utdallas.edu/~metin
1
Quality Management
Chapter 9-10
These slides are based in part on slides that come with Cachon & Terwiesch
book Matching Supply with Demand http://cachon-terwiesch.net/3e/. If you
want to use these in your course, you may have to adopt the book as a textbook
or obtain permission from the authors Cachon & Terwiesch.
utdallas.edu/~metin2
Learning Goals
Statistical Process Control
X-bar, R-bar, p charts
Process variability vs. Process specifications
Yields/Reworks and their impact on costs
Just-in-time philosophy
utdallas.edu/~metin3
Steer Support for the Scooter
utdallas.edu/~metin4
Steer Support Specifications
Go-no-go
gauge
utdallas.edu/~metin5
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
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
R
79.9
79.91
79.92
79.93
79.94
79.95
79.96
79.97
79.98
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
X-b
ar
Control Charts
utdallas.edu/~metin6
Statistical Process Control (SPC)
SPC: Statistical evaluation
of the output of a process during production/service
The Control Process
– Define
– Measure
– Compare to a standard
– Evaluate
– Take corrective action
– Evaluate corrective action
» Apply this to Global Warming or lack of it, or to Temperature Volatility.
utdallas.edu/~metin7
Not just the mean is important, but also the variance
Need to look at the distribution function
The Concept of Consistency:
Who is the Better Target Shooter?
utdallas.edu/~metin8
Statistical Process Control
Capability
AnalysisConformance
Analysis
Investigate for
Assignable Cause
Eliminate
Assignable Cause
Capability analysis
• What is the currently "inherent" capability of my process when it is "in control"?
Conformance analysis
• SPC charts identify when control has likely been lost and assignable cause
variation has occurred
Investigate for assignable cause
• Find “Root Cause(s)” of Potential Loss of Statistical Control
Eliminate assignable cause
• Need Corrective Action To Move Forward
utdallas.edu/~metin9
Statistical Process Control
Shewhart’s classification of variability:
– Common (random) cause
– Assignable cause
Variations and Control
– Random variation: Natural variations in the output of
process, created by countless minor factors
» temperature, humidity variations, traffic delays.
– Assignable variation: A variation whose source can be
identified. This source is generally a major factor
» tool failure, absenteeism
utdallas.edu/~metin10
Common Cause Variation (low level)
Common Cause Variation (high level)
Assignable Cause Variation
Two Types of Causes for Variation
utdallas.edu/~metin11
Mean and Variance
Given a population of numbers, how to compute the
mean and the variance?
deviation Standard
)(
Variance
Mean
},...,,{Population
1
2
2
1
21
N
x
N
x
xxx
N
i
i
N
i
i
N
utdallas.edu/~metin12
Sample for Efficiency and Stability
From a large population of goods or services (random if possible) a sample is drawn.
– Example sample: Midterm grades of OPRE6302 students whose last name starts with letter R {60, 64, 72, 86}, with letter S {54, 60}
» Sample size= n
» Sample average or sample mean=
» Sample range= R
» Standard deviation of sample means=
x
population theofdeviation Standard: where
n
x
utdallas.edu/~metin13
Sampling Distribution
Sampling distribution
Variability of the average scores of
people with last name R and S
Process distribution
Variability of the scores
for the entire class
Mean
Sampling distribution is the distribution of sample means.
Grouping reduces the variability.
utdallas.edu/~metin14
Normal Distribution
Mean
95.44%
99.74%
x
at x. cdf normal )1,_,,(normdist:functions lstatistica Excel
at x. pdf normal )0,_,,(normdist:functions lstatistica Excel
devstmeanx
devstmeanx
normdist(x,.,.,0)
Probab
normdist(x,.,.,1)
utdallas.edu/~metin15
Cumulative Normal Density
)_,,(norminv :prob""at cdf offunction Inverse
)1,_,,(normdist:at x (cdf)function Cumulative
:functions lstatistica Excel
devstmeanprob
devstmeanx
0
1
x
normdist(x,mean,st_dev,1)
prob
norminv(prob,mean,st_dev)
utdallas.edu/~metin16
Normal Probabilities: Example
If temperature inside a firing oven has a normal
distribution with mean 200 oC and standard deviation of
40 oC, what is the probability that
– The temperature is lower than 220 oC
=normdist(220,200,40,1)
– The temperature is between 190 oC and 220oC
=normdist(220,200,40,1)-normdist(190,200,40,1)
utdallas.edu/~metin17
Control Limits for Consistency not Correctness
Sampling
distribution
Process
distribution
Mean
LCL
Lower
control
limit
UCL
Upper
control
limit
Process is in control if sample mean is between control limits.
These limits have nothing to do with product specifications!
utdallas.edu/~metin18
Setting Control Limits:
Hypothesis Testing Framework
Null hypothesis: Process is in control
Alternative hypothesis: Process is out of control
Alpha=P(Type I error)=P(reject the null when it is true)=
P(conclude out of control when in control)
Beta=P(Type II error)=P(accept the null when it is false)
P(conclude in control when out of control)
If LCL decreases and UCL increases, we accept the null more easily. What happens to – Alpha?
– Beta?
Not possible to target alpha and beta simultaneously,
– Control charts target only a desired level of Alpha.
utdallas.edu/~metin19
Type I Error=Alpha
Mean
LCL UCL
/2 /2
Probability
of Type I error
st_dev)mean,,2/1norminv(UCL
st_dev)mean,,2norminv(LCL
-α
α/
The textbook uses Type I error=1-99.74%=0.0026=0.26%.
Sampling distribution
utdallas.edu/~metin20
Time
Process
Parameter
Upper Control Limit (UCL)
Lower Control Limit (LCL)
Center Line
• Track process parameter over time
- mean
- percentage defects
• Distinguish between
- common cause variation
(within control limits)
- assignable cause variation
(outside control limits)
• Measure process performance:
how much common cause variation
is in the process while the process
is “in control”?
Statistical Process Control: Control Charts
utdallas.edu/~metin21
Control Chart
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
UCL
LCL
Sample number
Mean
Out of
control
Normal variation
due to chance
Abnormal variation
due to assignable sources
Abnormal variation
due to assignable sources
utdallas.edu/~metin22
Observations from Sample Distribution
Sample number
UCL
LCL
1 2 3 4
utdallas.edu/~metin23
Number of
Observations
in Sample
Sample size (n)
Factor for X-
bar Chart
(A2)
Factor for
Lower
control Limit
in R chart
(D3)
Factor for
Upper
control limit
in R chart
(D4)
Factor to
estimate
Standard
deviation, (d2)
2 1.88 0 3.27 1.128
3 1.02 0 2.57 1.693
4 0.73 0 2.28 2.059
5 0.58 0 2.11 2.326
6 0.48 0 2.00 2.534
7 0.42 0.08 1.92 2.704
8 0.37 0.14 1.86 2.847
9 0.34 0.18 1.82 2.970
10 0.31 0.22 1.78 3.078
Parameters for computing UCL and LCL
the Table method
utdallas.edu/~metin24
Period x1 x2 x3 x4 x5 Mean Range
1 1.7 1.7 3.7 3.6 2.8 2.7 2 2 2.7 2.3 1.8 3 2.1 2.38 1.2 3 2.1 2.7 4.5 3.5 2.9 3.14 2.4 4 1.2 3.1 7.5 6.1 3 4.18 6.3 5 4.4 2 3.3 4.5 1.4 3.12 3.1 6 2.8 3.6 4.5 5.2 2.1 3.64 3.1 7 3.9 2.8 3.5 3.5 3.1 3.36 1.1 8 16.5 3.6 2.1 4.2 3.3 5.94 14.4 9 2.6 2.1 3 3.5 2.1 2.66 1.4
10 1.9 4.3 1.8 2.9 2.1 2.6 2.5 11 3.9 3 1.7 2.1 5.1 3.16 3.4 12 3.5 8.4 4.3 1.8 5.4 4.68 6.6 13 29.9 1.9 7 6.5 2.8 9.62 28 14 1.9 2.7 9 3.7 7.9 5.04 7.1 15 1.5 2.4 5.1 2.5 10.9 4.48 9.4 16 3.6 4.3 2.1 5.2 1.3 3.3 3.9 17 3.5 1.7 5.1 1.8 3.2 3.06 3.4 18 2.8 5.8 3.1 8 4.3 4.8 5.2 19 2.1 3.2 2.2 2 1 2.1 2.2 20 3.7 1.7 3.8 1.2 3.6 2.8 2.6 21 2.1 2 17.1 3 3.3 5.5 15.1 22 3 2.6 1.4 1.7 1.8 2.1 1.6 23 12.8 2.4 2.4 3 3.3 4.78 10.4 24 2.3 1.6 1.8 5 1.5 2.44 3.5 25 3.8 1.1 2.5 4.5 3.6 3.1 3.4 26 2.3 1.8 1.7 11.2 4.9 4.38 9.5 27 2 6.7 1.8 6.3 1.6 3.68 5.1
Average
3.81
5.85
n
xxxX n
...21
},...,min{
},...,max{
21
21
n
n
xxx
xxxR
• Collect samples over time
• Compute the mean:
• Compute the range:
as a proxy for the variance
• Average across all periods
- average of means,
- average of ranges,
• Normally distributed
The X-bar Chart: Application to Call Center
X
R
utdallas.edu/~metin25
• Define control limits
0.58 comes from the Table.
•Constants are taken from a table
• Identify assignable causes:
- point over UCL
- point below LCL
- many (6) points on one side of center
• In this case:
- problems in period 13
- new operator (CSR) was assigned
0
2
4
6
8
10
12
1 3 5 7 9 11 13 15 17 19 21 23 25 27
UCL=X +A2 ×R =3.81+0.58*5.85=7.19
LCL=X -A2 ×R =3.81-0.58*5.85=0.41
CSR 1 CSR 2 CSR 3 CSR 4 CSR 5
mean 2.95 3.23 7.63 3.08 4.26 st-dev 0.96 2.36 7.33 1.87 4.41
Control Charts: The X-bar Chart
The Table method
utdallas.edu/~metin26
Range Control Chart
ranges sample of average theof multipleA
ranges sample of average theof multipleA
3
4
RDLCL
RDUCL
Multipliers D4 and D3 depend on n and are available in the Table.
EX: In the last five years, the range of GMAT scores of incoming PhD class is
88, 64, 102, 70, 74. If each class has 6 students, what are UCL and LCL for
GMAT ranges?
079.6*0 159.279.6*2
.0D ,2D 6,nFor .6.795/)74701026488(
34
34
RDLCLRDUCL
R
Are the GMAT ranges in control?
utdallas.edu/~metin27
0
2
4
6
8
10
12
1 3 5 7 9 11 13 15 17 19 21 23 25 27
0
5
10
15
20
25
30
1 3 5 7 9 11 13 15 17 19 21 23 25 27
X-B
ar
R
Control Charts: X-bar Chart and R-bar Chart
For the Call Center
utdallas.edu/~metin28
X-bar and Range Charts: Which?
UCL
LCL
UCL
LCL
R-chart
x-Chart Detects shift
Does not
detect shift
(process mean is
shifting upward)Sampling
Distribution
utdallas.edu/~metin29
X-bar and Range Charts: Which?
UCL
LC
L
LC
L
R-chart Reveals increase
x-Chart
UCL
Does not
reveal increase
(process variability is increasing)Sampling
Distribution
utdallas.edu/~metin30
• Compute the standard deviation of the sample averages• stdev(2.7, 2.38, 3.14, 4.18, 3.12, 3.64, 3.36, 5.94, 2.66, 2.6, 3.16, 4.68, 9.62,
5.04, 4.48, 3.3, 3.06, 4.8, 2.1, 2.8, 5.5, 2.1, 4.78, 2.44, 3.1, 4.38, 3.68)=1.5687
• Use type I error of 1-0.9974
Control Charts: The X-bar Chart
The Direct method
8.531.5687)9987,3.81,norminv(0.
st_dev)mean,/2,-norminv(1UCL
-0.911.5687)0013,3.81,norminv(0.
st_dev)mean,/2,norminv(LCL
0.0026
Recall LCL=0.41 and UCL=7.19 in the Table method.
utdallas.edu/~metin31
Tolerances/Specifications
– Requirements of the design or customers
Process variability
– Natural variability in a process
– Variance of the sample means coming from the process
Process Capability
Let us Tie Tolerances and Variability
Process capability
– Process variability relative to specifications
Capability = Process specifications / Process variability
utdallas.edu/~metin32
Process Capability:
Specification limits are not control chart limits
Lower
SpecificationUpper
Specification
Process variability matches
specifications
Lower
Specification
Upper
Specification
Process variability well within
specificationsLower
Specification
Upper
Specification
Process variability exceeds
specifications
Sampling
Distribution
is used
utdallas.edu/~metin33
Process Capability Ratio
When the process is centered, process capability ratio
A capable process has large Cp.
Example: The standard deviation, of sample averages of the midterm 1
scores obtained by students whose last names start with R, has been 7.
The SOM requires the scores not to differ by more than 50% in an exam.
That is the highest score can be at most 50 points above the lowest score.
Suppose that the scores are centered, what is the process capability ratio?
Answer: 50/42
X
pC6
levelion specificatLower - levelion specificatUpper
utdallas.edu/~metin34
Process mean
Lower
specification
Upper
specification
+/- 3 Sigma
+/- 6 Sigma
3 Sigma and 6 Sigma Quality
Can you reduce variability so much that 12 Sigmas fit within specification limits?
utdallas.edu/~metin35
• Estimate standard deviation:
• Or use the direct method with the excel function stdev()
• Look at standard deviation relative to specification limits
= R/d2
3
Upper
Specification (USL)
Lower
Specification (LSL)
X-3A X-2A X-1A X X+1A X+2 X+3A
X-6B X X+6B
Process A
(with st. dev A)
Process B
(with st. dev B)
Cp P{defect}
1 0.33 0.317=31.7%
2 0.67 0.0455
3 1.00 0.0027=0.27%
4 1.33 0.0001
5 1.67 0.0000006
6 2.00 2x10-9
The Statistical Meaning of Six Sigma
L
e
s
s
V
a
r
i
a
b
i
l
i
t
y
utdallas.edu/~metin36
Another Chart: Use of p-Charts
p=proportion defective, assumed to be known
When observations can be placed into two categories.
– Good or bad
– Pass or fail
– Operate or don’t operate
– Go or no-go gauge
utdallas.edu/~metin37
• Estimate average defect
1 300 18 0.060
2 300 15 0.050
3 300 18 0.060
4 300 6 0.020
5 300 20 0.067
6 300 16 0.053
7 300 16 0.053
8 300 19 0.063
9 300 20 0.067
10 300 16 0.053
11 300 10 0.033
12 300 14 0.047
13 300 21 0.070
14 300 13 0.043
15 300 13 0.043
16 300 13 0.043
17 300 17 0.057
18 300 17 0.057
19 300 21 0.070
20 300 18 0.060
21 300 16 0.053
22 300 14 0.047
23 300 33 0.110
24 300 46 0.153
25 300 10 0.033
26 300 12 0.040
27 300 13 0.043
28 300 18 0.060
29 300 19 0.063
30 300 14 0.047
p =0.052
SizeSample
pp )1( ̂ = =0.013
pUCL= + 3 ̂
pLCL= - 3 ̂ =0.091
=0.014
Period n defects p
Attribute Based Control Charts: The p-chart
Think of printing defective pages
• Define control limits
• Estimate Standard Deviation
utdallas.edu/~metin38
0.000
0.020
0.040
0.060
0.080
0.100
0.120
0.140
0.160
0.180
13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Attribute Based Control Charts: The p-chart
utdallas.edu/~metin 39
Inspection
» Before a costly operation, PhD comp. exam before candidacy
» Before an irreversible process, firing pottery
» Before a covering process, painting, assembly
» After some use: now 30,000 flights for cracks in the fuselage;
was 60,000 flights before April 1, 11 Southwest incidence.
Centralized vs. On-Site, my friend checks quality at cruise lines
Inputs Transformation Outputs
Acceptance
samplingProcess
control
Acceptance
sampling
Where/When » Raw materials
» Finished products
utdallas.edu/~metin 40
Process Bottleneck
Based on labor and
material cost
MarketEnd of
Process
Defect
occurred Defect
detected
Cost of
defect
$ $ $Based on sales
price (incl. Margin)
Recall, reputation,
warranty costs
Discovery of Defects and the Costs
CPSC, Segway LLC Announce Voluntary Recall to
Upgrade Software on Segway™ Human Transporters
The following product safety recall was conducted by the firm in
cooperation with the CPSC.
Name of Product: Segway Human Transporter (HT)
Units: Approximately 6,000
Recall AlertU.S. Consumer Product Safety Commission
Office of Information and Public Affairs
Washington, DC 20207, September 26, 2003
Defect
detected
Defect
detected
utdallas.edu/~metin41
Examples of Inspection Points
Type of
business
Inspection
points
Characteristics
Fast Food Cashier
Counter area
Eating area
Building
Kitchen
Accuracy
Appearance, productivity
Cleanliness
Appearance
Health regulations
Hotel/motel Parking lot
Accounting
Building
Main desk
Safe, well lighted
Accuracy, timeliness
Appearance, safety
Waiting times
Supermarket Cashiers
Deliveries
Accuracy, courtesy
Quality, quantity
utdallas.edu/~metin42
The Concept of Yields
90% 80% 90% 100% 90%
Line Yield: 0.9 x 0.8 x 0.9 x 1 x 0.9
Assuming that yields are independent.
What if failure rate is higher at the second stage for the items that pass the first stage?
Yield of Resource = rate Flow
resource the atcorrectly processed units of rate Flow
Yield of Process = rate Flow
resources theof allat correctly processed units of rate Flow
utdallas.edu/~metin 43
Rework / Elimination of Flow Units
Step 1 Test 1 Step 2 Test 2 Step 3 Test 3
Rework
Step 1 Test 1 Step 2 Test 2 Step 3 Test 3
Step 1 Test 1 Step 2 Test 2 Step 3 Test 3
Rework:
Defects can be corrected
Same or other resource
Leads to variability
Examples:
- Readmission to Intensive Care Unit
Loss of Flow units:
Defects can NOT be corrected
Leads to variability
To get X, we have to start X/y units
Examples:
- Interviewing
- Semiconductor fab
ReworkRework Rework
utdallas.edu/~metin44
Why Having a Process is so Important:
Two Examples of Rare-Event Failures
Case 1: No rework loops
• Airport security
• Safety elements (e.g. seat-belts)
Case 2: Process has built-in rework loops
• Double-checking
1 problem every
10,000 units
99% correct
Indepen-
dence from
the first
Indepen-
dence from
the first two
“Bad” outcome happens
with probability (1-0.99)3,
That is (0.01) 3.
Good
Bad
99% 99%
99%
1% 1%
1%
You can add rework loops to improve quality.
If checks are dependent, this is less effective.
“Bad” outcome only happens
Every 100*10,000 units
That is probability of (0.01)3
1 problem
every 1 unit
utdallas.edu/~metin 45
Rare events are not so rare:Jetliner Crash due to Engine Icing
Engine flameout due to crystalline icing: Engine stops for 30-90 secs and hopefully starts again.
Suppose 150 1-engine flameouts over 1990-2005 and 15 2-engine flameouts over 2002-2005. What are the annualized 1- and 2-engine flameouts?
10=150/15 and 5=15/3
Let N be the total number of widebody jetliners flying through a storm per year. Assume that engines ice independently to compute N.
Set Prob(2 engine icing)=Prob(1 engine icing)2
(5/N)=(10/N)2 which gives N=20 ???
There are 1200 widebody jetliners worldwide. It is safe to assume that each flies once a day. Suppose that there are 2 storms on their path every day, which gives us about M=730 (=2*365) widebodyjetliner and storm encounters every year. How can we explain M=730 > N=20?
Engines do not ice independently. With M=730, Prob(1 engine icing)=10/730=1.37% and
Prob(2 engine icing)=5/730=0.68%.
Because of dependence,
Prob(2 engine icing) >> Prob(1 engine icing) 2 .
Unjustifiable independence leads to underestimation of the failure probabilities in operations, finance, engineering, flood control, nuclear power plants, etc.
May 11, 2011, the Emirates A-380
from Dubai was struck by lightning
as it approached Heathrow airport
utdallas.edu/~metin 46
Rare events are not so rare:
An Earthquake and a Power Loss
Unjustifiable independence leads to underestimation of the failure probabilities in nuclear power plants.
An earthquake in Japan or a tornado in Texas can cause
both a power outage and a structural damage in a nuclear reactor.Alabama’s Browns Ferry reactor had a faulty cooling valve replaced right before April 27 2011 tornados
that cut off the power to the reactor which was idled for a while (>2 weeks) afterwards.
Squaw Creek Reservoir
for cooling water
Pacific Ocean
for cooling water
utdallas.edu/~metin47
Just-in-Time Philosophy
Pull the operations rather than pushing them
– Inventory reduction
– JIT Utopia
» 0-setup time
» 0-non value added operations
» 0-defects
Discover and reduce process variability
utdallas.edu/~metin48
Push vs Pull System
What instigates the movement of the work in the system?
In Push systems, work release is based on downstream demand
forecasts
– Keeps inventory to meet actual demand
– Acts proactively
» e.g. Making generic job application resumes today (e.g.: exempli gratia)
In Pull systems, work release is based on actual demand or the
actual status of the downstream customers
– May cause long delivery lead times
– Acts reactively
» e.g. Making a specific resume for a company after talking to the recruiter
utdallas.edu/~metin49
Push/Pull View of Supply Chains
Procurement,Manufacturing andReplenishment cycles
Customer Order
Cycle
Customer
Order Arrives
Push-Pull boundary
PUSH PROCESSES PULL PROCESSES
utdallas.edu/~metin50
Direction of production flow
upstream downstream
Authorize
production
of next unit
Pull Process with
Kanban Cards
utdallas.edu/~metin51
Bro
wser
err
or
Number of
defects
Cause of Defect
Absolute
Number Percentage Cumulative
Browser error 43 0.39 0.39
Order number out of sequence 29 0.26 0.65
Product shipped, but credit card not billed 16 0.15 0.80
Order entry mistake 11 0.10 0.90
Product shipped to billing address 8 0.07 0.97
Wrong model shipped 3 0.03 1.00
Total 110O
rder
nu
mb
er
ou
t
off
seq
uen
ce
Pro
du
ct
sh
ipp
ed
, b
ut
cre
dit
card
no
t b
ille
d
Ord
er
en
try
mis
take
Pro
du
ct
sh
ipp
ed
to
billin
g a
dd
ress
Wro
ng
mo
del
sh
ipp
ed
100
50
Cumulative
percents of
defects
100
75
50
25
Pareto Principle or 20-80 rule
Errors in the shipping process
utdallas.edu/~metin52
•It is not enough to look at “Good” vs “Bad” Outcomes
•Only looking at good vs bad wastes opportunities for learning; especially as failures become
rare (closer to six sigma) you need to learn from the “near misses”
Reduce Variability in the Process
Taguchi: Even Small Deviations are Quality Losses
Lower
Specification
Limit
Target
value
Quality
Loss
High
Low
Performance
Metric Target
value
Quality
Loss
Performance
Metric, x
Upper
Specification
Limit
Traditional view of Quality loss Taguchi’s view of Quality loss
Performance
Metric
utdallas.edu/~metin53
• Double-checking
• Fool-proofing, Poka yoke (see Toyota)
• Computer plugs
• Set the watch 5 mins ahead
• Process recipe
• Recipes and checklists help standardize
Accommodate Residual (Common) Variability
Through Robust Design
• Team : Are all the nurse and surgeons present
• Learn names of the team
• Surgical tools clean
• Sponges new
• ….
• ….
Book to Read: The Checklist Manifesto:
How to Get Things Right
By Atul Gawande, Metropolitan Books
utdallas.edu/~metin54
Materials
MachinesSpecifications /
information
People
Vise position
set incorrectly
Machine tool
coordinates set incorrectly
Vice position shifted
during production Part clamping
surfaces corrupted
Part incorrectly
positioned in clamp
Clamping force too
high or too low
Cutting
tool worn
Dimensions incorrectly
specified in drawing
Dimension incorrectly coded
In machine tool program
Material
too soft
Extrusion stock
undersized
Extrusion die
undersized
Extrusion
rate
too high
Extrusion temperature
too high
Error in
measuring height
Steer support
height deviates
from specification
Ishikawa (Fishbone) Diagram
utdallas.edu/~metin55
Summary
Statistical Process Control
X-bar, R-bar, p charts
Process variability vs. Process specifications
Yields/Reworks and their impact on costs
Just-in-time philosophy
utdallas.edu/~metin56
Jesica Santillan died after a bungled heart-lung transplant in 2003. In an operation
Feb. 7, Jesica was mistakenly given organs of the wrong blood type.
Her blood type was 0 Rh+.
Organs come from A Rh- blood type.
Her body rejected the organs, and a matching transplant about two weeks later
came too late to save her. She died Feb. 22 at Duke University Medical Center.
Line of Causes leading to the mismatch
• On-call surgeon on Feb 7 in charge of pediatric heart transplants,
James Jaggers, did not take home the list of blood typesLater stated, "Unfortunately, in this case, human errors were made during the
process. I hope that we, and others, can learn from this tragic mistake."
• Coordinator initially misspelled Jesica’s name
• Once UNOS (United Network for Organ Sharing) identified Jesica,
no further check on blood type
• Little confidence in information system / data quality
• Pediatric nurse did not double check
• Harvest-surgeon did not know blood type
Process Failure in Healthcare:
The Case of Jesica Santillan
utdallas.edu/~metin57
- As a result of this tragic event, it is clear to us at Duke that we need to have
more robust processes internally and a better understanding of the
responsibilities of all partners involved in the organ procurement process.
William Fulkerson, M.D., CEO of Duke University Hospital.
- We didn’t have enough checks.
Ralph Snyderman, Duke University Hospital
Jesica is not the first death in organ transplantation because of blood type mismatch.
Process Failure in Healthcare:
The Case of Jesica Santillan
utdallas.edu/~metin58
Step 1: Define and map processes
- Jaggers had probably forgotten the list with blood groups 20 times before
- Persons involved in the process did not double-check,
everybody checked sometimes
- Learning is triggered following deaths / process deviations are ignored
Step 2: Reduce variability- quality of data (initially misspelled the name)
Step 3: Robust Design- color coding between patient card / box holding the organ
- information system with no manual work-around
- let the technology help
RFID tagged patients: Tag includes blood type and other info
Electronic medicine box: Alarming for the obsolete medicine
The Three Steps in the Case of Jesica
utdallas.edu/~metin59
1. Management Responsibility
2. Quality System
3. Contract review
4. Design control
5. Document control
6. Purchasing / Supplier evaluation
7. Handling of customer supplied material
8. Products must be traceable
9. Process control
10. Inspection and testing
11. Inspection, Measuring, Test Equipment
12. Records of inspections and tests
13. Control of nonconforming products
14. Corrective action
15. Handling, storage, packaging, delivery
16. Quality records
17. Internal quality audits
18. Training
19. Servicing
20. Statistical techniques
Examples: “The design process shall be planned”,
“production processes shall be defined and planned”
How do you get to a Six Sigma Process?
Do Things Consistently (ISO 9000)
utdallas.edu/~metin60
Zero Inventories
Zero Defects
Flexibility / Zero set-ups
Zero breakdowns
Zero handling / non
value added
Just-in-time Production
• Kanban
• Classical Push
• “Real” Just-in-time
Mixed Production
Set-up reduction
Autonomation
Competence and Training
Continuous Improvement
Quality at the source
Organization MethodsPrinciples
The System of Lean Production
Pardon the French, caricatures are from Citroen.
utdallas.edu/~metin61
• Avoid unnecessary inventory
• To be seen more as an ideal
• To types of (bad) inventory:
a. resulting from defects / rework
b. absence of a smooth process flow
• Remember the other costs of inventory (capital, flow time)
Invento
ry in p
rocess
Buffer argument:
“Increase inventory”
Toyota argument:
“Decrease inventory”
Principles of Lean Production:
Zero Inventory and Zero Defects
utdallas.edu/~metin62
71
2345
68
ITAT=7*1 minute
3
1
2
4
ITAT=2*1 minute
Good unit
Defective unit
ITAT: Information Turnaround Time
utdallas.edu/~metin63
• Flexible machines with short set-ups
• Allows production in small lots
• Real time with demand
• Large variety
• Maximize uptime
• Without inventory, any breakdown
will put production to an end
• preventive maintenance
Avoid Non-value-added activities,
specifically rework and set-ups
Principles of Lean Production:
Zero Set-ups, Zero NVA and Zero Breakdowns
utdallas.edu/~metin64
Push: make to forecast Pull: Synchronized production
Pull: Kanban
• Visual way to implement a pull system
• Amount of WIP is determined by
number of cards
• Kanban = Sign board
• Work needs to be authorized by demand
• Classical MRP way
• Based on forecasts
• Push, not pull
• Still applicable for
low cost parts
• Part produced for specific
order (at supplier)
• shipped right to assembly
• real-time synchronization
• for large parts (seat)
• inspected at source
Methods of Lean Production: Just-in-time
utdallas.edu/~metin65
InventoryInventoryInventory
CycleInventory
Production with large batches
End of
Month
Beginning of
Month
Cycle
Production with large batches
End of
Month
Beginning of
Month
Cycle
Production with large batches
End of
Month
Beginning of
Month
Cycle
Production with large batches
End of
Month
Beginning of
Month
Inventory
End of
Month
Beginning of
Month
Produce Sedan
Produce Station wagon
End of
Month
Beginning of
Month
Produce Sedan
Produce Station wagon
End of
Month
Beginning of
Month
Produce Sedan
Produce Station wagon
End of
Month
Beginning of
Month
Production with small batches
Methods of Lean Production:
Mixed Production and Set-up reduction
utdallas.edu/~metin66
• Create local decision making rather
than pure focus on execution
• Use machines / tools, but avoid the
lights-off factory
• Automation with a human touch
• Cross training of workers
• Develop problem solving skills
Organization of Lean Production:
Autonomation and Training
utdallas.edu/~metin67
• Solve the problems where they occur
- this is where the knowledge is
- this is the cheapest place
• Traditional: inspect and rework
at the end of the process
• Once problem is detected, send
alarm and potentially stop
the production
Own Process Next Process End of Line FinalInspection
End User
$ $ $ $ $
• Rework
• Reschedule
• very minor • minor
delay
• Significant
Rework
• Delayed
delivery
• Overhead
• Warranty
cost
• recalls
• reputation
• overhead
Defect found
Defect fixed
Organization of Lean Production: Continuous Improvement and Quality-at-the-source