1utdallas.edu/~metin
Quality Management
Chapter 8
2utdallas.edu/~metin
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
3utdallas.edu/~metin
Steer Support for the Scooter
4utdallas.edu/~metin
Steer Support Specifications
Go-no-go gauge
5utdallas.edu/~metin
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
6utdallas.edu/~metin
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
7utdallas.edu/~metin
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?
8utdallas.edu/~metin
Statistical Process Control
CapabilityAnalysis
ConformanceAnalysis
Investigate forAssignable Cause
EliminateAssignable 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 occurredInvestigate for assignable cause• Find “Root Cause(s)” of Potential Loss of Statistical ControlEliminate assignable cause• Need Corrective Action To Move Forward
9utdallas.edu/~metin
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
10utdallas.edu/~metin
Common Cause Variation (low level)
Common Cause Variation (high level)
Assignable Cause Variation
Two Types of Causes for Variation
11utdallas.edu/~metin
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
ii
N
ii
N
12utdallas.edu/~metin
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
13utdallas.edu/~metin
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.
14utdallas.edu/~metin
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)
15utdallas.edu/~metin
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)
16utdallas.edu/~metin
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)
17utdallas.edu/~metin
Control Limits
Samplingdistribution
Processdistribution
Mean
LCLLowercontrol
limit
UCLUppercontrol
limit
Process is in control if sample mean is between control limits. These limits have nothing to do with product specifications!
18utdallas.edu/~metin
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(out of control when in control) Beta=P(Type II error)=P(accept the null when it is false)
P(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 a desired level of Alpha.
19utdallas.edu/~metin
Type I Error=Alpha
Mean
LCL UCL
/2 /2
Probabilityof Type I error
st_dev)mean,/2,-norminv(1UCL
st_dev)mean,/2,norminv(LCL
The textbook uses Type I error=1-99.74%=0.0026=0.26%.
Sampling distribution
20utdallas.edu/~metin
Time
ProcessParameter
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
21utdallas.edu/~metin
Control Chart
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
UCL
LCL
Sample number
Mean
Out ofcontrol
Normal variationdue to chance
Abnormal variationdue to assignable sources
Abnormal variationdue to assignable sources
22utdallas.edu/~metin
Observations from Sample Distribution
Sample number
UCL
LCL
1 2 3 4
23utdallas.edu/~metin
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 LCLthe Table method
24utdallas.edu/~metin
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 mean - average range
• Normally distributed
The X-bar Chart: Application to Call Center
25utdallas.edu/~metin
• Define control limits
• 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 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 ChartThe Table method
26utdallas.edu/~metin
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 Table 8.2.
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?
27utdallas.edu/~metin
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 ChartFor the Call Center
28utdallas.edu/~metin
X-bar and Range Charts: Which?
UCL
LCL
UCL
LCL
R-chart
x-Chart Detects shift
Does notdetect shift
(process mean is shifting upward)
SamplingDistribution
29utdallas.edu/~metin
X-bar and Range Charts: Which?
UCL
LCL
LCL
R-chart Reveals increase
x-Chart
UCL
Does notreveal increase
(process variability is increasing)SamplingDistribution
30utdallas.edu/~metin
• 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 ChartThe 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
31utdallas.edu/~metin
Tolerances/Specifications– Requirements of the design or customers
Process variability– Natural variability in a process
– Variance of the measurements coming from the process
Process capability– Process variability relative to specification
– Capability=Process specifications / Process variability
Process CapabilityLet us Tie Tolerances and Variability
32utdallas.edu/~metin
Process Capability: Specification limits are not control chart limits
LowerSpecification
UpperSpecification
Process variability matches specifications
LowerSpecification
UpperSpecification
Process variability well within specifications
LowerSpecification
UpperSpecification
Process variability exceeds specifications
Sampling Distribution
is used
33utdallas.edu/~metin
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
34utdallas.edu/~metin
Processmean
Lowerspecification
Upperspecification
+/- 3 Sigma
+/- 6 Sigma
3 Sigma and 6 Sigma Quality
35utdallas.edu/~metin
• Estimate standard deviation:• Or use the direct method with the excel function stdev()• Look at standard deviation relative to specification limits
= R / d 2
3
Upper Specification (USL)
LowerSpecification (LSL)
X-3A X-2A X-1AX X+1A
X+2 X+3A
X-6BX X+6B
Process A(with st. dev A)
Process B(with st. dev B)
x Cp P{defect}
1 0.33 0.317
2 0.67 0.0455
3 1.00 0.0027
4 1.33 0.0001
5 1.67 0.0000006
6 2.00 2x10-9
The Statistical Meaning of Six Sigma
36utdallas.edu/~metin
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
37utdallas.edu/~metin
• Estimate average defect percentage
• Estimate Standard Deviation
• Define control limits
1 300 18 0.0602 300 15 0.0503 300 18 0.0604 300 6 0.0205 300 20 0.0676 300 16 0.0537 300 16 0.0538 300 19 0.0639 300 20 0.067
10 300 16 0.05311 300 10 0.03312 300 14 0.04713 300 21 0.07014 300 13 0.04315 300 13 0.04316 300 13 0.04317 300 17 0.05718 300 17 0.05719 300 21 0.07020 300 18 0.06021 300 16 0.05322 300 14 0.04723 300 33 0.11024 300 46 0.15325 300 10 0.03326 300 12 0.04027 300 13 0.04328 300 18 0.06029 300 19 0.06330 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
38utdallas.edu/~metin
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
39utdallas.edu/~metin
Inspection
Where/When » Raw materials» Finished products
» Before a costly operation, PhD comp. exam before candidacy
» Before an irreversible process, firing pottery
» Before a covering process, painting, assembly
Centralized vs. On-Site, my friend checks quality at cruise lines
Inputs Transformation Outputs
Acceptancesampling
Processcontrol
Acceptancesampling
40utdallas.edu/~metin
ProcessStep
Bottleneck
Based on labor andmaterial cost
MarketEnd ofProcess
Defectdetected
Defectoccurred Defect
detectedDefectdetected
Cost of defect
$$ $
Based on salesprice (incl. Margin)
Recall, reputation,warranty costs
Defectdetected
Discovery of Defects and the Costs
CPSC, Segway LLC Announce Voluntary Recall to Upgrade Software on Segway™ Human TransportersThe 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 Alert
U.S. Consumer Product Safety CommissionOffice of Information and Public AffairsWashington, DC 20207September 26, 2003
41utdallas.edu/~metin
Examples of Inspection Points
Type ofbusiness
Inspectionpoints
Characteristics
Fast Food CashierCounter areaEating areaBuildingKitchen
AccuracyAppearance, productivityCleanlinessAppearanceHealth regulations
Hotel/motel Parking lotAccountingBuildingMain desk
Safe, well lightedAccuracy, timelinessAppearance, safetyWaiting times
Supermarket CashiersDeliveries
Accuracy, courtesyQuality, quantity
42utdallas.edu/~metin
The Concept of Yields
90% 80% 90% 100% 90%
Line Yield: 0.9 x 0.8 x 0.9 x 1 x 0.9
Yield of Resource = rate Flow
resource the atcorrectly processed units of rate Flow
Yield of Process = rate Flow
correctly processed units of rate Flow
43utdallas.edu/~metin
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 units, we have to start X/y units Examples: - Interviewing - Semiconductor fab
44utdallas.edu/~metin
Why Having a Process is so Important:Two Examples of Rare-Event Failures
Case 1: Process does not matter in most cases• 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
“Bad” outcome happens with probability (1-0.99)3
Good
Bad
99% 99%
99%
1%
1% 1%
Learning should be driven by process deviations, not by defects
“Bad” outcome only happens Every 100*10,000 units
45utdallas.edu/~metin
Rare events are not so rare:Chances of a Jetliner Crash due to Engine Icing
Engine flameout due to crystalline icing: Engine stops for 30-90 secs and hopefully starts again.
Suppose 150 single engine flameouts over 1990-
2005 and 15 dual engine flameouts over 2002-2005. What are the annualized single and dual 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=700 widebody jetliner and storm encounter very year. How can we explain M=700 > N=20? The engines do not ice independently. With M=700, Prob(1 engine icing)=10/700=1.42% and Prob(2 engine icing)=5/700=0.71%. 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, etc.
46utdallas.edu/~metin
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
47utdallas.edu/~metin
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
48utdallas.edu/~metin
Push/Pull View of Supply Chains
Procurement,Manufacturing andReplenishment cycles
Customer OrderCycle
CustomerOrder ArrivesPush-Pull boundary
PUSH PROCESSES PULL PROCESSES
49utdallas.edu/~metin
Kanban
Direction of production flow
upstream downstream
Kanban
Kanban
Kanban
Authorize productionof next unit
Pull Process with Kanban Cards
50utdallas.edu/~metin
Bro
wse
rer
ror
Number ofdefects
Cause of DefectAbsolute 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 o
ut
off
seq
uen
ce
Pro
du
ct s
hip
ped
, bu
tcr
edit
car
d n
ot
bill
ed
Ord
er e
ntr
ym
ista
ke
Pro
du
ct s
hip
ped
to
bill
ing
ad
dre
ss
Wro
ng
mo
del
ship
ped
100
50
Cumulativepercents ofdefects
100
75
50
25
Pareto Principle or 20-80 rule
51utdallas.edu/~metin
•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 ProcessTaguchi: Even Small Deviations are Quality Losses
LowerSpecification Limit
Target value
QualityLoss
High
LowPerformance Metric Target
value
QualityLoss
Performance Metric, x
UpperSpecification Limit
Traditional view of Quality loss Taguchi’s view of Quality loss
Performance Metric
52utdallas.edu/~metin
• Double-checking (see Toshiba)• Fool-proofing, Poka yoke (see Toyota)
• Computer plugs• Set the watch 5 mins ahead
• Process recipe (see Brownie)• Recipes help standardize
Accommodate Residual (Common) Variability Through Robust Design
53utdallas.edu/~metin
Materials
MachinesSpecifications /information
People
Vise positionset incorrectly
Machine toolcoordinates set incorrectly
Vice position shiftedduring production
Part clampingsurfaces corrupted
Part incorrectlypositioned in clamp
Clamping force toohigh or too low
Cutting tool worn
Dimensions incorrectlyspecified in drawing
Dimension incorrectly coded In machine tool program
Material too soft
Extrusion stock undersized
Extrusion dieundersized
Extrusion ratetoo high
Extrusion temperaturetoo high
Error in measuring height
Steer support height deviates from specification
Ishikawa (Fishbone) Diagram
54utdallas.edu/~metin
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
55utdallas.edu/~metin
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
56utdallas.edu/~metin
- 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
57utdallas.edu/~metin
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 infoElectronic medicine box: Alarming for the obsolete medicine
The Three Steps in the Case of Jesica
58utdallas.edu/~metin
1. Management Responsibility2. Quality System3. Contract review4. Design control5. Document control6. Purchasing / Supplier evaluation7. Handling of customer supplied material8. Products must be traceable9. Process control10. Inspection and testing
11. Inspection, Measuring, Test Equipment12. Records of inspections and tests13. Control of nonconforming products14. Corrective action15. Handling, storage, packaging, delivery16. Quality records17. Internal quality audits18. Training19. Servicing20. 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)
59utdallas.edu/~metin
Zero InventoriesZero DefectsFlexibility / Zero set-upsZero breakdownsZero handling / non value added
Just-in-time Production• Kanban• Classical Push• “Real” Just-in-timeMixed ProductionSet-up reduction
AutonomationCompetence and TrainingContinuous ImprovementQuality at the source
Organization MethodsPrinciples
The System of Lean Production
Pardon the French, caricatures are from Citroen.
60utdallas.edu/~metin
• 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)
Inve
ntor
y in
pro
cess
Buffer argument:“Increase inventory”
Toyota argument:“Decrease inventory”
Principles of Lean Production: Zero Inventory and Zero Defects
61utdallas.edu/~metin
71
2345
68
ITAT=7*1 minute
3
1
2
4
ITAT=2*1 minute
Good unit
Defective unit
ITAT: Information Turnaround Time
62utdallas.edu/~metin
• 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
63utdallas.edu/~metin
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
64utdallas.edu/~metin
InventoryInventoryInventory
CycleInventory
Production with large batches
End ofMonth
Beginning ofMonth
Cycle
Production with large batches
End ofMonth
Beginning ofMonth
Cycle
Production with large batches
End ofMonth
Beginning ofMonth
Cycle
Production with large batches
End ofMonth
Beginning ofMonth
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
65utdallas.edu/~metin
• 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
66utdallas.edu/~metin
• 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