Guidelines for Choosing feedforward Control in Industrial Applications
feed forward control is best deployed in control systems design applications where the process or controlled variable behavior is well understood. A well understood process is one in which the first principal equations describing the process are known for the given application, and are correctly applied to adequately describe the process. It is also very useful in designs where the process is not understood at all, but the behavior of the process can be measured and experience has shown that it is replicable under known operating conditions. Usually a mix of both approaches is used. Following are some illustrative examples:
Example 1) feedforward speed regulation
The well understood process:
The inertia of a rotating mass can be computed (or measured: see the section II of this paper,). We know that the torque required to accelerate that inertia is equal to the product of inertia and the acceleration rate: (t = I.a). If we assume a frictionless system, and that the inertia and the desired acceleration rate is known, the torque required to accelerate the mass to a desired speed without closing a speed loop can be computed. Why is this? We know that acceleration is the derivative of speed. This implies that the derivative of a clean speed reference will yield the acceleration rate. That rate can be multiplied by the known inertia to yield the torque required to accelerate the inertia such that the rotational speed tracks the speed reference (known as inertia compensation ). The system will track the speed reference without closing a speed loop. This is considered an open loop speed controller, and it works based on the computed feedforward torque reference.
The notsowell understood process
Unfortunately the real world is full of nonlinearities that limit the effectiveness of such feed forward control schemes. In the above example, if we do not add some extra torque (or, over time, energy) to compensate for the nonlinear friction and windage losses, these losses will consume the energy that was intended to be used for accelerating the mass, and result in a difference between the actual speed and the speed reference. By using a feedforward loss compensation signal for friction losses, this can be overcome. A simple way to accomplish this is to set the speed reference of a temporarily closed speed loop controller to a set of specified values, and tabulate the timeaveraged value of the torque reference at each speed. This lookup table (known as friction compensation ) can then be used with an input pointer of speed reference, and an output of feedforward friction and windage loss torque to effectively compensate for these energy losses.
If you have setup the two above feedforward signals perfectly, and the friction losses are predictable and non timevarying, the actual speed will match the speed reference. However real world implementations cannot exactly compensate for time varying friction losses, and the actual speed will vary with the time varying friction and windage losses. Friction and windage in a given system will vary with temperature, humidity, bearinggrease age etc. Usually the above compensation terms are utilized to assist a closed speed loop by maintaining the speed "in the ballpark" and thereby minimizing the output of the loop PI regulator. This is very desirable in some applications, because the integrator (which is an energy storage device) in the loop is maintained with a minimal storage value, enabling it to be reset with a minimum disturbance to the system. See this paper for more insights into the weaknesses of feedforward control.
Example 2) Feedforward tension control
The well understood process:
The tension on the outflow side of a driven roll (or bridle) in a properly configured section of web (paper and converting industries) or strip (metals industries) is known to be: T_{OUT} = T_{IN} – t_{M}.GR/R. Or in other words: The tension on the OUTPUT side of a driven roll is equal to the tension on the INPUT side minus the motor shaft torque that is reflected to the driven roll surface. If the motor is "motoring" it will lower the tension on the output side, if it is "regenerating" or absorbing power, it will increase the tension on the output side. Here again, if we know (from measurement for example) what the tension on the input side of the driven roll is, and we know what tension we want on the output side, we can compute the torque we need to generate the desired tension control.
The notsowell understood process
Friction and windage losses will consume energy in tension control designs as well (see this paper). They will consume energy that was supposed to be used for holding tension. The result of this is that the actual tension will differ from the tension reference. This can be overcome by implementing the same feedforward friction/windage loss compensation signal described in Example 1 above. However, note that if there is no closed speed loop, the feedforward terms described above will be required to compensate for the energy needed to accelerate the roll inertia during line speed changes. If this is not done the system will not maintain tension and the web/strip may "bottom out" when discharging from winders/unwinders, or web breaks may occur during changes in line speed.
Example 3) feedforward speed regulation with pressure.
The well understood process:
Assume that a process is devised where, in order to accelerate a mass, air must be blown into a saillike device. We know that F = P.A (Force = pressure times area) and F = m.A (Force = mass times acceleration). If the effective surface area, the mass, and the desired acceleration rate are know, the Pressure needed to meet that desired acceleration rate can be computed. In addition, if the relationship between the pressure and the fan torque, and speed are known, the same techniques described above can be implemented to predict the fan motor torque that is required to meet the desired application acceleration rate. designs similar to these demonstrate the usefullness of feed forward controls techniques. They can be deployed to solve many difficult controls problems. See this report for an example.
The notsowell understood process
In these applications the design engineer must also develop feedforward compensation schemes to compensate for the energy losses associated with the movement of air in pipes. Since this kind of windage loss term is highly nonlinear and poorly understood it is best to close a speed loop around the feed forward pressure loop, and let the speed controller compute the difference.
Guidelines for Choosing Feedback Control in Industrial Applications
Feedback control is typically used to regulate a variable (or variables) in a control systems design which has time varying disturbances, and or operating parameters. It is also used when the accuracy afforded by feedforward controls is not adequate to meet the application performance specifications.
For example:
A customer installs a DC parallel current controller on an aluminium foil coating machine entrysection unwinder. After startup the customer complains about entry section strip breaks that occur with great regularity when a set of niprolls nips the thin material. He points at the drive and states that "the unwind controller is not regulating tension properly". The onsite service engineer points at the feedback display and states that "we are regulating torque perfectly", and indeed the estimated torque feedback is a flat line. It is clear that it is impossible for the process dependant tension disturbance attributed to the nip rolls biting the web, to be compensated by an open loop tension regulator. Unfortunately, in this application, the problem is not the current controller on the unwinder, but rather the source of the disturbance in the process itself. The parallel current controller by itself will not be able to solve this customer problem because it is an openloop tension regulator.
The big drawback to closed loop feedback control is the need to design the controller to be stable. Closed loop stability analysis is not easily understood. Most problems with closed loop controls encountered in the field are due to this lack of understanding. Often the individual who is tuning the loop is using guesswork based on past experiences. This may help them out for a particular setup and operating parameter set, but as soon as the operating parameters vary beyond the point for which the loop was tuned, the loop goes unstable. If the loop cannot be stabilized by the individual at startup, they almost always look for a feedforward solution. This is not a very sensible way to solve such problems, often the solution requires a simple adjustment, or an additional "pole" or "zero" scheduled at the correct frequency.
Some variables, due to the nature of the physics involved, will always have to be regulated with a closed loop. Speed, position, and temperature are good examples. Following are some examples of problems that can occur if you don’t do "due diligence" when closing these types of loops.
Example 1. Speed with feedback Control.
The bandwidth (see the introduction here) of a speed loop varies linearly with inertia, the bigger the inertia, the lower the bandwidth for a given fixed gain. If the application reflected inertia varies a from very small value to a very large value (for example: a paper machine winder) and the controls engineer implements a fixed gain regulator, the loop may be stable at core, but wildly unstable at full roll. Or the loop may be stable at full roll, but the drive gets very "noisy" and chatters and growls at core. In these applications, where the change in inertia is significant, the controls engineer will always have to design the speed controller to schedule the gain of the loop as a function of the roll inertia (which can be computed based on a known density, a measured diameter, and a roll width). In essence, the gain is increased proportional to the increase in roll inertia, or until a gain limit is reached (all systems are gain limited due to noise, read Section II of this paper for more information about gain limitations due to system noise, and torque jitter). When a proportional gain limit is reached the controller must be designed to decrease the integral gain as inertia increases in an inversely proportional manner, this will keep the loop stable over the entire roll buildup.
Example 2. Position with Feedback Control.
If a PI (proportional plus integral) position loop is closed around a torque loop, without a closed speed loop between the position and torque loops. This loop has two integrations in the forward path (from torque to speed, and then from speed to position), and then a third one from the PI regulator. That equals 270 [deg] of phase shift from DC to the first lead frequency. With only one "zero" (+90 [deg]); in the PI regulator, their will always be 180 [deg] of phase shift at the loop crossover frequency. A crossover that is 180 [deg] phase shifted, nomatter what the gain value is, will be unstable at or about the crossover frequency. If a derivative term (i.e. a PID regulator) is used to close the loop, the extra "zero" helps by adding phase margin (+90 [deg]). However for most untrained field service engineers, PID regulators are difficult to tune. A better approach is to close the position PI loop around a closed speed loop. The speed loop acts like a lowpass filter in the outer position loop, and a fixed gain position loop tuning (albeit an optimally "slow" tuning) can be found that will be stable at all inertia values. An even better approach is to use gain and frequency scheduling to guarantee that the controller is optimally "hot" for all inertia values. If you have trouble understanding "poles" and "zeroes". Try taking the first course of our 10 volume/20course Advanced Dynamic Simulation course. Or ask us to help you out with your particular controls problem.
From the above examples, it is clear that closed loop controller design can be difficult. It's always best to take the approach of understanding (i.e. modeling) the problem through analysis, and designing a controller based on that understanding. If you have controls problems that you have trouble comprehending, we would be glad to help you. Drop us an email and we'll get back to you.
When to use feed forward, when to use Feedback?
If you read the above two sections on feedforward, and feedback control, you should have a good idea of when to use the two approaches. Here are some basic guidelines that cover 99% of the designs found in industrial controls applications:
1) Use ONLY feedforward control if:
 The physics/chemistry of the application are well known
 You can easily measure or calculate the variables in the equations
 There are no significant process disturbances
 The accuracy of the measurements used is an order of magnitude better than the application specifications
2) Use ONLY feedback control if:
 Significant unmodeled process disturbances exist
 You cannot measure or calculate variables in the equations describing the physics/chemistry of the application.
 The accuracy of the measurements used is NOT an order of magnitude better than the application specifications.
3) Use both when:
 The physics/chemistry of the application are well known
 You can easily measure or calculate the variables in the equations
 The accuracy of the measurements used is on the order of the application specifications
 You want to prevent saturation of the controller integrators
 You want to improve trajectory tracking, but 2^{nd} and higher order dynamics in the plant limit your controller bandwidth
If you have a closed or open loop controls problem that you cannot solve, feel free to contact us at ApICS LLC. We would be glad to be part of the solution to your problem.
