In this paper we consider the segmentation problem of a texture image composed of different kinds of texture fields not as a pattern classification problem but a combinatorial optimization problem. We apply a probabilistic and effective successive search procedure of genetic algorithms to the clustering of small regions in a feature space. Moreover, we propose a new feature extraction scheme using two-dimensional wavelet transform, which can extract the hierarchical characteristics of the texture feature, and perform the accurate feature extraction upon the unstationary texture fields.
This paper is concerned with controller design and its experimental evaluation for active vibration control of a building structure in case of earthquake. Taking account of the mode-decomposition for the building structure model, we design an H∞ controller for active vibration control, where we adopt the linear matrix inequality approach because the standard assumption of H∞control is not satisfied in this case. Then we evaluate its effectiveness for a five-story building structure via experiments.
In this paper, we consider an iterative learning control method (for short, ILC method). With the iteration of experiments, the ILC method yields the desired input for tracking the target trajectory. Most of former ILC methods use the time derivative of the error signal or the passivity of systems. Contrary to these former methods, this paper proposes an alternative ILC algorithm which does not use such things. This algorithm has the following property; the input space is restricted in the prescribed subspace, and the iterative learning law uses the modified error signal, which is projected on this input subspace. The effectiveness of the proposal method is demonstrated by a numerical example and an experiment.
This paper proposes a new self-tuning controller having a new design parameter. In selecting the design parameter, the controller gives a strongly stable self-tuning controller, that is, not only the closed-loop system is stable, but also the controller itself is stable.The controller consists of a generalized minimum variance controller and a parameter identification law. The proposed controller has an extended minimum variance controller with a newly introduced design parameter. The parameter is introduced by applying the coprime factorization approach and Youla parametrization of stabilizing compensators to the design of minimum variance controller.
In this paper, we propose a method of optimum seeking in an uncertain environment by extending the conventional genetic algorithms (GA). The key point of our approach is to evaluate an individual not directly by an objective value of a corresponding solution currently observed, but by accumulating values which have been observed at preceding generations. Finally, we confirm the effectiveness of our extended GA through some computaitional experiments using simple function optimization problems.
A feasible discrete-time SAC (Simple Adaptive Control) algorithm extended from continuous-time system was proposed for a single-input single-output (SISO) system and removed an offset between a plant output and a reference model output, by inserting parallel feedforward compensators to both of the plant and the reference model. Here, we contrive the design parameters instead of adding the parallel feedforward compensator to the model to remove the above-mentioned offset, and propose a discrete-time SAC algorithm which is feasible and applicable to a multi-input multi-output (MIMO) system. We prove the stability of the system by using asymptotic output tracker theory instead of CGT (Command Generator Tracker) theory, which removes the conventional restriction between the plant and the model.