You are on page 1of 63

Example 1 - One Factor Design

Background
Consider the following experiment, which was performed at a pulp mill. Plant performance is
based on pulp brightness, as measured by a reflectance meter. Each of the four shift operators
(denoted by A, B, C and D) made five pulp handsheets from unbleached pulp. Reflectance was
read for each of the handsheets using a brightness tester. The data set is shown next.

The goal of the experiment is to determine whether there are differences between the operators in
making the handsheets and reading their brightness.
Experiment Design
The experimenters use DOE++ to create a standard one factor design. The design matrix and the
response data are given in the "Operator Study" Folio.
Analysis
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

Note that in the figure shown above, the columns have been renamed by double-clicking the
column header and entering the new name in the window that appears.
Step 2: The data set is analyzed with the default risk (significance) level of 0.1. Several tables of
results are provided on the Analysis tab, which is added to the Folio upon calculation. The first one
is the ANOVA table, which is shown next.

The p value is less than the risk level (0.021 < 0.1), indicating that there are significant
differences between these four operators.
Step 3: A Box plot is created, as shown next.

The plot shows that operators C and D are almost the same. This is confirmed by the Mean
Comparsions table on the Analysis tab, as shown next.

The table also shows that there is no significant difference between operators A and B. The small
p values displayed in red indicate significant difference between the compared operators.
Conclusions
From the ANOVA table, the experimenters determine that there is a significant difference between
the four operators. The paired comparison shows that the difference between A and B and the
difference between C and D are not particularly significant.

Example 2 - Two Level Full Factorial Design
Background
Consider that one of the initial steps in fabricating integrated circuit (IC) devices is to grow an
epitaxial layer on polished silicon wafers. The wafers are mounted on a six-faceted cylinder (two
wafers per facet), called a susceptor, which is spun inside a metal bell jar. The jar is injected with
chemical vapors through nozzles at the top of the jar and heated. The process continues until the
epitaxial layer grows to a desired thickness. The nominal value for thickness is 14.5 m with
specification limits of 14.5 0.5.
The current settings caused variations that exceeded the specification by 1.0 m. Thus, the
experimenters first need to find the factors that affect the process. Then further experiments will
be conducted to optimize the process. There are four experimental factors.

Experiment Design
To find the important factors, the experimenters use DOE++ to design a two level full factorial
design. The design-specific settings, the factor properties and the response properties used are
shown next.



The design matrix and the response data are given in the " 2 Level Full Factorial Design" Folio. To
view the data set, click here.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

The data set as entered in the DOE++ Folio.
Note that not all rows are shown in the figure above. There are 96 rows of data in the Folio.
Step 2: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms and including all effects up to four-way interactions in the analysis.
Step 3: A Pareto chart is created, as shown next.

The Pareto chart shows that effects B, D and CD are significant.
Analysis Part II
The results for the reduced model and the optimization are given in the "Optimization" Folio.
Step 1: The design Folio is duplicated and the copy is named "Optimization."
Step 2: Only the significant effects are selected to calculate the new model, as shown next.

Note that selections in the Effects window must be hierarchical (i.e., for any second order or
greater effect selected for inclusion, all related main effects must also be included). Thus, effect C
must be included in order to include CD in the model.
Step 3: The reduced model is calculated. The Regression Information table from the Analysis tab
is shown next.

Step 4: Optimization is performed using the settings shown next.

The optimization results are:

Analysis Part III
Although both optimum solutions give the same predicted thickness, the variability at these two
settings may be different. It is important to identify which solution is better in terms of variability.
To do this, the experimenters study the variability of each setting. In variability analysis, the
response is the standard deviation of the observations at each setting.
The results of the variability analysis are given in the "Variability Analysis" Folio.
Step 1: The design Folio is duplicated and the copy is named "Variability Analysis."
Step 2: The Thickness response is selected for variability analysis and all factors are selected to
be considered in the variability analysis, as shown next.

This results in a new response column called Thickness Std. being added to the Folio. The values
in this column are standard deviations of the observations at each factor setting.
Step 3: The Thickness Std. column is selected for inclusion in the analysis by selecting the
checkbox in the column header and the checkbox in the Thickness column header is cleared.
Step 4: The log transformation is applied to the Thickness Std. response, as shown next. This is
generally appropriate if a response is standard deviation.

Step 5: The model is calculated. The Regression Information table from the Analysis tab is shown
next.

Applying this model to the two optimal solutions in the analysis for the Thickness response shows
that the first solution has less variability.
The predicted results for the Thickness and Thickness Std responses are given in the "Optimal
Solution" Spreadsheet, as shown next.

Conclusions
The Pareto chart shows that effects B, D and CD are significant for thickness. Using the reduced
model for the optimization, the optimum settings for factors B, C and D are:

By considering the expected variability for these two solutions, the first solution is found to be the
best one.
For more discussion on how to optimize both mean response and variability of response, please
refer to Wu and Hamadas book, Experiments: Planning, Analysis, and Parameter Design
Optimization, John Wiley & Sons, New York, 2000.

Example 3 - Two Level Fractional Factorial Design
Background
Consider a manufacturing process for an integrated circuit.* The objective is to improve the
process yield. The five factors that may affect the process are:

It is too expensive to run a full factorial design, which has 2
5
= 32 runs. Therefore, the engineers
decide to run a half fractional factorial design using generator E = ABCD.
Experiment Design
The engineers use DOE++ to design a two level fractional factorial design. The design-specific
settings, the factor properties and the response properties used are shown next.



The design matrix and the response data are given in the "Fractional Factorial Design Model 1"
Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
engineers enter the data set into the Standard Folio, as shown next.

The data set as entered in the DOE++ Folio.
Step 2: In the Select Effects window, all effects up to two-way interactions are selected for
inclusion, as shown next.

Step 3: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms.
Step 4: An Effect Probability plot is created, as shown next.

The Effect Probability plot shows that effects A, B, C and AB are significant.
Analysis Part II
The results for the reduced model are given in the "Fractional Factorial Design Model 2" Folio.
Step 1: The design Folio is duplicated and the copy is named "Fractional Factorial Design Model
2."
Step 2: In the Select Effects window, only the significant effects are selected to calculate the new
model, as shown next.

Step 3: The reduced model is calculated. The coefficients for A, B, C and AB, found in the
Regression Information table in the Analysis tab, are:

Conclusions
As shown in the Regression Information table in the Analysis tab, each of the effects in the
reduced model has a positive effect on the yield. Therefore, in order to increase the yield, the high
level setting of factors A, B, and C should be applied.
* Montgomery, D. C. Design and Analysis of Experiments, 5th edition, John Wiley & Sons, New
York, 2001.

Example 4 - Plackett-Burman Design
Background
Plackett-Burman design is one of the so-called "screening designs." Such designs are traditionally
used for identifying important factors from among many potential factors. In the analysis of these
designs, usually only main effects are estimated.
Consider a life testing of weld-repaired castings.* The objective of the test is to identify the
important factors that affect the life and to improve the product life. There are seven factors that
may affect the life. A two level full factorial design will require 2
7
= 128 runs. It will be time-
consuming and costly. Therefore, an eight run Plackett-Burman experiment will be conducted.
For this example, the seven factors are:

The response is the failure time of each sample. The logarithmic transformation of the failure time
is used in the analysis.
Experiment Design
The experimenters use DOE++ to design a Plackett-Burman design. The design-specific settings,
the factor properties and the response properties used are shown next.



The design matrix and the response data are given in the "Cast Fatigue Experiment" Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

The data set as entered in the DOE++ Folio.
Step 2: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms.
Step 3: An Effect Probability plot is created, as shown next.

The Effect Probability plot shows that effect F is significant.
Conclusions
The effects of the factors are given below:

As shown in the Regression Information table on the Analysis tab, assuming that there is no
interaction, a higher product life can be achieved by setting A, B, C, D and E at their respective
low levels and F and G at their respective high levels. Otherwise, further experiments can be
conducted to study the interaction effects of those factors.
Factor F was found to be the most important factor.
* Wu, Jeff and Hamada, Michael, Experiments: Planning, Analysis, And Parameter Design
Optimization, John Wiley & Sons, New York, 2000.

Example 5 - General Full Factorial Design
Background
General full factorial design is used when there are several factors (<5) that have multiple levels.
If there are many multiple level factors, the size of a general full factorial design will be
prohibitively large. In such cases, Taguchi OA design should be used.
Consider that a soft drink bottler is interested in obtaining more uniform fill heights in the
bottles.* The filling machine theoretically fills each bottle to the correct target height, but in
practice, there is variation around this target, and the bottler would like to understand better the
sources of this variability and eventually reduce it. There are three control factors.

There are two replicates at each factor setting, making a total of 24 runs. The response is the
average deviation from the target fill height observed in a production run of bottles. Positive
deviations are fill heights above the target.
Experiment Design
The experimenters use DOE++ to design a general full factorial design. The design-specific
settings, the factor properties and the response properties used are shown next.



The design matrix and the response data are given in the "Soft Drink Bottling Experiment" Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

The data set as entered in the DOE++ Folio.
Step 2: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms. The ANOVA table from the Analysis tab is shown next.

This table shows that effects A, B, C and AB are significant.
Analysis Part II
The results for the reduced model and the optimization are given in the "Reduced Model" Folio.
Step 1: The design Folio is duplicated and the copy is named "Reduced Model."
Step 2: In the Select Effects window, only the significant effects are selected to calculate the new
model, as shown next.

Step 3: The reduced model is calculated.
Step 4: To identify which factor settings can provide the smallest height deviation, the
Diagnostics window is used, as shown next.

Step 5: These results are copied to a Spreadsheet for further analysis. The data are given in the
"Best Setting" Spreadsheet, as shown next.

Conclusions
The Spreadsheet shows that the runs at run orders 12 and 23 (i.e., standard orders 5 and 17) are
at the best combination of settings, which is A = 12, B = 25, C = 200. At these settings, the
expected deviation is lowest.
In general full factorial design, all factors are assumed to be qualitative factors, which means that
the factors can only take the discrete values defined in the design, which in this case are:

If the factors can be treated as quantitative factors, meaning they can take any value within a
range, further analysis using response surface methodology should be conducted to find the
optimal settings for the manufacturing process.
* Montgomery, D. C. Design and Analysis of Experiments, 5th edition, John Wiley & Sons, New
York, 2001.
Example 6 - Taguchi Orthogonal Array Design
Background
Taguchi orthogonal array (OA) designs are often used in design experiments with multiple level
factors. Taguchi OA can be thought of as a general fractional factorial design.
Consider an experiment to study the effect of four three-level factors on a fine gold wire bonding
process in an IC chip-package.* Taguchi OA L27 (3^13) is applied to identify the critical
parameters in the wire bonding process. The response is the ball size. The smaller the ball size,
the better the process.
For this example, the four factors are:

These four factors are assigned to columns 1, 2, 5 and 8 in L27. The remaining columns either
represent the interaction effects of these factors or are treated as dummy factors. For example,
columns 3 and 4 are the interaction of AB. Columns 6 and 7 are the interaction of AE. Columns 9
and 10 represent the effect of AH. Those columns can be deleted from the design matrix and will
not affect the analysis results. More detailed discussion on the properties of Taguchi arrays can be
found in the appendix of Taguchis handbook.**
Experiment Design
The experimenters use DOE++ to design a Taguchi OA design. The design-specific settings, the
factor properties and the response properties used are shown next.



Once the Standard Folio has been created, all unused factor columns are deleted, keeping only
the Force, Power, Time and Temperature columns in the design matrix. The Time column then
becomes factor C and Temperature becomes factor D.
The design matrix and the response data are given in the "Taguchi OA L27(3^13)" Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

The data set as entered in the DOE++ Folio.
Step 2: The following effects are selected for inclusion in the analysis:

Step 3: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms. The ANOVA table from the Analysis tab is shown next.

This table shows that effects A, B and C are significant.
Analysis Part II
The results for the reduced model and the optimization are given in the "Reduced Model" Folio.
Step 1: The design Folio is duplicated and the copy is named "Reduced Model."
Step 2: Only the significant effects are selected to calculate the new model, as shown next.

Step 3: The reduced model is calculated.
Step 4: To identify which factor settings can provide the smallest ball size, the Diagnostics
window is used, as shown next. The Fitted Value column is the expected ball size for the factor
settings under different runs.

Conclusions
From the Fitted Value column, it is determined that run order 4 (standard order 1) gives the best
result. The predicted ball size is 35.1667. The settings are Force = 5, Power = 40 and Time = 15.
In Taguchi OA Factorial design, all factors are assumed to be qualitative factors, which means that
the factors can only take the discrete values defined in the design, which in this case are:

If the factors can be treated as quantitative factors, meaning they can take any value within a
range, further analysis using response surface methodology should be conducted to find the
optimal settings for the manufacturing process.
* T. Hou, S. Chen, T. Lin and K. Huang, "An integrated system for setting the optimal parameters
in IC chip-package wire bonding process," Int. J Adv Manuf Technology, 2006, 30, 247-253.
** G. Taguchi, S. Chowdhury and Y. Wu, Taguchi's Quality Handbook, Hoboken, New Jersey,
Wiley, 2004.

Example 7 - Central Composite Design
Background
Central composite design is the most commonly used response surface methodology (RSM)
design. RSM design is usually used to study the quadratic effects of factors.
A chemical engineer is interested in determining the operating conditions that maximize the yield
of a process.* Two controllable variables influence process yield: reaction time and reaction
temperature.

A central composite design with five center points and alpha = 1.414 is used to conduct the
experiment. A full quadratic model is fitted to the data.
Experiment Design
The engineer uses DOE++ to design a central composite design. The design-specific settings, the
factor properties and the response properties used are shown next.



The design matrix and the response data are given in the "Central Composite Design" Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
engineer enters the data set into the Standard Folio, as shown next.

Step 2: All effects (i.e., full quadratics) are selected for inclusion in the analysis, as shown next.

Step 3: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms. The ANOVA table from the Analysis tab is shown next.

This table shows that effects A, B, AA and BB are significant.
Step 4: A Pareto chart is created, as shown next.

From these results, only effects A, B and AA and BB would be included in the reduced model. In
fact, term AB could also be included in the model, as it is only slightly below the critical value, as
shown in the ANOVA table and Pareto chart. The inclusion or exclusion of AB is a personal decision
that should be made based on the knowledge of the experiment and the statistical results. For this
example, the engineer decides that only A, B, AA and BB will be included in the model.
Analysis Part II
The results for the reduced model and the optimization are given in the "Reduced Model" Folio.
Step 1: The design Folio is duplicated and the copy is named "Reduced Model."
Step 2: Only the significant effects are selected to calculate the new model, as shown next.

Step 3: The reduced model is calculated. The coefficients for the parameters in the reduced
model are:

This model can be used as the final model to conduct optimization.
Step 4: Optimization is performed using the settings shown next.


The optimal solution is shown next.

Step 5: The contour and surface plot can also be used to visually identify the optimal settings for
factors A and B, as shown next.


Conclusions
The contour and surface plots show that the maximum yield occurs at Time = 86.8 and
Temperature = 176.3F, which is the same as the result from the optimization. The predicted
maximum yield is 80.1861. Keep in mind that it is necessary to conduct an experiment using
these settings to confirm this conclusion.
* Montgomery, D. C. Design and Analysis of Experiments, 5th edition, John Wiley & Sons, New
York, 2001.

Example 8 - Box-Behnken Design
Background
Box-Behnken design is a response surface methodology design. It is used to further study the
quadratic effect of factors after identifying the signficant factors using screening factorial
experiments.
Box-Behnken designs do not contain any points at the vertices of the experimental region. This
could be advantageous when the points on the corners of the cube represent factor-level
combinations that are prohibitively expensive or impossible to test because of physical process
constraints.
Consider a UV-light system that is used to inactivate fungal spores of Aspergillus niger in corn
meal.* Fungal contamination of grains during the post-harvest period has been a recurring health
hazard.
The response is the log
10
reduction of the fungal spores. Therefore, the goal is to maximize the
reduction (response).
Three process parameters in the UV-light system will affect the inactivation results. They are: A)
treatment time (number of pulses), B) the distance from the UV strobe and C) input voltage for
the UV lamp.

A 15 run Box-Behnken design with three center points is conducted. A full quadratic model is
fitted to the data. Using this model, the optimal setting that gives the largest reduction of fungal
spores was found.
Experiment Design
The experimenters use DOE++ to design a Box-Behnken design. The design-specific settings and
the factor properties used are shown next.


The design matrix and the response data are given in the "UV-light Treatment" Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

Step 2: All effects (i.e., full quadratics) are selected for inclusion in the analysis:

Step 3: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms. The ANOVA table from the Analysis tab is shown next.

This table shows that effects A, C, AC and AA are significant. The p value for factor B is 0.1481,
which is close to the risk level 0.1. Therefore, the experimenters decide that it will also be
included in the final model.
Analysis Part II
The results for the reduced model and the optimization are given in the "Reduced Model" Folio.
Step 1: The design Folio is duplicated and the copy is named "Reduced Model."
Step 2: In the Select Effects window, only the significant effects are selected to calculate the new
model, as shown next.

Step 3: The reduced model is calculated. The coefficients for the parameters in the reduced
model are:

This model can be used as the final model to conduct optimization.
Step 4: Optimization is performed using the settings shown next.


The optimal solution is shown next.

Conclusions
The optimal solution is found to be A = 100 s, B = 3 cm and C = 3800 v. Under these settings,
the expected logarithmic transformation of the reduction is 4.9. Keep in mind that it is necessary
to conduct an experiment using these settings to confirm this conclusion.
* S. Jun, J. Irudayaraj, A. Demirci and D. Geiser, "Pulsed UV-light treatment of corn meal for
inactivation of Aspergillus niger spores," International Journal of Food Science and Technology,
2003, 38, 883-888.

Example 9 - Taguchi Robust Design
Background
Taguchi robust design is used to find the appropriate control factor levels in the design to make
the system less sensitive to variations in uncontrollable noise factors (i.e., to make the system
robust).
Consider an experiment that seeks to determine a method to assemble an elastomeric connector
to a nylon tube while delivering the requisite pull-off performance suitable for an automotive
engineering application.*
The primary design objective is to maximize the pull-off force, while secondary considerations are
to minimize assembly effort and reduce the cost of the connector and assembly.
The controllable and noise factors are:


For the inner (control) array, Taguchi OA L9 (3^4) is used. For the outer (noise) array, a two level
full factorial design is applied.
Experiment Design
The experimenters use DOE++ to design a Taguchi robust design. The settings and factor
properties used for the inner (control) array are shown next.


The settings and factor properties used for the outer (noise) array are shown next.


The design matrix and the response data are given in the "Taguchi Robust Design Example" Folio.
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
experimenters enter the data set into the Standard Folio, as shown next.

The data set as entered in the DOE++ Folio.
Step 2: The larger-the-better ratio is chosen for the analysis.
Step 3: The data set is analyzed with the default risk (significance) level of 0.1, using individual
terms, for each of the three responses (Y Mean, Y Std and Signal Noise Ratio).
Step 4: The Main Effect plot is created for the Y Mean (pull-off force) response using the Data
Means setting, as shown next.

Step 5: The Main Effect plot is created for the Y Std response using the Data Means setting, as
shown next.

Step 6: The Main Effect plot is created for the Signal Noise Ratio response using the Data Means
setting, as shown next.

Conclusions
Since the goal is to maximize both signal-to-noise ratio and the pull-off force, some trade-off may
need to be made in the selecting of factor settings. Examining the main effect plots for Y Mean
and Signal Noise Ratio shows that the medium level for A is clearly the best choice for maximizing
signal-to-noise ratio (robustness) and the average pull-off force (Y Mean). It also has a relatively
low standard deviation value. For the wall thickness, B, the medium and high level are slightly
better than the low level for signal-to-noise ratio; however, the medium is preferred to high level
in order to maximize the average pull-off force. From the main effect plot, the better settings for
factors C and D can also be determined. In the final analysis, the best settings to maximize
signal-to-noise ratio are A (medium), B (medium), C (deep) and D (low), based on the
experimental results for maximizing pull-off force.
* G. Taguchi, "The Taguchi Approach to Parameter Design," Quality Progress, Dec. 1987, pp. 19-
26.

Example 10 - One Factor Reliability Design
Background
One factor reliability DOE can be used in design comparison and factor effect identification.
Suppose that there are three different materials that can be used in a product. The engineer
wants to know if there is a difference between these three choices and, if there is a difference,
which material is the best choice in terms of the product life.
Ten samples were tested for each material. The test was stopped at 500 hours.
Experiment Design
The engineer uses DOE++ to design a one factor reliability design. The design-specific settings
and the factor properties used are shown next.


The design matrix and the response data are given in the "3 Levels One Factor" Folio..
Analysis Part I
Step 1: After performing the experiment according to the design and recording the results, the
engineer enters the data set into the Standard Folio, as shown next.

Step 2: The Weibull distribution is chosen for the analysis.
Step 3: The data set is analyzed with the default risk (significance) level of 0.1. The Likelihood
Ratio Test table from the Analysis tab is shown next.

This table provides the test for the overall effect of the factor. A small p value indicates that the
factor (treatment) has a significant effect on the response.
The second table is the MLE Information table, as shown next.

This table gives the value of the estimated parameters in the model. A small p value indicates that
the parameter is significantly different from 0.
The third table, the Life Characteristic Summary table, displays the fitted life characteristic. For
the Weibull distribution, the life characteristic is eta, the scale parameter, as shown next.

The last table, the Life Comparisons table, shows paired comparisons of the life at each level, as
shown next.

Step 4: A Response vs. Level plot is created, as shown next.

From this plot, it is difficult to tell which material is the best.
Step 5: A Life Characteristic plot is created, as shown next.

This plot shows that material type C has the largest eta value, which means it has the longest
expected life.
Conclusions
The Likelihood Ratio Test Table shows that there is a significant difference between the three
types of material. Most notably, from the paired comparison, it is apparent that there are
significant differences between A and B and between A and C. The difference between B and C is
not particularly significant.

Example 11 - Two Level Fractional Factorial Reliability Design
Background
Consider an experiment to improve the reliability of fluorescent lights. The purpose was to find the
best factor setting to improve the life time. Five two-level factors denoted by A-E were studied
using a 2
5-2
design with the factor generator D = AC and E = BC. In addition to the main effects,
the experimenter also thought that the AB (= DE) interaction might be important.
Each treatment had two replicates (i.e. two lights were tested) and the experiment was conducted
over 20 days with inspections every two days (the failure times were short because the lights
were subjected to an accelerating factor that stressed the lights at higher than normal conditions).
If a light had not failed by the 20
th
day, its failure time was recorded as right censored at 20.
Experiment Design
The engineer uses DOE++ to design a two level fractional factorial reliability design. The design-
specific settings used are shown next.

The design matrix and the response data are given in the "Fluorescent Light Life Test" Folio.
Analysis
Step 1: After performing the experiment according to the design and recording the results, the
engineer enters the data set into the Standard Folio, as shown next.

Step 2: In the Select Effects window, the following effects are selected to include in the analysis:

Step 3: The Lognormal distribution is chosen for the analysis.
Step 4: The data set is analyzed with a risk (significance) level of 0.05, using individual terms.
The model coefficients, shown in the MLE Information table on the Analysis tab, are:

Step 5: A Pareto chart is created, as shown next.

Conclusions
From the MLE Information table, it is determined that the model for the ln-mean or the scale
parameter, m, in the distribution is:

The engineer determines that in order to improve the reliability, factors B, C and D should be set
to their low (-1) level, indicated by their negative coefficients, while A and E should be at their
high (+1) level, indicated by their positive coefficients. This setting is also good for the AB
interaction term. Under this setting, the predicted scale parameter in lognormal distribution is
3.7829. Therefore, the life distribution under this factor setting is a lognormal distribution with
standard deviation of 0.1589 and ln-mean of 3.7829.
For more advanced analysis, the data can be entered into ReliaSofts ALTA for accelerated life
data analysis for further investigation of the life stress relationship. Since factors A and C are not
shown to be significant, they can be removed in further analysis.

You might also like