This article was downloaded by: [182.74.116.

110] On: 01 July 2015, At: 00:51

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA

INFORMS Transactions on Education

Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org

An Interactive Spreadsheet-Based Tool to Support

Teaching Design of Experiments
S. T. Enns,

To cite this article:

S. T. Enns, (2008) An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments. INFORMS Transactions on
Education 8(2):55-64. http://dx.doi.org/10.1287/ited.1080.0008

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact [email protected].

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.

Copyright © 2008, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
 Vol. 8, No. 2, January 2008, pp. 55–64
issn 1532-0545
08
0802
0055 informs ®

doi 10.1287/ited.1080.0008
I N F O R M S © 2008 INFORMS
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

Transactions on Education

An Interactive Spreadsheet-Based Tool to

Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

Support Teaching Design of Experiments

S. T. Enns
Department of Mechanical and Manufacturing Engineering, University of Calgary,
Calgary, Canada, T2N 1N4, [email protected]

T his paper describes an interactive spreadsheet-based tool that can be used to generate data representative
of the type that might be obtained running a structured set of experiments. The purpose of this tool is to
help the user experience the iterative nature of design and analysis of experiments. The tool supports quick
and simple generation of data for one and two-factor problems. The underlying relationships are based on
queuing approximations for a single-stage batch production environment. Factor levels are related to product
lot sizes and the response is assumed to be average lot flowtimes. Variability due to replication is emulated by
sampling from a statistical distribution. Statistical software packages can be used to generate linear or quadratic
models from the results generated. Analysis can include the examination of main and interaction effects or the
optimization of lot sizes to minimize flowtimes.
Key words: design of experiments, central composite design (CCD), response surface methods
History: Received: July 5, 2005; accepted: January 23, 2006. This paper was with the authors 3 months for
2 revisions.

1. Introduction A logical alternative would be to have students

Experimentation and analysis is generally an itera- run actual experiments. However, this approach also
tive and interactive process in real life. This paper presents various challenges. First, it is likely to be
addresses the problem of teaching basic design of very time consuming. Second, proper laboratory facil-
experiments (DOE) methodology from this perspec- ities must be available to run either physical or
simulation experiments. Third, students may be over-
tive. In particular, a spreadsheet-based tool to gen-
whelmed by the process of running experiments, with
erate data representative of that which might be
attention being diverted away from the experimental
obtained using structured experimentation is pre- design and analysis aspects of the exercise. Fourth,
sented. Since the experimental design can be readily it is not that easy to design experimental scenar-
changed to generate new results, the tool supports ios that illustrate the intended behavior. For exam-
following an iterative path in which the analysis of ple, catapults are sometimes used to launch balls and
previous results is used to define further experimen- the distance thrown is then measured as a response.
tation requirements. Changing the catapult settings allows main and inter-
The most common approach in teaching DOE is action effects to be observed. Replication is easy to
to use textbook data sets. These are static and do perform but within-group variances are not likely to
not support capturing the true essence of experi- be equal at different factor settings, making analysis
mental design and analysis as a multiple-stage pro- using analysis of variance (ANOVA) techniques inap-
cess, with each stage being dependent on previous propriate. As well, response surface problems where
convex or concave behavior is exhibited may be hard
results. It is difficult to study the type of problem
to construct. However if resources can be committed
that requires a progression of steps. For example, the
to solving suitable problems, the approach of running
investigator may wish to the change values of the actual experiments can provide a high level of experi-
design points, alter the design type itself or simply ential learning. A comprehensive example of such an
run additional replications to increase confidence lev- exercise is given by Box and Liu (1999).
els. As well, randomization of experiments is gener- This paper suggests a third approach, meant to
ally ignored when solving textbook problems since support an iterative paradigm for teaching experi-
the actual experimentation component is missing. mental design and analysis. Box (1999) provides a
55
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
56 INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS

strong argument for the need to teach sequential Figure 1 Single Stage Production with Batch Arrivals
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

investigation if statistical education is to show rele- Lot flow time

vance to solving real world industrial problems. The Lot queue
Lot service
Q1 = ???
idea here is to generate data quickly with the aid of time
time
Processing
a user-friendly spreadsheet-based tool. The develop- station
ment and use of such a tool is described in the remain-
der of the paper. It is meant to support the study of
interactions effects, curvature and response surfaces
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

ca Machine cd
for single or two-factor scenarios. Mock experiments cs
can be quickly set up, evaluated and then altered for
another round of design and analysis. Qi = Lot size for part type i
ca = Lot interarrival time coefficient of variation
Q2 = ???
cs = Lot service time coefficient of variation
cd = Lot interdeparture time coefficient of variation
2. The Approach
The basic idea behind this approach is to use a set
of quantitative relationships in an underlying model period of time. This will again result in more work in
that exhibit the desired type of behavior. These rela- queue and an increase in flowtimes.
tionships should ideally exhibit linear or nonlinear Lot flowtime behavior is convex with respect to lot
behavior, depending on the factor settings. As well, sizes. A good objective is to minimize the weighted
they should support the demonstration of interaction mean lot flowtimes, W , across all part types by select-
effects. Furthermore, if the relationships are not well ing the best lot sizes, Qi , for each part type i. Since all
known or understood it makes the problem context part types have unique characteristics, the best lot size
more palatable with respect to the need for an exper- combinations are affected by their relative production
imental solution. characteristics as well as the variability of lot interar-
The problem context implemented in this spread- rivals. Therefore, the problem can be viewed as one
sheet model is that of lot size selection in a batch of lot size optimization to be solved using response
production system. A big problem in manufactur- surface methods. The analytical relationships used to
ing is establishing good lot (or batch) sizes for pro- describe this problem are approximate and are not
duction. In batch production facilities it is common well known. Therefore, this is the type of problem
to have multiple part types processed on the same that might well lend itself to experimentation.
machine (or resource). These are capacity-constrained The model embedded in this spreadsheet tool
machines that can process only one part type at a assumes a lot size selection scenario where a sin-
time. It is common for each part type to have unique gle machine is being used to produce batches of
part processing time, lot size and lot setup time char- two part types. The configuration of interest is illus-
acteristics. The machine is typically set up for one par- trated by Figure 1. The lot flowtime relationships
ticular part type and then a lot of parts is processed. embedded in the spreadsheet model are given in the
The lot processing time is equal to the part processing appendix. These relationships may be of interest to
time multiplied by the lot size. The lot service time is operations management or industrial engineering stu-
the lot processing time plus the lot setup time. dents. However, it is not essential to know anything
The arrival of lots of different part types is typically about the underlying relationships in order to use this
stochastic so lot flowtime behavior can be modeled tool if the primary interest is to learn DOE methodol-
using queuing relationships. If a lot of parts arrives ogy. In other words, experimentation can be done in
and the machine is busy, the lot will have to wait in a context-free manner.
queue. It is normal to assume that lots in the queue
will be processed first-come, first-served (FCFS). If
the lot sizes are too small, there will be many setups 3. The Spreadsheet Model
incurred and the utilization, defined as the proportion The spreadsheet implementation is designed to be
of time the machine is busy being set up or process- simple, transparent, and easily modified. An Excel®
ing parts, will be high. The result is that the average workbook1 serves as the user interface for specify-
number of lots waiting for processing may be high. ing inputs as well as for extracting the experimental
This means the average lot flowtime, defined to be results. The user inputs are specified in the Inputs
the lot queue time plus lot service time, will also be worksheet shown as Figure 2. The colored cells are
very high. This drives up total manufacturing times user-defined inputs.
and inventory costs. If the lot sizes are too large, the
machine utilization will be lower but the machine will 1
An example of such a workbook (DOE_Tools.xls) can be found at
be committed to producing one part type for a long http://ite.pubs.informs.org/.
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS 57

Figure 2 Example of Inputs Worksheet

Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

Cells in Range D10:E13 specify the production sce- E13 can be used as inputs to estimate behavior while
nario. The user must specify the demand per unit time designing the experiment, but the values in these cells
for each of the part types, D, the time required to set are over-written when the experiment is run.
up the machine for each specific part type, , the pro- The cells in Range I6:J7 are used to specify the low
duction rate per unit time, P , and the production lot and high factor settings for two-factor, two-level (22 
size, Q. Note that P is the production rate when the experimental designs. These are referred to as facto-
machine is steadily processing the specified part type. rial design points. To run a single factor experiment,
Since there will be some idle time as well as time for one would set the low (−1) and high (+1) settings
setups, the value of P must be larger than D in order equal for one of the lot sizes.
to have a stable system in which all demand is met. The cells in Range I11:I13 are used to specify the
The final user input describing the production sce- experimental design. These inputs represent the num-
nario is in Cell D6. This value specifies the amount of
variability in the stream of lot arrivals to the machine.
It is expressed as a coefficient of variation, defined to Figure 3 CCD with Coded Variables
be the standard deviation of interarrival times divided Q2
by the mean interarrival time. A higher value indi- 0, +1.41

cates greater variability in interarrivals and a longer +1, +1

–1, +1
average queue time. Values in the range of 0.10 to 1.0
Axial
are typically appropriate, with 0.30 often being a good
points
guess for practical purposes.
Formulas in Range D16:D20, which are based on Center
points 0, 0 +1.41, 0
the relationships found in the appendix, reference the –1.41, 0

user input values described. Of particular interest is Axial Q1

points
the estimated value of machine utilization. This value
must be less than 1.0 in order to obtain feasible perfor-
mance. When a structured set of experiments is run,
as described next, new lot size values will be written –1, –1 +1, –1

into Cells D13 and E13 automatically, based on the

0, –1.41
user-defined design. In other words, Cells D13 and
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
58 INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS

Figure 4 CCD with Actual Variables and Replications ation of the observed lot queue times, Wq , when mul-
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

275, 331 (2) Q2

tiple replications are run. Therefore, increasing this
value makes the observed outputs more variable.
200, 300 (2) 350, 300 (2) The “Run Experiments” button on the Inputs Work-
sheet runs the experimental design and generates the
output. This is done by activating Visual Basic for
Application (VBA) macros located within the Excel
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

169, 225 (2) 275, 225 (10) 381, 225 (2)

workbook. In order to make the macros available
Q1
when loading the workbook, the security level setting
should be set to Medium. Changing this setting can be
done by going into Tools on the Main Menu bar, then
200, 150 (2) 350, 150 (2) into Macro and finally into Security. As well, running
the macros requires that the Analysis ToolPak—VBA
275, 119 (2) add-in be available. This can be selected by going into
Tools and then Add-Ins.
ber of replications to run at the factorial, center and The results for the experimental design will be
axial points of the design, assuming a Central Com- generated in the Outputs worksheet. The order in
posite Design (CCD). Figure 3 shows the coded vari- which the results appear will be randomized. In other
able values for a CCD. Values of zero can be entered words, the results are presented as if the experi-
for the number of center or axial points if a full mental runs were made randomly. Randomization is
response surface model is not required. Figure 4 good practice in reality since it helps ensure indepen-
shows the actual design points that would be gener- dence by reducing the chance of systematic errors.
ated for the input scenario shown in Figure 2. The The “Sort” button in the Outputs worksheet can then
numbers in brackets identify the number of replica- be used to rearrange the output into a more structured
tions at each design point. form. The sorted data can be readily transferred to
Cell D5 specifies the degree of variability that will other software packages for analysis.Figure 5 shows
be observed in the average lot flowtimes, W . As an example of the sorted outputs generated using the
described in the appendix, it is the coefficient of vari- experiment specified in Figure 2.
Figure 5 Example of Outputs After Sorting
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS 59

Figure 6 Macro Code Within the VB Editor

Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

The macros can be accessed by going into the Visual values based on the experimental design point are
Basic Editor, located under Tools and then Macro. Fig- written into Columns C and D of the Outputs work-
ure 6 shows a view from within the VB Editor. In sheet. For each design point processed, the Exper-
this figure the Project Explorer is visible and the con- iment macro is also called. This macro writes the
tents of the workbook is shown in the upper left- values for the current design point into Cells D13
hand window. If this window is not visible, it can and E13 of the Inputs worksheet. It then calculates
be accessed within the VB Editor by going into View the machine utilization rate, average lot service time,
and then activating the Project Explorer. The Project average lot queue time, and average lot service time
Explorer window should include the atpvbaen.xls file for the given observation using the formulas in Range
and “funcres” references. If these are not present, they D16:D20 and writes these values sequentially into
can be added within the VB Editor to a list found rows in the Outputs worksheet.
under Tools and then References. This list should also This macro also uses the coefficient of variation in
include VBA and the Microsoft Excel Object Library Cell D5 of the Inputs worksheet to specify a multi-
as references. plier for adjusting the calculated queue time to come
All of the VBA code in the workbook is con- up with an observed queue time. A normal distribu-
tained within a module called Experimental Design. tion is used in generating this multiplier. Once data
This module contains three macros, called ExpDesign, has been generated for each of the design points, the
Experiment, and Sort. Part of the code in these macros experimental output is randomized. This occurs at
is shown in the large window of Figure 6. It is not nec- the end of the ExpDesign macro. The “Sort” button,
essary to understand this code unless the user wishes which activates the Sort macro, in the Outputs sheet
to modify it. However, a brief description is given as can then be used to put the data back in a structured
follows. form.
The “Run Experiment” button activates the ExpDe- In order to analyze the experimental results gen-
sign macro. This macro systematically chooses points erated, the appropriate columns from the Outputs
in the experimental design and writes the coded val- worksheet can be copied into statistical analysis
ues for the design points into Columns A and B of software, such as Minitab® or Design-Expert®
the Outputs worksheet. As well, appropriate lot size (Montgomery 2001). In some cases the user of
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
60 INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS

these statistical packages must specify the desired is not. Furthermore, it could be easily observed that
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

experimental design and then the software will the experimental region selected is unlikely to contain
automatically generate worksheet columns showing the lot size combination yielding minimum flowtimes
appropriate factor settings. This means the user must and that the ranges should be moved so both lot sizes
first generate the experimental design before the are reduced. Interaction or surface plots could be used
responses, W , found in column H can be copied and to verify this.
pasted into the analysis worksheet. Another set of experiments would then be run. The
steepest-descent algorithm could be used in determin-
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

ing what lot size combinations should be examined

4. An Example Problem next. Analysis and experimentation would occur iter-
Suppose the problem is to determine the lot sizes that atively until it appears in the region within which
will minimize weighted mean lot flowtimes given the experiments are being conducted is convex and may
demand rate, setup time and production rate informa- contain a minimum. Figure 7 shows analysis, using
tion shown in Figure 2. Initially the region in which Minitab, obtained with results from a 22 experimen-
the optimal lot size combination lies would not be tal design where the coded (−1 +1) lot sizes repre-
clear. In this case we might start by running a small sent actual lot size ranges of (200 350) and (150 300)
factorial design, using perhaps two replications for for Q1 and Q2 , respectively. The top window in this
each design point. For example, we might choose to figure shows the experimental design and responses,
use actual lot sizes of (400 600) and (300 500) for Q1 copied from the Outputs worksheet. The bottom win-
and Q2 , respectively. These values would represent dow shows the ANOVA results. The interaction term
the (−1 +1) coded variable settings. Using two repli- is shown to be statistically significant, indicating
cations at each design point would yield eight obser- there may be curvature in the response surface. Fig-
vations in total. ure 8 shows the interaction plot associated with these
Analysis of the results obtained (not shown) could results.
be done using ANOVA. Such analysis would prob- At this point we might decide to add center
ably indicate that the main effects for both lot size points to the design and determine if curvature is
factors are significant but that the interaction term statistically significant. For example, 10 center points

Figure 7 Minitab ANOVA for a 22 Design

 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS 61

Figure 8 Minitab Interaction Plot rerun with 10 additional center points, the curva-
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

Interaction plot (data means) for W ture will be shown to be statistically significant (not
0.24 shown).
Q1 A final step would be to add axial points. Figure 5
200
0.23 150 shows the Outputs worksheet obtained when rerun-
ning the experiment as a CCD with factorial and axial
0.22 points replicated twice and with 10 center point repli-
Mean

cations. Figure 9 shows the Minitab analysis when a

Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

0.21 full quadratic model has been specified. In this case

the design was initially set up using the menu path
0.20 of Stat>DOE>Response Surface. Results show a high
R2 value, statistically significant linear and interac-
0.19 tion terms at the 95% confidence level and a lack-of-
150 300 fit value that is not significant at the 95% confidence
Q2
level. To confirm the model fits well, the residuals
should also be examined. Figure 10 shows residual
plots for this example.
could be specified by entering this value in Cell I13
of the Inputs worksheet. If using Minitab for analy- Contour and surface plots can be used to confirm
sis, the experimental design must be initially set up whether or not a minimum actually exists within the
before the results can be pasted from the Excel work- region being modeled. Figures 11 and 12 show plots
sheet to the appropriate Minitab worksheet columns. for this example, obtained using Minitab. In this case
A Minitab menu path of Stat>DOE>Factorial can be it is obvious the minimum is within the region. The
used to create the design, including specification of optimal lot size combination appears to be some-
center points. If the number or order of the design where around 315 and 260 for Q1 and Q2 , respec-
points in Minitab differs from those in the Outputs tively. This could be confirmed using the Response
worksheet, the design in Minitab can be edited as Optimizer in Minitab or similar capability in another
appropriate. If the experiment shown in Figure 7 is package. If the region does not contain the minimum,

Figure 9 Quadratic Model of Response Surface

 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
62 INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS

Figure 10 Residual Plots

Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

Residual plots for W

Normal probability plot of the residuals Residuals versus the fitted values
99
0.02
90 0.01

Residual
Percent

0.00
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

50
–0.01
10
–0.02
1
–0.030 – 0.015 0.000 0.015 0.030 0.180 0.195 0.210 0.225 0.240
Residual Fitted value

Histogram of the residuals Residuals versus the order of the data

4.8 0.02

0.01
3.6
Frequency

Residual
0.00
2.4
–0.01
1.2
–0.02

0.0
– 0.02 – 0.01 0.00 0.01 0.02 2 4 6 8 10 12 14 16 18 20 22 24 26
Residual Observation order

it would be necessary to shift the design in the appro- As well, students sometimes have difficulty initially
priate direction and make further attempts to fit a accepting that quite different lot size combinations are
model that will identify a minimum. selected by different individuals or groups attempt-
ing to find the optimal. This happens because the
5. Discussion and Conclusions response surface may be very flat around the opti-
This spreadsheet-based tool has been used effec- mum. It is valuable for them to observe that while the
tively in laboratory and homework exercises in DOE lot size combinations may be different, the predicted
elective courses for engineering undergraduate and lot flowtimes are nearly equal.
graduate students. Students are given a handout In summary, more training of students with
describing the software, problem and relationships. appropriate skills in design of experiments and
The information provided is similar to that given in response surface methodologies is clearly required.
this paper. They are then asked to provide a writ-
ten report for a given production scenario, showing
their path of analysis, final results and conclusions. Figure 11 Contour Plot Showing Optimal
As well, they are asked to explore the effects of setup
Contour plot of W vs Q2, Q1
time reduction if setup times are cut in half. This
requires finding new optimal lot size combinations
and determining the impact on performance. 300 W
The exercise has been given to students with and < 0.20
without providing an instructional computer labora- 0.20 – 0.22
tory. In general, a hands-on computer lab in which 250
0.22 – 0.24
0.24 – 0.26
students are guided through the solution path for an 0.26 – 0.28
Q2

example problem is valuable. It not only facilitates > 0.28

faster and better understanding of the tool but also 200
improves understanding of the requirements when
doing the assigned analysis for evaluation.
Issues encountered have usually related to pick- 150
ing lot size ranges that are too small or too large.
Encouraging students to think about the problem and
evaluate behavior at the bounds of the design space 200 240 280 320 360
helps alleviate the selection of inappropriate ranges. Q1
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS 63

Figure 12 Surface Plot where Di is the demand rate, Pi is the production rate, Qi is
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

the lot size, and i is the lot setup time for part type i. Simi-
Surface plot of W vs. Q 2, Q1
larly, the utilization rate, , is determined by the following:
m    m 
 Di

Di Q D
= i + i = + i i 0 ≤  < 1
i=1
Qi Pi i=1
Pi Qi
0.275
Note that in analysis using rapid modeling relationships,
0.250  is usually constrained to be 0.95 or less. Although the
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

utilization must simply be less than 1.00 in order to attain

W

0.225 feasibility, it has been found that performance estimates

deteriorate very quickly as utilization levels approach 100%.
0.200 If it is assumed that there is no variation in the lot setup
times for a given part type, the standard deviation of lot
200
150
250 service times, s , will be a function of lot processing times
300
200
250 350 Q1 only. The value of cs2 can therefore be determined from the
Q2 300
following:
 2 Ex2  − x̄2
cs2 = s2 = 
x̄ x̄2
If it is further assumed that there is no variation in process-
The spreadsheet-based tool presented in this paper ing times for a given part type, the previous equation can
introduces a unique approach to support such train- be restated as follows:
ing. An obvious extension would be to allow sim- 
m 2

2 i=1 Di /Qi i + Qi /Pi  
ulated experimentation in other contexts besides cs =
− 1
 m i=1 Di /Qi x̄
2
production lot sizing.
Using the previous equations, the mean lot flow time for a
given part type, i, can be derived as follows:
Appendix. Modeling Relationships
A good objective for lot size selection is to minimize the Wi = Wq + xi
weighted mean lot flowtimes across all part types using a

m
single resource. Since all part types have unique production x̄2 m 2
i=1 Di /Qi ca − 1 +
2
i=1 Di /Qi i + Qi /Pi  
=
characteristics, the best lot size combinations are affected 21 − 
by the relative characteristics as well as the variability of  
Q
lot interarrivals. Furthermore, since there is no closed form + i + i 
Pi
solution for lot flowtimes under GI/G/1 queuing assump-
tions, we must resort to approximations. These approxima- The weighted mean lot flow time across all part types is
tions are found in the “rapid modeling” literature, where then given by the following:
queuing heuristics are used to approximate work flow in
W = Wq + x̄
complex manufacturing networks. The following literature
includes relevant discussions: Whitt (1983), Buzacott and The mean weighted lot flowtimes are convex with respect to
Shanthikumar (1993), and Hopp and Spearman (2001). the lot sizes. This makes the model useful for demonstrating
The mean lot flowtime for part type i, Wi , expressed as various types of models, including response surfaces.
the sum of the expected queue time, Wq , and the mean lot However, any given combination of lot sizes will always
service time, xi , can be approximated as follows: produce the same flowtime values when these approx-
imations are used. In other words, there is no uncer-
ca2 + cs2   tainty so replication would serve no purpose. Therefore,
Wi = Wq + xi = x̄ + xi the approach taken in using these relationships for DOE
2 1−
training purposes is to induce uncertainty. Uncertainty is
where x̄ is the weighted mean lot service time for all part induced by adding a random value to the queue time, Wq .
types, xi is the mean lot service time for part type i, and The value of Wq is first calculated for the center point of
 is the resource utilization rate. The ca and cs variables the experimental design, Wqcp . This value is then multi-
plied by a random variable drawn from a normal distribu-
are the coefficient of variation of the lot interarrival times
tion [N 1 CV ], with a mean of 1 and standard deviation
and lot service times, respectively. Note that it is com-
of CV, where CV is specified by the user in Cell D5 of the
mon to assume that the average queue time for lots of all
Inputs worksheet. The difference between Wqcp N 1 CV 
types of parts will be equal when using the rapid modeling and Wqcp is then the amount of induced uncertainty in
approach. Therefore, Wq is not indexed by part type. average observed lot flowtimes from replication to repli-
The mean lot service time across m part types, including cation. Therefore, the average observed lot flowtime, W ,


lot setup and processing times, is given by the following: obtained for any given replication is determined by the

m following:
i=1 Di /Qi i + Qi /Pi 
x̄ =
m W  = Wq + x̄ = Wq + Wqcp N 1 CV  − 1 + x̄
i=1 Di /Qi
 Enns: An Interactive Spreadsheet-Based Tool to Support Teaching Design of Experiments
64 INFORMS Transactions on Education 8(2), pp. 55–64, © 2008 INFORMS

Since the amount of expected uncertainty is always based Box, G. E. P., P. Y. T. Liu. 1999. Statistics as a catalyst to learning
Additional information, including supplemental material and rights and permission policies, is available at http://ite.pubs.informs.org.

on the center point of the design, it will be the same at all by scientific method part I—An example. J. Quality Tech. 31(1)
factor settings within a given design. This helps ensure the 1–15.
ANOVA assumption of equal variances at all combinations Buzacott, J. A., J. G. Shanthikumar. 1993. Stochastic Models of Man-
ufacturing Systems. Prentice-Hall, Englewood Cliffs, NJ.
of factor settings will not be violated.
Hopp, W. J., M. L. Spearman. 2001. Factory Physics. McGraw-Hill,
Boston.
Montgomery, D. C. 2001. Design and Analysis of Experiments, 5th ed.
References John Wiley and Sons, New York.
Downloaded from informs.org by [182.74.116.110] on 01 July 2015, at 00:51 . For personal use only, all rights reserved.

Box, G. E. P. 1999. Statistics as a catalyst to learning by scientific Whitt, W. 1983. The queueing network analyzer. Bell Systems Tech. J.
method part II—A discussion. J. Quality Tech. 31(1) 16–29. 62(9) 2779–2813.