Experimental Design
16.621
Elements of Designing an Experiment
? Goal is to assess your hypothesis
– The main thing is to keep the main thing
the main thing - [Tom Curran, AFRL]
? The design goal for your experiment is
to achieve your objective and satisfy
your success criteria.
? Your HOS are the touchstones for your
experimental design.
SOME FRAMING QUESTIONS
? Once the HOS is set we need to define in depth the path to get there
? Need to think through the whole process on an end-to-end basis
? Need to identify the hard parts
? Need to identify the key milestones (“mid term exams”)
? This will include (at some level) questions such as:
- What will we do?
- Why will we do it?
- Where will we do it?
- Who will help us?
- How will we do it?
- How well do we have to do it?
- When will we do it?
Project Specific Applications of these Questions
? What is the best overall approach to achieve my
objective?
– e.g., should I test children’s car seats in the lab or
in actual airplane environment?
? What are my choices of facilities?
– e.g., should I use a water tunnel or a wind tunnel
to determine the performance of a split-tip keel?
? How will I measure my independent variables?
– e.g., Should I measure rotational speed of my
airspeed indicator with a stroboscope or a shaft
encoder?
Observations About Design
? Design is an inductive process.
? Ways to learn design
– Study past designs that have been
successful and extract patterns
? Look at the online models
– Work with a “master” designer
? Involve your advisor
– Do it yourself and receive feedback
? Basically this is the 16.621 process
The 621 Design Process
Idea!
HOS
Conceptual Design
Detailed Design
Version II
Overall experimental
approach; “scope
and scale”; type of
model(s), materials,
software; data to be
taken; sources of
error,….
Version I Version III
Detailed drawings and
dimensions; pseudo
code; specific
equipment; make/buy
decisions; construction
methods; calibration
methods; detailed error
analysis, budget and
schedule,….
Concpetual Design Process
? Start with an open mind. Avoid preconceived approach.
? Brainstorm with your partner
– Think of multiple ways to accomplish your objective(s)
– Ask other people, keep your “eyes open”
– Avoid rejecting ideas, particularly wild ones, at this stage.
– Keep on the cusp of creative thinking.
– Write these ideas in your notebooks.
? Develop a set of desired attributes to use to compare your different
designs, keeping your HOS in mind.
? Sort through the different options to arrive at a best design to achieve
your objective.
? Verify your conceptual design supports your HOS
– Revise HOS if original set is not feasible
? Then proceed with “detailed design”.
A 621 Example
The potential strategies for creating the aerated ice are as follows:
1 Drill air holes into the ice.
2 To use a mixture of crushed ice and water close to freezing
point.
3 Suspend small pockets of air in the ice mix (using tethered
balloons)
4 Construction of a specialized rig that constantly blows air
through the ice mix.
5 The use of soda water or carbonated water.
6 A combination of these strategies.
The most effective method for creating aerating ice will be
determined during the preliminary testing stage of the project. The
method will be chosen based on the best combination of speed,
ease of production, and quality of ice produced.
Source: Bennett, W. “Simulating and Testing Ice Screw Performance in the Laboratory”, 16.621 Design
Proposal, Fall 2002
Fidelity of Your Approach
? An experiment is usually a “model” of the “real thing”.
– It may be too complex or expensive to test the real thing
– It usually will be difficult to have control over the variables
in a real operational environment
– e.g. a part task simulator will be used to test a cockpit
display instead of modifying a real cockpit.
? The “fidelity”* of your (model) experiment needs to be
assessed as part of the design process.
? Lack of fidelity is one source of errors.
? A good way to consider these errors in the design stage is to
write them down, assess their importance, and consider
mitigation strategies.
*Fidelity: Exact correspondence with fact or with a given quality,
condition, or event; accuracy. http://dictionary.reference.com/
A 621 Example
A number of potential error sources in the experiment have been identified. These are
summarized in Table 1 along with a brief description of the problem and how to decrease
or eliminate the error.
Table 1: Sources of Error
Source of Error Description Method of Mitigation
Hysteresis
Shape memory alloys have a
hysteresis effect that could
potentially cause the actuators to
get “stuck” in a particular state
In order to decrease the effects of
hysteresis, ample cooling time will be
given between consecutive test runs
Cyclic Loading
Depending upon the frequency and
duration of the perturbations, the
actuators could experience several
thousand cycles of heating and
cooling throughout the run of the
experiment
A low frequency of perturbations, and
periodic checks of a baseline
condition will be used to ensure that
the actuators have not deformed over
time.
Sensors
Commercially available
accelerometers will have a certain
offset, and will drift with time
These errors will be corrected during
data analysis by shifting and rescaling
raw data
Inconsistent
Release
If the actuators start functioning
before the parachute is released or
fully inflated, errors resulting from
initial conditions could occur.
In order to ensure that the parachute
has completely inflated before a test
function is sent to the actuators, a
time delay will be built into the
function generator based on the initial
trial run of the experiment.
Source: Wong, B. “Assessment of Shape Memory Alloys for Steered Parachutes ”, 16.621 Design
Proposal, Fall 2002
Summary
? Design your experiment to achieve your
objective, satisfy your success criteria and
assess your hypothesis.
? The conceptual design stage starts with
an open mind and brainstorming, and then
uses objective criteria to select best
approach, considering modeling fidelity
(errors).
? Continue to refer to or revise your HOS
Active Learning
? Working with your partner, pick one
aspect of your experimental design
(facility, measurement,…) and develop
two or more approaches.
? Pick two or three attributes you would
use to select one approach over the
other.
? Be prepared to report to the class at the
end of the hour.