If I use a TO92 transistor as a self-heating thermometer to sense air flow, I'd prefer the leads to be bad thermal conductors. Kovar is a terrible heat conductor, 17 w/mk.
I like exponential smoothing, out = out + k*(in-out)
which models a 1st order lowpass.
That's the reason to measure air flow!
I'd have a separate pcb temperature sensor, an LM35 maybe.
Yes, that helps too.
The problem is that a high-energy impulse takes a long time to die out, so direct impulse handling is usually worthwhile. Low-pass filtering of what survives median or middle half filters also work well.
Well, we are using temperature difference between RTDs (or transistors) as a proxy for air flow, itself a proxy for cooling capability. So we need to worry about places the proxy may fail to predict.
Also a good approach.
Joe Gwinn
An impulse is still real signal. I don't expect an air flow measurement to be precise and it can be lowpass filtered hard; the hazards to the system are long-term and thermal.
It's trivial to do a 2nd order lowpass in c or an FPGA, if you want spikes to wash out sooner.
I can't predict what direction air may flow in a card cage, so I prefer an omnidirectional sensor, namely a cylinder.
This would just be a bullet feature to help sell the controller, so it doesn't need to be super precise. Just an air flow warning. Maybe a fan failed, maybe a cat is napping on the intake.
If you really want something that is not analogous to a hot wire, then, depending on the expected air speed, and what sort of analog input capability you have, an ultrasound transducer and receiver can work. My experience is with ceramics for that purpose. But I suspect piezo film would work too. The trick is to avoid multi-pathing - short pulse and look at first signal only. And of course, and maybe even simpler, you might try using the distortion of a single piece of piezo film directly.
Ultrasound measurement will be directional. And big. And expensive.
Use a circular mode, or, two arms
About 1 or 2 cm at f>=30khz.
Traditional, simple ultrasonic transducers are available under $4 on Mouser.
Or, you might find the following datasheet interesting, though a little more expensive at $24. Circuits are included in the datasheet.
Remember to avoid multipathing or driving resonances, i.e. keep the pulse short. You can model the transducer and driver in spice, and play with the values for the passives until you get a short pulse, that essentially means low Q. With the RLC driver in the data sheet you can get it down to a one or two cylce pulse.
Sorry, it is a little bit of a project rather than just plug it in and use it.
The other option I mentioned was just letting the air pressure deform a small pieze of piezo film. I have not tried that for your application, but I do know from experience that they are sensitive to deformation.
The issue is that not all impulses are equal. Gaussian noise does have peaks, but it does average out to reasonably well behaved values.
But impulse noise (as from a pulse jammer or a sparking contact somewhere) does not follow Gaussian statistics, and can inject far more than the average power level. Mathematically, this is often modeled as the sum of two independent Gaussian noise signals, one of a far larger standard deviation than the other. This kind of signal causes long-tail effects in linear filters of all kinds, and it's pretty much universal to provide a non-linear pre-filter to clamp or blank those impulses before the are permitted to enter the linear filter stage.
Yes, one can do all this in C or a FPGA, but one still needs the non-linear impulse filter before the linear smoothing filter.
Yes. One thing that was learned in the GPS receiver realm was that outdoor antennas for precision receivers need to be impossible for a seagull to perch upon.
And then there is the Starlink Effect on cats:
.
Joe Gwinn
Perhaps another, even simpler solution
How about measuring power (or voltage and current) into the fan? That should surely tell you whether the fan is working and whether it is pushing air.
I'll be measuring Vbe of a diode-connected transistor to measure junction temperature. I don't expect a lot of impulse noise, and not much gaussian noise. Figure a 700 mV level that changes a couple of mV per degree C. Lots of signal compared, say, to a thermocouple.
Probably not.
Given the low impedance of a forward-biased PN junction, it may be herd for external EMI to get in, which would be nice. We'll soon know.
But in the big radar systems of my experience, we always needed a non-linear impulse blanker or interpolator. This was generally proven in the lab when they were trading various sign el-processing approaches off. For low sidelobes (in time), this mattered a lot.
Fault Detection Fault Isolation has a parallel problem, that many detected faults are in fact transient false alarms, so one always has some kind of M-of-N (often 3-of-5) filter to ignore blips.
Joe Gwinn
increased bearing friction will look like moving air even when it's not.
As noted, the flow sensor needs to be on a controller board that plugs into a generic VME crate. That board has no access to the fans.
That seems unlikely - that the load vs speed for bearing friction would really look the same as that for air.
It sort of doesn't make sense also, if they would be even similar magnitude, it means the fan is not very efficient. And even then, as long as the friction is not very much larger, you still have a curve for load vs speed that includes a contribution from pushing the air.
And if the bearings are failing, the curve will likely be changing - upward.
So overall, it seems like it could be a pretty nifty diagnostic. The concern for bearing friction doesn't seem like a good reason not to try it, quite the opposite.
Fine, then the choices would seem to be (a) thermal conduction of moving air, (b) propagation of sound, or (c) direct pressure on a deformable object. All three are cheap. The first and third are probably the most reliable, the third might be the more accurate with changes in temperature, humidity, and air pressure.
Another perhaps more exotic idea that comes to mind is deflection of electrons. But I think the first three might be more practical.
One more almost out of the box idea. Many MCUs and CPUs nowadays have temperature sensors. They are in effect thermal conduction type anemometers. Find a cheap one, and running a simple loop, clock it just fast enough to produce a small increment in temperature over the ambient when the fan is off. You have to admit, in some ways its a pretty relevant measurement for the purpose.
That is interesting, compute something useless to raise chip temperature. But I'd prefer to poke the sensor up off the board, into the air stream.
My plan so far is to measure the temperature of a TO92 transistor at two different power dissipations and so some math on that. A TO92 is about 200 K/W in still air, maybe half that at 200 LFPM.
OK, I need an ARM in a TO92 package.
Or if you want to do as little work as possible, get a cheap arduino, teensy, stm or whatever, put it on two rows of 0.1inch headers, and connect to it over whatever you like to get the data. You could use it to read your TO92 transistor too if you like.
You could put it in a tall header also, if you want it in the air stream - after a fashion
on a whim I put "MAF sensor LCSC" into google and that lead me to the LCSC page on flow sensors on that page I found
Goertek SF15M-001 and Honeywell AWM92205VPP
both less than a buck in small quantities, neither in stock.
Digikey has the frist one and even a leaked data sheet, which aeems to have been tranlated, badly, from some other language. Also the second as a marketplace product with no datasheet or photo...
the first seems to measure air flow via a vertical tunnel, . the second is mystery meat,
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.