Rules For Rounding Decimals 1

[ringster]

I have searched for rules of rounding and no luck. ANSI Z210.1 and ANSI/IEEE 268. Anyone have an idea where I might find a standard pertaining to rounding, that will satisfy ASME Y14.5 or Y14.41?

 

[CoryPad]

ASTM E 29 - 06 Standard Practice for Using Significant Digits in Test Data to Determine Conformance with Specifications

Regards,

Cory

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips Fora.

 

[KENAT]

Let's make sure I understand the question, do you mean something in the ASME standards that explcitly states that 1.50000000000000000... rounds up while 1.4999999999999... rounds down?

KENAT, probably the least qualified checker you'll ever meet...

 

[ringster]

I was thinking in terms of rounding fractional equivalents, like .0625, .01625,.078125 to either 2 or 3 places.

And no, I DO NOT propose using fractional equivalents on a NEW design unless driven by tool sizes or such.

The real problem occurs when the digit following the last place to be rounded to is 5. For example .625 rounded to 2 places. Do we make it .62 or .63? At one time there was a rule which apparently is no longer existant.

 

[fcsuper]

In school, I learned an unwritten rule that says you should round a "5" to the even number unless there is some specific reason to round to the odd. I was never given a reason for this rule. I also do not believe it is explicitly stated in any standard.

There are standards regarding this kind of rounding in the Statistics principle, but I do not know what the stadards are. Though, I do not believe they are applicable to engineering drawings since (from what I understand of them) they make some assumptions that may contradict ASME Y14.5M's rule regarding the absolute nature of dimensional tolerancing.

Matt Lorono
CAD Engineer/ECN Analyst
Silicon Valley, CA
Lorono's SolidWorks Resources
Co-moderator of Solidworks Yahoo! Group
and Mechnical.Engineering Yahoo! Group

 

[KENAT]

ringster, I think the same or very similar came up fairly recently, may have been a ringman post. Take a look.

KENAT, probably the least qualified checker you'll ever meet...

 

[KENAT]

thread1103-224494 might be relevant, I thought there was an even more relevant one but my memory may be playing up.

KENAT, probably the least qualified checker you'll ever meet...

 

[CoryPad]

ASTM E 29 answers the question of rounding a 5:

6.4.3 When the digit next beyond the last place to be retained is 5, and there are no digits beyond this 5, or only zeros, increase by 1 the digit in the last place retained if it is odd, leave the digit unchanged if it is even. Increase by 1 the digit in the last place retained, if there are digits beyond this 5.

Regards,

Cory

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips Fora.

 

[CheckerRon]

What corypad just wrote makes sense. When I fix decimal places on my HP scientific calculator it always rounds up when 5 is the last digit (programmed that way).

 

[fcsuper]

Checkerron, The ASTM quote says to round to the nearest even number, not to always round up. Granted, it does say this in a very round-about way. I'm glad to see that it does appear in some stardard, though I question the usefulness of this rule. Any thoughts?

Matt Lorono
CAD Engineer/ECN Analyst
Silicon Valley, CA
Lorono's SolidWorks Resources
Co-moderator of Solidworks Yahoo! Group
and Mechnical.Engineering Yahoo! Group

 

[btrueblood]

fcsuper,

I learnt it the way you did, and the ASTM standard says...A QC type a long while back, whose opinion I trust, noted that rounding up the 5 creates a statistical bias in the data, slight but measureable. Rounding only to the even number (i.e. rounding down on the odd numbers) means that half the time you round the 5 up, half the time down, resulting in no bias or at least a more randomized bias. Made sense to me at the time, and have used the technique ever since. Glad CoryPad could point us to the reference, a star for him on that.

 

[ctopher]

FYI, Some info.
Please see pg.101, 10.2.2.2.

http://mscweb.gsfc.nasa.gov/543web/files/GSFC-X-673-64-1F.pdf

http://books.google.com/books?id=k8...a=X&oi=book_result&resnum=2&ct=result#PPP1,M1

http://books.google.com/books?id=JR...X&oi=book_result&resnum=6&ct=result#PPA114,M1

Chris
SolidWorks/PDMWorks 08 3.1
AutoCAD 08
ctopher's home (updated Aug 5, 2008)
ctopher's blog
SolidWorks Legion

 

[btrueblood]

Wups. I miswrote what I do, the language in the ASME standard and in the GSFC spec. are both the method taught and the method I follow.

 

[drawoh]

ringster,

I accept whatever my CAD software does. Anything else would be a lot of work.

Presumably, I am rounding .625 off to .62 or .63 because I do not need the accuracy. If I actually cared about this, I would specify the extra decimals.

JHG

 

[fcsuper]

Well, that makes sense for measurements then; though kinda moot for drawings, where I agree with drawoh's last comment.

Matt Lorono
CAD Engineer/ECN Analyst
Silicon Valley, CA
Lorono's SolidWorks Resources
Co-moderator of Solidworks Yahoo! Group
and Mechnical.Engineering Yahoo! Group

 

[btrueblood]

Fc, drawoh - Absolutely. As a designer, you can round the number to whatever you want. Presumably, if the difference mattered, you would alter the drawing tolerances to reflect same. But, if somebody is checking a part to your drawing and its tolerances, they should be rounding as per the standard, or not rounding at all.

 

[KENAT]

drawoh, you have greater faith than I. I've seen all kinds of fun caused by the rounding settings on the CAD not being what someone thought.

Also, if I put .63 on my drawing, I like to model .6300000..., not model .625 and 'round' to .63 at the drawing, again I've seen accumulations of this kind of thing cause issues.

KENAT, probably the least qualified checker you'll ever meet...

 

[drawoh]

KENAT,

One of the nice things about GD&T position and profile tolerances is that you do not need to round off. You can apply a dimension of .625", and a position or profile tolerance of .05". The drawing correctly shows your model, and it shows the accuracy you want.

Basically, if I apply a dimension of .62"±.05", an error of .005" is irrelevant. I would be tempted to round the model and dimension off to .60".

JHG

 

[CheckerRon]

I share KENAT's frustration with model dimensions that do not match the corresponding drawing due to rounding.
Frankly .625 ±.050 is a gross example that we seldom find.
When you are dealing in 3, 4, and even 5 place tolerances which I know KENAT does as well as I in my job, sloppy rounding, (often the result of bad modeling) becomes a real problem.

 

[KENAT]

Thanks Ron, you said it far better than I.

KENAT, probably the least qualified checker you'll ever meet...