Robert F. Hart, Ph. D.
Marilyn K. Hart, Ph.D.
It is well known that a point outside the control
limits on a control chart tells the operator
when a special cause of variation has occurred.
In other words, it tells the operator when to
correct or adjust the process. What is not as
well known, the control chart also tells the
operator when to leave the process alone or
run the risk of incurring the following losses
due to overcorrecting:
| 1. |
When unnecessary adjustments
are made, there is an economic loss for
the labor required to make the adjustment
in addition to the loss which may be incurred
due to downtime. |
| 2. |
Unnecessary adjustments will
increase the variability of a stable process.
An excessive number of adjustments (over-correcting
or knob twiddling) will increase the 6 sigma
spread of the process by up to 41%. |
The reason that the process variability may
increase is as follows. If the process is originally
stable and normally distributed, it has an inherent
mean and inherent spread of 6 sigma. Readings
will naturally fall within 3 sigma above or
below the mean. When the process is correctly
centered on the target, attempts to readjust
the level of the process back to target after
getting a reading less than 3 sigma from that
target will result in moving the whole distribution
up and down, as illustrated in Figure 1. This
will result in an increased total spread of
the data.
The following is an example from a Midwest
electric components manufacturing company. The
part was a bracket that needed to be bent to
104° ± 5°. Every time the process
made a bend outside of the specifications an
adjustment was made in an attempt to bring the
process level back to 104°. A histogram
of the data from the "over-adjusted"
process is illustrated in Figure 2. Upon learning
that unnecessary adjustments increase the variability
of the process, they decided to let the process
run for a while without adjustments. A histogram
of the data from the process when it was left
alone is illustrated in Figure 3. Note that
the process is still not meeting specifications,
but the spread of the data is much less than
when the process is over-adjusted. You cannot
"force" a process to have less than
its inherent variability. To be able to meet
specifications, the company must improve the
process to decrease the variability, using 100%
inspection until that is accomplished. Knob-twiddling
will only make it worse.
Figure 1. Spread of the Data
Figure 2. Spread of the “Over-Adjusted”
Process
Figure 3. Spread of the Process When Left Alone
References:
Hart,
Marilyn K. and Robert F. Hart. Quantitative
Methods for Quality and Productivity Improvement.
American Society for Quality Control Quality
Press, Milwaukee, Wisconsin, 1989.
