|
CONTRIBUTIONS OF DR. TAGUCHI
Qi2 is
proud to have hosted Dr. Taguchi in-person in Los Angeles on May 27, 1998 and again on
April 22, 1999. Dr. Taguchi has both made the usage of Design of Experiments more practical for
engineers and other technical professionals, and he has added some very powerful methods
as well. Taguchi Methods are more than just DOE. Some of his contributions include:
DOE Simplification
Most of Taguchi's orthogonal arrays are easier-to-use rearrangements
of earlier designs (e.g., Plackett-Burman). Interactions can be designed in and analyzed
more easily, and the arrays can be modified for mixed-level designs with simple-to-follow
steps, also.
Parameter Design
By including noise factors in your experimental design (on the
outside of the design matrix as part of a 2-way or greater design to provide balanced
replicates), you can identify how to make your product design or process. Click here to
see a 2-way parameter design example. Parameter design optimizes the controllable
factors to be robust to sources of variation (as represented by the noise factors).
Signal-to-Noise (S/N) Ratios
By converting the raw measured data to a measure of variation, you
can select the preferred levels of the controllable factors to reduce variation. Using the
Target Value is Best Case S/N ratio, you can avoid the problem that the standard deviation
has -- when the mean increases, the standard deviation increases in the same proportion.
The Target Value is Best Case S/N ratio will allow you to study (and reduce) variation
relative to the mean.
Dynamic Characteristics
The dynamic case has multiple target values for the output response
characteristic. This approach involves adding a signal factor as an outside
"Group" as part of a 2- or 3-way design. As an adjustment factor chosen prior to
running the experiment, it should provide a range of input signals that have an ideal
relationship over a range of the output response. The signal factor must be tested at 3 or
more levels, so that linearity over the response range can be evaluated. There are several
types of dynamic S/N ratios developed by Taguchi depending on the problem being studied.
The ideal response function would have maximum linearity, maximum slope, and minimum
variability (noise + lack of fit) about the fitted straight line.
What is Quality?
Taguchi has said that "Quality is the loss imparted to society
from the time the product has shipped." Note that Taguchi takes the customers'
perspective; quality problems encountered before the product has shipped are referred to
as costs. "Society" refers to the global perspective, not the internal cost
center focus used by many companies. The losses are the customers'. The more traditional
"Goalpost" mentality of what is considered good quality says that a product is
either good or it isn't, depending or whether or not it is within the specification range
(between the lower and upper spec limits -- the goalposts). With this approach, the spec
range is more important than the nominal (target) value. But, is the product as good as it
can be, or should be, just because it is within spec? Taguchi says no to this.
The Quality Loss Function (QLF)
The QLF according to Taguchi quantifies the amount of global loss
from the customers' perspective. It captures his definition of quality. According to the
QLF, the quality loss increases as a function of the square of the deviation from the
nominal value, regardless of the location of the spec limits. Click here to see an
example of the Quality Loss Function). The simple underlying quadratic function, L =
k(MSD), is a continuous function. MSD stands for the mean square deviation of the
product's measured responses from the target value. The more consistent the product, the
smaller the MSD, and the smaller the quality loss (L). The slope of the parabola (the
graph of the quadratic function) varies according to the value of the product when it
fails; it is included in the k constant.
The QLF can be used to estimate the cost savings resulting from variation reduction and
also the break-even costs of tighter product tolerances. This latter application of the
QLF is called Tolerance Design. You can use it to justify the increased costs of
specifying the higher-quality levels (lower-variation settings) of the significant factors
identified by the DOE. The selected factor (variable) to be specified more tightly must be
significant in its effect on the performance quality characteristic, and the quality
characteristic must be important from the customer's perspective. |