Feedback in nonlinear systems
Few physical systems are truly linear over broad ranges of possible inputs. Many can be treated (approximately) as linear over a limited range. This is attractive because linear systems analysis methods are more fully developed. Feedback can cause some nonlinear open-loop systems to have linear closed-loop behavior. This is sometimes called dynamic linearization [Brogan, 1991]. Other systems are intentionally made nonlinear (e.g. by using on-off controllers) to improve performance. Adaptive and self-learning control systems use feedback and are inherently nonlinear. Feedback also plays an essential role in the training of artificial neural networks through back propagation, and in Kalman filtering and recursive least squares estimation. A predicted signal is compared with a measured signal. The difference is then used to adjust values of parameters or states to reduce future errors.
The Engineering Handbook 1996, IEEE Press, Article on Feedback by William L. Brogan, University of Nevada, Las Vegas pp. 1625-1634
Main page for Science and Engineering