Data-driven Response Estimation-based Tuning and its Validation Using a Ball-and-Beam System

Data science has been attracting a great deal of attention in recent years because of its promise to extract value from the data that abounds in society. In the control engineering field, data-driven design, in which control system design can be performed directly from data, can also be considered a part of data science. Using the data-driven approach, a control system can be optimized directly from controlled data. However, even if a system is optimally designed, its behavior cannot be verified until the system is actually controlled. Therefore, in the present study, a data-driven response estimation-based tuning (DRET) is proposed in order to design a control system based on the estimated time response. It is applied to the control system design of not only stable systems but also unstable systems. In the design of DRET, a finite impulse response filter is used and a model matching problem is solved directly from the control data, to compensate for the difference between an original objective function and a data-driven objective function. The proposed method is applied to the control system design of a ball-and-beam system, which is an unstable system, and its usefulness is verified.