Statistical Process Control

By :
Mahender Kumar
1
 Basic Introduction
Statistical Process Control (SPC)
• SPC is a statistical technique which compares current process output
against a historical template.

• The template is usually based around a centre value with ± 3 standard

measures of variation (of the characteristic being measured).

• The objective
j is to keepp the current p
process stable,, within these ± 3
control limits and to limit variation within them.

• Industry
Industry-standard
standard rules are adopted: If data from the current process
approaches these control limits or forms non-random patterns within
the limits, then appropriate action to control the process is taken.

• No reference to specifications are made in SPC

• SPC was developed by Walter Shewhart and others

others, and many
traditionalists refer to SPC charts as Shewhart charts 2
 Basic Introduction
Statistical Process Control (SPC)

• Statistical Process Control,

Control or SPC for short
short, has been
around since the 1920s although it didn't really gain
widespread
p use in industryy until the 1980s. Manyy
people are immediately turned off of SPC just because
it has "statistical" in its name. However, by simply
understanding a few basic concepts of variation (why
things are not ALWAYS made exactly the same) you
will
ill be
b able
bl to
t leverage
l the
th conceptst off SPC to
t monitor
it
and control your manufacturing processes.

3
 Variation: The Heart of SPC

• Variation exists in everything

• When we’re manufacturing products with customers
that expect and demand high quality and consistency in
our goods, variation can become a big problem.
• Too much variation leads to rework,
rework scrap,
scrap or customer
problems.
• A perfect process would be one with no variation. They
d ’t exist.
don’t it
• As the variation in our processes is reduced, the output
of our p
processes will be improved.
p
• That’s our goal with SPC-to reduce the variation in our
processes and then monitor the process to make sure the
variation doesn
doesn’tt increase.
increase
4
Measures of variations

5
 FFEXERCISE
EXERCISE

IMAGINE FOR ONE BRIEF MOMENT THAT EACH OF THE ONE HUNDRED
AND FORTY-ONE
O O WORDS
O S OF
O THIS S PARAGRAPH
G IS
S A SEPARATE
S
COMPONENT FORM A FIRST SHIFT RUN OF FOURTEEN-INCH
FLYWHEELS. YOU ARE ONE OF FIVE INSPECTORS PERFORMING THE
FINAL INSPECTION OF THSES FINSISHED COMPONENTS WHICH WERE
PRODUCED ON FOUR FAIRLY SMALL DIAL INDEX MACHINES THAT ARE
NOT BEING CONTROLLED BY THE USE OF STATISTICAL TECHNIQUES.
AS CAN BE EXPECTED FROM AN OPERATION OF THIS NATURE, THERE
ARE A NUMBER OF DEFECTIVES COMPONENTS BEING MADE. EACH
WORD THAT CONTAINS AN F REPRESENTS A DEFECTIVE COMPONENT.
HOW MANY OF THE DEFECTIVES ARE YOU ABLE TO FIND? CHECK AGAIN
AND INSPECT FOR THE PRESENTS OF F'S
F'S. WRITE YOUR FINAL COUNT
IN THE BOTTOM LEFT HAND CORNER OF THIS PAGE. THIS EXAMPLE
SHOULD GIVE YOU A FAIR IDEA OF HOW RELIABLE 100% INSPECTION
CAN BE
BE.
6
 INSPECTION
D
Draw sample
l

YE N
Meets spec. ?
S O

ACCEPT REJECT

How good was it? How bad was it?

zBarely meet spec? zJust outside spec?

zmiddle of spec? zWay out of spec?

zSame
S as before?
b f ? zSame
S as before?
b f ?

7
 INSPECTION

Lower Spec.

A
B

What's the difference between ball A and B?

Why
y is the spec
p there and not somewhere else?
What is the purpose of the spec? 8
 Great!!!
II'm
m in spec.

SPECIFICATION

9
 Hey!!!!!
But I'm in spec.
p

SPECIFICATION

10
 TARGET
Every specification has a TARGET
TARGET.
The upper and lower specification
is meant to serve as a
guide line. What you
really want is the stuff
that hits the TARGET.

11
LEANRING 1
Meeting
g specification
p
is not enough

we need a way to
communicate more.
What
a ???

12
 Average
Income

Country X Country Y

10,000
0,000 Rs/Month
s/ o t 11000
000 Rs/Month
s/ o t

Which country is ECONOMICALLY more stable ???

13
Country X Country Y

8000 46000
12000 3000
10000 1000
9000 3000
11000 2000

Avg.
g 10000 11000
Std dev. 1414 17516
14
 LEANRING 2
Meeting specification is not enough
we need a way to communicate

zHow close to target

zHow spread out the results were

15
 Wh t is
What i SPC?

„ Statistical Process Control

„ A monitoring
it i ttooll th
thatt let's
l t' us know
k when
h
a process is changing before products
b
become unacceptable
t bl
„ It is a prevention tool
– Inspection = defect detection
– SPC =detect process change
defect prevention

16
 WHY SPC?

„ IInspection
ti does
d nott assure quality
lit
„ inspection is too late, its after the fact

„ need to detect process change before

defectives are produced

„ Meeting specification does not go far

enough
g

17
 SPC how
SPC, h does
d it work
k

Quantitify the Mue and the Sigma of a

process and detects change from the
standard deviation by calculating the
control limit by estimating the R bar
over d2 tot estimate
ti t the
th inherent
i h t
variation of a process.

18
 DISCUSSION ON VARIABILITY

First order

lower Upper
spec.
size spec.

19
 Second order

lower Upper
spec.
size spec.

20
 After 6 orders

lower Upper
spec.
size spec.

21
 After 12 orders

lower Upper
spec.
size spec.

22
lower Upper
spec.
size spec.

Over the long run a pattern begins to develop. Notice

there is a large cluster in the middle. As further from
the middle you go, there are less and less
23
 lower Upper
spec.
p size spec.
p

If the source of the material is stable, over a long

time period,
period a bell like shaped ccurve
r e will
ill emerge
from the inspection. The Bell shape curve is also
commonly referred to as the Normal distribution
24
What is HITOGRAM?
Why we need it to understand?
What is this BELL shape and normal distribution?

Plot HISTOGRAM for following DATA

Data
D t
4.2 12.4 6
5.2 14.8 7.8
5.4 18 11
2.1 17 11.8
9 19 94
9.4
9.6 15.5 10.8
13 2 10
14 5 11
15 7 10.1
96
9.6 10 1
10.1 88
8.8
25
 Process vs.
vs control limits

Distribution of averages

Control limits
Specification limits

Variance of averages < variance of individual items

Distribution of individuals

Process limits

26
 LEANRING 3
CHARACTERISTICS OF A
NORMAL DISTRIBUTION
LOCATION
SPREAD:
The dispersion it is
usually expressed as
LOCATION:
SIGMA
The central tendency
it is usually
expressed as the
AVERAGE

27
SPREAD
 Distribution Patterns

Saw toothh N ti l Skewed

Negatively Sk d Positively Skewed

Sharp Drop Twin Peak Bell Shape

28
Average different
p
Spread same

29
A B
 Average same
p
Spread different

A 30
B
Average different
Spread
p different

31
A B
 LEANRING 4
SIGMA -measure
measure of spread

gma
sig

32
 +/-3 sigma

+/- 2 sigma

+/- 1 sigma

2% 2% 33
14% 32% 32% 14%
 +/-3 sigma
99.73%
+/-2 sigma
96.45%
+/-1 sigma
64.25%

2% 2% 34
14% 32% 32% 14%
 LEANRING 5
99.73%
+/-3 sigma

95.45%
+/- 2 sigma

68.26%
+/- 1 sigma
g

2.14% 13.6% 34.13% 34.13% 13.6% 2.14%

35
 LEANRING 6
Properties of a normal model curve :-
•It
It is symmetrical , unimodel and bell shaped.
shaped
•It is uniquely determined by the two parameters ,
namely mean and standard deviation
deviation.
•In the family of normal curves smaller the standard
deviation , higher will be the peak.
peak
•If the original observations follow a normal model with
mean mu and std dev sigma then the averages of random
sample of size n drawn from this distribution will also
follow a normal distribution
distribution.
•The mean of the new model is same as the original
model II.ee mu but the standard deviation gets reduced to
σ (sigma)/root "n" 36
 Sources of Variation
Common Cause

S i l Cause
Special C

37
 Causes of Variability
• Common Causes:
– Random variation (usual)
– No pattern
– Inherent in process
– adjustingg thee process
adjus p ocess increases
c eases itss variation
va a o
• Special Causes
– Non-random variation (unusual)
– M exhibit
May hibi a pattern
– Assignable, explainable, controllable
– adjusting
j g the process
p decreases its variation

SPC uses samples to identify that special causes have occurred

38
 LEANRING 7
If only common cause of variation are present, the
output
p of a process
p forms a distribution that is
stable over time and is
PREDICTABLE.

PREDICTION

39
 SO WHAT?

That's great, we can make prediction based

on sigma, So what?

Once we know
O k the
h sigma
i off a process then;
h
Process has not changed if it is inside +/- 3
sigma.
i
If outside +/- 3 sigma, process has changed

40
 LEANRING 8

SINCE WE CAN NOT SAMPLE 100 UNITS TO

DETERMINE IF OUR MANUFACTURING PROCESS
HAS CHANGED WE NEED A Q QUICK EFFECTIVE
WAY TO MEASURE THE TWO ATTRIBUTE OF
A PROCESS;; THE CENTER AND THE SPREAD

CENTER = AVERAGE
SPREAD = RANGE
= (MAXIMUM - MINIMUM)

Why Average ????

41
 How was our process behaving
over time? Let's calculate the
average and range of each set
CONTROL CHART TEMPLATE
1 23
2 23
3 24
4 26
5 27
Average
g
Range

average = (23+23+24+26+27)/5 Range = 27 - 23

= 24.6 =4
42
 CONTROL CHART TEMPLATE
1 23 22 23 22
2 23 25 23 24
3 24 25 24 24
4 26 26 27 25
5 27 27 27 26
avg 24.6 25.0 24.8 24.2
Min
Min.
Max Plot the average and
Range 4 5 4 4
the range on the control
chart template

Notice the center and the spread

of the process varies much like
when we looked at the histogram
43
 LEANRING 9
If you thought of the control charts as a stretched out
slinky,
li k it would
ld llook
k lik
like a hi
histogram
t if you collapsed
ll d
it. Since the control chart is nothing more than a
histogram e expressed
pressed o over
er time
time, what
hat wee said abo
aboutt
SIGMA applies to the control chart as well.

x-bar Chart
x x
x xx
x xxx xxx
x xxx xxx x
x xxx xxx
x xx
x xx
xx
x xx

44
We can calculate the sigma of all the points in the
control charts and draw lines at +/- 3 sigma. Since
99.7% of the vaules are suppose to fit in the line
we can say that a process has changed if it one of
the points are outside the +/- 3 sigma lines. We
will call the +/-3
/ 3 sigma lines the CONTROL
CO O LIMIT

xx
xxx
xxxxxxxx
xxxxxxx
xxxxxxx +/- 3 sigma
xxx
xxx
xx
xxx 45
HOW DO YOU CALCULATE CONTROL LIMITS?

IIn the
th pastt it was important
i t t for
f operators
t andd auditors
dit
to be able to calculate the control limit. Today, in most
manufacturing g plants
p the computer
p calculates the
control limits and people interpret them.

This makes sense because computers are excellent

at calculating number. However, computers are not
too intelligent. They can not reason and make good
d i i
decisions. People
P l are very capable
bl off reasoning
i and d
making good decision. However, people need good
information. SPC is a tool that converts pprocess data to
information allowing people to focus on what they do best.

46
 LEANRING 10
Control Limits for
Average and Range Chart
X+X+X+…X R+R+R+ R
R+R+R+…R
1 2 3 n 1 2 3 n
X = R =
n n

UCL = X + A2R UCL = D4R

CL =X CL =R
LCL = X - A 2R LCL = D 3R

47
 Setting up control charts:
Calculating the limits
1. Sample n items (often 4 or 5)
2
2. Find the mean of the sample (x (x-bar)
bar) x
3. Find the range of the sample R
4. Plot x on the x chart
5. Plot the R on an R chart
6. Repeat steps 1-5 thirty times
7
7. Average the x ’ss to create x(x-bar-bar)
(x bar bar)
8. Average the R’s to create (R-bar) R
48
 Setting up control charts:
Calculating the limits
9. Find A2 on table (A2 times R estimates 3σ)
10 Use formula to find limits for x-bar
10. x bar chart:

X ± A2 R
11. Use formulas to find limits for R chart:

LCL = D3 R UCL = D4 R

49
 WE USE STATISTICS EVERYDAY

HOW LONG DOES IT TAKE TO GET TO WORK?

50
 TYPES OF VARIABILITY
Common cause= normal, Inherent
Arrive work between 7:55 to 8:01due to number traffics lights
that you stopped at on your way to work.

Special cause = assignable

Arrived to work today at 8:45 because;
a) flat tire on the way to work
b) Accident on the interstate
c) I met up with an old drinking buddy and
I stayed out later than I should have
have.

51
 LEANRING 11

Cp = Tol band / 6 sigma

Cpk = Min of (Avg - LSL) or (USL - Avg) / 3 sigma

_ _ 2 _

σ √
2 2
( 1 ) + (x-x
(x-x ( 2) + … (x-x
( )
= n
(n-1)
(n - 1)

σ (R bar) =
R
d2
52
 Process capability
Good quality: defects are rare (Cpk>1)
Poor quality: defects are common (Cpk<1)

=
USL – x
= 24 – 20 =.667
3σ 3(2)
Cpk = min
=
x - LSL
= 20 – 15 =.833
3
3σ 3(2)

14 20 26
15 24
= =
3σ = (USL – x, or x – LSL) 53
 The control limits can be drawn around both the
average (x-bar) and the Range chart. Therefore,
you can detect several different types of change.
X-bar chart
xx
xxx
xxxxxxx
xxxxxxxx
xxxxxxx
xxx +/- 3 sigma
xxx
xx
xxx

Range chart
xxx
xx
xxx
xxx
xxxxxxxx
xxxxxxx +/- 3 sigma
xxxxxxx
xx
xxx
54
 LOCATION SHIFTS

Process spread
remains same
let's see what
while center
that looks like
increases
in a control chart

55
 Spread remains same
Center shifts up
X-bar chart

+/- 3 sigma

Range chart

+/- 3 sigma

56
 SPREAD CHANGE

Process spread
increase while
center remain
same let's see what
that looks like
in a control chart

57
 Spread increased
Center remain same
X-bar chart

+/- 3 sigma ?
Range chart

+/- 3 sigma

58
 Spread increased
Center remain same
X b chart
X-bar h t

+/- 3 sigma

Range chart

+/- 3 sigma

59
 LEANRING 12
Note that when the process variation increased
the Range chart points shifted to a higher le level.
el
However, the process center (X-bar) seems to
swing wildly going out of both Upper and Lower
control limit while the average is still the same.

Because of the tendency of the X-bar chart to

swing with increase variability
variability, the Range chart
must be reviewed first to determine if the process
variability increased prior to looking at the X-bar
chart to determine if the process shifted.

60
 TREND

target
+/- 3 ssigma

Rule of thumb
thumb, if there are 7 points in a row all higher or lower
than the preceeding point. In this case from the start of the
trend to the time a point went outside the control limit there were
12 samples.
l An
A experinced
i d operator/auditor
t / dit would ld b
begin
i llooking
ki
for assignable cause much sooner. 61
 SHIFT

target
+/- 3 ssigma

Rule of thumb, if there are 6 consequetive points above or below the

target line, a process shift has occurred. In this case, because the
process shifted to somewhere between the target and the upper control
limit, there is a good chance that a point will be outside the control limit
soon. In the above example, it took about 11 points to go outside the
control
t l lilimit.
it AAn experienced
i d operator/auditor
t / dit would ld h
have llooked
k d ffor
assignable cause sooner. 62
Summary Process changes
Summary..

Small shift .. in Center while Spread same

in Spread while Center same

Large shift in Center up or down while Spread same

Spread
p increase while Center same

Center slowly trending up or down while spead same

Center
C t shift
hift up or down
d att the
th same time
ti the
th spreadd
increase 63
 PROCESS CAPABILITY

So you now know how to detect change in a

process. You even know how to detect different
type of change to the process distribution. Up to
now we have not talked about the QUALITY of the
products being produced while the process is
controlled using SPC methods.

If we controll the
h process the h process will
ill produce
d
parts with variation as the equipment is CAPABLE
off producing.
d i W
We callll thi
this PROCESS CAPABILITY
CAPABILITY.

64
 C C
Cp Cpk.....
k S
Say what?
h t?

C k
Cpk Cp
Lower Upper
pp
spec. spec.

7:42 7:48 7:54 8:00 8:06 8:12

65
 Cpk = Target - lower spec or Upper spec - Target
3 sigma 3 sigma

Cpk looks at the likelihood of making product outside

either lower or upper specification

C k
Cpk C k
Cpk
Lower Upper
pp
spec. spec.

7:42 7:48 7:54 8:00 8:06 8:12

66
 LEANRING
IT IS IMPORTANT TO KNOW
WHAT YOUR MACHINE IS
CAPABLE OF PRODUCING.
OTHERWISE YOU MAY BE
CHASING YOUR TAIL
TRYING TO GET THE
MACHINE TO DO WHAT IT IS
NOT CAPABLE OF DOING.
67
 Each red x represents five individual
reading (blue x) that are spread out
more than the average (red x)

x UPPER SPEC X-bar chart

x x
x
xxx UpperControl Limit
x xx x x
xx x x xx x xx xx
x x
xx x x xx xxx x x
xx xxxxxxx
x xxx
xxxxxxx
x x x x = x
x x x xx xxxxxxx
xx x xxxxxxxx
x x x xx xxxxxxx
xxxxxxx
x xxxxxxx
xxx
x x xxx xx
x xxx
xx xxx xxx
x xx
xx xxxxxxx xx x
x
xxx
x xxx
xx LowerControl Limit
x
x LOWER SPEC Each point is an average
x of five indivdual points

Control chart will not differentiate a capable and

a not capable process
process. it will only signal change.
change
The control chart does not care what the spec 68is.
 LEANRING
If your process is not capable, then there is a
good chance that some of your sample will
have valuesoutside the specification. Chances
are if you are notrunning SPC control chart
chart, you
x may be tempted to make an adjustment. Let's
x see what would happen.
pp
x
xxx Upper U spec
x xx x
xx x xxx xxx
xx x
x xx xxxxx
x x xx xxxxxxx
x x
x x x xx xxxxxxx
xx x
x x x xx xxxxxxx x
xxxxxxx
x x xxx xx
x xxx
xx xxx
x
xx xxxxxxx
xxxx x
x xxx
xx
x
x Lower spec
x
69
 LEANRING
If by
b the
th luck
l k off the
th draw
d you gett a reading
di below
b l th
the
lower specification even though the process has not
changed,
h d andd adjusted
dj t d ththe machine
hi up.
The distribution will shift up.

x Upper Sample
p
x U spec
xx
x
xxxxx x
xx x
x xxx
xxxxxxx
x x x
x x x xx xxxxxxx x
x
x x x xx xxxxxxxxx x
x x
x xxx xxx xxxxxxx
xxxxxxx xx = x
x x xxxx x
x xxx
xx xxx
x xxxxxxx x x
x
xx x
xxx x
xx
x Lower spec
x
x
70
 LEANRING
After the
Aft th distribution
di t ib ti shifted
hift d up, there
th is
i now a
much greater chance of getting a value outside
upper specification.
ifi ti So
S the
th machine
hi iis adjusted
dj t d
down, slightly more than it was adjusted up.
Chance off outt off
Ch
spec was = 10%
x
x Chance of out of
xxx
x spec is now = 40%
xx xx x
x Upper spec xx x xxx
xxxxxxx
x x x x x
x x x xx xxxxxxx
x
xx x x x xx xxxxxxxxx
x x x x
xx xx x
x x xxx xxx xxxxxxx
xxxxxxx
xxx x xxx
xxxxxxxx x x x xxx x xxxxxx
x x x xx xxxxxxx x x x
xxxxx x
xxx xx
x
x
x x x xx xxxxxxxx x x x
x
x xxxx x xxx
xxx xxxxxxx
xxxxxxx xx
x x xxx x x
x xxx
xx xxx x
x xxxx
xxxxx x x x
x x
x xxx x
xx L
Lower spec
x
x 71
x
 LEANRING 13
The adjustments continues until, the actual products
produced varies much more than the capability of the
machine
machine.

x
x x
x x
xxx x
x x xx x
xx x x xx x xx xxx
x x xx x
xx Upper spec x x x
x
x
x xx xxx x x xx x x xx x xx
x
xx x x x xx
x xxxxxxxx xx x xx xxx x x
x xx x
xx x x xx x xx x x x xx xxxxxxx
x
xxxxxxxx x x xx
x
x xx xxxxxxxx
x x
x x x xx xxxxxxx x x xxx x xxxxxxxx
xx x xx xxx x x x x xxxx xxxxxx xx x x x xxx xxxxxxx
x xxxxxxx
xx xx xxxxxxx x
x xx xxxxxxxx x x xxx x xx xx
x x x xxx x xxx
xx x x xx xxxxxxx x x xxx x xx
x x x x xxxxx xxx xxxx
x x x xx xxxxxxx x xxx xx xx xxxx x xx x xxx xx
x x xxx x xxxxxx xx x x xx xxx x x x xxx
x x xxxxxxxxx x x xx x x xx xxxxxxx x
x xxx x
x x xxx x x
x xxx x xx xxxxxxxxx xx
xx Lower x x xxx x
x xxxxxxx
x xxxxxxx x
x L spec x x
x xxxx xxxxxx x
x x xx
x
xxx xx x
xxx
x xx
x x xxx
xx
x
x
x 72
 LEANRING 14
If you are controlling your process using SPC Method,
even if your process is not capable, no adjustment
would take place
place. Therefore
Therefore, the product you produced
is what the machine is capable of and not more.

x x
x
x
xx Upper spec UpperControl Limit
x
xx xx x x
x xx
xxx x xxx
xxxxxxx
x x
x xxx
x x x xx xxxxxxx = x xxxxxxx
x x x xx xxxxxxxxx xxxxxxxx
x x x
x xxx
x xxxxxxx
xxx xxxxxxx
xxxxxxx xxx
xxx
x x xxxx x
x xxx
xx xxx xx
x xxxxxxx x x x
xxx
x
xx x
xxx
xx Lower spec LowerControl Limit
x
x
x

73
74
 ATTRIBUTE CONTROL CHARTS

Defectives Defects

Variable
V i bl
Sample
pp-Chart
Chart c - Chart
Fixed
Sample np-Chart
np Chart u - Chart

75
 CONTROL CHART

When you need to discover how much variability in a process is due

to unique events/individual actions in order to determine whether a
PROCESS IS IN STATISTICAL CONTROL
Attribute
Variable Data
Data

X bar - R chart p Chart

X-s Chart np Chart
Ch
X- Mr Chart c Chart
u chart

76
THANKS

77