PID loops (again)

Thanks for all replys to my previous post.Very helpfull.Ive managed to tune all the loops for liquid level control.
Im still struggling though trying to tune one of the loops.Im trying to tune a loop controling dissolved oxygen in an oxidation ditch. PV is the measured DO and the output of the loop is controlling a variable speed air blower.Ive tried tuning it as with the other loops but there is a large lag in response and as result the blower speed is hunting from 0-100% whatever I try.The trial and error method is now quite exhausted.Should I be using just PI or would PID be better? also how do I get round this delay time?

rob allan said:

Ive tried tuning it as with the other loops but there is a large lag in response and as result the blower speed is hunting from 0-100% whatever I try.Should I be using just PI or would PID be better? also how do I get round this delay time?

Click to expand...

Sounds like the PID from hell. Use the PID. The differentiator provides damping. I would rely on the PID less and a openloop estimation more, and use it as a feedforward or bias.

Make a table of dissolved oxygen verses control output. The more accurate the better but if you can get with 10%, this will help a lot. Use the SP to find an estimated control value and use this as the bias. If you do this right your system should control the %oxygen within a few percent of desired with the PID gains turned off.

Now you can enable the PID gains, but they can be MUCH smaller because the estimation is doing most of the work. When I say smaller, I mean they should have less effect on the control output.

Don't make step jumps in the SP. The SP should be ramped smoothy from one SP to the other. Obviously your system cannot respond to fast transients.

You can use PI with a very high I value if you are using siemens.. or very low I if you are using allen bradley

D may not be required..

Please note the output goes to proportional valves and not to drives

Take a look at my website www.escor.org. I can't tell you the secret to my success, since I make my living selling DO control systems for wastewater treatment plants, but Peter is right - DO control is not the right application for PID loops. I abandoned PID control for DO about 17 years ago, developed my own algorithm, and have never regretted it.

Besides the very long process lags, the oxygen uptake rate (an indication of process loading) varies dramatically over time. The oxygen transfer efficiency of the diffusers varies with time, mixed liquor temperature, and air flow rate. The response of DO to air flow changes is generally 3 to ten minutes depending on tank configuration and mixing and hydraulic loading. I have often observed DO concentration continuing to rise even after air flow is reduced, and DO will continue to fall after air flow is increased. If the operator changes his MLSS by wasting too much sludge then the dynamics of the process change again. Add to that blower limitations and so on and PID is just about impossible to tune and get stability and reasonable response. Peter's table aproach is problematic at best because of all these variables. If you can control DO within ±0.5 ppm consistently with any conventional algorithm you are doing well. Fortunately the process is forgiving, and this level of control is adequate for most processes.

WARNING - BLATANT COMMERCIALISM COMING: If you want to contact me offline you can do it from my web site. We've done a couple of jobs in Spain repacing PID and other DO control systems with our proprietary system. I've always wanted to go to Great Britain!

Tom, the thread is old.

However, it will be interesting to see what Tom does and how he does it.

Aw shucks, fellers - Roxio's post kicked it up to the top of the list, and I didn't never notice thet the original post was old as the hills.

I wunders whut ole Rob ended up doin!

Thanks, Peter. I reckon I'll jist call up BA and cancel them there tickets. Heathrow is probably crowded this time of year anyhow!

Thought it was a bit strange getting an email telling me someone had replied to my post.I couldn't remember posting anything for a while.
Im back doing water work at the minute which makes a change from working in the motor industry for a while.
I never could sort that loop out though!
Didnt matter though as the system wasnt up to spec anyway and we ended up running both duty and standby at 100% and still couldnt achieve the clients spec -one of them jobs!

Still think this is great website,its helped me out plenty of times

Ill be going now as I have beer waiting and a missus who gets annoyed when I work all day then trawl the net when I get home

Gan Canny

Rob

It find this interesting.

Tom Jenkins said:

Take a look at my website www.escor.org. I can't tell you the secret to my success, since I make my living selling DO control systems for wastewater treatment plants, but Peter is right - DO control is not the right application for PID loops. I abandoned PID control for DO about 17 years ago, developed my own algorithm, and have never regretted it.

Click to expand...

The VFDs require analog set points so you couldn't be using simple on/off control. You must have a formula for a model that you use and you calculate the coefficients using the Excel's LINEST function or Mathcad's LeastSquares functions for predicting the response. Even better would be to have the PC calculate this every 10 minutes or so. That way your model could adapt to changing temperatures and all the other things you mention. I may be making this more complicated than it needs to be, but the math is the easy part.
Your systems are not trivial. If you are using a model then I bet you have many terms in the formula. Do you only use a model and update the model?

We use two different systems. The old standby is based strictly on feedback from dissolved oxygen and air flow, and uses a very elaborate variation of floating point control with things like rate of change, temperature compensation, deadband, delay times, and so on. It is much more stable than PID, and still responds rapidly to sudden loading changes and so forth. The tuning is much more intuitive to a typical operator than proportional band, resets per minute, etc.

We are in the process of patenting and commercializing a feedforward system with a process model. This was never done before because there was no instrumentation to measure oxygen transfer efficiency and oxygen uptake rate in real time. We developed the instrumentation, and the results of the model based control have been very gratifying. Here are the key equations used in the process model:

Cool, a very fancy P controller with a model that changes on-the-fly.

Did you develope those equations yourself? I expected the formulas to be about as complex the ones you presented. I can see why you have been successful at your type of control. Having the real time feedback is a big help too.

This backs up a point I have been trying to make about PIDs. A lone PID is not the best solution except for the simplest of temperature systems. Learning how to use the bias input for feed forwards is key.

Re: Cool, a very fancy P controller with a model that changes on-the-fly.

Peter Nachtwey said:

Did you develope those equations yourself?

Click to expand...

Don't I wish! There was a team of 5 guys involved in this, some of them from Europe. A couple of them go all over the world to do design, analysis, and testing of aeration systems. The role of Tim (my main engineer) and myself was to take their theoretical knowledge, turn it into hardware, develop the programming for the modelling calculations, add the ability to control the mechanical equipment, and build and test the whole system in the real world.

The various bits and pieces of the model, especially the equations with all the Greek letters, have been part of the standard aeration system design procedure for decades. Noone had ever atttempted to use them for control, though. Our team combined the various chunks into a coherent whole, added some terms for accommodating the non-steady state considerations, designed a system for measuring the necessary input data in real time, and added the experience with controlling air flows and blowers and coordinating it with the process model.

We'll know how clever we were when the various Patent Offices get done with us! I must brag, though, that it was gratifying when the plant operator told Tim, "You know, this control is really a lot better than we need."

And for those who care about such things, the PLC doing all of the process measurement math and control was an AutomationDirect.com 250 networked with three other AutomationDirect.com. PLCs for various machinery control functions over an DirectNet RS-485 serial link. The data collection and operator interface for setpoint entry and monitoring (no control or calculations!) was LookOutDirect on a standard PC with Win XP. All of the data was logged into CSV files and imported into MS Excel for anaysis.