Part II S /IEE Measure Phase from DMAICweb.eng.fiu.edu/leet/TQM/chap3_2012.pdf · 2014. 12. 17. ·...
Transcript of Part II S /IEE Measure Phase from DMAICweb.eng.fiu.edu/leet/TQM/chap3_2012.pdf · 2014. 12. 17. ·...
9/4/2012
1
Part II
S4/IEE Measure Phase
from DMAIC
Introduction
• Part II (Chapter 3-14) addresses process definition,
process performance, and the quantification of
variability.
• KPOVs and KPIVs are identified through consensus.
• Basic analysis tools are introduced:
• Six Sigma measures
• Measurement systems analysis (MSA)
• Failure mode and effects analysis (FMEA)
• Quality function deployment (QFD)
9/4/2012
2
Introduction
5 Components in S4/IEE Measure Phase:
• Plan project and metrics
• Baseline project
• Consider lean tools
• MSA
• Wisdom of the organization
Chapter 3
Measurements and
S4/IEE Measure Phase
9/4/2012
3
Introduction
• An objective of the measure phase is the development
of a reliable and valid measurement system of the
business process identified in the define phase. • How an inappropriate tracking methodology can lead to
the wrong activity (firefighting);
• Overview of basic descriptive statistics;
• Data gathering, presentation, and simple statistics;
• Introductory discussion of confidence interval and
hypothesis testing;
• Attribute vs. continuous data;
• Ineffectiveness of visual inspections;
• Examples of experiment traps.
3.1 Voice of the Customer
• VOC assessment is needed up front when executing
S4/IEE projects at the 30,000-foot level.
• Define your customer.
• Obtain customer wants, needs, and desires.
• Ensure that focus of project is addressing customer needs.
9/4/2012
4
3.1 Voice of the Customer
• Important customer key process output categories are
classified with regard to their area of impact: • Critical to quality (CTQ): flatness, diameter, etc.
• Critical to delivery (CTD): on-time, accuracy, etc.
• Critical to cost (CTC)
• Critical to satisfaction (CTS)
• Important key process input issues are classified as critical
to process (CTP)
• The format of a tree diagram can be useful to ensure a
linkage between customer requirements and process
performance metrics. NeeddriversCTQs
• Tree diagram can also be used to describe the hierarchy of
critical to (CT) categories at various level.
3.1 Voice of the Customer:
Critical to Satisfaction (CTS) Tree
Cleveland Service Shop Generates Most Defects with a
Failure Rate of 89% Late Deliveries
Sources of
Variation
9/4/2012
5
3.2 Variability and
Process Improvements
• Variability is everywhere.
• Common cause vs. special cause
• If variability is too large because of special causes,
attempts need to be made to avoid the source of special
cause.
• If variability is too large because of common causes, the
process needs to be changed.
3.3 Common vs. Special Causes
and Chronic vs. Sporadic problems
• J. M. Juran (Juran and Gryna 1980) considers the
corrective action strategy for sporadic and chronic
problems.
• W. Edwards Deming addresses this basic issue using a
nomenclature of special causes and common causes
(Deming 1986).
• Process control charts are tools that can be used to
distinguish these two types of situations.
• Sporadic problems are defined as unexpected changes in
the normal operating level of a process.
• Chronic problems exist when the process is at a long-term
unacceptable operating level.
9/4/2012
6
3.3 Common vs. Special Causes
and Chronic vs. Sporadic problems
• With sporadic problems the corrective action is to bring the
process back to the normal operating level.
• The solution to chronic problems is a change in the normal
operating level of the process.
• Solving these two types of problems involves different
basic approaches.
3.3 Common vs. Special Causes
and Chronic vs. Sporadic problems
The Juran’s control sequence (Juran and Gryna 1980) is
basically a feed-back loop that involves the following steps:
1. Choose the control subject (i.e., what we intend to
regulate).
2. Choose a unit of measure.
3. Set a standard or goal for the control subject.
4. Choose a sensing device that can measure the control
subject in terms of unit of measure.
5. Measure actual performance.
6. Interpret the difference between the actual and standard.
7. Take action, if any is needed, on the difference.
9/4/2012
7
3.3 Common vs. Special Causes
and Chronic vs. Sporadic problems
Breakthrough sequence for solving chronic problems 1. Convince those responsible that a change in quality level is desirable
and feasible.
2. Identify the vital few projects; that is, determine which quality problem
areas are most important.
3. Organize for breakthrough in knowledge; that is, define the
organization mechanisms for obtaining missing knowledge.
4. Conduct the analysis; that is, collect and analyze the facts that are
required and recommend the action needed.
5. Determine the effect of proposed changes on the people involved and
find ways to overcome the resistance to change.
6. Take action to institute the changes.
7. Institute controls to hold the new level.
3.4 Reacting to Data
76.2
78.2
74.1
74.1
75.0
74.5
75.0
75.0
71.8
76.7
77.8
77.1
75.9
76.3
75.9
77.5
77.0
77.6
77.1
75.2
76.9
78.3
72.7
76.3
78.5
76.0
76.8
73.2
78.8
77.6
75.2
76.8
73.8
75.6
77.7
76.9
76.2
75.1
76.6
76.6
75.1
75.4
73.0
74.6
76.1
79.3
75.9
75.7
77.9
78.0
9/4/2012
8
3.4 Reacting to Data
• Population parameters
• Sample statistics
• Simple Random Sampling
– Random variable (x): a response from sample
taken randomly from population
• Sampling error
• Confidence interval
3.5 Sampling
9/4/2012
9
• Dot Plot
– A simple procedure to illustrate data
positioning and its variability. Along a
numbered line, a dot plot displays a dot for
each observation.
• Histogram
– Discrete Data: Tally for each outcome
– Continuous Data: Frequency within each
class (group, cell)
3.6 Simple Graphic Presentations
3.7 Example 3.2: Histogram
2.2
2.6
3.0
4.3
4.7
5.2
5.2
5.3
5.4
5.7
5.8
5.8
5.9
6.3
6.7
7.1
7.3
7.6
7.6
7.8
7.9
9.3
10.0
10.1
9/4/2012
10
3.7 Example 3.2: Dot Plot
3.8 Sample Statistics (Mean, Range, Std. Deviation, and Median)
• Sample Mean (𝑥 )
– A measure of central tendency.
– It is the arithmetic average of the data values
𝑥 = 𝑥𝑖𝑛𝑖=1
𝑛
• Sample Median (𝑥50)
– A measure of central tendency.
– The number at (n+1)/2 rank
9/4/2012
11
3.8 Sample Statistics (Mean, Range, Std. Deviation, and Median)
• Sample Range (R)
– A measure of dispersion of data.
– It is the difference between the highest and
lowest value of a set of data
– Sample size, outliers
• Sample Standard Deviation (s)
– is the square root of the sample variance.
• Sample Standard Deviation (s)
𝑠 =1
𝑛 − 1 (
𝑛
𝑖=1
𝑋𝑖 − 𝑋 )2 =
1
𝑛 − 1 (
𝑛
𝑖=1
𝑋𝑖2 − 𝑛𝑋 2)
– Degree of freedom n (nu) = n-1
• Sample Variance (s2)
22
3.8 Sample Statistics (Mean, Range, Std. Deviation, and Median)
9/4/2012
12
• Continuous data can assume a range of
numerical responses on a continuous
scale. (e.g., micrometer reading)
• Attribute (categorical or qualitative) data
can assume only discrete levels. (e.g.,
operator, machine, number of defects,
pass/fail)
• Logic pass/fail
23
3.9 Attribute vs. Continuous
Data Response
• Transactional workflow metric (could similarly apply to
manufacturing; e.g., inventory or time to complete a
manufacturing process): Random sample of last year’s
invoices where the number of days beyond the due date
was measured and reported (i.e., days sales outstanding
[DSO]). If an invoice was beyond 30 days late, it was
considered a failure or defective transaction. The
number of nonconformances was divided by the sample
size to estimate the nonconformance rate of the overall
process. • Note that this procedure would not be the recommended strategy within
S4/IEE, since much information is lost in the translation between continuous
and attribute data.
24
3.9 Attribute vs. Continuous
Data Response: Examples
9/4/2012
13
• Transactional quality metric: Random sample of last
year’s invoices, where the number of defects in filling out
the form were measured and reported. The number of
defective transactions was divided by the sample size to
estimate the defective rate of the process.
• Transactional quality metric: Random sample of last
year’s invoices, where the invoices were examined to
determined if there were any errors when filling out the
invoice. Multiple errors or defects could occur on one
invoice. The total number of defects was divided by the
sample size to estimate the defect rate of the process.
25
3.9 Attribute vs. Continuous
Data Response: Examples
• Transactional quality metric: Random sample of last
year’s invoices, where the invoices were examined to
determined if there were any errors when filling out the
invoice or within any other step of the process. Multiple
errors or defects could occur when executing an invoice.
The total number of defects when the invoice was
executed was divided by the total number of
opportunities for failure to estimate the defect per million
opportunity (DPMO) rate of the process.
26
3.9 Attribute vs. Continuous
Data Response: Examples
9/4/2012
14
• Manufacturing quality metric: Random sample of parts
manufactured over the last year where the diameter of
the parts were measured and reported. If the part was
beyond specification limits, it was considered a defective
part or a "failure." The number of non-conformances
was divided by the sample size to estimate the
nonconformance rate of the overall process. • Note that this procedure would not be the recommended strategy within
S4/IEE, since much information is lost in the translation between continuous
and attribute data.
27
3.9 Attribute vs. Continuous
Data Response: Examples
• Manufacturing quality metric: Random sample of printed
circuit boards over the last year where the boards were
tested for failure. The number of defective boards was
divided by the sample size to estimate the defective rate
of the process.
• Manufacturing quality metric: Random sample of printed
circuit board over the last year where the boards were
tested for failure. Multiple failures could occur on one
board. The total number of defects was divided by the
sample size to estimate the defect rate of the process.
28
3.9 Attribute vs. Continuous
Data Response: Examples
9/4/2012
15
• Manufacturing quality metric: Random sample of printed
circuit board over the last year where the boards were
tested for failure. Multiple failures could occur on one
board. The total number of defects on the boards was
divided by the total number of opportunities for failure
(sum of the number of components and solder joints
from the samples) to estimate the defect per million
opportunity (DPMO) rate of the process.
29
3.9 Attribute vs. Continuous
Data Response: Examples
• Still remain a very large single form of inspection
activity.
• Characteristics often do not completely describe
what is needed, and the inspectors need to
make their own judgment.
• A visual inspection does not necessarily address
the real desires of the customer.
• Visual inspections often lead to the thinking that
quality can be inspected into a product.
30
3.10 Visual Inspections
9/4/2012
16
• Hypothesis tests are used when decisions need
to be made about a population.
• Hypothesis tests involve a null hypothesis (H0)
and an alternative hypothesis (Ha)
• Two risks (errors) can be involved within a
hypothesis test: and
31
3.11 Hypothesis Testing and Interpretation of ANOVA Computer Output
• Null (H0): Mean of population equals the criterion. Alternative
(Ha): Mean of population differs from the criterion.
• Null (H0): Mean response of machine A equals mean response
of machine B. Alternative (Ha): Mean response of machine A
differs from machine B.
• Null (H0): Mean response from the proposed process change
equals the mean response from the existing process. Alternative
(Ha): Mean response from proposed process change does not
equal the mean response from the existing process.
• Null (H0): There is no difference in fuel consumption between
regular and premium fuel. Alternative (Ha): Fuel consumption
with premium fuel is less.
32
3.11 Hypothesis Testing
Examples
9/4/2012
17
33
3.11 Hypothesis Testing Interpretation of ANOVA Computer Output
Source df Seq SS Adj SS Adj MS F P
A 1 0.141 0.141 0.141 0.34 0.575
B 1 31.641 31.641 31.641 75.58 0.000
C 1 18.276 18.276 18.276 43.66 0.000
D 1 0.331 0.331 0.331 0.79 0.395
E 1 0.226 0.226 0.226 0.54 0.480
Error 10 4.186 4.186 0.419
Total 15 54.799
Possible Experimentation Traps:
• Irrational subgroups
• Sampling scheme
• Measurement error
• Poor experiment design strategy
• Erroneous assumptions
• Data analysis errors
34
3.12 Experimentation Traps
9/4/2012
18
Consider that the data in Table 3.3 are the readings (in
0.001 of cm) of a functional gap measured between two
mechanical parts. The design specifications for this gap
were 0.008 0.002 cm. This type of data could be expected
from the measurements in a manufacturing line, when the 16
random parts were taken over a long period of time.
Need to consider:
• how samples are selected when making such an
assessment of the capability/performance of the
response;
• the precision and accuracy of the measurement system
are often overlooked.
35
3.13 Experimentation Trap: Example 3.3 Measurement Error and Others
36
3.13 Experimentation Trap: Table 3.3
Obs 1 2 3 4 5 6 7 8 9 10
1 2.99 7.88 9.80 6.86 4.55 4.87 5.31 7.17 8.95 3.40
2 6.29 6.72 4.04 4.76 5.19 8.03 7.73 5.04 4.58 8.57
3 2.65 5.88 4.82 6.14 8.75 9.14 8.90 4.64 5.77 2.42
4 10.11 7.65 5.07 3.24 4.52 5.71 6.90 2.42 6.77 5.59
5 5.31 7.76 2.18 8.55 3.18 6.80 4.64 10.36 6.15 9.92
6 5.84 7.61 6.91 3.35 2.45 5.03 6.65 4.17 6.11 4.63
7 2.17 7.07 4.18 4.08 7.95 7.52 2.86 6.87 5.74 7.48
8 4.13 5.67 8.96 7.48 7.28 9.29 8.15 8.28 4.91 8.55
9 7.29 8.93 8.89 5.32 3.42 7.91 8.26 6.60 6.36 6.10
10 5.20 4.94 7.09 3.82 7.43 5.96 6.31 4.46 5.27 6.42
11 5.80 7.17 7.09 5.79 5.80 6.98 8.64 7.08 5.26 4.46
12 5.39 2.33 3.90 4.45 6.45 6.94 1.67 6.97 5.37 7.02
13 10.00 3.62 5.68 5.19 7.72 7.77 7.49 4.06 2.54 5.86
14 9.29 7.16 7.18 5.57 3.53 7.12 6.14 10.01 6.69 4.80
15 4.74 9.39 7.14 4.42 7.69 3.71 2.98 2.20 7.89 9.60
16 5.19 7.98 2.36 7.74 5.98 9.91 7.11 5.18 5.67 5.92
𝑥 5.77 6.74 5.96 5.42 5.74 7.04 6.23 5.97 5.88 6.30
s 2.41 1.86 2.30 1.59 1.97 1.69 2.19 2.38 1.42 2.14
9/4/2012
19
• It is typically desirable to have a measurement system
that is at least 10 times better than the range of the
response that is of interest.
37
3.13 Experimentation Trap: Example 3.3 Measurement Error and Others
• Sources of the measured variability of parts (T2 ):
– Repeatability (12): the ability of an appraiser to get a
similar reading when given the same part again;
– Reproducibility (22): the ability of differing appraisers
obtaining a similar reading for the same part;
– Measurement tool-to-tool (32)
– Within lots (42)
– Between lots (52)
– T2 = 1
2 + 22 + 3
2 + 42 + 5
2
– The precision of the measurements of parts is
dependent on T2
– The accuracy depends upon any bias that occurs
during the measurements.
38
3.13 Experimentation Trap: Example 3.3 Measurement Error and Others
9/4/2012
20
39
3.14 Experimentation Trap: Example 3.4 Lack of Randomization
Test
Number
Duration
of
Pressure
(sec) Strength
(lb)
1 10 100
2 20 148
3 30 192
4 40 204
5 50 212
6 60 208
40
3.14 Experimentation Trap: Example 3.4 Lack of Randomization
Test
Number
Duration
of
Pressure
(sec) Strength
(lb)
1 30 96
2 50 151
3 10 190
4 60 200
5 40 210
6 20 212
9/4/2012
21
• The following strategy was used to determine if
resistance readings on wafers are different when
taken with 2 types of probes and/or between
automatic and manual readings.
41
3.15 Experimentation Trap: Example 3.5 Confused Effects
Automatic Manual
Probe Type 1 G1, G2, G3, G4 G1, G2, G3, G4
Probe Type 2 G5, G6, G7, G8,
G9, G10, G11, G12
G5, G6, G7, G8, G9,
G10, G11, G12
Initial Experiment Strategy
42
3.15 Experimentation Trap: Example 3.5 Confused Effects
Automatic Manual
Probe 1 Probe 2 Probe 1 Probe 2
G1
G2
G3
G4
G5
G6
G7
G8
G9
G10
G11
G12
Revised Experiment Strategy
9/4/2012
22
• A system under development had 3 different functional
areas. In each of these areas there were two different
designs that could be used in production.
• 8 special systems were built
43
3.16 Experimentation Trap: Example 3.6 Independently Designing and
Conducting an Experiment
System
Number
Functional Area
A B C
1 + + + 2 + + - 3 + - + 4 + - - 5 - + + 6 - + - 7 - - + 8 - - -
• It is important for the designer to have some involvement
in the details of the test activity.
• With communication, the designer may also find other
unknown characteristic of the design/process that is
important.
44
3.16 Experimentation Trap: Example 3.6 Independently Designing and
Conducting an Experiment
9/4/2012
23
• Random sampling plans are based on the assumption that
errors are independent and normally distributed.
• If appropriate, randomization is introduced as an
approximation alternative when conducting an experiment.
• If data dependence can be modeled, it is possible to
develop procedures that are more sensitive than those that
depend only on randomization.
45
3.17 Some Sampling Considerations
• Select what to measure: Consider the questions that need
to be answered and the data that will help answer these
questions. Consider also the final and internal customers to
the process and how the measures will be tracked and
reported.
• Develop operational definitions: Consider how to clearly
describe what is being measured to ensure that there is no
miscommunications.
• Identify data source(s): Consider where someone can
obtain data and whether historical data can be used.
46
3.18 DMAIC Measure Phase
9/4/2012
24
• Prepare collection and sampling plan: Consider who will
collect and compile the data and the tools that are
necessary to capture the data. Create a data-sampling
plan that addresses any potential data integrity issues.
• Implement and refine measurement: Consider what could
be done to assess an initial set of measures and
procedures for collecting the data before expending a lot of
resource collecting/compiling questionable data. To ensure
ongoing data integrity, consider what procedures will be
followed to monitor data-collection practices over time.
47
3.18 DMAIC Measure Phase