Who Needs an Early Warning System?
Warning Systems in our History
Keeping Watch to the North
The Distant Early Warning Line, also known as the DEW Line, was a system of radar
stations in the far northern Arctic region of Canada, with additional stations
along the North Coast and Aleutian Islands of Alaska, in addition to the Faroe
Islands, Greenland, and Iceland. It was set up to detect incoming Soviet bombers
during the Cold War, and provide early warning of a land based invasion. The DEW Line remained in operation until the
late 1980s; The United States and Canada replaced it with the North Warning
System, a program that involved refurbishing many existing DEW sites, and
decommissioning the remainder.[i] The
North Warning System still exists today.
Alert: Defective Product has Landed on
our Shores
To the casual
observer, a factory does not invoke the same dramatic imagery that guarding a
nation against invasion does. That is, of course, unless you happen to be the
person managing that factory.
Does any of this
sound familiar? A sales manager and plant manager are having a chat; let’s
listen in…
Sales Manager: I just
got off the phone with one of our biggest customers, Bob Buysalot, and boy is
he mad. He’s doing that big SRW job in the foothills and ran into trouble with
our product. He tells me that the heights are all over the place; that he has
to shim virtually every course. Now he wants us to pay for the shims and all
his extra labor. This is going to cost us a fortune! Don’t you do QC in your
plant? Don’t your guys check product height?
Plant Manager: Well
of course we do QC in the
plant,
and I’ve got the paperwork to prove it… see, look here, the operator has put a
checkmark and his initials in the “OK”
space on the form – he does these checks every 15 minutes, sometimes
even more often if he has time.
Okay… Let’s stop listening, because odds are these two are going to be at it for awhile.
Giving the
customer in our vignette the benefit of the doubt, let’s say a problem does exist
(the product height is too variable); what could have happened during
manufacturing to cause the problem? When did things go sideways? Who knows?
Truth is we may never discover what went wrong; not even if we bring in the
Crime Scene Investigators (CSI). Why? Well, simply put, there is not enough
evidence to be found… the operator’s paper checklist is too sparse; a checkmark
made with a stubby pencil is not much to go on. But, what if…
What if we had a
way to look back in time thanks to a metaphorical time machine made of data; or
better yet, what if we would have had some type of early warning system keeping
watch over our process, back when the product was made?
I can already
hear the murmuring… You’re saying, “…you are going to tell us about
statistical
process control or SPC… I’ve already heard about SPC and I’ve heard that it
requires a lot of math. My guys don’t do math…” Please, stop murmuring. You
have nothing to fear. Yes, I am going to tell you about SPC; however, I have
discovered a way to do SPC that does not involve learning tricky concepts, or
performing complex calculations. In fact, I am going to show you how to get the
benefits of SPC without having to do arithmetic.
But first, more
history: Dr. Walter Shewhart developed the basic principles of statistical
process control, or SPC, back in the 1920s. In the buildup to WWII, SPC proved
so important to the rapid expansion of manufacturing capabilities that the War
Department deemed it top secret. In fact, SPC made the notion of
interchangeable parts – from different manufacturers no less – a reality. Sadly
though, the use of SPC in American industry declined after the Second World
War, while Japanese companies readily accepted and implemented SPC under the
tutelage of Dr. W. Edward Deming, a former colleague of Shewhart. The success
of Japanese companies in the 1970s and 1980s is at least partially a result of
the widespread use of SPC. Statistical process control experienced a
renaissance in American industry in the 1990s, which has contributed to a
significant improvement in the country’s competitive position. There is no
question today that the use of SPC is an indispensable tool in world-class
manufacturing operations.
I hear more
murmuring… you’re saying “…I have a complete set of block height gauges, so I
am good to go… right?” Well, yes and no. These tools fit the classification of
attribute, or Go NoGo gauges. A Go NoGo gauge is a measuring tool that does not
return a size in the conventional
sense, but instead returns a state.
The state is either acceptable (the
part is within tolerance and may be used) or it is unacceptable (and must be
rejected). They are well suited for use in the production area of the factory
as they require little skill or interpretation to use effectively and have few,
if any, moving parts to be damaged in the often hostile production environment.
So far, so good… but here’s the problem: Attribute gauges cannot help the user
track how the characteristic (say, the height) is trending, relative to either
the upper or lower tolerance limit. So? Turns out, this can be a big deal…
Manufacturing
processes are inherently variable, and as a result, the odds of two consecutive
items coming from the product stream being identical are very, very, very
small. Let’s say your machine operator is checking product height every five
minutes with the block height gauge. Every unit that he checks is “in spec”
according to the marks on the gauge. SOP probably dictates that he keeps running.
No problem, yes? Not so fast. What if all those “in spec” measurements happen
to be bumping along close to one of the tolerance limits? This could be a problem
and a big one at that… Remember this: variation happens! While it may very well
be true that all of the units checked were “in spec” there is a good chance
that the units NOT checked will be out of spec – things like that just happen.
In short, because variation happens, SPC, and more specifically, control
charting deserves a second look.
Since you have
heard of SPC, then you probably know about “x-bar & r charts,” “c-charts,”
“p-charts,” “u-charts,” etc., etc. Please, put all that aside for a while. When
making CMU, SRW, or pavers, we should not use those charts – they are not the
right choice for controlling a multi-cavity molding process. Multi-cavity
molding is often referred to as a “family process.” This type of process poses
some special problems in SPC, as opposed to a continuous process like
extrusion.
A family process
consists of several statistically independent processes that are affected by
common factors; these processes are also referred to as “multi-stream”
processes. An example of a family process is injection molding using a
multi-cavity mold. The filling and cooling in some cavities may be affected by
factors not acting on the other cavities. If an operator samples live parts
from a thirty-two cavity mold, the probability of a cavity not being included
in the sample could be as high as 83.3%. If the samples are taken on an hourly
basis, production may continue for a full shift or even a day without sampling
one of the cavities. Thus, nonconforming parts may go undetected for a
substantial period.When a faulty cavity is part of a sample that leads to an out-of-control point (plotted on one of the more traditional control charts mentioned above), a common tendency is to adjust factors that affect all the cavities – these are the “global” factors. If the faulty cavity is not included in the next sample, the process may seem to have been properly adjusted. In actuality, the operator very well may have erroneously changed a process that was in statistical control.[ii]
Finding a Solution…
The dilemma posed by family
processes required an innovative solution. In their book, Statistical Quality Control, Grant and Leavenworth proposed
manually combining the results of median charts with individual measurements
charts. Individual measurements track local variations, while median charts
monitor global variation. Median charts require no calculations by the
operator, eliminating still another potential source for errors. Range
variations are inherently visible in the plot of each sample.
In a median chart, each sample or
observation consists of a unit from each member of the family. Sampling
frequency is process-dependent; initially, sampling should be frequent enough
to profile the process. Ongoing sampling frequency depends on the rate of
variation. Stable processes need to be sampled less frequently.
A median is a measurement of
central tendency similar to the mean or average. The median of a sample is that
point which divides the values of
the individual
measurement in half. 50% of the individual values are greater than the median
and 50% are less than the median.
Median/Individual
Charts for family processes are exceptionally easy to prepare and interpret. Analysis
is made much more efficient by differentiation between and identification of
global and local causes of variation; nonconforming product is detected much
faster and more reliably. Overall process variation can be reduced by centering
the individuals. Operator error, resulting from either observation or
calculation, is minimized or eliminated.
Interpretation
of a Median/Individual Measurement chart is simple. The median is represented
by a moving line, generally around the chart centerline (see the figure,
below). Individuals are represented by points vertically aligned according to
sample
Individuals are shown as dots in columns that
represent an entire sample subgroup; in the illustration, the sample subgroup
consisted of 20 measurements taken at the locations indicated by the
yellow-colored squares (on the illustration below).
Again, tolerance
limits help with interpretation. Individuals that fall outside these limits may
indicate the presence of local causes of variation. For a well-controlled
process, the dots should tend to cluster around the centerline (in this case,
the nominal specification of 8 inches, i.e. the green line)
As an aside… control charts have “control limits” – based on
statistical calculations – instead of
tolerances, and it is possible to calculate control limits
for M/I charts. But in the interest of keeping things simple, I’ll save the
topic, “control limits,” for another time…
Imagine for a
moment the benefit of using an M/I chart… As I see it, a machine operator (or
anyone else for that matter) can tell at a glance how the output of the process
is trending. Such a look at things could provide early warning of a developing
problem.
While knowing
that a problem is in the offing is important, so too is knowing that all is
well. A common problem that plagues manufacturing is a tendency for operators
to over-adjust their processes. This is no small problem. An over-adjusted
process becomes unpredictable and can spin out of control. When this happens,
waste ensues. Statistical techniques give operators the long view of things, a
perspective that tends to tamp down the propensity to make adjustments when
none are really
needed.
Getting Quality
Right – Right at the Source
Quality happens
in a relatively small space; it happens in the space bounded by operator,
his/her process, the methods for operating the process, and raw materials.
Intuitively, the leaders in our industry know this but repeatedly many opt to
focus on end-of-line inspection in a misguided attempt to control quality. This
is a problem because end of line inspection does not control anything. Think
about it… Whom do we typically place at the “QC” station? Moreover, how much
time has elapsed between making the product and inspecting it? If the “QC
person” does detect an error can he make any upstream adjustments to keep the
errant condition from getting worse?
Statistical
Process Control, in general, and the Median/Individual Measurement chart
specifically for multi-cavity molding, is a tangible way to make Quality at the
Source, a tenet of Lean, a reality in your operation. Quality at the source is
an organization-wide effort to improve the quality of a firm’s product by
having employees act as their own quality inspector, who never pass defective
units to the next stage.
Imagine… detecting an adverse trend in a
manufacturing process and taking action to correct the underlying cause before
making defective product… What’s that worth?
[i]
Distant Early Warning Line, Wikimedia,
Retrieved August 30, 2011
[ii]
Rauwendaal, Chris (2008), SPC Statistical
Process Control in Injection Molding and Extrusion – 2nd Edition. Hanser
Publishers, Munich ISBN 978-446-40785-5
