The Discovery of Precontrol

Precontrol was developed by the consulting company of Rath and Strong for a
major Fortune 500 company that had become disenchanted with cumbersome
and ineffective control charts. Precontrols founder, Frank Satterthwaite, is a
brilliant statistician who established its theoretical underpinnings in a
comprehensive paper over thirty years ago. Unfortunately, just as precontrol
was gaining recognition, the United States, flushed with its economic success
in the post World War II years, threw out all statistical methods in industry,
control charts and precontrol included. Then, as SPC became fashionable
again in the 1980s and the control chart became its centerpiece, precontrol
started to reappear on the statistical horizon. Three years ago, the ratio of
control chart users to precontrol practitioners was 99:1. Today the ratio is
85:15. In a few years, as the simplicity and effectiveness of precontrol is
better publicized, the control chart will be relegated to history, as in Japan, and
precontrol will become the principal maintenance tool in the SPC world.

The Mechanics of Precontrol in Four Easy Steps

The mechanics of precontrol can be taught to anybody in industry, including
line operators, in less than ten minutes! There are four simple rules to follow:
Rule 1: Divide the specification width by 4. The boundaries of the middle
half of the specification then become the precontrol (P-C) lines. The area
between these precontrol lines is called the green zone. The two areas
between each precontrol line and each specification limit are called the yellow
zones. The two areas beyond the specification limits are called the red zones.
Rule 2: To determine the adequacy of process capability at that time, take a
sample of five consecutive units from the process. If all five fall within the
green zone before two consecutive yellow zone results, conclude at the
moment that the process is in control. (In fact, with this simple rule, the usual
samples of fifty to one hundred units to calculate Cp and Cpk are not
necessary. By applying the multiplication theorem of probabilities or the
binomial distribution, it can be proved that a minimum Cpk of 1.33 will
automatically result.) Full production can now commence. If even one of the
units falls outside the green zone, the process is not in control. To determine
and reduce the cause of variation, conduct an investigation. Use engineering
judgement (which is not infallible by a long shot) or, better still, multi-vari
charts and/or paired comparisons, sometimes followed by variables search or
full factorials.
Rule 3: Once production starts, take two consecutive units from the process
periodically.

The following possibilities can occur:

1. If both units fall in the green zone, continue production.
2. If one unit is in the green zone and the other in one of the yellow zones,
the process is still in control. Continue production.
3. If both units fall in the yellow zones, adjust the process setting, if they are
both in the same zone. If they are in different zones, stop production and
conduct an investigation into the cause of variation.
4. If even one unit falls in the red zone, there is a known reject, and
production must be stopped and the reject cause investigated. When the
process is stopped, as in 3 and 4, and the cause of variation identified and
reduced (or eliminated), Rule 2, i.e., five units in a row in the green zone,
must be reapplied before production can resume.
Rule 4: The frequency of sampling of two consecutive units is determined by
dividing the average time period between two stoppages (i.e., between two
pairs of yellows) by six.* In other words, if there is a stoppage (two yellows),
say, at 9 A.M. and the process is corrected and restarted soon after, followed
by another stoppage at 12 noon (two yellows again), the period of three hours
between these stoppages is divided by six, to give a frequency of sampling of
every half hour. If, on the other hand, the period between two stoppages is
three days, the frequency of sampling is every half day.
Figure 15-3 is a graphical portrayal of precontrol, with a summary of the four
simple rules. (The normal curve on the right is a special case, where the
process width and the specification width are equal. It is used to calculate the
maximum risk and is shown only as an illustration.)

The Statistical Power of Precontrol

The theory behind the effectiveness of precontrol is based on the
multiplication theorem of probabilities and the binomial distribution.
Although the mathematical derivation of the and risks for precontrol is
beyond the scope of this paper, the following is a summary:
The worst risk the risk of overcorrection, i.e., stopping a process when
it should continue is around 2 percent of the time.
The worst risk, the risk of allowing a process to continue when it should
be stopped, expressed as product percent defective, is close to 1.5 percent.
These two risks are paper statistical maximums, established from a great
number of empirical in plant trial and error runs in which we achieved
real-world results of virtually zero variations (0.00 percent). Thus,
*

When precontrol was first developed, the time period between two stoppages had to be divided by 24 to
determine the frequency of sampling. This was later found to be much too conservative and the number was
changed to 6. However, if much greater protection against continuing a process that should be stopped is
desired, the frequency of sampling can be increased to 10, 12, 15, even up to 24. In most industrial
applications, however, such an increase is not at all desired.

reducing these risks further would make precontrol needlessly less costeffective.
When the process width is greater than the specification width, generally
Cpks of 0.8 or less, precontrol is so sensitive that it will stop the process at
least ninety-nine times out of one hundred and force an improvement
investigation.
When the process width is 75 percent or less of the specification width
Cpks of 1.33 or more the use of precontrol becomes most productive.
The process is in control and precontrol will keep it there.
When the process width is 50 percent of the specification width Cpk of
2.0 precontrol will allow hundreds and thousands of units to be produced
without a single reject.

The beauty of precontrol, therefore, is that it is an ideal incentive/penalty tool.

It penalizes poor quality by shutting down the process so often that problem
solving with the use of design of experiments becomes imperative. It rewards
good quality by sampling more and more infrequently. Control charts, on the
other hand, have no sampling rule or built-in flexibility to deal with this
important feature.

Charting Precontrol: Easing the Operators Burden

As opposed to control charts, a graphical record with a chart is not mandatory
in precontrol. The machine or process operator has the simplest of rules:
Two greens or one green and one yellow: continue. Two yellows in the same
zone: adjust; two yellows in opposite zones or one red: stop. There is no
need to distract the operator with long and painful data entries. However, if a
precontrol chart is required for historic purposes or as proof of control to a
customer or from a supplier, the operator can just make easy slash-mark
entries prepared on forms with green, yellow, and red zones. Figure 15-4 is an
example of a precontrol chart used to control the thickness of chrome, nickel,
and gold deposits on glass in a sputtering machine. The actual readings are
recorded for each precontrol sample of two units. From these readings, Cp
and Cpk values are easily calculated and histograms plotted, if required.
There is no need for manual calculations, handheld calculators, or expensive
computer programs, an important advantage for small supplies who do not
want or need money thrown at a process with expensive control charts.

Tackling Multiple Quality Characteristics

Precontrol is also a far more economical tool in controlling multiple quality
characteristics on a product or process. As an example, if the variations in a
thirty-six-cavity mold must be monitored in an injection molding machine, the
number of readings required to establish even trial control limits in a control
chart would average 3,600. By contrast, precontrol would determine process
capability with 5 readings for each cavity, or a total of 180 readings.

It is not necessary to monitor all parameters of a product or process

continually after initial process capability is confirmed on each parameter.
Only the most important or the most troublesome parameters need to be
sampled. As an example, in the case of the thirty-six-cavity mold, only the
single largest and smallest cavities may need ongoing surveillance. The rest
can be monitored infrequently or not at all.

The Versatility of Precontrol

The rule of six samplings between two stoppages need not be associated with
time alone. In a very fast operation, where several hundred units are produced
in a minute, the quantity of units between two stoppages (two pairs of
yellows) can be divided by 6 to determine sampling frequency.
Precontol is also a versatile technique in determining when to adjust a process
or change a tool, rather than a technique tied to fixed or arbitrary time periods.
The time to change is when the process drifts into one or the other yellow
region. Two yellows give the signal. It is as simple as that.
Another feature of precontrol is its applicability to one-sided specification
limits. As an example, consider a product with a minimum requirement of 10
volts, but no maximum. Instead of two precontrol lines drawn in the middle
half of the distance between the lower and upper specification limit, there
would now be only a single precontrol line. It would be located midway
between a desired target or design center and the lower specification limit. A
less desirable method would be to locate the single precontrol line midway
between the product average and the lower specification limit. In the above
example, if the design center were at 15 volts, the single precontrol line would
be drawn at 12.5 volts. If the average of the product was 16 and the
alternative method used, the single precontrol line would be at 13.

Converting Attributes Into Variables: The Bo Derek

Scale
Because precontrol deals primarily with the measurement of continuous
values, called variables, it may appear to the uninitiated that it cannot be
applied to attributes that deal in discrete numbers, such as accept or reject,
pass or fail. They may argue that, whereas precontrol is a better substitute for
and R control charts, the use of p and c control charts* is unavoidable.
Attributes, however, need not be any limitation whatsoever to precontrol.

P and c charts suffer from the same fundamental weaknesses as do X and R charts, and because they deal
with attributes they lack the discriminatory power of variables. They are even worse than X or R charts and
for this reason they will not be discussed in this chapter.

The trick is to convert an attribute into an artificial variable, using a

numerical scale from 1 to 10. In what is humorously called the Bo Derek
scale from the actresss movie Ten, the number 10 would correspond to
perfection in the quality characteristic being measured and the number 1
would equate with the worst possible reject. There would be other grades of
quality ranging between these two extremes. As an example, to convert
attributes such as paint, color, scratches, gouges, dents, pinholes, etc., into
variables, a committee consisting of the customer, sales, engineering,
manufacturing, and quality would grade ten physical samples, ranging from a
totally unacceptable 1 to a perfect 10. These physical samples would, then,
constitute a variables scale to determine the appropriate number for a
similar defect in production. In case physical samples are too difficult or too
inflexible to develop and maintain, color photographs can be substituted. (It
may not be necessary to use a full 1 to 10 scale; a truncated 1 to 5 scale can
sometimes suffice.) In a numerical scale of this kind, let us assume that the
specification limits are from 4 (worst acceptable) to 10 (perfect). Because this
is, in reality a one-sided specification, a single precontrol line would be drawn
at 7.

Weighting Attributes in Terms of Importance

A weakness of p and c control charts is the practice of lumping all types of
defects into a single total for attributes. Such charts lack the sensitivity
associated with grading defects by importance. In precontrol, every defect
mode is assigned a weight in proportion to its importance. This weight,
multiplied by the number of defects in that mode, gives the score for each
defect type. These scores, when added up, result in a weighted total for a
particular time, and a precontrol chart can then be plotted. Such charts have
the effect of a magnified scale and are far more sensitive and pinpointed in
terms of defect modes than is a catchall p or c chart.
To enable the operator to adjust the production process whenever two
consecutive yellow results are encountered, several companies have elected to
precontrol each important quality characteristic (defect mode) separately. This
procedure pinpoints the need for a particular adjustment.
Figure 15-5 is an example of such a precontrol chart. It expands solder
defects, each with its own weight based on the importance of the defect in
terms of potential field failure. The precontrol chart highlights the various
types of defects found at a particular time, trends with time, and whether or
not the weighted defect scores exceed the tight precontrol lines of 50 defects
per million connections (i.e., 50 ppm).

A Capsule Comparison of Control Charts vs. Precontrol

Table 15-4 is a comparison of control charts and precontrol. It clearly shows
the weaknesses of control charts versus the strengths of precontrol in terms of
simplicity, use by operators, mathematical complexity, applicability to small
production runs, recalibration of control limits, machine adjustments,
frequency of sampling, discriminatory power, conversion of attribute charts,
and most important of all economy.
The operator can react instantly when two consecutive yellows (still good
parts!) are encountered. Precontrols advantage here makes the attainment of
zero defects an actual possibility rather than just a verbal goal.
As a postscript, if precontrol, rather than control charts, had been used in
Table 15-3 to control bushing wear, the very first sample at 8 A.M. would
have indicated that at least one of the five units checked would have fallen
beyond the precontrol lines of 5.001 and 4.999, indicating lack of process
capability. This shows the sensitivity of precontrol in stopping a bad process,
whereas a control chart would have led one down the garden path all day,
declaring the bad process to be good.
Further, if precontrol had been used in Figure 15-2, it would not have shut this
process, confirming productions contention of zero rejects among thousands
of units.
Among the unconverted who hang onto control charts, Ford is one of the most
vociferous and tyrannical. Its quality auditors refuse to accept precontrol
because it is not in the book! Its professionals, on the surface, appear to be
somewhat more logical. They claim that precontrol with fixed specifications,
and, therefore, fixed precontrol limits, does not permit continual, never-ending
improvement; control charts supposedly produce improvement, with narrower
and narrower control limits. However, these professionals miss the point
altogether. Fords argument is tantamount to saying that the temperature in a
room can be controlled by a thermometer! Neither control charts nor
precontrol are tools for improvement, which can best be achieved through
DOE.

Practice Exercise on Precontrol: The Wire Bonder

An automatic wire bonder, which bonds a wire the thickness of a human hair
to the die and post of a transistor, is put on precontrol to maintain statistical
control. The integrity of the wire bond is checked with a destructive pull test
on the wire. The specification for the bond strength, before the bond is lifted
on either side, is a minimum of 6 gm and a maximum of 14 gm. The initial
sample of five units had the following readings: 8.7 gm, 9.0 gm, 9.4 gm, 8.9
gm, 10 gm.
Question 1. What are the values for the precontrol lines?
Question 2. On the basis of the initial sample of five units, is the process in
control?
Question 3. How frequently (initially) should a sample of two units be tested
for bond strength, assuming that when full production began the period
between stoppages (two pairs of yellow) averaged 12 hours?
Question 4. During subsequent production, the results of two sample units,
drawn periodically from the process, were as follows. (For the purposes of
this exercise, waive the rule requiring five units to fall within the green zone
in order to reestablish process capability.)
Sample No.
1
2
3
4
5
6
7
8
9
10
11
12

Unit 1
9.4
9.0
8.9
8.5
8.4
8.0
8.0
7.5
13.0
12.0
11.6
11.0

Unit 2
9.0
8.8
8.6
8.1
8.0
8.0
7.6
7.3
13.0
12.0
11.4
10.8

Action
____
____
____
____
____
____
____
____
____
____
____
____

What action would you take on each sample: Would you continue production
or stop?
Question 5. What nonrandom trends do you detect in the data in Question 4?
Explain your answer.
Question 6. Assuming that there was no upper specification limit of 14 gm,
where would you draw a single precontrol line if the target (or desired) bond
strength was at 11.0 gm?

Here are the answers to our six questions:

Question 1. Precontrol lines are at 8 and 12 gm.
Question 2. Process in control. All five units in the initial sample are within
the precontrol lines (green zone).
Question 3. Sampling frequency should be 12 hours/6 = 2 hours.
Question 4. Action on sample numbers 1 through 7 and 10 through 12:
Continue production. Action on sample numbers 8 and 9: Stop production.
(In actual practice, the process must be adjusted or corrected after sample
number 8, and then five more units must be drawn and determined to be
within the green zone before the sampling of two units is resumed. The same
holds for sample 9.)
Question 5. There are three nonrandom trends in the data associated with
Question 4: (1) Bond strengths are getting lower and lower until a correction is
made after sample number 8. Probable cause is bond contamination or loss of
bond energy. In any case, the trend needs to be investigated, quite apart from
whether production continues or stops. (2) The second reading in each sample
is almost always lower than the first. The probably cause may be in the
measuring instrument or in the fixture. (3) Sample number 9 indicates an
overcorrection (pull strengths too high).
Question 6. With the target at 11.0 gm and only a single (lower) specification
limit, the precontrol line should be midway between the target of 11.0 gm and
the lower specification limit of 6.0 gm, i.e., at 8.05 gm.