摘要:
Iterative feedback tuning (IFT) enables the tuning of feedback controllers based on measured data without the need for a parametric model by performing 1 + (# inputs) × (# outputs) experiments. The method is based on optimization, and in each iteration, an unbiased gradient estimate is employed. Due to unbiased gradient estimates, the method converges to a stationary point of the control criterion, provided that the closed-loop signals remain bounded throughout the iterations. Recently, a new extension to this approach has been developed called Stochastic Approximation Iterative Feedback Tuning (SAIFT). The aim of SAIFT is to efficiently implement IFT for multivariable linear time-invariant systems such that the required number of experiments reduces to two for any Multi-Input Multi-Output (MIMO) system. The first experiment consists of a normal experiment, as in regular IFT. In a second experiment, a randomization technique is employed, resulting in an unbiased gradient estimate from a single dedicated gradient experiment, regardless of the size of the MIMO system. This gradient estimate is used in a stochastic gradient descent algorithm. Simulation results prove that the approach reduces the number of experiments required to converge but the method has yet to be experimentally validated. The aim of this work is to implement the SAIFT method on a mechatronic MIMO system. Particular attention is given to the issue of keeping experiment time to a minimum, and methods to increase the efficiency and performance of the algorithm are considered. For example, it is shown that in case of a diagonal controller structure, the variance of the gradient estimates can be reduced significantly. Checking feedback controller stability is automated based on a nonparametric model through the generalized Nyquist theorem. These nonparametric models are usually simple, fast, and accurate to obtain. It is shown that, for tuning an arbitrary linear time-invariant MIMO controller, two expe