January 2011

By Douglas C. Wood

Have you some measure you depend on? A nurse takes patient temperatures, a business measures a critical process; all of us have measures we use to get our work done.

A mistaken measure can cost you. One manufacturer had been tracking extruder pressures religiously for years. Each hour a measure was taken by the operator. There were occasional issues with color variation of the extruded polypropylene. Deeper evaluation showed the pressure variability of the process was one tenth the variation recorded by the operator. The color was actually altered by these small changes to pressure, too small to be seen in the recorded metric! The process had been running 15% waste; the measure was useless to control it. After changes were made, waste dropped to 3%. Millions were saved.

We intend to measure the changes or consistency of a process, but how do we know the numbers collected actually tell us about the process? Truly critical metrics should not depend upon faith.

When we measure there are two areas of variability:

• the process itself (that we want to measure)
• the process of taking the measure (that we don’t want to see)

The first is what we think we are seeing; the second is what we unintentionally measure. Do you know which is larger? How do you tell which contributes more to the variability of your measurement?

Measurement systems analysis (MSA) is how to answer this. It is a simple process, if done using statistical software. The first step is to want to know your measure’s reliability. The second step is to consider the elements that contribute to measurement variation.

MSA used to be called gage repeatability and reproducibility, or gage r & r for short. It is part of the quality tool set for data collection. To collect quality data, you need to know what sources contribute to good and bad variability.

Here are the main steps to MSA:

• Decide the subject you are measuring, what data you will collect, and the tools to take the measure.
• Decide how many tests need to be run, based on the degree of variability and the needed precision. (It is usually 5-10 tests or ‘trials.’ A trial may be a part, a patient, etc.)
• How many people are running the measures (2-3 is usual)
• Each person needs to repeat their measurements to have reliable numbers. (2-3 repeats per person is usual)
• Make sure the instrument(s) is (are) calibrated, and the measurement procedure is clear to all. Everyone needs to use the same procedure.
• Arrange the trials in a random sequence. The 1st person then measures and records the data
• Set a different random sequence, and the 2nd operator measures and records the data. Do not let the different operators see each other’s results, or tell what the answer ‘should be.’ When all operators have done one set of measures, one trial is completed.
• Repeat 6 and 7 for all people..
• Perform data analysis using desired software

 

Each step is relatively simple. In the end, our software tells us if the measure is repeatable (by the same person) and reproducible (by other people.) it tells us how much of the variability on the measure is due to the process (the good variation) and how much is in the measurement (the bad variation.)

For example, Minitab software can provide a result summary like this:

In this example, the part-to-part (or the actual subject variation) is much larger than the Gage R & R, Repeatability or Reproducibility variation. This is a good thing, for the good part is the major part of our measure. If it had been the other way around, we would need to do some work on reducing the bad variation.

Besides making a reliable analysis in seconds, Minitab offers detailed help that guides an analysis. For example, MSA may be done in stages, doing a short MSA to establish the general variability, then more extensive work to home in on the key components of variation.

The result is knowing, not guessing, at the sources of variation. To guess is to take measures on faith, and risking being misled or unpleasantly surprised.

How to learn more

We offer a three hour webinar on Quality Tools for Data Collection, including an MSA exercise. Offered on various dates as a live, instructor led internet course from our website: DC Wood Course Registrations You may contact us for more information:

DC Wood Consulting, LLC
13817 Bradshaw, Suite B
Overland Park, Kansas USA 66221

email: click here
phone (913) 669-4173
fax (913) 273-1611

Copyright 2011 DC Wood Consulting, LLC