On 2023-06-14 10:33, John Larkin wrote:
> On Wed, 14 Jun 2023 08:24:54 -0400, Phil Hobbs
> <pcdhSpamMeSenseless@electrooptical.net> wrote:
>
>> On 2023-06-13 22:18, John Larkin wrote:
>>>
>>> https://www.dropbox.com/scl/fi/zppy7en9noljcfdo0qyv6/Darl_Flow_Sensor_1.jpg?dl=0&rlkey=v7jk899m8l1jxz91r05lj6byx
>>>
>>> (I hope that link works)
>>>
>>> The darlington is a TO-92 sticking up from the PCB into the air
>>> stream.
>>>
>>> The idea is to turn on the mosfet for a while to heat up the
>>> darlington, then turn it off and take a few measurements as it cools
>>> down. A few points should be enough to calculate the cool-off time
>>> constant. The thermal tau should drop by about 2:1 from zero air
>>> flow to 200 LFPM.
>>>
>>> This is even simpler:
>>>
>>> https://www.dropbox.com/scl/fi/1c779s23si0q6ly4ssu72/Darl_Air_Flow_Sensor_DC1.jpg?dl=0&rlkey=75logvs33gcon8xzmtfpmvf7e
>>>
>>> We'd have to turn off the fans and make a baseline ADC measurement
>>> to use as the zero-flow point, which we could do. The more air flow,
>>> the cooler the transistor and the lower the ADC voltage will be.
>>>
>>
>> I like the general trend.
>
> The more one thinks about a problem, the simpler it should get.
> Luckily, the production tests and dummy load boards aren't a rush to
> get done.
>
>>
>> The problem is a bit on the squishy side anyway, ISTM. It's framed as an
>> airflow measurement, when what you actually care about is whether the
>> boards in the crate get adequate cooling.
>>
>> Air flow is one ingredient, but both it and its effects will change in
>> some dimly-known way, depending on the boards that get put in the crate,
>> how the dissipation is distributed, the ambient conditions, junk piled
>> on top, dust inside, and so on and so forth.
>
> Sure, but it helps to know that, if the fans are cranked all the way
> uo, each board will get a guarenteed amount of air, namely 200 LFPM
> over its surface.
>
> I might put an air flow sensor on every board always, to flag the
> various conditions you have named, although each will have temperature
> sensors too. Spme of the plugin boards could dissipate 200 watts.
>
>>
>> So although it would be nice for datasheet purposes, even a
>> super-accurate, unambiguous measurement of airflow at one position
>> wouldn't make the cooling that much easier to calculate accurately.
>>
>> So abandoning hot wires, propellers, and so on, in favor of a
>> measurement of actual board cooling, seems to be at least as valuable.
>> (You can use some simple approximation to back out an airflow number for
>> the datasheet.)
>>
>> So that's all pretty sensible, I think.
>>
>> As the National Semi Temperature Measurement Handbook (2007) says, "For
>> sensors in plastic packages like TO-92, SO-8, and SOT-23, the leads
>> provide the dominant thermal path. Therefore, a board-mounted IC sensor
>> will do a fine job of measuring the temperature of the circuit
>> board (especially the traces to which the leads are soldered)."
>
> I'd spec that the assemblers keep the leads max length and put tiny
> pads and traces on the board. The leads can be half an inch long and
> are probably kovar, a terrible thermal conductor. I need to do the
> math on that, or some measurements.
Nah, Kovar is for glass-metal seals. TO92s are all tin-plated
copper--take a look at the cut-off ends. You can also find out right
quick with a magnet.
>>
>> That's also true for transients, because thermal diffusion in plastic is
>> almost as poor as its thermal conduction. (The thermal diffusivity of
>> copper wire is about 6e-5 m**2 s, vs. 2e-7 or so for plastic, so roughly
>> 300x.)
>
> We can wait a minute or two to let things stabilize when we do the
> per-slot air flow measurement, at production test of the boxes. The
> boards themselves would be running at essentially zero power then.
Sure. My point is that the coupling between the plastic package and the
air flow is the pits, and doing it as a transient doesn't make it a
great deal better. A typical way of localizing thermal measurements is
to use AC at a frequency that can't diffuse as far as the board, but
then it won't make it to the surface of the plastic either.
Of course you've got as long as you like to do the measurement, which
certainly helps.
>> So as long as you don't expect the thermal transient to actually be an
>> exponential, some sort of calibration like that should work OK.
>>
>> OTOH since it's the board you'll be measuring anyway, you could just use
>> an I2C temperature sensor, and heat that with a nearby resistor. Our
>> fave is the Sensirion SHTC3, which has a very nice humidity sensor as
>> well--great for dew point estimation.
>
> I have a separate use for the air flow sensor, in a case where our
> board could be in someone else's crate and the surface flow direction
> is not known. Flow measurement would be a sales feature there.
As long as what you're measuring is actually board cooling, you can put
an approximate fpm number on it based on some lab correlation. Worse
things than that are done every day.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510
http://electrooptical.net
http://hobbs-eo.com