very large tanks control strategy quesion 1

[martinrelayer]

hello,

I would like to ask if someone was involved in a level control of a *very* large tank or some advice.

We have a water treatment plant, very large, and they want to control a very large output tank level automatically with a water pump
The water demand may change and this causes problems to the level control.
As a constraint we have water turbidity, and "speeding up the pump" causes troubles, so the increments shall be moderate.
The designer made a PI control, that didn't work well, after that someone did like a step logic, working better.
But still the error in the level is too much (setpoint at 95%, you can see oscillating from 80 to 100%) and this is not a PI control, only adding or substracting steps in flow setpoint.
Also you can see turbidity going up when flow goes up.

I was thinking of adding like a correction factor, a PID correction to the controlled variable, like plus minus 10 litres/second to the steps with the same setpoint as the tank level, the same process value and the output or CV adding or substracting to the flow setpoint.

but before I would like to ask for you opinion!!

please see attached graph

kind regards,
martin.

 

[martinrelayer]

sorry forgot the file

 

[PNachtwey]

This is the 3rd post I have seen like this. The answer isn't PI, PID or step control. All of these will cause rapid changes to the flow that will increase your turbidity. The question I have does the turbidity increase if the flow changes slowly or is just a matter of flow rate? If it is just a matter of flow rate there isn't much you can do since the flow out must be equal to the flow in over time. If the change in flow rate is the problem then steps or anything that causes abrupt changes in flow is the problem.

The solution is to make your system into a low pass filter to flow where the inflow can vary but the outflow changes slowly like a low pass filter. The solution is simple, since you are NOT controlling level use a proportional only controller or a proportional band controller. This is easy to setup. Now the flow will be proportional to the measured level but the level can change quickly so the trick is to use a low pass filter. The measured level is run through the low pass filter to get a filtered level. Use the filtered level for your proportional only controller.

Adjust the time constant to get the desired results. If you chose a time constant that is too long the actual level will lag the filtered level by too much and there may be an over flow so here is the next trick.
Make the time constant for the low pass filter shorter as the tank level increases or gets past a certain point but make the low pass filter longer while the tank levels are low.

About time constants. Time "constants" are not always constant. The instantaneous time constant is something divided by the rate of change in something. In your case the time constant is the level divided by the rate of change in level or volume divided by the rate of change in volume.

This method will not control the level to any particular level but it will keep the level within the proportional bands if the in flow doesn't exceed the out flow for too long or the in flow doesn't increase too rapidly relative to the time constant but the time constant should be short using my definition of a time "constant".


Peter Nachtwey
Delta Computer Systems

http://www.deltamotion.com

http://forum.deltamotion.com/

 

[martinrelayer]

thank you very much for your post.

turbidity depends on flow rate as you can see in graph but also need to take care on flow changes because too much change "makes peaks" in turbidity.

the trick you say is very interesting, i will try make modificatons.

regards,
martin.

 

[martinrelayer]

question: 1st order filter or moving average??

 

[martinrelayer]

i'll tell why, in process, in the middle between pump and tank we have the treatment plant, including sand filters that need a backwash from time to time, so that backwash water is stored in a tank and slowly pumped to the imput....

 

[djs]

Perhaps a slightly different control scheme would be helpful. Consider regulating the flow as an inner loop. The flow would remain constant across most of the tank level. However when the tank level reaches an extreme, the flow can be shifted up or down to keep the tank level within bounds. Under most conditions the flow into the tank will remain constant the average of the outgoing flow.

 

[PNachtwey]

question: 1st order filter or moving average??

I would use a 1 order low pass filter, IIR. It is easy to change the coefficients of the filter as a function of level. If you use a moving average filter, FIR, you will need to change the length of the array to get the same effect. That could be a problem if you need to expand the length of the array. What data to you put there?

Also, is there some optimal outflow where the ratio of flow to turbidity is the highest or is it alway proportional? This is another place to optimize things. Do the level of the tank affect the turbidity? If the same amount of stuff is mixed in a shallow tank the concentration of stuff is greater. This may mean the keeping the water level up a little keeps the turbidity down. Just thinking. I would be looking at your data with an excel spread sheet getting answers to the questions. If the flow is constant will the turbidity change as the level changes?
Does the inflow affect the turbidity. If the in-flow affects the turbidity more than the out-flow there are other "weighting" factors for optimization.

@djs, what you suggested is what I think other people do now based on previous threads. I think your plan would work well IF the flow rate is ramped up slowly. From what I have seen the flow is ramped in steps. The key would be to ramp the set point slowly but that is what the low pass filter is doing. I suggest using a P only system because that is all that is really required and a proportional band is easy to implement. The trick is ramping the set point in an optimal way.

Peter Nachtwey
Delta Computer Systems

http://www.deltamotion.com

http://forum.deltamotion.com/

 

[martinrelayer]

Peter,

yes the lenght of the array I cannot change in this DCS...
for the flow, I guess changing the flow value is more important than how much flow is in. Operators always say "speeding up" is worse than anything.
well i have something to work on, lets see what happens.

regards,
martin.