(Return to ICNN97 Homepage) (Return to ICNN'97 Agenda)
ICNN'97 FINAL ABSTRACTS
INTELLIGENT ESTIMATION & CONTROL (Return to Top)
ICNN97 Intelligent Estimation & Control
Session: EC1A Paper Number: 582 Oral
Intelligent flight controller for helicopter control
Saleh Zein-Sabatto, Yixiong Zheng and Wen-Ruey Hwang
Keywords: Intelligent flight control helicopter control genetic algorithm
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC1B Paper Number: 403 Oral
Radial basis function network based power system stabilizers in multimachine power systems
M. A. Abido and Y. L. Abdel-Magid
Keywords: Radial Basis Functions Power system stabilizer
Abstract:
A Radial Basis Function Network (RBFN) based Power System Stabilizer (PSS) is presented in this paper to improve the dynamic stability of multimachine power systems. The proposed RBFN is trained over a wide range of operating conditions in order to re-tune the parameters of the PSS in real-time. Time domain simulations of a multimachine power system with different operating conditions subject to a three phase fault are studied and investigated. The performance of the proposed RBFN PSS is compared to that of conventional power system stabilizer (CPSS). The results show the good damping characteristics of the proposed RBFN PSS over a wide range of operating conditions.
_____
ICNN97 Intelligent Estimation & Control
Session: EC1C Paper Number: 470 Oral
Synthesis of a discrete-time feedback neural controller
P. Couturier, A. Johannet and M. Betemps
Keywords: discrete-time feedback neural controller real plants nonlinear functions
Abstract:
Real plants, generally nonlinear, are not satisfactorily controlled by linear methods. For this reason, a great part of the research centred on control theory is currently devoted to nonlinear and adaptive control. In this context, neural network are attractive because of their intrinsic abilities to identify nonlinear functions. After a presentation of neural networks and of learning methods, we propose an original control scheme inspired by "specialised learning". This scheme allows an "on-line" computation of a neural controller that cancels tracking error in accordance with an imposed reference model. The ways in which learning is performed are discussed and a canonical representation of the control scheme is presented. Tested in simulation, this control scheme exhibits satisfactory properties especially for time convergence and nonlinear abilities.
_____
ICNN97 Intelligent Estimation & Control
Session: EC1D Paper Number: 27 Oral
Pulp digester level prediction using multiresolution networks of locally active units
A. Fern, J. Miranda, M. T. Musavi and D. R. Coughlin
Keywords: Pulp digester level prediction multiresolution networks locally active units
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC1E Paper Number: 181 Oral
Neural Network adaptive control of the penicillin acylase fermentation
Mei-J. Syu and C. B. Chang
Keywords: Adaptive control fermentation
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC2A Paper Number: 243 Oral
Relaxation oscillator networks with time delays
Shannon Campbell and DeLiang Wang
Keywords: Relaxation oscillator networks time delays excitatory chemical synapses
Abstract:
We study relaxation oscillators with couplings that mimic excitatory chemical synapses. Such oscillator networks have been shown to synchronize quickly without time delays. We present analytic results for a pair of oscillators showing that loose synchrony occurs for a wide range of initial conditions and time delays. Simulations indicate that locally coupled networks in one and two dimensions also exhibit loose synchrony. To characterize loose synchrony we introduce a measure of synchrony, the maximum time difference. We obtain histograms of this measure for one and two dimensional oscillator networks. Also, we conjecture that there is a range of initial conditions for which the maximum time difference remains bounded as the system evolves.
_____
ICNN97 Intelligent Estimation & Control
Session: EC2B Paper Number: 542 Oral
Meanfield neurodynamics with changing values of threshold logic
Harold Szu and Charles Hsu
Keywords: Meanfield neurodynamics threshold logic Hebbian synaptic weight
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC2C Paper Number: 634 Oral
Experimental analysis of behavior stability in neuron gain domain in recurrent complex-valued neural networks
Akira Hirose and Hirofumi Onishi
Keywords: neuron gain domain recurrent complex-valued neural networks behavior stability
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC2D Paper Number: 136 Oral
Numerical studies of resonance in a sinusoidally driven chaotic neuron model and its global coupling network
Shin Mizutani, Takuya Sano, Tadasu Uchiyama, Noboru Sonehara, and Katsunori Shimohara
Keywords: Stochastic resonance Chaotic neuron global coupling network
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC2E Paper Number: 370 Oral
Detecting oscillations in neural networks via frequency domain analysis
Jorge L. Moiola, Daniel Berns, Guanrong Chen and Haluk Ogmen
Keywords: neural networks oscillations frequency domain analysis periodic solutions
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC2F Paper Number: 278 Oral
Equivalent dynamics in different neural oscillator models
Jorg Berdux and Rainer Malaka
Keywords: neural oscillator ordinary differential equations harmonic oscillators
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC3A Paper Number: 220 Oral
Adaptive filtering and prediction based on Hopfield neural networks
Mariko Nakamo-Miyatake and Hector Perez-Meana
Keywords: Adaptive filtering Adaptive prediction Hopfield neural networks
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC3B Paper Number: 510 Oral
The analytical solution of gamma filter model
Tai-Wu Chiang
Keywords: gamma filter functional approximator numerical training
Abstract:
Gamma filter has been proposed by De Vries and Jose C. Principe [1] as a functional approximator . In order to get the (sub) optimal values of weights and (temporal resolution) mu, numerical training algorithms have been derived by Steve Lawrence [3] and Principe. This paper presents an analytical approach to this optimization problem. Using the classical Laguerre Polynomial, we have successfully uncoupled and derived the optimal exact solutions for the weights and mu. Finally, we have a geometrical interpretation of the whole idea as an inner product preserved transformation between linear vector spaces.
_____
ICNN97 Intelligent Estimation & Control
Session: EC3C Paper Number: 645 Oral
A nonlinear model predictive controller using neural networks
Ozgur Karahan, Canan Ozgen, Ugur Halici and Kemal Leblebicioglu
Keywords: nonlinear model predictive control artificial neural networks
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC3D Paper Number: 627 Oral
Neural network identification and control in the presence of noise
O. Olurotimi, R. McDonald and S. Das
Keywords: Neural network identification Neural network control noise
Abstract:
This paper examines the performance of a control system design in the presence of noise. An architecture from the seminal work of Narendra and Parthasarathy is modified to institute recurrence in the neural net and the recurrent system performance is compared to the feed forward system response. The process of comparing the feed forward to the recurrent system is repeated for ten networks each having unique weights. The weights of each network are processed to obtain certain previously derived performance measures. The results of the experiments show that bias and variance performance of neural network control and identification systems can be improved by using the performance measures in the design process.
_____
ICNN97 Intelligent Estimation & Control
Session: EC4A Paper Number: 516 Oral
Internal feedback neuron networks for modelling of an industrial furnace
Anil K. Gobbak, H. Raghavendran and Anand M. Tapas
Keywords: neural networks recurrent neural networks system identification industrial furnaces
Abstract:
Most of the physical processes are dynamical in nature. Modeling of such systems using process physics is a complicated task involving a lot of time and effort. Artificial neural networks with their self learning capabilities offer promise for an alternative way of modeling.
This paper presents a dynamic neural network architecture which has the potential for use in dynamic system identification. In this network, the behaviour of each neuron is made dynamic by incorporating feedback connections in it. The delayed outputs of each neuron are fedback to itself as additional inputs through weights. This particular network architecture is termed as Internal Feedback Neuron Network (IFNN).
The networks is trained using a specially derived gradient based training algorithm. Simulations have been carried out on an Industrial Furnace and satisfactory results have been obtained.
_____
ICNN97 Intelligent Estimation & Control
Session: EC4B Paper Number: 170 Oral
Backpropagated adaptive critic neurofuzzy controller for nonlinear dynamic system
Z. Gherari and Y. Hamam
Keywords: Neurofuzzy controller prediction rule-based refinement
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC4C Paper Number: 545 Oral
Training strategies for critic and action neural networks in dual heuristic programming method
George G. Lendaris and Christian Paintz
Keywords: Training strategies critic neural networks action neural networks dual heuristic programming method
Abstract:
This paper discusses strategies for and details of training procedures for the Dual Heuristic Programming (DHP) methodology.
This and other approximate dynamic programming approaches (HDP, DHP, GDHP) have been discussed in some detail in the literature, all being members of the Adaptive Critic Design (ACD) family. The inverted pendulum problem is used as the example application. As reported elsewhere, this "plant" has been successfully controlled using DHP. The present paper suggests and investigates several alternative procedures for training the DHP, and compares their performance with respect to speed and quality of resulting controller design. A promising modification is to introduce a real copy of the criticNN (criticNN#2) for making the "desired output" calculations, and very importantly, this criticNN#2 is trained differently than is criticNN#1. The idea is to provide the "desired outputs" from a stable platform during an epoch while adapting the criticNN#1. Then at the end of the epoch, criticNN#2 is made identical to the then-current adapted state of criticNN#1 and the actionNN can be simultaneously trained on-line during each epoch, with a faster overall convergence than the older approach. Further, the measures used herein suggest that a "better" controller design (the actionNN) results.
_____
ICNN97 Intelligent Estimation & Control
Session: EC4D Paper Number: 306 Oral
Neural network isolation of system inputs for transient modelling and control
Anya Tascillo
Keywords: neural network isolation system inputs transient modelling and control
Abstract:
A neural network is used to predict the sensitivity of a complex nonlinear system such as an automobile to input variation, which will aid greatly in the effort to model the system and the effects of changes to its controllers. A blend of signal processing techniques is used to provide maximum resolution neural network inputs for various drivers, vehicles, engine technologies, transmissions, velocity traces, and operating temperatures. The neural net predicts what four different vehicle outputs will be, given a sample of driving inputs.
_____
ICNN97 Intelligent Estimation & Control
Session: EC4E Paper Number: 355 Oral
The application of wavelet neural networks to nonlinear predictive control
Dexian Huang and Yihui Jin
Keywords: wavelet neural networks nonlinear predictive control least squares learning algorithm
Abstract:
An identification and predictive control strategy for nonlinear processes based on orthogonal wavelet basis function networks is proposed. In this paper, a wavelet neural network with a linear least squares learning algorithm is developed for a process model. This can be used with nonlinear programming to implement nonlinear model predictive control strategy. Since simplified on-line optimization method has been developed, this control strategy is very easy to implement. Using the proposed identification and control strategy ,a control system of bilinear process is simulated. It shows excellent performance superior to a standard PID controller for the nonlinear processes.
_____
ICNN97 Intelligent Estimation & Control
Session: EC4F Paper Number: 223 Oral
a neural network algorithm using wavelets and auto regressive inputs for system identification
J.R. HUll, H.P. Pendse and M.T. Musave
Keywords: wavelets auto regressive inputs system identification
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC5A Paper Number: 1 Oral
Modeling of neural networks in feedback systems using describing functions
Peter C. Y. Chen and James K. Mills
Keywords: feedback systems describing functions SISO neural networks
Abstract:
In this article, a novel approach to modeling of neural networks in feedback systems using describing functions is proposed.
Results on using describing functions for modeling of single-input single-output (SISO) neural networks with respect to an exponential input are presented. Through a simple example, it is then demonstrated that the resulting models of neural networks can be used to analytically calculate values for network weights such that the transient behavior of a feedback system embedded with a neural network can be ``shaped'' as desired. These results suggest that the proposed approach of using describing functions for modeling of neural networks could facilitate further theoretical analysis and synthesis of neural networks in feedback systems. Simulation conducted to verify the analytical results are described. Sources of approximation error in this proposed approach are examined, and potential applications and possible extension of the work reported in this article are discussed.
_____
ICNN97 Intelligent Estimation & Control
Session: EC5B Paper Number: 363 Oral
A new method for the analysis of neural reference model control
Michael Wigbers and Martin Riedmiller
Keywords: neural reference model control MRAC-control backpropagation
Abstract:
In recent years there has been much effort to develop the theoretical aspects of neural MRAC-Control, that is to find conditions under which an unknwon process can be identified by an input-output model and controllers can be trained by gradient descent. On the other hand, the application of neural network techniques to real world control of nonlinear dynamical systems has been of substantial interest. Since the theoretical conditions that ensure controllability and the applicability of indirect adaptive control are hard to verify in practice, the success of controller training is mostly shown by testing relevant situations. We trained a controller for a subsystem of a spark ignition engine by dynamic backpropagation and various truncated gradient algorithms. Afterwards we related the neural MRAC-approach to Pole placement and Linearization techniques in order to show the successful training by Pole analysis of the completely trained loop. This is a new method to verify the plausibility of the adaptation process and the trained regulator.
_____
ICNN97 Intelligent Estimation & Control
Session: EC5C Paper Number: 512 Oral
Inverse mappings of continuous functions using feedforward neural networks
Hatem M. Deif, Jacek M. Zurada and Wojciech Kuczborski
Keywords: Inverse mappings continuous functions feedforward neural networks
Abstract:
In this paper we present a methodology for solving inverse mapping of continuous functions modeled by multilayer feedforward neural networks.
The methodology is based on an iterative update of the input vector towards a solution, which escapes local minima of the error function.
The update rule is able to detect local minima through a phenomenon called ``update explosion.'' The input vector is then relocated to a new position based on a probability density function (PDF) constructed over the input vector space. The PDF is built using local minima detected during the past search history. Simulation results demonstrate the effectiveness of the proposed method in solving the inverse mapping problem for a number of cases.
_____
ICNN97 Intelligent Estimation & Control
Session: EC5D Paper Number: 41 Oral
An escape method from local minimum by orbital correction method at controller learning
Kotaro Hirasawa, Masanao Ohbayashi and Hiroto Takata
Keywords: escape method local minimum orbital correction method controller learning
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC5E Paper Number: 305 Oral
Partitioning input space for reinforcement learning for control
Dean F. Hougen, Maria Gini and James Slagle
Keywords: input space Partitioning control-learning temporal domains
Abstract:
This paper considers the effect of input-space partitioning on reinforcement learning for control. In many such learning systems, the input space is partitioned by the system designer. However, input-space partitioning could be learned. Our objective is to compare learned and fixed input-space partitionings in terms of the overall system learning speed and proficiency achieved. We present a system for unsupervised control-learning in temporal domains with results for both fixed and learned input-space partitionings. The trailer-backing task is used as an example problem.
_____
ICNN97 Intelligent Estimation & Control
Session: EC5F Paper Number: 81 Oral
Improving tuning capability of the adjusting neural network
Yoichi Sugita, Masahiro Kayama and Yasuo Morooka
Keywords: tuning capability adjusting neural network parameter tuning
Abstract:
The Adjusting Neural Network (AJNN) we proposed in previous papers has a remarkable capability for parameter tuning of a control model. Namely it can complete parameter tuning accurately with small tuning numbers. However, when parameter errors are relatively large, its tuning capability may occasionally deteriorate, which leads to an increase of tuning numbers. In this paper, we discuss two ways of overcoming this weakness of the AJNN. We propose a new learning algorithm for the AJNN and we develop the AJNN architecture. We simulate the effectiveness of both approaches and compare their results with results from our previous AJNN using the problem of temperature control for a reheating furnaces plant.
_____
ICNN97 Intelligent Estimation & Control
Session: EC6A Paper Number: 437 Oral
The use of feedforward neural networks to cancel nonlinearities of dynamic systems
Edilberto P. Teixeira, Elmo B. Faria, and Adriano Breunig
Keywords: linearization direct estimation feedback
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: EC6B Paper Number: 526 Oral
Fixed-weight controller for multiple systems
L. A. Feldkamp and G. V. Puskorius
Keywords: Fixed-weight controller multiple systems closed-loop system
Abstract:
We demonstrate here a perhaps unexpected result: the ability of a single fixed-weight time-lagged recurrent network, properly trained, to act as a stabilizing controller for multiple (here 3) distinct and unrelated systems, without explicit knowledge of system identity.
This capability, which may be regarded as a challenge to the usual understanding of what constitutes an adaptive system, seemed plausible to us on the basis of our earlier results on both multiple time-series prediction and robust controller training. We describe our training method, which has been enhanced toward enforcing stability of the closed-loop system and dealing with process noise, and provide some results.
_____
ICNN97 Intelligent Estimation & Control
Session: EC6C Paper Number: 492 Oral
The state estimation of the CSTR system based on a recurrent neural network trained by HGAs
Lei Jia, Guangdong He and J. P. Jiang
Keywords: BP algorithm genetic algorithm hybrid genetic algorithms recurrent neural network
Abstract:
The CSTR system (The Continuous Stirred Tank Reactor system) is a typical nonlinear system. At present, one of its states, reaction consistence, can not be measured. In this paper, a recurrent neural network is used to estimate the value of the state. Nevertheless, due to the strong non-linearity of the system, traditional training method such as BP algorithm usually converges in local optimum. Genetic Algorithms(GAs), as a global optimization search method, can solve the problem, but the conventional GAs converge very slowly. To improve the learning speed of the neural network, a Hybrid Genetic Algorithm (HGA) is employed. The results demonstrate the proposed HGA can get very good effect.
_____
ICNN97 Intelligent Estimation & Control
Session: EC6D Paper Number: 489 Oral
The model reference adaptive control based on the genetic algorithm
Lei Jia and J. P. Jiang
Keywords: model reference adaptive control genetic algorithm PID control
Abstract:
A new control method that is a Model Reference Adaptive Control method (MRAC) based on the combination of PID control and the Genetic Algorithm (GA) is introduced. It implements the characteristic of the Genetic Algorithm's global optimization to optimize the PID's three control parameters: Kp, Ki, Kd, to obtain the best control effect. This paper gives an example using this method to control a nonlinear system --- Continuous Stirred Tank Reactor system (CSTR). Because the state of the CSTR system con not be gotten, a neural network is used to estimate the value of the state. This neural network is training by the GA. The simulation results are given.
_____
ICNN97 Intelligent Estimation & Control
Session: EC6E Paper Number: 45 Oral
Nonlinear control system with radial basis function controller using random search method of variable search length
Ning Shao, Kotaro Hirasawa, Masanao Ohbayashi, Kazuyuki Togo and Mitsuo Ileuchi
Keywords: Nonlinear control system radial basis function controller random search method variable search length
Abstract:
In this paper, a new optimization method which is a kind of random searching is presented. The proposed method is called RasVal which is an abbreviation of Random Search Method with Variable Search Length and it can search for a global minimum based on the probability density functions of searching, which can be modified using informations on success or failure of the past searching in order to execute intensified and diversified searching. The features of RasVal are such that it does not require differential calculation as the gradient method, therefore, it takes a shorter calculation time than the gradient method, and the random search with intensification and diversification is carried out in order to solve the local minimum problem. By applying the proposed method to a nonlinear crane control system which can be controlled by the Universal Learning Network with radial basis function(R.B.F.), it has been proved that RasVal is superior in performance to the commonly used back propagation learning algorithm, and it has also been shown that the RasVal has better performance of the generalization capability than the gradient method.
_____
ICNN97 Intelligent Estimation & Control
Session: EC6F Paper Number: 225 Oral
Multi-layer neural network controller for trajectory tracking control of a pneumatic cylinder
David C. Gross and Kuldip S. Rattan
Keywords: Multi-layer neural network controller trajectory tracking control pneumatic cylinder
Abstract:
A Feedforward MNN Controller for Pneumatic Cylinder Trajectory Tracking Control
Pneumatic cylinders are used in many industrial applications to position loads using a rectilinear motion. Pneumatic cylinders are limited to a narrow range of applications because their nonlinear dynamics are difficult to control with linear controllers. Conventional linear control techniques can not compensate for both the internal friction and the compressible air flow present in the cylinders. Multilayer neural networks (MNNs) are nonlinear mappings which can be used to compensate for the nonlinear nature of these dynamic systems. A model of a pneumatic cylinder was developed to provide training data for the feedforward MNN controller. The MNN was designed to cancel the cylinder dynamics and was used in conjunction with a proportional feedback controller to control the cylinder motion. The MNN was trained over a range of constant velocity cylinder trajectories, and the resultant controller allows the model to follow a constant velocity trajectory within the trained state space.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 43 Poster
Non-linear system control using learning Petri network
Masanao Ohbayashi, Kotaro Hirasawa and Singo Sakai
Keywords: Non-linear system control learning Petri network brain science
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 610 Poster
Frequency-domain vibration control using adaptive neural networks
Gary G. Yen
Keywords: vibration suppression adaptive neural control Frequency-domain structural control
Abstract:
This paper proposes a frequency-domain adaptive neural controller appropriate for on-line system identification and real-time adaptive control. After survey of existing literature involving controls of structural vibration, we report new developments carried out under the adaptive neural control program for the USAF Phillips Laboratory. The new results include a frequency-based neural control architecture suitable for MIMO systems subjected to tonal disturbances in the presence of sensor and actuator failures. This architecture is demonstrated on the ASTREX test facility.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 18 Poster
Global and local optimization in adaptive neurocontrol
Tomas Hrycej
Keywords: Global optimization local optimization adaptive neurocontrol
Abstract:
Global optimization algorithms are more adequate for neurocontrol applications than presently used local algorithms. If embedded in an appropriate scheme providing for training data accumulation, they can be used also for on-line adaptation. The results of a computational experiment show that the danger of divergence of adaptation with help of local algorithms is not only theoretical - it is significant also for relatively simple plants.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 563 Poster
Comparison of CMACs ans radial basis functions for local function approximators in reinforcement learning
R. Matthew Kretchmar and Charles W. Anderson
Keywords: CMACs radial basis functions local function approximators reinforcement learning
Abstract:
CMACs and Radial Basis Functions are often used in reinforcement learning to learn value function approximations having local generalization properties. We examine the similarities and differences between CMACs, RBFs, and normalized RBFs and compare the performance of Q-learning with each representation applied to the mountain car problem. We discuss ongoing research efforts to exploit the flexibility of adaptive units to better represent the local characteristics of the state space.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 126 Poster
Analytical and computational results on recurrent dynamic neural network for signal representation
Marc Karam and Mohamed A. Zohdy
Keywords: Recurrent neural networks Signal representation signal modeling
Abstract:
This paper presents a recurrent dynamic neural network to solve signal representation and processing problems. The neural network is essentially composed of feedback-type connections, and arrays of integrators, linear gains, and nonlinear activation functions. By seeking a minimum energy state, the neural network solves for the sets of representation coefficients required to model a given signal in terms of elementary basis signals. An analytical model of the recurrent neural network was obtained through discretization of the integrator blocks and linearization of the activation function. Continuity of the algorithm when segment boundaries are crossed is made possible by varying the slope of the linearized activation function. The proposed approach results in a closed analytical form of the recurrent neural network solution. The perceived advantages are estimation of robustness, prediction of convergence by examining the eigenvalues of the analytical state matrix, and increase of computational speed. Moreover, unlike classical traditional methods, the approach offers the possibility of handling time-varying signals with uncertainties and considerable noise.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 522 Poster
Non-monotone network dynamics
Thomas Trappenberg
Keywords: recurrent neural networks Non-monotone transfer functions network dynamics
Abstract:
Recurrent neural networks with non-monotone transfer functions have the interesting feature that they do not settle into a state if the start configuration is too far from the stored patterns. Instead, the network shows strong fluctuations. A first attempt to analyze the nature of this mode is made.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 466 Poster
Yield improvement in GaAs IC manufacturing using neural network inverse modeling
Jacek M. Zurada, Andrzej Lozowski and Aleksander Malinowski
Keywords: Yield improvement GaAs IC manufacturing neural network inverse modeling
Abstract:
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 182 Poster
A design method of optimal control system by use of neural network
Hiroaki Nakanishi, Takehisa kohda, and Koichi inoue
Keywords: Neural net controller Powell's conjugate direction
Abstract:
This paper proposes a method to design an optimal control system by use of neural network. Because the normal Back-Propagation(BP) method cannot be applied to this case, we choose the Powell's conjugate direction algorithm for training the neural network. In a regulator problem, the neural network functions as a state feed-back controller, and in a servo problem it functions as both feed-foward and feed-back controller. The proposed method can be applied to various problem where conventional methods cannot be applied. Simulation results show the effectivity of the proposed method.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 493 Poster
Non-linear PID controler using neural networks
Toshiharu Matsukuma, Atsushi Fujiwara, Junbo Song and Yoshihisa Ishida
Keywords: Non-linear controler PID controller pneumatic servo system
Abstract:
This paper describes a non-linear PID control scheme for pneumatic servo system. The PID controllers are being widely used in industrial applications because of simple, cheap and excellent performance. However, the requirement for control precision becomes higher and higher, as well as the plants become more and more complex. In order to achieve the satisfied control performance, we have to consider the affection of nonlinear factor contained in plant, such as pneumatic cylinder. Hence, the conventional PID controller is limited in some applications. In this paper, a nonlinear PID algorithm and its application to the pneumatic servo system are introduced.
The test is carried out in practical pneumatic servo system, the experimental results suggest its superior performance and disturbance rejection.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 478 Poster
An application of MNN trained by MEKA for the position control of pneumatic cylinder
Junbo Song, Xiaoyian Bao and Yoshihisa Ishida
Keywords: multiple extended Kalman algorithm multilayer neural network pneumatic cylinder
Abstract:
As an important driving element, the pneumatic cylinder is being widely used in industrial applications because of simple, cheap and excellent performance. However, along with the developing of control technology, the requirement for control precision becomes higher and higher, In many case, in order to achieve the satisfied control performance, we have to consider the affection of nonlinear factor contained in pneumatic cylinders. In order to solve the nonlinear problems of pneumatic servo system, in this paper we proposed a control scheme based on Multilayer Neural Network(MNN) trained by Multiple Extended Kalman Algorithm (MEKA). The test results of MNN controller trained by MEKA in a practical pneumatic servo system suggests the superior performance. As well as the experimental results also show that the proposed method has less sensitivity than the neural network trained by simple gradient descent training algorithms.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 476 Poster
Perturbation analysis of a class of neural networks
Anke Meyer-Base
Keywords: Perturbation analysis robustness stability associative memory artificial neural networks
Abstract:
We establish robustness stability results for a large class of artificial neural networks for associative memories under parameter perturbations and determine conditions that ensure the existence of asymptotically stable equilibria of the perturbed neural system that are near the asymptotically stable equilibria of the original unperturbed neural network.
The proposed stability analysis tool is the sliding mode control and it facilitates the analysis by considering only a reduced--order system instead of the original one and time--dependent external stimuli.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 417 Poster
Universal neural network controllers
Chu Kwong Chak, Gang Feng, Jian Ma, and Marimthu Palaniswami
Keywords: Neural network controller Stabilization
Abstract:
Neural networks have been developed for many years. Recently there appeared a number of results on the study of neural networks applying to dynamic system control. To enrich the fundamental basis of analysis, this paper shows that neural networks are universal controllers. That is, if the system to be controlled is stablized by a continuous controller, there exists a neural network which can approximate the controller such that the controlled system by the neural network is stablized with a given bound of output error.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 75 Poster
Neural spectral composition for function approximation
Andrea Pelagotti and Vincenzo Piuri
Keywords: Neural spectral composition function approximation minimal dimension
Abstract:
An innovative neural-based approach for function approximation is proposed by means of the spectral analysis of the function y(x) to be approximated. Approximation is obtained by the spectral composition of the approximating function y'(x) performed by a neural network. The synthesis procedure for the neural network ensures the minimal dimension of the network itself, according to the chosen approximation error.
Parameters adaptation is very fast. Since the most of the structure is independent from the particular approximated function, the circuit architecture implementing the network can be easily modularized for architecture adaptation.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 165 Poster
Modeling liquid phase oxygenation process using neural networks
Jiabao Zhu, Zhengzhi Han, Jin Xiao and Ming Rao
Keywords: liquid phase oxygenation process neural networks backpropagation
Abstract:
This paper describes a modeling application for LPO (Liquid Phase Oxygenation) process using a backpropagation neural network. An integrated neural professional IITM based software for developing neural network models is developed for this project. In this software, a visual data processing method for improving the effects of training , a correlation analyzer that can be used to help simplify the model's inputs/outputs structure and a training kit are all included. The satisfactory result is reached by selecting different structures and variables of the neural networks. At last, an on-line upgrading scheme is proposed and has shown the great potential in simulation.
_____
ICNN97 Intelligent Estimation & Control
Session: ECP1 Paper Number: 442 Poster
A local connected neural oscillator network for sequential character segmentation
Hiroaki Kurokawa, Chun Ying Ho and Shinsaku Mori
Keywords: local connected neural oscillator network sequential character segmentation learning
Abstract:
This paper proposes local connected neural oscillators network using new local connection updates method for the sequential character segmentation. In the network, each neural oscillator is applied a learning method to control its phase and frequency. Since the learning method has an ability to control the phase adjustment of each neural oscillator, the information expression in the phase space is achievable. Furthermore, it is considered that the network has the ability to achieve real time image processing and information processing in a time series. The learning method is possible under the assumption that synapses have plasticity. However, since it is supposed that only the feedback synapse has plasticity, we can construct a network with high simplicity. Simulation results show the efficiency of this proposed network to realize the sequential character segmentation.