Statistical Method
Jakarta, 29-30 March,2016
Speaker: Heru Purnomo
Module Outline
• PART 1: Introduction
• PART 2: Capability Studies
• PART 3: Cp, Cpk and Six Sigma • PART 4: Use of Capability Studies • PART 5: Technical Considerations • PART 6: Control Charts
• PART 7: Use of control chart
PART 1:
What is Statistical Process Control?
• A tool that allows us to manufacture products per our customer’s requirements on a consistent basis. This is achieved by preventing defects at each stage of the manufacturing process.
• By itself will not insure our products, processes and services meet our customer’s expectation. Therefore, we assume that all initial work has been done to meet those requirements such as:
– Establishment of specification – Process Characterization – Standards and SOPs – Validation
Where should we use SPC?
• Use as a Method for Preventing Defects
• Use With the Assumptions that Requirements Have Already Been Established:
– Customer – Process – Product
So, how do we prevent defects, and build products,
processes and provide services with consistently good quality?
To answer this question, we will use a hypothetical filling operation to illustrate various methods of preventing defects.
How Does One Prevent Defects?
Two Possible Causes
Cap Fell Off Due to Low Removal Force
Cap Never Put Onto Vial
CAP NEVER PLACED ONTO VIAL
• This is an Error in the form of an Omission. • Errors are Prevented by Mistake Proofing.
• Mistake Proofing means ensuring that the defects either cannot occur or cannot go undetected.
Reference
Shingo, Shigeo (1986). Zero Quality Control: Source Inspection and the Poka-yoke System. Productivity Press, Cambridge, Massachusetts.
Low Removal Forces
• Removal Force is Affected by Many Factors.
• Having low removal forces is an Optimization (Targeting) and Variation Reduction problem.
• Low removal forces are prevented by identifying and controlling the key inputs and establishing proper targets and tolerances.
Reference
Taylor, Wayne A. (1991). Optimization and Variation Reduction I Quality. McGraw-Hill, New York and ASQC Quality Press, Milwaukee.
Optimization & Variation Reduction
(O.V.R)
Lower Spec Limit Target Upper Spec LimitTypical O.V.R Problems
Larger the Better
Lower Spec Limit
Target
Smaller the Better Upper Spec Limit
• microbial level • particulate level • contamination level
Closer to the target the Better
Target Lower Spec Limit Upper Spec Limit
• Vial fill volume • Potency, Assay
Two Approaches to Preventing
Defects
• Mistake Proofing
Reducing Variation
“ If I had to reduce my message for management to
just a few words, I’d say it all had to do with
reducing variation.”
Tools for SPC
• SPC provides two tools for reducing variation:
1) Control Charts 2) Capability Study
• The primary tool for managing variation is the capability study.
• An effective program for reducing variation must also incorporate many other tools including:
Design Experiments
Variation Decomposition Methods Taguchi’s Methods
Statistical Variation
The differences, no matter how small, between ideally identical units of product.
2.3 ml
DIFFERENCES
Displaying Variation
24.0 23.2 22.4 21.6 6 5 4 3 2 1 0Cap Removal Force in PSI
Nu m be r M ea su re d
The Bell-Shaped Curve
24.0 23.2 22.4 21.6 Normal CurveThe Standard Normal Curve
• Contains an Area equal to One• Corresponds to 100% of ALL possible Outcomes in a Stable System. • Has a Mean = 0, and SD = 1
μ
σ
-3
Standard Deviations from the Mean
3 1
-1
-2 2
Slide 20
A STABLE PROCESS
Total Variation
Slide 21
AN UNSTABLE PROCESS
• Total
Slide 22
A CAPABLE PROCESS
NOT CAPABLE CAPABLE Spec LimitsSlide 23
Objective of SPC
To consistently produce high quality
products by achieving stable and
Slide 24
WHY
REDUCE
VARIATION?
Slide 25
Reducing Variation Reduces Defects
Upper Spec Limit Lower Spec Limit Upper Spec Limit GOOD
PRODUCT PRODUCTGOOD
•Defective Product •No Defective Product
Lower Spec Limit
Slide 26
Reducing Variation Widens
Operating Windows
Lower Spec Limit Upper Spec Limit Lower Spec Limit Upper Spec LimitSlide 27
Bad Product
Reducing Variation Improves
Customer Value
Lower Spec Limit Bad Product Target GOOD PRODUCT L O S S $10 $20 $0 Lower Spec Limit Upper Spec Limit Target L O S S $10 $20 $0 •Taguc hi’s LossLoss equals value
Upper Spec Limit
Slide 28
Better Management
• Provides facts on process performance to allow:
Prioritized improvement projects Tracking progress
Demonstrating results
• Provides data on process capability required in
product design.
Slide 29
Maximize Process Capability
• Replace existing equipment only if necessary.
• Improve capability make new product feasible
.
Slide 30
Benefits of SPC
• Fewer Defects
• Wider Operating Windows
• Higher Customer Value
• Better Management
Slide 31
PART 2:
Slide 32
Variation Reduction Tools
• There are many tools that help to achieve
stable and capable processes:
Scatter Diagrams
Screening Experiments Multi-Vari Charts
Analysis of Means (ANOM) Response Surface Studies
Variation Transmission Analysis Component Swapping Studies Taguchi Methods
Slide 33
Capability Study
• Determines if a Process or System is Stable and
Capable i.e., can it consistently make good product.
• Can Measures the Progress and Success achieved
after changes or improvement
Pre-Capability Study Variation Reduction Tool Post Capability Study
Slide 34
Capability Study
Process capability measures statistically summarize
how much variation there is in a process
relative to
costumer specifications.
Too much variation Hard to produce output within costumer requirement
(specification)
Low index value (e.g. Cpk < 0.5)
(Process sigma between 0 and 2) Moderate Variation Most output meets costumer
requirement
Middle index value
(Cpk between 0.5 and 1.2)
(Process sigma between 3 and 5) Very little variation Virtually all of output meets
costumer requirements
High index value (Cpk > 1.5)
35
Short-Term vs Long Term Sigma
Long-Term Sigma
This is the variation is due to both common and special
causes. This variation is calculated based on all of
individual readings (population). Used for Pp and Ppk
calculations
Short-Term Sigma
This variation is due to common causes only. This
variation is estimated from the control chart data. Used for
Cp and Cpk calculations
36
C
p
and C
pk
vs. P
p
and P
pk
• These metrics are calculated the same way, but they
use a different way to estimate standard deviation
• C
pand C
pkare considered “short-term” measures
The estimated StDev (Within) is average of for
each subgroup
• P
pand P
pkare considered “long-term” measures
The estimated StDev (Overall) is calculated using
the standard deviation (n-1) for all the data set
points
2
d R
The conservative approach is to report whichever set of statistics is smaller. Typical data sets are usually too small to be able say with certainty that one metric is better than the other.
37
Process Capability Ratio – C
p
(Cont.)
+3σ -3σ Process Width T LSL USL
•or
process the of iationNormalAllowedvar variation (spec.)
Cp = 99.73% of values
σ
6
LSL
-USL
C
p=
Where σis “within”38
Subgroup Measured Values Average Std. Dev.
1 52.0 52.1 53.0 52.3 51.7 52.22 1.3 2 51.7 51.5 52.0 51.7 51.3 51.64 0.7 3 51.7 52.2 51.9 52.6 52.5 52.18 0.9 4 51.3 52.2 51.8 52.5 51.4 51.84 1.2 5 50.8 50.9 51.7 51.8 51.4 51.32 1.0 6 52.6 51.4 52.9 52.6 52.4 52.38 1.5 7 53.0 52.9 52.5 52.5 51.8 52.54 1.2 8 52.5 52.7 51.2 53.7 51.3 52.28 2.5 9 51.9 51.6 51.6 52.7 51.7 51.90 1.1 10 52.2 52.7 52.3 51.8 53.2 52.44 1.4 11 52.4 52.6 52.1 51.8 51.9 52.16 0.8 12 51.3 51.2 51.9 53.1 52.9 52.08 1.9 13 51.7 51.6 51.4 51.4 51.1 51.44 0.6 14 51.8 51.0 52.4 51.2 51.6 51.60 1.4 15 52.0 51.7 52.6 51.8 52.7 52.16 1.0 16 52.0 52.3 51.8 52.0 51.5 51.92 0.8 17 51.8 51.8 51.8 51.9 52.0 51.86 0.2 18 52.0 51.9 51.4 51.8 53.3 52.08 1.9 19 51.5 52.6 52.8 52.4 52.0 52.26 1.3 20 51.5 51.8 50.8 51.3 52.5 51.58 1.7 51.99 1.22
Slide 39
Is the process Stable?
21 19 17 15 13 11 9 7 5 3 1 53.0 52.5 52.0 51.5 51.0 Observation In di vi du al V al ue _ X=51.994 UCL=53.043 LCL=50.945 21 19 17 15 13 11 9 7 5 3 1 1.2 0.9 0.6 0.3 0.0 Observation M ov in g Ra ng e __ MR=0.394 UCL=1.289 LCL=0
Slide 40
Is the process Capable?
55 54 53 52 51 50 20 15 10 5 0 H1 Fr eq ue nc y Histogram of H1 • LS L • US L Potential Capability Cp = 1.55 Actual Capability Cpk = 1.23
Slide 41
Calculating the Grand Average
• If
Χ
1,
Χ
2,… ,
Χ
20are the subgroup averages, the
grand average is:
X =
Χ
1+
Χ
2+… +
Χ
2020
Grand Average Tim eSlide 42
Calculating the Average Range
• If R
1,R
2,… ,R
20are the subgroup ranges, the
average range is:
R = R
1+R
2+… +R
2020
• Estimates the average within the subgroup
variation.
Slide 43
Between and Within Subgroup Variation
Other names for within subgroup variation:
• Noise • Unexplained • Inherent • Error Total Variation Within Subgroup Variation Between Subgroup Variation
44
Process Capability Ratio – C
p
• Ratio of total variation allowed by the specification to the total
variation actually measured from the process
• Use C
pwhen the mean can easily be adjusted (i.e., plating,
grinding, polishing, machining operations, and many
transactional processes where resources can easily be added
with no/minor impact on quality) AND the mean is monitored
(so operator will know when adjustment is necessary – doing
control charting is one way of monitoring)
• Typical goals for C
pare greater than 1.33 (or 1.67 if of
considerable importance)
If Cp < 1 then the variability of the process
45
Different Levels of C
p
The C
pindex reflects the
potential of the process if
the mean were perfectly
centered between the
specification limits.
For a Six Sigma Process,
Cp = 2
The larger the Cp index, the better!
− = 6 LSL USL C p
σ
C
p= 1
USL LSLC
p> 1
C
p< 1
46
This index accounts for the dynamic mean shift in the
process – the amount that the process is off target.
Calculate both values and report the smaller number.
Notice how this equation is similar to the Z-statistic.
Process Capability Ratio – C
pk
−
−
=
σ
LSL
x
or
σ
x
USL
Min
C
pk3
47
Process Capability Ratio – C
pk
(Cont.)
• Ratio of the distance to the closest spec to ½ of the estimated process variation • Use when the mean cannot be easily adjusted (i.e., stamping, casting,
plastics molding)
• Typical goals for Cpk are greater than 1.33 (or 1.67 if of considerable
importance)
• For sigma estimates use:
• R/d2 [short term] (calculated from X-bar and R chart) – use with Cpk
• s = Σ (xi -x) 2 [ “longer” term] (calculated from (n-1) data points) – use
with Ppk
• “Longer” term: When the data has been collected over a sufficient time period that over 80% of the process variation is likely to be included
48
Actual Process Performance (C
pk
)
Unlike the Cp index, the Cpk index takes into account off-centering of the
process. The larger the Cpk index, the better.
6σ LSL USL 6σ LSL USL
C
p= 1
C
pk= 1
C
p= 1
C
pk< 1
49
Calculating C
p
, C
pk
and P
p
, P
pk
How did Minitab calculate these values?
602 601 600 599 598 USL LSL PPM Total PPM > USL PPM < LSL PPM Total PPM > USL PPM < LSL PPM Total PPM > USL PPM < LSL Ppk PPL PPU Pp Cpm Cpk CPL CPU Cp StDev (Overall) StDev (Within) Sample N Mean LSL Target USL 6367.35 39.19 6328.16 3631.57 10.51 3621.06 10000.00 0.00 10000.00 0.83 0.83 1.32 1.07 * 0.90 0.90 1.42 1.16 0.620865 0.576429 100 599.548 598.000 * 602.000
Exp. "Overall" Performance Exp. "Within" Performance
Observed Performance Overall Capability
Potential (Within) Capability Process Data
Within Overall
Cp and Cpk values are calculated based on estimated
StDev(Within).
The minimum of CPU (capability with respect to USL) and
CPL (capability with respect to LSL) is the Cpk.
If a target is entered, then Cpm, Taguchi’s capability index,
is also calculated.
Cp and Cpk values are considered to be “short term.”
50
C
p
, C
pk
vs. P
p
, P
pk
How did Minitab calculate these values?
602 601 600 599 598 USL LSL PPM Total PPM > USL PPM < LSL PPM Total PPM > USL PPM < LSL PPM Total PPM > USL PPM < LSL Ppk PPL PPU Pp Cpm Cpk CPL CPU Cp StDev (Overall) StDev (Within) Sample N Mean LSL Target USL 6367.35 39.19 6328.16 3631.57 10.51 3621.06 10000.00 0.00 10000.00 0.83 0.83 1.32 1.07 * 0.90 0.90 1.42 1.16 0.620865 0.576429 100 599.548 598.000 * 602.000
Exp. "Overall" Performance Exp. "Within" Performance
Observed Performance Overall Capability
Potential (Within) Capability Process Data
Within Overall
Pp and Ppk values are calculated based on estimated
StDev(Overall).
The minimum of Ppu (capability with respect to USL) and
Ppl (capability with respect to LSL) is the Ppk.
Pp and Ppk values are considered to be “longer term.”
•Process Capability Analysis for Supp1
If the Cp and Pp values are significantly different this is an
Slide 51
PART 3:
Slide 52
Process Should be Stable before Checking Capability
A
Stable Process
NOT a Stable Process
UCL
LCL
UCL
Slide 53
Stable Process are Predictable!
Distance from Average
(d) Percentage out of Spec.
-5.0σ 0.3/million -4.5σ 3.4/million -4.0σ 31/million -3.5σ 233/million -3.0σ 0.135% -2.5σ 0.6% -2.0σ 2.3% -1.5σ 6.7% -1.0σ 15.8% -0.5σ 30.9% 0.0σ 50% 0 1σ -3σ 3σ d LSL 2σ -2σ -1σ Individuals Distribution Percent out of Spec
Slide 54
Cp
Compares the Specification Range to
the Width of the Process, (±3
σ each
side of the mean):
C
p
= USL-LSL
6S
LSL
Cp = 1.5
Cp = 2
Cp Does Not Consider Centering
Cp = 1 Cp = 0.5
Slide 55
Cp = 1
Allows a ±1.5
σ operating window, worse case
is 6.7% defective.
LSL USL ±1.5σ 6.75% Defective • -3σ • 3σ • 3 – Sigma ProcessSlide 56
Cp = 1.5
Allows a ±1.5
σ operating window, worse case
is 1350 defects per million.
LSL ±1.5σ 1350 Defects/Million -4.5σ 4.5σ 4.5 – Sigma Process USL
Slide 57
Cp = 2.0
Allows a ±1.5
σ operating window, worse case
is 3.4 defects per million.
•LSL ±1.5σ •USL
3.4
Defects/Million
-6σ 6σ
Slide 58
What has Changed?
6 – Sigma Process LSL ±1.5σ USL -6σ 6σ LSL USL ±1.5σ -4.5σ 4.5σ 4.5 – Sigma Process LSL USL ±1.5σ -3σ 3σ • 3 – Sigma Process
Slide 59
Operating Windows
• Don’t forget to include them in your
process design.
• Stable process can hold a ±1.5
σ operating
window.
• Automated processes may be able to hold
a ±1.0
σ operating window.
• If a process is not stable, a ±1.5
σ window
may not be enough room for the average.
Why?
UCL, LCL = X ± 3σ/√n
When, n = 5
then,
Slide 60
Cp Does Not Consider Centering
LSL USL
Cp = 2
Slide 61
Determine Cpk
LSL
X
Cpk =
Distance from X to the nearest Spec 3S
3S Average - LSL
Slide 62
C
pk
= 1
USL
LSL
Slide 63
C
pk
= 2
USL
LSL
Slide 64
C
pk
When C
p
= 2
USL LSL • Cpk = 1/2 Cpk= 1/2 Cpk= 1 Cpk= 1 • Cpk= 1.5 • Cpk = 1.5 • Cpk = 2.0Fix the variability, then move the average
When the process is perfectly centered,
C
pk= C
p.
Slide 65
Interpreting Cpk
Table below gives the corresponding defect level of
various Cpk’s with Cp = 2.0:
Cpk Defect Level 1.5 3.4 dpm 1.167 233 dpm 1 1350 dpm 0.83 0.6% 0.5 6.7%Slide 66
PART 4:
Slide 67
Uses of Capability Study
• Identifying processes needing improvement.
• Tracking process performance.
• Verifying the effectiveness of fixes.
• Determining the ability of suppliers to consistently make
good product.
• Qualifying new equipment.
Slide 68
Identifying Processes Needing Improvement
• Unstable processes of processes with poor Cp’s and/or
Cpk’s are target for improvements.
• If the process is unstable it is a good candidate for control
chart.
• If the process is stable, but not capable, one should first
look for obvious sources of variation.
• If no obvious sources exist, then you should perform
designed experiments to uncover them.
Slide 69
Verifying Effectiveness of Fixes
• Use a capability study to demonstrate the
effectiveness of fixes.
• New estimates of Cp and Cpk should be at least
15% greater than the pre-fix estimates.
• True changes are unlikely when pre and post
capability estimates are within ± 15% of each
other
.
Slide 70
Assessing the performance of Suppliers
• Materials and components from our suppliers make up one
or more inputs in our manufacturing process.
• Our final quality is only as good as our supplier’s quality.
• All suppliers need to provide good product on a consistent
basis.
• “Consistency” requires a stable process of manufacturing.
• If the process is not stable, the products produced will not
be stable in quality.
• “Stability” can only be assessed by looking at time ordered
samples.
Slide 71
Qualifying New Equipment
• Want to demonstrate the equipment can
consistently make good products.
• Should use a capability study to demonstrate
“consistency”.
• Consider requesting a capability study when
purchasing a new equipment.
Slide 72
Determining the Manufacturability of New
Product
• Capability studies measure the match between
product specifications and process variation.
• A process may be capable of manufacturing one
product, but not another.
• For new products, use capability studies to
determine how well the product design adapts to
the manufacturing process.
Slide 73
PART 5:
Slide 74
The Normality Assumption
• Common Misconception:
• Control charts work well even when the data are not
normally distributed.
• The normality assumption was originally introduced from the
control chart constants, i.e. d
2, A
2, D
4, etc,
• Even the control chart constants do not change appreciably
when the data are non-normal*.
“The data has to be normally distributed to be control charted”
Slide 75
Why 3 Standard Deviation Limits?
• Not established solely on the basis of probability theory.
• Outcomes in most stable processes generally occur
between ±3 S.D.’s from the average.
• Originally designed to minimize the time looking
unnecessarily for shifts in the process average.
• Additionally concerned with missing an actual process shift
as it occurs.
Slide 76
Rational Sub grouping
• Organizing the data into rational subgroups allows
us to answer the right questions.
• The variation occurring within the subgroups is
used to set the control limits.
• The control chart uses the within subgroup
variation to place limits on how much variation
should naturally exist between subgroups.
Slide 77
Rational Sub grouping
Some Guidelines:
• Try not to place unlike things together into the same
subgroups.
• Organize in a way that produces the lowest variation within
each subgroup.
• Maximize the opportunity to observe the variation between
subgroups.
Slide 78
PART 6:
79
Control Charts
Control charts are one of the most commonly
used tools in our Lean Six Sigma toolbox
• Control charts provide a graphical picture of the
process over time
• Control charts are both practical and easy-to-use
• Control charts help us establish a measurement
80
What Do Control Charts Tell Us?
• When the process location has shifted
• When process variability has changed
• When special causes are present
• Process not predictable
• A learning opportunity
• When no special causes are present
• Process is predictable
• No clues to improvement available; may need to
introduce a special cause to effect a change
Why Use a Control Chart?
81
• Statistical control limits are another way to separate common cause and
special cause variation
• Points outside statistical limits signal a special cause
• Can be used for almost any type of data collected over time
• Provides a common language for discussing process performance
When To Use:
• Track performance over time
• Evaluate progress after process changes/improvements • Focus attention on process behaviour
82
Control Chart Selection
• Control chart selection should be based
upon:
• Data type
• Number of observations
• Sample size
• Subgrouping
• The primary determinant in control chart
selection is Data Type
83
Data Types
• There are many different types of data
• Each type of data has its own unique control chart
• The basic format and underlying concepts are the same
across the entire family of control charts
• A basic understanding of the different data types is
important to increase the successful use of control charts
• How many different types of data are there?
Two General Kinds of Data
84
• Attribute – The data is discrete (counted).
Results from using go/no-go gages, or from the
inspection of visual defects, visual problems,
missing parts, or from pass/fail or yes/no
decisions
• Variable – The data is continuous (measured).
Results from the actual measuring of a
characteristic such as diameter of a hose,
electrical resistance, weight of a vehicle, etc
.
Continuous Data
85
• Continuous data is a set of numbers that can potentially take on any value
• Also known as variable data
• Examples: 0.1, 1/4, 20, 100.001, 1,000,000, -3.26, -10,000 • Common Applications
• Dimensions (lengths, widths, weight, etc) • Time (seconds, minutes, hours, etc)
• Finance (mills, cents, dollars, etc) • Distribution Types
• Normal • Uniform
• Exponential
• Because continuous data has more discrimination, go for continuous data whenever possible
Control Charts for Individual Values
86
Time ordered plot of results (just like time plots)
Statistically determined control limits are drawn on the plot. Centerline calculation uses the mean
20 18 16 14 12 10 8 6 4 2 53.0 52.5 52.0 51.5 51.0 Index UCL=53.043 Avg=51.99 LCL=50.945 LCL= X + 2.66mR Centerline = X UCL= X + 2.66mR
Attribute Data
87
• Attribute data has two main subsets, Binary data and Discrete
Data
• Binary Data is a characterized by classifying into only two
outcomes
• Examples: Pass/Fail, Agree/Disagree, Win/Loss, defective/conforming
• Common uses: Proportions and ratios • Distribution: Binomial
• Key assumptions
• Events are independent of each other • Mutually exclusive outcomes
Attribute Data (Cont.)
88
• Discrete Data is a set of finite outcomes, usually
integers, and is measured by counting
• Also known as Ordinal data • Common uses and examples:
• Number of product defects per item
• Number of customer requirements per order • Number of accounting errors per invoice • Distribution: Poisson
• Poisson characteristics and assumptions • Unlimited number of defects per item • Constant probability of defect per item • Probability of defect per unit is low • Defects are independent of each other
89
Control Chart Selection Tree
TYPE OF DATA •Poisson Distribution Count or Classification (Attribute Data) Count Defects or Nonconformance Fixed Opportunity C Chart Variable Opportunity U Chart Classification Defectives or Nonconforming Units Fixed Opportunity NP Chart Variable Opportunity P Chart Subgroup Size of 1 I-mR Subgroup Size < 9 X-bar R Subgroup Size > 9 X-bar S Measurement (Variable Data)
•Binomial DistributionBinomial Distribution •Normal DistributionNormal Distribution
90
Individuals and Moving Range
Charts
• Display variables data when the sample subgroup size is
one (And in certain situations, attribute data)
• Variability shown as the difference between each data
point (i.e., moving range)
• Appropriate Usage Situations:
• When there are very few units produced relative to the opportunity for process
variables (sources of variation) to change
• When there is little choice due to data scarcity
• When a process drifts over time and needs to be monitored
• I-mR is a good chart to start with when evaluating
continuous data
91
Calculations for Individuals Charts
1. Determine sampling plan2. Take a sample at each specified interval of time
3. Calculate the moving range for the sample. To calculate each moving range, subtract each measurement from the previous one. There will be no moving range for the first observation on the chart
4. Plot the data (both individuals and moving range)
5. After ‘30' or more sets of measurements, calculate control limits for moving range chart
6. If the Range chart is not in control, take appropriate action
7. If the Range chart is in control, calculate limits for individuals chart 8. If the Individuals chart is not in control, take appropriate action
92
Why Use Subgroups?
It allows us to examine both within
sample variation and between sample
93
X-Bar & R Chart
• The X-bar & R chart is the most commonly used control chart due to its use of subgroups and the fact that it is more sensitive than the ImR to process shift • Consists of two charts displaying Central Tendency and Variability
• X-bar Chart
• Plots the mean (average value) of each subgroup
• Useful for identifying special cause changes to the process mean (X)
• X-bar control limits based on +/- 3 sigma from the process mean are calculated using the Range chart
• R Chart
• Displays changes in the "within" subgroup dispersion of the process • Checks for constant variation within subgroups
94
Calculations for X-Bar & R Charts
1. Determine an appropriate subgroup size and sampling plan
2. Sample: (Take a set of readings at each specified interval of time) 3. Calculate the average and range for each subgroup
4. Plot the data. (Both the averages and the ranges)
5. After ‘30' or more sets of measurements, calculate control limits for the range chart
6. If the range chart is not in control, take appropriate action
7. If the range chart is in control, calculate control limits for the X-bar chart 8. If the X-bar chart is not in control, take appropriate action
95
Rational Subgrouping
• An important consideration in using the X-bar & R (and X-bar & S) chart is the selection of an appropriate subgroup size
• Rational Subgrouping is the process of selecting a subgroup based upon “logical” grouping criteria or statistical considerations
• Subgrouping Examples
• “Natural” Breakpoints:
• 3 shifts grouped into 1 day; • 5 days grouped into 1 week,
• 10 machines grouped into 1 dept
• Wherever possible, both natural breakpoints and homogenous group considerations should be combined together in selecting a sample size
96
Attribute Control Charts
• Attribute control charts are similar to variables
control charts, except they plot proportion or count
data rather than variable measurements
• Attribute control charts have only one chart which
tracks proportion or count stability over time
• Chart Types
• Binomial: P chart, NP chart
• Poisson: C chart, U chart
97
Attribute Control Charts
• Binomial Distribution Charts
• Use one of the following charts when comparing a product to a standard and classifying it as being defective or not (pass vs. fail):
• P Chart – Charts the proportion of defectives in each subgroup • NP Chart – Charts the number of defectives in each subgroup
• Poisson Distribution Charts
• Use one of the following chart when counting the number of defects per sample or per unit
• C Chart – Charts the defect count per sample (must have the same sample size each time)
• U Chart – Charts the number of defects per unit sampled in each subgroup (using a proportion so sample size may vary)
Slide 98
PART 7:
Slide 99
Uses of Control Charts
• Evaluation
• Improvement
Slide 100
Uses of Control Charts
• Evaluation: Determine if the process is both stable
and capable, as part of a capability study.
• Improvement
Slide 101
Uses of Control Charts
• Evaluation
• Improvement: Identify changes to the process so
that the causes may be investigated and
eliminated.
Slide 102
Improvement
• Control Charts search for differences over time.
• Observing a change on the control charts means a key
input variable has changed.
• The pattern observed on the control chart provides clues
about the key variable that changed:
• Timing of the change
• Shape or pattern
• Trends
Slide 103
Maintenance
• Control charts can help us to decide when to make
adjustments to the process.
• Using control charts we can make better decisions, and
minimize the chance of making two possible errors:.
• 20% - Failing to adjust when the process needs adjustments
• 80% - adjusting when the process does not need adjustment
• When maintaining processes using control charts, try to
center the average around the desired target.
Slide 104
Statistical Step to Establish Control Limit
1. Collect the data 30 or more
2. Prepare Individual moving range chart (I-MR) using appropriate
statistical software
3. Review the moving range chart, if any data point beyond are beyond
the UCL, the data the data point must be evaluated and excluded if there is an assignable cause, then replot the moving range chart.
4. Review the individual chart, if any data point beyond are beyond the
UCL and LCL, the data the data point must be evaluated and excluded if there is an assignable cause, then replot the individual chart.
Real Time Evaluation
105
Rule 1 The data outside of control limit: One point of outside the control limit 20 18 16 14 12 10 8 6 4 2 56 55 54 53 52 51 Index UCL=53.043 LCL=50.945 Avg=51.99
Real Time Evaluation
106
Rule 2 Trend Shift: 8 consecutive point on same side of center line
27 24 21 18 15 12 9 6 3 53.0 52.5 52.0 51.5 51.0 Index UCL=53.043 Avg=51.99 LCL=50.945
Real Time Evaluation
107
Rule 3 Trend drift: 6 consecutive points that trend in the same direction (all increasing or all decreasing)
24 21 18 15 12 9 6 3 53.0 52.5 52.0 51.5 51.0 Index UCL=53.043 Avg=51.99 LCL=50.945
Slide 108
PART 8:
Opening a new project in Minitab
109 Menu bar Toolbars Session window Data window Project Manager window (minimized)Overview of Minitab
Worksheet
• Each Minitab worksheet can contain up to 4,000
columns, each column is identified by a number
• The letter after the column number indicates the
data type:
D : date / time
T : text (alphanumeric)
If no letter appears, the data are numeric
Example 1
Problem
Supervisor of medical company is preparing a sales report for a new line of facial cream that the company intends to distribute nationally. In a pilot launch, the company sold facial cream at various stores in Jakarta and Bandung for three months.
Data Collection
The supervisor recorded the daily revenue for two locations during the three months and stored them in minitab project
Example 1
Tools
• Dotplot
• Time Series Plot • Graphical Summary
• Display Descriptive Statistics • Layout Tools
Open Project
1. Choose File > Open project 2. Choose ISPE_Example 1.MPJ. 3. Click open.
Creating Dotplots
• Choose Graph > Dotplot
• Complete the dialog as shown below, then click OK
• In graph variable, enter ‘Jakarta Sales’ and ‘Bandung Sales’ by highlighting them and double clicking each variable, then click OK
Interpreting your result
The graph shows the sales data during the three month period for both location. On average, Bandung sales appear higher than Jakarta sales.
Correcting the Outlier
After checking with the person who entered the data, your discover that the sales information for data n=20 is missing.
Instead of entering 0, you should enter an asterisk (*) to indicate that the value is missing.
• Click project manager toolbar
• In the Bandung column, highlight the cell in column 3 and row 20 as show below.
• Press [DELETE]
Updating a graph
• To choose the dotplot, click in the project manager toolbar • Click the graph to make it the active window
• Choose Editor > Update > Update graph now
Time Series Plot
• Choose Graph > Time Series Plot • Choose Multiple, then click OK
• In series, enter ‘Jakarta Sales’ ‘Bandung Sales’ • Click Time / Scale
• Complete the dialog as shown below
Graphical Summary
• Choose Stat > Basic Statistic > Graphical Summary • Complete the dialog as shown below
Display Descriptive Statistic
• Choose Stat > Basic Statistic > Display Descriptive Statistics.
• In variable, enter ‘Jakarta Sales’ ‘Bandung Sales’ • Click statistics
• Complete the dialog box as shown below, then click OK
Creating a multiple graph
• Click graph folder, then click the dotplot in the project manager. Click the graph to make it the active window. • Choose Editor > Layout tool
• Double click all graph have been created to place the graph in the layout window.
Problem
The validation supervisor want to evaluate the consistency of the fill weight for hydrocortisone cream. The cream is packed in tube. The target weight is 1150grams. The specification limit are 1100 and 1200 grams.
Earlier evidence indicate this process is stable with a mean of 1150 grams and a standard deviation of 8.6 grams
Tools
I-MR
126
I-MR
• Open ISPE_Example 2.MPJ
• Choose Stat > Control Charts > Variable Charts for Individuals > I-MR
• Complete the dialog box as shown below
I-MR
• Click Scale, under X scale, choose stamp
• Under Stamp columns, enter date/time. Click OK • Click I-MR Options.
• In Mean, type 1150; in standard deviation type 8.6, then click OK
129 Individual chart shows that the process is clearly not in
statistical control also process operated consistently above the mean.
130
Next Step
Problem
With previous data analyse normality and capability process
Tools
Probability
Capability Six Pack
131
Probability Plot
• Open ISPE_Example 2.MPJ
• Choose Grap > Probability Plot > Single
• Complete the dialog box as shown below
• Complete dialog as shown below
Normality Check
• Check normality data (P>0.05)
133 1200 1190 1180 1170 1160 1150 1140 1130 99.9 99 95 90 80 70 60 50 40 30 20 10 5 1 0.1 Mean 1164 StDev 8.576 N 60 AD 0.293 P-Value 0.591 Fill Weight Pe rc en t
Probability Plot of Fill Weight
Normal - 95% CI
Normal Data P > 0.05
Capability Analysis
• Open ISPE_Example 2.MPJ
• Choose Stat > Quality Tools > Capability Analysis
• Complete the dialog box as shown below
• Complete dialog as shown below
Capability Analysis
• Cp/Cpk > 1.33
135 1200 1185 1170 1155 1140 1125 1110 LSL 1100 Target * USL 1200 Sample Mean 1163.58 Sample N 60 StDev(Overall) 8.57554 StDev(Within) 8.34686 Process Data Pp 1.94 PPL 2.47 PPU 1.42 Ppk 1.42 Cpm * Cp 2.00 CPL 2.54 CPU 1.45 Cpk 1.45 Potential (Within) CapabilityOverall Capability
PPM < LSL 0.00 0.00 0.00
PPM > USL 0.00 10.81 6.39
PPM Total 0.00 10.81 6.39
Observed Expected Overall Expected WithinPerformance
LSL USL
Overall Within Process Capability Report for Fill Weight