Improved performance of universal integral regulators (Q1411372)

The paper is concerned with the design of an integral control for mi\-ni\-mum-phase nonlinear systems modeled by \[ \left\{\begin{aligned} \dot x&=f(x,\theta) + g(x,\theta)u,\\ y&= h(x,\theta), \end{aligned}\right. \] where \(x\in\mathbb R^n\) is the state, \(u\in\mathbb R\) is the control input, and \(y\in\mathbb R\) is the measured output. The functions \(f,g,h\) depend continuously on \(\theta\), a vector of unknown constant disturbances and parameters that belongs to a compact set \(\Theta\in\mathbb R^l\). It is assumed that, for all \(\theta\in\Theta\), \(f\) and \(g\) are sufficiently smooth vector fields on \(U_\theta\), an open connected subset of \(\mathbb R^n\) that could depend on \(\theta\), and that \(h\) is a sufficiently smooth function in \(x\) from \(U_\theta\to\mathbb R\). In this work, the author is interested in input-output linearizable, minimum-phase nonlinear systems where \(f\) and \(h\) do not necessarily vanish at the origin, i.e., \(f(0,\theta)\neq 0\), \(h(0,\theta)\neq 0\). The case is considered where the system has a well-defined normal form and possibly nontrivial zero dynamics. Earlier work by the author and coworkers has shown how integral control can be combined with robust control techniques to achieve nonlocal and semiglobal regulation results. The results are applicable to output feedback control problems through the use of high-gain observers and saturation nonlinearities. When used with sliding-mode control, the technique produces universal integral regulators that have been shown to coincide with PI and PID controllers, followed by limiters, for relative-degree-one systems and relative-degree-two systems. The use of integral action improves the steady-state response at the expense of degrading the transient response. The present paper summarizes two recent techniques to modify the universal integral regulator so as to improve its transient performance. The first approach modifies the integrator equation in a way that can be looked at as a special design of the traditional anti-windup schemes. The second approach uses nonlinearities to shape the error signal before integration. The two approaches produce very close results when applied to the pendulum example. It is also interesting that, in both approaches, the integrator is driven by an augmented error, composed of the tracking error and its derivatives, rather than the tracking error by itself. However, there are fine differences, between the theoretical results in the two cases.