This paper proposes an anomaly cause identification technique in the case of a single anomaly occurring in a plant. The technique 1) defines anomaly modes based on a functional model and the structure for each component or system of plants, 2) uses estimation rules collected previously as a knowledge base to judge the occurrence of each anomaly mode, and 3) identifies the anomaly cause by expanding the search space using the component connection information from the part model including the component where an anomalous behavior is detected. A simple technique is also formulated and implemented to estimate the values of unobserved flow rates based on mass balance equations. The applicability of the proposed anomaly cause identification technique is demonstrated through the applications to several anomaly scenarios for a system around a main fractionator of an oil refinery plant.
In this report a design method for a servo controller with a disturbance feedforward is presented. A conventional controller with feedforward often generates an excess output as a result of the addition of the integrator signal and the feedforward signal. The additional feedback can be inserted to canncel the integrator signal. The controller can be implemented by the same configuration as a two degrees-of-freedom LQI servo system. Experimental results show availability and practicablity to a real plant.
This paper proposes a synthesis method of a supervisor based on a reinforcement learning. In discrete event systems, a supervisor controls disabling of controllable events to satisfy control specifications given by formal languages. However a precise description of the specifications is needed to construct the supervisor. In the proposed method, the specifications are given by rewards, and the optimal supervisor is derived under uncertain environments by considering rewards for occurrence of events and control patterns through learning. By computer simulation, we examine an efficiency of the proposed method.
Support vector machines (SVMs) are known to have high generalization ability for pattern recognition. But since conventional SVMs are originally formulated for binary classification problems, for multiclass problems, the regions in which data cannot be classified exist in the feature space. In this paper, we propose decision-tree-based support vector machines, in which we recursively calculate a hyperplane that separates a class (or some classes) from the others. This can resolve the problem of the conventional SVMs, that is the existence of unclassifiable regions, but a new problem arises. Namely, the division of the feature space depends on the structure of a decision tree. To prevent degradation of generalization ability, more separable class should be separated atthe upper level of a decision tree. We propose four types of decision trees by taking into account the distribution of data in the input space. Using the Euclidean distances between class centers, and Mahalanobis distances, we determine one-to-the-others decision trees and some-classes-to-the-others decision trees. We show the performance of this algorithm using benchmark data sets.
This paper is concerned with the local stability and the quadratic performance of a piecewise affine system. In terms of piecewise quadratic Lyapunov functions, we derive new conditions that explicitly characterize inner approximations of the domain of attraction and the domain of quadratic performance for the piecewise affine system. Furthermore, we apply these analysis conditions to a saturating system which can be encountered in many practical control problems. It turns out for the saturating system that the present stability condition is not as conservative as the local circle criterion. We also propose two numerical methods for maximizing the inner approximation of the domain of attraction. A numerical example is included to show the effectiveness of the present results.