|
SPC - statistical process control
Statistical Process Control chart introduction for non-mathematicians.
Statistical Process Control = SPC.
SPC Concept
> All processes vary.
> The aim of SPC is to minimise variation.
> Variation is expressed in "sample standard deviations"
(SD)
> SD is a mathematical term, based on probability theory.
Consider the process of driving a car along a dry,
straight, level road.
Look at your hands on the steering wheel - they move as you
make slight adjustments.
Q: Why are the adjustments necessary ?
A: Because the road is not totally flat,
there's some play in car's suspension, etc.
If you draw a graph of the movements you will have a
"normal distribution curve"
In a normal distribution, most of the time your hands are
in the middle
(In the secret language of SPC the
"middle" is called the "average" or "mean" or "x-bar")
But sometimes your hands are a
little to the left,
at other times they're a little to the right
And on rare occasions they're on the
extreme left or
extreme right.
SPC Application
Using the "sample standard deviation" (SD), you
can then calculate and chart the usual extremes of a process under a given set of
conditions. This variation is "normal" and we cannot reduce the variation
without making fundamental changes to the process (eg going on an
advanced drivers course, buying sports suspension, etc.)
Because of the properties of the normal distribution it is
usual in SPC or quality control applications to multiply the standard deviation by
three and then add that to and take it away from the average which gives a
confidence level which encompasses 99.73% of the observations or "population".
For example:
average = 20
1 SD = 1.3
3 SD = 3.9
average + 3 SD = 23.9
average - 3 SD = 16.1
therefore 99.73% of all that we see will be between 16.1 and
23.9
Back in the car...
A child runs out in front of you and you swerve to avoid him.
The chart now has a "blip" which is outside the normal curve. This indicates that something unusual has happened.
In SPC applications it is often useful to
recognise what is "normal" and what is "non-normal". This case is clearly
"non-normal".
In some situations also helps to know "when" things happen.
Therefore, an SPC control charts also has a time axis.
SPC Control charts
We can then apply these concepts in an SPC chart for monitoring customer
complaints.
Having collected the basic information we can then add the
average and +/- 3 SD markers.
What
we are saying is this:
1. Given the normal variation of the
process we will always have a level of complaints. There is nothing we can
do without making fundamental changes to the process.
2. As long as the individual plots are
between the red lines then the process is exhibiting normal variation.
3. BUT clearly something unusual
(non-normal) happened in March - we had far more complaints than usual. We
need to investigate and find out why.
4. THIS IS THE CLEVER BIT -
February's low result is equally unusual. We need to find why it happened and build that in to
the process. Do it often enough and the average line will go down, and
eventually become zero.
To help the investigation, use appropriate
problem solving tools
The problem with traditional SPC
Most SPC control chart techniques are based around the
+/- 3 SD method, which covers 99.73% of a population. 99.73% sounds good
until you realise that this means that 0.27% are not accounted for, there is 1/4
percent chance of something slipping through the net.
If you are trying to control critical processes or very
high volume (eg aerospace, building nuclear reactors, making silicone
chips, etc.) then you will get it wrong once in every 370 attempts.
Such critical industries tend to use Six Sigma (+/- 6 SD)
methods as it gives a 99.9997% (3 in a million failures) confidence level.
Three in a million sounds much better than one in three
hundred and seventy.
Need more ?
Click here for on-line
SPC
training. And check
their toolbox for other quality improvement
tutorials
Or buy R H Caplen's excellent book
A Practical Approach to Quality Control.
|