How to do flow linearization?

Hi Znago,

You mean you want to make your own controller?

What make and model is the flowmeter?

What type of output does it have (4-20mA, 0-10V, -10V to +10V)?

I would think that most flowmeters nowadays will output a linearized flow anyway. If it is not linearized, you must take measurements over the range that the flowmeter is likely to encounter to determine the transfer function. Once you have the data, you can fit the data to a curve and determine your coefficients. From there, simply use the math functions in your programming language.

Mike

Hi Mike,

Yes, I need to make my own controller. The meter we were thinking of using is from Liquid Controls. 0-12v output of two pulse channels.

I don't think that one would need to actually do measurements since the graph should/would be provided by the meter manufacturer.

What I am trying to understand is, at what time intervals to determine the flow rate and how to make the corrections. I mean, you can't determine flow rate at every pulse and adjust the K factor for every pulse, right? So, is it that you take a fixed time duration (say 5 seconds?) and determine the flow rate for that and then adjust the flow rate for the next sample?

Hope this clarifies my doubt.

Thanks

Znago,

There isn't any single answer for this question. It depends on the conditions and the precision you are looking for.

If the flow changes very slowly, the correction period will be long and vice versa.

You basically want to minimize the reading error. Try to split your curve into piecewise linear section with specific slopes and offset and change set when your raw flow cross a boundary. The more boundaries you choose, the smaller the error but higher processing capability will be needed.

OK, I understand. And I have also seen such a graph that takes the meter curve and divides it in to a number of linear sections. Now the graph is a graph of the K-Factor (on Y axis) vs. the Frequency of the meter pulser (on the X axis).So as the frequency changes, different K-Factors are applicable.

My question is, with respect to how fast or slow one must calculate/apply these changes, is it enough to sample the pulses for one second and then get the frequency? Is that too slow?

The way I understand the algorithem needs to work is as follows:

• Count the pulses (for a given sample time of.....?) and derive the frequency
• Find the appropriate K-Factor for that frequency
• Apply that K-Factor to the sample (i.e. sampled number of pulses) and derive the volume
• Log the volume (i.e. increment the totalizer) and process next sample of pulses

Would this be correct?

It depends on what you are doing with the info. If it is a process feedback, make sure that it is updated much faster than your loop response time. If it is only shown to an operator, it can be refreshed every seconds.

It also depends on the pulse rate. If you get a pulse per second, you might want to switch to a period measurement instead of a pulse counting method. But 1 Hz pulse counting it might be OK if you only need a daily report.

Actually, the typical pulser which is connected to a flow meter will output at several kilohertz.