• Journal of Advanced Marine Engineering and Technology
• /
• v.20 no.3
• /
• pp.286-286
• /
• 1996

A neural predictive tracking system for the control of structure-unknown dynamic system is presented. The control system comprises a neural network modelling mechanism for the the forward and inverse dynamics of a plant to be controlled, a feedforward controller, feedback controller, and an error prediction mechanism. The feedforward controller, a neural network model of the inverse dynamics, generates feedforward control signal to the plant. The feedback control signal is produced by the error prediction mechanism. The error predictor adopts the neural network models of the forward and inverse dynamics. Simulation results are presented to demonstrate the applicability of the proposed scheme to predictive tracking control problems.

• Journal of Advanced Marine Engineering and Technology
• /
• v.20 no.3
• /
• pp.154-161
• /
• 1996

A neural predictive tracking system for the control of structure-unknown dynamic system is presented. The control system comprises a neural network modelling mechanism for the the forward and inverse dynamics of a plant to be controlled, a feedforward controller, feedback controller, and an error prediction mechanism. The feedforward controller, a neural network model of the inverse dynamics, generates feedforward control signal to the plant. The feedback control signal is produced by the error prediction mechanism. The error predictor adopts the neural network models of the forward and inverse dynamics. Simulation results are presented to demonstrate the applicability of the proposed scheme to predictive tracking control problems.

• Communications of the Korean Mathematical Society
• /
• v.11 no.4
• /
• pp.1147-1157
• /
• 1996

Nonlinear ill-posed problems arise in many application including parameter estimation and inverse scattering. We introduce a least squares regularization method to solve nonlinear ill-posed problems with constraints robustly and efficiently. The regularization method uses Trust-Region approach to handle the constraints on variables. The Generalized Cross Validation is used to choose the regularization parameter in computational tests. Numerical results are given to exhibit faster convergence of the method over other methods.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.34 no.7
• /
• pp.679-688
• /
• 2010

The major objective of the present study is to extend the applications of inverse analysis to more realistic engineering fields with a complex combustion process rather than the traditional simple heat-transfer problems. In order to do this, the unknown initial mass fractions of $CH_4/O_2$ are estimated from the temperature measurement data by inverse analysis in the practical diffusion-controlled turbulent combustion problem. In order to ensure efficient inverse analysis, the repulsive particle swarm optimization (RPSO) method, which belongs to the class of stochastic evolutionary global optimization methods, is implemented as an inverse solver. Based on this study, it is expected that useful information can be obtained when inverse analysis is used in the diagnosis, design, or optimization of real combustion systems involving unknown parameters.

• Journal of the Korean Mathematical Society
• /
• v.37 no.6
• /
• pp.1031-1042
• /
• 2000

The equations prescribed in Ω⊂R(sup)n are called loaded, if they contain some operations of the traces of desired solution on manifolds (of dimension which is strongly less than n) from closure Ω. These equations result from approximations of nonlinear equations by linear ones, in the problems of optimal control when the control when the control actions depends on a part of independent variables, in investigations of the inverse problems and so on. In present work we study the nonlocal boundary value problems for first-order loaded differential operator equations. Criterion of unique solvability is established. We illustrate the obtained results by examples.

• The KSFM Journal of Fluid Machinery
• /
• v.14 no.1
• /
• pp.34-41
• /
• 2011

Fire characteristics can be analyzed more realistically by using more accurate properties related to the fire dynamics and one way to acquire these fire properties is to use one of the inverse property estimation techniques. In this study two optimization algorithms which are frequently applied for the inverse heat transfer problems are selected to demonstrate the procedure of obtaining pyrolysis properties of charring material with relatively simple thermal decomposition. Thermal decomposition is occurred at the surface of the charring material heated by receiving the radiative energy from external heat sources and in this process the heat transfer through the charring material is simplified by an unsteady 1-dimensional problem. The basic genetic algorithm(GA) and repulsive particle swarm optimization(RPSO) algorithm are used to find the eight properties of a charring material; thermal conductivity(virgin, char), specific heat(virgin, char), char density, heat of pyrolysis, pre-exponential factor and activation energy by using the surface temperature and mass loss rate history data which are obtained from the calculated experiments. Results show that the RPSO algorithm has better performance in estimating the eight pyrolysis properties than the basic GA for problems considered in this study.

• Kyungpook Mathematical Journal
• /
• v.57 no.2
• /
• pp.211-222
• /
• 2017

In this paper, we consider three inverse eigenvalue problems for a special type of acyclic matrices. The acyclic matrices considered in this paper are described by a graph called a broom on n + m vertices, which is obtained by joining m pendant edges to one of the terminal vertices of a path on n vertices. The problems require the reconstruction of such a matrix from given partial eigen data. The eigen data for the first problem consists of the largest eigenvalue of each of the leading principal submatrices of the required matrix, while for the second problem it consists of an eigenvalue of each of its trailing principal submatrices. The third problem has an eigenvalue and a corresponding eigenvector of the required matrix as the eigen data. The method of solution involves the use of recurrence relations among the leading/trailing principal minors of ${\lambda}I-A$, where A is the required matrix. We derive the necessary and sufficient conditions for the solutions of these problems. The constructive nature of the proofs also provides the algorithms for computing the required entries of the matrix. We also provide some numerical examples to show the applicability of our results.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.35 no.7
• /
• pp.657-664
• /
• 2011

This study proposes an inverse method for estimating the boundary temperature in a gas-filled, onedimensional parallel domain enclosed by parallel plates. The distance between the plates is considered submicron to one mm. In the current method, it is assumed that the conditions of both heat flux and temperature are simultaneously applicable to one boundary, while no conditions are applicable to the other boundary The temperature on one of the boundaries should be inversely determined from the known temperature and heat flux on the other boundary. This study proposes a procedure for estimating the unknown boundary temperature through Monte Carlo simulation. Both the forward and inverse problems employ the Monte Carlo approach. The forward (direct) problem is solved by using the direct simulation Monte Carlo while the inverse solution is obtained by the simulated annealing.

• 제어로봇시스템학회:학술대회논문집
• /
• 1989.10a
• /
• pp.144-148
• /
• 1989

We solve the inverse kinematics problems in robotics by employing a neural network. In the practical situation. it is not easy to obtain the exact inverse kinematics solution, since there are many unforeseen errors such as the shift of a robot base the link's bending, et c. Hence difficulties follow in the trajectory planning. With the neural network, it is possible to train the robot motion so that the robot follows the desired trajectory without errors even under the situation where the unexpected errors are involved. In this work, Back-Propagation rule is used as a learning method.

• Journal of applied mathematics & informatics
• /
• v.7 no.1
• /
• pp.29-40
• /
• 2000

We provide semilocal convergence theorems for Newton-like methods in Banach space using outer and generalized inverses. In contrast to earlier results we use hypotheses on the second instead of the first Frechet-derivative. This way our Newton-Kantorovich hypotheses differ from earlier ones. Our results can be used to solve undetermined systems, nonlinear least square problems and ill-posed nonlinear operator equations.