find significant factors.
5.1.4 Pareto Plot
Figure 7 Pareto Chart of the Effects
Then we abandoned some insignificant factors like D, high order terms like
ABCD and so on. In the end, we got the following Half Normal Standardized Effets.
5.1.5 Half Normal Standardized Effects:
Figure 8 Half Normal Standardized Effects
We can find that NaOH, Water_Temp and Water_to_Flour are significant factors,
and no interaction left.
5.1.6 ANOVA
Next figure is our ANOVA table of the experiment:
Figure 9 ANOVA Table
In the ANOVA Analysis, the three factors are significant according to the P value
of them. Then we can formulate our model of the experiment:
To replace the binary variable with the actual ones, we get:
The P value of “Lack of Fit” indicates that our model fit the linear model well
since it isn’t significant.
5.1.7 Residual Plot
Next step is to analyze the residual of our model and see whether our model is
good or not.
Figure 10 Residual Plots for the model
The residual plots tell us that even though our sample size isn’t large, our residual
is good and our model fits well.
5.1.8 Contour Plots of Result
In order to further explain there is no significant interaction among the factors,
we have the contour plot and the response surface plot below:
Figure 11 Contour Plots of the Response Variable
All the three plots indicate that there aren’t significant interactions so in our
model we can abandon them without hesitation.
5.1.9 Response Surface Plot
Figure 12 Response Surface Plot between NaOH and Water_Temp to Response
The surface plot is generally a plane plot, no significant curvature is available.
This also means that interactions among factors are not significant.
5.2 Further analysis
According to our previous data analysis, there are two interesting findings. First
is that the coefficient of NaOH solution’s concentration is positive. This implies that
more condensed NaOH solution will increase the stickiness of our home-made paste.
However, this contradicts with our research result of literature review. Most literature
recommends us to use dilute resolution. One possible explanation is that the linear
assumption is not satisfied. Our experiment data need to be analyzed more carefully.
Besides, the levels’ setting of phenol is similar to that of a categorical variable.
However, we regard it as a continuous variable in our data analysis.
Thus, we add three center points in our experiment to check the linear assumption
of the factors’ levels. Though new data were collected several days later after the first
16 observations, we still assume these data are supportive for our assumption
checking as the experiment condition was generally the same. The original data set in
the previous sections have already contained the center points obtained afterwards.
The data analysis output from Minitab is listed below.
Figure 13 ANOVA table after adding three points
The p value of curvature shows that the quadratic is not significant. Our linear
assumption is generally satisfied.Our explanation for why literature recommends
dilute solution is that it is safer to use as condensed NaOH solution is corrosive.
More words to add for the follow up experiment. Actually when we have more
data, we generally need to incorporate them in our model construction. However as
they were not collected at the same time, we drop them in our model construction.
The final model is only based on the previous 16 observations.
5.3 Optimization design
Our experiment is not intended for an optimization research. Specifically, there
are several optimization criteria for the home-made paste, such as low manufacturing
cost at a target stickiness level, high stickiness etc. Different optimization criterion
requires different experiments. However, our collected observations are appropriate
for us to analyze the optimized treatment combination for higher stickiness.
The output from Minitab is listed below.
Figure 14 Response optimization
Figure 15 Response optimization with desirability
We can imply that the treatment of high level of NaOH, Water_temp,
Water_to_Flour provides highest stickiness.
6. Conclusion
From our experiment, several useful conclusion can be obtained.
The concentration of NaOH, temperature of the water and mass of water to
flour affects the stickiness of home-made paste a lot.
One of the best way to produce high stickiness of the paste is use high level of
NaOH, temeperature of water and mass of water to flour.
7. Possible improvement for our experiment
Our experiment follows the basic design method of full factorial. However,
basically it can be regarded as a DIY experiment as resources are limited. Several
improvements can be made to gain better result for our research.
Better Gauge R&R System
Initial Design with center points
Besides, what we want to emphasize is that safety is the most important issue in
experiments, especially when chemicals are involved.
8. Reference
Design and Analysis of Experiments (2005), Sixth Edition, Douglas C. Montogomery,
John Wiley
实验设计与分析及参数优化 (2003), C. F. Jeff Wu, Michael Hamada 著 张润楚等
译 中国统计出版社