Scaling for Feedforward-Feedback Control


Thread Starter


Reference Bela Liptaks PROCESS CONTROL, chapter titled "Feedforward & Feedback Control", there is an illustration of this control scheme applied to a heat exchanger. However, there is no specific discussion in respect of the SCALING required for the measured variables. In this chapter, there is an illustration of this control scheme applied to a heat exchanger. However, there is no specific discussion in respect of the SCALING required for the measured variables. I wish to implement this scheme on a PLC, but it will be useless unless the scaling of all variables is done correctly. I would be grateful to hear from someone that could advise me regarding this. Many thanks.

Bruce Durdle

Liam, The scaling of transmitters etc will be done according to the normal rules - in other words, try to keep the maximum expected value of the measured parameter below about 90% of range, and watch that you don't get too low for the minimum controlled turndown value. You then need to sort out the feedforward equation in engineering units. This will be something like: Steam flow = Liquid flow x Change in liquid temp required x Liquid specific heat/(Steam flow x change in steam enthalpy) (Note that the Change in liquid temp required is found from (Outlet liquid temperature setpoint - inlet liquid measured temp) The first step is to check the actual values you will be reading at the relevant inputs. (For Modicon Quantum PLCs, a 4-20 mA range can come across as a register value of 6400-32000, for instance: you should convert this to say 0-10000) You now have to get the various scales lined up. For the temperature difference, you need to make sure either that both inlet and outlet temperatures are to the same scale, or convert them to the same scale - eg, if inlet is 0-50 degC and outlet 0-150 degC, the effective inlet signal as a % of range must be divided by 3 to get it into a 0-150 range as well. If you are measuring steam enthalpy, or calculating it from pressure/temperature, you will need to consider that as well. The multiply operations do not give the same problems. However, the divide operation can give trouble - but different PLCs handle it differently, so you need to check it all out. The maximum value of flow calculated as setpoint to the flow controller must then be scaled so that it corresponds to the measured range from the steam flow transmitter. This is a quick summary - if you want more detail, I can find some worked examples from my collection of notes and send it to you direct. Cheers, Bruce
I like the remark from Bruce :...from my collection of notes. The paradoxe in Instrumentation is this: Process Control and Instrumentation requires more maths, physics and chemestry ...than any other discipline. Is there a global tool package ? not that I know. How much can be found in the Web: so little is almost nothing. Most black box hardware/software contain nearly nothing. Through 30 years carriere as consultant and or employe I never used the ISA "Bible". Never used Liptack. The most supporting info were from Miller and Considine. And from some (I say some because there are few) suppliers. I suggested the latest Liptack edition to include a movable Appendix binder (something like the french collection "Thechniques de l'Ingénieur) So that every three months (for instance) material could be added/renewed. This appendix should be based on the design calculations and tutorial. It would be soon a large size volume, never outdated. My suggestion had no echo.I don't mind, I'm retired from the profession, not without desolation. My collection is sizable, not in inches but in feet !!!

Bruce Durdle

The issue of scaling of feed-forward calculations is one of the many "implementation" issues that are not generally discussed. The academically-trained get to know all about the high-level techniques, and will give a block diagram using all the correct symbols - but when it comes to actually making something that will actually do the job in the real world, are usually not to be seen. The poor old technician has to somehow figure out how to put the abstract ideas into reality. (An engineering technician of my acquaintance said that professional engineers had a lot of bright ideas - the technicians had ideas that could be built. As a professional engineer myself, I find myself agreeing with him.) So if I pick up a Moore SLDC352 controller - I know, that's going back a few years, but so am I - it has a "summer" that does: OUT = GX * (X + BX) + GY * (Y + BY) + GZ * (Z + BZ) Another similar item may do it using: OUT = G0 * (G1 * (X + BX) + G2 * (Y + BY) + G3 * (Z + BZ)) + B0 and it is quite an exercise to actually work out suitable values for the various configurable parameters. Add in a couple of stages where the intermediate result is bounded to less than 100% and it becomes a major task. Yet this is quite conveniently glossed over in most of the standard texts - Liptak and Considine included. Start to play with multiplication and especially division and it gets even worse. So I guess I am issuing a challenge to all academics (especially those at the professional levels) on the List - don't just cover the abstract academic concepts - spend some time looking at what is involved in actually putting those ideas into practice in actually available operating systems. And don't forget the poor old panel operator, who has to somehow start up and shut down a plant with a control scheme that is operable only above 70% load! Cheers, Bruce. (PS - I have put my money where my mouth is, and have covered this particular topic in the set of notes I referred to).

Arnold Dillon

Scaling is always an interesting problem, and Feedforward presents a couple of special challenges. Normalizing all variables from 0 to 100% can sometimes help as long as you keep in mind that 75% on one measurement may be more or less of the other process measurement. --However, one thing that I haven't seen discussed here is the fact that a lot of Feedforward loops act on the output from the control algorithm to the final control element. i.e., The Feedforward directly affects the control valve output. This means that you have to take into consideration not so much the scaling of all of the measured variables, but you have to know what the flow profile of the control valve looks like. Is the valve linear, equal percentage, etc.? The Feedforward portion of this type of control is basically open loop, with the feedback from the affected measured variable acting to trim the final output to get to a final desired value. In this case ask yourself how a five percent (or some convenient number) change from your Feedforward signal to the control valve will change the process variable that you want to control. You may need some field data to determine this and once you have that data you may need to apply a multiplier to the Feedforward signal to get something close to the desired effect. Hope this helps. Arnold Dillon I.C. Thomasson Associates