TEXAS TORQUE |
|
In PID, we learned about a feedback controller. Feedback in this case refers to how the control system responds to error. This works most of the time, however, this is reactive, not proactive. For a PID to work, your system must have error to begin with; which isn’t ideal.
In contrast, a feed forward controller uses model information to instantly set the approximate desired state. This reduces error time, and it can help minimize oscillations by reducing the PID’s importance in the system.
A normal feed forward controller takes in two to three values for the model: $k_s$ for static application, $k_v$ for velocity application, and (optionally for the most part) $k_a$ for acceleration application.
It is important for those implementing control systems to have a healthy understanding of physics. As such, this will not go in depth for the reason.
In a nutshell, however, two connecting objects that contain a normal force exert a frictional force. There are two types of frictional forces, static and kinetic. Each type has its own coefficient of friction, μ. Static friction exists at which point kinetic friction takes over. This is due to intermolecular forces soft-binding objects at the subatomic level.
For many applications of PID, you may have witnessed the error term existing but nothing happening. This is due to static friction: despite putting in energy, the static friction is preventing movement. The $k_s$ term overcomes this by always keeping the model near or at the slipping point to provide fast changes in velocity.
Feed forward values should not be calculated emperically. Proper characterization is needed. Additionally, dimensional analysis must be used to insure proper unit conversion, as output values are unit dependent.