This paper presents a method for shapes modeling of 3-D objects using a B-spline surface model based on a projection expression. The surface model is locally defined on each patch that is effectively arranged depending on object shapes. In order to assure the continuity of the resultant surface, 4th order NURBS surfaces with embedding functions are employed. Consequently, this model can be considered as a hybrid model that is consist of polygon patches and NURBS surfaces. In order to estimate the surface function, a regularization problem is solved by an iterative algorithm. In the algorithm, since the domains where the surface has to be estimated are restricted to only the error patches, the computation time can be reduced. However, the accuracy of estimation is never inferior since a subdivision operation of NURBS is employed for the other domains. Through several experiments using real data of range image, the effectiveness of the proposed method has been confirmed.

抄録全体を表示

This paper presents a maximum likelihood (ML) identification and restoration method for noisy blurred images. The unitary discrete sine transform (DST) is employed to decouple the large order spatial state-space representation of the noisy blurred image into a bank of one-dimensional real state-space scalar subsystems. Assumed the noises are Gaussian distributed processes, the maximum likelihood estimation technique using the expectation maximization (EM) algorithm is developed to jointly identify the blurring functions, the image model parameters and the noise variances. In order to improve the computational efficiency, the conventional Kalman smoother is incorporated to give the estimates. The identification process also yields the estimates of transformed image data, from which the original image is restored by the inverse DST. The experimental results demonstrate the effectiveness of the proposed method.

抄録全体を表示

This paper presents experimental results of pattern (character and obstacle) recognition for controlling a mobile vehicle by adaptive resonance theory (ART) neural network. ART, a kind of competitive neural network, can self-organize and self-stabilize in response to complex input vectors. New memories are stored in such a fashion that existing ones are not forgotten of modified. The experiments are presented to show the ability of ART to control the mobile vehicle avoiding the obstacles. Finally, the experimental results and some future considerations are given.

抄録全体を表示

In this paper, we investigate the idea of using Artificial Neural Network (ANN) and Genetic Algorithms (GAs) to detect upright, frontal views of faces in complex gray-scale scenes. The first component of our system is a neural network that receives as input a 20x20 pixel region of the image, and generates an output ranging from 0 to 1, signing the presence or absence of a face. The neural network is trained with many face examples to learn the concept of face. GA is employed to search efficiently the input image by extracting sub-windows from the input image. The sub-window is sent to the NN, what is called the face filter, to see how closely it resembles a face. Fitness is given according to the output of the NN. Based on its fitness the sub-window is retained by the GA in subsequent steps. Experimental results show that significant speedup can be achieved by applying the proposed approach.

抄録全体を表示

The combustion process is a complex phenomenon. Under the gravitational field, the fuel is supplied by convection, however, under the micro gravitational field, the fuel is supplied by diffusion. So, it seems that the combustion process under micro gravity is different from one under gravity. In this paper, in order to study the combustion process under micro gravity, we propose the nonlinear stochastic reaction diffusion model. And the characteristics of the combustion process under micro gravity is analized through numerical simulations.

抄録全体を表示

Many spatio-temporal patterns such as target and spiral patterns in BZ reaction are observed in the various fields of engineering, including fluid and chemical engineering. The analysis of such patterns is one of important problems of engineering. In this paper, focusing our attention on the bacteria colony patterns, the stochastic modeling of behaviors of the bacteria called Bacillus subtilis and analysis of the created patterns are studied through numerical simulations.

抄録全体を表示

Many software systems have been produced under host-concentrated development environment. In such host-concentrated one, the progress of software development technologies has caused several issues that the increase of response time influences to software productivity and all of development management activities are suspended when the system is down. However, from the late 1980s, personal computers have been spread on our daily life instead of conventional mainframe machines, because the price and performance of personal computers have been extremely improved. At present, computer systems which aid the software development management have been also changing into UNIX workstations or personal computers to reduce the development cost. And Client/Server Systems (CSS's) have come into existence as a result of the progress of networking technology by UNIX systems. Such CSS's have been used more and more in the period of network computing. Thus, the software development environment has been changing into distributed one because of the progress of network computing technologies. In this paper, we introduce a software reliability growth model (SRGM) for distributed development environment by applying a mathematical technique of stochastic differential equations. We propose the estimation method for the unknown parameters of our model.

抄録全体を表示

In this study, we analyzed two merge-type transfer line systems by the Markov process at discrete times. The first system is the merge-type transfer line system of three parallel processes. That is, it is a manufacturing process in which three kinds of parts, A , B and C , are assembled in parallel and are completed at the same time. Next, the second system is the merge-type transfer line system with two stages in which three kinds of parts completed in the first stage are supplied to the next stage in series. Examples of numerical calculations were given for the obtained results.

抄録全体を表示

The authors previously proposed an on-line pipe length measurement for straight and curved pipes with closed ends using stationary waves which were generated in the pipes. The system adopted several modes of the stationary wave as state variables of a linear dynamic system and reduced the length measurement to a parameter estimation problem, and thereby realized a highly accurate pipe length measurement. This paper presents a new length measurement system for straight pipes with open ends. Experiments will show that a highly accurate pipe length measurement is realized without any trouble on the annoying open end compensation.

抄録全体を表示

A method is presented for clustering signal space generated by information intensive decentralized systems. By identifying signal space with locally square summable functions, a universal dynamics is designed for successively approximating feature parameters in Fourier images. The approximation sequence is not monotone. However, we can design aggregation scheme for ordering Fourier images governed by feature parameters to be detected. The clustering rule was verified through simulation studies.

抄録全体を表示

The hybrid active noise control (ANC) system consists of the combination of feedforward and feedback type ANCs. The filtered-X LMS algorithm is used to adapt the controllers. The averaging method is applied to the hybrid ANC system with the frequency domain expression to derive a stationary point of the adaptive algorithm. Then, it is found that though the two adaptive filters of both ANCs are operating simultaneously, the stationary point of each adaptive filter is same with that for the case where each ANC system is operating separately. Then the local convergence condition is derived. Finally, the convergence condition is explicitly examined for a simple case and its validity is shown by some simulations.

抄録全体を表示

A novel method of subspace identification via distribution-based approach is proposed for a class of time-varying, continuous-time stochastic systems. Efficacy of the method is demonstrated by simulations.

抄録全体を表示

This paper is concerned with an application study of model-based fault detection method to a ship propulsion system. When modeling the object system, Quasi-ARMAX model with multi-model form is used. In the model, the system non-linearity is incorporated into model parameters by using non-linear non-parametric models (NNMs). Kullback discrimination Information (KDI) is introduced as fault detection index to evaluate the distortion in identified model, which is caused by a fault. The effectiveness of the method is verified through simulation studies on the ship propulsion system.

抄録全体を表示

In this paper, we will expand and generalize the orthogonal functions as basis functions for dynamical system representations. The orthogonal functions can be considered as generalizations of, for example, the pulse functions, Laguerre functions, and Kautz functions. A least-squares identification method is studied that estimates a finite number of expansion coefficients in the series expansion of a transfer function, where the expansion is in terms of recently introduced generalized orthogonal functions. The analysis is based on the result that the corresponding linear regression normal equations have a block Toeplitz structure. It is shown how we can exploit a block Toeplitz structure to increase the speed of convergence in a series expansion.

抄録全体を表示

The determination of model order is an important problem in system identification. It is often determined by statistical test and information criteria. Recently a test method based on Lower-Diagonal-Upper triangular(LDU) decomposition was proposed in [1]. In this method the covariance matrix of input-output data is decomposed by LDU decomposition. It is well known that the rank of a covariance matrix equals the number of nonzero elements in the diagonal pivot matrix in the decomposition, and also equals model order. Thus the rank test in [1] is based on the asymptotic distributions of the pivots. But in some cases, the rank test can not provide the correct order of the model. Our purpose is to improve the accuracy of determination of model order using Minimum AIC Estimate(MAICE) and LDU decomposition.

抄録全体を表示

This paper presents a new method for the signal separation of interleaved pulse trains consisting of several independent signal sources, some of which possibly exist for only a short period of time. The separation is based on the differences in the pulse repetition intervals (PRIs) of the signal sources. To detect each signal source, we derive a map from the input signal to a function of two variables: time and PRI, which we call the PRI map. By taking the absolute value of the PRI map, we obtain a short-time PRI spectrum, from which we can use thresholding to estimate the number of signal sources, their time duration, and the PRIs. The PRI map is an extension of the PRI transform, which was proposed by the author previously. The construction of the PRI map is similar to that of a wavelet transform. The difference is that the PRI map is a nonlinear transform, while the wavelet transform is a linear one. Simulation results are presented on the performance of the signal separation of interleaved pulse trains using the PRI map.

抄録全体を表示

In this paper, a unit root of cumulative call rate is investigated with Monte Carlo experiments. call rate is shown to be I(0) with a linear trend in [1970, 1998]. Vector Error Correction model consists of only I(1) variables in a frame work of Johansen's cointegration analysis, and stationary variables like a call rate are often added to the RHS of the resultant VEC model without considering cointegration. However, if cumulative call rate is I(1), this variable should be contained in cointegration analysis. Since call rate has a linear trend, a unit root of cumulative call rate should be checked around a quadratic trend. MacKinnon's critical values for unit root are extended to a quadratic trend case with Monte Carlo simulation experiments It is shown that cumulative call rate has a unit root with a quadratic trend. Cointegration properties are investigated among cumulative call rate, GDP, money supply and bank loan and VEC model is constructed.

抄録全体を表示

We have proposed a speech-displaying system [2] [7] based on the spectral estimation of speech signals. The system, named KanNon system, can display a speaker's individual difference and feeling such that it will be useful for a communication tool for a hardness-of-hearing person. In this paper, we show identification algorithms of two time series models with white noise and impulse inputs for spectral estimation of vowels and consonants of speech signals. Further, we apply the likelihood ratio test to distinguish vowel from consonant of a frame of speech data.

抄録全体を表示

This paper presents an Adaptive Neural Network Controller (ANNC) to control the speed of a photovoltaic (PV) supplied, separately excited dc motor driving a centrifugal pump load. The ANNC is also used to track the maximum power point (MPP) of the PV generator by controlling the converter duty ratio D. Since the controller has two tasks, then two objective functions must be existed. But to make the design of the controller easy and simple, the objective function related to the MPP is converted as a constrained to the other objective function by approximating some parts in the system. An Adaptive Neural Network Identifier (ANNI) is used to compute the system Jacobian, which is used to update the weights and biases of the ANNC. Both of the ANNC and the ANNI are trained on line using the BP algorithm with adaptive learning rate. Simulation results of the motor speed, current and voltage at steady sate operation are obtained for a step change in the insolation and reference speed (ω_ref) by two different cases depending on the calculation of the system Jacobian.

抄録全体を表示

In this paper, a new hybrid supervised learning control scheme is presented for continuous stirred tank reactor (CSTR) systems. The control system works at distal supervised learning control mode or extreme control mode, respectively. Neural networks controller is trained by using extreme control signal as supervised signal when system works at extreme control mode and there is no supervised signal for training neural networks controller when system works at distal control mode. The precise mathematical model of the CSTR is not necessary for designing hybrid supervised control system. Simulation results show the presented control system with satisfactory dynamic and stationary performance and strong adaptability.

抄録全体を表示

In this paper, some popular approaches to combine neural networks and fuzzy logic systems are briefly surveyed. Then, a novel combination model is presented for synthetic integration of rainfall. The presented model consists of four network layers: input layer, membership function constructing layer, inference layer and defuzzification layer. The combination model is trained using three kinds of forecasted rainfall data, produced by gradual regression method, periodic analysis plus multi-layer method and model output statistics method, as inputs and real rainfall as outputs in Zhejiang province from 1980 to 1997. Then the trained model is employed to integrate/forecast the rainfall of Zhejiang province from 1998 to 2000. Integration results show that the presented model can achieve satisfactory forecast performance.

抄録全体を表示

This paper is concerned with considering an optimal strategy of controlling traffic signals by introducing a new concept of flow stability for multiple crossings as well as a single crossing. Compared with a fixed-type signal control strategy, an optimal strategy is proved to be the more effective under several computational examples.

抄録全体を表示

Noise-induced chaotic bifurcations of random attractors arising in a Duffing oscillator with harmonic and random excitations are investigated. We first follow the standard pullback procedure to obtain random invariant measures and then intuitively construct Poincaré maps of them. The results on the stability and fractal analysis show that the noise-induced chaotic behavior which is not strange is generated from strange but non-chaotic invariant measures.

抄録全体を表示

In order to visualize 3-D (three-dimensional) distributions of gas temperature and flow in chambers, an algorithm for CT (computed tomography) based on the MAP (maximum a posteriori) estimation has been developed. The algorithm can be regard as the extended Kalman filter having local iterations, to cope with non-linear observation processes of acoustic time-of-flights. In this report, derivations of the proposed algorithm as well as the performance of its application are introduced.

抄録全体を表示

Orthogonal type forcefree control by considering following locus is proposed for industrial articulated robot arm. The tip of the robot arm is moved on an objective locus according to the exerted force without gravity and friction. Contouring work of industrial articulated robot arm is, therefore, carried out with forcefree control. Effectiveness of the orthogonal type forcefree control by considering following locus was guaranteed by simuslation and experimental work on a two-degree-of-freedom articulated robot arm. Applications of orthogonal type forcefree control by considering following locus was also discussed.

抄録全体を表示

We consider the existence of the local strong solutions of the zero Dirichlet problems of the coupled Navier-Stokes equations which govern the incompressible magnetic fluid in three dimensional bounded domain. We derive the local existence in time of a unique strong solution for boldmath u₀ belonging in IH^1/2. The proofs have been shown to apply the contraction mapping theorem by using the theory of the sub-differential operators.

抄録全体を表示

The problem of constructing optimal curves for given set of data at discrete data points is considered. Both equally-spaced and non equally-spaced data points are treated. The curves are constituted by using B-splines as basis functions, namely as weighted sum of shifted B-splines of degree k. It is then shown that an optimal approximation can be solved without any boundary conditions, wherein explicit solution formulas are presented. A problem of optimal interpolation is also considered in parallel. Some numerical examples are included to demonstrate the usefulness of the results.

抄録全体を表示

Support Vector Machines (SVMs) have become a subject of intensive study in statistical learning theory. They have been applied to successfully to classification problems and recently extended to regression problems. Support vector machines for regression problems is called Support Vector Regression (SVR). In this paper, a brief introduction to SVR is presented and then a new system identification method based on SVR is proposed for linear in parameter models. The effectiveness of the proposed method is examined through numerical examples.

抄録全体を表示

A new adaptive predistortion method for compensation of unknown linear and nonlinear distortion in a high power amplifier (HPA) is investigated by a databased approach using the support vector regression (SVR). The proposed approach can construct an inverse model of the nonlinear HPA by learning procedure using training input-output data samples. An on-line adaptive algorithm for updating the parameters of the SVR machines is also given to reduce the computational burden of the batch processed optimization. By comparing with an ordinary look-up table scheme, it is shown that the proposed data-based approach can attain improved suppression of the power spectrum of the amplifier output outside the input frequency band caused by the nonlinear distortions.

抄録全体を表示

A model-set identification algorithm is proposed in a probabilistic framework based on the leave-one-out technique. It provides a nominal model and a bound of its uncertainty for a provided plant assuming that the effect of the past inputs decays with a known bound. Since it does not require further assumptions on the true plant dynamics or on the noise, a risk to make inappropriate assumptions is small. The number of assumptions is shown to be minimum in the sense that identification is impossible after removing the assumption made here. An algorithm similar to the proposed one is constructed based on a mixing property. A simple plant is identified by means of the proposed algorithm for illustration.

抄録全体を表示

In this paper, an optimal model predictive control (MPC) method is proposed for a class of high order plants which are modeled by first-order linear systems with time-delay. First, the mathematical model of the plant is described by the CARIMA model. Secondly, based on the state space model the optimal MPC is obtained incorporated with Kalman predictor/filter to get the present and future estimates of the state. Finally, a simulation example is given for comparing our method with the generalized predictive control method.

抄録全体を表示

We consider the term structure modeling by using a appropriate stochastic parabolic system with boundary noises. After finding a sufficient condition for the no arbitrage opportunity, we solve the mean-variance optimal control problem in the incomplete market.

抄録全体を表示

Controlled variables in multivariable systems interact each other in general, therefore, it is necessary to remove their interaction sufficiently for the design of control systems. A cross controller was used for removing the interaction thus far mainly, however, it was complicated and the controller was high-order. On the other hand, the precompensator by the inverse Nyquist array method was also used for noninteraction of the controlled variables. However it is difficult to remove the interaction of the controlled system, which has very strong interaction, when the precompensator is used only one. This paper proposes the series and/or parallel connections of the precompensators for noninteracting PID control.

抄録全体を表示

The problem of closed-loop identification and control with finite impulse response (FIR) models has been considered in this paper. FIR models are easy to analyze and the coefficients are easily determined by solving a linear equation even when the optimality is required. Several methods for the closed-loop identification and designing the controller are compared each other to select suitable ones for FIR models. Numerical examples are given to illustrate the comparison.

抄録全体を表示

In this paper, we present a recursive algorithm for a subspace identification. Using the results of the subspace extraction via Schur complement (SES) approach, we derive a recursive formula of an error covariance matrix in subspace methods. This recursive procedure requires the eigenvalue decomposition (EVD) of the covariance matrix at each step. Therefore we introduce the rank-k modification algorithm with the spectrum slicing theory and achive to update the EVD recursively. The algorithm improves a computational complexity from O(N³) to O(N²k) class where N x N is the size of the covariance matrix.

抄録全体を表示

We derive two blind identification algorithms for the FIR channel based on oblique projections. The one is a deterministic method that recovers the channel parameters completely under the noise-free condition. The other one is a second-order statistics-based method that eliminates the noise effect by means of the SVD technique, followed by application of the deterministic method. Some numerical results are included.

抄録全体を表示

From the viewpoint of stochastic inverse problem, identification of transfer functions in feedback systems has been examined by the previous papers[1][2], and applied to EEG analysis[3]. Some transfer functions between measurement regions are identified by our method of inverse problem in a stochastic feedback system. However, there are still open problems of our method for stochastic inverse problem: One problem is that there occurs a chaos-like behavior in matrix Riccati equation and it affects a property of identified innovation model. The other is a filtering effect on transfer functions for pre-processing of high pass filter. These problems are discussed from the viewpoint of statistical inverse problem.

抄録全体を表示

In this paper, we analyze two closed loop identification methods by simulation studies. The one is the two-stage prediction error method based on the least-squares estimate. The other is a subspace identification method based on the orthogonal decomposition. The bootstrap method is applied to estimate the standard error of identified models. Simulation results are also included to show that the accurate estimates of standard deviation of the estimated frequency responses are obtained.

抄録全体を表示

We consider a sensor allocation problem for the Kalman-Bucy filter within an information theoretic framework. For the signal and the observation of Kalman-Bucy filter, the mutual information between them is determined by the power of the signal component in the innovations process, and we cannot make the mutual information larger without increasing the power of this term in the innovations process. Under a constraint that the mean square power of the term takes a preassigned value, a set of equations which were derived in our previous paper is applied obtain a recursive algorithm to compute the optimal gain matrix for the sensors which minimizes the least-squares estimation error.

抄録全体を表示

The carrier phase ambiguity resolution is vital for relative GPS (Global Positioning System) positioning using carrier phase measurement. In this paper, a practical GPS positioning algorithm including on the fly integer ambiguity resolution is presented. In the algorithm the Kalman filter with re-configuration scheme of the state vector is efficiently utilized to resolve integer ambiguities and continuously provide a precise positioning even if there are variation of visible GPS satellites and cycle slips. The experimental results of positioning using real receiver data are shown.

抄録全体を表示

A state estimation problem is considered for linear systems driven by white noise. The observation noise is assumed to be a second-order process fully unknown except for the total mean power. The Kalman-Bucy filter is adopted here as state estimator. The main result is a necessary and sufficient condition, stated in terms of the Hamiltonian matrix, for the Kalman-Bucy filter to guarantee error covariance less than that performed under the nominal condition where the observation noise is white.

抄録全体を表示

The separate-bias estimation is well known for the unknown constant bias case. In this paper, the modified separate-bias Kalman filter is investigated for a continuous-time dynamical linear system in the presence of randomly time-varying bias. The certain general conditions are derived under which the modified separate-bias estimator is equivalent to the augmented state Kalman filter. However it can be shown that the resultant conditions for the continuous-time linear system with a stochastic bias are more complex than the discrete-time situation, and that the optimal separate-bias estimation may not provide any computational advantage.

抄録全体を表示

In this paper we propose a new fault detection method for robust control systems. Our technique is based on the singular value analysis of system matrices which is intermittently updated by recursive subspace method. A threshold of system fault has been derived with respect to robust stability of the closed loop system independently of the measurement noise. The numerical examples demonstrate the validity of the proposed method.

抄録全体を表示

This paper is concerned with a design of a collision avoidance robust controller, based on a genetic algorithm. The robust safety controller with additional integral type forces the plant output of a linear multivariable time system to converge to that of a reference model without any collision even if there are variations of the the plant parameters. Here it is applied to a robot manipulator system.

抄録全体を表示

This paper discusses a design method of a robust adaptive stabilizing controller for nonlinear systems with uncertainty. The control scheme proposed in [5] for nonlinear systems with relative degree 1 is extended to nonlinear systems with relative degree more than 1. The robust adaptive stabilizing controller is obtained using the control scheme in [5] recursively.

抄録全体を表示