ISOM 2700: Operations Management

Session 4. Quality Management I

Statistical Process Control

Lijian Lu
Dept. of ISOM, HKUST

1
 Financial Crisis of 2008

The crisis represents a failure of quality, and solving it will require, among other
things, careful management of quality in financial institutions and across financial
supply chains.

Paul Zipkin. 2009. Quality Snags in the Mortgage-Finance Supply. Quality

Management Journal 16 (3)

2
 Cost of Poor Quality
Source: Wall Street Journal

Source:
The New York Times

3
 Dimensions of Quality
• Quality of design
– Performance
– Features
– Aesthetics
– Perceived quality
– Reliability / Durability
– (Example) Laptop

• Quality of conformance
– Deviations from the target
– The degree to which the product or service
design specification are met
– (Example) Label says content = 500 ml, but…
– (Example) Uber says driver arrives in 5 mins, but…
4
 Major Concepts of Quality Management
• Customer satisfaction
– Understand, evaluate, define, and
manage customer expectations
"What the business thinks it
produces is not of first importance;
what the customer thinks he is
• Management responsibility buying and considers "value" is
decisive; it determines what a
business is.”
Peter Drucker
• Continuous improvement The Practice of Management,
1989

• Prevention over correction

“Do the right things right the first time, every time.”

5
 Ways to Manage Quality
Make sure process
satisfies reqs Make sure these
parts satisfy reqs

Process control Inspection (Next class)

• Objective: Control the process to • Objective: Sort out bad products
improve the quality of product before they reach customers
• Procedure • Procedure
– Track quality metrics
• Acceptance sampling
– Control charts • Six sigma
– Find root causes and improve • Total Quality Management (TQM)
continuously

Prevention Detection 6
Variations are “Normal”

7
 Two Causes of Quality Variation
• Common-cause variation (natural variation)
– Natural (inherent) variation in process
– Affects all output
– Cannot be eliminated without change in process
itself Low level High level
– Examples: product design, work method, ect

• Assignable variation
• External to the normal operation of a process
• Affects only some but not entire output
• Causal factor can be detected and potentially
eliminated
• Example: unskilled workers, poor material, ect
8
Process with only Common-cause Variation
• Track the output of the process
– If it forms a stable distribution (frequency) over time, the
process is in statistical control
– If it does not, the process is out of control

Prediction
Frequency

e Process in statistical control:

Tim • Distribution is stable over time
Weight • All the variation is common-cause
Example: weight of boxes of cereal
9
 Process with Assignable Variation
• The distribution (frequency) is not stable over time if one (or
more) of the following are changing
– Location (mean of the distribution)
– Dispersion (width of the distribution)
– Shape ?
?? ??
? ?
? ?
? ?
? ?
? ??
?? ?

Prediction
Frequency

i m e
T
Weight

Example: weight of boxes of cereal

10
 Deming’s Funnel Experiment
Policy 1: Keep y fixed at x*
Frequency

x*
y
Policy 2: After each drop, reset
position of funnel using:
0
x* xi ynew = yold - (xi - x*)
Frequency

Deming’s
Experiment Video
https://www.youtube.com/watch?v=t5DV69_2VeQ
x*
11
 Results
Policy 1: do nothing Policy 2: reset position
140 100
90
120
80
100 70
80 60
50
60 40
40 30
20
20
10
0 0
1

10

13

16

19

22

25

28

31

34

37

40

10

13

16

19

22

25

28

31

34

37

40
When only common-cause variations exist, the process
is performing at its best. So leave it alone!

But how do we know a process has only common-cause variations?

12
 Implications
• Two types of mistakes
– Assume a variation or a mistake is due to an assignable
cause when in fact the cause is due to the process
• Can lead to over adjustment and lower levels of quality

– Assume a variation or a mistake is due to the process

when in fact it has an assignable cause
• Can result in missed opportunities to correct mistakes / make
improvements

• Need a method for detecting assignable versus

common-cause variation so that one can respond
intelligently
13
 Conformance Analysis

• A process is in control if it is
operating without assignable
cause variation.

• Otherwise, the process is out of

control, and corrective action
should be taken

14
 Monitoring Process: Control Charts
• Objective
– Use data to detect variabilities due to assignable causes
• A statistical technique to determine how the process
distribution has shifted over time and if it’s in control

15
 Example: LCD Display

• Test the quality and performance of LCD displays

before shipping

• Test for
– Contrast
– Color scales
– Dead pixels
– … and more

16
 Types of Control Charts

Control
Charts Binary value:
good or bad

Continuous Attribute
Variables Variables

X-bar Chart P Chart

17
 P-chart

• Attribute variables:
– good or bad
– defective or non-defective

• P-chart
– Controls the percentage of defective items
– Type of data
• A series of samples over time
• for each sample, number of defective (non-defective) items

18
 Construction of P-chart
• Step 1. Identify sample size n (within a subgroup)
• Step 2. In each sample i, compute:
– pi: fraction of defective items
• Step 3. Compute means of sample:
– p-bar: mean pi

p (1 - p )
• Step 4. Compute standard deviation: s p =
n

• Step 5. Compute control limits

p (1 - p )
UCL p = p + 3s p = p + 3
n
p (1 - p )
LCL p = max{ p - 3s p ,0} = max{ p - 3 ,0}
n 19
 Example: Red Bead Experiment
Number of red
beads p
1 7 0.1400
2 4 0.0800
3 1 0.0200
4 5 0.1000
5 6 0.1200
6 4 0.0800
7 9 0.1800
8 3 0.0600
9 3 0.0600
10 3 0.0600
11 5 0.1000
12 6 0.1200
13 6 0.1200
Dr W. Edwards Deming 14 6 0.1200
15 9 0.1800
• 50 beads per draw 16
17
10
11
0.2000
0.2200

• bead is white or red 18

19
4
6
0.0800
0.1200
20 1 0.0200
– White is good 21 8 0.1600
– Red is bad 22
23
6
3
0.1200
0.0600

Red Bead p = 0.1072 24

25
7
1
0.1400
0.0200

Game p (1 - p ) pbar 0.1072

sp = = 0.0438 sigma p 0.0438
20
n
https://www.youtube.com/watch?v=geiC4UgpDyw
 P-chart

UCL p = p + 3s p = 0.1072 + 3 ´ 0.0438 = 0.2384

0.2500

0.2000
Up trend
Percentage red

0.1500
p = 0.1072

0.1000

0.0500

0.0000
1 3 5 7 9 11 13 15 17 19 21 23 25

sample

LCL p = max{ p - 3s p ,0} = max{0.1072 - 3 ´ 0.0438,0} = 0

21
 Types of Control Charts

Control
Charts Binary value:
good or bad

Continuous Attribute
Variables Variables

X-bar Chart P Chart

22
 X-Bar Chart

• Suitable for measurable characteristics of quality

– Example: The weight of an item, the diameter of a bottle
cap, the thickness of a circuit board, and so on

• X-bar chart
– Tests for change in mean of distribution over time
– Location test
– Type of data:
• A series of samples over time
• For each sample, measures taken on each unit

23
 X-bar Chart: An Example
Sample x1 x2 x3

• The thickness of a 1
2
0.0629
0.0630
0.0636
0.0631
0.0640
0.0622

printed circuit 3
4
0.0628
0.0634
0.0631
0.0630
0.0633
0.0631
5 0.0619 0.0628 0.0630
board is an 6 0.0613 0.0629 0.0634
7 0.0630 0.0639 0.0625
important quality 8
9
0.0628
0.0623
0.0627
0.0626
0.0622
0.0633
parameter 10
11
0.0631
0.0635
0.0631
0.0630
0.0633
0.0638
12 0.0623 0.0630 0.0630
• Data on board 13
14
0.0635
0.0645
0.0631
0.0640
0.0630
0.0631
thickness (in inch) 15
16
0.0619
0.0631
0.0644
0.0627
0.0632
0.0630

are as given. 17
18
0.0616
0.0630
0.0623
0.0630
0.0631
0.0626
19 0.0636 0.0631 0.0629
20 0.0640 0.0635 0.0629
21 0.0628 0.0625 0.0616
22 0.0615 0.0625 0.0619
23 0.0630 0.0632 0.0630
24 0.0635 0.0629 0.0635
25 0.0623 0.0629 0.0630

24
 Construction of X-bar Chart
• Step 1. Identify sample size n (within a subgroup)

• Step 2. In each sample, compute:

– X-bar: mean value
– R: range (= max – min)

• Step 3. Compute mean of sample means:

– x-double bar: mean of X-bar
– R-bar: mean of R

• Step 4. Compute control limits

LCLx = x - A2 R UCLx = x + A2 R
25
 Table for A2
Sample size

n A2
-------------------------
2 1.880
3 1.023
4 0.729
5 0.577
6 0.483
10 0.308
15 0.223
20 0.180
25 0.153
26
 X-bar Chart: An Example
• Data on circuit Sample
1
x1
0.0629
x2
0.0636
x3
0.0640
x-bar
0.0635
R
0.0011

board thickness 2
3
0.0630
0.0628
0.0631
0.0631
0.0622
0.0633
0.0628
0.0631
0.0009
0.0005
4 0.0634 0.0630 0.0631 0.0632 0.0004
Size of subgroup: n = 3 5
6
0.0619
0.0613
0.0628
0.0629
0.0630
0.0634
0.0626
0.0625
0.0011
0.0021
7 0.0630 0.0639 0.0625 0.0631 0.0014
8 0.0628 0.0627 0.0622 0.0626 0.0006
9 0.0623 0.0626 0.0633 0.0627 0.0010

x = 0.0630 10
11
0.0631
0.0635
0.0631
0.0630
0.0633
0.0638
0.0632
0.0634
0.0002
0.0008
12 0.0623 0.0630 0.0630 0.0628 0.0007
R = 0.0009 » 0.001 13
14
0.0635
0.0645
0.0631
0.0640
0.0630
0.0631
0.0632
0.0639
0.0005
0.0014
15 0.0619 0.0644 0.0632 0.0632 0.0025
16 0.0631 0.0627 0.0630 0.0629 0.0004
17 0.0616 0.0623 0.0631 0.0623 0.0015
18 0.0630 0.0630 0.0626 0.0629 0.0004
19 0.0636 0.0631 0.0629 0.0632 0.0007
20 0.0640 0.0635 0.0629 0.0635 0.0011
21 0.0628 0.0625 0.0616 0.0623 0.0012
22 0.0615 0.0625 0.0619 0.0620 0.0010
23 0.0630 0.0632 0.0630 0.0631 0.0002
24 0.0635 0.0629 0.0635 0.0633 0.0006
25 0.0623 0.0629 0.0630 0.0627 0.0007
mean 0.0630 0.0009
27
 X-bar Chart
Circuit Board Thickness
X Bar Control Chart

0.0645

UCLx = x + A2 R = 0.063 + 1.023 * 0.001 = 0.064

0.064

0.0635
x = 0.063
Thickness

0.063

0.0625

0.062

0.0615 LCLx = x - A2 R = 0.063 - 1.023 * 0.001 = 0.062

Out of control point
0.061
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

Group

28
 Using Control Charts

• Objective: diagnose the state of the process (in or out

of control)
• Take corrective action when
§ outside the control limits (out of control)
§ 8 or more consecutive points on one side of center line
§ Trends of 8 or more consecutive points (decreasing or
increasing)
UCL

LCL

29
 Detect Abnormal Variation in the Process:
Detect – Stop - Alert
Toyota Video
https://www.youtube.com/watch?v=r_-Pw49ecEU

Jidoka Andon Board / Cord

If equipment malfunctions / gets out of A way to implement Jidoka in an assembly line
control, it shuts itself down automatically
to prevent further damage Make defects visibly stand out
Requires the following steps:
Detect Once worker observes a defect, he shuts down
Alert the line by pulling the andon / cord
Stop
The station number appears on the andon board
Source: www.riboparts.com, www.NYtimes.com
 Takeaways

• Understand the difference between common-cause

and assignable variation

• Understand a key quality philosophy: Never tweak a

process that is in control

• Know how to use X-bar chart and P-chart

31