Closing the Loop: Feedback, Feedforward, and Adaptive Control
Process automation relies on precise control systems to operate production equipment for everything from lumber to pastries. Using control algorithms produces stable, accurate, and often delicious results.
Controlling Automated Systems
Process control is the label used to describe a number of techniques used to manipulate a system using outputs from a controller. For example, if you want to heat water to a specific temperature and then hold that temperature perfectly steady, you will need a variable heater (the output), a temperature sensor (the feedback), and a way to control the heater (control system) so that the temperature doesn’t go over the set point.
This style of control system is referred to as a closed-loop system and is the basic structure for feedback, feedforward, and adaptive control systems.
Figure 1. Process control models are used for nearly every kind of production industry. Image used courtesy of Unsplash
Feedback control is an extremely common method for driving a system, usually consisting of an input data point or setpoint, an output, a process, sensors, and a controller. With these components, we create a closed-loop system that senses the effects of the output from the controller. An algorithm is used to control the output based on the input and the sensors.
Figure 2. Feedback control model.
If we use the above example, controlling the water temperature is the process, the heater is the output, a temperature sensor in the vessel would be the sensor, we could use a PLC as our controller, and the setpoint would be the input.
The goal is for the water temperature to reach the set point without overshooting. As the control system changes the output, perhaps in this case, the change could be the percentage of power to the heater, the sensor reports back to the control system any change in the system. As the temperature of the water reaches the set point, the algorithm in the control system will adjust the output. Essentially, the feedback controller compares the set point with the actual temperature and tries to reduce the difference.
The actual algorithms used for this control (variations of PID controllers) are outside the scope of this article.
Feedback control is a great way to control a system, and therefore has been successfully used for many years, but some systems experience disturbances that could affect the feedback control system. The disturbances for this water heater system might include such things as the water level being reduced, or maybe the water contains a reactive material that is increasing the temperature along with a heater.
Figure 3. Feedback control model with feedforward.
A disturbance affects the system in a negative way, but if these disturbances are known in advance, a feedforward control system can assist the feedback control system to reject or compensate for the disturbances before they happen, reducing the chance for overshoot and corrective action by the feedback controller.
Another way to compare the differences is that a feedback system controls the output based on the measured effect, but disturbances can place large negative impacts on the process. A feedforward system controls outputs based on all of the known disturbances before they can affect the process, but it cannot adjust the output mid-process based on measured feedback if the disturbances are different than normal.
Because of these benefits and drawbacks of both systems, a feedforward controller is typically used in conjunction with a feedback controller. By using both models, you will almost certainly increase the accuracy of your control system.
Whether you are using purely feedback or combined feedback/feedforward control systems, there will be parameters that need to be tuned or adjusted to create the type of system you require. Some control systems need to react very quickly while others react very slowly. By adjusting parameters, you can tune exactly how the control system will respond to changes in the process from the output signal or from external disturbances.
Figure 4. Feedback control model with both feedforward and adaptive control.
With a new system, the parameters are tuned by engineers and technicians. As the system ages or as conditions around the process change (changing seasonal temperatures, batch mixes, etc) the process may not respond in the same way, so the parameters might need to be adjusted again.
With adaptive control, the system is monitored and automatically adjusts its own parameters so that the feedback control system will run as expected, even if the system degrades over time. The adaptive controller is given a reference model to compare the in-control process to and a set of parameters that can be adjusted. The process output is monitored and the parameters are adjusted to keep the process operating at optimal control.
In some circles, this ability to monitor conditions over long periods of time and adapt to changes is the basis for so-called ‘artificial intelligence’ (AI) that drives some modern systems.
Combined Control System Models
In most systems, a simple feedback controller will work quite well, but in cases where you want greater accuracy or where your system could degrade or change dramatically over time, you might want to consider combining all three types of controls.
Figure 5. CNC machines must rely on extremely precise, high-speed control models. Image used courtesy of Adobe Stock
As a final example, a computer numerically controlled (CNC) machining center uses highly accurate ball screws and servo motors to spin cutting tools at very high speed with very high precision. These machines can make use of all three control systems to accurately cut metal.
The servo motors have encoders that report the motor position and speed, and the control system uses that information along with the pitch of the lead screw to determine how far the cutting tool has moved; an example of feedback control.
Sensors in the spindle inform the control system of increased vibration, data which is used to determine if cutting tools are wearing down, requiring a reduction in cutting speed or a tool offset. These sensors could produce both feedforward and adaptive control ability because they are measuring disturbances that affect the process over time.
By combining all three types of control, you can build a control system that is not only highly accurate but able to maintain its optimal stability over time and with variations affecting the process.