DIY Laser levelling using webcam and laser level

[Neale]

Forgive me jumping in (and apologies if I've missed something) but is the geometry of the measuring setup ideal? As mentioned, the measurements are dependent both on the straightness of the rail and of the beam itself and it's not really possible to separate out the two. If you turned the sensor through 90deg so that it looked along a line parallel to the gantry, and placed the line source just off the end of the gantry but with the laser line intersecting the sensor, you would then be moving the target down the length of the beam (straight apart from gravitational bending!) rather than relying on the optical quality of the cylindrical lens generating an accurately straight line. However, you would need to get the laser line horizontal to reasonably close limits both on the roll and pitch axes (if that makes sense).

When I first started following this thread I had a picture in my mind of a point laser shooting a beam parallel to the rail being measured and striking the sensor in the middle. What I have described above is really only the same thing but using a line source to make lateral alignment less critical.

Or I might have missed the point but I do find this whole exercise very interesting to follow.

[devmonkey]

Forgive me jumping in (and apologies if I've missed something) but is the geometry of the measuring setup ideal? As mentioned, the measurements are dependent both on the straightness of the rail and of the beam itself and it's not really possible to separate out the two. If you turned the sensor through 90deg so that it looked along a line parallel to the gantry, and placed the line source just off the end of the gantry but with the laser line intersecting the sensor, you would then be moving the target down the length of the beam (straight apart from gravitational bending!) rather than relying on the optical quality of the cylindrical lens generating an accurately straight line. However, you would need to get the laser line horizontal to reasonably close limits both on the roll and pitch axes (if that makes sense).

When I first started following this thread I had a picture in my mind of a point laser shooting a beam parallel to the rail being measured and striking the sensor in the middle. What I have described above is really only the same thing but using a line source to make lateral alignment less critical.

Or I might have missed the point but I do find this whole exercise very interesting to follow.

Hi Neale,

That works fine for checking something is straight, what I'm trying to do here is build a system that can check things are planar, e.g. for bringing the two X rails into plane, hence the line laser.

One idea to separate the gantry error in this test from the laser error is to run the test twice with different sections of the laser line.

I've just tried a much better detection algo, what it does is find the simple maxima of a moving average window of image rows, it then fits a gaussian to this window only, ideally it would just fit a parabola but I haven't coded that. Readings are now much more stable across the entire dynamic range and it works better with wobbly beam cross sections. Just need to migrate to the cold garage to test it.

I think a reasonable measure for success would be for the system to meet din 876 grade 1, which for a surface plate 600mm long means 16um variation in height. The cost of such a plate for my machine is £5500 !
https://shop.mitutoyo.eu/web/mitutoy...08/index.xhtml

[John McNamara]

Hi Joe (Devmonkey)

Firstly... Happy New Year! I hope 2020 is a good year for all.

That new algo sounds interesting.

I wonder if it will also improve the resolution for stretched wire as well?

While you are at it it would help if there was some sort of adjustment to compensate for the image brightness and or contrast. I found this was quite sensitive when backlighting the stretched wire. some sort of adjustment would be helpful.

Anyway looking forward to testing the new algo and comparing the results to my old tests as disclosed here in this thread using the previous algo when you have the time to upload the new one.

Regards
John

[devmonkey]

Hi John,

The algo changes are committed so you can re-download and test. In settings the there is a new 'model' field, this takes a number 0->4 inclusive.

0: Old global gaussian fit,
1: Simple global energy maxima
2. New global parabolic fit
3: Local energy maxima then gaussian fit around this maxima
4: Local energy maxima then parabolic fit around this maxima

For laser the best is 4, please let me know for the wire, you may find that since the wire shadow is much less noisy and narrower than the laser that the original model works best still. Also note I have increased the default smoothing (settings->smoothingFactor from 3 to 20), you may want to reduce this for the wire shadow as it is narrower.

[devmonkey]

So I ran another set of tests directly on my granite island without the gantry, all tests were with the cheap Bosch laser. I ran 3 tests, each measuring 10 stations over 500mm with the laser position moved between each test. Each test consisted of 4 runs, 2 forwards, 2 reverse.

The laser was positioned as follows for the 3 tests

1. Planar test with laser orthogonal to ruler taped to island from a distance of 1.5m so the laser line spanned the ruler.
2. Planar test with laser in same position as (1) but rotated 45 degrees to test a different section of the laser line.
3. Straightness test with laser inline with the rule, i.e. shining along the rulers length.

The idea was that 1 and 2 would test different parts of the laser line arc (say 45 degrees of arc per test), whilst test 3 would use only a tiny part of laser arc, less than 1 degree. The test surface is a granite island that most likely isn't flat but is stable, flatness doesn't matter here as we are testing the laser not the island.

Consistency between the runs of each test was very good, with an peak-peak repeatability error of a few microns, likely due to me eyeballing the station locations against the ruler markings.

The chart below shows the results, overlayed on each other. You can see at a glance the 4 runs for each test are highly internally consistent but that the results vary considerably between tests 1-3. All are parabolic suggesting the granite island has a parabolic surface error of around 100 microns. The jagged line group is test 1, the slightly lower group is test 2 and the lowest group is the straightness test 3.



So this is telling me that the laser line generator is not very straight and has a parabolic nature to it with more noise in the test 1 arc than the test 2 arc.

Since test 3 used a tiny part of the laser arc and light travels in a straight line I think we can assume the test 3 line accurately reflects the true surface straightness of the granite island.

Separated out this is the chart for test 3.



So (although I cannot be 100% sure without a calibrated reference surface) I think this system is suitable for verifying something is straight. This means it can substitute for the usual straightness test of using a precision level to straighten a rail in the horizontal plane or a tight wire to straighten a rail in the vertical plane. However it needs a better laser or a different approach for checking two things are planar.

I have done a bit of research on how this is done in the optical metrology industry and although they use exactly the same approach of measuring height deviations from the test piece to a planar sheet of laser light they also without exception use a rotating laser rather than a line generating lens. Infact they use something called a penta-prism which bends a laser beam 90 degrees irrespective of the the angle of incidence of the beam to the prism. This means so long as the laser is fixed solidly and somewhat vertical as the prism is rotated above the beam in a solid bearing you will cast a perfect plane irrespective of bearing runout or laser prism miss-alignment.

Unfortunately penta-prisms with a few arc-seconds of accuracy are rather expensive. If you use a cheap less accurate one then you could well be casting a cone rather than a plane as you rotate the prism.

[devmonkey]

One final test, I repeated the straightness test on my milling machine table, again 10 stations over 500mm, 3 runs again eyeballed off a ruler bolted to the table. The table is pretty straight within +/- 0.01mm over 500mm.



It is certainly much straighter than my kitchen island !

I did this test very quickly and it wasn't performed perfectly, for example the laser was really too close sensor meaning that the sensor was saturated or nearly saturated around the beam, this would lead to a little inaccuracy as the maxima is effectively on a flat peak, it would be better if the laser was further away reducing its brightness.

[devmonkey]

Forgot to add you can now press 'z' to zero and 'm' to take measurements during a run, this is much easier at least for me rather than fiddling with the mouse.

[devmonkey]

So I have 'borrowed' a much higher quality line laser with decent optics, and believe I have verified that line lasers are not straight. Setup was with the sensor on my milling table moving along 10 stations over 500mm against a ruler that was clamped to the table. Note the sensor was moved, the table was stationary to avoid any rocking on the saddle block.

The laser was setup in a fixture about 2m away so that it projected its line across the top of the table hitting the sensor. The fixture allowed the laser to be rotated so I did 4 runs then flipped the laser 180 degrees axially (so the projected line was now inverted wrt the first runs) and did 4 more runs.

I think this suggests that this laser has a parabolic error in its cylindrical lens or the lens alignment, what do you think? Chart shows all 8 runs, you can clearly see two groups of 4 parabolas, one group the inverse of the other, this parabolic error is greater than any error in table straightness.

[Kitwn]

Joe,
I doubt I will ever be measuring to a few microns in my own workshop but I love the way you are making such precision achievable in a DIY environment. The revelation about the non-straight laser is important but presumably the ability to invert it means the error can be quantified and allowed for. Does this mean a cheap line laser mounted in a suitable rotating/inverting housing could be virtually as accurate as the high quality one?

Kit

[Kitwn]

Joe,
Whilst replying on your build thread an idea came to me. I'd be interested in comments on it's practicality:

Could your laser leveling camera and software, possibly using a taught wire as John is doing, be used to measure the height errors as a gantry machine moves along it's imperfect rails and edit the Z axis G-code to correct for them in much the same way as a probe is used to measure and correct flatness errors in PCB cutting?

In this way a machine of modest accuracy could be used to build parts for a better one. I'm thinking specifically of jobs like grinding the top of gantry profile as you have done for your build.

Initially I'm thinking of only flattening a linear cut as above but since your stated aim is making planar measurements is it reasonable to think of correcting over an area in order to turn plain old ally plate into something approaching tooling plate or even turn a piece of granite worktop into a reasonably accurate surface plate?

Kit