Control.com - Nerds in Control

I've got an RO system which gets its reclaimed water from a a large holding pond, where it is pumped into a tank, then pumped through multi-media filters, then prefilters, then into high pressure feed pumps and into the first pass ROs.

The bio load of the reclaimed (sewer water) is quite high, especially during Arizona summers, so we inject bleach before entering the multimedia filters. Then, because we're using thin film membranes, we dechlorinate using sodium bisulfite.

The problem is poor control of the free chlorine levels, which results in poor dechlorination control due to the large swings in free Cl, which is bad for the membranes. The DCS measures the free Cl after the multimedias using a Hach Cl17, which is well downstream of where it's injecting ahead of them. It's using a Bailey Infi90 advanced PID control algorithm, but we can't tune the oscillations out of it.

I'm getting better results by feedforwarding the chlorine right now with metering pump speed based on a simple ratio of the feed flow to the MMFs, but it's still not tight enough and is subject to uncomfortable excursions.

Is this high dead time PID loop (with variable transportation lag plus the Cl17's 2.5 minute sample rate) tunable at all, or I should switch to a different control scheme and what should that scheme be?

Deadtime is the biggest enemy of any control loop. Imagine trying to drive your car if the wheels only turned 2.5 minutes AFTER you turned the steering wheel!!

Realistically, the "absolute best" performance you can expect from a feedback (PID) loop is to correct an upset within approximately 10 deadtimes. In your case, if you have a 2.5 minute sample period, plus "some" transport delay, you have an effective deadtime of 3+ minutes, so it will take over 30 minutes to correct for any upset.

If you try to correct faster than this, the loop will oscillate. So de-tuning the feedback control will reduce the oscillation. But still, it may not be enough...

The best way to fix the problem is to reduce the deadtime. (Easier said than done!) Here are some ways to reduce the deadtime:

1. Increase the sample speed. 2. Relocate either the instrument or the point of injection, so they are closer to each other.

Feedforward is a good idea. If you can measure the disturbance (the demand for chlorine), and respond to it, then you will have reduced the upsets to the process. This means there is less work for the (slow) PID loop to do. Because feedforward is never perfect, you still need the PID loop to "trim" the control.

Here's an article you may want to review: "How to Control Processes with Large Dead Times". Just follow the link below: http://www.expertune.com/r2p.asp?f=AList <http://www.expertune.com/r2p.asp?f=AList&l=artdt.html> &l=artdt.html

Good luck!
George Buckbee
<http://www.expertune.com/>

A long time ago, I was the principal author of the Manual of Practice on Disinfection of Wastewater for the WEF. And I engineered, built and sold disinfection systems, as my brother-in-law and sister still do. I still remember a lot.

The fact that the law requires that you measure the residual after the contact tank means generally that you have impossibly long lag times for feedback or PID control.

What you're doing, with feed forward control and a residual trim is what we've been doing for the past 30 years, both for gas chlorine feed and hypochlorite feed.

One of the problems you have, also, is that the concentration of the hypochlorite solution itself is changing, because of the heat.

You should certainly try to get an analyzer with faster response time than two and a half minutes! If you do, you can then use the chlorine analyzer to trim.

Another big problem is that chlorine is not very soluble in water, and therefore the disinfection activity is slow and not instantaneous. That is probably why you're still having excursions, in conjunction with the variable bioloading of the reclaimed water.

If you want to talk about this more offline, give me a call.

Walt Boyes
Editor in Chief Control magazine www.controlglobal.com blog:Sound OFF!! http://www.controlglobal.com/soundoff _________________

Putman Media Inc. 555 W. Pierce Rd. Suite 301 Itasca, IL 60143 630-467-1301 x368 wboyes@putman.net

You can genereally correlate free chlorine to ORP, and that can be measured much faster than every 2.5 minutes. There is not a direct correlation. It
has to be done empirically so you still need the chlroine monitor as a final check.

Greg,

In addition to the responses from other list members, I would like to say that model predictive control might be of help in this application. Large deadtimes are not a problem as long as the prediction horizon is long enough. Besides, the influence of measured disturbances could be incorporated into its multivariable (possibly nonlinear) model. Please, feel free to contact me
if you need more information on this.

Regards,
Andrey Romanenko
// Ciengis
http://www.ciengis.com
+351-239700353

I am a product specialist in the chemical division at Norweco, Inc. (http://www.norweco.com). We have run into many problems with this similar type of application before. I will agree with some other replies stating that the liquid chlorine being fed will have an inconsistent strength, so you have to rely heavily on the reactiveness of your RO system. In many cases, we have found that the RO systems work wonderfully in some applications and may hinder in others. You may consider removing the use of the RO system entirely on the dechlorination process, and instead use a gravity flow tablet feeder system. The feeder works in conjunction with 92% sodium sulfite dechlorination tablets to completely remove chlorine to non-detectable levels without the need for RO systems, measuring or mixing devices. If you would like to learn more about these types of systems, please feel free to contact me at (419) 668-4471.