This post is intended to provoke discussion on the thorny subject of blending. Let me first say that of the modelers I have used ViaCAD is proving to be one of the the least frustrating in this respect. That said, there are some caveats!

Attached is an incomplete model of a real casting, these were made in Germany in the 1940s and 50s. The purpose of my model is to scale down to 5 and 7 1/4 inch gauge and to produce accurate patterns for the scaled down castings. The point here is that more than 50 years ago a pattern maker made and BLENDED patterns from which beautiful accurate castings were actually made. The model is derived from the ORIGINAL drawings.

As with all pattern making there is always a certain amount license in what the drawing means when specifying blend radii. An example is that the 60mm rad cannot be extended to apply all the way up under the top flange because the resultant blend would then continue right up onto the top, where the blend is specified as 10mm. Ok, so the pattern maker would start with 60mm at the bottom and continue up with an equivalent, probably, constant width (approximately 20mm), because this looks right. So far so good, I can do this with ViaCAD.

As in the excellent tutorial (elsewhere) I blend the corners at right angles to obtain tangency then chain. In my case constant width up the two sides followed by 10mm radial around under the top flange. This latter aught to be constant width also (about 14mm) because it's going to run out into the acute angle between the flange and the horizontal body. Again this kind of works in ViaCAD, but takes forever, effectively breaking my model.

Note I want two fixed width blends at right angles to each other. This kills my computer which goes into 98% processor bound for ViaCAD, nothing else running except the system monitor at 2%. I have had this work, but takes hours! I'm running on a 700Mhz Mac G4 under OS X 10.4.11. If I save the resultant file and attempt to re-load then I observe a similar processor bind while re-calculating curves during the load.

Just to be bloody minded I even tried a third constant width blend between the top flange and the horizontal part of the body (should be 10mm). I've no idea how long this might take. I don't have the patience!

My solution (workaround) is to avoid constant width all together. Going completely against my experience, I come up with the result as shown in the attachment. First SINGLE edges under the flange using fixed radii, 10mm and guess at 5.5mm followed by 'By position' up the sides (simulating fixed width) and finally 10mm fixed at the front. The final result is not entirely satisfactory because there is an un-blended edge remaining under the flange which I can't get rid of without introducing nasty cusps. Incidentally the check box in options dialog to de-select chaining should be modal (thats what a check box is), not have to be un-checked every time a single edge is required. This throws me for every time.

Questions:

Is the exponentially increasing computation time when using constant width blends at right angles to each other a bug or a feature?

It often doesn't work blending mutually perpendicular edges in a particular order because acis refuses to cap or complains about radius. Error messages, in this respect seem, to me, almost random! I can often put an arbitrarily smaller radius where I'm sure a larger one would work, but acis says I can't! Is there some obscure acis rule? The pattern makers test would be 'can I roll a ball of the specified radius around the in internal corner without the ball jambing between the tangent faces'. The equivalent for an external corner would be a virtual ball rolling around inside the material.

As for capping, the most astonishing things seem to work while something simple to the eye gets rejected. Again, what are the rules? In my example the 'by position' blend which runs into the blend under the flange amazes me:eek:. I tried a fixed width here, but that failed, acis saying the blend radius may be too big. Well it isn't because I can do a it manually by position. Whats more it takes no time and looks right!

What then is the difference between constant width and by position. If only I had a computer and enough time, patience and geometry I could calculate enough points to give me constant width using by position :rolleyes:

Added:
What about the second example ... graphics_crash. This has the two 'fixed width' blends and appears to work at a reasonable speed until I select a shaded view. Now my graphics crashes. A bug here I think :eek:

Hi jdi
Not exactly but close. Using constant radii works fine for me too.

The 60mm radius up the sides should be something like 20mm wide and the 14 rad under the flange 14 wide, this should produce 10 rad under the flange.

What happens if you try to use fixed width blends? When I try it takes forever to do the second (14 wide), the 3rd (10mm rad in your model) just won't compute if I the others are constant width!

I note that you are doing the blends in a different order. The order chosen seems to have much influence on the ability of acsis to compute!

>>The order chosen seems to have much influence
>>on the ability of acsis to compute!

Yes, not just for ACIS though.

Blend large radius before small radius especially when doing variable radius blends.

Constant radius blends automatically use blend reordering to deal with the above.

I've requested Spatial add reordering to variable radius blends.



Tim

Hi Paul

No the file is not CORRUPT, just broken:eek: I you select 'display in wireframe' before opening it, it will load.

If you then try to display with shading or hidden line it will blow.

>>>>>>>>> The file is BUGGY
Brad

Hi Steve,

Correct me if I am wrong but doesnt VC quote that 'shared' edges are supposed to be G1 in any case (see manual) ?. In Rhino one can select edges by position, tangency, curvature.....this is how i would hope VC to be.

Funny thing I cannot get the error to repeat in Windows Vista. I can fillet 2,3, or all 4 edges with the automatic chain... option selected and no problem on any.



Jason
ViaCAD v6b810

>> This example clearly demonstrates the impossibility of chaining around >>non-tangent corners.

How about the linked movie below? I'm selecting an edge to roll about plus a face connected to the vertex to create what ACIS calls a vertex roll blend.

http://www.csi-concepts.com/Dem...ages/VertexRollBlend.mov


Tim

>>I admit I am still in the early learning process with VC/Acis, but do find >>your method and the need for such a method a little long winded for >>what I would normally do in selecting a rad size then 4 mouse clicks.

Yup:) First time someone has brought up a rolling blend since I started development. If rolling blends are something folks want, we can certainly improve the UI.

>>What as in the example you shown, the boss was 5 sided or 7 or 15 sided >>etc, then there would be a need to split the model for all sides?

That would be a problem as the current implementation for rolling blends is very limited. I got around the limit by splitting in the case of four sides.

>>Is this all down to Acis limitation?

No, I think I can come up with a better user interface to support rolling blends if this is useful.

>>Please also advise on the crashes I see as mentioned above.

Those were crashes inside ACIS which I will need to report to Spatial.


Tim

ACIS can blend across different continuities. Where it can have a problem though is blending across nearly G1. This is true for booleans and shelling as well.


Tim

Rolling blends are essential lost wax pattern making for complex castings. The corners produced, otherwise, often prevent easy extraction of the wax from the die. Also they just don't look right.