What's new
What's new

OT-Precison vs acccuracy

Joe Gwinn

Stainless
Joined
Nov 22, 2009
Location
Boston, MA area
Generally you need good accuracy and ​precision if you want to do good work.

Yeahbut, what about the official definition of a meter, when it was two scratches on a long polished metal bar? Its accuracy was 100%, by definition.

So what's the precision? Basically, the width of the scratches. But but but - what about measuring things less than a meter long, and the like? Maybe 100 subdivisions - we'll call it a centimeter. So, the precision is one centimeter?

Uhh, no. That's the resolution then. Huh? Everybody calls it precision.

Lost down rabbit hole.

So what to do? All these terms have been used, generally loosely, and their definitions have become blurred, even if Marketing wasn't involved.

Anyway, the definitions I prefer are that precision is basically the number of significant decimal digits, and accuracy is how close the measurement comes to what it claims to be. Accuracy is always relative to some standard, which can be local and not official. Or it could be the international standard meter (which is no longer defined by a scratched metal bar).

 

EPAIII

Diamond
Joined
Nov 23, 2003
Location
Beaumont, TX, USA
Well, speaking about "resolution" I have to observe that your image is of poor resolution. Therefore it can not be read and it makes no point other than higher resolution would be a good thing.



I dislike the word "precision" because it is used so loosely. I was raised using 3 words: repeatability, resolution and accuracy. I made a chart similar to eKretz's in an attempt to teach high school students. The context was a discussion of the Phalanx anti-missile gun on the USS Stark, and its failure against the Iraqi exocet missile, circa 1987 (so many things to learn: Saddam was our friend, France makes impressive weapons, stuff doesn't work if it isn't turned on). Here are some of my slides. Summary is: precision = resolution.
View attachment 339444
 

CalG

Diamond
Joined
Dec 30, 2008
Location
Vt USA
The way I remember it...

Accuracy is how close you can get to the desired size, and...
Precision is how often you can do it. :D

Dennis


the way you remember these qualities is flawed. Popular, but flawed.

Precision= group size
Accuracy= how near to the bull are the groups
Repeatability =can you get the same results across the entire sheet of bulls

Is a jig borer more accurate than a bridgeport? Or is it more precise?
Is a digital display that reads out to 4 places right of the decimal more "accurate" than a handwheel dial, or is it more precise?
Is a basket of thermometers that display values "no two alike". inaccurate? of imprecise.

A cock that is broken, is Accurate twice per day!

Accurate takes a target. Precision does not. I can be precisely off the target. (intentionally)
What is improved via a "clean up pass"? Accuracy or precision. And can it be repeated? How?

Think on these things!


I'm not understanding why "common misunderstanding" should be so prevalent in this group of professionals.
 

mhajicek

Titanium
Joined
May 11, 2017
Location
Minneapolis, MN, USA
My Haas has a resolution of .001mm. Yes, it will move about two and a half clicks per tenth on an indicator, back and forth, with no perceptible backlash under zero load (though I can push the table a few tenths by hand.) When I throw in a tool and cut, with no offsets, it will cut precisely and repeatably the same amount oversize, within about a tenth, until I dial it in. Adjust the cutter offset to the midpoint of tolerance, and then it will cut accurately and repeatably until the tool begins to wear.
 

TDegenhart

Diamond
Joined
Mar 26, 2011
Location
Geneva Illinois USA
I would try to explain it this way to someone who isn't in our line of work:
Precision - to make a part exactly to size one time.
Accuracy - the ability to repeat that size on a part over and over and over.

Actually you have it backward. Accuracy is to meet a standard value. Precision is measured by the statistical variation from that standard.

Tom
 

SeeFair

Aluminum
Joined
Oct 20, 2015
Well, speaking about "resolution" I have to observe that your image is of poor resolution. Therefore it can not be read and it makes no point other than higher resolution would be a good thing.

Fair comment. So here's my definitions, with regard to measurement:
Resolution: the smallest resolvable change using the device. For a ruler with 10 tick marks per inch, the resolution for the average person is 0.01", or one tenth of a tick mark (estimating between ticks).
Repeatability: the tolerance band for repeated measurements of the same object. For the example of the ruler, it might be 0.02" (do you always estimate the same?). For several people using the same device, the repeatability number maybe gets larger.
Accuracy: how close the measurement is to the "true" value. Not as good for a harbor freight scale as for a starrett, even worse for the stretchy suspenders with the inch marks on them.

I used laser feedback on a machine in open air a long time ago. Superb resolution, but repeatability and accuracy not so much, because it measured lots of things besides distance (air pressure, air temperature, air composition, etc. anything that changed air index of refraction). Putting shrouds on the beam path helped a lot with repeatability, and temperature compensation helped with accuracy. For an accurate, repeatable laser, it needs to run in a vacuum bellows (unless a lot has changed since then).
 

Joe Gwinn

Stainless
Joined
Nov 22, 2009
Location
Boston, MA area
Mostly (not always) a gaussian distribution. This is not well-understood by those who specify unilateral tolerances.

Uniform distributions are also common in practice.

The weirdest distribution commonly seen in practice is Gaussian or uniform, but with the central ~tenth of the range punched out.

This most often happens when there is a tighter-tolerance grade sold for more money - it's done by cherry-picking the run of production.

A common pattern is that the run of production is Gaussian, and what is sold as the lower-precision grade is Gaussian with central region missing, and the higher-precision grade has a uniform distribution the same width as the hole in the Gaussian.

If there are more than two precision grades, uniform-with-hole distributions will appear.
 

hanermo

Titanium
Joined
Sep 28, 2009
Location
barcelona, spain
Great thread.
Imho it conflates the 2, like always tends to happen.


Accuracy is the ability to make, or measure, a part or feature of a part to an established standard, like a perfect inch or meter or whatever.
A perfectly accurate meter is the exact same length as the one in the museum (arts y metiers iirc)I visited 3 years ago in france.

Precision implies accuracy but does not guarantee it.
Precision is making a needle valve for an ic engine, or a lens for a home telescope.
The lens is probably precise in its curvature to about 0.05 microns, or 0.0005 mm, or 0.0000025".

But the lens, or valve, is not necessarily accurate, because it does not have an established size, measurement, radius, or curve profile.
They can have excellent fits, and fantastic resolution, and amateurs routinely fit stuff to better than 1 micron, often better than 0.1 micron in telescopes etc.

Precision is usually badly used.
It should be == accuracy, and resolution used for machine tool stuff.
Resolution is the min. size you can move, or make a feature like a divot, whatever.
If you can make a cut of 0.001 mm your machine tool has a resolution of 1 micron.

This does not mean that your machine tool can cut a piece to 10.001 mm - not at all.
Or that it can make 3 pieces with 0.001 mm ledges, of size 10.001 mm, for example.

Resolution is a better word than precision.
Accuracy is a good word. It says anyone anywhere can measure the piece and come up with a result that others can repeat if needed.

Most / many here could make flat ledges in steel of 0.001 mm feature size.
But they mostly would only moderately match ledges cut by someone else.
They would have precision and resolution, but would not be accurate to 0.001 mm levels.

Many here could also work the ledges to be generally accurate to 0.001 mm.
This would be specialist work, and a customer would pay a lot of money for it.
 

jim rozen

Diamond
Joined
Feb 26, 2004
Location
peekskill, NY


Uniform distributions are also common in practice.

The weirdest distribution commonly seen in practice is Gaussian or uniform, but with the central ~tenth of the range punched out.

This most often happens when there is a tighter-tolerance grade sold for more money - it's done by cherry-picking the run of production.

A common pattern is that the run of production is Gaussian, and what is sold as the lower-precision grade is Gaussian with central region missing, and the higher-precision grade has a uniform distribution the same width as the hole in the Gaussian.

If there are more than two precision grades, uniform-with-hole distributions will appear.

That used to be carbon composition resistors - 20, 10, 5, and two percent tolerance. They would paint on a red, silver or gold last band - or no band at all.
 

CalG

Diamond
Joined
Dec 30, 2008
Location
Vt USA
I suggest we all apply the three "quality" terms to a round of golf. Linear measurement is not the only field of endeavor.

Precision
Accuracy
repeatability.
 

Lewie

Cast Iron
Joined
Jun 7, 2018
Location
Albuquerque NM
That used to be carbon composition resistors - 20, 10, 5, and two percent tolerance. They would paint on a red, silver or gold last band - or no band at all.

:-) As a retired Electronic technician, and later, engineer , That takes me back some 50/60 years. :-)
...lew...
 








 
Top