How to Compensate for Asymmetric Flow in a Square Duct?

Whether it is for gas or liquid

In similar situations (that you have to linearize a problematically unlinear signal) I've used A/D conversion, manipulated data to my needs (10 bit data has .1% error threshold) and then back to analog. A single low-end microcontroller with at least 3 A/D channels (the second to sample output and pump to it with a RC network from a digital output so to eliminate the need for separate D/A and the 3rd to get fluid temp if you need mass measurement also) can do the trick. Cost is generally moderate. Of course you 'll need some reference fixture to calibrate the whole thing. Just an idea. S.M.

And to make my post a little more comprehensible, the only condition the position of your (single) flow sensor must meet for my proposal to work, is that the sensor output is monotonic to the fluid flow. S.M.

The original location was really squirrelly. The new location is 4d downstream of the second elbow so I may be able to do the look-up table if the profile is consistant. (Re is 5x10^6 at 3200 ft/minN2) (We try to calibrate today.) I think calibration at the new location has a better chance of working. The first straight run is 6d, which is close to the point that the blower profile should have smoothed out, before hitting the elbow.

If the profile is still variable, I'll take my best guess after getting everyone's input. I'll post what we do and the results.

The volume being moved is 7200 CFM, which may make direct volume measurement impractical.

What is the geometry of the inlet to the fan? Does the fun draw from the process or the room? If from the room can you put the necessary straight length of round duct in?

The flow is a loop.

The process is blower--> heater--> dryer--> filter--> cooler and back to the blower. All of the duct is 18x18 square. None of the duct has a decent straight run.

The connection between the cooler is neither a bell, a duct straight run nor plenum. It is a sharply converging transition, w/ half angle of 80 deg, so it is not a suitable location either.

How big is the duct and what is the Reynolds Number in the duct?

The duct is 18"x18" & Re is ~ 6.2 x 10^5 (~ 7100 acfm, N2 w/ 3% air, 55 F dew point, 10" wc static pr.)

I wonder if there might be a section of duct that, taken in total, can serve as a consistent restriction. This would assume that this duct section remains relatively clean. Then static ports upstream and downstream could measure a differential across this "restriction".

A temperature probe (or a few) with some thermal mass (or processing) to damp out local temperature variations temperature variations could be used for temperature corrections.

Then the flow would be calculated from pressure drop corrected for temperature. My theory here is that the static pressure (especially in the boundary layer) would be less sensitive to duct turbulence. (I could be completely full of it, however.)

In any event, I think an empirical approach would have to be used (with the solution perhaps including long term averaging from a group of pitot tubes, etc. ) In wind tunnels flow straightening is essential -- and for similar reasons it might be essential for your purpose too.

Other possibilities:

Measuring fan motor electrical power?

Heated-wire mass airflow sensors in a grid pattern? (four vertical wires, four horizontal, etc)

Measure drag on a grid (a piece of honeycomb that spans the entire duct, for example)

I considered motor power, but think it would be a relative indication. As I expected, I was unable to get the developed pressure Vs flow rate from the blower curve to match a traversed flow. The blower is not ducted in a way that I can apply a known SEF to the blower curve. (Service Effect Factor - a derating of the blower published by AMCA).

Interesting idea to do the measuring around a "restriction" but doesn't that suffer from the same issue of flow profile and inlet/outlet straight runs? I imagine that is true even for an elbow meter. Also does anyone do gas proving at such low dp, to "prove" to some standard (NIST?)? I couldn't add much permanent pressure drop or it changes the blower requirements.

I considered motor power, but think it would be a relative indication.

Definitely. In most of these suggestions I have been thinking that the actual flow would need to be measured by another method (that could required some disassembly, insertion of flow straighteners, turbines, etc). Then the process control measurement method would be calibrated against this flow determination.

Does the total flow (as averaged from the individual flows from a 5x5 matrix) remain constant, or does even that seem to vary?

Not sure how to answer your question. At steady state (same ~ flow) two sets of traverses (meter in the first straight run near the blower) gave essentially the same answer, however the traverses were performed without moving the system flow away and back to the same operating point.

Plotting the Pitot flow Vs the thermal dispersion flowmeter, did NOT produce a reasonably straight line for a set of 5 data points. Inspection showed the profiles could be grouped, as skewed or nearly fully developed & symmetrical. plotting the profiles that were similar generated two nearly parallel lines. Since the thermal dispersion meter is a single point thermal dispersion meter, I could not see how I could approach 5% accuracy if the profiles depend on how steady state was approached.

As a result I have now moved the flow meter away from the blower, which is the original disturbance, but I still am too close to a double elbow to meet normal L/D guidelines. so I cannot claim fully developed

Currently I'm working through new Pitot calibration data with the meter in the straight run downstream of the elbows (no flow straighteners). This time the Pitot taps are on top, allowing me to extend the grid to have two more data points in the vertical, to assess what is happening above and below the normal 5x5 grid. (e.g. sampling closer to the top and bottom walls)

With the meter in the new location, the profiles are more consistent, but the profiles are still skewed to the bottom. The extra data point confirmed that the 5th grid point is the peak flow. I can't say whether this was true in the first data set, since the taps were oriented horizontally.