Statistical Process Control

Training Manual
 Shift in Customer Expectations

Customer’s at
1970’s Today’s Customer
Lower
Improved Product Costs
product
quality
Longer Higher
Delivery Product
times Costs

Shorter Improved
Delivery product
times quality
 Why SPC?

Customer Expectations (Q,C,D)

Process Output
(Q,C,D) GAP
 Why this gap?

Variation Waste
 Variation

Take note of variation in data

Distinguish between “Common” and “Special” Cause
variation.

Observe the famous “Dice Experiment” to understand the

impact of variation.
 Causes of Variation

Special Cause Common

Cause

UCL
Plan
_
X
LCL

Do
Special Cause
Check and Act
 Steps for SPC

1. Develop Operational Definition.

2. Conduct MSA study
3. Develop Sampling Plan
4. Collect Data as per Sampling plan and plot the
appropriate control chart.
5. Look for Special causes
6. CAPA on the Special causes.
7. Assess process capability and improve capability, if
required.
8. Hand over to process owner for DWM.
 Steps to develop Operational Definition

• Identify the characteristic of interest.

• Select the instrument
• Describe the method of inspection /test
• State the decision criteria
• Document the operational definition
• Test the operational definition
Operational Definition ( Format)
 Types of Data

1. Attribute data: Value of countable quality characteristics

Example - number of defects, number of defectives.

2. Variable data: Value of measurable quality

characteristics

Example – Tread Length, Tread Weight, Fabric gauge,

Booking Temperature.
 Sampling Plan

Inputs / Type of Data Data Who will Data Method

Outputs (Variable / Source Sample Size Collect Collection of Collecting
Discrete) & Location the Data? Frequency data

         

           

           

           
 Need for MSA before SPC
Xbar chart
Temporary: UCL=2.69, Mean=2.40, LCL=2.10 (n=5)
2.7 UCL

2.6

2.5

2.4 Mean

2.3

2.2

LCL
2.1

Range chart
Temporary: UCL=1.07, Mean=0.50, LCL=none (n=5)
UCL

1.0

0.8

0.6
Mean

0.4

0.2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Process variation + measurement variation = resultant data

 Basic Statistics

Measures of Central Tendency and Dispersion –

Mean & Standard Deviation
Different Mean but Same Mean but
same spread (variation) different spread (variation)

Different Means,
different spread
(variation)
 Control Charts

• Purpose
– To help people manage systems to make the right
decisions.
• Benefits :
– Prevents over control
– Prevents under control
– Shows how the system reacts to the change.
Selection of Charts
Control Chart Format
X bar-R Chart
Calculating Control Limits for Average & Range Chart

UCL  X  A2 R
Control limits calculated in the normal way:
LCL  X  A2 R

UCL  D4 R
LCL  D3 R
 Control Chart Constants

Sample
size A2 d2 D3 D4 E2

2 1.880 1.128 -- 3.268 2.660

3 1.023 1.693 -- 2.574 1.772
4 0.729 2.059 -- 2.282 1.457
5 0.577 2.326 -- 2.114 1.290
6 0.483 2.534 -- 2.004 1.184
7 0.419 2.704 0.076 1.924 1.109
8 0.373 2.847 0.136 1.864 1.054
9 0.337 2.970 0.184 1.816 1.010
10 0.308 3.078 0.223 1.777 0.975
 Rules for Special Cause Identification

Rule 1: Any point outside the limits

 Rules for Special Cause Identification

Rule 2: 7 Points in a row above or below the

mean line
 Rules for Special Cause Identification

Rule 3: 7 points in a row descending or ascending

Common Vs. Special Causes
 Control Chart & Normal Distribution

1/2 the data

is either side
of average

2/3 of the
data is
contained in
the centre 1/3
 Normal distribution curve

Compare to the normal distribution.

Characteristics of normal distribution

Mean = Median = Mode

Distribution is symmetrical
Normal distribution curve

Normal distribution.
Most common distribution.
Bell shaped.
 Normal distribution curve

Lower Spec Upper

Spec

-  - 1σ
µ
+ 1σ
+ 
68.26%

- 2σ 95.44% + 2σ

- 3σ + 3σ
99.73%
99.9937% + 4σ
- 4σ

- 5σ + 5σ
99.99943%

- 6σ + 6σ
99.999998%
Exercise: Histogram
 Exercise: Histogram

Class Width = 2
Classes Frequency Total
34-35 8
36-37 25
38-39 31
40-41 35
42-43 20
44-45 5
46-47 1
Exercise: Histogram
 Calculating Capability Indices

R USL  LSL Z min

ˆ  Cp  Cpk 
d2 6ˆ 3

USL  X X  LSL
Z upper  Z lower 
ˆ ̂
0.0668 0.00023
Total out of specification = 6.68% + 0.023% =
6.703%
30-31
Capability Analysis
 Calculate Cp/Cpk

USL LSL Target Process Mean [Link] Cp Cpk

48 36 42 44 1
48 36 42 43 1
48 36 42 42 1
48 36 42 41 1
48 36 42 40 1

Cp

Cpk
A

Is Cp < = > Cpk?

B

Is Cp < = > Cpk?

C

Is Cp < = > Cpk?

 A
In these three
charts the process
variation is the
same, but as the B
process becomes
more centred, Cpk
gets? C
A

Is Cp < = > Cpk?

B

Is Cp < = > Cpk?

C

Is Cp < = > Cpk?

 A
In these three
charts the process
average is the
same, but as the B
process variation
is reduced, Cpk
value goes?
C
 Capability for One-sided specifications

• Can be either a maximum or a minimum.

• Complete as for the normal method.
• Only one specification is used -
• So can only calculate Zupper or Zlower
• Cp cannot be calculated.
• Can only calculate Cpk (Cpu or Cpl)
 Calculating Capability during new machine trials

New product or service.

Entire trial lot data ( No sampling).
Trial run, with data collection.
Complete control chart and histogram.
Complete analysis as before but use
actual standard deviation instead of
estimated standard deviation.
 Standard Deviation during Trial Runs

Calculate the actual standard deviation.


 (x  x) 2

n 1
 Capability during Trial Runs

Calculate and interpret the

capability indices.
Cp becomes Pp
USL  LSL
Cp becomes Pp Pp 
6
USL  LSL
Pp  Interpret as Cp
6
Interpret as Cp
Cpk becomes Ppk
Z min
Cpk becomes Ppk Ppk 
3
Z min
Cpk  Interpret as Cpk
3
Interpret as Cpk
 For On going process with samples at regular intervals

R
ˆ 
d2

USL  LSL Z min

Cp  Cpk 
6ˆ 3
14.00  0.00 2.00
 
6  2.00
3
 1.17
 0.67
 New
process

Take a sample of at least

50 consecutive parts

Complete a variables
Find and remove SPC for
special cause(s)
control chart
manufactured
parts - 1
Is the
process No
stable?

Yes
Improve
Complete a capability process
study using Pp and Ppk

Is the
process No
capable?

Yes
 Yes
Complete an on- Find and remove
going control chart special cause(s)

Is the No
SPC for
process
stable? manufactured
parts - 2
Yes
Complete a capability Improve
study using Cp and Cpk process

Is the
Yes No
process
capable?
 Understanding Control Vs. Capability

• Control • Capability
– Assess statistical – Process output meets
stability specifications
– No ‘special cause’
variation
– Only ‘common cause’
variation
 Kume Matrix: Stability vs. Capability

(a) Unstable and Incapable: long way to go

(b) Stable but Incapable: Reduce natural variation
(c) Unstable but Capable: Dangerous situation
(d) Stable and Capable: Ideal state
What is the result of SPC ?