| Design Of Experiments
The purpose of designed
experiments is to identify the causes of variation in a
process. There are a number of ways to do this
including cause and effect diagrams, Pareto charts,
control charts, and designed experiments. The first
three methods observe the process in its “natural
state”. By contrast, designed experiments perturb
relevant factors in a pre-determined way.
Observing a process in its natural state may fail to
reveal major causes of variation. This is because there
are three sources of difficulty facing the investigator
and designed experiments address them all. The
difficulties are experimental error (noise), confusion
with correlation and causation, and complexity of the
variables in the process. Other advantages of designed
experiments include time. Observing the process in its
natural state may take much longer to identify sources
of variation than a designed experiment. Second
advantage is cost. When there is more than one variable
of interest as is often the case, designed experiments
are much more efficient than varying one factor at a
time. Third advantage of designed experiments is that
this may be the only feasible approach for problems such
as factor screening, optimization, and when interaction
among factors is present.
Designed experiments include a range of experimental
situations. Early stages of investigation often have
many possible variables and screening experiments are
useful to narrow the list. Full and fractional
factorial designs with linear effects are usually used
to further investigate the reduced set of variables.
Once the variables of interest are condensed to 4 or
less, response surface designs may be used to construct
a nonlinear model in the experimental region of
interest. The response surface model allows
optimization of the factor variables for the response
variable of interest.
Design of experiments (DOE) can be effectively applied
during new product development, manufacture of existing
products and in any other situation where multiple
potential variables can affect output. In R&D, robust
product design techniques can be used to design products
which are “robust” to variation during manufacture and
customer usage. Robust products function with less
variation in the presence of uncontrolled variation in
manufacture and usage. In marketing, DOE can be useful
in identifying customer preferences and displeasers. In
this application, the variables in the designed
experiment could include various customer demographic
characteristics.
A
mixture experiment is a special type of experimental
design problem in which the response depends only on the
relative proportions of the design factors (or
components) and not on the absolute amounts of these
components. Chemical processes are typical applications
for mixture experiments. Changing one component in the
mixture necessarily affects the proportion of the other
components. Mixture designs take this dependency among
components into account. Mixture experiments are a
special type of response surface experiment.
|
