fan feedback algorithm

As usual, John Popelish has sorted it out. It may be useful to add that the pressure drop across the box is probably proportional to the square of the airflow/airspeed through the box, and there is a fair risk that the flow will be laminar at low flow rates and turbulent at high flow rates (search on Reynolds number) which will give you a nasty kink in the linearisation table as the flow goes through the laminar to turbulent transition or back again.

--
Bill Sloman, Nijmegen

I would begin by creating an experimental linearization table that predicts PWM outputs that produce linear flow steps. Then when the flow control has an error, its output gets translated to PWM by interpolation from the table. This may not solve all your problems, but it will linearize motor's part in the loop gain. You may have to do the same thing with the flow sensor's signal. After that, you should be able to apply a time sampled PID (or just PI) controller that updates every second or, perhaps a little faster. If the flow signal is noisy, you may want to filter the flow signal so that each time the controller computes a next output, it is using something like the average flow signal since the last update. Then you can tune the controller gain and integral (and possibly derivative) to achieve faster response and better stability.

^^^^^^^^^^^^^^^^^^

^^^^^^^^^^^^^

???

Are you using PWM or not?

You're doing your PWM all wrong.

Sounds like you need to find somebody who can write a PID algorithm.

And whatever you're doing to get the PWM, it's wrong if the nonlinearity is as bad as you say.

What do you have so far? I've done PWM to the gnat's ass, maybe I can help.

Hmmm, in the few seconds of review before hitting send, it's struck me - are you doing PWM by division? That's not right, somehow - there's a lot more difference between 1/2, 1/3, 1/4 and so on than between 1/4093, 1/4094, and 1/4095.

Just count clocks - that should linearize it some. :-)

Cheers! Rich

A biplar supply could provide braking. :-) (or, of course, an H-bridge.)

If it _were_ any faster, I'd worry! ;-)

This is a good thing - you can get "virtually" "instantaneous" data. ;-)

Cheers! Rich

Do you need precise air flow or is this a situation where you need a minimum airflow but would prefer not to outrageously overdrive the fan (such as for audible noise reduction)? This approach would be easier.

Is the fan true DC such that compensating for the fan's resistance could produce a nearly load-independent analog speed control? (The PWM would be lowpassed or a DAC substituted, driving an amplifier with negative resistance to cancel the fan coil resistance) Some "DC" fans - brushless fans - have electronics that modulate the drive coils which wouldn't apply to this approach. If this compensation could be included, the load-independent speed control makes the control system much easier to manage.

snipped-for-privacy@ieee.org wrote: (snip)

I would want to use an independent air flow test like a hot wire anemometer or other thermal mass flow meter or a propeller flow meter to create the table for the orifice pressure drop sensor to include all those factors.

On 18/01/2006 the venerable snipped-for-privacy@wwc.com etched in runes:

I am doing a very similar project, but using an air velocity sensor on the input duct. I found these references very useful.

The electronics is a copy of the circuit in the Maxim app note. The fan has a

24VDC brushless motor and the op-amp (TL082) runs from +24 & -5 with a TIP122 as the follower. A 4k7 dummy load is fitted across the fan to help stability.

--
John B

Delete \'spam blocker\' to reply direct

At any given time, there is a pressure error. This is the difference between the desired pressure and the actual pressure.

When you change the fan speed, there is some amount of time which passes before the pressure becomes more or less steady. We'll call that the settling time. It is essential to what follows that you have some idea of what the settling time is. You should be able to measure it by performing some experiments. It may be a function of the step size or of the actual pressure, or even of fan speed. Just try to find the worst case settling time.

OK. So I would suggest that you use a two-mode algorithm. When the pressure error is large, you go into a seek mode designed to rapidly shrink the pressure error. Once you get the pressure error below a certain level, you go into a track mode. Let me elaborate.

In seek mode, you set all PWM bits to zero, producing minimum pressure. You wait for one settling time, then check the pressure. If the pressure is too low (it should be) you set the most significant bit (msb) of your pwm word high. Wait one settling period and check the pressure again. If the pressure is now too high, set the bit back to zero. If it is still too low, leave the bit set. Now go to the next most significant bit, and do the same thing. Do this for all twelve bits. Assuming the desired pressure does not change, and the physical characteristics of the system do not change rapidly, this will get you to the optimal setting very fast (well, it will take 12 settling periods). Once you have gone through all twelve bits, go into track mode.

In track mode, check the pressure. If the pressure error is too large (you will have to decide what "too large" means), go back into seek mode. If the error is not too large, then just increment or decrement the PWM word by one, depending on whether the pressure is high or low. Then wait one settling period, and repeat.

It may be that track mode isn't even necessary. Certainly, if the mechanical part of the apparatus is static, I don't see why you would ever have to adjust the PWM once you find the optimum setting. In that case, you can use track mode to monitor the pressure error and re-start seek mode if necessary.

I hope that helps.

--Mac

I suggest a classical PID (Proportional, Integral, Differential) algorithm to start with. In your description of the current algorithm I saw no mention of the equivalent of what would be called the Differential term in the PID algorithm. That is where you create feedback based on the rate of change of the error signal. In systems will large amounts of inertia, such as your, differential control can be very useful in stabilizing a feedback loop.

Beyond a standard PID loop, your description of the non-linearity suggests that a modification is in order. Perhaps the P, I, and D coefficients cannot be constant but should vary depending on the operating point. Perhaps determine what that dependence is by separately optimizing P, I, and D at several operating points. Then contruct a look-up table or a simple continuous function to develop these coefficients in actual operation.

-Robert Scott Ypsilanti, Michigan

Fuzzy has been promoted for this purpose, but it's certainly not the only way.

Do you have tachometer feedback on the fan? I'd be inclined to look at using two loops-- inner and outer-- and some linearization. Also, if you can use proportional only control (the main consequence being that some steady state error, perhaps all on one side of the 'setpoint' is acceptable) I'd use that.

Yes, I assume your filtering out the noise from the fan turbulence. There's the inertia of the fan blades and compressibility of the air in your box to contend with.

Best regards, Spehro Pefhany

--
"it\'s the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

[snip]

Maybe linearize the fan in terms of flow?

...Jim Thompson

--
|  James E.Thompson, P.E.                           |    mens     |
|  Analog Innovations, Inc.                         |     et      |
|  Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
|  Phoenix, Arizona            Voice:(480)460-2350  |             |
|  E-mail Address at Website     Fax:(480)460-2142  |  Brass Rat  |
|       http://www.analog-innovations.com           |    1962     |
             
     It\'s what you learn, after you know it all, that counts.

Your comments on my proposed solution are definitely correct. I was shooting for something which would be stable (despite any non-linearities in the system), simple, and reasonably fast.

I got the impression that the current solution (the one the OP doesn't like) is a badly tuned PID type algorithm. I could well be wrong.

Anyway, I think the OP disappeared. ;-)

--Mac

Are you sure the basic idea is okay? Measuring pressure seems like a bizarre way to measure flow rate. What happens if the exit path gets clogged by dust? Pressure goes up, but flow goes down! What happens if the inlet temperature goes up or down? If the goal is to keep something in there cool, you're measuring the wrong thing.

Also bothersome is the bizness about 12-bit resolution. Heat radiation goes as something like the fourth power of temperature, so at low temperatures you're talking about very low dissipation. Sounds like you could use a log converter in there somewhere to compress the range.

Now you may not have the authority to re-engineer the whole loop, but maybe a word to the right person may take this problem off your back.

I would go further than that, the orifice implies constant flow vs pressure so that you can tabulate the PWM vs pressure characteristic. When he gets a large step change in pressure setpoint he can steer the PWM open loop to within a safe *ratio* of the tabulated value before he lets the linearized PID model take over, and this linearized model, and even the ratio switchover, will be paramterized by the projected operating point. E.g. less gain at low speeds, much more gain at high speeds. Note it oscillates at low speeds because of the greater gain in velocity vs PWM. This type of thing should reduce that 15 minute settling to 1.5 minutes, but not much better than that with a 30 second fan speed lag.

ISTR that fan power required runs on the order of 3rd to 4th power of flow rate. College was a time ago and i do not remember everything, but some surprising results stuck. maybe two fans is the correct answer, one for upto about 10% of full power and a second for the rest of the curve.

--
JosephKK