With two way digital communication what are your modding plans?

[quertz]

Hey Slatye,

I think a camera would have just a way too small resolution.
If the image would cover the build area and nothing else, you would theoretically get HD resolution.
This won't be the case as you also have to analyse the image (the laser point will look a lot larger that it actually is) plus you will have calibration errors (skew camera, ...).

My current problem is that everything I can think of requires precise optics...

[jontelling]

Hey Slatye,

I think a camera would have just a way too small resolution.
If the image would cover the build area and nothing else, you would theoretically get HD resolution.
This won't be the case as you also have to analyse the image (the laser point will look a lot larger that it actually is) plus you will have calibration errors (skew camera, ...).

My current problem is that everything I can think of requires precise optics...

I originally thought this but I’ve done some numbers (some made up).

If you had a CCD (or CMOS) of 15M and every pixel is sensitive to 405nm (they're not as they have filters for each colour) that would give you a square of 3872pixels.

The peachy at 18bit per channel is 262143 bit per channel. So, the dynamic range would need to be 67points per CCD pixel.

From what I could find based on my old Canon 40D the CMOS is 16bit.
So even if we only listen to the blue pixels the location of the laser could be worked out as we know there’s only one laser per image, that’s ignoring focusing of the camera..

food for thought.

[Slatye]

The nice thing about using cameras is that there's a bunch of research already into making cheap-and-nasty cameras (with horrible optics that cause odd distortions) behave like "ideal cameras" (zero distortion).

I was thinking about using them along the lines of how the compass is used in UAVs - it's not updated very frequently (10 - 80Hz or so, compared to 400Hz+ for the main control loops and inertial measurements) and when it is updated it's not very accurate (to within a couple of degrees if there's not too much magnetic interference), but you can be fairly confident that it doesn't drift over time (unlike inertial and barometric measurements). There are a couple of filter designs that very neatly integrate this information with that from gyros and accelerometers (both very fast and very accurate in the short-term, but prone to drift in the long-term) to produce fast and accurate and drift-free outputs. In the case of the Peachy, the high-speed control is done with the drip sensor (which gives you immediate information about incremental water level changes, but may well drift over time if the drops are not quite the expected size relative to the container) and straight feed-forward on the mirrors (also prone to drift, from things like the DACs and lubricants warming up). The cameras won't immediately fix drift there, but over the course of a print they'll prevent anything drifting too far.

With regards to camera resolution: take a standard (cheap) 640x480 webcam with a 60-degree field of view horizontally. That should allow you to measure angles down to 0.1 degrees, give or take a bit. If the cameras are mounted 30cm above the printing surface then they'll be able to measure the laser dot position to an accuracy of roughly half a millimetre (if I've done the maths right). As Quertz said, the laser dot won't just be one pixel, but it wouldn't surprise me if you could consistently lock down the laser position to within a few square millimetres. No good for actually positioning the laser in realtime (especially because cheap cameras can't go anywhere near fast enough) but fine for slower control (eg. sensing that the design appears to be a little bit bigger than it should be at this stage, and therefore the output amplitude should be slightly reduced).

Using twin cameras for height estimation appears to (again, if my maths is right) give you 0.2mm height resolution, assuming cameras 30cm above the print surface, separated by 20cm. Height resolution will change over the build height, but getting it accurate to within 1mm should be straightforward. Again, it's not a perfect high-speed feedback system, but after the resin has risen a couple of millimetres you should be able to make a decent estimate of whether or not the "vertical rise per drop of water" is being calculated correctly.

[mike_biddell]

I think the best option for mirror position feedback would be to use a patterned disk attached to the thread and bounce a laser off it to focus on an optical mouse chip. And there you have it 2000 dpi resolution for about £1 sterling (just cannibalise a mouse)

[jontelling]

I think the best option for mirror position feedback would be to use a patterned disk attached to the thread and bounce a laser off it to focus on an optical mouse chip. And there you have it 2000 dpi resolution for about £1 sterling (just cannibalise a mouse)

I think you're on to something, but not how I read your post. Rather than use the sensor directly from an optical mouse maybe we could use the whole unit, after all the mouse just sends movement information back down the USB not the RAW data it collects.

I'll put my thinking hat on as I've never worked with data from a mouse.

EDIT: Just the thing!
http://www.instructables.com/id/Opti...obot/?ALLSTEPS

[mike_biddell]

Interesting article.... 1,900 counts per rev... pretty good resolution

http://www.sciencedirect.com/science...24424709003483

[jontelling]

Interesting article.... 1,900 counts per rev... pretty good resolution

http://www.sciencedirect.com/science...24424709003483

I think I've mentioned this, even using 5000points around 360deg it isn't enough, the peachy is 18bits, 262K points which is uses over 40deg. That's 2,358K per 360. I'll find my post...

[mike_biddell]

errr not sure the number of A to D bits is relevant for this approach. We are reading digital values (x and y) serially. There is no analogue to digital conversion involved as the signal from a mouse is already digital. We can store values in long words if necessary. I think there are some very high resolution mice, so I think if I was monitoring mirror position, I would try this optical approach.

[Slatye]

Mike - the problem is that you can't measure the position of the mirrors as accurately as the Peachy can drive them. As jontelling has said, the Peachy can apparently do 262144 counts over a range of about 40 degrees, giving it an angular resolution of 0.00015 degrees. The encoder you linked manages just under 0.2 degrees angular resolution. If you're trying to use all of the Peachy's precision, you might drive the mirror to 8.20305 degrees - but the encoder will only be able to tell you that it's somewhere between 8.2 and 8.4 degrees. You can't verify that it's at exactly the right position.

With that said, the encoder could be used as a relatively coarse check that the mirrors are in the right place. With 0.2° angular resolution, you can guarantee that (with the distances I assumed for the cameras) the laser is within about 1mm of the target.

[mike_biddell]

Yes.... understand..... 5000 dpi is about the best resolution from the current breed. But optical is the way to go..... I think it would require a custom chip..... expensive !!!!!

Thinking this through, with the mirror mounted at right angles to, but rigidly fixed to a 2 inch diameter disk. Monitoring the disk at its periphery with an optical mouse type setup, we could get 30,000 steps. Pi*D = roughly 6 inches and at 5000 dots per inch, that is 30,000 dots. Therefore steps per degree = 30000/360 = 83 steps per degree. An angular resolution of .012.