I've done a short bit on PID Control Loops and Gas Flow Control at:
http://iecfusiontech.blogspot.com/2007/ ... alves.html
Criticism, advice, or unwarranted disdain welcome.
PID Control Loops
-
jlumartinez
- Posts: 143
- Joined: Thu Jun 28, 2007 7:29 pm
- Location: Spain
The classical equation of PID is almost useless nowadays in the chemical industry. The error in the derivative term means a bump in the valve opening in every setpoint change (because in that moment the derivative of the error becomes infinity). For that reason is used a derivative term over the process variable (PV) instead of the error.
Also the error computed in the proportional term is less and less used now. A PID with proportional over the error gives a strong movement in the valve in every setpoint changes (proportioanal kick). In the process industry where 99.9% of the time the setpoint is not changed is better to tune the PID to reject disturbances instead to follow the setpoint. And with a PID with the proportional on the process variable (PV) you can get a more robust tuning, avoiding instabilities generated in setpoint changes for using the error in the P term.
Cohen-Coon is a very good method but is too agressive. It creates a 25% overshoot in the response. Since we are interested in obtaining a robust, smooth and no-overshoot response it is good practice to reduce the proportional gain by 25% and increase the integral time by 25% . It gives a nice response
I recommend you to see http://bestune.50megs.com/typeABC.htm where you can find the industrial equations of the PID controller.
Also the error computed in the proportional term is less and less used now. A PID with proportional over the error gives a strong movement in the valve in every setpoint changes (proportioanal kick). In the process industry where 99.9% of the time the setpoint is not changed is better to tune the PID to reject disturbances instead to follow the setpoint. And with a PID with the proportional on the process variable (PV) you can get a more robust tuning, avoiding instabilities generated in setpoint changes for using the error in the P term.
Cohen-Coon is a very good method but is too agressive. It creates a 25% overshoot in the response. Since we are interested in obtaining a robust, smooth and no-overshoot response it is good practice to reduce the proportional gain by 25% and increase the integral time by 25% . It gives a nice response
I recommend you to see http://bestune.50megs.com/typeABC.htm where you can find the industrial equations of the PID controller.
All excellent points.jlumartinez wrote:The classical equation of PID is almost useless nowadays in the chemical industry. The error in the derivative term means a bump in the valve opening in every setpoint change (because in that moment the derivative of the error becomes infinity). For that reason is used a derivative term over the process variable (PV) instead of the error.
Also the error computed in the proportional term is less and less used now. A PID with proportional over the error gives a strong movement in the valve in every setpoint changes (proportioanal kick). In the process industry where 99.9% of the time the setpoint is not changed is better to tune the PID to reject disturbances instead to follow the setpoint. And with a PID with the proportional on the process variable (PV) you can get a more robust tuning, avoiding instabilities generated in setpoint changes for using the error in the P term.
Cohen-Coon is a very good method but is too agressive. It creates a 25% overshoot in the response. Since we are interested in obtaining a robust, smooth and no-overshoot response it is good practice to reduce the proportional gain by 25% and increase the integral time by 25% . It gives a nice response
I recommend you to see http://bestune.50megs.com/typeABC.htm where you can find the industrial equations of the PID controller.
I also like to play tricks with the integral term to reduce windup to zero.
I think you make a very good point about none of this being set in stone. You adjust the filter for the kind of response you want.
My short exposition was to give an overview and give some very simple tuning rules that get you in the ball park.
Since most of the controls that are going to be used will be on stemless valves and the control of currents and voltages, limitations on change are not really necessary. Continuous adjustment is not a significant wear out mechanism.
In many cases you can take advantage of the plants higher frequency response for small signal changes by putting rate limiters on change of control valve position and also limiters of various types on the control algorithm.
That way you can get fast response for small changes and slower response to large changes.
You can also add hysteresis to the valve change (small changes not allowed). Or you can do things like continuous adjustment allowed until within a certain error band and then only minimum steps allowed. To minimize valve wear from hunting.
Doing stuff like this digitally gives lots of options.
That way you can get fast response for small changes and slower response to large changes.
You can also add hysteresis to the valve change (small changes not allowed). Or you can do things like continuous adjustment allowed until within a certain error band and then only minimum steps allowed. To minimize valve wear from hunting.
Doing stuff like this digitally gives lots of options.
Right on track.
Simon, Its good to see your making progress on the controll side of things. After reading the full WB6 technical report its becoming clear exactly what areas the guys at EMC2 will need help with. This is going to be one of them. Control systems aren't my area of expertise. But even if a bullet proof control system in place what do you think of the actual valves/actuators. Is it just me or are we going to need very special customized hardware, or could we get away with existing technology like if we pulse width modulated a bosch fuel injector.
Fuel management in jet aircraft seems like it has a parallel in what we are trying to do here, ill try and dig up some info on FADEC but even in this day in age only some of the higher end Pratt and Whitney and Rolls Royce Turbines have full digital management. Even then alot of the information is proprietary.
Fuel management in jet aircraft seems like it has a parallel in what we are trying to do here, ill try and dig up some info on FADEC but even in this day in age only some of the higher end Pratt and Whitney and Rolls Royce Turbines have full digital management. Even then alot of the information is proprietary.
Purity is Power
-
jlumartinez
- Posts: 143
- Joined: Thu Jun 28, 2007 7:29 pm
- Location: Spain
Basically, for tuning purposes you adjust:
* The proportional gain to give more or least speed to your control depending on the process variable, PV, response to a change in the valve opening (output process signal, OP). For huge responses you need an small gain, for little response you need a high gain to move quicker the valve. It is defined by the steady state: the ratio of change of the PV to a change in the OP. the PID gain is usually correlated to the inverse of the process system gain.
* The integral time is adjusted to match (aprox.) the time constant of the system. So the integral time shuouldn´t be used for getting a faster response. Its meaning is related to the dynamic response of the system.
* The derivative time is used just in the case of having a dead time (always I am using a process model as being a first order plus dead time). The derivative time is used to compensate this dead time, so it is another parameter defined by the dynamic of the system. Usually a good derivative time is half of the dead time (this is the rule applied inside Ziegler- Nichols tuning method). The filter is just used to get a nice and non-noise signal. Getting rid of the noise is fundamental to have a proper derivative of the signal. If you are using derivative time is always good to put a small filter to the PV.
Another good tuning method is called IMC. I prefer Cohen-Coon with relaxed gain and integral time
* The proportional gain to give more or least speed to your control depending on the process variable, PV, response to a change in the valve opening (output process signal, OP). For huge responses you need an small gain, for little response you need a high gain to move quicker the valve. It is defined by the steady state: the ratio of change of the PV to a change in the OP. the PID gain is usually correlated to the inverse of the process system gain.
* The integral time is adjusted to match (aprox.) the time constant of the system. So the integral time shuouldn´t be used for getting a faster response. Its meaning is related to the dynamic response of the system.
* The derivative time is used just in the case of having a dead time (always I am using a process model as being a first order plus dead time). The derivative time is used to compensate this dead time, so it is another parameter defined by the dynamic of the system. Usually a good derivative time is half of the dead time (this is the rule applied inside Ziegler- Nichols tuning method). The filter is just used to get a nice and non-noise signal. Getting rid of the noise is fundamental to have a proper derivative of the signal. If you are using derivative time is always good to put a small filter to the PV.
Another good tuning method is called IMC. I prefer Cohen-Coon with relaxed gain and integral time
Keegan,
I think we are going to need some pretty special controls.
If the reactor is operated at the resonance peak, I see a phase locked loop for that (HV control). One operating at a different frequency for POPS.
I like operating at the the resonance peak because although system gain is lower, power gain is higher. It means lower voltage and power for the power supplies. It may also mean a slightly larger reactor to compensate for the lower system gain.
The gas control stuff is going to have to be pretty custom.
I took a whack at the control valve/orifice problem in
http://iecfusiontech.blogspot.com/2007/ ... izing.html
jl,
Thanks. A welcome addition.
I think we are going to need some pretty special controls.
If the reactor is operated at the resonance peak, I see a phase locked loop for that (HV control). One operating at a different frequency for POPS.
I like operating at the the resonance peak because although system gain is lower, power gain is higher. It means lower voltage and power for the power supplies. It may also mean a slightly larger reactor to compensate for the lower system gain.
The gas control stuff is going to have to be pretty custom.
I took a whack at the control valve/orifice problem in
http://iecfusiontech.blogspot.com/2007/ ... izing.html
jl,
Thanks. A welcome addition.