The following content is
provided under a Creative Commons license. Your support will help
MIT OpenCourseWare continue to offer high quality
educational resources for free. To make a donation or to
view additional materials from hundreds of MIT courses,
visit MIT OpenCourseWare at ocw.mit.edu. PROFESSOR: OK, so we're going
to talk about Six Sigma. Now, if you've seen,
the course is heavily weighted towards Lean with
a little bit of Six Sigma. And we think that's the
appropriate place to start. But we need to get you
started on Six Sigma. And so learning
objectives here are that Six Sigma is a valuable
approach for improving process quality. OK? And we'll try to explain a
little bit more about that. We're going to have you
be able to interpret a basic statistical
process control chart. That's what the active
learning experiment's about. We're going to talk about the
difference between process control limits and
specified control limits. And you'll be able to describe
what a capable process is. OK, so Six Sigma is a strategy
to improve process quality by identifying and
eliminating defects, and defects in the broadest
possible sense, not just defects in the size of
a hole being drilled, but defect in something that you
want to deliver your customers, not what they want. That's a defect. OK? It's a very
data-driven approach. So if you've seen
Lean-- in fact, one of the comments we got
back on the feedback sheet is it's kind of a lot
of touchy-feely stuff you're covering. And Lean is, in some ways,
a lot of touchy-feely stuff, because of the
importance of people. Six Sigma, if you
read the literature on Six Sigma, at least
the literature I've read, people are never mentioned. It's about a
problem-solving process. And that's important, but
it has its limitations. And the important thing
is to know when to use it and how it fits in with a
more holistic Lean approach. And it's a very structured
implementation approach with certified experts. You probably have heard of
black belts and green belts, and things like that, OK. So Six Sigma has this
hierarchy of experts. And that's really not
the mindset of Lean. The mindset of Lean is the
experts, the coach, not the person who knows the most. It's the person who gets the
most out of the people there. So there's some big
cultural differences between Six Sigma and Lean. But they've come together,
and the important thing is to try to understand that. Now, where does the
word sigma come from? Well, this is a pretty
analytical group, so you probably all know this. But a normal distribution,
something like what you just saw with the M&M's-- although
that wasn't quite normal, but it was somewhat
close to that-- is defined by the standard
deviation, which is sigma. And in this chart-- so this
is one standard deviation from the mean, two standard
deviations, and three. And Six Sigma would be
six standard deviations. And just keep this in mind. We're going to deal with some
Three Sigma in a little bit, coming up here. Three Sigma is
99.73% of the area under a normal distribution,
would be within a Three Sigma band. OK. An important part of
Six Sigma is defects. Because the goal of Six Sigma is
to reduce the number of defects per million opportunities. And a defect is defined as any
process output that does not meet the customer
specifications, as I mentioned earlier,
in a very broad sense. OK? And so you've got
two things here. You've got opportunities
and defects. So you want to
reduce the defects, but you also, in
many ways, you want to reduce the number of
opportunities to have defects. OK, so reducing the
number of hand-offs-- Sue mentioned the baton. Hand-offs between
people are always a potential source of mistakes
or errors or misunderstandings. Just think about how many
miscommunications have you had with your boyfriend
or girlfriend or parents or best friends. So hand-offs of any
kind are something that you want to minimize. I just read some
literature I'm going to share this with you, since
we're in the aviation field-- I'm in the aviation field. The data from last year's
commercial air transport travel in the US was there was 0.02
fatalities for every million boardings, 0.02. I think that's a nine sigma. Yeah, OK. So that gives you kind of
a benchmark of something being really good. But how good is good? Some people say, you
know, 99% is good. I mean, If my car
starts 99% of the time, and it's an old
car, that's good. But that's three
days out of the year, I'm not going to get
out of my garage, and I'm going to miss
getting to my appointment. OK, that's not so good. So 99% you might think is good. But that's 20,000 lost
article of mail per hour; unsafe drinking water
for 15 minutes a day; in this audience, 5,000
incorrect surgical operations per week; two short
or long landings at a major airport each
day; 200,000 wrong drug prescriptions a year. That's not really
very good, 99%. This is much better, Six Sigma. This is what you're looking for. OK? 1.7 incorrect operations per
week-- even that's too much. But boy, that's a lot
better than 5,000. If you're one of
those 1.7 people, you don't care what sigma it is. You just got-- [LAUGHTER] Same here. And as I mentioned,
airline travel is nine sigma for safety. OK, so a basic tool
in Six Sigma is called the statistical
process control chart. So a chart shows--
so it's a sequence. This is a time chart. And we're going to
build one just shortly. It's time, and this
is some measurement. And this shows how stable
it is, how much it's tending to be around
the center of the chart. It's a very common chart. And in both fields, if you go
into a production facility, an engineering, production
facility, now the operator will be showing you
their control chart. I couldn't believe, when I first
went in and saw the machinist proudly or the machine operators
showing me their control chart. I see control charts
all through health care. So let's make a control chart. OK, so we're going to
do a little experiment, a little thing here. We're going to put it in
the framework of a pharmacy. For those of you who are not in
the framework of health care, you could think of
another application where you're getting something. But our pharmacy here dispenses
a medicine called a White Bean Medicine. And we get it in bulk from
two different suppliers. This is our brand
name supplier, Goya. And it appears to
come from Spain. It's a little hard to tell what
the original source is here. But at the end of it,
there's some Spanish, and so I presume it
comes from Spain. And this is our
generic supplier, [? Shaw's. ?] So we get
these two suppliers. And if you read the back
of the [? Shaw's-- ?] well, you can read the back of the
label, if you need to, later. So our pharmacy gets it
and puts it in these bins. And then they've decided to
measure dosages by volume. Now, usually a
dosage is by weight. OK? But it's much easier just
to scoop this up and fill it than to scoop it up and fill
it and weigh it, and so on. So we're doing it by volume. And what we want to do is to
track our pharmacy dispensary, to be sure that we're
getting consistent weight. And so our pharmacy-- this is not a critical
medicine, but it's widely used in our facility. So we dispense a
lot of it each day. And so what we've
done in our pharmacy is our pharmacy is going to take
three samples of this medicine each day and weigh it. And the next day, it will
take three more samples and weigh it. And we're going to build this
chart to see what it's like. And the pharmacy has
been very cooperative. They've given us 20 samples,
that are already on your table. OK? And then I'm going to
tell you what to do. We're going to enter this data. So we've got a three-cup sample. We're going to enter data
into control charts-- one for the average weight and one
for the range of the weight of the three cups. And we're going to
use it for 20 days to establish our
process capability. So we're going to
measure for 20 days, and then that's
going to tell us, do we have a stable process? And we can monitor after that. And so that's what
we're going to do. So our first phase is now each
of you are going to weigh this. You each have four days. In the center of your
table, you have 12 cups. And the cups, there are
three cups for each day. It says right on
there, this is day 16. There are three day 16 cups. And you have a sheet. And your sheet
says, here's day 16. So you measure cup A, B, C.
You compute the average-- we can get some calculators,
if you need it-- the max and the min
and then the range, which is the max minus the min. And then you bring that
up here to data central. And we'll record it. And you have a scale. And you want to use grams. This is in pounds and kilograms. You want to use the kilograms
scale to get it in grams. And don't use this now. But when we're all finished,
we'll dump the beans in there. But don't do it
now, because we may need to go back and do
some second measurements. OK, so please now start at your
tables measuring these samples, filling out the chart,
and bringing it up here. [INTERPOSING VOICES] STUDENT: 98. STUDENT: Oh, my goodness. STUDENT: Excellent. Very [? well done. ?] PROFESSOR: OK. So we have all the data. And my helper over here
is going to enter it day by day, so it unfolds like
you would see a normal control chart, if you're
watching it day by day. So he's now on day five. Now, what we have-- this
is the mean average. This is the average,
and this is the range. And this is the
mean of the average, and this is the
mean of the range. OK? And then what we have
here are these red lines are the Three Sigma lines. Remember, on a
normal distribution, I said Three Sigma. So the way these control charts
work is you get this data. And it's been established
that the processes-- you look at a three
sigma deviation. And as long as you're within
that three sigma deviation, your process is under control. OK, that's just sort
of what's evolved. And of course,
these red lines will change as he enters
the data, because he's getting more and more samples. So this is like watching
the election returns on election eve. [LAUGHTER] STUDENT: We're all
[? down here. ?] PROFESSOR: OK. STUDENT: [INAUDIBLE] PROFESSOR: So
we're up to day 10. STUDENT: [INAUDIBLE] PROFESSOR: OK, and what's
going on in the background-- one of our colleagues
at Cal Poly put this spreadsheet together. So this shows the [? app, ?]
the mean, what's called x bar. This is x bar,
and this is r bar. X is the average,
and r is the range. And then we've got the upper
and lower control limits. These are called upper control
limits and lower control limits. We've got the upper and lower
control limits for the average and for the range. And usually, the way this
field works is 20 samples is considered a
good enough baseline to establish the
process capability. So oh. STUDENT: Oh-oh. PROFESSOR: [INAUDIBLE] STUDENT: I'm
double-checking the data. PROFESSOR: Can you
double-check that? STUDENT: Yeah, I'm
double-checking. [INAUDIBLE] what's on the sheet? We may have to go [INAUDIBLE]. PROFESSOR: OK, well,
let's finish it off. Let's keep going. STUDENT: OK, so that's
it for 20 data points. PROFESSOR: OK, so something's
obviously funny with day 15. So when you have some
funny data, what do you do? [INTERPOSING VOICES] PROFESSOR: Go to the
[? gemba. ?] So who has 15? OK, so let's go
take a look at 15. What's going on here? STUDENT: [INAUDIBLE]
Which sample is that one? PROFESSOR: Can you
weigh it again? STUDENT: It's cup C. PROFESSOR: It doesn't
look quite full, does it? Why don't you put some
more beans in there. [LAUGHTER] Just fill it up kind of
like the rest of these. [INTERPOSING VOICES] PROFESSOR: What's now? STUDENT: 73. PROFESSOR: OK, so
now, on day 15, something happened
in the pharmacy. And all three cops are
shy by 1/4 inch, 1/8 inch, something like that. So what we would probably do is
go back to the pharmacy records and see what happened. And we might find
out, for example, that some employee
had been ill that day, and they'd hired a temp. Or a fill-in wasn't
cross-trained properly, and they thought this was
a full cup, a full dosage. So we can eliminate. Let's delete that point. We know that that point is no
good, because for certain-- we can see it. We can verify it's no good. So-- STUDENT: What happens
if I [? blank it? ?] PROFESSOR: OK, that's fine. STUDENT: Going to zero? PROFESSOR: OK. STUDENT: Is it
going to go to zero? PROFESSOR: No, that's fine. STUDENT: [INAUDIBLE] PROFESSOR: No, no, that's fine. It's fine. Yeah. Thank you, sir. STUDENT: All right. PROFESSOR: Yep. OK, so now we've got our
process capability established. We've got 20 days
minus the 1 bogus day. And our upper and lower
control limits are-- we know now, for our
process, it's 75 and 69. And we got them for the range. OK? So good. Thank you. So that's how we would
build a control chart. So now we're going to do
a little bit more slides. And then we're
going to come back to see how we use
this control chart. So I'm going to change now
from the basic control-charting to a little bit about the basic
Six Sigma method, called DMAIC. And DMAIC is another
Deming cycle. It's like plan, do, study, act,
except it's slightly different. It's called Define, Measure,
Analyze, Improve, and Control. And so DMAIC is the
easy thing to remember. So if you're doing a
Six Sigma intervention, like in our pharmacy,
we would have to define, who are the customers? And what are their requirements? So in the case of the
pharmacy, the customers would be whoever has
ordered the medicine, and they have some requirements. And then we want to know what
the key characteristics are important to the customer. Well, the customer
for medication is really interested in weight. They don't care too
much what it looks like. They really don't
care how it's packed in the cup, things like that. But they're really
interested in weight. And of course,
we're using volume, because it's a little
quicker, but they're interested in weight. So OK, then you measure. So once we know the key
characteristics, then what are the key input and
output characteristics? And you want to verify
the measurement system. So here are output
characteristics, is the weight of the cup. That's what we're measuring. Later on, we might
get more sophisticated and measure the input
characteristics. Like, we might weigh
these two samples, or something like that. But right now,
we're just focused on the output characteristics. And we collect the
data and establish the baseline performance. That's what we've done. OK, and then we've
done that so far. Then we're going to
look at the raw data. We convert the raw
data information and get insights
into the process. Well, we already found
out we had something wrong with poor Molly. On day 15, she got the flu. And we had a good temp, but
we didn't train the temp. So we've learned now that
training is important so we don't dispense
the wrong medication. OK, so we've improved. We've developed a solution. And now we want to put our
process under process control. I [? mean, ?] under
process control, we're going to now
monitor day by day the medication dispensing and
see if it stays within control. That's DMAIC. So here's a simple example. You have a process. So you have a defined process
with an input and an output. You measure the
weight with a scale. You analyze the data
with the control chart. Then you want to
improve the process. We just talked about
training is one improvement. And then you put it
under process control. So that's the basic
Six Sigma cycle. This is one area where I said,
in Lean, when you came out of this course,
you could probably pick up any book in Lean,
and you could go for it. If you pick up Six Sigma
books, they do a deep dive into a lot of math. So it's a lot of
statistical mathematics. So you may need more than
what we've taught you, because we're not
focusing on math. OK. It's easy to see in a process
control application like this. It may get more complicated
if you're looking at something that's not quite so visual. But anyway, that's
the way it is. OK, now there are certain kind
of variation in the process. And actually, our control chart
was a good example of that. There's what's called
Common Cause Variation. And this is just the
sort of randomness in the system, the things
that affect all the samples. And then you have
Special Cause Variation. And we just had an
example of that. Common Cause was the
different variations of 19 of the 20 days. And then we had one case which
was a Special Cause Variation. We sort of went out of bounds. In that, we could go in, and we
could intervene fairly quickly and correct. To get our Comma
Cause Variation down, we're going to have to do
something different here. We're going to have to be much
more careful about how we fill the cups, and stuff like that. We might have to have
more careful procedures. But the Special
Cause Variation, we want to drive out of the system. And the control
charts give you a way to quickly see the
differences between these two. So let's look at an example. This is actual data
for patient falls. I've forgotten the facility. Anyway, the reference is just
off the back of the bottom of the screen here. But this is patient falls. And we've heard sue mention this
morning how patient falls are-- you had it on a slide. It was something about there
being a source of waste, or something. STUDENT: [? Patient ?] falls are
considered [? to have ?] never have been something [INAUDIBLE]. PROFESSOR: [? They've ?]
never [INAUDIBLE].. Great-- never [? met, ?] OK. And here we've got, in this
facility, this is by month. And we got an average of about
six patient falls a month. And they've been charting it. And we have an upper
control limit of about eight and a lower control
limit of about four. Now, how many patient
falls would you want? STUDENT: [INAUDIBLE] PROFESSOR: None. OK, that's a
customer-specified limit. There's a difference
between your process limit. This is your as is process. In this case, your
process is not very capable in the
eyes of the customer. So we're going to get into
this towards the back end of the module about the
difference between the process capability, the upper
and lower control limits, and the upper and
lower specified limits. So I just introduce
it here as a teaser so you don't fall asleep. OK, so now, in this facility,
they tracked it for 20 days. And then they kept tracking it. And then something
starts going kind of bad here towards the end of
the around 28th month. We've got something
that's kind of getting up to the control limits. And then we've got
something that's going outside the
control limits. And this is when the red
flags start saying, OK, we're outside the normal bounds
we consider acceptable. We better go investigate
what's happening here. And that's what we did
with our pharmacy thing. So now what we're going to
do is continue our exercise. And the first thing I'd like
you to do is take the cups. And there's a
white tray in here. And dump all the beans
that you have into there. And just nest the
cups back together. And then the pharmacy's going to
bring you a new batch of cups. So let's get these
out of the way first. OK, so while we're
getting some help here, what's going to happen is, each
of the stations on their easel has a chart for the
average and the range. And what I want you to do is
to come put a scale on this where you're going to put the-- OK, the upper control
limits, we don't have to be accurate to the
second decimal place here. It's about 75.4 and 69.5. So we want to do
something like this. First of all,
let's get the mean. I'll just do one, and I'll let
the other tables do theirs. So this is 72.5. That's the mean. And then the upper control
limit is about 75.4. So we'll just put that here. And the lower control
limit is 69.5. OK, so you're going to
want to finish that. This is for the average,
and this is for the range. And I'll let each
team finish their own. And you can come up here and
get the data if you need to. And then-- STUDENT: [INAUDIBLE] PROFESSOR: So we want
to draw the control limits on our chart. That's now fixed. That's our process capability. And now we're going
to start moderating it for days 21, 22, 23, 24, but
you're each going to do it. You're going to
do 21, 22, 23, 24. You're going to
do 21, 22, 23, 24. OK? And you're going to measure
three samples each day and plot the data on
your control charts. So whatever the data is
for the average on day 21, you put here in the range. On day 21, you put here. OK? And you're going to look at it. And you're going to see
whether your process stays under control or whether it-- is it doing this, or is
it doing that, or that, or something else? And if it starts
going out of control, once you've decided
it's out of control-- you may have to
watch a little bit-- then it's time to
stop to investigate in what the root cause is. Just like we did here,
we went to the table and found out the root
cause was it wasn't filled. But we want you to do
this in a structured way. OK? The tendency you're
all going to have is to say that I know
what the root cause is. Somebody didn't fill it. But maybe or maybe not. So we're going to ask you to use
a fishbone diagram that Analisa just told you about. And that's on the second page. And start thinking
about all the reasons that the behavior you're
observing might be happening. For instance, could
it be that Aubrey, who was thinking about something
else, didn't weigh it right? Maybe it's a measurement thing,
so that would be a personnel. Or maybe you got a
crummy weighing scale. That's a machine. So start filling out
all the things you can. And then, when you
get those filled out, then start looking at
them systematically. Which one do you
think is most likely? OK? So it's a detective operation. So now you're going to actually
do a root cause analysis. OK? OK. Does anybody have any
questions about-- yeah? STUDENT: So the
upper and lower limit and the average that we're
using are from our first sample? PROFESSOR: We've established
our baseline, yeah. So that's not going to
change [? anymore. ?] STUDENT: Do we all
use the same one? PROFESSOR: You all
use the same one. Would it help if I write
down those numbers? STUDENT: Yes, especially
there on the left [? in the screen. ?] PROFESSOR: Yeah. Can you see them? STUDENT: Yeah, there it is. STUDENT: 75.4 and 69.5. PROFESSOR: Yeah. Oh, you've got good eyes. OK, it's nice to be
with young people. [INTERPOSING VOICES] And then now, you just
operate on your own. You're all independent. Think of yourself as the
independent measurement quality control group of a pharmacy. OK? STUDENT: [INAUDIBLE] for A. STUDENT: 0 to 7 and 1/2. STUDENT: 74. [INAUDIBLE] There's
a rock in mine. [? Turn ?] [? it ?] around. STUDENT: Yeah, [INAUDIBLE]
[? wrong ?] with it. [INAUDIBLE] STUDENT: Average is
going to be [INAUDIBLE].. [INTERPOSING VOICES] STUDENT: So it will
be [INAUDIBLE] 16. STUDENT: [? RG ?] is-- OK. STUDENT: [? And ?]
are the machines that you use for
calculating [INAUDIBLE] [INTERPOSING VOICES] STUDENT: So let's talk about how
those might have been issues. STUDENT: OK. [INAUDIBLE] materials. STUDENT: The cup itself? STUDENT: Yes, sir. [INTERPOSING VOICES] STUDENT: You want
to write those down? STUDENT: Quality of [? scale? ?] STUDENT: Yeah, quality of
[? scale ?] [INAUDIBLE] machine [INAUDIBLE]. [INTERPOSING VOICES] STUDENT: It could be not leaving
it long enough on the scale. STUDENT: Well, I just
sampled three different ones. [LAUGHTER] STUDENT: [INAUDIBLE] STUDENT: So bean variations. [INTERPOSING VOICES] STUDENT: I think
it's pretty full. PROFESSOR: OK, well let's see. STUDENT: [? If ?]
we dump it out? STUDENT: Well, I'll try dumping
it out and see [INAUDIBLE].. [INTERPOSING VOICES] PROFESSOR: [INAUDIBLE]
[? It's ?] [INAUDIBLE].. STUDENT: Why are
they [INAUDIBLE].. STUDENT: Oh. Look at that. PROFESSOR: OK, so
going around the room, we had a range of a root cause
analysis things, ranging from-- this table just said, we
think this is the cause. And they tested it,
and that was the cause. So they just jumped
right to a conclusion. We had a very structured
process over here. But this is a way
to make it visible. If you go into a facility
that's under statistical process control, the employees
are trained on this. And they're trained to recognize
this kind of deviation. And then they do
something about it. OK, here's a chart for resident
falls in a long-term care facility, control chart. And here's their 20 days. And this was their control. Now, in this case,
they were intervening to try to reduce patient falls. So this is not now
just doing nothing. They're doing a bunch
of interventions. And these are different
interventions. Actually, what
they did here was-- it's not on the
chart, but what they did was they just
simply identified the patients in this
long-term care facility who were susceptible to falling. And they put stickers on their
wheelchairs or their walkers in their room and
on their charts. And they picked stickers
that were so interesting, that the residents
who weren't in danger wanted the stickers, too. But just being conscious of-- those were at-- the at-risk
patients started to drop it. And you can see their
interventions lowered the mean and lowered the control
limits and lowered them again. And this was a process
improvement study that I actually heard
at the last IHI meeting. OK, so now our last topic-- this is Process Capability. Process Capability is defined
as the ability of a process to meet the customer
expectations. And what we have here is we
know what the capability is, but we don't know whether
it meets the customer expectations. And to get the customer
expectations on there, we have to find
out what they want. HUGH MCMANUS: The
customer-determined limit, the spec limits, are essentially
what the customer wants. The upper and lower
values between which a process must be controlled--
that's what the customer wants. The bounds in which we choose
to control the process, or which we try to
control the process, are the upper and lower
control limits of the process. So how do we measure, given
those two definitions, the process capability? One way to do it is to
think about the process. And again, this is
an approximation. But it's a pretty good one,
under a lot of circumstances. The process is having a
normally distributed behavior. So if we collect
some data, we can assume a normal distribution,
do some statistics using our statistical
tools, and decide how good or bad the process is. And if we have a normally
distributed process, its behavior is going
to be characterized by its standard
deviation, sigma. And if we draw our spec
limits around that normal distribution-- here is
our lower spec limit, and here's our
upper spec limit-- we can define a quantity
CP, the process capability, which is essentially this
distance here, the upper limit less this lower spec limit-- so that the range
which is acceptable divided by the Six
Sigma of the process. And if we have, say, a CP
of 2, that's a very tight. Basically, the spec limits are
twice Six Sigma of the process. Then it's a very
tight distribution inside the spec limits. And the chance of something bad
happening are very, very small, the chance of going out. So a CP of 2 is a
very good process. That, in fact-- STUDENT: [INAUDIBLE] Six
Sigma is [INAUDIBLE]---- HUGH MCMANUS: Six Sigma-- STUDENT: --standard
deviations [INAUDIBLE].. HUGH MCMANUS: That's right--
six standard deviations on either side of
the mean, right. So a CP of 2 would be
a Six Sigma process. In that sense, it would
be a very good one. And just graphically,
you can just see there's basically no tail
sticking out of the bad zone. CP of 1 means you've got
three sigma on each side. Three sigma, you've got 97 point
whatever percent of the process falling in that range. So the chance of something
falling outside the range is quite small. But it's not 0. We can see a little
bit of ink here outside of the acceptable range. And once we get below 1,
it starts getting ugly. We really don't want to
even talk about that-- so from a statistical process
control point of view. Another issue which
one can ignore is the fact that we
may not be centered. And we can actually redefine
our process capability using a metric called CPK,
which takes into a fact that we may drift off mean. And it's defined this way. It's basically the distance to
the closer boundary from where we are divided by three sigma. And it's whichever
one of those is worse. And those same curves
shifted over 1 and 1/2 sigma look a lot less pretty. Our former Six Sigma process
now has a CPK at 1.5. But it's still OK. There's a tiny, tiny--
there's a couple, like two pixels worth of
ink sticking out there. The chance of a defect is
still very, very small. But our formally
OK-looking CP of 1 process now has quite a
bad-looking tail. And our process that
was bad is now horrible, our relatively
uncontrolled process. So that's a different way of
measuring process capability, which takes into account the
fact that the mean may not be on the center. What we do with these two
measures is different. Here we have a archery example. This is a classic statistically
characterize-able process. We have shots that
are both widely dispersed and off-center. So how would we
characterize that? It's basically low on
both metrics, right? It's widely dispersed,
and it's off-center. Here's another archer. It has a tight distribution,
but it's way off-center. STUDENT: So CP is high. HUGH MCMANUS: So CP is
high, and CPK is low. That's right. So that's the difference
between those metrics. On this metric, it looks great. On that metric, it does not. And this, of course,
is what we want. If you're an archer or
any other person who needs to control their
process, what do you do first? Yeah, I'm cheating a
little bit on this one. I do archery sometimes,
not very much. But you have to do
this first, actually. You have to get your
process repeatable, even if the meat is way off. Because if you're
correcting every time, then it just goes
all over the place. But if you just go, OK,
same thing every time, OK. Five times in a row, I've
hit up and to the right. Now I'm going to adjust. So often, you concentrate on
knocking your CP down first, to understand your process,
to see where the mean is. Because you can't
really measure the mean in a situation like-- you can
statistically, but it's hard. OK, enough of that. This actually-- remember
we said we would-- I'd tell you where the
definition of Six Sigma comes from. Interestingly, it
comes from a process with a mean shift
of 1 and 1/2, which would give you three defects
per million opportunities. So if you have a
very good process, but you still don't control
your mean to more than 1 and 1/2 sigma, you still have
a very low defects per million. That's where that
idea comes from. Those are some of
the concepts that we use to get processes under
control and keep them there. And Six Sigma, as
an overall method, is very useful for taking
variation down, especially in critical applications like
large-scale manufacturing or health care, where you want
your defects to be very low, your variation to be very low. We've talked to you about
control charts, which is a great place to start. It's not the end. And we've glossed over
all the statistical tools you need to really
understand the numbers. But the visual is very
powerful in and of itself. Even if you don't
do the statistics, having the visual evidence
of what your process can do and whether it's deviating
is very powerful. And once we understand
what our process can do, we can compare that to what the
customer wants and understand the capabilities of our process
and whether they're acceptable or not.
