Survey of a controller design method based on experimental data and a proposal of data conversion method

Several methods have been proposed for designing a controller using experimental data directly without identifying a controlled plant. Among them, optimal controller design methods from one-time experimental data, such as Virtual Reference Feedback Tuning (VRFT), Fictitious Reference Iterative Tuning (FRIT), and Noniterative Correlation-based Tuning (NCbT), are especially expected to reduce the time and cost of designing a controller. VRFT and FRIT make a reference signal from experimental data. In contrast, NCbT is a method that removes noise influence by determining parameters for the controller so as not to have the interrelation of the correlation function between a reference signal and noise. However, VRFT and FRIT cannot deal with data that include noises. Although NCbT can be used to design an optimal controller from experimental data with noise, it is limited to cases where there is no interrelation between a reference signal and noise. In this paper, we show a summary of the conventional method and argue about a problem where there is noise. We also propose a data conversion method using linearity after applying spline fitting instead of using experimental data directly. In addition, we discuss the advantages of the conventional method and the proposed method.