When you start a new job in quality, the first thing you should do is evaluate the company's measurement system.
Measurement provides the feedback to your process control systems. (Ever hear of garbage in, garbage out?) The control chart method is an excellent means for estimating measurement error. To use
this method, plot the repeat readings on an X bar and r chart using the number of repeat measurements as the subgroup size. The r chart should be in control and the X bar chart should be out of
control. In most of the studies I've done, the measurement system was either very good or very bad, and it wasn't hard to tell which. However, there's a problem with
measurement system evaluations: There's a strong tendency to shoot the messenger. People often don't understand the concept of measurement error. They seem to have an implicit faith in digital
readouts, believing that once something's been meas-ured, its value is known. Additionally, many people don't understand the concept of random variation. As such, when you tell them that their
measurement system is exhibiting random variation, you've mixed two unfamiliar concepts that generate a lot of misunderstanding and confusion. So people often get angry at this point. I think
this anger results from fear and perplexity.
Figure 1: A Cylinder, as from a CAD Program
|
Let me give you an example. Say your task is to measure the length and diameter of a cylinder, which sounds pretty straightforward. If you use a computer-aided drawing program,
you probably envision a round, straight cylinder with two parallel faces like the one shown in Figure 1. But in the real world, the faces aren't
parallel, and the cylinder isn't straight or round. So which height should you report? Do you want the height from the low point of
face A to the high point of face B, the height from the high point of face A to the low point of face B or the average? And what do you do if the cylinder is out of
round (which it will be)? Should you report the maximum outer diameter, the minimum or both? Once you start asking questions like these, feathers get ruffled. The calibration
department will bristle at the suggestion that something might be wrong with their gages. (Calibration people often understand accuracy but not repeatability).
High-level management will be horrified by the suggestion that their production people may not know exactly what's going on. The database management people
will be uneasy about the integrity of their database, and the production operators will become even more convinced that the engineers don't really know what they're doing.
Measurement systems should be evaluated with regard to accuracy, repeatability, reproducibility and stability. Most effort directed toward measurement systems
concerns itself with accuracy. Vast sums of money and time are spent each year calibrating gages. However, I've never seen a bad part shipped to a customer
because a gage was out of calibration. It seems to me that repeatability is vastly more important than accuracy.
It reminds me of the old story of three umpires. The youngest umpire says, "Some is strikes and some is balls, and I calls 'em like they is."
The second umpire, who is a little older and more experienced, says, "Some is strikes and some is balls, and I calls 'em like I sees 'em."
But the oldest and wisest umpire says, "Some is strikes and some is balls, but they ain't nothin' until I calls 'em."
In theory, a strike is a ball thrown in a certain region in front of the batter. Near the boundary areas, it becomes a little confusing to determine if a pitch was inside or
outside the invisible lines. Such classifications are somewhat arbitrary. In the same way, making measurements always involves a degree of classification and, to some
extent, these classifications are also arbitrary. If a shaft is out-of-round, an experienced mechanical inspector will find a way to incorporate that fact into the
inspection report. Perhaps the average diameter will be reported, or the high and low readings. Regardless of the method used, the report is not a complete reflection of
reality. Measurement error studies begin to reveal some of that discrepancy. Conducting a measurement error study is easy. Take one of your parts and
measure it several times. One way to do this is to follow these three steps: 1. Select five or more parts and prepare them for gaging.
2. Choose two or more people to measure the parts and have them note any problems during the study. 3. Have each person measure the parts in random order without seeing the other
person's readings. 4. Report the difference between these repeat readings.
Although it's easy to conduct, explaining the study to someone who's surprised and upset by the results is a task better suited to a diplomat than to an engineer.
About the author Gregory P. Ferguson is senior quality engineer at Global Solar Energy in Tucson,
Arizona. He has published numerous technical articles and assisted in the publication of two books. Comments can be e-mailed to him at gferguson@qualitydigest.com . |
