Makishi Nakayama

Optimal Control of Linear Systems with Stochastic Parameters for Variance Suppression: The Finite Time Horizon Case

Abstract In this paper, we consider an optimal control problem for a linear discrete time system with stochastic parameters. Whereas traditional stochastic optimal control theory only treats systems with deterministic parameters with stochastic noises, this paper focuses on systems with both stochastic parameters and stochastic noises. We derive an optimal control law for a novel cost function by which the designer can adjust the trade-off between the average and the variance of the states. Furthermore, a numerical simulation shows the effectiveness of the proposed method.

Parameter Identification for Systems under Disturbance

Abstract Since system deterioration will cause severe accidents, performance degradation and/or large energy loss, control system should be maintained periodically. If such maintenance can be replaced by parameter identification based on input-output data obtained through normal control operations, the cost and the time required for maintenance is reduced. In normal control operations, input-output data are affected by disturbance and thus the existing parameter identification methods yield identification errors. This paper proposes a method to identify the parameters for systems with disturbance. The method assumes that two sensor outputs are available and employs a parameter dependent filter to annihilate the effect of disturbance.

Modeling of Plant Dynamics and Control based on Reinforcement learning

The dynamics modeling of a plant was developed by using Q-learning, which is one method of reinforcement learning. We thought the modeling of the dynamical system to be the function approximation problem for the system output response signal, and enhanced reinforcement learning to the modeling method of the dynamical system. We describe that this modeling method guarantee to offer highly accurate dynamics models by numerical samples, which deals with incinerators combustion. Results of numerical simulation show that the predictive control method using these models has robust tracking property

Combustion Control for Energy Recovery Furnace Using Model Predictive Control

Abstract This paper describes a recovery steam stabilization systems for a fluidized bed incinerator with a high efficiency energy recovery system based on multivariable MPC(Model Predictive Control). It is composed of a combustion control system and a recovery steam control system based on multivariable MPC, which is able to control superheated steam generation rate, temperature and pressure stably by manipulating a steam flow valve and primary air flow into the energy recovery zone. Experimental results for a real fluidized bed incinerator show the usefulness of the proposed method.

A Robust Width Control Design for Hot Strip Mills: An LPV System Approach with Time Variable Transformation

Abstract A robust width control for hot strip mill systems in a steel industry is designed based on LPV system theory. Taking account of varying equilibrium point, parameter uncertainties, and a varying time delay, the design problem is formulated as a robust control problem for an LPV system, where the number of rotation of the exit roller is taken as “time.” The effectiveness of the design is shown by numerical simulation.

On-line Diagnosis Technique for Electro-Hydraulic Servo Dynamics of Rolling Mills Based on Parameter Estimation

This paper deals with a condition diagnosis technique to maintain a dynamic performance of electro-hydraulic servo system and to assure thickness accuracy.The diagnosis technique has following two features. One is to estimate the parameters of servo dynamics during rolling without specified input signals for testing. The other is to diagnose the degradation trend of the servo system and carry out the predictive maintenance based on the on-line monitoring of the estimated parameters, such as the resonant frequency and the damping coefficient.For the plate rolling, the servo signals at biting condition are used, because these signals have high frequency components enough for the estimation. The parameters are estimated such that the frequency characteristics of the model is optimally matched to the actual data. On the other hand, servo data at steady state rolling is used for the strip rolling. The parameters are calculated recursively by the least square estimation.These methods have been successfully applied to the actual rolling and the thickness variation caused by the servo degradation can be reduced by servo gain adjustment according to the estimated parameters.