Surfacing question

Do you really need the patches or wouldn't it be easier to somehow regenerate the whole surface from the measured map ? Why did you say "I would hate to have to go the route of complex sweeps and lofts" ? How about creating curves from points calculated from formulas and topomaps, then sweep, with a macro ? That's basically the approach in our MathSurf add-in

) which might help.

I find that using insert>surface>fill is a pretty good way to get a smooth patch.

You can do this by using a split line then delete face to get rid of the bump in question, and then surface fill with tangent edges.

If you do end up surfacing I suggest using splines as the quality of the surfaces seems much better.

hope some of this helps!

Lee

Hello Philippe,

The problem I see with going directly from a single topo map, which is developed with MathCad, is that we need to look at several maps to be sure we find a reliable, repeating pattern before we try to implement any corrections. Besides, the topo maps show hundreds of small errors that there is no need to correct. We are looking for the large scale, smooth, gross errors. How could formulas be generated from such a complex bitmap? Send me an address that I can upload a typical topo bitmap to - for your appraisal. I can also send a SWX model of one of our panels with the thin patches. Like I said, these do correct for the main errors we are concerned with, but the abrupt edges of these patches cause other problems.

I was thinking of something like a surface with zero offset to the paraboloid that had 'handles' I could 'pull' and 'push' portions of, up or down a few microns, in a controllable way.

Maybe sweeping the whole panel at one time could work as you suggest - its not something I thought of.

Sincerely, Jerry Forcier Forcier Machine Design snipped-for-privacy@comcast.net

Philippe Guglielmetti wrote:

Hello Lee,

This idea sounds promising. I have a million questions, not having messed with surfaces much. I guess I should experiment a bit to reduce the confusion.

I currently use split line to generate the boundary of the patches followed by offset surface of a tenth or so and then thicken back. Do you mean to delete the top surface of my current patch. If it is necessary to also delete the vertical wall of the patch, I could be in trouble, because it is so small that I can't zoom in far enough to be able to select it!

Sincerely, Jerry Forcier

Lee Bazalgette wrote:

alternatively, you could select the entire boundry of the panel and surface fill that. You can add control lines if you find that it isn't filling as you want. Sometimes this can be great, and other times it creates things that I suppose could only be considered as ART since they are of no use to a product designer!

Get Unigraphics or CATIA .....

.0002" seems like a small amount to me-maybe its going to distort/warp by more than that when its assembled and 'hung',and also change shape with temperate-are you sure you aren't being overly concerned with accuracy-this maybe is due to variations in the cnc itself and therefore patching is not going to help?? I understand large scientific bowls have many automated adjustment points on the back to compensate for this...dunno just a thought...

How large?

Did they get stress relieved? Aged?

Not on a lathe as the dieties intended?

How repeatable is it?

How much does it vary with thermal drift?

Did you rough and then remove all clamping and then do another setup to finish (thus removing clamping & stock stresses)?

How repeatable is the CMM? How did you calibrate the CMM to your process?

What temp are you holding for machining and measurement?

How accurate do you need to be? What is the maximum *accuracy* of your machining process?

Most CNC machines are programmable in a smallest increment of .0001: ... so the actual expected accuracy can be no greater than that. It might be a little better if you programmed in metric as .001 MM is smaller than .0001:

Can you just create your surface with many knot/control points and probe at those points? Then just adjust them by the error ...

BTW, What degree is the surface? A parabolic surface should be of degree 2 X 2 IIRC ... and so should any patches I suspect.

Lee,

How would you select the vertical wall if it is only .0001" tall? I zoom in but still can't 'see' it to select.

Jerry Forcier

Lee Bazalgette wrote:

Jerry, What frequencies are you dealing with here? Is this in the terrahertz? Smaller?

Anyways, SW can draw it. Can you find a machine shop that can make it? If so, I want their phone number cuz I have some jobs for them.

Jerry, I sent you an e-mail.

Sorry George,

I didn't get it. Or I may have trashed it, thinking it was spam. Please resend.

Jerry Forcier snipped-for-privacy@comcast.net

George wrote:

Hello Mark,

Of course the errors are not in Solidworks. They are a combination of many things; Mostly panel flex from tool pressure and stress relaxation during the cut, (even though the panels were thoroughly stress relieved, there is still some stress change). Tool galling is not the issue as the cutting edge contact is natural crystal diamond running at 4000 surface feet per minute and the metal slides off the diamond cleanly. The dishes I work on are between 12 feet and

100 feet in diameter. There are many individual panels (mostly trapezoidal) that make up a complete dish. I machine the panels (about 36" size), one at a time, in a tilted orientation so that the spindle is never normal to the workpiece surface. The machine is running constantly, in a temperature controlled room, so machine growth is not a big problem. Everything, and more, that both you and I have mentioned affects the measured accuracy. What I try to do is find repeatability in the sum of all these errors and carefully apply a correction factor. These panels are currently very good, but there is a repeatable error, and that is the .0001" to .0002" I speak of. I want to change my machining model to compensate for this. If I couldn't see this error in a repeatable way, it would be folly to go further, but I CAN see it and it calls me to fix it.

Sincerely, Jerry Forcier

Mark Mossberg wrote:

Does the machine's control have an internal errormap? If so calibrate it ..... and switch to metric programming.

Some controls used to have an error map capability ... calibrated every so many units of travel. Usually used with lasers to calibrate. The backlash issue is a different thing ... but speaking of same I trust you are always cutting from the same direction ..... if not then doing so may aid in getting a little more machine precision.

I know of them. Paul Elliot at nwdesigns.com.

I'd check you math though. In some cases the arc fitting and increase in decimal points MAY increase error. Certainly if your CLDATA file uses fewer I'd suspect. Truncation & round off issues then the arc fitting ... some assumptions

*may* be being made by the software (I don't know).

Did you mention what CAM software you are using? I don't recall.

It sounds like your surface should be, at it's root, a simple revolved parabola of the analytic form Z = X**2 + Y**2 .... knowing that it's a power of 2 should help with the surface, as I mentioned before.

Check your CAM side tolerance settings !!! It may *look* smoother due to the arcs but actually be worse in tolerance. The arc fitting is generally used to reduce the amount of data (program size) and smooth out "jitter" when the axis feed rates change too fast for the machine. But they add tolerance otherwise IMHO.

In general the CAM side should have an inside and an outside tolerance. Inside controls how much stock may be left uncut (compared to the true surface) by a linear chord (G01) across the surface and outtol how much the surface may be gouged bu overcutting. The combination of these two controls determines how long a linear motion can be between two points on the curved surface (dependiing on how it's curved).

For your class of surface, BTW, I'd probably try to get tool motion in arcs about the axis of the parabola. Then most of yor output would be G02 (oe G03) moves at a constant Z. More sources of error removed, samallest programin memory, no need for arc fitting, Z depth precision limited by the number of decimal points in Z and the X & Y locatyions calculated by the controlto it's *internal* precision which is always (AFAIK) much higher then the input program's.

Did the 'surface fill' option work in the end?

inaccuracies.

Jerry,

Now I understand, your trying to correct a repeatable machine related phenomenon in geometry. SW doesn't really give you that kind of control over surface curvature. What you need is to locally deform an (irregular ??) area, without loseing the mathematical accuracy of the surrounding surface ? The only program I (personally) ever used that's capable of that is Catia. Somebody else may want to chime in about Rhino.

Sounds like a hell of a set up. I'd be very interested in seeing it. I've seen micron finishes done on a lathe with diamonds, but never on a VMC. Does the surface come out finished ? Or does it require polishing ? Even with the enhancments to your machine, it's amazing to me that you can come as close as you do on a part that big.

These are trapaziodial "radial" sections ? Like a slice of pie with the pointy end cut off ? How thick are they, and how do you dampen them ? Is the back side conssistent enough to use a poured urethane vacuume nest, or something similar ?

Allot of what Cliff said about arc fitting is true. It can actually make the surface less accurate, but smoother. However, with proper tolerance settings this shouldn't be a problem. I use arc fitting all the time in Mastercam (Basically the Northwood codes, but integrated).

Chasing tenths on a VMC is kinda like chasing your tail sometimes, I know. About all you can do is try to eliminate variables up front, untill you're only dealing with machine accuracy. problem is, at that scale, somebody farting in the next room seems to affect it.

Regards

Mark

trapezoidal)

Jerry,

Now that I think of it, you may be able to use "fill surface" if the error boundary is regular.

You would need to use parabolic curves as constraints and fuss with it a bit. You could use 3D sketch to create a grid of points on both surfaces, and constrain them in the X-Y to match your CMM points. Then you could tweak the curves untill the point to point numbers equaled the error. May work,,, dunno

Mark

You should use the "grid points" as the points at which the CMM measures the surface IMHO if you do things this way.

Then you need a *lot* of faith in what happens between & beyond them .

As a test I'd try (on scrap) measuring outside the points to be modified as well and compare those locations to the nominal math surface and then modify & recut & remeasure to see how the areas were affected by the modifications. Lots of measurements. And I'd NOT trust the CMM without a lot of proof .... you need the standard deviation of the CMM process (on blind data sets of similar geometry) too .....