SPC Made Easier, Simpler, More Statistically Powerful
SPC Made Easier, Simpler, More Statistically Powerful
SPC Made Easier, Simpler, More Statistically Powerful
Statistically Powerful
Keki R. Bhote
n the search for world class qualiImanagement,
ty, there must be synergy between
customers, suppliers,
and employees. Organization, systems, and training of the right kind
are essential catalysts. A blueprint
of improvement in these general
areas is beyond the scope of this
article. Instead, it concentrates on
easy, simple, cost effective, and statistically powerful tools that can improve product quality, not by an
inconsequential 10 percent, 50
percent, or even 100 percent, but by
factors of 10:1 and 100:1 in a short
period of time. If quality is the centerpiece for a company's resurgence, these statistical tools are the
piece de resistance, within quality.
The Traditional, Ineffective Quality
Tools
It is common knowledge that
Japan drew even with the U.S. in
quality progress by the mid 1960s
and has gone on to command an
ever-widening lead. What is less
well-known are the quality tools
used by the two countries. The U.S.
persevered in the traditional tools of
inspection and sorting, detection
and correction, fire fighting, exhortation, and the crutch of sampling
plans, at least until 1980. These
tools consume costs but add little
value.
Conventional SPC Tools - Too
Little, Too Late
The Japanese, on the other
hand, abandoned these ineffective
tools and adopted statistical process
control (SPC) in the 1950s, under
the coaching of Dr. W. Edwards
Deming. They rode the crest of the
SPC wave until the mid 1970s,
when they realized that SPC in production was too little and too late.
Paradoxically, the U.S. rediscovered
SPC at the same time that the Japanese were discarding it. (SPC was
widely used in the U.S. during World
War II, but then the baby was
thrown out with the bathwater, in a
revolt against the mysteries of the
world of statistics.)
"If Japan Can, Why Can't We"
What re-ushered the SPC age
in America was the airing of the
NBC White Paper-"If Japan Can,
Why Can't We." It gave the American public at-large its first glimpse of
the reasons behind Japan's success: Quality, in general, and SPC
in particular. Deming was rescued
from the U.S. quality wilderness and
elevated to a prophet within his own
country. Major companies scurried
to jump on the SPC bandwagon.
Unfortunately, SPC has become
synonymous with control charts.
Control charts are complex, costly,
and almost useless in their ability to
12
Target
.. ~
~ .................................~~.~~~~~-"'""""""'."""'"'"~".""'."'."".""'.",""""""""
Design of Experiments
Characteristic
Classical
Taguchi
Shainin
Technique
Fractional factorials,
EVOP, etc.
Orthogonal arrays
Multi-vari,
factorials
Effectiveness
Moderate
(20 percent to 200 percent
improvement)
Low to moderate
(20 percent to 100 percent
improvement)
Retrogression possible
Retrogression likely
Extremely powerful
(100 percent to 500 percent
improvement)
No retrogression
High
High
Low
Many experiments
Many
experiments
Few experiments
Moderate
High
Low
Low
Poor
High
No randomization
Cost
Complexity
Statistical Validity
Applicability
Requires hardware
Main use in production
Ease of
Implementation
search, full
Requires hardware
Can be used as early as
prototype and engineering
run stage
Moderate
Difficult
Easy
Fall 1987
vari~ble
13
14
CPK = (1 - K) C p
(Where K is an absolute value, without regard to its sign)
(Design Center, D, need not be at
the mid-point of the specification
width.)
When the process average, x,
and the design center, D, coincide,
K is reduced to zero, so that Cp and
CPK are equal. Fig. 3 shows two examples of non-centered processes.
Process A has a narrow spread,
with a respectable Cp of 2.5. But
because it is skewed toward one
end of the specification Iim~s, ~s CPK
is reduced to 1.0. By contrl;lSl, Process B has a wider spread and a
poorer Cp of 1.67. Yet, it is more
toward the design center and, therefore, has a higher C PK of 1.33. The
objective, again, is to aim for high
Cp.'s of 5, 10, 15, and 20 without
adding recurring costs.
Measuring Variation - The Easy
Way
Traditional process capability
studies require 50 units to be measured on the process, from which
the average, x, and the standard deviation(s) are calculated. The process limits are then conventionally
defined as 3s. (Sometimes,
where production is limited or measurements are difficult, 30 units can
suffice as a statistical minimum.)
From these measurements, Cp and
CPK can be calculated.
However, there is a statistically
sound short-cut to process capability studies. The rule is to measure
five units in a row taken from the
process. If all five units fall within
the middle half of the specification
limits, it can be proved, using the
binomial theorem, that a C PK of 2.00
and over, exists. If even one of the
five units falls outside the middle
half of the specification limits, the
process has excess variation, which
must be reduced using engineering
judgment, or preferably, the design
of experiments.
Target
20
40
20
20
20
20
Cp~25 ~0.8
Cp~-~1.0
20
40
20
20
15
40
20
Cp~- ~1.33
Cp~-~1.66
12
20
40
20
10
40
20
40
20
2.5
FIg. 2. Six frequency distributions comparing the specification width to the process width.
Cp~- ~2.00
Fall 1987
Cp~- ~8.00
15
Variation Reduction
Process A
A Roadmap
Prioritization of
Customer Requirements
10
12
14 15
LSL
(1)
20
USL
I--
Paired Comparisons
Process B
Multi-Attribute
Competitive Analysis
(2)
Multi-Vari Charts
I--
Components Search
Market Research
(3)
Cp=2.5; C =1.0
PK
Despite narrow distribution, poor CR because
.-t
Variables Search
Full Factorials
10 11
LSL
14
X
15
D
17
20
USL
Cp=1.67; Cp .1.33
K
Wider distribution than Process A, but closer
to design center, so acceptable Cp .
K
B vs. C
PATH 1
If mathematical relationships!
formulas between input variables
are known
Positrol
PATH 2
Prototype stage: If only two to
six units are available
PATH 3
At the eng. run, pilot run or full
production, with 30 or more
units available
Process Certification
Control Charts!
Pre-Control
FIll. 4. DOE tools taught by Dorian Shainin are reflected in this illustration.
.-
16
Target
Speclflcatlons- Arbitrary,
Overbellrlng, Wrong
The diagnostic journey begins
with a challenge to engineering
specifications that non-engineers assume are God-given. There are several reasons for poor specifications:
Customers are not consulted
-the difference between marketing and selling
The importance of a product feature to a customer versus the cost
of that feature is not value researched
The engineer's ego to create a
"state of the art" design
The engineer's reliance on prevIous designs, boilerplate requirements, or suppliers' published
specifications
Reliability is seldom a specification
In translating product specifications into component specifications, a reliance on formulas or
guesses, rather than design of experiments.
It is this author's experience that a
good percentage of both quality
problems and design cycle time can
be reduced if quality techniques are
used to determine specifications.
Some of these include: Value Research, Quality Function Deployment (a Japanese tool to translate
the "voice of the customer" into
hard specifications), Multi-Attribute
Analysis and, of course, DOE tools.
Computer Techniques
Route 1 in Fig. 4 uses Monte
Carlo simulation and other computer
methods to determine the effect of
independent variables upon a given
output. However, this is only useful
if the formula governing the respective variables is known and can be
programmed. More often than not,
these mathematical relationships are
not known, especially when second
order and higher order relationships
between the independent variables
can cause totally unexpected results. This is also one of the main
statistical weaknesses of the
Taguchi methods, since the orthogonal array uses fractional factorials,
Fall 1987
Multi-vari Chart
Bam
lOam
9am
11 am
12am
0.2510 f-----.---__+-----+----+-+-~-~--+-----
1\
I- \
~ ~
0.2500
Xl
1-\-l-----J.----\H---+l~~F+_~~~+_-J-V__\l_---l----:\~l__+\.......+~
~
I:
""
i5
...
.~~
0.2490 [--;:;:::=:::;:::::::;----+-t--4----1~----+_----
~Max.
ttr-!Min.
Right
0.24S0
p===t~~~___l_~~~--+~~~__+~~-
... A cause-and-effect
diagram will only indicate the
numerous variables or causes,
without any due of where to
start or what interactions exist
between those variables. It
may also miss the boat
altogether. The real cause may
not even be on the list of
causes drawn on the diagram.
clue of where to start or what interactions may exist between those
variables. It may also miss the boat
altogether. The real cause may not
even be on the list of causes drawn
on the diagram.
17
Temporal
Hour-to-hour; shift-to-shift; day-today; week-to-week or month-tomonth variations.
Multl-Varl Case Study: Rotor
Shaft
A manufacturer, producing cylindrical rotor shafts, with a diameter
requirement of 0.250 in. 0.001 in.,
was experiencing excessive scrap.
A process capability study indicated
a spread of 0.0025 in., against the
requirement of 0.002 in., that is, a
C PK of 0.8. The foreman was ready
to junk the old turret lathe and buy a
new one, for $70,000, that could
hold a tolerance of 0.0008 in., that
is, a C PK of 1.25. The plant manager
directed that a multi-vari study be
conducted before the purchase of
the new lathe, even though the payback period for the new lathe would
be only nine months.
Fig. 5 shows the resu It of a
multi-vari chart. The positional (within each shaft) variations describe
taper changes, from the left side of
the shaft to the right, and out-ofround conditions (maximum diameter and minimum diameter) on each
side of the shaft. The cyclical variations, from one shaft to the next, are
shown by the thin connecting lines
between each shaft. The temporal
variations, from one time period to
the next are also shown. Analysis of
each variation indicated:
18
Target
P-C Line
p-c Line
zone).
Condition
12
14
86%
Action
Continue
Continue
unit in Yellow
3. Two units in Yellow
1
14
7%
Slop
Slop
Red
l<-"7'qYe~lI!Qlowti..l._ Green Zone --of""Y~.I~'O~~..I Red
4. Frequency of sampling:
_
__z_o--,n.;crt'=--~z~on~.~=:::::====~zosn~e
_~~z~o~n.=---__ Divide the time interval between two
30"
LSL= LPL
1.50"
Target Area
1.S<J
30"
stoppages by six.
USL.UPL
Fall 1987
this arti~le. On the other hand, precontrol IS only now coming into the
consciousness of quality practitioners and needs an explanation of at
least its mechanics. Fig. 6 shows
the application of pre-control with
four simple rules to: (1) establish
pre-control lines; (2) determine process capability at the start of production; (3) monitor production on an
on-going basis and (4) determine
the frequency of sampling.
The simplicity of pre-eontrol is
now obvious. Its mechanics can be
taught to managers and blue-collar
workers alike in five minutes. Even
the least sophisticated line or machine operator can use it and more
important, make quick adjustment to
assure defect-free' production on
thousands upon thousands of units.
~hile there is no need for charting
In pre-control, plots of the two unit
samples versus time can be maintained as a record. From such plots,
frequency distributions or Cp 's can
be derived easily. Pre-contr01 can
also be used for tightened specifications (relative to broader customer
~pecification for one-sided specifications); or for attributes (by weighing
I>
19
Control Charts
Complex-calculations of control
Pre-Control
Simple~pre-controlare
width
unclear
3. Mathematical
4. Small production
Useless for production runs below 500 unitssampling of 80-150 units before even trial limits
can be established
5. Re-calibration of
control limits
moved inward
6. Machine adjust-
7. Frequency of
'sampling
Vague, arbitrary
8. Discriminating
2. Use by operators
runs
ments
power
10. Economy
Attribute charts can be converted to precontrol charts by weighting defect modes and
an arbitrany rating scale
runs
and the
2.0-the a and
Author:
Keki A. Bhote is senior corporate consultant, quality and productivity improvement, Motorola, Inc., Schaumburg,
IL. He is the author of Supply Management, a management briefing published
by American Management Association
Membership Publication Division, 1987.
2For those interested in the theory of precontrol, there is a considerable body of literature available in A.S.Q.C. annual confer-
20
Target
'''''''''''''''''''''''''''''''',"'J_}~~.",!,'