The author previously proposed a functional subspace method in the framework of the generalized-likelihood-ratio (GLR) technique for fault detection in linear discrete dynamic systems. This method detects the fault by estimating the unknown anomaly function in the system as a linear combination of adequate basis functions. Since it is important to decrease the number of the basis functions to raise the detectability, system informations (signals), such as the state and the input of the system, are utilized to construct the basis functions. To construct specifically effective basis functions, a unified approach based on a sensitivity analysis is taken ; that is, based on a sensitivity analysis in innovation sequence, optimal time-variant aggregation vectors are first sought for each system information, and then the linearly-independence and the magnitudes of the resultant basis functions are taken into account. Finally, the robustness of the proposed method to some system uncertainties is mentioned.
The purpose of this study is to construct a new model for handwritten character generation using neural networks. This model can generate not only trained characters as the handwritten-character-generation-models proposed formerly, but also untrained characters by combining radicals. The proposed model uses the neural networks with back-propagation training for generating the radicals of characters and also for estimating the features of untrained characters which signify the peculiarities of each individual hand-writing. The basic experiments verified the validity of the model.
This paper presents the design and evaluation of a discrete time adaptive control scheme for mechanical manipulators. In general, the dynamic models of a mechanical manipulator are described by nonlinear differential equations and many adaptive control schemes based on those models have been proposed. In practical apprications, however, these continuous time control schemes are usually implemented using difference approximation of the differential operations. In this study, a new type of non-linear discrete time model is adopted as the approximate model of the mechanical manipulators. This model is obtained by applying numerical discretization techniques to the minimization problem of the action functional. The validity of the proposed adaptive control scheme is evaluated by the control performance in experiments. Extensive experimental results for a direct driven (DD) robot show the effectiveness of the proposed scheme.
An optimal regulator problem is considered for descriptor systems. The performance index is quadratic, in which the weight matrix for the descriptor variable is chosen so that only the dynamic part of the system is observed. The necessary and sufficient condition for the existence of an optimal stabilizing feedback is presented in terms of positive semidefinite solutions of a generalized Riccati equation. The optimal feedback gain is computed without transforming the descriptor equation into a state equation.