On Validation of Extended State Observer Through Analysis and Experimentation

This paper is concerned with the question of, for a physical plant to be controlled, whether or not its internal dynamics and external disturbances can be realistically estimated in real time from its input–output data. A positive answer would have significant implications on control system design, because it means that an accurate model of the plant is perhaps no longer required. Based on the extended state observer, it is shown that, for an nth order plant, the answer to the above question is indeed yes. In particular, it is shown that the estimation error converges to the origin asymptotically when the model of the plant is given. In face of large dynamic uncertainties, the estimation error is shown to be bounded. Furthermore, it is demonstrated that the error upper bound monotonously decreases with the bandwidth. Note that this is not another parameter estimation algorithm in the framework of adaptive control. It applies to a large class of nonlinear, time-varying processes with unknown dynamics. The solution is deceivingly simple and easy to implement. The results of analysis are further verified through simulation and hardware tests.