Modelling the influence of non-minimum phase zeros on gradient based linear iterative learning control

The subject of this paper is modeling of the influence of non-minimum phase plant dynamics on the performance possible from gradient based norm optimal iterative learning control algorithms. It is established that performance in the presence of right-half plane plant zeros typically has two phases. These consist of an initial fast monotonic reduction of the L2 error norm followed by a very slow asymptotic convergence. Although the norm of the tracking error does eventually converge to zero, the practical implications over finite trials is apparent convergence to a non-zero error. The source of this slow convergence is identified and a model of this behavior as a (set of) linear constraint(s) is developed. This is shown to provide a good prediction of the magnitude of error norm where slow convergence begins. Formulae for this norm are obtained for single-input single-output systems with several right half plane zeroes using Lagrangian techniques and experimental results are given that confirm the practical validity of the analysis.