When SPC is properly understood
and implemented, it is a very powerful tool that assists both workers
and management in controlling and improving the process. Unfortunately,
many SPC applications are full of problems.
 |
Control limits are adjusted
too often. If the process is under statistical control and is stable,
there is little need to adjust the limits. Limits should only be
adjusted if: a) process improvement can be verified, or b) the process
inputs have changed, resulting in a change in process capability
(as measured by CPk). Also, be cautious of computer programs that
automatically adjust control limits based on some pre-defined protocol.
|
 |
Control limits are incorrectly
set. Common errors are: a) forcing limits to be symmetrical about
the product nominal value (or some other, arbitrary target value);
b) failure to use process data to establish the limits; c) arbitrarily
setting the limits to some predetermined values, such as 50% of
the product tolerance. (The formulas for establishing control limits
do not make use of product specifications!)
|
 |
A range chart is not
used. Some people have the misconception that all that is needed
to control a process is to keep the process average centered. This
misses the entire issue of piece-to-piece variation.
|
 |
Points plotted beyond
a control limit are ignored. The usual excuses include: a) "I checked
the next few parts and they were okay." b) "It was only one point!
I never get excited over only one point!" Obviously, the whole concept
of assignable cause has been lost in these instances. (A quick method
to evaluate the significance of these beyond-limit points is to
perform a short CPk analysis. If the CPk value has not changed significantly,
the point was a true, statistical "freak". If, however, the CPk
value has significantly changed, the point was not a freak and,
in fact, the process has changed.)
|
 |
Very little attention
to out-of-control conditions that occur within the control limits.
There is common misconception that the only time a process is out-of-control
is when a point plots beyond one of the limits. In reality, of several
common out-of-control conditions, only one of them is a point beyond
a limit. Missed are the "within-limits" out-of-control patterns,
such as unexpected trends, unexpected cycles, shifts, stratification,
increased variation and few points near the center line.
|
 |
Charts get completed,
filed, but never or only rarely analyzed. This gets back to the
issue of management understanding and responsibility. If SPC charts
are only used to keep records or to satisfy outside auditors, there
is a tremendous loss of valuable data and knowledge that can and
should be used as part of the quality improvement process.
|
 |
Charts are maintained
by someone other than the process operator. While there are specific,
rare cases where this is the best (and perhaps only) way to implement
SPC, such as where laboratory analysis is required, this practice
misses the whole point of giving the process operator knowledge,
authority, responsibility and ownership of the process. As long
as someone else is checking the product characteristic or process
parameter, it becomes that person's responsibility to control the
process.
|
 |
SPC training given without
regard to the timing of implementation. This is a sure way to waste
your training dollars: train in large groups; implement whenever
it is convenient, even if it is weeks or months later. By then,
most, if not all, of the new-found knowledge will have been forgotten.
|
 |
Applying SPC to the
process, but measuring the wrong characteristic or parameter or
measuring the right characteristic or parameter but at the wrong
place. This practice sends the absolutely wrong message to the workers.
It states loud and clear that management does not understand SPC
nor the process. Having the operator measure something over which
he or she has absolutely no control is senseless and downright frustrating.
|
 |
Mixing more than one
process on one chart. The most common situation occurs on machines
with multiple stations, such as a multiple-head drill press. Even
though this may appear to be one machine, each spindle must be thought
of as a separate process. The same issue applies to multiple cavity
molds.
|
 |
Control limits are never
reviewed and adjusted. This is almost as bad as adjusting the limits
too often. The process will change. These changes must be monitored
and their effects evaluated. Monitoring should be performed by someone
very knowledgeable in SPC. Based on the analysis, charts may be
added or deleted; sample frequency may be adjusted. CAUTION: All
apparent changes to the process should be validated before any changes
are made to the control charting process, limits, etc.
|
 |
Using SPC charts without
control limits. This is little more than record keeping. It is okay
if, in fact, you are in the early phases of an SPC application and
you are just gathering data. However, if management believes that
this is an adequate and acceptable way to use "statistics", they
have missed the point completely.
|
 |
The process is set up
well off target. This is a certain way to assure that the process
will "run out-of -control". It also demonstrates a lack of discipline
and understanding of what the limitations of an SPC chart are and
what is required to properly use and interpret a control chart.
|
 |
Limits are based upon
product specification. This is a very common error, generally made
with good intentions. The idea is to force the target to be the
nominal specification value and set the limits at, say, 50% of the
product tolerance. Of course, this misses the whole point that SPC
is based on controlling the capability of the actual process, not
wishful thinking.
|
 |
Corrective action not
noted & recorded. This is a fatal flaw. All that knowledge of what
was done to correct a problem is lost. No one but the person involved
in the corrective action benefits. It is permissible to "code" typical
causes and corrective actions, provided some sort of analysis (pareto?)
is periodically performed to determine if there are any recurring
problems.
|