MEIE5224
Six Sigma Methodology
Process Capability – Non-Normal
Distributions
outline
I. Process Capability – Non-Normal Variable
II. Process Capability - Attributes
– Binomial (defective/not-defective)
– Poisson (1 or more defects)
III. Process Capability and Sigma Level
IV. Control Charts
-Attribute data ( Pass/Fail)
How to calculate the sigma level?
• Suppose that a process has 24000 DPM, the
what is its sigma level
Ans:
Convert the DPM to defect area which is
φ(Z)=24000/1000000= 0.024
You may use excel normsinv(.024)= -1.977
Then you add 1.5 which means that the
process is 3.477 sigma
Process capability non-normal
commission fees
Since the P-value <0
Reject Ho
Data is NOT normally
distributed
ISO Method
Non-normal Distributions (e.g., Weibull, Exponential)
– Find XL with Pr()=0.00135,
USE inverse of the
– Find XU with Pr()=0.99865 CDF
– Estimate 6σspread using XU - XL
General ISO method
Steps for ISO method
1. Test for Normality
2. Fit Best Distribution
3. Find XL and XU
4. Find 6σspread and Ppk, DPM
Example : Commission Fee (Loan)
Minitab : Stat >> Basic Stats >> Graphical Summary
Mintab Command Output
Minitab Commands:
– Stat > Reliability > Dist Analysis
>> Parametric Distribution
Analysis
– Variable: FeeStudy; Assumed
Distribution = Weibull
– Under Estimate Option in the
Tool Menu: Go to “Estimate
Percentiles for these
additional percents” and type in >>
0.135 99.865
Using Minitab
You can fit wiebull distribution to ISO!
Box-Cox Transformation
Transform data to look normal.
• Compute Pp, Ppk, DPM, etc. for transformed data.
• Box-Cox Method in Minitab:
– Step 1: Find an optimal Lambda
• Control Charts >> Box-Cox Transformation to find “optimal lambda”
for data set (lambda transformation exponent W=Y^Lambda )
• For these data, optimal lambda ~0.45
– Step 2: Process Capability Analysis (Normal)
• Under options – click on Box-Cox Transformation enter value for
lambda (for these data – use 0.45 from above)
Control Charts
Defining statistics Control Chart
Examples of Out-Control Points and Nonrandom
Patterns
• One or more points plots outside the control limits.
• 9 consecutive points above or below the center line.
• A run of 6 consecutive points increasing or decreasing.
CL Vs. SL
• Control limits are used to evaluate process
stability or "in a state of statistical control”.
• Specification limits are used to evaluate process
capability – how well product (or service) meets
customer expectations.
• Ultimately, we will evaluate if process is stable or
not, separate from capable or not.
– Example: A process may be unstable but in-
specifications, or stable but not in-specifications.
Statistical Process Control (SPC)
• Common Statistical Process Control Methods
– Variable Control Charts
• X-bar/Range Chart
• Individual/Moving Range
– Attribute Control Charts
• P-charts - % non-conforming product (e.g., % defective, yield)
• U-charts – each unit may have 1 or more defect opportunities
• Other methods of control:
– Automatic Source Inspection (e.g., poka-yoke)
– Acceptance Sampling (not covered in class)
– Numerous other types of control charts ( moving average control
chart)
Monitoring chart
Two typical control charts.
1) Monitor within subgroup variation (such as
subgroup range or standard deviation).
• Range or Moving Range Charts
2) Monitor between subgroup variation (such as the
mean or median)
Determination of CL and Limits
• To determine center line (CL) and limits of a control
chart, we use the following
– Individuals, X X-bar (mean) +/- 3σx
– X-bar X-double bar (grand mean) +/- 3σx-bar
– Range, R Average Range (R-bar) +/- 3σR
– Moving Range, RM Average RM (RM-bar) +/- 3σMR
– Fraction non-conforming, p Average Fraction Non-conforming
(p-bar) +/- 3σp
• Most of the above sigma terms cannot be calculated directly and
therefore must be determined using statistical assumptions and
estimation techniques.
Example
What is your conclusion?
Over-controlling a Process: The Funnel Experiment
“There are two types of variation. Common cause variation is variation
due to the system. Reducing common cause variation is the
responsibility of management. Dr. Deming has estimated that 94%
of problems are due to common cause variation, i.e., due to faults
in the system. Special cause variation is variation due to sporadic,
unnatural events. Special cause variation is the responsibility of the
front-line personnel to correct. Dr. Deming has estimated that 6% of
problems are due to special cause variation.”
“Control charts help us determine the type of variation present. This is
important to know because the strategies for improving a process
that is in statistical control is different from the strategies for
improving a process that is out of statistical control.”
Funnel Experiment
• Dr. Deming has said "If anyone adjusts a stable
process for a result that is undesirable, or for a
result that is extra good, the output that follows
will be worse than if he had left the process
alone.
• An excellent example of over-controlling is the
funnel experiment described by Dr. Deming. The
objective of the funnel experiment is to drop a
marble through a funnel and hit a target.
• http://www.spcforexcel.com/overcontrolling-
process-funnel-experiment#rule1
Results
Case Study- Attribute Data
• Problem
– High Supplier Defect Rate of parts
Brief Info. about the case study
Manufacturer of Fiber Optics has a concern with their
supplier of their couplings.
– 5 Defect Categories for Couplings at Incoming Inspection
Process
• Short fiber (scrap), dirty connector (rework), damaged
protective material (scrap), excess scratches (rework),
miscellaneous (rework)
– Receive shipments daily from their supplier ~ 250 units
per day.
– Couplers may have 1 or more defects per unit
• If at least 1 defect, unit is considered DEFECTIVE.
Attribute Quality Characteristics
Attributes to assess stability
– % Non-conforming (% defective) to standard (binary).
• Pass/ Fail
• Leaks/ Does not leak
• Present/ absent
– # of non-conformities (# defects)
• # of scratches,
• # of paint defects
– Or # non-conformities per unit
• # of scratches per table
Common Attribute Charts
Counting “Defective Units”
• p-chart – plot fraction non-conforming per
sample.
• np-chart – plot # of non-conforming per sample –
requires equal sample size per subgroup.
Counting “Defects”
• c-chart – plot # of non-conformities per inspection
unit
• u-chart – plot average # of non-conformities per
inspection unit.
What happens when n is large?
