In this segment, we're going to talk
about operationalizing design decisions via design vectors.
A design vector is a set of design variables
that specify the aspects of a design choice
that are under our control.
Now these designs can take on two different forms, one,
where we have control or influence
over alternative novel solutions--
these are traditional designs--
and then we also have choices, where
we may have existing or off-the-shelf solutions.
In practice, we often have a mixed set of different designs
and choices that we need to trade and include together.
These can be parameterized via design vector,
and the design vector must then be enumerated,
where we list the possible combinations
of different designs, and then we need to sample.
The proposed designs and choices need
to drive the value attributes.
We call this value-driven design.
This ensures that we're focusing on answering the right question
and not needlessly looking at design trade-offs
that may have no impact on the ultimate, benefit-cost trades.
Now the design vector specifies the space
of designs that are going to be considered in the study.
It's important to recognize this.
The span of variables include all
of the aspects of the design that we're going to consider.
Every design variable has units and a range
that's going to be considered, and possibly sampling levels.
Good design variables capture the range
of possible solutions.
They're realistic, either physically
or in terms of available technology or components.
They're under the control of the designer,
and they impact attributes.
An example design vector for the space tug
includes three elements-- the manipulator mass, which
specifies the size of the payload,
measured in terms of kilograms and has four discrete levels.
This was chosen to drive that capability attribute.
The second design variable is propulsion type.
This is a categorical variable that
included storable biprop, cryogenical biprop, electric
and nuclear thermal.
The propulsion type was chosen so it would drive the delta-v
and the response time attributes.
And lastly, the third design variable
was amount of fuel on board, which
was specified at eight different levels
to get across a large range of different possible masses.
The amount of fuel was chosen to drive, again,
delta-v and response time.
Supporting the development of the design variables,
we developed an approach called design value mapping, which
is a matrix based approach to ensure that design variables
actually drive the attributes.
We start off by taking our attributes, previously listed,
and putting them along the columns of a matrix.
These often include our units and range.
Next, we put the design variables along the rows.
This could be a very large, brainstormed list.
Each design variable should have associated
with them units and range.
So we have a good idea of the kinds of designs
that we're going to be considering.
Next, either individually in groups,
we put a 0, 1, 3, or 9 in the cells,
at the intersection of each row and column.
These represent the degree of impact
that a design variable has on an attribute.
0 means no impact, 1 means light impact,
3 means moderate impact, and 9 means strong impact.
This is our first order model.
Filling out the 0, 1, 3, 9 allows
us to understand the degree of impact
that a design variable has on the attributes.
After we've entered the 0, 1, 3, or 9,
we sum across the rows and the columns.
Columns whose sums are high means that that attribute
is being driven strongly.
If there is a low sum down a column,
that means that that attribute is only weakly driven and poses
a risk that there's some aspects of the value proposition that
cannot or may not be met.
Looking at row sums, we see the design variables
that most strongly impact value and those
that only have weak impact.
Design variables that have weak impact
likely should not be included in a study.
The more design variables and acceptable levels
for those design variables, the larger the tradespace.
The bigger the tradespace, the more
effort required to develop the models and simulations
to evaluate them-- the longer it might take as well.
So there's a tension between having a large enough design
space that drives the values and a small enough design
space that could be explored and analyzed
in the resources allowed for the study.
The benefits of a design value mapping process
include focusing attention on driving value,
identifying preliminary design drivers,
and providing documentation and justification for an inclusion
and exclusion decisions on factors
affecting a tradespace study.
I've also found that it's incredibly effective
as a boundary object for cross-disciplinary
conversation.
Often in complex systems, you have many domain experts
that don't talk to one another, or they locally optimize,
according to their own domain.
By collaboratively filling out a DVM,
these domain experts can identify points
where there may be mismatch in mental models,
where their decisions might negatively
affect other decisions.
A DVM also motivates creative proposing
of new and different design variables,
and therefore concepts, and helps to break away
from anchoring on prior concepts that may not actually
make sense going forward.
A precaution of using a DVM includes the fact
that it's a poor first-order model.
It's really only a screening model
that should be used for focusing on design variables
that most affect the value space.
It also does not take into account context dependence
and complex interactions among the design variables.
These will be taken into account to the extent
possible in the actual modeling and evaluation
phase that happens later.
You will be using DVMs in this week's project.