One opinion- Do NOT use 2007 for curvy models

I was working on my presentation for SWx World 2007 tonight (the basics
of loft) and I ran into a major, I believe undocumented, change in how
lofts work.

Just to be sure, I upgraded to SP2.1, deleted the loft and recreated it
with the same curves (those of you in the know will understand why) and
the issue is still evident.

The issue is a little esoteric, though folks dealing with lofts will
know right away what I am talking about and know it is HUGE. It has to
do with how tangency is handled at the beginning and the end of a loft.

In previous versions, when executing a loft tangent to existing faces,
the loft edges (or when merging tangent faces, the UV lines) would
follow the vector of model edges adjacent to the profile (as long as
that angle was greater than 60 deg - if it was less, the vector would
be perpendicular)

So, for instance, if your start and end profiles each consist of two
segments, but there is an angle between the junction of the two profile
segments and the adjacent edges on the tangency face that created that
junction, when using 'tangency to face' as a start and end
condition the loft would follow that adjacent edge's angle and make a
very pretty surface/face (respecting the undocumented 60 deg limit -
over 60 deg it follows the adjacent edge, under 60 deg it just goes
perpendicular).

However, it wasn't perfect yet - when using 'curvature' instead
of 'tangent' as the start/end condition, the loft always starts
perpendicular to the loft profiles regardless of the angle of adjacent
edges, and that is one of the reasons we can so rarely use curvature as
a tangency condition- in most cases that (at least I) run into, that
initial vector causes self-intersection problems.

I have made this issue known to the developers at SWx in the past, and
to their credit, they have addressed it in 2007... but only for
'boundary surfaces'!!
There is an option in the dialog to 'align with other geometry' in
the boundary surface PM.. However, in typical SWx fashion, they only
got part of the message - it is a drop down option that is no option
- you can open the drop-down , but there is only one option to choose
from (align with other geometry). So if you use boundary, it follows
edges adjacent edges regardless, even when curvature is invoked. But
it no longer obeys that 60 deg limit - I took one test down to 45 deg
and it still followed the edges.

Boy, that will be hard to explain at SWx World

In the real world, as I tried to explain to the developers, there are
times when you want your new curvy face to follow the vectors of
adjacent edges, and there are times when you don't - that arbitrary
60deg limit was always crap, but at least it could be relied upon.
Instead they left it for boundary and took it away for loft - instead
of fixing the issue, tangency is always perp to the profiles unless you
use boundary, where it is always tangent to adjacent geometry (at least
in my initial tests)

So here's the summation -
Short term: I now have to risk being that presenter 'bashing' the
new version at SWx world, where I am ethically obligated to let folks
know that there is an issue that they need to be aware of that is
solved by not upgrading - using surfacing techniques is Ok because
you can use boundary, but solid lofts I tested (or surface lofts) are
simply FUBAR. In addition, I have to make the samples in 2006 and 2007
so I can show the difference, and take time to show that difference
because of an inopportune change
..Long term: I will have to continue to advise my clients not to upgrade
to 2007. Always good advice before a few service packs get through,
but here a mission critical functionality was lost when it comes to
modeling curvy shapes.

Please let me know that I missed something. I really hope that htisis
a f*** up on my part, and that the software I rely on didn't take an
inoportune nose dive on something this important

Ed.

P.S. -why wasn't this caught in Alpha and beta? Simple - most
folks just open existing models, CTRL+Q, and see if there are any
issues. And I have to admit, SWx 2007 is exceptional - it rebuilt a
2006 database two weeks ago in a test that I ran that I would have
never expected to pass through a version change.
However, when it comes to lofts (and likely other features, but I only
know for sure about lofts) when you upgrade SWx it still uses the old
algorithms to calculate those lofts (with exceptions that I am sure
Paul Salvador will provide).
I will note this quirk in my presentation in New Orleans.
For an example, if you open up a SWx 99 loft in 2007, it still uses the
SWx 99 loft function - evident by the lack of options for start and
end tangency and guide curve influence that have happened since. And I
freely admit that this is smart on SWx part - it is a way to preserve
legacy data, where tools have been cut and whatnot off of the legacy
data
Unfortunately, if you do alpha or beta testing and just rebuild
existing parts, you won't find issues in how the features have
fundamentally changed because if you just Ctrl+Q, you are still using
the old stuff, not the new stuff.
I'm a bit of a pain in the ass when I test because I always go 'off
script' - if they ask me to test one thing, I try to jigger the
test to what I would actually do. If I am testing a part I did before,
I try to rebuild it sketch by sketch, using the new features. But I
have to admit I missed this tangency issues - and this one is HUGE.
Based on what I have seen, I cannot use 2007 nor recommend it for curvy
modeling (even with the groovy new features)

 

Oh Geeze,this is embrassing - after my angry tirade, I didn't make
clear what the issue was. Here it is- at least in my tests, all
'tangent' lofts in 2007 behave like 'curvature lofts' in that edges and
UV lines are perpendicular to the profiles. Tangent lofts no longer
follow adjacent edges, which was critical to loft quality and
preventing 'self-interstion'. Only 'Boundary' seems to have this
important code in 2007

 

Oh Geeze,this is embrassing - after my angry tirade, I didn't make
clear what the issue was. Here it is- at least in my tests, all
'tangent' lofts in 2007 behave like 'curvature lofts' in that edges and
UV lines are perpendicular to the profiles. Tangent lofts no longer
follow adjacent edges, which was critical to loft quality and
preventing 'self-interstion'. Only 'Boundary' seems to have this
important code in 2007

 

Hello Ed,

I'm trying to get my pea brain around the problem, but I'm at a loss.
As I do a LOT of curvy stuff, is there anyway you can send me a simple
model of the condition that you are talking about? I would like to try
it for myself and get some understanding.

Thanks,
Muggs

muggs
828
@
comcast
..
net

 

Yes, Pictures please.

 

Thanks Muggs... this is something that is hard to describe with words.
I got up early this morning to make a simple side-by-side loft to send
to SWx, but I still suspect that the 'simple' sample won't completely
communicate the significance of the problem. But I will be happy to
send the model to you later after a meeting I have (and Ken too - I
think I have your new email).
My weekend is booked, but next week/weekend I will likely come up with
more substantial samples.
Man, i hope I jsut f***ed something up, because if I didn't I will have
to use boundary for just about everything that requires start/end
tangency to multiple faces.
Ed

 

It is more than possible I f**ed comething up.

I still need to root through old models, but this mornings simple tests
show that this IS NOT an issue particular to 2007. I withdraw my
warning about 2007, with great chagrin

But, boy will this will be an interesting topic in new orleans.

It has to do with UV lines and what happens when you model one face at
a time vs what happens when you try to get a few faces out of the same
loft (be it surface or solid).

At this point it looks like I was completly blind to how much I rely on
modeling one face at a time. Just flat-out blind to it, and blinded by
it..

Too much to go into with words in a post - it will be worked out
intensely next weekend and what I learn will be (hopefully) distilled
into a few clear samples in my presentation (and of course posted on
our website) in February.

Ed 'stupid egg on face' Eaton

 

I read something once regarding lofts in the API documentation that was
a bit unnerving...

There was a note that said that SW has changed the algorithm to
interpolate lofts in nearly every release, and that part of the data
included in a loft feature is the version in which it was made, so that
SW can use the correct year-specific algorithm to regenerate the loft.

What's disturbing is that one can loft between identical sections with
identical guides in different versions of SW and get different results
in each version.

 

Yup, we hammer on SW about this. Makes it hard to edit or revise work
done in a previous version.

 

a lot of famous people have died recently LOL (tease,tease) )

 

Hey, quick question...
Is there any mention in the documentation about the algorithms being
preserved when a service pack makes a big change?
I know that it is preserved between releases, and plan to make a few
points about that in this years presentation. But I do not know if
this caching is restricted to just 'major releases'.
If you have any insight that you could share, it will be helpful
(credit to you if you want it).
If not, no big deal - I will only talk about what I can prove with
samples, and mention that I have no evidence about service packs.
Thanks,
Ed

BTW - I am torn on what to think about this issue. If they had to use
the same algorithm for ever and ever, they could not improve. If they
used the current one on every rebuild, legacy data could be fouled up.
On the flip side, the unacknowledged use of old algorithms can lead to
a lot of frustration - you expect things to work one way because of
current experience, but when editing older models they work in
unexpected ways because they use the old algorithms. I will point out
some things that will tell you if the old one is in use, but it is not
definitive.

This (and similar issues between releases) is also why I have adopted
the habit of saving a parasolid copy of my models when released on the
assumption that we lock in time a geometrically exact copy of what was
sent out the door, regardless of when it is reopened down the line - I
don't think that parasolids can change when opened in new versions (but
am not even sure about that)
Ed

 

Ed,

Is something wrong here when you have to spend hours experimenting with
the software just to figure out how to get it to do what you want it to
do? Is it possible that SW doesn't even know what it is doing?

 

Huh?
You don't want to have to redo a model using and logging subtle
variations of input geometry to see their effects on the result? And of
course constructing as both solid and surface geometry, and when doing
the surface version doing it all in one surface or breaking into
individual faces and multiple features, then log all of those results
against variables like guide curve influence, start and end tangency
magnitude, then redo everything again as a boundary surface and surface
fill to see what happens with those?
Then repeat the process on a completely different model to be sure you
aren't being misled by a red herring in any of the above (as I was
earlier int his thread)? And repeat everything again when a new
version comes out because anything might have changed?

Man... some people...

 

Funny thing Ed, I was in a cave this last week and had just that desire
to explore all the nuances of being inside one of the biggest examples
of 3D modeling ever created.

Fortunately the park Ranger wouldn't let me do that because we only
paid for the two hour tour. The two hour tour.

Happy New Year Gilligan.

 

Ed,

NURBS mathematics, and the algorithms for sweeps, lofts, etc., have been
"perfected" for well over a decade. They are pretty much standards, and are
available as C++ source code libraries. Some even for free. Solidworks has
been reinventing the wheel, and as a result, we can't rely on our shapes
being consistent from one release to the next. As a kluge fix, they
incorporate the previous stuff, but in some versions, (2001+ to 2003 comes
to mind) a simple CNTRL-Q rebuilds to the new algorithims. I nearly ruined a
$200K tool because of that one. The only thing that saved me was I compared
the Parasolid, sent to the tool maker, to the new model. I had to use 2001+
to model the additions because the difference in local curvature, in the
area in question, was over .010" when resolved in 2003. This means the data
has to stay in 2001+ format until they build a new tool. If I have to do any
further work, and I just might, I have to reload 2001+ to do it. This is so
ridiculous it's laughable. Because I've learned that I can't trust it, every
curvy project that goes out for tooling remains in the original format. This
is unbelievably clumsy to administer, and adds up to alot of non value added
labor.

I don't know of any other system that has "this" type of problem at this
magnitude. It's been going on for years, and we just bite the bullet and put
up with it.

Mark

 

This phenomenon has so little to do with development it would be laughable if
it wasn't so irritating, not to mention costly. Just like removing the
curvature comb trace curves instead of adding a $5 option button. Next
release will have the curvature comb option button and the one after that a
loft side curve influence option button. What we have here is deliberate,
sometimes not very well thought out milking excercises. At least Solidworks
has managed to stay on course instead of changing direction every few years
and calling that development. Functional Design. What a hoot. Maybe that's
the problem. If Solidworks had someone nipping at their heels they'd have put
the buttons in.

 

This issue cropped up in this forum a few weeks ago and I commented back
then that it doesn't have to be this way. I was a long-time I_DEAS user,
and it allowed you to explicitly migrate each individual feature to the
latest algorithm if you wanted to take advantage of new functionality. All
old algorithms are still incorporated in the software, so that old parts
will always rebuild correctly if you choose not to migrate.

I think this and many other "failings" of SWX are a function of the
corporate mindset of Solidworks. It was created on the basis of "forget
about those expensive unix based systems used by big organisations, here's
our new cheap offering to be used at home and by one-man jobbing shops". I
doubt they ever expected it to be so successful as a mission-critical system
in larger organisations, and everything about it smacks of a lack of rigour
in this regard.
It very much reminds me of the differences between Windows and *nix OS's -
look how hard it is for Microsoft to now make Windows a stable and secure
platform, where they end up releasing patches for patches.

John H

 

John....the example you use sounds like a nice feature of Ideas but UG
doesn't have it. Its a "big" player that owns Ideas now. UG handles it
much the way Solidworks does.....the older feature uses the old code.
No way to upgrade it that I know of.

 

Jason,
I was told at a UGS presentation that all the old I-DEAS algorithms were
being ported to NX, because otherwise it would not be possible for 100%
migration of all I-DEAS models to NX, which is their stated aim.

If the facility to migrate individual features has not yet been incorporated
into NX then I imagine it will eventually be, as all the best I-DEAS
functionality is gradually being integrated.
Examples of where they've already done this is with a complete re-write of
the NX sketcher, & incorporation of assembly configurations.

John

 

They may bring over the old algorithms but what I'm talking about is
the option you say Ideas has to "upgrade" the feature to the newer
algorithm. You criticize Solidworks for not having that option when UG
doesn't even have it either. Will they add that ability...I don't know.

Also....if the UG sketcher is the best they can do....then they need
more help. It still lags far behind the Solidworks sketcher. And where
are Assembly configurations? Between Part/Assy families, Arrangements,
Reference sets, and Mating Alternates....they don't play well together.
Then try throwing UGs layers in the mix and you have a very confusing
mix of tools that all work differently. Solidworks configurations are
way easier and more powerful.