In the previous Loop Signature article it was shown how feedforward operates. It will be necessary to generate models of the process and of the load transfer functions in order to tune the feedforward compensator. As will be shown in the next article, feedforward control tuning is not nearly as critical as feedback tuning, and fairly simple models are usually fine for the purpose in hand.
It is relatively easy to obtain such models. Figure 1 shows the connections needed for recording the tests that must be performed to do this. One needs to record the signals coming from:
• LD, the load variable, which is the process fluid flow passing through the heat exchanger as measured by the flow transmitter. Feedforward can only be employed if it is possible to measure the load variable. To generate the models none must be able to generate a step change in the LD.
• PD, the temperature feedback controller output.
• PV, the process variable, which is the output of the temperature transmitter.
Once the connections have been made to the recorder, the feedback controller must be placed in manual, and the feedforward (FFW) must be switched off, or frozen. An easy way of accomplishing this is to switch the FFW gain to zero, but be careful not to bump the process.
Tests are then performed. Firstly, a step change is made on the PD, and the PV is allowed to stabilise. The LD is then stepped, and once again, one must wait for the PV to stabilise. Figure 2 shows an example of such tests. These tests must be performed in the area where the process is normally operated.
One must then make simple models of the first order lag and deadtime type. The first model will be the PV vs the LD, and the second one the PV vs the PD.
It may be queried why only single steps (up and down) were made when doing the tests, as in the Loop Signature articles in the previous series, it was stressed that on a self-regulating process, at least 2 steps up, 3 down, and one back should be made. In this series however, it will be assumed that the practitioner will have done all the basic troubleshooting and analysis as detailed in the previous series, so that elementary loop problems are ironed out before starting on more advanced practices like feedforward. This is essential.
One exception to a single step change would be if the system is nonlinear, and if the load changes are going to occur over a very wide range. It may then be necessary to establish models at different points, as various feedforward tunings may be required as the load varies over the range.
Once the models have been obtained, the feedforward tuning can be calculated as per the formula derived in the last article, which was:
The values to be inserted in the feedforward module will be:
• PGL/PGP is the gain of FFW
• The FFW’s lead is set equal to TCP
• The FFW’s lag is set equal to TCL
• (DTL – DTP) is the value to which the delay is set to in the FFW module
Two points should be noted very carefully:
• If DTP > DTL (which would result in a negative delay), then set the FFW delay to zero. You cannot have negative time.
• It is very important that the FFW’s action is set correctly to drive the valve in the right direction on a load change. If it is set incorrectly, the valve will move in the wrong direction, and then the feedback control will have to also correct for this, which means that the system will be worse with feedforward than without.
One of the more difficult things to understand when implementing a feedforward system is how to practically set up the adder where the feedback and feedforward signals are combined before the valve, as seen in Figure 1. Essentially it must be realised that we have two completely separate control systems, one being the feedforward, and the other, the feedback. Both of these will be simultaneously operating the same valve.
The problem with this is very practical. We have two process demand (PD) signals each with a range of 0 to 100%, and the valve itself can only move 0 to 100%. Therefore, if both PDs were at 100%, and we used a straight adder, then its output would be 200%, and the valve can only move to 100%. How can we deal with this?
The best method of combining the two PD signals is given in Figure 3. This avoids inserting anything additional in the loop that can multiply or divide, which would change the loop gain and hence affect the control responses to both of the controllers.
It should be noted that in general, one should be careful of using multipliers and dividers in control loops as it can result in non-linearity. For example, in ratio control, people sometimes divide the one PV by the other to get a ratio, and then feed this signal to the controller. This causes non-linearity. It is better to put the ratio signal on the setpoint of the controller and not inside the loop.
Going back to Figure 3, one sees that a third input, which is an adjustable bias, has been placed on the input to the adder. To try and understand it, imagine a perfect world where the feedforward was so perfectly tuned that it could absolutely cancel out the effects of load changes. If there were no other types of load changes occurring from anything else in the whole system, and if the valve operated perfectly and was completely linear, then there would be no need at all for a feedback control. The feedforward would correct perfectly every time as the load changed. However, a bias would still be necessary to be able to get the process to any particular setpoint.
In the real world of control, things are far from perfect. Valves, as we know, have all sort of problems associated with them. The feedforward models will never in reality be completely correct, however, simple the process dynamics, and in fact, most process dynamics in systems where you need to use feedforward are likely to be more complex than the simple ones we are using here. Models seldom remain constant for all sorts of reasons, and systems are seldom completely linear. This means that the feedforward will never in the real world give perfect cancellation. There are all sorts of things that may also cause other load disturbances. As in this case of the heat exchanger, there may be changes in the steam pressure and in the temperature of the process fluid.
Therefore, as mentioned in the previous article, feedforward control cannot be used by itself, and is in fact always used in conjunction with feedback control. One should think of the feedforward as being the master control system, and the feedback as being a trim.
Returning to setting up the bias, one would try and operate at mid-range load conditions. At such a condition it would be nice to have the feedback controller’s output at about 50% as to allow it to apply trim adjustments equally in both directions. Therefore, one would adjust the bias to get the feedback controller’s output to about 50%.
The final control system is shown in Figure 4. One change in it is the addition of a secondary cascade steam flow controller. This is to ensure that all valve problems are eliminated from the control schemes, and that the control valve will in fact deliver the steam flow as demanded by both the feedback and feedforward controllers. The reasons for this are given in Loop Signature 17 in the above referenced CD. It is mandatory to use a cascade flow loop in sophisticated and important control systems that have slow process dynamics.
One point to note is that the feedforward must be added to the output of the feedback controller so the adder’s output goes to the setpoint of the cascade secondary flow controller. I have on occasion seen people adding it onto the output of the secondary controller. This is wrong. It is equally important that the feedforward control’s output also feeds a ‘perfect’ valve.
The next article in this series will show some tests done on feedforward systems, and will illustrate how well feedforward can work.
About Michael Brown
Michael Brown is a retired expert in control loop optimisation, possessing over five decades of experience in process control instrumentation. His primary focus has been consulting and delivering instruction on practical control loop analysis and optimisation. He has conducted training and optimised control systems at numerous facilities internationally. Mr. Brown continues to author articles reflecting his extensive work, and his courses remain accessible for purchase in PDF format. Additionally, he manages the distribution of Protuner Loop Optimisation software and welcomes inquiries regarding loop-related issues.
Contact details: Michael Brown Control Engineering CC,
| Email: | michael.brown@mweb.co.za |
| www: | www.controlloop.co.za |
| Articles: | More information and articles about Michael Brown Control Engineering |
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version
