I already have a diagram where I can show the process, but i do not know how to attach it.

I have this problem since I started this job, we had problems of water, then reactives, and so on.

We add milk of lime to the SAG mill (automatic valve). The discharge goes to a common sump for BALL and SAg mills discharge. We add to this sump ZnSO4. A pump sends the fluid to the cyclone, overflow goes to the sump2, underflow goes to the Ball mill( to the ball mill we add NaCN). From sump 2 we send the fluid to the unit cell where we add MIBC, then the fluid goes to the conditioner where we add CYTEX, MIBC, NaCN, Xanthate, and at this point we do the measurement of pH.

I hope you will understand my problem, it is a non-linear loop (control) because i have dead time between the automatic valve and the measurement point. the equipment is new, because recently I changed the pH electrodes.

I tried to read the pH in the sump 1, but it did not work because after that the pH is changing.

I think that may be a cascade control would fix my problem but i am not sure. I need some help in this, so if you could help me.

Or may be modelating (maths) the process ( really difficult i think) and working with the model.

Thanks.

 

What causes the pH shift? Is it a natural consequence of the process or is it due to changes in the feedstock?

If you can measure or estimate the parameters that will affect the PV, and thus the CV, you may be able to "feed forward" and be in the right neighborhood before you can measure the resulting change in the PV.

Alternatively, if the changes are random (say induced by semi-random variation in the feedstock) and it's not possible to measure the changes as it's introduced to the mill, then you may have to add several points of measurement along the entire liquid stream. May not be a bad idea anyway. That way you get a better picture of the overall state of this particular process (pH maintenance from front to back, not the grinding process per se).

 

http://www.expertune.com/ArticlesFull.html#PIDOpt

Start here. There's a bunch of stuff about PID with large deadtimes and pH control loop linearisation.

Cascade Loops don't directly help dead time dominant processes. Feed forward control only works if you have an absolutely defined and repeatable model - ie prediction - of the process you are trying to control. So the most important "feedforward" should be that you add chemicals at a rate ( ie weight of chemical per weight of total plant thruput) not liters per minute so that you have constant chemical addition rates.

You should ALWAYS try to keep the control strategy as simple as possible and get the equipment and loop tuning right before trying to implement a more complex control strategy.

If your deadtime is too large for a PID loop to deal with then you may require a deadtime compensator like a Dahlin controller or Smith predictor. For these controllers to work you will need a repeatable and predictable dead time followed by a first order response.

If your process is reasonably stable but not very predictable ( and since its pH then it isn't very predictable ) then the simplest solution may be a sample and hold controller. Basically this is an Integral only controller that executes once each time the deadtime has passed and the process settled.

Rob
www[.]lymac.co.nz

 

Hello and thanks for you help.

I would like to explain in detail the process and the disturbances that I have. Actually I have a drawing where I show the reagents, process water, feed, and so on. Al this parameters affect directly to my control loop.

Actually as you know it is really difficult to measure the disturbances that affect the PV. I can not make a measurement of the feed because it is always changing and working.

I could install another pH probe in cyclones (close point) and stand the pH probe in the conditioner.

I do not have instruments enough to get values of reagents, but I could get manually the "gr/t" of each reagent that actually is affecting the process. The process water is another problem because is changes from ph 11 to ph 7 depending of the source.

I thought that cascade control may be solve my problem by installing a pH probe in cyclones and the other in the conditioner, but i am no sure due to the non linearity (dead time).

As far as I know feed forward control is good dealing with disturbances, but I have to know the model and have measurable disturbances, but I do not. What do you think if I can use a lead-lag into the feed forward or may be use a combination of feedforward and feedback control?

I would like to solve this problem working with the equipment that I have at hand (pH probes and transmitters, ball valve (with positioner), and the ControlLogix processor (1756-L63 Rev. 16.21). The ControlLogix does cascade control, has instructions like lead-lag, deadtime, and feedforward into the PIDE. If it is not possible to solve the problem with this processor, I would find other options.

To be honest I would like that you understand my problem and based on your experience, may be suggest the best way to solve this problem. As you said as simple as it can be. From this point i would work thinking in solve the problem.

best regards.

 

try turning off the integral and derivative functions first, then adjust your proportional band to get stead valve position, gradually reduce the proportional band until the valve starts moving in a cyclical fashion, once you get things stable, turn on the integral function and adjust to you find a stable point.

with such long dead times, integral may not work at all and needs to be left off.

The dead time compensator algorithms work best in servo where you know all the variables, but in your case you have to many wild variables and the dead time compensation is no better that proportional. Do not use derivative at all, it wont be stable.

 

Hmm, you should probably have a better idea of the mill throughput than it sounds like you have. The last SAg I worked on had a really bad throughput control strategy that meant the whole system was ridiculously unstable with a 30 or 40 minute cycle. Be sure you don't have the same issue. Also, you should really try to get some control over the pH of the feed water as this is going to be a big source of disturbances for you.

Regarding turning off Integral and trying to tune the loop to oscillate. With a dead time of 15 minutes, the loop is very prone to oscillation when the disturbance cycle is around 15 minutes so using a Ziegler Nichols type tuning method here is going to be extremely difficult. You're also going to need Integral to correct for the non-linearities in the pH control, so a P only controller will only work for you if you want stability, have an already reasonably stable system and don't care about a constant error.

If you want you can contact us off-forum at the web address below

Rob
www[.]lymac.co.nz