Stability
The
American Heritage on-line dictionary defines stability
as the following:
"The
state or quality of being stable, especially:
1.
Resistance
to change, deterioration, or displacement.
2.
Constancy
of character or purpose; steadfastness.
3.
Reliability;
dependability"
Measurement
systems have to reliable and dependable.
Before you can run an analysis of your
measurement system, it must be stable.
This means that your measurement system must be
in statistical control - it must be consistent and
predictable. The
only way to know this by monitoring the measurement
system using a control chart.
This is how the stability of the measurement
system is determined.
The
procedure for determining the stability of a
measurement system using control charts is given
below.
1. Obtain
a standard.
Determine
what the expected range of results are for a given
measurement system.
This the range you would expect from production
samples over time.
Ideally,
you would like to compare your measurement system to a
traceable standard.
For example, scales often have weights that
serve as standards.
Quite often, however, this is not possible.
In that case, select a part from production
that falls in the center of the expected range of
results and designate it as the master sample for the
stability analysis.
If
you are in the process industries, you do not make
parts. You
produce liquids, gases and solids.
For example, you may produce polyvinyl chloride
resin. In
these situations, select a certain amount of material
that will serve as the standard for the stability
analysis. You
might select a 50 pound bag of resin to serve as the
master sample. Again,
be sure it is in the center of the expected range of
results. Thoroughly
mix the material to make it as homogeneous as
possible.
If
you believe that the measurement system may not be
stable over the range of the expected results, you may
want to select standards at the high and low ends of
the expected range as well as the middle.
In this case, you will have three control
charts.
2. Measure
the master sample on a regular basis.
The
master sample is tested over and over on a regular
basis. How
often depends on the how often the measurement system
is used. For
example, if the measurement system is used on each
shift, run the master sample once on each shift.
If it is used daily, run the master sample each
day. The
master sample should be run just as any other sample
would be run. This
means that different operators will be testing the
master sample and that it will be run at different
times of the day.
The key is to be sure that all the potential
sources of variation in the measurement system have an
opportunity to be present over time.
3. Plot
the results using an X-mR control chart.
The
master sample results are plotted on an individuals
(X-mR) control chart over time.
After about twenty samples have been taken, the
control limits can be calculated.
4. Bring
the measurement system into statistical control by
finding and eliminating special causes.
The
control chart on the master sample results is then
interpreted for out of control situations (for
example, points beyond the control limits, seven in a
row above or below the average).
For more information on interpreting control
charts, see our April 2004 newsletter available on the
website. Any
special causes indicated on the control chart need to
be investigated. The
root cause should be found and eliminated.
Once there are no special causes, the
measurement system is in statistical control.
You are ready to do further analysis on the
measurement system.
Below
we examine two examples: one of a stable measurement
system and one that is unstable. If you have an
unstable measurement system, you can not trust the
results are you getting.
