relative/absolute accuracy

[borf]

i think i understand relative accuracy now (pro-e strongly recommends it) and am left saying "what were they thinking?"

http://www.eern.com/proe/tips/accuracy/index.htm



after trying to diagnose many geometry "errors", im tempted to do two
things: switch to absolute accuracy and use a higher accuracy
altogether. i would ask this first though:



1 what type of accuracy do those (who are aware) use (absolute/relative)

2 is increasing the accuracy sometimes necessary or a sign of bad modeling.

4 accuracy affects the model display. do display opions affect accuracy?

5 are there any drawbacks to increasing accuracy except higer regeneration time.



its hard to approach a class "A" surface (what im trying to imitate)
when gaps falsely appear, or a dihedral analysis reports false
information, lines go to zero length or dont connect ect ect...and all
on an inconsitant basis. im just wondering how others have dealt with
it.

 

[jeff4136]

I don't claim to "know" much about the subject, but some suppositions I rely on...

I think the rel accuracy works well for a majority of purposes, but it does cause problems given certain sets of circumstances.

I have all my commonly used templates setup for an abs accuracy of .001". I sometimes use .01" for "class A ish" surface work as I'm looking for "lighter" surfaces and surfaces are created in a manner that does not tend to pose difficult to calculate edge coincidence problems. (I will typically toss these out to Rhino to evaluate surface structures and edge coincidence, though.)

Accuracy settings do not affect display and vise versa. Look at display settings; curves and shade quality for display issues.

Increasing accuracy will increase regen times and file size due to denser, higher precision polynomial (NURBS vs. analytic) definitions.

To fall back on a Rhino philosophy; set accuracy an order of magnitude smaller than the smallest feature unless some specific requirement dicates otherwise (works for me, anyway). There shouldn't be any reason or benefit for the "average" modeler to set accuracy to values smaller than somewhere in a 4E-4 in., 1E-2 mm abs range. (Contrary to myths propogated by Autodesk and the like. ACIS defaults to a .001 resfit value and "tolerant" edges go wider than that. The 1E-06 stuff is pretty meaningless.).

I would first get a handle on the display issues (simply set to highest quality) and then do a little experimentation, evaluating surface and co-edge characteristics, downstream operations, etc.

It might be worth noting that accuracy settings do not have a simple, straight forward effect on geomety, e.g. a .12345 rad circle (hole on a planar face) at 12.12345 will be exacly that to approx 10 significant digits of accuracy regardless of user accuracy settings. Accuracy does affect coincident point evaluation and polynomial curve calculations.

 

[Brian_Adkins]

The benefit of relative accuracy is that Pro/E will only try to be as accurate as it needs to be. It would be very inefficient to treat every part with the same degree of mathematical precision, because parts like washers and plates simply don't require it.


PTC has a good article on "Detailed Information Regarding Model Accuracy" here: http://www.ptc.com/appserver/cs/view/solution.jsp?n=32869


The article on EERN is a pretty good one (I asked for Oleg's permission before posting it there long ago). There have been some threads that dispute some of that article's technical accuracy, but I still think it does a good job at helping people understand the difference.


I would say we use relative accuracy on about 95-99% of our models. This is because we model mostly sheet metal, hardware, tubing and other 'simple' shapes. For our stylized surface parts (hoods, cowlings, external fuel tanks, etc.), we recommend absolute accuracy. This is especially true when using any type of 'master-model' techniques where multiple parts of different sizes are related back to a single 'master' surface model.


-Brian Adkins

 

[jeff4136]

Thanks, Brian. I'm always on the lookout for info of this sort.

Where you use abs accuracy; what value, why that particular value (if you have the time and inclination)?

 

[borf]

thanks, i wanted to be sure before using it.

ill try to give an example of the false display mentioned. this happens with low accuracy settings in general not just relative
accuracy.



zoom in on a surface (display on high) you will normally see the
boundry curves a little off thier surfaces in some places. you might
even see surface gaps. you might get a dihedral analysis reporting
angles where there are none (im assuming tangent surfaces have no
dihedral). these "fix" themselves when raising the accuracy -
ususally .0001. so you see its hard to diagnose problems with this
information. right now im trying to get two surfaces to merge.


Edited by: borf

 

[chf65]

From a manufacturability point of view, I cannot disagree more.


By recommending theuse of relative accuracy, we are totally forgetting its effect on manufacturability.


Relative accuracy of 0.001 means absolutely nothing when it comes to required machining features. Of course, it hasan "equivalent absolute accuracy", and if we know what is the largest diagonal dimension we can derive a meaningful relative accuracy. But this is not how people think about it. Accuracy setting has great effects on edges and surface intersection algorithms, which are usually of great concern for manufacturing.


For high precison parts, it is best to think about setting absolute accuracy to 1/2 the manufacturing capabilities. Usually, 0.0005 for mm and 0.00005 for inch.


For lower precison parts, you can get away with less.

 

[lbotez]

My 0.00002 cents: if all your parts are not dimensionally spread all over then use absolute accuracy. I am using 0.001 (mm) for all my parts and everything works just fine. Most of them are automotive engine components so you can figure the dimensional range that I'm working with.

 

[albino]

my accuracy absulote is 0.007 who i can change for 0.001?