Precision square tester

[rhb]

Does anyone have experience with short distance photodiode distance sensors such as this:

https://www.digikey.com/product-detail/en/on-semiconductor/QRE1113GR/QRE1113GRCT-ND/965713

I'm wondering if a pair of these could be used to construct a high precision (0.0001"  over 10" or better) reference square for checking other squares using a 10-12 bit ADC as typically found in MCUs.  With a full range of 0.5 mm, 0.0001" is about one part in 200, so even an 8 bit ADC should suffice.

I have in mind a square with a pair of spring loaded probes.  So long as the calibration square is within  range and the sides of the blade are parallel, reversing the square should allow adjustment to true.  A  temporary square made from a set of precision parallels can be used for initial setting.

What's the short and long term stability of distance measurements with these?

[cellularmitosis]

How would it be used in-circuit?  Would you be sending out pulses of light and timing their response, or would you be trying to sense changes in intensity?

[rhb]

Take a look at the current vs distance plot on page 5 of the datasheet.  It's nearly linear from 0 to 0.5 mm.  Measure the voltage drop across a stable resistor with an MCU ADC.

Press gauge against reference "square", press button.  Reverse reference "square" and repeat. Calculate midpoint voltage and it's ready to use.  The reference square need only be within 0.25 mm of true over the separation of the sensors.

If you've not dealt with dimensional metrology you won't appreciate what sub 0.0001" means. The temperature coefficient of steel is 6e-6/C.  So a fraction of a tenth  makes voltage or resistance to a few ppm look like duck soup.   There are *no*  zero tempco metals.  Fused silica and Zerodur are the lowest tempcos I know of.

Check out this baby:

https://www.edmundoptics.com/optics/windows-diffusers/interferometry-windows/8-dia.-lambda20-fused-silica-dual-surface-flat/

It's only 8".  I'm sure they would be very pleased to discuss a larger one should you need one.

[Conrad Hoffman]

Super Invar is good enough at room temperature that you probably won't be able to measure it, but I digress. Any of the reflective sensors tend to be useless if you change targets. Too much difference in reflectivity and scattering. There's a super sensitive one that uses a random fiber bunch, half to send, half to receive. Probably Keyence. Great analog device, but only with one target.

I've been fascinated by squareness for a long time. The traditional height gage with a ball in the base still baffles me. Unless the indicator on top, and the ball on the bottom, have the exact same radius, it would seem sensitive to angle, thus worthless for really precision work. I'm always reduced to a DIY cylindrical square and a piece of white paper. I don't have room to set up a collimator and use my optical square.

[rhb]

Varying reflectance would ruin things, that's why I suggested spring loaded probes with the sensor looking at a constant target.  Changes in the reflectance can be calibrated out.

Accurate angles are *really* hard.   A right angle is the easy case. Integral fractions of 360 aren't too bad, but non-integral fractions are *very* difficult.

I'd been trying to find a linear photodiode array when I ran into these senors.  One look at the current vs distance graph and I was excited.

There are a lot of *simple* problems in dimensional metrology which are incredibly difficult.  Consider an 18" test bar with 1/4" wide collars every inch.  Even a very worn lathe can make one. But try to make one to a tenth on even a new lathe.  It will take a lot of coolant and time.

[Conrad Hoffman]

359 tooth gears for sidereal drives are really annoying. 360 not so much. On the test bar, it's easy to take a piece of center-less ground stock and bring it round and true with a cast iron lap. The trick is putting the center holes in.  I'm thinking they could be bored if the end was supported in a spring loaded V-block or steady-rest of some sort. Going about it the other way, cutting the collars, is a PITA. I've made several 2-collar bars and those are quite easy to get right. It's usually good enough to just reverse the bar on the centers and use a shear tool to cut identical size collars. I suppose one could cut a multiple collar bar a tenth or three over size, then equalize everything with the lap. Long ago I had access to a small cylindrical grinder. I miss that.

A while back I made some small squares, actually rectangles. I put a hole in the center and mounted them to my Harig Grind All #1. Rotated them 90 degrees for the finish grind and done. That's supposed to be good to 10 seconds. No good for NIST, but plenty good for me.

One of the mysteries of life is why sine bars have no provision to attach anything to them.

[MadTux]

I recently bought an old autocollimator to make a perfect squareness reference for machine alignment/scraping (and check flatness of my yet to be acquired surface plate, so nobody sells me some old thombstone  ).

My idea is to get a more or less square cast iron block, scrape 2 faces opposite to each other to flat and parallelness (should be easy to check with good dial indicator, best electronic one I have has 300nm resolution ;-) )

Now, attach 2 flat (cheap mirrors aren't exactly optically flat, which gives bad/ghost images in autocollimator image) and parallel mirrors (mirror surface is parallel to bottom surface) to these surfaces and use the autocollimator to measure the angles of these mirrors. If both mirrors are at the same angle, the bottom surface should be square to both mirror surfaces. For perfect cube, do the same to all surfaces.

[rhb]

Double faced flats are expensive whether coated or not.

A good variant is to use a  10 second or better level to scrape two faces parallel, then rotate 90 and scrape the other faces level using a leveled surface plate.  I'd surely want an old piece of cast iron if I were going to do all that work.

I've never seen it mentioned, but I use a printer's brayer (rubber roller) to spread spotting paste so as to get a thin uniform coat.  In the past I've used steel scrapers, but the next time I do any I'll make up a carbide tool.

But making squares is not the same as testing them.  One can use a pair of tenth or better indicators, but setting them is a chore.  The goal here is to have something that simplifies fine comparisons.  At present I'm limited to a reference and white paper.

I'm going to get 10 of the NXP sensors and test them on a micrometer stage using a 34401A to measure the current and see how repeatable they are.  I was hoping to hear from someone else who had tried this,  But for under $8 it's pretty low cost entertainment.

[Conrad Hoffman]

I used to use similar photo sensors in a feedback system for piezoelectric actuators. The problem with phototransistors is/was noise. In a really low bandwidth situation you can filter it to death, but there may still be low frequency jumps and wander. Or not. It might be possible to select them if they're cheap enough. I've always wondered why they don't sell them with photo diodes, which seem to be more linear and lower noise than photo transistors. You might find these interesting, especially the fiber optic types that can do very high res proximity sensing. https://www.keyence.com/products/sensor/fiber-optic/fu/index.jsp

[Henrik_V]

Accurate angles are *really* hard.   A right angle is the easy case. Integral fractions of 360 aren't too bad, but non-integral fractions are *very* difficult.

What tolerances do you get with a good sine bar and gauge blocks?
 

[rhb]

Accurate angles are *really* hard.   A right angle is the easy case. Integral fractions of 360 aren't too bad, but non-integral fractions are *very* difficult.

What tolerances do you get with a good sine bar and gauge blocks?
 

Depends upon the sine bar, gauge blocks and skill of the user.  For a standard set of blocks in 0.0001" increments and a 5" sine bar the resolution is about 4 arc seconds. So with a cheap set of blocks and sine bar, 15 seconds is probably the best one can hope for.  And that's being optimistic.  A more realistic expectation would be double that and one minute if you did not have good temperature control and were in a hurry.

Just wringing the blocks is a bit of an art form to get consistent results.  And then you have to wait for them to cool off from handling.

A traditional and very effective method for odd angles is the use of toolmakers buttons.  These have a hole in the center and look like thick washers.  A piece if plate has a hole in the center.  A button of the appropriate diameter is placed in the center and the number of divisions of a circle desired are laid out with buttons of the appropriate diameter to just touch the center button and the adjacent buttons,  Then everything is clamped down with the button screws.

This sort of work is why geometry, algebra and trigonometry are so important in secondary education.  Unfortunately, at least in the US there is a huge disconnect between education and reality.  If you have no concept of why you are learning something, you have very little motivation for learning it.

I ordered some ON QRE1113 sensors from Digikey, so they should be here in a couple of days.

BTW This will only get you a trapezoid.

A good variant is to use a  10 second or better level to scrape two faces parallel, then rotate 90 and scrape the other faces level using a leveled surface plate.  I'd surely want an old piece of cast iron if I were going to do all that work.

I was thinking of squares made with four faces and pivots at the corners which you adjust so that the faces are parallel and when you turn it 90 degrees the faces are level.

[rhb]

My QRE1113s came so I set up a simple test using the spindle of a micrometer as a mirror target.  I did not attempt to exclude external light and the setting of the mike was not very precise. The mike will read to 0.0001",  but you have to be able to see the vernier scale.  Readings were taken with a 34401A.  The QRE1113 is epoxied to the head of an 8d finish nail filed down to the size of the sensor.

I made one run at steps of 0.005" and then a second run with other sizes in addition to the 0.005" readings.  Because the calibration is done by reversals, the non-linearity is not a problem.  Obviously gain will be needed. Voltage was 5.1 V with a 750 ohm resistor in series with the LED.

Repeatability is good.  Using a magnifier I set 0.010", moved away to around 0.015" and reset 0.010".  On all 3 trials I got 86.3 uA at 0.010".

Edit: I made a pass at 0.001" increments and Vcc = 10 V

[cellularmitosis]

Nice work!   

[tomato]

Accurate angles are *really* hard.   A right angle is the easy case. Integral fractions of 360 aren't too bad, but non-integral fractions are *very* difficult.

What tolerances do you get with a good sine bar and gauge blocks?
 

Depends upon the sine bar, gauge blocks and skill of the user.  For a standard set of blocks in 0.0001" increments and a 5" sine bar the resolution is about 4 arc seconds. So with a cheap set of blocks and sine bar, 15 seconds is probably the best one can hope for.  And that's being optimistic.  A more realistic expectation would be double that and one minute if you did not have good temperature control and were in a hurry.

Just wringing the blocks is a bit of an art form to get consistent results.  And then you have to wait for them to cool off from handling.

No offense, but it seems unnecessary to degrade the resolution from 4 arc seconds to 1 minute. Gage block accuracy, temperature, and errors due to wringing aren't serious issues at that level. And impatience ... that is certainly not a legitimate limiting effect.

[rhb]

Accurate angles are *really* hard.   A right angle is the easy case. Integral fractions of 360 aren't too bad, but non-integral fractions are *very* difficult.

What tolerances do you get with a good sine bar and gauge blocks?
 

Depends upon the sine bar, gauge blocks and skill of the user.  For a standard set of blocks in 0.0001" increments and a 5" sine bar the resolution is about 4 arc seconds. So with a cheap set of blocks and sine bar, 15 seconds is probably the best one can hope for.  And that's being optimistic.  A more realistic expectation would be double that and one minute if you did not have good temperature control and were in a hurry.

Just wringing the blocks is a bit of an art form to get consistent results.  And then you have to wait for them to cool off from handling.

No offense, but it seems unnecessary to degrade the resolution from 4 arc seconds to 1 minute. Gage block accuracy, temperature, and errors due to wringing aren't serious issues at that level. And impatience ... that is certainly not a legitimate limiting effect.

A cheap set of gauge blocks is only good to 50 millionths. A cheap sine bar to 100 millionths.  Just handling the stuff is a serious temperature problem.   Obviously you have not sat waiting for the temperatures to normalize if you think patience doesn't matter.

I suggest you go read a bit about work practices in jig boring rooms.  I have a very limited amount of experience in dimensional metrology.  But enough to understand just how difficult it gets below 0.001". When I think about what it takes to make 30 holes in a die set to punch 0.001" thick gaskets it just blows my mind. The tolerances are of the order of 50 millionths of an inch or less.

As an exercise, calculate the effect of a 1 C temperature change on a 0.0001"  3-4" iron frame micrometer.

Lacoste & Romberg developed the airborne gravimeter.  Everyone else said it could not be done.  I saw a scrap damper at a geophysical convention  and was chatting with the guy at the booth.  One of their interview questions for machinists was, "How long do you think it would take to make this part?"  Typical answer was a week or so.  Actual company experience over many years was it took a month.

In the late 40's John Strong built a ruling engine at Johns Hopkins.  The machine was submersed in a temperature controlled oil bath in a room inside a room.  Staff could only remain in the outer room to inspect the operation for very brief periods of time because their body heat would distort the ruling engine and ruin the diffraction gratings.  There is a great set of articles about this in Scientific American circa 1948.

[tomato]

A cheap set of gauge blocks is only good to 50 millionths. A cheap sine bar to 100 millionths.

Those are pretty crappy tolerances, yet they do not degrade the measurement in your example to 1 arc-minute.

Just handling the stuff is a serious temperature problem.

Again, even careless handling does not degrade the measurement in your example to 1 arc-minute.

Obviously you have not sat waiting for the temperatures to normalize if you think patience doesn't matter.

Patience definitely does matter.  The failure to exercise patience, however, is not a legitimate excuse for degrading the measurement limits.

Lacoste & Romberg developed the airborne gravimeter.  Everyone else said it could not be done.  I saw a scrap damper at a geophysical convention  and was chatting with the guy at the booth.  One of their interview questions for machinists was, "How long do you think it would take to make this part?"  Typical answer was a week or so.  Actual company experience over many years was it took a month.

In the late 40's John Strong built a ruling engine at Johns Hopkins.  The machine was submersed in a temperature controlled oil bath in a room inside a room.  Staff could only remain in the outer room to inspect the operation for very brief periods of time because their body heat would distort the ruling engine and ruin the diffraction gratings.  There is a great set of articles about this in Scientific American circa 1948.

I'm not quite sure what the relevance of this is. 

[Conrad Hoffman]

Temperature can be a killer, or not matter much at all. I work with small parts, usually under 0.5" and no heroics are needed in terms of temperature control. If you get up to several inches it can matter a lot. Any analog sensors have to be characterized for changes in gain and offset with temperature. Long ago I built interferometers that used re-entrant construction. With the right amount of steel going one way, and a bit of aluminum going the other, it was possible to get a very stable gap over moderate room temperature changes. Obviously not good for rapid changes.

[rhb]

What tolerances do you get with a good sine bar and gauge blocks?

A person who asks this question is not familiar with either the tools or the work practices required to get the best performance.  I see no benefit to suggesting that the tolerances achievable by a master toolmaker using the best tooling made are relevant to such a question.

My response was based on what I thought I could reasonably expect to achieve with the $500 or so of Chinese tooling I have for doing such a setup.   I might well do better, but I'd have to have another way of making the same measurement to have any confidence.

How accurately spaced are the rolls on the sine bar?  I don't know.

How accurately round are the rolls? I don't know.

How straight and flat is the sine bar?  I don't know

How parallel are the rolls? I don't know

How flat is that part of my surface plate? I don't know.

How accurate are these particular gauge blocks? I don't know.

How uniform are the temperatures?  I don't know.

Most of the tolerances for the factors i mentioned are 0.0001".  Moreover, a sine bar is limited to setting small angles.  So there is an additional error term as the angle gets larger .  IIRC my angle blocks are supposed to be good to 5 minutes.  Things get very difficult if 1 C causes significant expansion. They are extremely difficult if 0.01 C causes significant expansion as is the case for a ruling engine.

As for the topic of this thread, it occurs to me that the hump in the 0.001" increment measurements is likely error in the screw pitch of the micrometer.  So I shall be investigating that with a more robust setup by changing my initial reference position.  I also have a few hundred nanoAmperes of noise to investigate and hopefully eliminate.

The optical windows appear to be clear epoxy fill and are anything but flat.  So another experiment will be to lap the face of a sensor  flat and test it.  Figuring out a way to do that will take quite a bit of thought.  They are quite difficult to handle because of the small size.

[tomato]

My response was based on what I thought I could reasonably expect to achieve with the $500 or so of Chinese tooling I have for doing such a setup.   I might well do better, but I'd have to have another way of making the same measurement to have any confidence.

How accurately spaced are the rolls on the sine bar?  I don't know.

How accurately round are the rolls? I don't know.

How straight and flat is the sine bar?  I don't know

How parallel are the rolls? I don't know

How flat is that part of my surface plate? I don't know.

How accurate are these particular gauge blocks? I don't know.

Most of the tolerances for the factors i mentioned are 0.0001". 

So, do you know the tolerances or not? 

[rhb]

Most of the tolerances for the factors i mentioned are 0.0001". 

So, do you know the tolerances or not?

Mine have never been calibrated, so I have only claimed accuracies and tolerances made by the manufacturer or seller.  Usually only the latter.

Please post current calibration certificates for your gauge blocks, sine bar and surface plate and what it cost. for each calibration.  If you don't have those, then you don't know either.

As you disagree with my estimate of accuracy, you can then provide a tutorial for the OP on how to calculate the tolerance band for a sine bar setup using your tooling.

In any case, you can't set up 90 degrees with a sine bar.  So none of this has anything to do with the original thread on testing squares.

[tomato]

As you disagree with my estimate of accuracy, you can then provide a tutorial for the OP on how to calculate the tolerance band for a sine bar setup using your tooling.

Here's why I disagreed with your first estimate:  You initially calculated a figure of 4 arc-seconds using a specified accuracy of .0001" for your gage blocks and sine bar.  I have no qualms with that.  But you then degraded the figure by a factor of 15 (to 1 arc-minute), without just cause. That is bad metrology. 

Now you are saying that the specified accuracy of your gage blocks and sine bar are not .0001" and are, for all intents and purposes, complete unknowns. That's fine, but that is a completely different situation.  If that accurately reflects reality, then go with that.

[rhb]

I'm finding that alignment is tricky.  I'm definitely going to have to devote considerable thought to how to mount these.  I'm getting repetition errors  of up to 1- 2% going back and forth between 0.005" and 0.010" with my test setup.   The linearity between 0.001" and 0.011" is remarkable.  I suspect most of the error in this plot is experimental.

The line fit was from 0.000" to 0.012".  So it's a slightly better fit over the 0.001" to 0.011" range if I omit the endpoints.

[Henrik_V]

What tolerances do you get with a good sine bar and gauge blocks?

A person who asks this question is not familiar with either the tools or the work practices required to get the best performance.  I see no benefit to suggesting that the tolerances achievable by a master toolmaker using the best tooling made are relevant to such a question.
[...]

@rhb: This is metrology here    I asked this question out of curiosity. If I'm in need to measure small angles, I asked my fellows for an autocollimator   
(and now I going to ask them the same question)

 

[rhb]

For ordinary workshop tooling the typical specs are:

5" Sine bar:

roll diameters  0.0001" x 2  =  8 arc seconds
roundness  0.0001" x 2 = 8 arc seconds
parallelism 0.0001" = 4 arc seconds

Gauge blocks

2 B grade blocks 2*0.00005 = 0.0001" = 4 arc seconds

B grade surface plate 0.0002" = 8 arc seconds

The total assuming all tooling is in calibration is 32 arc seconds  tolerance range.  That's assuming no errors in the setup and thermal equilibrium.  If the gauge blocks are 15 F warmer from wringing then add another 4 arc seconds that's 36 arc seconds.  If there are more than two blocks add 2 arc seconds per block.  I've not included warping of the sine bar from temperature differences, perpendicularity of the rolls to the bar, flatness of the gauge blocks and a lot of other factors.  I'd say that 1 minute is pretty optimistic using Chinese tooling and sensible using Mitutoyu, Starrett or Brown & Sharpe tooling which has been recently calibrated.

I'd be very interested to hear what the professionals  you work with say to the same question.  I didn't actually go through all the calculations listed above.  I made a rough count of how many dimensions were involved.   I took it as a question from a hobbyist with some exotic project and limited budget.  There are some pretty amazing people around here.  And I'm crazy enough to have spent a lot of time trying to find ways to measure a fraction of an arc second cheaply that wasn't near horizontal.  The best option I know of is a large disk with a steel tape and a micrometer to move the tape.

I'm trying to find a source for some 2" x 12" x 1" glass slabs so I can try my hand at making some optical flats for checking straight edges.

[tomato]

For ordinary workshop tooling the typical specs are:

5" Sine bar:

roll diameters  0.0001" x 2  =  8 arc seconds
roundness  0.0001" x 2 = 8 arc seconds
parallelism 0.0001" = 4 arc seconds

Gauge blocks

2 B grade blocks 2*0.00005 = 0.0001" = 4 arc seconds

B grade surface plate 0.0002" = 8 arc seconds

The total assuming all tooling is in calibration is 32 arc seconds  tolerance range. 

No.  One does not simply add individual uncertainties in a linear fashion to get the overall uncertainty.