For tuning controllers, various methods had been proposed. In this paper, data-driven prediction methods based on data-driven controls are proposed. Proposed methods can predict closed-loop responses of controllers before experiment verifications. Compared to conventional methods, proposed methods can predict control responses in the case where the updated controller has unstable zero point. In this paper, the validity of the proposed method is verified through experiments.
In the atomic-level deposition process for semiconductor manufacturing, control of the pressure and gas concentration in the process chamber has become important, and its fast response is required to improve the throughput. At the same time, sensor signals are severely corrupted by noise and there exist constant disturbances which may cause some off-set of the target pressure. Hence this paper proposes a new pressure control system for semiconductor processes, which achieves required fast responses in the presence of heavy sensor noises and constant disturbances. First, a plant model is constructed, and an output estimation mechanism based on the plant model is employed which is effective in the presence of both unknown constant disturbances and heavy measurement noises. Second, we propose a new feedback control system using the output estimation mechanism instead of the output measurement, which can achieve fast response without steady state error and reduce the sensor noise effect. The effectiveness of the proposed control method is demonstrated by some simulations. Finally, experimental validation is performed.
Recently, a control method has been proposed to guarantee the second moment stability, which is a stability index of stochastic systems, for plants with stochastic parameters. This method can be easily coupled with conventional LMI techniques, and thus can be used for advanced control system design. In this paper, we modeled the communication delay with probability distributions and applied the method to a network control system that controls vehicles from a remote location. In addition, we confirmed its effectiveness through simulations and experiments using a driving simulator.
In general, model predictive control (MPC) requires to solve the optimization problem within the control period. To overcome this limitation, we had proposed a novel MPC method based on the prediction of disturbances using an echo state network (ESN) in active vibration control of hybrid electric vehicle (HEV) powertrains. To achieve real-time control in this method, ESN predicts the future disturbances and applies them to MPC, instead of completing the real-time optimization within the control period. However, the prediction accuracy of ESN will decrease when patterns are learned additionally in the previous method and will bring worse control performance. To avoid this weak point, we developed a new learning method of ESN with clustering technic by self-organizing map (SOM) in the proposed method. The performance of the proposed method is verified through the simulations under the different combustion conditions of the internal combustion engine (ICE).