Professional Documents
Culture Documents
Continuous Improvement
Continuous Improvement
1
Manufacturing Excellence Evolution
“The past causes the present,
and so the future…….”
1984
100 % Sampling
Integrated
Lean Theory of
Quality Control Japan TQC TQM 1988 Lean Constraints,
and Statistical Quality BY In Western Six Lean and Six
Theory (SPC) Circles JAPAN Companies SIX ơ Sigma Sigma (TLS)
1924 1962 1968 1980 1986 2001 2014
1947 – ISO
came Iso 9001-1987
TQM + BUSINESS EXCELLENCE MODELS
into being 1990 onwards
2
Variation Reduction: Variation is
the Enemy of Quality
3
The Traditional definition that Quality is meeting
requirements is not good enough.....
Loss of business......
Loss of Jobs.....
Better Definition of Quality
Mean Target
How do we improve consistency of performance?
Minimize variation around the target by reducing the distance of the mean to the target ( m) and by
reducing the standard deviation
Q m
5
Dr Genichi Taguchi is the Originator of this
World Class Definition of Quality
• The common thinking around specification limits is that the customer is satisfied as
long as the variation stays within the specification limits. If the variation exceeds the
limits, then the customer immediately feels dissatisfied. The specification limits divide
satisfaction from dissatisfaction. For example, if the lower limit is 10, and the upper
limit is 20, then a measurement of 19.9 will lead to customer satisfaction, while a
measurement of 20.1 will lead to customer dissatisfaction ???
• Taguchi states that any variation away from the nominal (target) performance will
begin to incur customer dissatisfaction. As the variation increases, the customer will
gradually (exponentially) become dissatisfied. In the previous example, if the
measurement is 19.9, the customer will be dissatisfied more than a measurement of
19.8. If the measurement is 20.1, the customer will be slightly more dissatisfied than
the measurement of 19.9.
• Taguchi states that the specification limits do not cleanly separate satisfaction
levels for the customer. Any variation away from the nominal (value of 15 in the
example above) will start to incur customer dissatisfaction. The quality does not
suddenly plummet once the limits are exceeded, rather it is a gradual degradation as
the measurements get closer to the limits. The goal of a company should be
to achieve the target performance with minimal variation. That will
minimize the customer dissatisfaction.
6
Example: Quality
Dissatisfaction is not Stepwise
You purchase the orange on Day 1, but if you eat the orange you will be very
dissatisfied, as it is not ready to eat. This would fall below the lower limit. On
Day 3 it would be acceptable to eat, but you are still dissatisfied because it
doesn’t taste as good as eating on the target date. If you wait for Day 5, you
will be satisfied, because it is eaten on the ideal date. If you wait until Day 7,
you will be slightly dissatisfied, because it is one day past the ideal date, but it
will still be within the limits provided by the supermarket. If you wait until Day
9, you will be very dissatisfied, as it will be too far past the ideal date.
7
World Class Quality
• Variation always creates costs. Actions to deal with variation after the fact will
inevitably increase costs. Actions to reduce variation at the source will reduce
costs while increasing product quality
• The traditional approach to the problem of product variation has been that of
specifications. As long as the quality characteristic X falls within specification
limits, the product is said to be satisfactory. When outside specification the
product is suddenly deemed to be unsatisfactory
• The sudden shift in the way product is treated will essentially create a step-
function in any cost curve
• The very nature of the Specification Approach fosters periods of neglect
followed by periods of intense process scrutiny and do not support Continuous
Improvement
8
HOW FAR HAVE YOU PROGRESSED
ON THE ROAD TO QUALITY
IMPROVEMENT???
Determine and Maintain
Optimum Control Factor
SPC
Ideal
Settings in the Presence of + Taguchi DOE State
”Noise”
Innocence
Is this where
You are currently?
State of Chaos
9
The Only Way Out of the Cycle of Despair
(LOSS OF HOPE)!!!
Ideal State
Process in Control
100% conforming product
Control Charts
PROCESS IMPROVEMENT
Threshold State o Maintain process in control
o Give timely warning of any
Process in Control
troubles
Some nonconfroming
product
Musr either…
Real Time SPC Charts
ENTROPY
o Change process, or
o Change
specification
Sorting is only a
temporary fix
Used Correctly
Control Charts (not after the fact / “grave yard” SPC)
o Maintain process
in control ***is the tool to get you
o Evaluate efforts of out of the Cycle of Despair***
improvement
dominate
Random fluctuations due
to Assignable Causes will
Cycle of Despair
eventually frustrate
efforts at process
Entropy Slide
improvement
The only way out of chaos
is to first eliminate the
Assignable Causes
10
The Effect of Entropy
11
Cycle of Despair
12
The Only Way Out
• If a process is out of control the Control Chart does nothing but prompt
• The users have to identify and remove Assignable Causes of Uncontrolled
Variation
• If the process is in control, the Control Chart can be used to evaluate the results
of process experimentation, but it is the producer who must perform
experiments and then follow up with removal of faults of the system
• The power of the Control Chart depends on how effective the company is
organised to utilize the insights offered by the Control Chart
• Whenever the information provided by the SPC Charts is not effectively used,
the cycle of improvement is broken – these breaks allow
Entropy to degrade the process
14
Each and every production process can be in one(1) of four (4) states…
New Definition of
Trouble
Conforming and Non trouble
Predictable
Nonconforming Product trouble
and Predictable Double Trouble
Conforming yet Process trouble
Unpredictable
Nonconforming Double trouble
and
Unpredictable
15
Important Question: What is your
company’s SPC Level of Maturity?
• Level 1: Neglect
•
• Control Charts either don’t exist or do exist…but nobody pays attention to them
• SPC has a low credibility rating
Level 2: Vulnerability
?
• After the fact analysis of Control Charts by Specialists to determine if something has gone wrong
• The production department mainly respond to out-of-spec situations
16
Remarks on Control Charts
17
More on Statistical Process Control
Charts
• When prediction is not feasible due to Uncontrolled Variation, the Control Chart
will warn of this instability
• A process is predictable, when it is in a state of Statistical Control
• Control Charts help people understand when they can perform predictions
about the future behaviour of the process
18
The Shewhart Concept of Variation
19
The Shewhart Concept of Variation
20
The two types of variation
• statistically predictable
• mathematical certain
• source; the system itself; common
causes
21
Two Ways to Improve a Production
Process
• When a process displays Controlled Variation, it should be thought of as
stable and consistent. The variation present in the process consist only
of that which is inherent in the process itself. Therefore, to reduce the
variation, the process itself must be changed.
• If a process displays Uncontrolled Variation, it is changing unpredictable
from time to time. It is both inconsistent and unstable. Therefore, the
first step in improving the output of the process is to identify the
Assignable causes for excessive variation and to remove it from the
process.
• The first step in any attempt to improve a process is to determine if the
process displays Uncontrolled Variation.
• The tool for the detection of Uncontrolled Variation is the Shewhart
Control Chart (SPC Control Chart used correctly)
22
Common Cause (Controlled)
Variation
23
Where does Common Cause
Variation Come From?
24
Stable Process
25
Reduction of Common Cause
Variation
26
Special Cause (Uncontrolled)
Variation
27
Where Does Special Causes Come
From?
28
The “Out-Off Control”/ Unstable
Processes
29
Common Versus Special Causes
30
How to use Control Charts to “Stabilise”
Processes?
• Shewhart’s control charts give very few false alarms, When a point is
outside a control limit it is most likely due to presence of some
Assignable Causes
• Whenever the running record stays within the three sigma limits, and
varies above and below the central line in a random manner, it is safe to
say that the process appears to be stable
• But what if the running record does not vary between control limits in a
random manner?
If the pattern can be meaningfully interpreted in the context of the process, it should
be taken as an Assignable Cause
If the “pattern” makes no sense in the context of the process, and it does not persist
over an extended period of time, then it needs no interpretation
31
Use Control Chart Pattern Analysis – to identify
non-random patterns
32
Blast Furnace Silicon Measurements
File: UNNAMED.DAT
SILICON X-bar
ucl
200 v
v
l v v
l
l v l l
l
150 v l
l l l cl
l v v l
l
l l l l l
l
l l l l
100 l
l
lcl
Range
ucl
l
150
l
l l v
l l
100 l
l
l l
l l l l
l l l l cl
l
50 l l
l l l l
v
l l l
l
l
l
0 lcl
5 10 15 20 25 30
X-bar: cl: 144.5253 ucl: 215.693 lcl: 73.35753 Subgrp Size: 3
Range: cl: 69.54545 ucl: 179.0511 lcl: 0 v Rule Violation
33
Control Chart Pattern Analysis – to identify non-random patterns
34
Control Chart Pattern Analysis – to identify non-random patterns
35
Four Rules for Defining Lack of Control
• Detection Rule 1:
A lack of Control is indicated whenever a single value falls outside the 3 sigma control limits
• Detection Rule 2:
A lack of control is indicated whenever 2 out of 3 successive values fall on the same side of,
and more than 2 standard deviations away, from the central line
• Detection Rule 3:
A lack of control is indicated whenever 4 out of 5 successive values fall on the same side of,
and more than 1 standard deviations away, from the central line
• Detection Rule 4:
A lack of control is indicated whenever at least 8 successive values fall on the same side of
the central line
36
Assessing the Capability of a Stable Process
An Unstable Process
is unpredictable and
does not have a
Capability
37
Capability, Stability and World Class Quality
38
The Capability of a Stable Process
39
What is a Reasonable Degree of
Statistical Control?
40
Example: Capability of Stable Process (a Multi Cavity Mold
Process)
ucl
7 hour to hour variation of cavities
6 l l l l
l
l
l
5 l l l l
l l l cl
l
l l
4
l l
l l l
3 l l
2 lcl
Range
ucl
8 cavity to cavity variation within subgroup l
6 l
l l l
4 l l l l l
cl
l l l l l l l l
2 l l l l
0 lcl
5 10 15 20
X-bar: cl: 4.65625 ucl: 7.358131 lcl: 1.954369 Subgrp Size: 4
Range: cl: 3.708333 ucl: 8.462608 lcl: 0 v Rule Violation
These four parallel operations are sub-grouped together, because they have
been adjusted so that the four cavities are essentially equivalent
41
Example: Capability of Stable Process (refer to Multi
Cavity Mold Process)
30 Mean
0
0 5 10 15 20
Samples: 96 Spec Lim: (0, 15)
Mean: 4.65625
Std Dev: 1.868453
No matter how the data behave. Virtually all of the Individual Values will fall
𝑅
within the Natural Process Limits: 𝑋± 3 = 4.66 ± 3(3.71/2.059) = -0.75 to 10.07
𝑑2
43
Specified Tolerance
44
Distance to Nearest Specification Limit (DNS)
DNS = MIN{𝑍𝐿 ; 𝑍𝑈 }
Based on the Empirical Rule, its is generally desirable for the DNS to be
larger than 3
The process must additionally display a ***reasonable degree of statistical
control*** (there is no way of knowing this without a Control Chart!!)
Show that DNS of the above Process = 2.59 sigma – is this desirable?
Re-aiming the process will resolve this issue.
45
Capability Ratio’s 𝑪𝒑 𝒂𝒏𝒅 𝑪𝒑𝒌
𝑈𝑆𝐿 − 𝐿𝑆𝐿
𝐶𝑝 =
6𝑋𝑅 𝑪𝒑 and 𝑪𝒑𝒌 should both be > 1
𝑑2
These indices are dimensionless
𝑈𝑆𝐿 − 𝑋 And confusing!!!
𝐶𝑝𝑘𝑢 =
3𝑋 𝑅
𝑑2
𝑋 − 𝐿𝑆𝐿
𝐶𝑝𝑘𝑙 = The better to use is “The Specified Tolerance” is
3𝑋 𝑅
𝑑2 8.33 sigma units…and the Distance to Nearest Spec
(DNS) = 2.59 sigma units
𝐶𝑝𝑘 = min{𝐶𝑝𝑘𝑢 ; 𝐶𝑝𝑘𝑙 }
46
Proof of the “uselessness” of Capability Indice 𝑪𝒑𝒌 Over
Time
10 Subgroups of size 5 from a stable process (50 values) were used to calculate the 𝐶𝑝𝑘
on a daily basis. Spec: [3;17] – (no adjustments to the process!!)
ucl
1.6
l
1.4 l l
l
l cl
1.2 l
l
l
lcl
0.8
2 4 6 8 10
Individual: cl: 1.302 ucl: 1.721481 lcl: 0.8825193
v Rule Violation
50
The Hypothetical Capability (Process Spread) of an
Unstable Process
51
The Hypothetical Process Limits of an Unstable
Process (Process Aim is hard to change)
52
Hypothetical Capability of Socket Cavity 4
53
Hypothetical Capability of Socket Cavity 4
File: UNNAMED.DAT
CAVIT4_THICKNESS X-bar
12
u
u
u
10 u u ucl
l
l l l
l
8 l cl
l l
v l l
6 lcl
u u
u
4 u
Range
v ucl
6 l l v l
4 l
cl
l l l l l l l
2 l l l l
l l l
0 lcl
5 10 15 20
X-bar: cl: 7.86 ucl: 9.792345 lcl: 5.927655 Subgrp Size: 5
Range: cl: 3.35 ucl: 7.083572 lcl: 0 v Rule Violation
54
Hypothetical Process Spread for Socket Cavity 4
File: UNNAMED.DAT
CAVIT4_THICKNESS
Mean
15
LSL USL
10
0
0 5 10 15 20
Samples: 100
Mean: 7.86
Std Dev: 2.589879
6(3.35)
Hypothetical Process Spread = = 8.6 𝑠𝑖𝑔𝑚𝑎 𝑢𝑛𝑖𝑡𝑠
2.326
55
Hypothetical Capability of Socket Cavity 4 (Removed
out of Control Averages)
File: UNNAMED.DAT
CAVIT4_THICKNESS
AVERAGE_THICKNESS From >6 To <10 X-bar
ucl
10
l
l
9 l
l l
l
8 cl
l
l l
7
l l
6 lcl
Range
8 ucl
6 l l l
4 cl
l l l l
2 l l
0 lcl
2 4 6 8 10
X-bar: cl: 8.109091 ucl: 10.31149 lcl: 5.90669 Subgrp Size: 5
Range: cl: 3.818182 ucl: 8.073542 lcl: 0 v Rule Violation
56
Hypothetical Process Limits for Cavity 4 –
(instability removed)
3𝑥𝑅
Hypothetical Process Limits = 𝑋 ±
𝑑2
3.35
= 8.11 ± 3 = [3.79 to 12.43]
2.326
File: UNNAMED.DAT
CAVIT4_THICKNESS
AVERAGE_THICKNESS From >6 To <10
Mean
12
10 LSL USL
0
0 5 10 15 20
Samples: 55
Mean:
Std Dev:
8.109091
1.94053
57
Short Term Capability Studies
• A short-term study will typically generate between 25 to 50 values. These values might be
dimensions of parts produced in sequence, or repeated measurements of the same thing,
or values obtained from repeatedly cycling through the same operation
• A naïve approach to short-term capability places these values in a histogram and
computes some numerical summaries
• Since Average and Range Charts can miss some signals, especially
when used with finite amounts of data, it is wise to look at a running
record of the individual values
• One begin with a running record of these 50 values plotted in their time order sequence.
The running record should be examined for any obvious patterns such as trends or
sustained shifts
• If nothing is seen in the running record one may continue to plot a XmR chart for the
Individual Values. The Individual Values Chart is examined for out-of-control points and the
Moving Range Chart is examined for evidence of breaks in the sequence of values
• If the XmR Chart does not show evidence of lack of control, then it may be
assumed that the machine or process has a defined capability
• An estimate of capability is obtained from calculating the Natural Process Limits
• These limits only apply to the machine or process for the conditions under which the data
was collected. Conditions normally involve one operator, using one batch of raw material,
at a time when the machine has just been adjusted to operate properly
• If the capability is insufficient, the machine or process should not be purchased 58
Note
Don’t tamper with the process when it is not
capable and the problem is in the system.
Over adjustment of the process makes
things worse.....
LSL USL
59
Note
60
Advantages to attaining
Statistical Control (Stability)...
61
Note
62
Responsibility Assignments...
63
The responsibility matrix for process
improvement
Statistical Control?
Yes No
No Action Local Workforce
needed now but must find the special
Yes should continually causes and initiate
improve corrective action
Are the results
Satisfactory?
Management Local Workforce
must correct must find the special
No the system causes and initiate
corrective action
64
Five (5) steps for improving any
process......
65
Step1: Create the Environment
66
Step 2: Identify key measures and set up the control
plan
67
Step 3: Assure Capable and Controlled Measurement
Systems
68
Step 4: Eliminate special causes
69
Step 5: Improve Process Common Cause Variation
70
ISHIKAWA’s message
71