1. CED RClemson Engineering Design Applications and Research
Varun Kumar
kumar3@clemson.edu
Committee:
Dr. Gregory Mocko (Chair)
Dr. Joshua Summers
Dr. Cameron Turner
UNDERSTANDING THE ROLE AND IMPORTANCE
OF DESIGN PROBLEMS IN CREATIVITY RESEARCH
2. 2 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Key contributions
Research questions and tasks
Design problem reuse pattern
Design problem similarity
o Based on structural elements
o Based on Latent Semantic Analysis
Effect of problems on example interventions
Research conclusions
Research overview
Recommendations and limitations
OUTLINE
Contributions RQs Problem reuse Structural sim. LSA sim. Concl.Regression Overview
3. 3 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
KEY CONTRIBUTIONS
Contributions RQs Problem reuse Structural sim. LSA sim. Concl.Regression Overview
1. Identified pattern of design problem usage
• Problems not reused often
• Evaluated 15 years published literature
2. Provided two methods to compare conceptual problems
• Structural similarity based on count of elements within the
problem
• Semantic similarity based on LSA
3. Identified design problems as potential moderator in effectiveness of
examples as intervention
• Meta-regression to study problem size on results
4. Initiated creation of experiment design problem repository
• Developed structure and DB for experiment and problems
4. 4 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Statements of need, requirement or function desired [1]
Conceptual problems used in creativity research
WHAT ARE DESIGN PROBLEMS
Contributions RQs Problem reuse Structural sim. LSA sim. Concl.Regression Overview
Conceptual design problem example [2]
Design and build a low-cost, easy to manufacture peanut
shelling machine that will increase the productivity of
the African peanut farmers. Target throughput is
approximately 50 Kg per hour. The goals include: a)
Must remove the shell with minimal damage to peanuts
b) Electrical outlets are not available as a power source
c) A large quantity of peanuts must be quickly shelled.
5. 5 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
RESEARCH QUESTIONS AND TASKS
RQ1: How can the pattern
of design problem usage in
creativity research be
identified?
Task 1:
Graph representation of
design problems
RQ2: How can design problem
be assessed for similarity
based on the problem
statement?
Task 2: Similarity assessment
based on structural elements
Task 3: Similarity assessment
based on Latent Semantic
Analysis
Contributions RQs Problem reuse Structural sim. LSA sim. Concl.Regression Overview
6. 6 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
RESEARCH QUESTIONS AND TASKS
RQ3: How can natural
language representation be
used to compare problems
for similarity?
Task 2: Similarity
comparison based on
structural elements
Task 3: Similarity
comparison based on
Latent Semantic Analysis
RQ4: Does the choice of
design problem influence
effectiveness of interventions
(presenting examples)?
Task 4:
Meta-regression analysis of
user studies
Contributions RQs Problem reuse Structural sim. LSA sim. Concl.Regression Overview
7. 7 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
DESIGN PROBLEM REUSE: GATHERING STUDIES
1. Creativity + engineering design 644,000
2. Creativity + engineering design + ideation 17,600
3.
Creativity + engineering design + ideation +
experiment
11,400
4. Between 2000 and 2014 8080
5. Design studies, design society, ASME 392
6. Abstract review for presence of experiment 129
7. Reported design problem 93
8. Removal of case studies 41
9. Removal of duplicate experiments 34
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
8. 8 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
DESIGN PROBLEM REUSE: GRAPH CODING
Study Author Problem name
S1 Toh (Student), Miller (Faculty) Milk frothing device (DP11)
S2 Linsey (Faculty), Markman (Faculty), Wood (Faculty) Peanut shelling machine (DP9)
S3 Daly (Faculty), Christian (Faculty) Solar device (DP12)
S4
Glier (Faculty), Schmidt (Student)
Linsey (Faculty), McAdams (Faculty)
Peanut shelling machine (DP9)
S6 Acuna (Faculty), Sosa (Faculty) Counter top stand (DP13)
….. …… ……
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
9. 9 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Legends
Researcher
Problem Used Design problem-researcher use
• Some problems used
multiple times.
E.g: 59 (Peanut sheller)
used by 10 (Linsey) 7
times
• Problem choice depends
on researcher’s
requirements
• Problem reuse not
widespread across
researchers
DESIGN PROBLEM REUSE: GRAPH
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
10. 10 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
• Is it possible to draw conclusions between studies when design
problems are different?
• Can we control one source of difference between studies?
• Can we compare problems to enable reuse?
DESIGN PROBLEM REUSE
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
11. 11 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY
• Question originated from work by Durand and coauthors [2]
o Size of the problem in terms of functional units
o Connectedness of the problem
o Participants' familiarity with the design problem
o Participants' familiarity with the design solutions
o Size (number of variables) of the potential solution space
o Are there assumed constraints due to known solutions
o The effort required to solve the problem
o The degree to which existing solutions will cause fixation
o The domain of the design problem
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
12. 12 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY
Design
Problem
Goals of the problem
Functional
Requirements
Non-functional
requirements
Information about end
user
Reference to an
existing product
• 5 elements of identified based on
o Literature
o Assessment of design
problems collected
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
13. 13 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY: PROTOCOL
Characteristic Questions asked
Number of goals • What is the final objective of the problem
statement?
• Does the problem statement ask only to design
one object or more?
Number of functional
requirements
• How many primary functions can you find?
• How many action verbs can you identify?
Number of non functional
requirements
• How many non-action and non-functional
aspects can you identify?
• How many performance and usability aspects
can you identify?
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
14. 14 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY: PROTOCOL
Characteristic Questions asked
Information about end
user
• Can you identify who is going to use the
product?
Reference to existing
product
• Do you know if the product that needs to be
designed exists?
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
15. 15 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY: PROCEDURE
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
Adapted from [3]
16. 16 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY: OUTCOME
Design problems Goal FR NFR
Ref. existing
product User
Sketch Ideas for subway improvement. 1 1 0 1 0
There is a need of designing a new drawing
table……..
1 1 2 1 1
Tubular map cases consist of a system for…….. 1 4 0 1 1
Design a new system for gathering together………. 1 2 3 1 1
It is asked to design a new table for offices………… 1 1 1 1 1
Design a wearable binocular which satisfies……….. 1 1 3 0 0
Design an urban (bi or tri) cycle for use by “white-
collar workers”.
1 1 0 1 1
Design a water lifting device. 1 1 0 1 0
Design and build a low-cost, easy to
manufacture………
1 3 4 1 1
…….. ….. ….. ….. ….. ….
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
Krippendorff’s
Alpha
1.0 1.0 0.50 0.18 0.60
17. 17 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL SIMILARITY: COMPARISON
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
Design problems Goal FR NFR Ref User
DP18 Devise an innovative solution for a remote controller
that meets……. 1 12 3 0 0
DP17 The Air Force wants to increase the functionality of
their Unmanned Air Vehicle technology…….. 1 11 1 1 1
DP3 Tubular map cases consist of a system for storage
and transportation of maps…………
1 4 0 1 1
DP13 Design a counter top stand to display and dispense
candy and chocolate snacks……... 1 4 5 1 1
DP9 Design and build a low-cost, easy to manufacture
peanut shelling machine that………………….
1 3 4 1 1
DP26 Design a machine that registers a bottle to a capping
station, caps it, and allows……………….
1 3 0 0 0
DP4 Design a new system for gathering together and
hiding the wires of the electronic
equipment………………….
1 2 3 1 1
…… ………….
…. …. …. …. ….
18. 18 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
5 elements analogous to a vector space
Cosine between vectors gives separation
Average structural similarity of 0.74
STRUCTURAL SIMILARITY: COSINES
DP10 DP11 DP12 DP13 DP14
DP10 1 0.70 0.80 0.80 0.70
DP11 0.70 1 0.61 0.60 1
DP12 0.80 0.61 1 0.99 0.61
DP13 0.80 0.60 0.99 1 0.60
DP14 0.70 1 0.61 0.60 1
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
19. 19 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Extracts contextual meanings from large text corpus.
Generates similarity scores between -1 and 1
37 design problem statements tested for LSA similarity
SEMANTIC SIMILARITY: LSA
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
20. 20 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
SEMANTIC SIMILARITY: RESULTS
DP10 DP11 DP12 DP13 DP14
DP10 1 0.05 0.14 0.09 0.07
DP11 0.05 1 0.43 0.42 0.12
DP12 0.14 0.43 1 0.38 0.33
DP13 0.09 0.42 0.38 1 0.28
DP14 0.07 0.12 0.33 0.28 1
A snippet of the LSA result matrix
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
22. 22 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
DP12
Develop products that utilize sunlight for heating and cooking food. The products
should be portable and made of inexpensive materials. It should be able to be
used by individual families, and should be practical for adults to set up in a sunny
spot. Specific materials requirements for a targeted temperature can be
postponed to a later stage [4]
DP13
Design a counter top stand to display and dispense candy at convenience
stores. The requirements of this task are: a. The stand must be easy to use both
by the final user to grab the product and by the shop attendant to refill the
product. b. The stand must contain and visually identify one specific target brand
and product presentation. c. The stand must be built in one single material to
choose between cardboard or laminated plastic (PVC, PS or PETG).d. The stand
must be innovative, yet simple to manufacture and assemble [5].
Structural LSA
0.99 0.38
STRUCTURAL AND SEMANTIC SIMILARITY
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
23. 23 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL AND SEMANTIC SIMILARITY
Design an automatic clothes-ironing machine for use in hotels. The purpose of
the device is to press wrinkled clothes as obtained from clothes dryers and fold
them suitably for the garment type. You are free to choose the degree of
automation. At this stage of the project, there is no restriction on the types and
quantity of resources consumed or emitted. However, an estimated 5 minutes
per garment is desirable [6]
Design an automatic recycling machine for household use. The device should
sort plastic bottles, glass containers, aluminum cans, and tin cans. The sorted
materials should be compressed and stored in separate containers. The amount
of resources consumed by the device and the amount of space occupied are not
limited. However, an estimated 15 seconds of recycling time per item is desirable
[6]
Structural LSA
0.81 0.52
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
24. 24 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL AND SEMANTIC SIMILARITY
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
Structural LSA
0.81 0.52
25. 25 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL AND SEMANTIC SIMILARITY
Alarm clocks are essential for college students, however often times they will wake up a
roommate and those around them as well. Design an alarm clock for individual use that will
not disturb others. The clock should be portable for use in a variety of situations such as on
the bus, in the library, or in a classroom. Customer Needs: a) Must wake up individual with
no disturbance to others b) Must be portable and lightweight c) Electrical outlets are not
available as a constant power source d) Low cost [2]
Corn is currently the most widely grown crop in the Americas with the United States
producing 40% of the world’s harvest. However, only the loose corn kernels are used when
bought canned or frozen in grocery stores. An ear of corn has a protective outer covering
of leaves, known as the husk, and strands of corn silk threads run between the husk and
the kernels. The removal of husk and silk to clean the corn is known as shucking corn.
Design a device that quickly and cheaply shucks corn for mass production. Customer
Needs: a) Must remove husk and silk from corn cob with minimal damage to kernels b) A
large quantity of corn must be shucked quickly. c) Low cost [2]
Structural LSA
0.70 0.30
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
26. 26 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STRUCTURAL AND SEMANTIC SIMILARITY
Structural LSA
0.70 0.30
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
• Durand and coauthors conclusions:
o Quantity: No difference
o Quality: Alarm has higher quality solutions than corn
o Novelty: Corn higher than alarm
o Variety: No difference
27. 27 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Used for identifying potential covariates
Relationship between effect of intervention and covariate
Use of examples in user studies
Does design problem have a relationship with this intervention?
META-REGRESSION
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
Experiment
Design problem
Examples Results
28. 28 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STEPS FOR META-REGRESSION
Collect studies
Extract information
Calculate effect size and variance
Generate regression model
between effect size and covariate
Verify significance of model
• 9 studies out of 34 contained examples as
interventions.
• Additional search conducted using keywords ‘design
+ creativity + examples + experiment’ in Google
Scholar
• Filtering was done to collect between-subject
experiments
• Study should report statistics
• Studies which measure quantity or fluency, quality,
variety or novelty
• English language publications between 1995 - 2015
• Total N = 17 studies
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
29. 29 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STEPS FOR META-REGRESSION
• Method of assignment to condition
• Mean, standard deviation, sample size
• Design problem used and its problem
statement.
• Different datasets for 4 metrics
Collect studies
Extract information
Calculate effect size and variance
Generate regression model
between effect size and covariate
Verify significance of model
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
30. 30 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STEPS FOR META-REGRESSION
Collect studies
Extract information
Calculate effect size and variance
Generate regression model between
effect size and covariate
Verify significance of model
𝑆𝑀𝐷 =
𝑋𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 − 𝑋𝑐𝑜𝑛𝑡𝑟𝑜𝑙
𝑠 𝑝
𝑠 𝑝 =
𝑛 𝐺1 − 1)𝑠 𝐺1
2
+ (𝑛 𝐺2 − 1)𝑠 𝐺2
2
𝑛 𝐺1 + 𝑛 𝐺2 − 2
𝑆𝑀𝐷′ = 1 −
3
4𝑁 − 9
× 𝑆𝑀𝐷
𝑆𝐸′ =
𝑛 𝐺1 + 𝑛 𝐺2
𝑛 𝐺1 𝑛 𝐺2
+
𝑆𝑀𝐷′ 2
)2(𝑛 𝐺1 + 𝑛 𝐺2
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
31. 31 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STEPS FOR META-REGRESSION
Study SMD’ SE’ Problem size
Tsenn -0.61 0.41 8
Smith,S 0.07 0.21 3
Shorachi 0.86 0.38 3
Lujun -0.21 0.33 2
Cardoso 0.15 0.33 2
Agogue -0.39 0.22 2
Chan -0.46 0.24 4
Goncalves -0.01 0.33 2
... … … …
Collect studies
Extract information
Calculate effect size and variance
Generate regression model between
effect size and covariate
Verify significance of model
𝑷𝒓𝒐𝒃𝒍𝒆𝒎 𝒔𝒊𝒛𝒆 = 𝑵𝒐. 𝒐𝒇 𝒈𝒐𝒂𝒍𝒔 + 𝑵𝒐. 𝒐𝒇 𝑭𝑹𝒔 + 𝑵𝒐. 𝒐𝒇 𝑵𝑭𝑹𝒔
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
32. 32 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
STEPS FOR META-REGRESSION
Collect studies
Extract information
Calculate effect size and variance
Generate regression model
between effect size and covariate
Verify significance of model
where 𝛿𝑖 = predicted SMD
𝛽0, 𝛽1 = regression coefficients
εi = within study variance
ηi = residual variance
𝛿𝑖 = 𝛽0 + 𝛽1 × 𝑠𝑖𝑧𝑒 + 𝜀𝑖 + 𝜂𝑖
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
33. 33 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
RESULTS OF META-REGRESSION: QUANTITY
Variable Coefficient SE 95% CI p
DP size -0.09 0.02 -0.14 to -0.04 <0.01
Constant 0.34 0.11 0.12 to 0.56 <0.01
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
34. 34 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
RESULTS OF META-REGRESSION: QUALITY
Variable Coefficient SE 95% CI p
DP size -0.07 0.07 -0.21 to 0.08 0.38
Constant 0.80 0.45 -0.08 to 1.67 0.08
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
35. 35 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
RESULTS OF META-REGRESSION: NOVELTY
Variable Coefficient SE 95% CI p
DP size 0.11 0.10 -0.08 to 0.30 0.25
Constant -0.27 0.39 -1.04 to 0.49 0.48
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
36. 36 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
SUMMARY OF REGRESSION RESULTS
Creativity metrics
Quantity Quality Novelty
Number of sample points 44 13 34
tau 0.15 0.56 0.95
R2 48.02% 0% 0.59%
Coefficient of problem size -0.09* -0.07 0.11
* p < 0.05
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
37. 37 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Structural similarity
o Offers a way for comparing two problem structures
o Five characteristics between problems can be compared
o Compares problems based on representation
LSA
o Similarity assessed based on contextual meaning
o No human interpretation
Element similarity and LSA
o Can be used together to generate a better similarity estimate
RESEARCH CONCLUSIONS
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
38. 38 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Meta-regression
o Problem size affects effectiveness of examples presented
o Other differences can also affect effectiveness of examples
o Regression results should be verified through experiments
o Need for verification and validation of experimental findings
RESEARCH CONCLUSIONS
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
39. 39 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
RESEARCH OVERVIEW
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
Graph of
problem and
author
Structural
similarity
Meta-
regression
Semantic
similarity
RQ1
Problem usage
pattern
RQ2
Using
natural
language for
similarity
RQ4
Impact of
problem
choice
RQ3
Similarity
comparison
between
problems
Low problem
reuse
Similarity between
problems
Problems can
moderate
effectiveness of
intervention
ContributionsResearch
tasks
Research
questions
40. 40 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Problems similar to existing ones can be chosen
LSA and five elements can be used for similarity comparison
Repository of benchmarked problems with similarity scores
Validation framework needed
Common methodology for reporting studies
RECOMMENDATIONS
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
41. 41 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Similarity assessment
o Do not address problem solvability
o LSA similarity and engineer’s interpretation may be different
o Element similarity needs robust protocol for identification
o Cosine similarity shows high similarity for most problems
Meta-regression
o Hypothesis generating tool
o Results depend on reported data
LIMITATIONS
Contributions RQs Problem reuse Struct. sim LSA sim. Concl.Regression Overview
42. 42 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
[1] Cross, N., 2008, “The nature of design,” Engineering design methods: strategies for
product design, John Wiley & Sons, Chippenham, Wiltshire, pp. 11–12.
[2] Durand, F., Helms, M. E., Tsenn, J., McAdams, D. A., and Linsey, J. S., 2015, “In
search of effective design problems for design research,” IDETC & CIE, American
Society of Mechanical Engineers, Boston, pp. 1–10.
[3] Cardoso, C., and Badke-Schaub, P., 2011, “The influence of different pictorial
representations during idea generation,” J. Creat. Behav., 45(2), pp. 130–146.
[4] Daly, S., and Christian, J., 2012, “Assessing design heuristics for idea generation in an
introductory engineering course,” Int. J. Eng. Educ., Vol. 28(2), pp. 1–11.
[5] Acuna, A., and Sosa, R., 2011, “The Complementary Role of Representations in Design
Creativity: Sketches and Models,” Design Creativity 2010, Springer, pp. 265–270.
[6] Patel, A., Kramer, W., Summers, J. D., and Shuffler-Porter, M., 2016, “Function Modeling:
A Study of Model Sequential Completion Based on Count and Chaining of
Functions,” International Design Engineering Conferences and Computers in
Engineering Conference (ASME IDETC/CIE, ASME-AMER SOC MECHANICAL
ENG, Charlotte, NC, pp. DETC2016–59860.
References
44. 44 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Maximum LSA score
DP29
Design the next generation of breakthrough alarm clock.
DP31
People generally use their mobile phone or a traditional alarm
clock in order to wake themselves up every morning. Yet, they
are not always effective and can sometimes cause
oversleeping. What would be other possible ways to wake up
in the morning in a certain time without using any form of
alarm clock? Generate as many ideas as possible.
LSA
0.53
45. 45 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Low LSA score
DP28
Redesign an electric toothbrush for increased portability.
DP30
Design a litter collection system for use by student groups in
volunteer "Adopt-A-highway" activities.
LSA
0
46. 46 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Appendix A: Design problem used for experimental verification
of protocol
Problem 1: Design a new system for gathering together and hiding the wires of the electronic equipment in an
office table. Currently the work in the field of design, architecture and engineering needs of a personal computer,
printers, and scanners. Each of these devices needs of electrical supply and the wires on table surface are annoying.
Actually, there are simple solutions to gather them, but it is difficult to extract or introduce a wire, or they leave the
wires hanging behind the table.
Problem 2: Design an automatic recycler device that can automatically sort plastic bottles, glass containers,
aluminum cans, and tin cans. The major differentiation between types of materials lay with the given dimensions of
the products: plastic bottles are the tallest, glass containers are very short and heavy, and aluminum cans are
lightweight. Devices are given strict requirements to adhere to such as volume and weight constraints, safety
requirements, and most importantly, have to operate autonomously once a master shutoff switch is toggled.
Problem 3: A mechanical system is required which, in the event of a fire, will enable people to escape from a
six-storey building by lowering themselves to the ground from windows. The system, which might make use of a 5
mm diameter steel cable, must be capable of lowering either a small child or a heavy adult at approximately the
same constant speed.
Problem 4: Design and develop an artifact to facilitate grocery shopping in a typical French/Singapore
Chinatown fresh market. The artifact should facilitate carrying of groceries from fresh market to home.
47. 47 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
S.No. Characteristic Definition Scoring system
1.
Reference to existing
product
Whether or not the problem statement contains a
reference to an existing product.
Questions to be asked
Does the problem statement contain reference to
an existing product?
Does the problem statement ask you to re-design
an existing product?
If a reference to existing product exists in the
problem statement, assign a score of 1, else 0
2.
Functional
requirements (FR)
These are the things which the product needs to do, or
the tasks that you want the product to perform without
any reference to how it should be done. Judgement
should be based only on explicitly stated texts, and not
on the implied meaning of a sentence. To identify FRs,
look for:
1. Action verbs like move, work etc. associated with
objects (objects include nouns on which the
action verbs act like throw stones, gather fruits
etc.)
2. Primary functions of the design (eg. move
objects, lift, transport etc.).
3. These could also be nouns derived from verbs
(eg. washing machine, toaster etc.)
4. If there are two objects associated with one
primary function, count it as 2 separate FRs (eg.
move object X & object Y is counted as 2 FRs)
Questions to be asked
What are the primary functions of the product?
What are the expected outputs/tasks which the
product needs to perform?
Count the number of functional requirements
given in the problem statement. There can be 2
cases:
1: When a new product design is desired: In
this case, FR should be specified in the
problem statement. Else, give a score of 0.
2: When a re-design or a new design for an
existing product is desired: FR count in this
case is already 1 to start off, since atleast 1
product function is known.
Appendix B: Element definitions
48. 48 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
S.No. Characteristic Definition Scoring system
3.
Non-functional
requirements
These are 'non-functional' requirements, which do not determine
the primary functions of the product, but cast a bound on the
overall shape, size, cost, operation and selection of the design.
Judgement should be based only on explicitly stated texts, and not
on the implied meaning of a sentence. Typical NFRs include:
1. Any restrictions on what the product or system shall or shall
not do apart from the primary functions (eg. should not
overheat, should be easy to use, should be manufacturable
etc.)
2. Any restriction on how the product shall fulfil its intended
functions (eg. device should move using rollers, device
should work using sliding mechanism etc.)
3. Any qualities that the product must possess (eg. easy to
make, easy to use, cheap, etc.)
Questions to be asked
• What things cast a limit or a bound on the solution space?
What are the qualities that the product should possess as a
whole?
How the overall product 'shall be' like?
Count the number of Non-functional requirements
given in the problem statement.
4. Number of goals
These are goals or final objectives associated with the design
problem.
Questions to be asked
What is the final objective of the problem statement, or
Does the problem statement ask only to design or do
something else?
Count the number of goals or objectives mentioned
in the problem statement.
5.
Information about end
user
Information about who is going to use the product or who is the
customer. It should be explicitly stated in the problem statement.
Questions to be asked
Who is going to be the end user of the product?
Check the problem statement to see if any
information about the end user is provided or not. If yes,
give a score of 1, else a 0
Appendix B: Element definitions
49. 49 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Problem statements
DP10
Redesign a traffic light using LED instead of an incandescent bulb. Ideas
generated must address the barred vision due to accumulation of snow as well
maintain the energy savings of LEDs.
DP11
Develop concepts for a new, innovative, product that can froth milk in a short
amount of time. The product should be able to be used by the consumer with
minimal instruction.
DP12
Develop products that utilize sunlight for heating and cooking food. The
products should be portable and made of inexpensive materials. It should be
able to be used by individual families, and should be practical for adults to set
up in a sunny spot. Specific materials requirements for a targeted temperature
can be postponed to a later stage.
DP13
Design a counter top stand to display and dispense candy and chocolate
snacks at convenience stores. The requirements of this task are:a. The stand
must be easy to use both by the final user to grab the product and by the shop
attendant to refill the product.b. The stand must contain and visually identify one
specific target brand and product presentation.c. The stand must be built in one
single material to choose between cardboard or laminated plastic (PVC, PS or
PETG).d. The stand must be innovative, yet simple to manufacture and
assemble.
DP14
Design a device to transport a ping-pong ball the farthest distance powered only
by a standard issue compression spring. The device is to be constructed with a
limited set of given materials (e.g. balsa wood, wire and Styrofoam).
50. 50 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
Regression assumptions
Quantity
Quality
54. 54 of 43
kumar3@clemson.eduCED RClemson Engineering Design Applications and Research
List of studies
Meta regression dataset
Links to different data tables
Hinweis der Redaktion
Good afternoon everyone. Thanks for being here.
My research was on understanding the role and importance of problems in creativity research.
Dr. Mocko was my advisor and my committee members include Dr. Summers and Dr. Turner.
I would like to thank Dr. Turner for agreeing to join my committee at the last moment.
-Outline of things I will be talking about today.
By generating a graph of problems used by researcher helped me show that design problems are not frequently used in experiments.
This is a source of difference between studies which may be due to requirements but we can try to reduce this variability in creativity research
By using 5 problem characteristics and LSA, I have provided the community two methods by which they can compare problem statements for similarity.
Through meta-regression, I have identified that problem choice may be influencing the effectiveness of experiments where examples are presented as interventions.
I also initiated an experiment repository which contains different design problems used in the past and the experiments using them.
They are statements of need, requirement or function desired in the product.
Conceptual problems are problems used in creativity experiments. They are different from real world design problems since they are designer to suit the experiment’s requirement and aim to enhance idea generation in participants.
Detailed specifications are generally not a part of such problems,
Here is an example of one such problem, the peanut sheller which has been used in creativity experiments in the past
The first research question that I had was if people use different design problems for their research, how can we show this fact.
Do we have people who share and reuse problems in their experiments? To accomplish this task, I used a graph representation between experiment researchers and design problems used by them between the years 2000 and 2014
The second and third research question was inspired from a recent work published by researchers from Georgia Tech. Their idea was that problems which yield similar results in creativity experiments are similar to each other. My idea was the results are also influenced by the participants who produce them. Why not compare them based on their statements.
To accomplish this task, I have used two approaches. The first approach is based on identifying five elements in a problem model. The second approach is based on semantic similarity between statements.
My third research question was if we are comparing problems for similarity based on their statement, we could use natural language elements to compare them. Problems are generally presented to participants in experiments in statements written in natural language which contain certain elements. If you think of a problem, you can realize that it needs to contain a requirement, goal at least.
To accomplish this task, I used Latent semantic analysis and elements such as requirements and goals to compare two problems.
My fourth question was does the choice have an impact on the interventions that are being tested and if yes, can we show it. While collecting studies, I read few studies that said presenting examples is good and inspires participants to generate creative solutions. Some said examples may in fact reduce the creativity aspects of solutions generated by participants.
Since most studies use different problems, the question that I had was : Could there be a relationship between design problem used and effect seen on participants who were exposed to examples in these studies. To accomplish this task, I chose to use meta regression which is a method commonly used in medical research for identifying potentially hidden covariates in published literature.
To address my first research question about limited reuse of design problems, I collected user studies which have been published between 2000 and 2014 in the area of engineering design community.
I intersected keywords ‘……’ and searched through two databases google Scholar and WebofScience
I narrowed down the results by years of publication between 2000 and 2014 since an earlier search between years 1980 and 1999 did not yield satisfactory results.
I focused on user studies published in mechanical engineering journals like Design studies, design society and ASME.
With about 392, I went over their abstracts to see if they were relevant or not.
I went over the content of about 129 studies to see if they use a design problem in an experiment or not.
Finally I removed case studies and duplicate experiments to get a set of 34 studies with 37 design problems.
In order to generate a graph to show how people have used the design problems in these 34 studies.
The authors who had used the problem were assigned to the problems that they had used. I searched the web to see if the authors were researchers or faculty who have been working in the area of creativity or not.
I removed authors who were students on these papers and I had only the faculty researchers left which I associated with the design problem used by them in an Incidence matrix to generate a bipartite graph between authors and problems.
This is the graph which I obtained which shows not many problems have been reused.
There is some reuse of problem shown by node 59 by author shown by node 10.
The problem is the peanut machine problem that I showed as an example earlier and the researcher Julie Linsey has used in 7 experiments.
Thus, we see there are researchers who use the same problem but we do not see different authors using the same problem
This may be due to their requirements but we are not sure if the problems were actually not usable in other experiments as of now.
Thus, what can be concluded from this graph is the fact that different design problems have been used by researchers based on their requirement making them a source of difference between studies that are published.
It may not be possible to use the same problem everywhere due to experiment’s constraints. Instead what we can try to do is to use similar problems for our experiments.
The alternative can be to use a problem which is similar to an used problem in some respect. If an old problem has been found to be suitable by researchers, using a similar problem in similar conditions can help reduce the chances of problem affecting the results.
For instance, if the peanut shelling problem has worked for undergraduate mechanical students, can we possibly use a problem which is similar to it if we are testing on undergraduates.
This is the graph which I obtained which shows not many problems have been reused.
There is some reuse of problem shown by node 59 by author shown by node 10.
The problem is the peanut machine problem that I showed as an example earlier and the researcher Julie Linsey has used in 7 experiments.
Thus, we see there are researchers who use the same problem but we do not see different authors using the same problem
This may be due to their requirements but we are not sure if the problems were actually not usable in other experiments as of now.
Thus, what can be concluded from this graph is the fact that different design problems have been used by researchers based on their requirement making them a source of difference between studies that are published.
It may not be possible to use the same problem everywhere due to experiment’s constraints. Instead what we can try to do is to use similar problems for our experiments.
The alternative can be to use a problem which is similar to an used problem in some respect. If an old problem has been found to be suitable by researchers, using a similar problem in similar conditions can help reduce the chances of problem affecting the results.
For instance, if the peanut shelling problem has worked for undergraduate mechanical students, can we possibly use a problem which is similar to it if we are testing on undergraduates.
My next research question was to address how to compare design problems based on their representation.
If I have two problem statements, can I compare them somehow and say problem 1 is similar to problem 2 in some respect.
In the work presented by Durand and coauthors, the authors identified 9 characteristics which they hypothesized as design problem characteristics.
They chose two problems and tested them to see if they produce similar results across 4 creativity metrics or not.
What they found was that the two problems were not similar or dissimilar across all 4 metrics based on the results obtained.
Hence the question remains-how do we know if two problems are similar or not?
This work provided me the opportunity of trying to compare problems based on problem characteristics, some of them which were also identified by Durand and coauthors.
So I went through the collection of 37 different problems I had and found these 5 characteristics to be present in most problems.
Some of these characteristics were also reported by Durand. Other researchers had also reported similar characteristics in the past.
Goal of the problem is about what the problem wants you to do. If the problem states ‘Design a peanut shelling machine’ there is one goal described in the problem. A problem may seek design of more than one thing at times in case of which there are two goals mentioned.
The functional requirements are primary tasks required in the product. These are general action verbs associated with an object or a noun. They describe the function that the device needs to accomplish.
Non functional requirements are quality related requirements. These are non action and non-function aspects which determine the shape, size, usability aspects of the product.
Information about end user is whether the problem states who the end user is going to be or not. This factor helps related the problem solver to who is going to use the product. This is a yes/no type variable.
Reference to existing product is the participant’s familiarity with product that he/she needs to design. For instance, if the problem states ‘Design a hair dryer’, it can be assumed that most participants would be familiar with the product. However, this at times can be a conundrum since participant’s background may also influence whether he/she knows about the product or not. This is also a categorical choice.
My next research question was to address how to compare design problems based on their representation.
If I have two problem statements, can I compare them somehow and say problem 1 is similar to problem 2 in some respect.
In the work presented by Durand and coauthors, the authors identified 9 characteristics which they hypothesized as design problem characteristics.
They chose two problems and tested them to see if they produce similar results across 4 creativity metrics or not.
What they found was that the two problems were not similar or dissimilar across all 4 metrics based on the results obtained.
Hence the question remains-how do we know if two problems are similar or not?
This work provided me the opportunity of trying to compare problems based on problem characteristics, some of them which were also identified by Durand and coauthors.
So I went through the collection of 37 different problems I had and found these 5 characteristics to be present in most problems.
Some of these characteristics were also reported by Durand. Other researchers had also reported similar characteristics in the past.
Goal of the problem is about what the problem wants you to do. If the problem states ‘Design a peanut shelling machine’ there is one goal described in the problem. A problem may seek design of more than one thing at times in case of which there are two goals mentioned.
The functional requirements are primary tasks required in the product. These are general action verbs associated with an object or a noun. They describe the function that the device needs to accomplish.
Non functional requirements are quality related requirements. These are non action and non-function aspects which determine the shape, size, usability aspects of the product.
Information about end user is whether the problem states who the end user is going to be or not. This factor helps related the problem solver to who is going to use the product. This is a yes/no type variable.
Reference to existing product is the participant’s familiarity with product that he/she needs to design. For instance, if the problem states ‘Design a hair dryer’, it can be assumed that most participants would be familiar with the product. However, this at times can be a conundrum since participant’s background may also influence whether he/she knows about the product or not. This is also a categorical choice.
This is the protocol that was tested to see the agreement between different evaluators in identification of five elements
the procedure for similarity comparison is simple. You identify the 5 elements in a problem statement and record either their number or presence or absence.
Based on the elements that have been identified, we can begin to make a tabulated list which shows how different problems compare to each other.
In order to check whether the process can be used by other, I conduced a protocol test to see how well people agree in identifying the elements.
People agreed reasonably while identifying functional requirements and goals in the problem, moderately for characteristics end user info and non functional requirements while showed poor agreement for characteristic reference to existing product.
Based on the elements that have been identified, we can begin to make a tabulated list which shows how different problems compare to each other.
In order to check whether the process can be used by other, I conduced a protocol test to see how well people agree in identifying the elements.
People agreed reasonably while identifying functional requirements and goals in the problem, moderately for characteristics end user info and non functional requirements while showed poor agreement for characteristic reference to existing product.
We have five elements which can been seen as 5 information vectors in space.
Cosine angle between the two vectors can be used to determine how far the vectors are spaced apart from each other
However, the variables end user information and ref to existing are categorical and not on the same scale as the other 3.
Hence, values for characteristics FR and NFRs need to be normalized on a scale of 0 to 1 in order to bring them to the same scale.
After normalization, the cosine angle between two vectors can be evaluated to see if they are close or far apart in space.
With the 37 problems I had collected, I evaluated the cosine similarity between each other.
What I could see was that the cosine value for most problems was high which seemed to indicate they are similar.
Min value of 0.46
The second approach that I used to evaluate similarity between problem statements was LSA.
LSA extracts similarity between texts based on their contextual similarity from a large text corpus.
Based on cosine similarity between vector spaces.
I tested the 37 problem statements that I had for sematic similarity.
I used the online LSA tool available on the University of Colorado website.
This is a snippet of the 37x37 matrix that I obtained from LSA.
The maximum LSA score that I saw was 0.53 which was for two problems which asked the participants to design alarm clocks. Low scores of 0 and less were also seen.
With two measures for similarity, we can begin to see how they compare against each other.
The table shows that problem 12 and 13 have very high structural similarity and some what good semantic similarity as well.
If we look at the two problems, we can see that contextually. the two problems are somewhat related since both have reference to material in them.
The structural similarity is rather high, although the two problems have different number of elements. This is due to the scaling down of the original vectors which changes the angle between the normalized vectors as compared to original vectors.
Here are two examples which has been used recently in a study on functional modeling.
The authors have chosen these two similar problems based on the number of requirements contained in both, similar solution requirements and similar number of words contained in them.
Both LSA and structural similarity score show that the problems may indeed be semantically and structurally similar to each other.
Here are two examples which has been used recently in a study on functional modeling.
The authors have chosen these two similar problems based on the number of requirements contained in both, similar solution requirements and similar number of words contained in them.
Both LSA and structural similarity score show that the problems may indeed be semantically and structurally similar to each other.
These two problems were used by Durand and others in their experiment to determine problem similarity based on results.
LSA shows a somewhat low contextual similarity between problems
These two problems were used by Durand and others in their experiment to determine problem similarity based on results.
LSA shows a somewhat low contextual similarity between problems
My fourth research task was to see if design problem choice has any relationship with effectiveness of examples used as interventions during experiments.
For this, I used meta-regression used for identifying potential covariates in literature already published.
Some studies have reported examples as a method of inspiring participants, some report them to be a cause of fixation.
My objective was to see if the design problem used in these studies has any impact on these conclusions which have been drawn.
My collection of 34 studies had 9 studies which studied the influence of examples in experiments.
I extended my search and collected some more experimental studies which had used a between subject design with a control group
In order to do a regression, we need to calculate the effect size of the treatment. Effect size is defined as the difference between means of treatment and control group which has been reported divided by the pooled standard deviation.
Since this effect size is biased for small sample sizes, we use a correction factor to estimate the unbiased mean effect size for the treatments.
The standard error helps estimate how much weight should be assigned to a study.
With the calculations we can start assimilating the effect sizes reported from different studies.
To build a regression model, problem size was used which is a sum of goals, FRs and NFRs which can be identified from the problem statement.
Residual variance is the variance which cannot be explained by the regression variable.
This may be due to other variables which might be causing the variation in effect sizes observed.
Results of the regression model.
For metric quantity, we can see that the with higher problem size, the using examples in fact reduces the quantity of ideas generated by participants who were presented with example solutions.