Nicely nonlinear modelling and control (Q4026327)

[For the entire collection see Zbl 0745.00051.] The author introduces the notion of proper linearizability by output- feedback, which basically means that feedback linearization can be found, such that the relationship between the new input and the output is linear. A bounded-state-bounded-output (BSBO) nonlinear element affects the linear element, which in turn affects a bounded-input-bounded-state (BIBS) nonlinear element. The nonlinear elements do not affect the linear behavior between the new input and the output. Definitions of BIBS and BSBO are given in Appendix A. The linearizing compensator is termed normal if the resulting linear system is externally equivalent to a linear system which is tangent to the original nonlinear system at an operating point. A model is a nicely nonlinear model, if it is tangent to the real nonlinear system at an operating point and properly linearizable by output-feedback. The author then remarks, that \(a\) if a normal linearizing compensator is applied in an inner loop and then a typical linear compensator designed by classical design methods (based on linearizing the plant at an operating point, etc.) is applied in an outer loop, then the system should perform well over the nonlinear operation of the nicely nonlinear plant model. Further, that the existence of a smooth linearizing compensator implies the existence of a normal linearizing compensator for a nicely nonlinear plant model. An outline of the proof in Appendix B. The author then states three main objects involved in the nicely nonlinear modelling problem for a given plant, namely: the data set and set of experiments, a parameterized class of nicely nonlinear models and the model selection criterion. He also notes that these three objects cannot be chosen independently. The author finally cites two examples. The first one deals with a nonlinear system composed of polynomials preceding apparently linear transfer functions. This system can be feedback linearized in a rather trivial way, assuming one of the polynomials is invertible. The second example deals briefly with the describing function approach and discusses the well-known assumption of low pass filtering effects of the plant, thus leading to possibly inaccurate results for very low frequencies. The reviewer finds the paper and the notion of nicely nonlinear modelling and control interesting. Some of the points discussed are rather trivial from an engineering point of view.