How to accurately measure mass flow through server vents using an anemometer

[thejollyroger]

I have to work out the air mass flow coming out each vent on a server and have proposed to use a hot wire anemometer to measure the air velocity at each vent by placing it slightly inside one outlets of the vent (it is a grid pattern), then multiply that by the vented area and air density to find mass flow.

My concern is that I am instructing technicians to do it on site, and I am worried that the measurements may vary significantly depending on where on the vent they are taken from i.e. if they take it from the corner or side of the vent it may be different from the centre.

This is my first time trying to measure something like this and was wondering how others in the industry do it. Is there a better way of doing this?

 

[IRstuff]

Please refrain from multiple posting.

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

 

[LittleInch]

The second post has now gone, but I think this was discussed a while ago and it was clear that a single velocity measurement x area is pretty inaccurate.

Look at this:

http://www.tsi.com/uploadedFiles/_S...pplication_Notes/AF-106 Traversing a Duct.pdf

and also scroll down this rather rambling thread

http://www.eng-tips.com/viewthread.cfm?qid=424482

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

 

[itsmoked]

You are correct in that the probes would have to be PRECISELY placed uniformly. AND, it wouldn't be even close to correct to do this and 'the math' and think that's correct. You would need capture all the exit air into a duct and measure it in the center while checking with the calculations provided by the sensor supplier to correct for the shape and dimensions of the duct and a single central sensor. Using this fairly close result to calibrate your 'simple scheme' the simple scheme would then be able to provide a realistic flow number.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[thejollyroger]

See my issue seems to be that we are trying to measure the flow coming out of a grid shaped vent, and not a duct. If it was a duct I could work out (similar to the procedure in the above PDF) the velocity profile and hence average velocity. But being a vent on a much larger square box, a fully developed velocity profile would not be present and it would be difficult to calculate the average velocity unless we took many measurements across each vent.

I could get the technician to take several measurements across each vent, but there are a total of 14 vents and he has to place the anemometer in place then shut the rack doors for each measurement, which will be quite tedious.

I was thinking about getting the tech to take a measurement from the middle of each vent to find the maximum velocity. We have a similar server here at head office that I could then take some measurements from to develop a rough velocity distribution across each vent, then scale that up or down with the maximum velocity the tech found.

Does anyone have a better idea on how we could do this?

 

[DrRTU]

A hot wire Is not the correct tool. You will need to build a section of duct out cardboard then do a traverse. Buy or rent a hood.

https://www.raecorents.com/tsi-alnor-balometer-ebt731-electronic-balancing-tool/

Hire a TAB contractor. Why invent the wheel? An independent will give you data with certs on the tester and instrument.

 

[LittleInch]

Ok, I think you mean something like this?

https://19power.co.uk/9u-19inch-wal...with-tempered-glass-door-black-with-lock.html

(a photo would help) and/or a drawing of the "vent"

So you have a fan driven entry for cooling air and this exhausts through a series of metal holes?

I think the variance will be too difficult to average and will vary too much depending on what is in the cabinet and nd how it is fixed, type of tray inside, how well the door seals etc etc

Why don't you measure air in? would seem much easier?

Why are you doing this?, maybe there's a better way.

The flow rate and velocity of these things must be pretty low - you will get a large error trying it your way. IMHO.

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

 

[willard3]

DRTU is right, a hotwire anemometer is the wrong tool.

A TAB contractor will have a hood which will give you much better information.

 

[DrRTU]

Post a photo of the server/rack. We have measured cabinets from raised floor supplies and found large velocity gradients from top to bottom. If you are trying to read the small fans on the server - good luck. Reading the discharge of small diameter axial fans will be challenging. These fans normally have a perf plate on outlet and do not like static added. If it’s a small fan, I would add short section of round and traverse with micro pitot. Room & cables at the server will be the key. Start with 4 diameters upstream and 4 diameters downstream.

 

[itsmoked]

I agree with Little; what is the point of doing this? There may be much better methods if we understand the point.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[thejollyroger]

This is for a server rack in defence, so unfortunately I can't post any photos and we can't hire subcontractors in to do the work. My job is to measure the airflow coming in to the server rack and each individual server, then to measure how much air is coming out of each of the server outlets (it is partitioned for different cards). We need to get an idea of the air flow going through each partition inside the server and how much heat each partition is producing so we can justify altering the server or server rack design (we have had several overheating issues resulting in shutdown of the servers).

The photo below is a photo I've pulled off Google to show similar server venting.

The server that is having issues is not at my site so I am instructing a technician to take the readings. We do have an identical server configuration in our office, but a different server rack setup (which means different mass flows going through each server). My plan is to measure the velocity of each individual square of each vent we have in our office (yes, hundreds of them) and then place the readings in Excel. I have a TSI 9535 and a low velocity pitot tube to take the readings. From there I can build a velocity profile for each vent, which means the technician will only have to take one reading from a certain square and I can scale that value according to the velocity profile that I have for that vent. The reason for doing it this way is because the technician will not have enough time to take hundreds of readings in the field. The velocities will obviously be different values, but the relative velocity profiles of the vents should be the same.

 

[danw2]

Doesn't the air flow through the multitude of tiny vent holes have to be pushed or pulled through those all those tiny vents by the two fans?

Wouldn't measuring air flow of the two fans provide a decent measure of combined air flow?

I'd think that thermal convection is minimal compared to the forced draft from the fans.

 

[thejollyroger]

Those two fans on the server in the picture are just for the power supplies. There are more inlet fans in the front.

I need to measure the airflow going in each server and coming out of each server vent, along with the temperature at each in order to calculate the heat produced by each card. Only then can we decide what needs more airflow and how we can achieve it.

 

[LittleInch]

I think you're doing this wrong. The fans on those units are puny and air flow would be impacted by multiple things which are slightly different to each server. The air flow from your server in the office could be quite different from an identical one in a rak with other things around it.

I would have thought what you need to do is create a grid of thermocouples all over the server cabinet and monitor temperature gain over time. Then you will find the literal "hot spots" and can do something about it.

An admirable attempt to do this from first principles, but I can't see it providing you with the real life information you need.

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

 

[thejollyroger]

I don't think you are understanding my posts, I'll try to clarify.

That server in the photo is not the server I am measuring off - it is a generic image from Google. Yes those fans are small, I don't see why that is relevant though. There are many more fans in the system.

The servers we have in the office are the same as the servers in the field. The rack setup is different, yes. This means different magnitude airflows through each server. However I would have thought that the relative velocity profile would be approximately the same on each server i.e. if you record a single velocity reading from a specific spot on the field server, we can compare it to the same measurement taken on the office server to see how they differ. See below.

Then it is should just be a matter of calculating the average velocity, calculating the mass flow and then calculating the heat flow.

We have already setup temperature loggers at each air outlet. Originally the technicians just recorded temperature, however temperature alone won't tell us how much heat is being produced by the cards. To find this we need the mass flow, hence the velocity meter readings. Again, the reason for doing this is to understand exactly how much airflow is going through each partition in the server, how much heat is being produced in each partition, and how much airflow there actually should be. This will govern how we make changes to the current setup and how we also design future servers and racks.

 

[DrRTU]

Purchase a Flir or Fluke IR camera. Reading flow on the little fans is impossible outside a lab. Cables and racking will cause your test methods to fail. Heat and dirt is the failure means, don't chase the heat removal medium but look at the heat via IR. Cards, drives, inlet / outlet obstructions and "holes" in the cabinet all greatly impact flow. You will have too many combinations to provide repeatable data. Look into the reporting software. With IR you don't physically touch an operational server.

 

[LittleInch]

I think we're completely understanding your posts. I think the general point is that your very nice picture of the velocity profile from your office reading will not bear any resemblance to your field velocity profile due to the many variables you have. The fan size is important because they are so weedy that any slight variation in heat output, internal configuration etc would impact on your venting and flow. I also don't think the system will scale that much and any change beyond 10-15% in any one chosen vent hole would change the shape of your dome.

I think DrRTU is correct - what you're interested in is heat , so go find the heat / temperature. You can worry about the air flow later when you're trying to solve the heat / temperature issue. So instead of just measuring the heat at the outlet, measure many more locations within the cabinet.

Remember - More details = better answers
Also: If you get a response it's polite to respond to it.

 

[itsmoked]

You don't need to measure the temperature of the massflow to get the total heat
rejection you use a wattmeter(s) to give you the precise heat load. There is no
other heat being dumped in the box than the heat brought in by the electrical
power. If this was some huge room with burners and electrical loads and heat
conveyance piping you'd need the complicated air flow volume and temperature to
get your heat rejection number but not with a server.

Keith Cress
kcress -

http://www.flaminsystems.com