In many industrial applications, a mathematical description of the plant is not available or the modeling study is too expensive and time-consuming. Data-driven methods overcome these problems, allowing the control engineer to quickly select the best model of the process in a specific class, in order to use it for design purpose. Nevertheless, in classical system-identification procedures, it is difficult to find a model which is both simple and reliable. Moreover, detecting which dynamics is relevant for the final control objective is not a simple task. In the so-called “direct” data-driven methodologies, a controller is directly selected from data, without need to identify a model of the system. In this way, process dynamics are automatically considered relevant or not, depending only on their weight on the final control index.
In this work, existing direct data-driven design methods are extended to a larger class of industrially relevant control problems, including some critical nonlinear issues. Statistical efficiency of the methods is also analyzed and improved via optimal input design. Finally, some real-world complex engineering applications are dealt with to prove the effectiveness of the extended methods.