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. 1 utdallas.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

2 utdallas.edu/~metin Steer Support for the Scooter

3 utdallas.edu/~metin Steer Support Specifications

Go-no-go gauge

4 utdallas.edu/~metin Control Charts

79.98

79.97

79.96

79.95

79.94

X-bar 79.93

79.92

79.91

79.9 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

0.09

0.08

0.07

0.06

0.05

R 0.04

0.03

0.02

0.01

0 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 5 utdallas.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

» Apply this to Global Warming or lack of it, or to Temperature Volatility.

6 utdallas.edu/~metin The Concept of Consistency: Who is the Better Target Shooter?

Not just the mean is important, but also the variance

Need to look at the distribution function

7 utdallas.edu/~metin Statistical Process Control

Capability Conformance Analysis Analysis

Eliminate Investigate for Assignable Cause 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 8 utdallas.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

9 utdallas.edu/~metin Two Types of Causes for Variation

Common Cause Variation (low level)

Common Cause Variation (high level)

Assignable Cause Variation

10 utdallas.edu/~metin Mean and Variance

 Given a population of numbers, how to compute the mean and the variance?

Population  {x1, x2 ,..., xN } N  xi Mean    i1 N N 2 (xi  ) Variance   2  i1 N Standard deviation   11 utdallas.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= x » Sample range= R » Standard deviation of sample means=

   where  :Standard deviation of the population x n

12 utdallas.edu/~metin Sampling Distribution

Sampling distribution is the distribution of sample means.

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 Grouping reduces the variability. 13 utdallas.edu/~metin Normal Distribution

normdist(x,.,.,1) normdist(x,.,.,0)

Probab

  Mean x  

95.44%

99.74%

Excel statistical functions : normdist(x,mean, st _ dev,0) normal pdf at x. Excel statistical functions : normdist(x,mean, st _ dev,1) normal cdf at x.

14 utdallas.edu/~metin Cumulative Normal Density

1 prob

normdist(x,mean,st_dev,1)

0 x norminv(prob,mean,st_dev)

Excel statistical functions : Cumulative function (cdf) at x : normdist(x,mean, st _ dev,1) Inverse function of cdf at "prob": norminv( prob,mean, st _ dev) 15 utdallas.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)

16 utdallas.edu/~metin Control Limits for Consistency not Correctness

Process is in control if sample mean is between control limits. These limits have nothing to do with product specifications! Sampling distribution

Process distribution

Mean

LCL UCL Lower Upper control control 17 utdallas.edu/~metin limit limit 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.

18 utdallas.edu/~metin Type I Error=Alpha

Sampling distribution

/2 /2

Mean

Probability LCL UCL of Type I error LCL  norminv(α/ 2,mean, st_dev) UCL  norminv(1-α / 2,mean, st_dev)

The textbook uses Type I error=1-99.74%=0.0026=0.26%. 19 utdallas.edu/~metin Statistical Process Control: Control Charts

Process Parameter • Track process parameter over time - mean Upper Control Limit (UCL) - percentage defects

• Distinguish between Center Line - common cause variation (within control limits) - assignable cause variation (outside control limits) Lower Control Limit (LCL) • Measure process performance: how much common cause variation Time is in the process while the process is “in control”?

20 utdallas.edu/~metin Control Chart

Abnormal variation Out of due to assignable sources control UCL

Mean Normal variation due to chance LCL Abnormal variation due to assignable sources

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Sample number

21 utdallas.edu/~metin Observations from Sample Distribution

UCL

LCL

1 2 3 4 Sample number

22 utdallas.edu/~metin Parameters for computing UCL and LCL the Table method

Number of Factor for X- Factor for Factor for Factor to Observations bar Chart Lower Upper estimate in Sample (A2) control Limit control limit Standard Sample size (n) in R chart in R chart deviation, (d2) (D3) (D4) 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

23 utdallas.edu/~metin The X-bar Chart: Application to Call Center

Period x1 x2 x3 x4 x5 Mean Range • Collect samples over time 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 • Compute the mean: 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 x1  x2  ...  xn 6 2.8 3.6 4.5 5.2 2.1 3.64 3.1 X  7 3.9 2.8 3.5 3.5 3.1 3.36 1.1 n 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 • Compute the range: 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 R  max{x1, x2 ,...xn} 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  min{x , x ,...x } 16 3.6 4.3 2.1 5.2 1.3 3.3 3.9 1 2 n 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 as a proxy for the variance 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 • Average across all periods 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 - average of means, X 24 2.3 1.6 1.8 5 1.5 2.44 3.5 - average of ranges, R 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 • Normally distributed Average 24 utdallas.edu/~metin 3.81 5.85

Control Charts: The X-bar Chart The Table method • Define control limits

12 UCL=X +A2 ×R =3.81+0.58*5.85=7.19 X R 10 LCL= -A2 × =3.81-0.58*5.85=0.41 0.58 comes from the Table. 8 •Constants are taken from a table 6

4 • Identify assignable causes: - point over UCL 2 - point below LCL

0 - many (6) points on one side of center 1 3 5 7 9 11 13 15 17 19 21 23 25 27 • In this case: - problems in period 13 - new operator (CSR) was assigned

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 25 utdallas.edu/~metin Range Control Chart

UCL  D4R  A multiple of the average of sample ranges

LCL  D3R  A multiple of the average of sample ranges

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?

R  (88  64 102  70  74) / 5  79.6. For n  6, D4  2, D3  0.

UCL  D4R  2*79.6 159.2 LCL  D3R  0*79.6  0 Are the GMAT ranges in control?

26 utdallas.edu/~metin Control Charts: X-bar Chart and R-bar Chart For the Call Center

12

10

8

6

Bar - X 4

2

0 1 3 5 7 9 11 13 15 17 19 21 23 25 27

30

25

20

15 R

10

5

0 1 3 5 7 9 11 13 15 17 19 21 23 25 27

27 utdallas.edu/~metin X-bar and Range Charts: Which?

(process mean is shifting upward) Sampling Distribution

UCL

x-Chart Detects shift

LCL UCL

Does not R-chart detect shift

LCL

28 utdallas.edu/~metin X-bar and Range Charts: Which?

Sampling Distribution (process variability is increasing)

UCL

x-Chart Does not reveal increase

LC LUCL

R-chart Reveals increase

LC L 29 utdallas.edu/~metin Control Charts: The X-bar Chart The Direct method

• 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   0.0026 LCL  norminv(/2, mean, st_dev)  norminv(0.0013,3.81,1.5687)  -0.91 UCL  norminv(1-/2, mean, st_dev)  norminv(0.9987,3.81,1.5687)  8.53

Recall LCL=0.41 and UCL=7.19 in the Table method. 30 utdallas.edu/~metin Process Capability Let us Tie Tolerances and Variability

 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 – Process variability relative to specifications

Capability = Process specifications / Process variability

31 utdallas.edu/~metin Process Capability: Specification limits are not control chart limits

Lower Upper Specification Specification Sampling Distribution is used Process variability matches specifications Lower Upper Specification Specification

Process variability well within Lower Upper specifications Specification Specification

Process variability exceeds 32 utdallas.edu/~metin specifications Process Capability Ratio

When the process is centered, process capability ratio Upper specification level - Lower specification level Cp  6 X

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

33 utdallas.edu/~metin 3 Sigma and 6 Sigma Quality

Lower Upper specification specification

Process mean

+/- 3 Sigma

+/- 6 Sigma Can you reduce variability so much that 12 Sigmas fit within specification limits?

34 utdallas.edu/~metin The Statistical Meaning of Six Sigma

Lower Upper Specification (LSL) Specification (USL)

L Process A e s (with st. dev A) C P{defect} s p V a 1 0.33 0.317=31.7% r  X-3A X-2A X-1A X X+1A X+2 X+3A i 2 0.67 0.0455 3a  b 3 1.00 0.0027=0.27% i  l 4 1.33 0.0001 Process B i (with st. dev B) t 5 1.67 0.0000006 y  6 2.00 2x10-9

X-6B X X+6B

• Estimate standard deviation:  = R/d2 • Or use the direct method with the excel function stdev() • Look at standard deviation relative to specification limits 35 utdallas.edu/~metin 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

36 utdallas.edu/~metin Attribute Based Control Charts: The p-chart Period n defects p 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 • Estimate average defect 7 300 16 0.053 p =0.052 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 • Estimate Standard Deviation 14 300 13 0.043 15 300 13 0.043 p(1 p) 16 300 13 0.043 =0.013 17 300 17 0.057 ˆ = Sample Size 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 • Define control limits UCL= + 3 ˆ =0.014 25 300 10 0.033 26 300 12 0.040 LCL= - 3 ˆ =0.091 27 300 13 0.043 28 300 18 0.060 29 300 19 0.063 30 300 14 0.047 37 utdallas.edu/~metinThink of printing defective pages Attribute Based Control Charts: The p-chart

0.180 0.160 0.140 0.120 0.100 0.080 0.060 0.040 0.020 0.000 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

38 utdallas.edu/~metin Inspection

 Where/When » Raw materials » Finished products

Inputs Transformation Outputs

Acceptance Process Acceptance sampling control sampling

» 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 utdallas.edu/~metin 39 Discovery of Defects and the Costs

End of Process Bottleneck Process Market

Defect occurred Defect Defect Defect detected detected detected

$ $ Cost of $ defect Based on labor and Based on sales Recall, reputation, material cost price (incl. Margin) warranty costs

CPSC, Segway LLC Announce Voluntary Recall to Recall Alert Upgrade Software on Segway™ Human Transporters U.S. Consumer Product Safety Commission The following product safety recall was conducted by the firm in Office of Information and Public Affairs cooperation with the CPSC. Washington, DC 20207, September 26, 2003 Name of Product: Segway Human Transporter (HT) Units: Approximately 6,000 utdallas.edu/~metin 40 Examples of Inspection Points

Type of Inspection Characteristics business points Fast Food Cashier Accuracy Counter area Appearance, productivity Eating area Cleanliness Building Appearance Kitchen Health regulations Hotel/motel Parking lot Safe, well lighted Accounting Accuracy, timeliness Building Appearance, safety Main desk Waiting times Supermarket Cashiers Accuracy, courtesy Deliveries Quality, quantity 41 utdallas.edu/~metin The Concept of Yields

Flow rate of units processed correctly at the resource Yield of Resource = Flow rate

Flow rate of units processed correctly at all of the resources Yield of Process = Flow rate

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?

42 utdallas.edu/~metin Rework / Elimination of Flow Units

Step 1 Test 1 Step 2 Test 2 Step 3 Test 3 Rework: Defects can be corrected Rework Same or other resource Leads to variability Examples: Step 1 Test 1 Step 2 Test 2 Step 3 Test 3 - Readmission to Intensive Care Unit

Rework Rework Rework

Loss of Flow units: Defects can NOT be corrected Leads to variability Step 1 Test 1 Step 2 Test 2 Step 3 Test 3 To get X, we have to start X/y units Examples: - Interviewing - Semiconductor fab utdallas.edu/~metin 43 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) “Bad” outcome only happens 1 problem every Every 100*10,000 units 10,000 units That is probability of (0.01)3 99% correct

Case 2: Process has built-in rework loops • Double-checking

1 problem 99% Good every 1 unit “Bad” outcome happens 99% 99% with probability (1-0.99)3, That is (0.01) 3. Indepen- Indepen- 1% 1% dence from dence from Bad the first the first two 1%

You can add rework loops to improve quality. 44 utdallas.edu/~metin If checks are dependent, this is less effective. Rare events are not so rare: Jetliner Crash due to Engine Icing

May 11, 2011, the Emirates A-380 from Dubai was struck by lightning as it approached Heathrow airport  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) widebody jetliner 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. utdallas.edu/~metin 45 Rare events are not so rare: An Earthquake and a Power Loss

Pacific Ocean Squaw Creek Reservoir for cooling water for cooling water 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 utdallas.edu/~metin that cut off the power to the reactor which was idled for a while (>2 weeks) afterwards. 46 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

47 utdallas.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 48 utdallas.edu/~metin Push/Pull View of Supply Chains

Procurement, Customer Order Manufacturing and Cycle Replenishment cycles

PUSH PROCESSES PULL PROCESSES

Customer Order Arrives Push-Pull boundary 49 utdallas.edu/~metin Pull Process with Kanban Cards

Direction of production flow

upstream downstream

Authorize production of next unit

50 utdallas.edu/~metin Pareto Principle or 20-80 rule Errors in the shipping process

Absolute Cause of Defect 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 110 100 Number of Cumulative defects percents of 100 defects 75

50 50

25

Browser error

Order entry Order mistake

Wrongmodel shipped

Order number Order out offsequence billingaddress Product shipped Product to 51

utdallas.edu/~metin

Product shipped, Product but not card credit billed Reduce Variability in the Process Taguchi: Even Small Deviations are Quality Losses

Traditional view of Quality loss Taguchi’s view of Quality loss

Quality Quality Loss Loss Performance Metric, x

High

Low Upper Performance Performance Lower Target Target Specification Metric Metric Specification value value Limit Limit

•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”

52 utdallas.edu/~metin Accommodate Residual (Common) Variability Through Robust Design

• Double-checking • Fool-proofing, Poka yoke (see Toyota) • Computer plugs • Set the watch 5 mins ahead • Process recipe • Recipes and checklists help standardize

• 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

53 utdallas.edu/~metin Ishikawa (Fishbone) Diagram

Specifications / information Machines

Cutting tool worn Vise position set incorrectly Dimensions incorrectly Clamping force too specified in drawing high or too low Machine tool coordinates set incorrectly Part incorrectly Dimension incorrectly coded positioned in clamp Vice position shifted In machine tool program during production Part clamping surfaces corrupted Steer support height deviates Extrusion temperature from specification too high

Extrusion stock Error in undersized measuring height Extrusion Extrusion die rate Material undersized too high too soft

People Materials

54 utdallas.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

55 utdallas.edu/~metin Process Failure in Healthcare: The Case of Jesica Santillan

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 types Later 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 56 utdallas.edu/~metin Process Failure in Healthcare: The Case of Jesica Santillan

- We didn’t have enough checks. Ralph Snyderman, Duke University Hospital

- 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.

Jesica is not the first death in organ transplantation because of blood type mismatch.

57 utdallas.edu/~metin The Three Steps in the Case of Jesica

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

58 utdallas.edu/~metin How do you get to a Six Sigma Process? Do Things Consistently (ISO 9000)

1. Management Responsibility 11. Inspection, Measuring, Test Equipment 2. Quality System 12. Records of inspections and tests 3. Contract review 13. Control of nonconforming products 4. Design control 14. Corrective action 5. Document control 15. Handling, storage, packaging, delivery 6. Purchasing / Supplier evaluation 16. Quality records 7. Handling of customer supplied material 17. Internal quality audits 8. Products must be traceable 18. Training 9. Process control 19. Servicing 10. Inspection and testing 20. Statistical techniques

Examples: “The design process shall be planned”, “production processes shall be defined and planned”

59 utdallas.edu/~metin The System of Lean Production

Principles Organization Methods

Zero Inventories Autonomation Just-in-time Production Zero Defects Competence and Training • Kanban Flexibility / Zero set-ups Continuous Improvement • Classical Push Zero breakdowns Quality at the source • “Real” Just-in-time Zero handling / non Mixed Production value added Set-up reduction

Pardon the French, caricatures are from Citroen. 60 utdallas.edu/~metin Principles of Lean Production: Zero Inventory and Zero Defects

Buffer argument:

“Increase inventory” Inventory in process in Inventory Toyota argument: “Decrease inventory” • 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)

61 utdallas.edu/~metin ITAT: Information Turnaround Time

7 6 8 1 Defective unit 5 4 3 2

Good unit

ITAT=7*1 minute

4 1

3 2

ITAT=2*1 minute

62 utdallas.edu/~metin Principles of Lean Production: Zero Set-ups, Zero NVA and Zero Breakdowns

Avoid Non-value-added activities, specifically rework and set-ups

• Maximize uptime • Flexible machines with short set-ups • Without inventory, any breakdown • Allows production in small lots will put production to an end • Real time with demand • preventive maintenance • Large variety

63 utdallas.edu/~metin Methods of Lean Production: Just-in-time

Push: make to forecast Pull: Synchronized production

• Classical MRP way • Part produced for specific • Based on forecasts order (at supplier) • Push, not pull • shipped right to assembly • Still applicable for • real-time synchronization low cost parts • for large parts (seat) • inspected at source

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 64 utdallas.edu/~metin Methods of Lean Production: Mixed Production and Set-up reduction

Production with large batches

Cycle Inventory Cycle

Inventory

Beginning of End of Month Month

Production with small batches

Produce Sedan

Produce Station wagon

Month Month 65 utdallas.edu/~metin Beginning of End of Organization of Lean Production: Autonomation and Training

• Automation with a human touch

• Create local decision making rather than pure focus on execution • Use machines / tools, but avoid the • Cross training of workers lights-off factory • Develop problem solving skills

66 utdallas.edu/~metin Organization of Lean Production: Continuous Improvement and Quality-at-the-source

• Solve the problems where they occur - this is where the knowledge is - this is the cheapest place

Defect found Own Process Next Process End of Line Final End User $ $ $ $ Inspection $

• very minor • minor • Rework • Significant • Warranty delay • Reschedule Rework cost • Delayed • recalls Defect fixed delivery • reputation • Overhead • overhead

• Traditional: inspect and rework at the end of the process

• Once problem is detected, send alarm and potentially stop the production

67 utdallas.edu/~metin