DMM fans - noisy friends?

[Amaruk]

Are you fans of DMMs with fans? Ideally I prefer my instruments to be quiet but I fully understand that fans are often required in order to manage a good thermal balance in a high-precision instrument.

I have two bench DMMs, a Siglent SDM3045X which does not have an internal fan, and a Keysight 34461A which indeed does have an internal fan. The latter is a 6.5 digit DMM where as the former is only a 4.5 digit DMM. Today I was curious to find out how much these two DMMs drift over time after they are powered up. I know that you are supposed to turn them on at least one hour or so before you use them for maximum accuracy. But how important (or not) is it to do that? Since I am developing my own custom instrument control software that has data logging capabilities I decided to run a little experiment with two the DMMs and a 5V voltage reference. The results were very interesting…

The experiment: I powered up the DMMs for 5 minutes before starting the drift test. This was to allow the DMMs time to fully connect to the network as all data acquisition is done over LAN. Both DMMs were then connected in parallel to the 5V voltage reference (AD584) and the test started. During the test my software measured the voltages from the DMMs (5 sample averages) every 30 seconds for about two hours. I then computed the voltage drift over time in PPM (parts-per-million) based on the first reported value from each DMM.

The results are shown in the images below. It should be noted that during the test the ambient temperature in the lab was held constant within 1F (0.5C).

While I did expect to see drift in the results, I did not realize that the difference between the two DMMs was going to be this big. Very interesting.

[TiN]

What is has to do with fans?

[Amaruk]

That is a very good point. I just assumed that the fan played a role in this - together with the voltage reference used. But I don't know what the cause of the drift is. As for voltage references, the one I use in my test (AD584) is not very expensive and from this test it looks very stable right from the start. So, what do you guys think is the reason some DMMs drift less than others?

[TiN]

It is a complex question and all parts, like physical, circuit and component design are important contributors into DMM stability, warmup time and other performance parameters. Some circuits working better and more stable then others. Many instruments with fans are more stable than ones without fans, and the opposite. You are comparing two instruments of very different goals and performance specifications. Higher sensitivity instruments, say 8.5d meters would often need weeks and months of time to get stable to their best design capabilities. This is why you often see requirement never to turn off equipment in calibration labs, even when not in use at the moment. Same reason why you have high-end frequency standards always on 

[bdunham7]

Are you fans of DMMs with fans? Ideally I prefer my instruments to be quiet but I fully understand that fans are often required in order to manage a good thermal balance in a high-precision instrument.

Fans may cause a DMM to reach thermal equilibrium sooner and also reduce internal temperature gradients.  This may improve performance or they may be used to allow for less-expensive components to suffice for a given spec.  This doesn't really have anything to do with the differences between your two meters because they are in entirely different performance categories.

I somehow don't have any DMMs with fans.  Of the two that I have that are comparable with your 34461A, one (8846A) has less warmup drift and one (34401A) has more.  The 34401A has a long warmup time, it doesn't totally stabilize until 3 hours or so--and it looks like your 34461A isn't much quicker, unless there is some other cause for the small drift you have observed.  You'd have to leave them for a longer time--like 24 hours--to see how much they wander around even after fully warmed up.

[floobydust]

I dislike a fan in a multimeter, the 34461A's fan moans and wails because they put in a high airflow fan under low voltage so it's noisy. It should be a low airflow/low? pressure drop fan.
As if it has a Cray inside, I still laugh about it because a RPi has much more computing power and needs no fan.

Problem I found is the lab HVAC causes air temperature to move up and down a few degrees, either from A/C cycling or furnace heat.
So the 34461a's temperature is tracking that- and it's a myth it settles to anything in that environment. Oh, and the fan airflow is right in-line with the ref IC instead of tucked away in the corner like the 3458A.

[Kleinstein]

The fan can help reaching a stable temperature and thus stable reading faster, has the is thermal contact from the interals to the environment and thus a shorter thermal time constant.  In the comparison it looks the KS meter with the fan is not faster though. There are other factors than just the temperature and some dirft parts depends on more local temperatures and don't care about the other end of the meter (e.g. the power supply part).
Another factor effecting the time needed to reach a certain fraction of the initial shift is, if there are multiple componenets that in part compensate, as in the 34461 (first going up and than down) this makes the initial part look fast and the later part look slow.

The 34461 has two more digits and about 1/100 the drift - so no really big surprise.
Part of the difference is that 34461 can use its prime range without a gain stage for 5 V. The SDM3045 preferred range is more like 2 V as it used a low voltage range ADC. So it needs to use an extra divider stange. I don't konw the details on the SMD3045 internals and they are a bit strange (the HW seems to support different ranges than the SW) - so not sure which range is actually good and which range is poor.

A comparison at some 1 or 2 V, or with the 34461 in the 100 V range may look different.

The SDM3045 is internally relatively similar to the larger brother SDM3055 that has a fan, though likely not significant higher power consumption.

Problem I found is the lab HVAC causes air temperature to move up and down a few degrees, either from A/C cycling or furnace heat.
So the 34461a's temperature is tracking that- and it's a myth it settles to anything in that environment. Oh, and the fan airflow is right in-line with the ref IC instead of tucked away in the corner like the 3458A.

That is an interesting point. So a slow thermal response also has its positive side.

[Amaruk]

Good insights into DMM design here. Thanks!

I should just mention that the Siglent SDM3045X was in its 6V range and the Keysight 34461A was in its 10V range during the measurements. As a result, the Keysight DMM only shows one more decimal place in its readings.

While drift, accuracy and precision are three different things I find these things all interesting. The Keysight DMM has a quoted accuracy of 35ppm and the Siglent has a quoted accuracy of 100 ppm (or 200ppm depending on where you get the numbers). Thus, the ratio we are looking at here is in the order of 3 to 6 and not 100. Again, these are different figure of merits. It is just interesting to compare them.

I really appreciate that the Keysight DMM has been designed with such a small drift as observed here. It can be used almost immediately after it has been turned on. Way to go! 

[Rydda]

To answer the OPs original question: I really, really hate instruments with noisy fans. Especially instruments that according to the dissipated power should be able to manage without a fan.  I know that I (with ambient noise level next to none in my lab space) am probably part of a small minority of customers. It seems that manufacturers ( and reviewers on YT ) all base their noise level judgements on industrial environments with a large amount of ambient noise.
 
My latest hate object was a 34461A, which I like everything about, except the noise.

To prepare, I used a termocamera to be able to judge the effect of any modifications on the internal temperature
I replaced the fan with a  Noctua NFA4x10-5V,  selected a speed so that it is barely audible and modified the fan mounting to remove air flow restrictions.

After the changes, the temperature map inside the case is practically the same.
I have tried to compare instrument readings before and after modifications, and I cannot see any difference.

I now have a multimeter that does not pollute my environment with fan noise. It can run all day and be no more obtrusive than a handheld one.

[Fungus]

I really, really hate instruments with noisy fans.  I know that I (with ambient noise level next to none in my lab space) am probably part of a small minority of customers.

I'm not sure the minority's so small that the manufacturers couldn't make an effort. They don't seem to care though and just slap the cheapest possible fan in there, job done.

To prepare, I used a termocamera to be able to judge the effect of any modifications on the internal temperature
I replaced the fan with a  Noctua NFA4x10-5V,  selected a speed so that it is barely audible and modified the fan mounting to remove air flow restrictions.

After the changes, the temperature map inside the case is practically the same.

I remember one of the forum members opened up his Rigol DS1054Z, disconnected the fan, and forgot to connect it again. It ran just fine for months until he noticed it. I'm not recommending doing that but the spec of the DS1054Z says ambient temperatures up to 50 degrees Centigrade so there's probably some margin for a fan with less airflow.

[LaurentR]

To answer the OPs original question: I really, really hate instruments with noisy fans. Especially instruments that according to the dissipated power should be able to manage without a fan.  I know that I (with ambient noise level next to none in my lab space) am probably part of a small minority of customers. It seems that manufacturers ( and reviewers on YT ) all base their noise level judgements on industrial environments with a large amount of ambient noise.

Same here. My bench is in a very quiet location and I hate noisy equipment.

I haven't dared fan-swapping the 3446x but maybe I should. The sound level is not super high but the whine is really annoying.

I did fan-swap the 33522B function gen (less worried about losing accuracy). That one is louder than the 3446X and the fan is hard to access.

I also got a Instek PEL-3031E electronic load last year. It has a decently large fan which is thermally control, but even at "idle", it sounds like a jet engine. Fortunately, they use a common off-the-shelf fan and I managed to find a quieter model in the same series. It's also very easy to replace.

I had fan-swapped all my Rigol gear when I still had it.

The only recent piece of TE I have that's appropriately quiet is the E36313A supply. It is basically silent at idle and not too objectionable at full power. They still managed to screw up the "off" mode where the supply periodically wakes up and the fan spins for a few seconds...Not a real annoyance but you don't expect anything to wake up and turn the fan on when it's off.

The most annoying part is that there is rarely an indication of noise levels in the datasheets and technical support usually has no more information.

[guenthert]

     Fans might be necessary for ATE which might end up in racks exposed to uncomfortable temperatures.  For bench instruments they are silly.  I don't buy the 'reaching thermal equilibrium quicker' argument.  For that to happen the fan needs to be temperature controlled (with a well adjusted PID controller to avoid oscillations), which sadly (and incomprehensibly) they typically are not.

     There doesn't seem to be much rhyme and rhythm to which DMMs have a fan either.  First generation of HP3456A had them, later revisions not (I'm not aware of other major changes between the revisions).  The popular HP34401A doesn't have one, yet the successors (including 34410A) do.  Top end HP3458A has one,  yet the top end models of Datron, Fluke achieve comparable precision (not speed though) without.  And even the 8.5 digit Datron 1271, markedly intended for industrial ATE, doesn't have one.

 

[bdunham7]

For bench instruments they are silly.  I don't buy the 'reaching thermal equilibrium quicker' argument.  For that to happen the fan needs to be temperature controlled (with a well adjusted PID controller to avoid oscillations), which sadly (and incomprehensibly) they typically are not.

The issue is internal temperature gradients, not the overall internal temperature.  And for that, a fixed-speed fan does achieve equilibrium much faster.  This mainly allows the manufacturer to get away with some less expensive parts.  The Keithley engineer that used to be here in the DMM6500 thread explained that.  If not for fans, any DMM in the sub-50ppm arena needs some very carefully selected and trimmed resistor networks and that gets expensive.  Otherwise you just have to tolerate a 3-hour or so warmup for best accuracy.

[Kleinstein]

The time to reach thermal equilibrium depends on the heat capacity and thermal resistance to the environment. The fan lowers that heat resistance and thus speeds up the time to reach equilibrium. In addition the overall temperature rise is usually somewhat smaller and thus a smaller step to start with.

Another point is that without a fan the air exchange in / out of the case can be fluctuating, e.g. driven by drafts in the room or pressure from someone opening or closing doors or wind gusts. This gives extra temperature fluktutions a fan can reduce.

A temperature regulated fan has the problem that the speed of the fan also effects the temerature gradients. So ideally one would need calibration depending on the fan speed. To speed up the warm up, the option would be to start the fan only with a delay, like after some 10 minutes. In principle simple, but could be confusing for the user.

The need for a fan also depends on the power level in the instrument: ideally a DMM would be relatively low power (e.g. 5 W)  and could get away without a fan. A somewhat longer warm up is usually not that important, as one rarely needs the best accuracy. Than a longer power on may be needed anyway, not just for the temperature, but also for other effects (stress from thermal mismatch and humidity effects) a fan can not speed up much.

[guenthert]

      Well, yes, for voltage dividers temperature gradient is relevant rather than the temperature per se, but that's not the case for other components, e.g. shunt resistors.

      I also don't think air speed itself needs to be taken into account when calibrating.  Presumably the fan controller would adjust air speed so that a target temperature within the chassis is maintained, regardless of heat generation and ambient temperature, i.e. for warmer ambient air, more air will be pulled which should then also equalize the heat being carried away from components (same for more humid air).

     Clearly a (controlled) fan would not completely eliminate warm-up time.  The fan ought to start immediately at high speed in order to i) overcome friction and ii) assure that a mixed air reaches the sensor(s) and (perhaps quickly) slow to the appropriate speed, but never stop (so that the sensors aren't isolated).

     The chassis-internal target temperature of a controlled-fan equipped device (necessarily) would be higher than any expected ambient temperature, which might accelerate drift or aging of components and hence be considered undesirable.  Further the control-loop will cause some fluctuation in the chassis-internal temperature depending on load (e.g. due to high-speed measurements or high-current across a shunt resistor) which might be worse then using a high, steady air flow when ambient temperature is stable and within specified range.  I wonder if it has been investigated whether that's the case or the engineers just chose the simple, cheap solution ...

[alm]

First generation of HP3456A had them, later revisions not (I'm not aware of other major changes between the revisions).

They moved a TO-3 power transistor from inside the case to the back. Presumably moving some of the heat dissipation outside the case.

Top end HP3458A has one,  yet the top end models of Datron, Fluke achieve comparable precision (not speed though) without.  And even the 8.5 digit Datron 1271, markedly intended for industrial ATE, doesn't have one.

I believe the high-speed comparators necessary to give the 3458A it's high-speed digitizer feature get quite hot. This may well have played a role in the use of a fan. I don't believe the Datron/Fluke designs have this same high speed.

Well, yes, for voltage dividers temperature gradient is relevant rather than the temperature per se, but that's not the case for other components, e.g. shunt resistors.

Temperature gradients are also responsible for thermal EMFs.

[mapleLC]

I've had success so far with the following and Noctua fans:

Keithley 619 - also switched fan from AC to DC
Keithley 237 - also switched fan from AC to DC

Agilent 546xx oscilloscope (x 2)

Several network switches

Its situational.  I don't plan to run any of these at industrial levels, so they don't get hot.