Statistical

Process Control
Kvalitetsflow 2015

1
 2

Suggested Program
Day 1
08.00-09.50 8.00-8.30 Welcome – Presentation and introduction
8.30-8.55 Over view of SPC – The basics
9.00 – 9.50 The reason for SPC – flow chart Variation

Coffee
10.10.-12.10 10.10-11.00 SPC – Cpk – Six Sigma – short discussion
11.05-12.10 What type of chart in different situations

Lunch
12.40-15.00 12.45-13.15 Setting up SPC Systems
13.20-14.15 Checklist : Ready to measure ?
14.20-14.40 Discussion – What do need the most ?
14.50-15.00 Planning for day 2
 3

Suggested Program
Day 2
08.00-09.50 8.00-8.30 Setting up SPC in QDA
8.30-8.55 Make QDA fit to your data
9.00 – 9.50 Case: try it on your data

Coffee
10.10.-12.10 10.10-11.00 Case: try it on your data
11.05-12.10 Analysis and reporting

Lunch
12.40-15.00 12.45-13.15 Analysis and reporting
13.20-14.15 Additional theory?
14.20-14.40 Analysis on your data – setting up reports
14.50-15.00 Evaluation and actions agreed
Training in SPC
and Setting up
SPC in QDA

4
 5

Setting up a SPC in general

 Proces variables – how to find the relation
between the proces variables and our
SPC measure system.
 First things first – MSA (Measure system
analysis) to approve the system before we
collect all data make decision on ”shaky
ground”
 Collecting basic information about the
proces
 USL, LSL
 UCL, LCL
 CP and Cpk values and targets
 6

Setting up SPC i QDA

 7

Six Sigma as a Metric


Sigma =  = Deviation
( Square root of variance )

Axis graduated in Sigma

-6
-5
-4
-3
-2
-1
-7

1
2
3

5
6
7
0

4
between + / - 1 68.27 % result: 317300 ppm outside
(deviation)
between + / - 2 95.45 % 45500 ppm

between + / - 3 99.73 % 2700 ppm

between + / - 4 99.9937 % 63 ppm

between + / - 5 99.999943 % 0.57 ppm

between + / - 6 99.9999998 % 0.002 ppm

 8

Effect of 1.5 Sigma Process

Shift
 Drift and combinations of PPM
and sigma
• part per million.
• Green, indicates data that are normally used
Sigma Drift 3 3.5 4 4.5 5 5.5 6
0 2,700 465 63 3.4 0.57 0.034 0.002
0.25 3,577 666 99 12.8 1.02 0.1056 0.0063
0.5 6440 1382 236 32 3.4 0.71 0.019
0.75 12228 3011 665 88.5 11 1.02 0.1
1 22832 6443 1350 233 32 3.4 0.39
1.25 40111 12201 3000 577 88.5 10.7 1
1.5 66803 22800 6200 1350 233 32 3.4
1.75 105601 40100 12200 3000 577 88.4 11
2 158,700 66800 22800 6200 1300 233 32
9
 10

Problem Definition
• What do you want to improve?
• What is your ‘Y’?
Reduce
Complaints
(int./ext.)

Reduce Reduce
Defects Cost

What are the Goals?

Problem Definitions need to be based on
quantitative facts supported by analytical data.
 11

Map the Process

Identify the variables - ‘x’

Measure the Process

Understand the Problem -

’Y’ = function of variables -’x’
Y=f(x)

To understand where you want to

be, you need to know how to get
there.
 12

Control Phase
How to create a Control Plan: (one
suggestion...)

 Select Causal Variable(s). Proven vital few

X(s)
 Define Control Plan
 5Ws for optimal ranges of X(s)
 Validate Control Plan
 Observe Y
 Implement/Document Control Plan
 AuditControl Plan
 Monitor Performance Metrics
 13

Control Phase
Control Plan Tools:

 Statistical Process Control (SPC)

Used with various types of distributions
Control Charts
 Attribute based (np, p, c, u). Variable
based (X-R, X)
 Additional Variable based tools
 -PRE-Control
 -Common Cause Chart (Exponentially Balanced
Moving Average (EWMA))
 14

Control Phase
How do we select the correct Control Chart:

Attributes Variables
Type
Data
Measuremen
Individuals t of
Defects Graph defects Defectives
of defectives
Ind. Meas. or subgroups
subgroups

Yes
Oport. Area Yes
Normally dist. Interest in Yes
constant from
C, u data X, Rm sudden mean
sample to
changes
sample
No No
No
If mean is big, X
u and R are effective MA, EWMA or X-R
too CUSUM and Rm
Yes
Size of the
subgroup p, np Use X - R chart
constant
with modified
More efective to
rules
No detect gradual
Ir neither n nor p
changes in long
are small: X - R, X
term
p - Rm are effective
15
 16

Statistical Process Control

(SPC)
 Invented by Walter Shewhart at Western
Electric
 Distinguishes between
 common cause variability (random)
 special cause variability (assignable)
 Basedon repeated samples from a
process
 17

Statistical Process Control

(SPC)
A methodology for monitoring a process
to identify special causes of variation and
signal the need to take corrective action
when appropriate
 SPC relies on control charts
 18

Variability
8
7
 Deviation = distance between 10
observations and the mean (or 8
average) 9

Emmett
 Results for “Emmett”

Observations Deviations
10 10 - 8.4 = 1.6
9 9 – 8.4 = 0.6
8 8 – 8.4 = -0.4
8 8 – 8.4 = -0.4
7 7 – 8.4 = -1.4 Jake
averages 8.4 0.0
 19

Variability
 Deviation = distance between
observations and the mean (or average)

Emmett

 Results for “Jake”

Observations Deviations
7 7 – 6.6 = 0.4 7
7 7 – 6.6 = 0.4 6
7 7 – 6.6 = 0.4 7
6 6 – 6.6 = -0.6 7
6 6 – 6.6 = -0.6
6 Jake
averages 6.6 0.0
 20

Variability 8
7
 Variance = average distance between 10
observations and the mean squared 8
9

Emmett

Observations Deviations Squared Deviations

10 10 - 8.4 = 1.6 2.56
9 9 – 8.4 = 0.6 0.36
8 8 – 8.4 = -0.4 0.16
8 8 – 8.4 = -0.4 0.16
7 7 – 8.4 = -1.4 1.96 Jake
0.0 1.0
averages 8.4
Variance
 21

Variability
 Variance = average distance between
observations and the mean squared

Emmett

Observations Deviations Squared Deviations

7 7
7 6
7 7
6 7
6
6 Jake
averages
 22

Variability
 Variance = average distance between
observations and the mean squared

Emmett

Observations Deviations Squared Deviations

7 7 - 6.6 = 0.4 0.16 7
7 7 - 6.6 = 0.4 0.16 6
7 7 - 6.6 = 0.4 0.16 7
6 6 – 6.6 = -0.6 0.36 7
6 6 – 6.6 = -0.6 0.36
6 Jake
0.0
averages 6.6 0.24 Variance
 23

Variability
 Standard deviation = square root of
variance

Emmett

Variance Standard Deviation

Emmett 1.0 1.0
Jake 0.24 0.4898979

Jake
 24

Variability
The world tends to be bell-shaped

Even very rare Most outcomes Even very rare

outcomes are occur in the outcomes are
Fewer Fewer
possible in the
middle in the
possible
(probability > 0) “tails” “tails” (probability > 0)
(lower) (upper)
 25

Variability
Even outcomes that are equally likely (like dice),
Here is why: when you add them up, become bell shaped
 26

“Normal” bell shaped curve

Add up about 30 of most things

and you start to be “normal”

Normal distributions are divide

up into 3 standard deviations
on each side of the mean

Once your that, you know a lot

about what is going on
 27

Causes of Variability
 Common Causes:
 Random variation (usual)
 No pattern
 Inherent in process
 adjusting the process increases its variation
 Special Causes
 Non-random variation (unusual)
 May exhibit a pattern
 Assignable, explainable, controllable
 adjusting the process decreases its
variation
 28

3 Sigma and 6 Sigma Quality

Lower Upper
specification specification

1350 ppm 1350 ppm

1.7 ppm 1.7 ppm

Process
mean
+/- 3 Sigma

+/- 6 Sigma
 29

Statistical Process Control

 The Control Process
 Define
 Measure
 Compare
 Evaluate
 Correct
 Monitor results
 Variations and Control
 Random variation: Natural variations in the
output of a process, created by countless minor
factors
 Assignable variation: A variation whose source
can be identified
 30

Sampling Distribution
Sampling
distribution

Process
distribution

Mean
 31

Normal Distribution

Standard deviation

3 2 2 3

Mean
95.44%

99.74%
 32

Control Limits
Sampling
distribution

Process
distribution

Mean

Lower Upper
control control
limit limit
 33

SPC Errors
 Type I error
 Concluding a process is not in control when
it actually is.
 Type II error
 Concluding a process is in control when it is
not.
 34

Type I Error

/2 /2

Mean

Probability LCL UCL

of Type I error
 35

Observations from Sample

Distribution
UCL

LCL

1 2 3 4
Sample number
 36

Control Chart
 Control Chart
 Purpose: to monitor process output to see if
it is random
 A time ordered plot representative sample
statistics obtained from an on going
process (e.g. sample means)
 Upper and lower control limits define the
range of acceptable variation
 37

Control Chart
Abnormal variation Out of
due to assignable sources control
UCL

Mean
Normal variation
due to chance
LCL
Abnormal variation
due to assignable sources

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Sample number
 38

Control Charts in General

 Are named according to the statistics
being plotted, i.e., X bar, R, p, and c
 Have a center line that is the overall
average
 Have limits above and below the center
line at ± 3 standard deviations (usually)

Upper Control Limit (UCL)

Center line

Lower Control Limit (LCL)

 39

Histograms do not take into

account changes over time.

Control charts can tell

us when a process
changes
 40

Control Chart Applications

 Establishstate of statistical control
 Monitor a process and signal when it goes
out of control
 Determine process capability
 41

Commonly Used Control

Charts
 Variables data
 x-bar and R-charts
 x-bar and s-charts
 Charts for individuals (x-charts)

 Attribute data
 For “defectives” (p-chart, np-chart)
 For “defects” (c-chart, u-chart)
 42

Developing Control Charts

 Prepare
 Choose measurement
 Determine how to collect data, sample
size, and frequency of sampling
 Set up an initial control chart
 Collect Data
 Record data
 Calculate appropriate statistics
 Plot statistics on chart
 43

Use QDA it is
much easier
 44

Next Steps
 Determine trial control limits
 Center line (process average)
 Compute UCL, LCL
 Analyze and interpret results
 Determine if in control
 Eliminate out-of-control points
 Recompute control limits as necessary
 45

Limits
 Process and Control limits:
 Statistical
 Process limits are used for individual items
 Control limits are used with averages
 Limits = μ ± 3σ
 Define usual (common causes) & unusual
(special causes)
 Specification limits:
 Engineered
 Limits = target ± tolerance
 Define acceptable & unacceptable
Process vs. control limits
46

Distribution of averages

Control limits
Specification limits

Distribution of individuals

Process limits

 Variance of averages < variance of individual items

 47

Variables Data Charts

 Process Centering
n


X bar chart
X bar is a sample mean X i
X i 1
n
 Process Dispersion (consistency)
 R chart
 R is a sample range

R  max( X i )  min( X i )
 48

X bar charts
 Center line is the grand mean (X double
bar)
m
 Points are X bars

x  / n
X j 1
j

X
m
UCL  X  z x LCL  X  z x
-OR-

UCL  X  A2 R LCL  X  A2 R
 49

R Charts
 Center line is the grand mean (R bar)
 Points are R
 D3 and D4 values are tabled according
to n (sample size)

UCL  D4 R LCL  D3 R
 50

Use of X bar & R charts

 Charts are always used in tandem
 Data are collected (20-25 samples)
 Sample statistics are computed
 All data are plotted on the 2 charts
 Charts are examined for randomness
 If random, then limits are used “forever”
Attribute Charts
ccharts – used to
count defects in a
constant sample size
n

c
c i 1
 centerline
m

LCL  c  z c

UCL  c  z c
Attribute Charts
p charts – used to track
a proportion (fraction)
defective
m

p x n

x
p j 1
  centerline
i
ij p 
i
i 1
n
m nm

p (1  p ) p (1  p )
UCL  p  z LCL  p  z
n n
 53

Control Charts for Variables

 Mean control charts
 Used to monitor the central tendency of a
process.
 X bar charts
 Range control charts
 Used to monitor the process dispersion
 R charts

Variables generate data that are measured.

 54

Mean and Range Charts

(process mean is
shifting upward)
Sampling
Distribution

UCL

x-Chart Detects shift

LCL

UCL

Does not
R-chart
detect shift
LCL
 55

Mean and Range Charts

Sampling
Distribution (process variability is increasin

UCL

Does not
x-Chart
LCL
reveal increase

UCL

R-chart Reveals increase

LCL
 56

Control Chart for Attributes

 p-Chart - Control chart used to monitor
the proportion of defectives in a process
 c-Chart - Control chart used to monitor
the number of defects per unit

Attributes generate data that are counted.

 57

Use of p-Charts
 When observations can be placed into
two categories.
 Good or bad
 Pass or fail
 Operate or don’t operate
 Whenthe data consists of multiple
samples of several observations each
 58

Use of c-Charts
 Use
only when the number of
occurrences per unit of measure can be
counted; non-occurrences cannot be
counted.
 Scratches, chips, dents, or errors per item
 Cracks or faults per unit of distance
 Breaks or Tears per unit of area
 Bacteria or pollutants per unit of volume
 Calls, complaints, failures per unit of time
 59

Use of Control Charts

 Atwhat point in the process to use control
charts
 What size samples to take
 What type of control chart to use
 Variables
 Attributes
 60

Run Tests
 Run test – a test for randomness
 Any sort of pattern in the data would
suggest a non-random process
 All points are within the control limits - the
process may not be random
 61

Nonrandom Patterns in
Control charts
 Trend
 Cycles
 Bias
 Mean shift
 Too much dispersion
 62

Typical Out-of-Control Patterns

 Point outside control limits
 Sudden shift in process average
 Cycles
 Trends
 Hugging the center line
 Hugging the control limits
 Instability
 63

Shift in Process Average

 64

Identifying Potential Shifts

 65

Cycles
 66

Trend
 67

Final Steps
 Use as a problem-solving tool
 Continue to collect and plot data
 Take corrective action when necessary
 Compute process capability
 68

Process Capability
 Tolerances or specifications
 Range of acceptable values established by
engineering design or customer
requirements
 Process variability
 Natural variability in a process
 Process capability
 Process variability relative to specification
 69

Process Capability
Lower Upper
Specification Specification

A. Process variability
matches specifications
Lower Upper
Specification Specification

B. Process variability
Lower Upper
well within specifications Specification Specification

C. Process variability
exceeds specifications
 Process Capability Ratio

specification width
Process capability ratio, Cp =
process width

Cp = Upper specification – lower specification

6

70
 71

Improving Process Capability

 Simplify
 Standardize
 Mistake-proof
 Upgrade equipment
 Automate
 72

Taguchi Loss Function

Traditional
cost function
Cost

Taguchi
cost function

Lower Target Upper

spec spec
Meet the Guru: The ”old” philosophy
Taguchi Lower tolerance Upper tolerance

Taguchi
Quality
through
design
1) Product must
bee robust
to variation The ”Taguchi” philosophy
Lower tolerance Upper tolerance
in process
2) Loss is equal
to distance
from
nominal
 But in reality it looks like this ?
The OK /NOK quality
Lower tolerance Upper tolerance
 75

Limitations of Capability
Indexes
 Process may not be stable
 Process output may not be normally
distributed
 Process not centered but Cp is used
Process Capability
 Theratio of process
variability to design Text Text Text Text Text Text

specifications Natural data

spread
Title

The natural
-3σ -2σ - µ +1σ +2σ +3σ spread of the
1σ data is 6σ
Lower Upper
Spec Spec
 77

Empirical Rule

-3 -2 -1 +1 +2 +3
68%

95%

99.7%
 78

Gauges and
Measuring Instruments
 Variable gauges
 Fixed gauges
 Coordinate measuring machine
 Vision systems
 79

Examples of Gauges
 80

Metrology - Science of
Measurement
• Accuracy - closeness of agreement
between an observed value and a
standard
• Precision - closeness of agreement
between randomly selected individual
measurements
 81

Repeatability and
Reproducibility
 Repeatability (equipment variation) –
variation in multiple measurements by an
individual using the same instrument.
 Reproducibility (operator variation) -
variation in the same measuring
instrument used by different individuals
 82

Repeatability and
Reproducibility Studies
 Quantify
and evaluate the capability of a
measurement system
 Select m operators and n parts
 Calibrate the measuring instrument
 Randomly measure each part by each
operator for r trials
 Compute key statistics to quantify
repeatability and reproducibility
 83

Reliability and Reproducibility

Studies(2)
M easurement (M ) made by
Op erators (i from 1 to m) on

Use QDA it is
Parts (j from 1 to n) in
Trials (k from 1 to r)
 
   M ijk 
 
xi   j k  average for each op erator

much easier
nr
xD  max ( xi )  min ( xi ) difference (range) of op erator averages
i i

R ij  max ( M ijk )  min ( M ijk ) range for each p art for each op erator
k k

 
  Rij 
 
Ri   j  average range for each op erator
n
 
  Ri 
R  i  average range of all
m
 84

Reliability and Reproducibility Studies(3)

Control limit of ranges Rij  D4  R

Use number trials (r) for n in table. Check
for randomness of errors.

Use QDA it is
Rep eatability or Equip ment Variation
EV  K1  R K1 is a constant tied to # of trials
Rep roducibility or op erator (ap p raisal) variation
 EV 2 
K 2  xD 2

much easier
AV   nr  K 2 is a constant tied to # of op erators
 
Rep eatability and Rep roducibility
R&R  EV 2   AV 2

Results are in actual units measured. Customary to exp ress

as p ercentages.
Under 10% - Accep table
10 - 30% - ? based on imp ortance and rep air cost
Over 30% - Unaccep table
 85

R&R Constants

Number of Trials 2 3 4 5

K1 4.56 3.05 2.50 2.21

Number of Operators 2 3 4 5

K2 3.65 2.70 2.30 2.08

 86

R&R Evaluation
 Under 10% error - OK
 10-30% error - may be OK
 over 30% error - unacceptable
 87

How do we know our process?

Process Map

Fishbone

Historical Data
 88

RATIONAL SUBGROUPS
Minimize variation within subgroups
BLACK NOISE Maximize variation between subgroups
(Signal)
WHITE NOISE
(Common Cause
Variation)
RESPONSE
PROCESS

TIME
RATIONAL SUBROUPING Allows samples to be taken that
include only white noise, within the samples. Black noise
occurs between the samples.
 89

Visualizing the Causes

Within Group

Time 1

Time 2

Time 3

Time 4

•Called  short term (sst)

 st + shift = •Our potential – the best we
can be
total
•The s reported by all 6
sigma companies
•The trivial many
 90

Visualizing the Causes

Time 1

Time 2

Time 3

Time 4
•Called shift (truly a
measurement in sigmas of how
far the mean has shifted)
•Indicates our process control
 st + shift = total •The vital few
Between Groups
 91

 Assignable Cause

 Outside influences
 Black noise
 Potentially controllable
 How the process is actually performing
over time

Fishbone
 92

 Common Cause Variation

 Variation present in every process

 Not controllable
 The best the process can be within the
present technology
 93

Gauge R&R

2Total = 2Part-Part + 2R&R

Recommendation:
Resolution £ 10% of tolerance to measure
Gauge R&R £ 20% of tolerance to measure
R&R
• Repeatability (Equipment variation)
Part-Part
• Variation observed with one measurement device when used several times by one
operator while measuring the identical characteristic on the same part.
• Reproducibility (Appraised variation)
• Variation Obtained from different operators using the same device when measuring
the identical characteristic on the same part.
•Stability or Drift
• Total variation in the measurement obtained with a measurement obtained on the
same master or reference value when measuring the same characteristic, over an
extending time period.
 94

In many cases, the data sample can be transformed so that it is approximately normal. For example, square roots,
logarithms, and reciprocals often take a positively skewed distribution and convert it to something close to a bell-
shaped curve
 95

What do we Need?
LSL USL LSL USL

Off-Target, Low Variation On Target

High Potential Defects High Variation
Good Cp but Bad Cpk High Potential Defects
No so good Cp and Cpk
LSL USL

Variation reduction and process


centering create processes with
less potential for defects.
 The concept of defect reduction
applies to ALL processes (not just
On-Target, Low Variation manufacturing)
Low Potential Defects
Good Cp and Cpk
 96

Eliminate “Trivial Many”

• Qualitative Evaluation
• Technical Expertise
• Graphical Methods
• Screening Design of Experiments Identify “Vital Few”
• Pareto Analysis
• Hypothesis Testing
• Regression
Quantify • Design of Experiments
Opportunity
• % Reduction in Variation
• Cost/ Benefit
Our Goal:
Identify the Key Factors (x’s)
 97

Graph>Box plot Graph>Box plot

DBP
Without X values 1

75% 9

DBP 1

10 0 Day DBP

9 4 1

10 50% 9
0

9
4 9

1
DBP
25%
99 9
0 Operator
14
94
4
0

9 9

9
1

0 Shift 9

Box plots help to see the data distribution 4

 98

Statistical Analysis

Apply statistics to validate actions & improvements

Hypothesis Testing
7 30

5
20
Frequency

Frequency
4

3
10
2

0 0

0.000 0.005 0.010 0.015 0.020 0.025 0.000 0.005 0.010 0.015 0.020 0.025
New Machine Machine 6 mths

Regression Analysis
Regression Plot
Y = 2.19469 + 0.918549X • Is the factor really important?
R-Sq = 86.0 %

• Do we understand the impact for

60

50

40

the factor?
Y

30

• Has our improvement made an

20

10 Regression
95% PI
0

5 15 25 35 45 55
impact
X

• What is the true impact?

 99

M.A.D

Stop
Adjust
Technical process &
design
Requiremen
t
Con- Preliminary Identify
Obtain Data on Calculate Z Rev 0
sumer Drawing/Databas Critical
Similar Process values Drawing
Cue e Process s
Identity
CTQs

Stop
Fix process
1st piece & design
inspection
Z<3

Prepilot Recheck
Data ‘Z’ levels
Obtain data
Z>= Design
Intent M.A.I.C

Pilot data
 100

Excercise