nanoDLP

Not yet. I can try that later this afternoon (it's 5:30am).

My sensor read values from 0.6 to 3.81 across the surface of a DLP grid. I found that the sensor readings were non-linear and I had to experiment and use a regression generator to get a formula to get the right greyscale values. Here is the formula:
greyscale = 20* (-0.1282*x^5+1.6196*x^4-7.843*x^3+17.8275*x^2-16.3317*x+4.75)

Even that formula was not perfect.

The problem is neither you nor the user will know the linearity formula for their sensor.

Using my method, you do not have to know the formula because the user will enter values (1-256) in a cell to get the right greyscale value so their sensor shows the right value.

If you use your method, you will need a way for the user to enter a linearity formula. Then they will have to experiment over and over entering different formulas until they get the right formula.

I suggested my method to avoid this challenge of having to determine a linearity formula.

So I gave it a shot. Here's my feedback...
* I clicked on the [Turn Projector On], then [Display] projector calibration grid, then [Mask Generator] button
* It stayed on the projector calibration grid
* then clicked in a visual grid cell, then the grid came up with that cell highlighted, then a couple of seconds later the mask came up
* then clicked in a visual grid cell, then the grid came up with that cell highlighted, but the grid stayed up this time
* then typed 256 into the grid cell (I used the middle cell) and hit [enter]
* I expected the square on the DLP to go black, but it did not seem to change intensity at all
* Then I clicked [Preview], nothing happened, I expected the mask to display on the browser or the DLP.
* I waited for about a minute and then the mask came up with a square cell darkened in the middle of the projection (I was expecting a dark point circularly fading about 1/2" in diameter)
* The DLP flickered several times while I was typing this out

Suggestions:
* Darken/lighten the square of light real-time each time as the user types a value in a corresponding cell (the user will have their light sensor on that spot to see if it is the right value before adjusting the next cell value)
* Highlight the selected cell by some other indicator than brightness (for example, put a red dot by it, or a red rectangle around it). In this way, each square on the DLP reflects the brightness value the user indicates in each cell. With this capability, the user can verify adjacent cells, or compare adjacent cells.
* The up/down arrows are really nice, but they are so small on my screen, I can't click on them (I have a 4K screen)
* When generating the mask, assume the center pixel on the square is the value in the cell, then interpolate from that point to the values defined by the center of each adjacent cell). The excel sheet I provided has an interpolation algorithm that you may be able to use.
* Have some sort of progress bar to indicate the mask is being generated
* Kill the mask generation if they move off the mask generation form (so it doesn't pop up suddenly when they are doing something else)
* Maybe allow mask generation to be cancelled?

I hope these suggestions are helpful!
I'm really excited to see this feature being implemented!

I understand now. Thank you.
Sounds like you have a good plan.
I'm really anxious for this feature and I think it will put nanodlp at the cutting edge of DLP mask creation!

Note on #5: The user could press a [next] button or enter a grid position number
Note on #6: displays grid position #1 (if [next] is pressed) or the grid position number entered
Note on #7: may need to be 1 ~ 1024 (I'm not sure)

I just saw the plan. I've been out-of-town. The challenge I have had is determining a mask from the measured points. I have found that the formula to go from measured points to a working mask is non-linear. I ended up using a 5th order regression to try to get a mask that makes all the points the same measured value.

To avoid all of that, I would suggest the following:
1. User decides how many points she/he wants to measure
2. Nanodlp displays grid based as an image (with no mask applied) on the projector, and displays a edit field for entering the dimmest values on the UI
3. User measures points to find the dimmest value
4. User enters dimmest point value into interface
5. User selects next input (or chooses a particular grid position - there may only be one wrong position that needs to be corrected)
6. Nanodlp displays only grid position #1 & user places sensor on grid position #1
7. User enters a value between 1 to 256 (1=full brightness, 256=black) to dim the grid position until the sensor reading matches the dimmest value
8. User continues steps 5 ~ 7 for all points
9. User could press display mask anytime and mask will be displayed (allow to choose with or without grid)
10. User could check if whole surface brightness is same or continue to tweak values
11. User click on export and mask will be saved

The difference here is that when you calculate the mask, you already know how much to dim at each grid position. You will not need to figure out some special formula to create the mask.

Thanks for such a quick reply! Apparently that's exactly what happened (firefox). I refreshed the browser and voila it shows the right mask.

I suppose one could measure the stumps on the build plate with a caliper and enter those values in.

The grid sample is the uvcal.stl document that can be downloaded here: http://www.buildyourownsla.com/forum/vi … 684#p11719

Three data samples (DS1, DS2, DS3) are this OpenOffice document: bicubic-interpolation-larry-v1.ods

That file also gives an example of bi-cubic interpolation. It was really slow, and I honestly don't remember if I left it in a working state.

Very Cool! These will be great features!

If when entering the data you can tab sequentially from field to field, I could potentially modify the uno to be a HID keyboard. When you measure a value by clicking an uno button, it would "type" the measured value in the field and then "press" [tab]. So it could automatically sequentially fill out the fields. http://mitchtech.net/arduino-usb-hid-keyboard/

I like the possibility of the following:
* Low resin detection
* Uneven platform detection <= Maybe make this a homing detection too? On muve3d printer this is a common issue and you could potentially automatically home both sides.
* Fallen part detection <= This is pretty common and wastes a ton of time waiting for a print that is already ruined.

Interestingly, projecting red (600nm) may be a better relative approximation of the UV band (400nm) than white light (multiple components across the spectrum). See the spectrum output for the muve3d projector: http://www.muve3d.net/press/projectors/

On banding...
I wonder if the vat bottom is acting as a prism and splitting the white light into its component values as it is reflected from both the bottom and top surfaces of the glass? See http://science-edu.larc.nasa.gov/EDDOCS … olors.html. If this is the case, then projecting red may prevent the banding.

Interesting. Thank you for that description. That's smart.

Great! Thanks! I'd be glad to help too when the wiki gets posted.