Gregory Smith
Design Engineer
Michelin Americas Research Corporation
Morphological Charts:
e-mail: greg.smith@us.michelin.com
An Experimental Study
Jenkins Richardson
Product Engineer This paper presents two design experiments to evaluate the impact of the size and shape
E-Z-Go (A Textron Company) of a morphological chart used to identify promising concepts within those charts. In each
e-mail: jlrichardson@textron.com experiment, students designed either a burrito folder or a pet feeder. The first experiment
used two rectangular morphological charts; one vertically oriented with more functions
Joshua D. Summers and one horizontally oriented with more means. The second experiment used four mor-
Associate Professor phological charts, two charts similar to the first experiment and two nearly square with a
Clemson University similar number of functions and means. Participants from similar populations were en-
e-mail: jsummer@clemson.edu rolled and similar protocols were followed to align the results for comparison. A compar-
ison of the significant aspects of each experiment emphasized the problem, the
Gregory M. Mocko participants, the morphological charts employed, and the experimental procedures. Fur-
Assistant Professor
thermore, a critique of the two experiments is provided and joint conclusions are drawn.
Clemson University Experimental results suggest that reducing information in the morphological charts, such
e-mail: gmocko@clemson.edu
as the number of functions, enhances the likelihood for developing higher quality inte-
grated concept solutions. While accurate for most configurations, this benefit is limited in
that the largest sized chart failed to show worse results compared with smaller counter-
parts. [DOI: 10.1115/1.4006261]
Number Concepts Receive tortilla Receive filling Combine materials Fold tortilla
Design problem Device to prepare a burrito Device to feed a cat or dog Possibility of overlap of
participants between two experiments
(overlap did not occur)
Evaluation criteria Low cost (9) Easy to clean (9) Low cost (7) Reduce impact of weightings
(weight) Easy to clean (3) Low number of parts (1)
Low number of parts (1)
Number of evaluation 3 4 Greater granularity
levels
Suggested evaluation values 9 (good) 3 (fair) 1 (poor) 9 (very good) 6 (good) 3 (ok) 1 (poor) Reduce impact of weightings
Evaluation method Decision matrix Not applicable
combinatorial multi-objective optimization problem. Multiple cri- Research was limited to the morphological chart size and shape
teria used to judge each combination determine that which is opti- as related to the number of functions and means per function in
mal for each criterion. Positive outcomes include a minimal the charts that designers control in construction. The number of
computational effort, use of a multitude of information from the functions and the number of means per function influence the
designer, quick feedback to the designer from many candidate sol- number of integrated concepts produced by the chart. Additional
utions, and consistent result events if uncertainty is in the inputs impact areas, such as participants creating charts or participants
(i.e., a range of values rather than a single value). The limitations with varying degrees of experience, were not investigated to avoid
to this approach include the amount of parameters requiring input, introducing experimental variation, and complicate statistical
the lack of representation of nonbehavioral characteristics (i.e., analysis without increasing the participant numbers. A data com-
aesthetics), a consistency of inputs (multiple designers introduce parison of each experiment is summarized in Table 5.
input variations), and the obtained solutions may not be globally
optimal.
Weas et al. [17] discuss using the analysis of interconnected de- 2.1 Design Problem. The design problems for the two
cision areas (AIDA) to combine solution principles (means). experiments were developed to create an innovative solution.
AIDA is a systematic approach for managing complex urban plan- Each problem entailed creating a burrito folder and an automatic
ning problems. This work describes an experiment with under- pet feeder. The design problem was taken from an assignment in
graduate students who applied the AIDA principles to a design the course from a previous semester to ensure that it was suitable
problem. Here, AIDA utilizes an interaction matrix that helps for the participants’ level of knowledge and expertise. By review-
designers to identify compatible sets of solution principles. Posi- ing project reports from the previous assignment, the functional
tive outcomes entailed the automated exploration of the design decomposition and the morphological chart were created using
space, allowed designers to quickly view the effects of design means extracted from the students’ work. The automatic pet
decisions, and the creation and evaluation of unique design con- feeder problem was designed based upon positive results of the
cepts. Limitations include only allowing one performance metric first experiment. In each problem, existing solution implementa-
per iteration (multiple performance criteria with user-defined tions were not obvious to any participant. The problems were
weights being more efficient) and only one binary compatibility developed such that they required multiple functional operations
value (a scale being more appropriate). and in domains requiring no expertise or familiarity beyond that
Others have tried to develop automation tools to help explore of a sophomore mechanical engineering student. Postexperiment
design space [18–20], yet these are not directly related to mor- interviews with participants found no unease about the lack of
phological charts. While each approach provides tools for requisite expertise. This design problem has been used in two sub-
exploring the design spaces within morphological charts, they sequent research experiments [25–28].
cannot exhaustively explore the solution concepts. The objective
here is to understand how to reduce the design space to improve
navigation without sacrificing the solution quality. Thus, guide- 2.2 Morphological Charts. Various sized morphological
lines are needed to help designers learn to construct morphologi- charts were created. As seen in Tables 6 and 7, experiment one
cal charts that effectively explore the design space represented examined two sized “rectangular” morphological charts with sig-
therein. nificant disparity between functions and means. This was the first
attempt to determine experimentally the effect of amount of com-
binatorial information on the quality of solutions generated.
2 Experimental Method Though the number of information units (total number of means)
The hypothesis addressed in this research is: Designers can cre- within each chart was held constant at 15, they were distributed to
ate higher quality integrated conceptual design solutions using either three or five functions. The authors note that the means for
morphological charts with fewer integrated conceptual design sol- populating both charts were quite similar as the chart size and the
utions. More specifically, designers in a certain time can explore design space orientation were of interest, not individual means
more easily a morphological chart with less combinatorial infor- within the charts. Means in both charts are shaded, indicating
mation than one with combinatorial possibilities. Previous studies overlap. For common functions (fill tortilla, fold burrito, and dis-
suggest that designers are challenged by the design space and that pense burrito), three means were common and distributed
reducing it would enhance exploration [21–23]. Using the strategy throughout the charts to prevent bias associated with mean selec-
of Shah et al. [24], the authors found that while many aspects tion position. Means selected in each chart was done randomly to
regarding the use of morphological charts are worthy of study, prevent inclusion of high quality potential means in the easier
examining the impact of morphological chart size regarding the navigated smaller chart. Because the purpose is to compare chart
quality of integrated concepts generated can yield the best results. size, each has different means.
Other items affecting the quality of integrated concepts include Because results suggested a possible effect, in the second
the amount of time to consider the morphological chart, designer experiment additional relationships between morphological chart
experience, and the presence of known good or bad solutions in sizes were studied. Four chart sizes (Tables 8 through 11) tested
the charts, all of which are beyond the scope of this research. the combinations of two different function numbers and two
Function Means
Store and Extrude through Pour from hopper Transfer with spoon Unwrap individual serving Sprinkle in powder form
dispense filling tube from sack from bucket size and drop then rehydrate
Fold burrito Plate under tortilla is Spatula lifts the edges Flexible work Work-surface edges lift up and Tortilla punched through
hinged and folds of the tortilla surface rolls up slide toward center hole in work surface
Dispense burrito Slide on conveyor belt Slide down chute Push off to side Drop from elevated area Grab with mechanical hand
Function Means
Store filling Multiserving package Bulk filled hopper Single serving package
Position tortilla Physical stop Visual marker Work on top of a stack of tortillas
Fill tortilla Extrude filling through tube Pour filling onto tortilla Spoon filling onto tortilla
Fold burrito Spatula lifts edges Roll into tube Punch through opening in table
Dispense burrito Gravity Conveyor belt Mechanical hand
Means
Function 1 2 3 4 5
Fill bowl Screw Rotary pocket Conveyor belt Solenoid Piston
Regulate food Weight Windup–spring Fixed volume container Fluid displacement Trip laser
Signal (to fill bowl/ inform owner) Weight Lever in bowl Camera/image processing Proximity sensor Radar
Power source Battery Gravity Air pressure—compressed air Windup–spring Engine
Means
Function 1 2 3 4 5 6 7
Fill bowl Screw Rotary pocket Conveyor belt Solenoid Piston Blower Robotic arm
Regulate food Weight Windup—spring Cam Fixed volume Fluid displacement Plunger Trip laser
container
Signal (to fill Switch Weight Lever in bowl Camera/image Proximity sensor Radar Pressure sensor
bowl/inform owner) processing
Power source Battery Gravity Air pressure—compressed air Plug–AC Windup—spring Flywheel Engine
Means
Function 1 2 3 4 5
Fill bowl Screw Rotary pocket Conveyor belt Solenoid Piston
Quantity of food Weight Butterfly valve Fixed volume container Wheel—fixed volume Trip laser
Regulate time Cam Pendulum Windup—spring Reversible chemical reaction Fluid displacement
Record keeping Analog–dial Graduation etched on Dial-on dispenser Measure and display weight Rotary counter
transparent (moves if dispenser
container moves!)
Signal owner to Buzzer (electronic) Light SMS/email Emit odor Play recorded
fill the reservoir message
Signal itself to fill bowl Weight Lever in bowl Camera/image processing Proximity sensor Radar
Error check Capacitance Laser LVDT Electric contact—bowl Weight
and dispenser
Power source Battery Gravity Air pressure—compressed air Windup—spring Engine
different means numbers. Here, the means of the largest chart to over 5 million possible combinations). Table 12 shows morpho-
were populated and pruned, without considering the quality of logical chart comparison of experiments one and two.
means to fit the smaller charts. Essential functions were retained
upon pruning. 2.3 Participants. Participants in both experiments were en-
The second experiment used morphological charts with several rolled in the same second year mechanical engineering course
concepts several magnitudes larger than the first experiment (625 (Table 13) with none participating in both experiments, while the
Means
Function 1 2 3 4 5 6 7
Fill bowl Screw Rotary Conveyor Solenoid Piston Blower Robotic
pocket belt arm
Quantity Weight Butterfly Cam Fixed Wheel— Plunger Trip laser
of food valve volume fixed volume
container
Regulate Timer— Cam Pendulum Bimetal— Windup— Reversible Fluid
time microprocessor heater spring Chemical displacement
reaction
Record Analog— Graduation Barometer Gauge stick Dial on Measure and Rotary
keeping dial etched dispenser (moves if display counter
on transparent dispenser moves!) weight
container
Signal owner Buzzer Light Vibrate SMS/email Emit odor Ring bell Play recorded
to fill the (electronic) (mechanical) message
reservoir
Signal itself to Switch Weight Lever Camera/image Proximity Radar Pressure sensor
fill bowl in bowl processing sensor
Error check Hall effect Capacitance Laser LVDT Electric contact— Weight Photo detector
bowl and dispenser
Power source Battery Gravity Air pressure– Plug—AC Windup—spring Flywheel Engine
compressed air
Experiment #1 Experiment #2
(burrito folder) (pet feeder)
Table 13 Participants
Experiment #1 Experiment #2
(burrito folder) (pet feeder)
Participants’ class Sophomore level mechanical engineering course in design and kinematics
Number of class sections 2 1
Number of participants 25 29
Minimum experience with morphological charts Morphological charts lecture to class before the experiment
Morphological charts review at start of experiment
Additional experience with morphological charts Unknown
Number of repeat participants Unlikely due to the time separation (1 year) between experiments
relevant characteristics remained identical. Although the first For the first experiment, the evaluation criteria were low cost
experiment encompassed two sections of the same class with dif- (9), easy to clean (3), and number of parts (1). For each solution,
ferent professors, both performed equivalently and results were the participants were instructed to score their integrated solution
considered as a single group (see Sec. 4.1 for proof of statistical against these criteria as either good (9), fair (3), or poor (1). It was
equivalence). As the second experiment encompassed a single determined that low evaluation concepts (poor, 1) for the highest
class section, pooled population analysis was not required. The weighted criteria (low cost, 9) never achieved similar scores to
minimum experience with morphological charts was normalized high evaluation concepts (good, 9) regardless of the evaluations
in which students were required to build charts interactively. for the other criteria (number of parts and easy to clean). An inter-
mediate evaluation level (6) was established and low cost-7 and
2.4 Procedure. Both experiments proceeded similarly easy to clean-9 criteria were reweighted. To correct this problem
(Table 14). Each participant was provided with a prepopulated in experiment two, evaluation criteria and levels were adjusted to
morphological chart and solution form, in which participants reduce the weighting impact, and changed to easy to clean (9),
listed either eight (first experiment) or ten (second experiment) low cost (7), and number of parts (1). Evaluation solution values
integrated conceptual design solutions and rated them against pro- were very good (9), good (6), ok (3), and poor (1). Participants
vided criteria. The prepopulated morphological chart was used to using other values were not excluded and these nonstandard val-
control the variability of possible creativity within the students. ues were accepted as is. Participant ratings and criteria weightings
The only related activity tested here was the integration of means- were used in a decision matrix to determine final scores for each
to-form solutions. solution. From this matrix the top three integrated solutions,
Experiment #1 Experiment #2
(burrito folder) (pet feeder)
Morphological hart training in advance Lecture on morphological charts presented class before the experiment
Morphological chart training on day of experiment Review of morphological charts presented at start of experiment
Time allowed to generate concepts 30 min
Number of concepts required to generate 8 10
Number of concepts chosen for judging 3 (or more if a tie existed) Exactly 3
Method for accommodating a tie in Evaluate all tied concepts. Participant score A set of exactly three concepts that included the
individual scoring of concepts determined using Eq. (3) highest ranked concepts was selected
Judge distribution system Concepts divided equally among five judges Concepts divided among two
judge panels of three judges
Judge evaluation method Evaluate the concepts using identical decision matrix and weightings as participants
defined and evaluated by participants, were collected for further the average of the top three solutions identified. In cases of no
analysis by judges. clearly identified third place, third place group scores were aver-
Participants were directed as to how to record and score their aged to create a composite third score. This score was averaged
chosen integrated conceptual design solutions. After the introduc- with first and second place scores (see Eq. 2).
tion, they were allowed 30 min to develop and evaluate their
design solutions, and instructed to form the highest quality con- solution3a þ solution3b
solution1 þ solution2 þ
ceptual combinations. The time limit for the concept generation score ¼ 2 (2)
phase was fixed by the need to conduct the experiment without 3
exceeding the allotted time. To choose the number of concepts for
the first experiment, participants generated and evaluated eight For the second experiment, three concepts were selected that
concepts in less than 30 min. Only one participant failed to gener- best represented the participant, including the three highest ranked
ate the full number of concepts; therefore, the second experiment concepts. Each judge ranked concepts using the same evaluation
required more concepts to encourage more novel solutions. levels as participants. The final participant scores were the aver-
age of nine scores, consisting of three judges’ scores for each of
the three selected concepts. The authors note that other evaluation
3 Analysis scales may be used with different results. This exploration of the
3.1 Jury. Each experiment employed a different system for role of evaluation scores and quality is beyond the scope of this
judging the integrated concepts; five individual judges or two study. A best practice baseline for creating a differentiated con-
groups of three individual judges. All judges were drawn from the cept set using the geometric scale was also used [30].
second graduate students in the CEDAR (Clemson Engineering The choice of evaluation criteria and their respective weights
Design Application and Research) Lab. To limit the time invested influenced the integrated concepts created by the participants. To
by each individual judge, results were divided amongst judges for ensure that all participants were working toward a common goal,
each experiment. In the first experiment, each judge was assigned they were required to score concepts against these criteria instead
approximately 60% of the concepts generated from each chart. of a personal preference. This experiment is used to study only the
Each integrated concept was also evaluated by three of the five process of exploring the design space to generate integrated con-
judges. While cumbersome, a uniform distribution of integrated cepts. While evaluation of the integrated concepts is not the focus
concepts among the judges was achieved. Intrarater consistency of this experiment, creating a consistent set of criteria for the par-
found no bias in judging [29]. ticipants and judges reduced the variation between what partici-
The second experiment used a simpler method for assigning pants identified as the best-integrated concepts without these
integrated concepts to the six judges. The collection of integrated specific criteria.
concepts was divided in half, with each evaluated by a three judge
panel working individually. While easier to establish, it was nec-
essary to elucidate any consistency between the two judging pan- 4 Results
els to determine the validity of combining evaluations from both
4.1 First Experiment Results. Of thirty participants in the
panels. An insufficient overlap in one morphological chart size
first experiment, four misunderstood the directions, creating their
prevented a full intra-rater analysis. The use of a system similar to
own means for each function instead of making selections from
the first experiment to spread the concepts evenly among judges
the morphological chart. They were excluded from the analysis.
may have mitigated the impact of differences between panels.
Another failed to rank all of their integrated conceptual design
Care must be taken when assigning evaluations with the ideal
solutions and was excluded also. As such, it can be inferred that
having each judge evaluate all integrated concepts. The logistical
there was ample time to complete tasks and that the scope of the
concern here is that a corresponding growth in the number of so
design problem was acceptable.
evaluations requires each judge to make that many more. Any
At the end of the exercise, all handouts were collected. The
assignment of evaluations among judges must minimally impact
three highest rated design solutions from each participant were
statistical analysis of the results. The additional effort to create
combined into a list of design solutions for evaluation. Absent no
evaluations in the first experiment eliminated the need to deter-
clear group of three or if there were a tie for third place, extra
mine if the judging could be pooled for statistical analysis, while
design solutions were added, and the list was purged of duplicate
reducing the uncertainty of the results.
solutions. This list of the three highest rated concepts covered
14% and 25% of the design spaces within the morphological
3.2 Score Calculation. Once the students had judged their charts (Table 15). The number of unique combinations generated
solutions, judges evaluated them accordingly. A final score was for the two charts were roughly equivalent (31 and 33). Further,
assigned by averaging the three scores provided by the panels, nearly half of student defined concepts were repeated by others in
which were used to evaluate participant-generated solutions. For each morphological chart (56% and 47%). The major difference
the first experiment, each participant was assigned the score from between the design space explorations was the actual coverage of
Table 16 Summary of ANOVA results: comparing quality of means generated for 3 3 5 and 5 3 3 morphological charts
Chart #1 (3F 5M) Chart #2 (5F 3M) 0.018 Means are not equal
Section 1 Section 2 0.42 Means are equal
Chart #1 for section 1 only Chart #1 for section 2 only 0.93 Means are equal
Chart #2 for section 1 only Chart #2 for section 2 only 0.44 Means are equal
Morphological chart size (F M) 45 47 85 87 All 4 function All 8 function
Number of participants 8 7 8 6 15 14
Number of possible concept combinations 625 2401 390625 5764801 2401 5764801
Number of unique concept combinations Quantity 19 21 24 18 38 42
in set of “top 3” % of generated concepts 1.3% 0.29% 0.0020% 0.00010% 0.62% 0.00024%
a
0.10 < p 0.20.
b
p 0.10.
c
0.20 < p.
d
There were insufficient data for the 8 7 size to evaluate Judge panel #1.
e
It was unnecessary to evaluate individual judge panel relationships here. Panels were consistent for both sizes (see above).
set of notional relationships between the four morphological panels scored them consistently (Table 20), and they can be com-
charts. pared based upon combined scores. ANOVA comparison showed a
difference in the means of the two groups (Table 21). A comparison
4.2.3 Rectangular Morphological Charts. Consider the verti- of average scores for each chart shows that the vertical chart pro-
cal (4 7) and horizontal (8 5) charts above, because both judge duced higher scores than the horizontal chart. Therefore, reducing