Hello all. For whatever reason, the fillet engine refuses to make a .008" fillet on the edge of certain letters on the wheel pattern seen below.

The circular text was created in Adobe Illustrator, converted to outlines, and exported as an AI v8 EPS file. It was then brought into Moi, I ran the "rebuild" command with a tolerance of .001 and deleted the original objects. The text was then extruded, booleaned into the wheel and then the top face was booleaned to match the curvature of the wheel.

I filleted the top and bottom edges of the letters .008 using circular or constant radius as needed. Some letters run perfectly while others don't. Very odd.

Any ideas?

The 3DM model lives here: (6mb)

http://rapidshare.com/files/455115117/Washburn_wheel_profile_v6R.3dm.zip

Thank you!

Chris

Hi Chris, I'm taking a look at it and I don't see anything really obviously wrong with it.

The one issue that you may be running into is a fair discrepancy in scale - your main object being about 6.8 units across, trying to then put a fillet of 0.008 units on that is a difference in scale of 850 times between those features of the model.

A difference in scale approaching that kind of magnitude can cause some difficulties - like the edge tolerance for the boolean was probably done at 0.001 units and that may not have been tight enough for the 0.008 fillet to get a good result on top of that.

You might try modeling something like this starting at a bit larger scale, like 10x its current size and then scale the final result down. That may help gain some extra accuracy because things like the 0.001 unit accuracy that was done with the boolean union at that 10x larger size will have stuff calculated a bit tighter.


The only other thing I noticed was a little bit of wiggling in some of the shapes, like here things are a little wavy (subtle but a bit of undulation in here):



It could be better if these shapes were actually more segmented, with some more plain straight line pieces in there.


But I'm not really sure what is exactly going wrong though.


Do you happen to have a version of this model before doing the boolean union?

- Michael

I deleted the verticle surfaces of those letters, then ran a loft with the "straight" option, then rejoined them and they would fillet at .008.. It did work much better when scaled to 10X.

In each of them, one of the little corner patches of the fillet, would come out with a poorly defined surface and I would dlete it and then fill it with a sweep function, then rejoin to a solid....

FYI

I will look into both and report back.

Thanks!

C.

Hi Chris, I've still been looking at your model to try and figure out what's going wrong with the fillet. I haven't got it figured out yet but it seems to be failing in the phase of building fillet surfaces, I'll be looking at it some more to see if I can figure out what is going wrong. So far it looks like some kind of bug in the filleting mechanism though and not really any problem with your model.

If you need it finished right now doing some alternate construction method in those areas where it failed like as Burr mentioned deleting the side walls and putting in a loft there sounds like it will be enough of a different structure to avoid this particular bug.

- Michael

Hi Michael. Burrman's suggestion only worked on some of the letters. Baffling! LOL

Thanks for trying. Luckily, the deadline is far enough off on this piece to hopefully figure it out. I seem to be really taxing the fillet engine with my work!

Chris

************************
Earlier versions of the model:

http://rapidshare.com/files/455499205/Washburn_wheel_profile_v5R.3dm
http://rapidshare.com/files/455499223/Washburn_wheel_profile_v4R.3dm
http://rapidshare.com/files/455499266/Washburn_wheel_profile_v3R.3dm

Face used to Boolean the tops of the letters
http://rapidshare.com/files/455499310/Washburn_wheel_face_only.3dm

Hi Chris, thanks for posting those previous versions, those are helping me when examining this. I'm still working on it.

I think that one strategy that may help a lot when doing this kind of filleted raised text stuff in the future would be to actually pre-fillet the sort of "side wall" edges of the text before you even boolean the text into the other pieces. Ideally do it actually on the curves before even extruding them, and use a slightly larger radius than the final fillet will use.

That will likely make the last fillet on the final booleaned result to be quite a bit more simple in structure because it will be working with all smooth chains of edges and take a lot of multiple-way junctures out of the picture.

By "side wall" edges I mean these ones here:



I bet having those prefilleted before the booleans and also just slightly larger than the other final fillets would help quite a bit. On the next project like this you should probably give that method a try first and see if it goes well.

I'm still working on your current model here though - there seems to be a couple of different bugs that are coming into play with some of those particular shapes.

- Michael

Hi Chris, I've attached a filleted version of your last revision. I think it should work for you but please take a close look over it to make sure.

So in digging through what was going on, it seems like the filleting engine is deciding on some of those letters that are made up of longer surfaces to calculate the wrong fillet, one that doesn't match up with the others.

Imagine for example if you have 2 surfaces that extend through each other like an "X" type shape, there are 4 different possible fillets that can be put into an X like that, like:



It's not quite exactly like that but it's similar to that - and when the surface/surface fillet is wrongly chosen for a particular edge it disrupts the whole fillet because that piece does not match up with all the other parts that were calculated correctly.

Anyway, this ends up happening in your cases here where the surfaces are kind of long, so to get it to fillet right now I extracted the faces for the pieces that did not fillet in your last revision file, and untrimmed them so that they were in the pre-booleaned state, then filleted the side wall directions with a slightly larger 0.01 unit fillet to do that technique I wrote about just above. Then unioned it into the main piece and trimmed off the outside part. Some of them would now fillet at 0.008 after that, but some of them needed to be split up into a few smaller surface pieces in order to avoid that bug I was describing here. So for example the Y letters I split at the tips of their rounded ends.

It's going to be a longer and more involved process to try and track down and fix the actual bug, and I'm probably going to wait to pursue that further until after I merge in the most up to date version of the geometry library for MoI v3 which has some various filleting fixes in it as well, which hopefully will generally improve things but I have yet to see about that.

The bug may also be some tolerance thing related to the small size of the fillets, I'm not entirely sure about that though. But it could possibly be good to build this at 10x your final size and then scale the final result by 1/10, if you are going to have a lot of fillets at around 0.01 units in size.


But anyway the attached version should get you your model done for now, and it's probably a good idea to use that "pre fillet the side wall edges before booleaning" technique as described previously when dealing with this kind of text filleting in the future, I think that technique does help simplify things for the filleter.


- Michael

Michael, as usual, I cannot thank you enough. I genuinely appreciate the amount of time you have invested into tracking down the source of this problem.

Also, as usual, my workflow has now been improved. Previously, I added the "rebuild" command to my working-with-text-and-curves arsenal. I will now add "pre-fillet" to that as well.

I've said it before, but it bears repeating: I will make changes to MY workflow to suit the software. That is how Moi will stay on target and not wander down the misguided path of Solidworks, etc.

Best,

Chris