Tachometers 2

[Sparweb]

Hi,
People don't often post about solved problems, but here I am. I'm hoping this will be a fun post for the group. I think have solved a problem and struggling to figure out why I had to spend 7 years working on it when it was so obvious at the end. Several years ago, I was trying to solve a problem that had daunted me for years, came up with a solution, and moved on. Since then I've looked back at what I did and wondered "why hasn't anybody ever thought of that before?"

I can't find any patents for the device, and I can't find any articles in journals (IEEE) that I have searched.
Once I settled on the idea, it seemed "obvious" to me which breaks one of the principles of the patent - but then how come nobody does it...?
I'm not really interested in getting a patent - the rigamarole, the cost, and the futility of trying to defend it are all disincentives. But I really do think I should publish about it - if this truly is a unique idea. It would be nice to have a useful article to my credit. It's an idea I would like to share, but I want to find out if it's really a new idea, first. So I'd like to formulate a little test and see if anybody else finds this "obvious" with a few clues and conditions. If I just told you, then any smarty pants can come on and say "yes it's obvious" and publish it for themselves.

I promise to tell the rest of my story but only if there's interest, or if the answer is so obvious that somebody guesses right away. I'm a mechanical engineer with an interest in all engineering disciplines, dabbling in many others, so I could have missed an obvious solution or the one I came up with is indeed obvious and electrical engineers do it all the time (for some reason they don't talk about it?).



First I have a confession to make: I built a wind turbine. (No apologies to those who are offended) I built it out of personal interest but I have been approached by a person who wants to buy one just like it from me. Never considered this a commercial product before, but if I am going to look at it that way, then I need to work on some of the control and safety aspects.

Its design is fairly elaborate but its construction is simple. It has had good performance for many years, but the obvious missing piece of information is how fast it's turning. I needed a tachometer.

Options that I considered: encoder, hall sensors, tach generator, frequency measurements on AC powerline.

Location of the turbine is at the top of the tower, which is stating the obvious. I point it out because that makes adding little finnicky gadgets difficult. To support some types of tachometer, signal wires going up the tower are needed. Some even need their own power supply to work. Sure, I can add little signal wires but it won't be robust. To make it robust would be costly. Slip rings, custom cables, etc. This wind turbine has run well for almost 10 years without a flaw and I would hate to have a thing like a tachometer in it that forces maintenance yearly. Despite the reluctance, I did actually install a hall sensor into it (two for redundancy) and started bench-testing but I got really frustrated with the difficulty of re-designing the mount of the generator itself to accommodate either a slip-ring assembly that would somehow pass small-signal currents without noise of their own.

I resolved to try to build a tachometer that relies only on the AC on the powerlines. However, the turbine is a customized 3-phase generator and the output is rectified to charge batteries. Off-grid stuff. The rotor blades are direct-drive on the generator, with a variable speed. The speed varies from 0 to 150 RPM (charge current cut-in) to about 400 to 500 RPM where the tail's furling mechanism folds everything out of the wind for protection. The AC coming from the turbine is definitely not a sine wave. It's more of a clipped trapezoid with harmonics and noise. The variable speed and DC rectifier switching puts spikes in that vary in intensity at different speeds. There are also flat spots crossing zero that you can see on an oscilloscope. It's a mess. This does not affect the power output of the turbine. Remember, the load is a DC rectifier, so the noise and spikes are inconsequential to power conversion.

I built the obvious thing with a LM2917 IC and tested several variations but they all failed. The 2917 got its input from one of the three AC lines from the generator to the rectifiers. It would pick up the harmonics at certain speeds, the baseline at others. I tried filtering but that was very tricky to get right. I thought I was close but then made a change in the the generator wiring and found myself back at the beginning. I tried an optoisolator and fed the pulses to a microcontroller and got exactly the same thing. The microcontroller could be programmed with some extra code to "guess" the bad data but always discarded huge ranges of records.

When I solved it, I was able to dispense with any filtering at all. I still used a microcontroller, so I was able to build a more elaborate data-logger around it, so I now have current, voltage temperature and a number of other things measured simultaneously. Every second if I want a mountain of data. Sampling and averaging is easy in the microcontroller programs.

So now that I've described my failures trying to solve it, before I had my "great idea", what would you do? Faced with the need for a tachometer, how would you get a good solid signal proportional to the speed of a machine like this?

http://www.sparweb.ca

 

[LionelHutz]

SparWeb - No filtering at all? To clarify, you input the signal directly into a microcontroller using resistors and have no software filtering?

I was confused about your single phase use comments too. Using L-G isn't a good plan unless the generator is Y connected with the center point grounded.

3DDave - Please post links to some commercial examples that use such an arrangement.

 

[MacGyverS2000]

If this is going into a micro, surely some VERY basic time-averaging of the signal can remove momentary spikes that confuse straight hardware zero-crossing detectors. It's not like the windmill can go from 0 to 100mph in an instant... there will be some rise/fall time as it speeds up and slows down, so averaging just a handful of crossings should be adequate to figure out the "instantaneous" speed.

Dan - Owner

http://www.Hi-TecDesigns.com

 

[IRstuff]

Boxcar averagers have traditionally been used to extract noisy sinusoidal data, so it's a possible approach

https://www.thinksrs.com/products/boxcar.html

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[LionelHutz]

The comment made was "with any filtering at all" which to me means no hardware or software filtering.

 

[waross]

You can drive a frequency dependent small motor from either one or three phases and then monitor the speed of the motor.

Bill
--------------------
Ohm's law
Not just a good idea;
It's the LAW!

 

[Sparweb]

Bill - that's another neat idea!
It's probably more of a power-hog than necessary but it has a brute-force beauty.

Something to say about averaging: If the baseline signal is distorted by harmonics that are multiples greater than 1 of the baseline, then all of the errors are > baseline. The average would be heavily skewed to a value greater than the baseline. Averaging only works when the error can be > or < than the baseline. That's what wrong with the chart I posted with the blue and red points. The baseline speed is about 125RPM at 0 Amps and goes up in a straight(ish) line to 400 RPM at 15 Amps. All the noise adds pulses, giving the data a seemingly higher RPM. It helped a little to select the median value in a time interval. I had measured the power curve independently using a lathe to drive the generator before everything was assembled into a wind turbine, so I do have a proper power curve for the generator to compare with.

Lionel, measuring line to line is better and that's in the schematic I used years ago - but I was so confused at the time I couldn't get good readings that I tried a lot of things including line to ground (and transformers and big capacitors and other things) just for the sake of trying it. There's a big jumble of things I tried using and failed. I accept full responsibility for being a dummy with electronics despite my determination to learn it the hard way.

To better address my comment about filtering, which was taken as a boast and not really intended that way, here's the input circuit that I'm using now. So, yes there is rudimentary filtering by tying each voltage divider to ground with a capacitor. But that's it. The three data lines (red green blue) are limited to 25mA by the resistors because the generator only delivers 30 VAC max. That's all they need to work well with my current scheme.

From here I am using a microcontroller, but I am using averaging only to slow the data stream down. There's very little need to record rotor speed 10 times per second when other parameters being measured by the datalogger need a second or two to collect a steady average.

I haven't exactly given away the key function that the microcontroller does that is making this work, yet. It's not just counting microseconds between pulses, it's doing something else that makes all the difference. I guess I'm dragging out the guessing game a bit too long. I assume you are all still game to try. By now I figure if it is obvious enough that "everybody knows", then this is the piece of info that should give it all away.

http://www.sparweb.ca

 

[Sparweb]

Boxcar averaging:
I hadn't heard that term but it's exactly what I do when stacking dozens of grainy photos to extract a noiseless night-time scene.

NGC891

http://www.sparweb.ca

 

[LionelHutz]

OK, more or less a type of low pass filtering. Your filtering is still there. It's just digital instead of circuitry.

 

[waross]

Spar. What RPM range does your tach cover?

Bill
--------------------
Ohm's law
Not just a good idea;
It's the LAW!

 

[IRstuff]

We normally call that frame averaging or frame stacking, but the principle is pretty much the same. Box car averaging is typically applied to a time varying, but repetitive, signal; raw SNRs of 1:1 would not be unusual.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Sparweb]

Almost zero to almost 1 Mhz. For 2 decimal accuracy let's say 10 kHz. That's because it's a microsecond counter that's running between cycle counts. In a 4-pole machine there are 2 cycles per revolution. That would result in 1 million RPM I think. I say "almost zero" because there has to be events to actually count. After 10 seconds waiting I've programmed the timer to give up and print zero although "12 RPM" would be a valid result.

Accuracy is limited to the accuracy of the counter and the cycle-to-cycle shift in trigger timing. If you were trying to measure something really fast it would matter. I'm measuring in hundreds of RPM, so I really don't care and cannot notice.

http://www.sparweb.ca

 

[Sparweb]

IRStuff,
How do you determine the frame window time (either duration or repetition rate)? Do you guess with one of those analyzers until a signal appears?

http://www.sparweb.ca

 

[MacGyverS2000]

Powering a motor and then measuring that is simply another type of filter... analog and mechanical based, rather than digital or software. Pick your poison.

Dan - Owner

http://www.Hi-TecDesigns.com

 

[IRstuff]

Never really used a box-car averager, personally, but my understanding is that there was something like a phase-locked loop that would keep hunting for the signal and the box-averager window would hunt around until the signal was actually found.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Sparweb]

This thread deserves an update, but primarily, "thank you" to all those who replied.

Your answers helped me formulate a more clear explanation of the tachometer device I was working on. Also inspired me to do more research on the matter. I was very surprised that the tacho method I've been using cannot be found in the engineering literature anywhere I look (electronics textbooks, IEEE papers, Forrest Mims, the CMOS cookbook, you name it).

 

[MacGyverS2000]

I was very surprised that the tacho method I've been using cannot be found in the engineering literature anywhere I look

Quite possibly because your approach is overly complicated for the problem it is trying to solve, i.e., other solutions exist that are more simple... I run into that all of the time in my designs, and I add the knowledge to my bag of tricks.

Dan - Owner

http://www.Hi-TecDesigns.com

 

[MRSSPOCK]

I must confess most of what has already been discussed up to this point is way beyond anything I understand, so I don't even know if understand correctly what we are meant to be thinking might just be common knowledge. But here goes my twopence worth. If this turbine is charging a bank of batteries, I guess somewhere down the line there is a rectifier, just before the batteries. Would a simple measurement at the recitifier bridge provide a clean signal, the frequency of which can be easily read? Am I barking up the wrong straw completely?

 

[MacGyverS2000]

I think rectified or not, Spar's issue was the signal was just too noisy to make direct use of it in measuring frequency, hence the following "50 million ways to skin a cat" discussion where we all came at it from different perspectives (and solutions). Some basic filtering is required, no doubt about it, and the method chosen would be based mostly upon Spar's level of comfort with designing / programming / using any of the above suggestions.

Of course, I still think a simple, slow-filtered op-amp circuit with a DC offset would be the best (easiest?) solution

Dan - Owner

http://www.Hi-TecDesigns.com

 

[waross]

One of the problems that Spar faced was the relationship between signal frequency and signal amplitude.
A simple filter that would work well at higher frequencies and signal levels would possibly filter out the signal along with the noise at lower signal levels.
A filter that worked well at lower frequencies may filter out the signal with the noise at higher frequencies.
With Spar's system, as I understand it, a zero crossing permits a one shot signal from the proceeding phase.
It matters not whether the trigger is from signal or noise zero crossing as the proceeding phase will be at a point in the wave form that is above the noise level.

Bill
--------------------
Ohm's law
Not just a good idea;
It's the LAW!

 

[MRSSPOCK]

I know I'm out of my depth here, but I'm going to stick my head above the parapet anyhow.
Would the voltage drop across a rectifier diode not provide a pretty regular pattern, varying from zero to whatever the diode characteristics produce, or would the noise still show up here also?