Scan to machine?

[revwarguy]

I need to be able to quickly take a part outline, cut in black cardboard to the correct dimensions, place it on a scanner, and have the computer scan in the image and create the gcode to cut out that part at the same scale. Can sheetcam do this? If it can't talk to a scanner directly, can it import a jpg from a scanner and do this? Ideally, it should have some controls over how much fuzziness to accept or reject in its line detection.

The idea here is to go from a scan to a cut part with few steps involved, especially avoiding making a dxf.

If not sheetcam, who does do this?

TIA,

[ger21]

Originally Posted by revwarguy If not sheetcam, who does do this? TIA,

No one that I'm aware of. I don't believe that SheetCAM does raster to vector conversion.

You can import a scan into V-Carve Pro, and then you can convert it to vectors and create toolpaths, but I think you'd need to scale it to the correct size, as I don't think the image contains any size info. Some formats do, but I don't think that V Carve would use that info. Also, the raster to vector conversion may change the sizes slightly.

Imo, drawing a .dxf involves fewer steps than you're talking about. You need to lay out your design, cut it, scan it, convert to vectors, then create toolpaths. For me, I'd draw it, and then create toolpaths. Most likely I could have it drawn before you can have it cut. Unless, you're talking about complex freeform shapes. But even then, how are you laying out the esign to be cut? Ther must be some drawing involved? Or are they just cut freehand?

There is another option. MeshCAM can load an image and create a 3D part. The image should preferably be 2 colors (white background). In MeshCAM, you can use the pencil toolpath option to try to just cut the perimeter of the part. This may work, but I would NOT recommend it, as it's not what the software was designed for.

[revwarguy]

Thanks, Gerry for the response.

My situation is that there already exists a bunch (hundreds) of part templates, cut from full scale drawings, for airplane part fixtures. Some are simple shapes, to be sure, but many are free form, complex shapes. Most of them will later be combined to form 3d fixtures. The idea of drawing them all in a .dxf is pretty daunting.

Thing is, I know I saw a guy do this once for use on a plasma CNC, back when I wasn't really sure just what I was seeing. I was so mesmerized by what was happening, you know - thinking of the possibilities, feeling overwhelmed by considering my own applications - just getting plain ole "hooked" on this CNC stuff - that I didn't notice exactly which steps were being taken. He's long gone, so I am still trying to find out what he used.

I do find it surprising that since there exists so many paper scaled drawings of things that haven't been e-drawn that this capability is rare! Not to mention the ability to just lay an actual 2D part on the scanner and reproduce it!

[Torchhead]

You have to understand the difference between a scanned image/photo/bitmap and what CNC machines cut. It may look like a smooth line when you draw it but blow it up in a paint program look at all the jaggies! When you get outside the "normal" size of scanners (typically 8 1/2 X 14) you are talking about a lot more bucks. I assume all these patterns are not that small. Bitmaps are thousands of dots that are individual objects. Regardless of what you see in the slick marketing of some vendors, very few items can be just simply scanned and cut. Usually the cleanup of the auto-trace (CorelDraw and Inkscape have built in auto-tracing programs) takes as long as if you were to just draw it.

One shortcut way we have used for years is to scan or photograph a shape (it can be as complex as you want) and import it into CorelDraw and put a layer over it and hand trace the shape using the vector drawing tools in Corel to get nice smooth curves and round circles. It's easy once you have it vector (line format) to measure the actual part and scale the drawing to any size you want. About the only thing that works pretty good for autotrace is simple high contrast B & W (line drawings to begin with).

The downside of the hand trace is it's slow going until you learn how to use the tools to "digitize the image".

It's not that the capability is rare but the results are often not equal to the time and money you put into the programs.

I'll bet you can find someone in your area with the skillset to do it on a contact basis for a reasonable price. It's an easy function once you know how to do it. The up side is you will get dimensionally correct results because you can size the vector (or change parts of it) at you leasure

A lot of users have been pulled in by the marketing of "just scan it and cut it...simple" only to find that is like the concept of the "Easy" button.

There are probing programs (MACH has one) than will go all around an object and over several hours give you a very accurate 3D model (which you can flatten into a 2D) There are also some lower cost laser scanner devices that will do the samd thing but they are pretty limited in what size they will scan.

[revwarguy]

Well, I thank you for the reply, but I suspect I do have a fairly good handle on the difference between bitmap image files and gcode ASCII files, having written software that uncompresses, rotates, does some edge detection using several different image formats.

My cheapie scanner easily does 9600 dpi scanning, allowing each pixel to easily represent a point along an edge within a thousandth. As I mentioned earlier, it is a black and white situation, so the color depth can be small and contrast thresholds can be high. As long as we don't use a image format that allows non-idempotent compression (i.e., don't use jpeg - did you know that just opening a jpeg, and saving it without making any overt modifications alone causes a loss of image fidelity! You can save yourself much of the "clean-up" from tracing by just not using a compressed image, and not "blowing it up.") we should have a good full scale representation of the part in the image file.

There are several image processing applications that easily does threshold edge detection that yields a clean line in this case, averaged by a radius of just a few pixels, so we know it can be done, just not by most CAM programs, I guess. And yes, the parts all fit on my scanner.

I believe placing a high contrast object on a scanner to produce a gcode file (or, as an option, an intermediate .dxf if modifications are needed) that reproduces this part to scale would be useful, and I haven't heard any argument that would prevent such software from working, as well as I think I saw it in action once, but just wish I had been paying better attention.

[ger21]

Originally Posted by revwarguy My cheapie scanner easily does 9600 dpi scanning, allowing each pixel to easily represent a point along an edge within a thousandth.

The problem is converting those "points" into lines and arcs. You really don't want a gcode file made up of .001 moves.

There are several image processing applications that easily does threshold edge detection that yields a clean line in this case, averaged by a radius of just a few pixels

Can you post a .dxf of what you consider a good raster to vector conversion?

[revwarguy]

"The problem is converting those "points" into lines and arcs. You really don't want a gcode file made up of .001 moves. "

Of course we don't. As mentioned earlier, it's a bit more involved than just connect the dots. Averaging, among other techniques, is used to make up the edge, or line. For instance, if there is a straight edge to the part, the "points" would either fall on or off the line by a few points in either direction (say five or six). However, if the software detects that there is this slight variation over an inch (say five or six over about 10,000 points - one inch's worth) then they are averaged, and you get a one inch straight line vector that is very close to the original, not a series of vectors made up of .001, .002 or .003 etc. moves. Non linear edges also have several published curve fitting techniques to do the same thing for more complicated shapes. Sharp corners can be off a bit more, but we are still talking only by a few thousandths.

"Can you post a .dxf of what you consider a good raster to vector conversion?"

I'm writing this away from my usual haunts right now, but I will keep my eye out for something. In general, though a resultant .dxf shouldn't be that much different than one drawn by hand, except the dimensions would be subject to the rounding and averaging techniques above. In any case we are still talking about a resolution error less than or close to what many CNC machines can cut. Complex curves may be comprised of a few more vectors than one might have done by hand, i.e., if you recognized that one curve was a perfect hyperbolic form and hand specified it, but so what? Multiple vector curves might complicate the toolpath a bit and increase cutting time on the machine a bit, but that's why I built a robot - to save my time over its time.

The above techniques tend to worsen as the image resolution gets less [which is why many find the process wanting - they weren't using enough dots, or they started from an image with 65,000 levels of contrast instead of a just a few, or they started with an image with lots of induced noise from compression - like jpegs!] or as the features on the parts get small. For instance, I wouldn't consider this for doing PC board layouts unless the original is *much* larger than the end part. For many parts, having things like notches that are 10-50 thou on a side are pretty rare, and tracing would probably fall down there.

What's the difference between using a scanner, which uses steppers to position a detector to see a change in light edges, or using steppers to position a probing switch to follow an edge around a part - except that the scanner is much more accurate and tremendously faster?

PS - good discussion!

[ger21]

Originally Posted by revwarguy In general, though a resultant .dxf shouldn't be that much different than one drawn by hand,

From my experience, they're usually not even close. I guess with more primitive shapes I may have seen some halfway decent results, but not good enough for me. That's why I requested a sample.

[iamstink2k]

adobe illustrator can take a scanned image and create a vector image.
you can set parameters to set the quality of the vector, to what you need.

then you can save as a dwg. and clean it up with a cad prog then create a toolpath (g-code)

[revwarguy]

adobe illustrator can take a scanned image and create a vector image.
you can set parameters to set the quality of the vector, to what you need.

Yes, it sure can. I was hoping a CAD program could as well. Some try, but the general opinion is that the results are poor.


Gerry, this discussion has nothing to do with Sheetcam now, but I am interested in pursuing it and posting some examples. Can you suggest a better forum and explain what is needed to move it? It will be a few weeks before I am back at my workstation to do this, however.

[notselrach]

I am building a jig for bending the sides of a ukulele. I traced the uke on a piece of white paper, took it to fedex and had it reduced by 50% so I could work with on my scanner at home. Once scanned I used photoshop to clean up the image and saved it as a .bmp I used the same resolution at all stages 300 dpi. I vectorized it with Win Topo and saved it as a dxf. When I used Cut2D to route out the shape it was not the same. I don't understand what part of the process has possibly changed resolution or dimension? I don't know what to change to make a more accurate result. Any help would be appreciated.....Thanks, BC

[mcphill]

Can you post one of your scans? I have a couple programs that convert bitmaps to vectors, and by tweaking the parameters it can smooth it more or less.

Though "tedious" the best way I find to do this work is to import the bitmap, and use drawing tools to recreate the part. Splines work great on the "organic" shapes, and circles and straight lines are "guaranteed" to be circles and straight lines. The importance of the accuracy of what you need will dictate how precise you will need to be, as you are fully aware.

Anyway, if you can post one of your parts, we can all "try" to convert it in to something usable to you, and you can determine if any of those options will work for you.