After I got my BSc in aeronautical engineering, I had a choice of going into structures, aerodynamics or control. I chose control because it was closest to flying aircraft, and because it was known to be the hardest subject. I also knew it had a ruthless and domineering professor, but I thought I could probably handle that. I came to regret that decision.
But there it is, control engineering is my chosen profession. I think I never really got the hang of it, and I certainly never learned to like it. But maybe that has given me a somewhat "off" perspective on control that may be worth listening to.
Control means taking a system to a target, either in the literal sense in warfare, or in the figurative sense of taking a system to a new setpoint and keeping it there. I will mainly discuss flying an airplane (or pointing a spacecraft) somewhere, and stopping and moving a haptic device.
The obvious way to steer a system to a new place is to adjust some input, like a glorified steering wheel, and wait for the result. In the simplest case, that's it. You know approximately what the result will be, and that's good enough for you. This method of control is called feedforward. If you know enough about your system, this may be all you need. If you want to sell five bags of rice, you put five bags on the counter. No need to weigh them.
The outcome of your adjustment may not always satisfy you. The result may be off, because :
- you do not know your system well enough.
This is called the "modelling error".
- there may be unknown external inputs, like a gust of wind.
These are called "external disturbances".
If you can measure these disturbances,
they can be corrected by feedforward.
If you can't, these are unmodelled inputs.
To remedy this you can measure the outcome and correct it by an extra input. This is called feedback.
Feedback can reduce the error, but it introduces two new problems :
- the outcome is always delayed a bit with respect to the input.
The delay may come from the system, or from the measurement, or both.
- the measurement may be imperfect in other ways.
It may have a fixed offset, or it may fluctuate randomly.
If your feedback is wrong or out of sync, then the outcome will be wrong, or even unstable.
TBW
TBW