SixSigma is a
ProcessImprovement methodology promoted by Motorola.
The method was designed for
ManufacturingEngineering situations,
where defects are caused by
ProcessParameters varying too much.
The method is very useful for checking if a manufacturing process
is capable of producing a design, because it compares the process variability
to the design requirements. It is also useful for adjusting
manufacturing processes to minimize their defect rates.
However, it is often used in other situations (like processing paperwork),
where defects either exist or don't exist, and the reason for the underlying
variability is unknown.
I have seen SixSigma lumped together with TotalQualityManagement. Are they very much related?
The name
SixSigma refers to a metric for predicting defect rates.
In many manufacturing processes, important parameters
(like temperature of a furnace, or length of machined part)
can be measured. This data can be analyzed statistically.
For example, the
StandardDeviation can be calculated.
Sigma is the symbol for
StandardDeviation.
The higher the
StandardDeviation, the more the parameter varies --
and the less likely we are to achieve the
CustomerRequirements.
The
StandardDeviation can be estimated, even with very little data.
For example, if 3 randomly chosen parts are 8.000 inches long,
8.005 inches long, and 8.010 inches long, then the average is 8.005 inches,
and the
SampleStandardDeviation is 0.005 inches. (Yes, I really did get out my calculator for that!)
Using just 3 data points makes this a very rough estimate. You need at least 30 data points for a statistically "valid" estimate.
The
CustomerRequirements can often be translated
(more or less painfully) into a tolerance for the parameter being measured.
For example, if the furnace temperature is 2 Celsius degrees above average,
a coating might be too thick. Or, the part must be between 7.980 inches
and 8.020 inches long in order to fit in the next assembly.
The tighter the tolerances, the less likely we are to achieve the
CustomerRequirements.
The metric used in
SixSigma calculations is:
Abs(
ClosestSpecificationLimit -
ProcessAverage) /
StandardDeviation
For example: (8.020 inches - 8.005 inches) / (0.005 inches)
Equals: (0.015 inches) / (0.005 inches)
Equals: 3 Sigma
A higher
SigmaNumber is better.
One of the first lessons of the
SixSigma methodology
is to
CenterTheProcess. In this example, we want to be equally likely
to make parts that are too short or too long.
Sometimes centering the process is a bad idea. The CenterTheProcess page has counter-examples.
If we (somehow) change the
ProcessAverage to 8.000 inches, we get
(8.020 inches - 8.000 inches) / (0.005 inches)
Equals: (0.020 inches) / (0.005 inches)
Equals: 4 Sigma
Because this StandardDeviation was estimated from just 3 data points, we don't really know whether the process was already centered at 8.000 inches. Maybe it was a fluke that the 3 data points were on the same side of 8.000 inches. Making adjustments based on too little data is actually counter-productive -- though it might make sense to double-check the calibration of the part cutting machine.
On a good day, a 1 Sigma process has a 1/3 chance of producing a defect
in any given opportunity. A 2 Sigma process has a 5% defect rate (on a good day). A 3 Sigma process has a 0.3% defect rate (on a good day), etc.
However, many days are not good days. Motorola assumes that the process can shift by 1.5 Sigma before someone notices the problem and re-centers the process.
So a process that is at 2 Sigma on a good day may be at 0.5 Sigma on a bad day
(and have a 2/3 defect rate). A process that is at 1 Sigma on a good day
can have a 100% defect rate on a bad day!
The goal of the
SixSigma methodology is to reduce the variation
and/or loosen the
CustomerRequirements enough to achieve
a 6 Sigma quality level. A 6 Sigma process has 3 defects per 1,000,000
opportunities (even on a bad day).
But 0 sigma means that the average is right at one of the spec limits, so half the parts are in spec and half out. 0.5 sigma would mean better than half the parts are in spec, not 2/3 out.
No, 0 sigma means that all of the parts are more than 0 sigma away from the mean - in either direction. It is like having the upper and lower spec limits equal the mean exactly.
Because many managers want to apply this method,
even in non-manufacturing areas that can't measure their process variation,
there are tables for converting defect rates into Sigma values.
Most of
Discussion and below moved to
SixSigmaDiscussion, except for the following clarification:
When a company says it is using SixSigma, it is referring to the SixSigma methodology for defect reduction, which includes several statistical tools in addition to the metric described above. But, when a company claims that a process has reached SixSigma, it is referring to six standard deviations, as calculated above.
The
SixSigma methodology for defect reduction is known by the acronym DMAIC:
Define
Measure
Analyze
Improve
Control.
There are many books on Six Sigma which describe the methodology and tools (I won't list any here; just search Amazon, bookpool, or your favorite bookseller). I have found so far only one published book specifically on "Six Sigma Software Development", which seemed to be more of an overview of Six Sigma with brief references to software, than a thorough discussion of the considerations of applying Six Sigma tools to software process improvement. A better resource is the Measurement and Analysis program at the
SoftwareEngineeringInstitute (
http://www.sei.cmu.edu/sema/). Their team includes a Six Sigma black belt, Jeannine Siviy, who is leading their effort to explore applying Six Sigma for software.
See
SixSigmaDiscussion for thoughts on the relative merits or demerits of applying Six Sigma for software ;)
KarenSmiley, 8/14/2003
Resources related to SixSigma
Beginner
Newbie material ->
http://www.isixsigma.com/library/content/six-sigma-newbie.asp
ICT related material and links ->
http://www.isixsigma.com/bp/it/ and
http://www.softwaresixsigma.com/SixSigma_Main_SixSigma.htm
Other
See also:
ProfoundKnowledge,
CenterTheProcess
CategoryMethodology CategoryManufacturing CategoryStatistics