IEEJ Transactions on Electronics, Information and Systems
Online ISSN : 1348-8155
Print ISSN : 0385-4221
ISSN-L : 0385-4221
Volume 116 , Issue 7
Showing 1-22 articles out of 22 articles from the selected issue
  • Shigeki Nakauchi, Shiro Usui
    1996 Volume 116 Issue 7 Pages 727-733
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (1194K)
  • Takehisa Hara, Noboru Hiraiwa, Kenichi Hirotsu, Sadao Fukunage
    1996 Volume 116 Issue 7 Pages 734-740
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In this paper, the system of detecting corona discharges automatically with an artificial neural network is examined and a network which can distinguish between corona and noise patterns occurring in power cables is investigated. A feed-forward type of a neural network with three layers, i.e., input, hidden, and output layers is used. It is found that the network which learns only corona and no noise patterns does not show a good performance. This means that the network should learn both corona and noise patterns even for recognizing only corona discharges. The network which uses frequency spectra of the waveforms obtained by fast fourier transform (FFT) method as input patterns is also investigated. The network with FFT pretreatment is found to show a better performance than the one without FFT pretreatment.
    Download PDF (1718K)
  • Yuko Osana, Motonobu Hattori, Masafumi Hagiwara
    1996 Volume 116 Issue 7 Pages 741-747
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In this paper, we propose Chaotic Bidirectional Associative Memory (CBAM). Most of the conventional associative memory models can not deal with one-to-many associations because the stored common data cause superimposed patterns. Although only a few models enable one-to-many associations, they require manual controls or extra networks to separate the superimposed patterns. As a result, they have very complicated structures. In contrast, the structure of the proposed CBAM is very simple because it merely uses chaotic neurons in a part of the conventional Bidirectional Associative Memory (BAM). In the CBAM, each training pair is memorized together with its own contextual information. Since the chaotic neurons corresponding to the contextual information change their states by chaos, the one-to-many associations can be realized in the CBAM. A series of computer simulations shows the effectiveness of the proposed CBAM.
    Download PDF (1596K)
  • Makoto Kinouchi, Masafumi Hagiwara
    1996 Volume 116 Issue 7 Pages 748-754
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    To deal with temporal sequences is very important and difficult problem for applications of neural networks. In this paper, we propose a Multilayer Network using Complex neurons with local Feedback (MNCF). A complex neuron can keep previous information more easily than a real neuron because of the phase component. We derive simple learning algorithm based on the back-propagation for temporal sequences. It can be considered as a generalized original back-propagation. It is shown in computer simulations that the proposed network has better ability than the conventional real ones, including Elman's network.
    Download PDF (1053K)
  • Motonobu Hattori, Masafumi Hagiwara
    1996 Volume 116 Issue 7 Pages 755-761
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper proposes Intersection Learning algorithm for Bidirectional Associative Memory (ILBAM). The condition which guarantees the recall of all training data in the BAMs is given by a system of linear inequalities. In order to solve the system of linear inequalities, some learning algorithms using relaxation methods have been proposed. However, the weight renewal times of the algorithms depend on the correlation of training data. In this paper, we derive a relaxation method based on geometrical consideration and apply it to the learning of the BAMs. A number of computer simulations show the following effectiveness of the proposed ILBAM algorithm: (1) It can guarantee the recall of all training data. (2) It requires much less weight renewal times than the conventional methods. (3) It becomes more effective in case there are many training data to be stored. (4) It is insensitive to the correlation of training data. (5) It contributes to the noise reduction effect of the BAMs.
    Download PDF (1578K)
  • Kyoko Makino, Kazuo Nishimura, Hideki Hayashi, Masahiko Arai
    1996 Volume 116 Issue 7 Pages 762-768
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper proposes a new method to optimize the operation of a power transmission system, based on a structured neural network. Because the structured neural network is built by prewiring, the time to build the network is very short. Simulations of transmission loss reduction have been carried out. It has been shown that the optimization performance of the proposed method is satisfactory in practice.
    Download PDF (1374K)
  • Takeshi Iwasa, Noboru Morizumi, Sigeru Omatu, Kouji Fukui
    1996 Volume 116 Issue 7 Pages 769-775
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper considers a temperature control for a chemical plant which produces polythlene in a batch reactor. In this plant, the reaction in a tank is complex and has many nonlinear factors. The temperature control is realized by valves manipulation. Since the valve control has been performed by using a conventional PID controller, it generally needs much effort and time to tune the PID gains. Thus, a control sytem is preferable if we can tune PID gains without operator's decision under various conditions of the plant. For that purpose, we use the self-tuning neuro-PID control method. This method has the characteristic of tuning PID gains automatically by neural networks. First, we construct a simulation model of the plant. Regarding the simulation model as a true plant, we construct a self-tuning neuro-PID controller for the model in order to find the initial values. Using the tuning method and initial values, we performed the temperature control for the batch reactor for polythlene. From the experiment results, we show the effectiveness of the proposal algorithm to improve the control performance of the temperature in the batch reactor
    Download PDF (2080K)
  • Makoto Ohki, Hitoshi Miyata, Mikihiro Tanaka, Masaaki Ohkita
    1996 Volume 116 Issue 7 Pages 776-784
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper proposes to use membership functions (MSFs) in a form of piecewise linear functions for the fuzzy control rules. For determining a manipulated variable with an aid of fuzzy control so as to match a required output of a plant, tuning of fuzzy rules is needed. So far, this kind of tuning is performed by technique of trial and error. However, the tuning method needs much time for its tuning, and a suitable tuning cannot be always expected. So, we have tuned fuzzy rules based on a performance function in a neural network, where shapes of the MSF can be changed flexibly. Hence an expressiveness of the fuzzy reasoning is expanded. For its learning computation, a learning algorithm using coefficients modified by global searching (LACG) is derived. This LACG serves to avoid the local optimum in its numerical solution. An advantageous feature of the LACG is described with numerical examples.
    Download PDF (1142K)
  • Kazuo Kiguchi, Toshio Fukuda
    1996 Volume 116 Issue 7 Pages 785-793
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Robot manipulators are expected to perform sophisticated tasks in many fields. In order to perform sophisticated tasks, robot manipulators have to control both position and force simultaneously. Robots are also expected to work not only in limited environments, but also in unknown environments. However, conventional controllers have some problems to control force on unknown environments, because they do not have ability of adapting to the unknown environment. Furthermore, friction compensation is difficult, especially for an unknown environment, because the friction between the robot manipulator and the environment varies when the applied force to the environment changes. In this paper, neural networks and fuzzy-neural networks are applied to position/force hybrid control in order to solve these problems.
    The effectiveness of the proposed controller is verified with a 3DOF planar robot manipulator by computer simulations.
    Download PDF (1441K)
  • Kotaro Hirasawa, Masanao Obayashi, Hirofumi Fujita, Masaru Koga
    1996 Volume 116 Issue 7 Pages 794-801
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In this papaer, Universal Learning Network(U.L.N.) is proposed, which can be used as a fundamntal tool in modeling and control of large-scale complicated systems such as economic, social and living systems as well as industrial plants.
    The basic idea of U.L.N. is that most of the large scale complicated systems can be modeled by the network which consists of nonlinearly operated nodes and branches that may have arbitrary time delays including zero or minus ones. Therefore, U.L.N. can be applied to many kinds of systems which are difficult to be expressed as ordinary first order difference equations with one sampling time delay.
    In this sense, U.L.N. is a natural extention of recurrent neural network. It is also shown from simulation results of nonlinear identification that U.L.N. with arbitrary time delays can model nonlinear systems more efficiently than recurrent neural network.
    Download PDF (1297K)
  • Kazuhiro Kohara, Yoshimi Fukuhara, Yukihiro Nakamura
    1996 Volume 116 Issue 7 Pages 802-808
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    We investigate ways to use prior knowledge and network learning techniques to improve neural multivariate prediction ability. The prediction of daily stock prices was taken as an example of a complicated real-world problem. We make use of prior knowledge of stock price predictions and newspaper information on domestic and foreign events. Event-knowledge is extracted from newspaper headlines according to prior knowledge. We choose several economic indicators according to prior knowledge and input them together with event-knowledge into neural networks. Also used are selectively intensive learning techniques for improving the ability to predict large changes by neural networks. In the selective-presentation approach, the training data corresponding to large changes in the prediction-target time series are presented more often. In the selective-learning-rate approach, the learning rate for training data corresponding to small changes is reduced. The effectiveness of our approach is shown through experimental stock-price prediction.
    Download PDF (1424K)
  • Eitaro Aiyoshi, Atsushi Yoshikawa, Takuya Wakutsu
    1996 Volume 116 Issue 7 Pages 809-818
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    The Hopfield neural networks with continuous state transition have inefficient convergence in the neighborhoods of 0 or 1 as well as local convergence in their applications to the continuous versinos of 0-1 combinatorial optimization problems. Meanwhile, in the penalty approach to the constrained problems, it is feared that a lot of feasible combinatorial states satisfying the constraints corrupt into the local minima unconstrained problems. In order to settle these questions, “linked state transition neural networks with constraint satisfaction”, in which plural neurons transit simultaneously between discrete states so as to satisfy the constraints, are proposed from the standpoint of “discrete solutions to discrete problems by discrete neural networks”. In this paper, the linked state transitions in their applications to the traveling salesman problems realize automatical permuting procedures in a traveling route in the k-Opt method which is one of useful local search methods to combinatorial problems, and their numerical performance in experimental results are mentioned for relatively large-scale bench mark problems. These findings enhance practical values of the neural networks, and mediates between the fields of combinatorics and the neural networks closely.
    Download PDF (1133K)
  • Yoshifumi Arai, Shigeki Nakauchi, Shiro Usui
    1996 Volume 116 Issue 7 Pages 819-825
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    There exist several attempts to develop a system for device-independent color reproduction in color desktop publishing. However, the difference of a color appearance between the original and the reproduced colors occurs when illuminant light is changed. This is because conventional device-independent color matching is calibrated only by the colorimetric values under the standard illuminant such as D50, that is, it is based on the metamerism.
    In this study, we present a device-independent color reproduction method which can predict and correct the color shifts due to the illuminant changes. This method adopts the principal components (PCs) of the spectral reflectance as an intermediate color representation. We trained three layered neural networks to perform the transformation from CMY values of the proof printer to PCs, and the transformation from these PCs to C'M'Y' values of dye-sublimation printer. After the learning process, we evaluated the accuracy of the transformation from CMY to C'M'Y' through PCs by the trained networks. As a result, we found that our proposed method has an ability not only to predict but also to correct the color shifts due to the illuminant changes.
    Download PDF (1274K)
  • Mahito Fuji, Takayuki Ito, Sei Miyake
    1996 Volume 116 Issue 7 Pages 826-834
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    A neural network model with population coding for extracting binocular disparity is proposed. The population coding is expected to represent binocular disparity with hyperacuity. In order to solve the false target problem in stereo vision, some constraints, such as compatibility, uniqueness and smoothness, are effective. In general, the uniqueness constraint is realized by selecting an appropriate binocular disparity among some candidates. Accordingly, it is difficult that the uniqueness constraint is realized in a system with population coding.
    The presented neural network model realizes the uniqueness constraint and population coding simultaneously. The model also interpolates binocular disparities at pixels where features to extract binocular disparity do not exist. The model is based on physiological findings.
    It is shown by computer simulations that the model is able to extract almost correct binocular disparities in terms of center of gravity of distributed binocular disparities from random-dot stereogram forming a sphere and a plane, as well as two planes. In other words, the model is effective to extract binocular disparity from curved objects by means of population coding.
    Download PDF (3938K)
  • Noboru Masuda, Teruo Negishi
    1996 Volume 116 Issue 7 Pages 835-842
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper shows determination of magnetic bias for InSb semiconductive magnetoresistance elements including magnetic sensor.
    The semiconductive magnetoresistance element has two kinds of linear and square characteristics in magnetic resistance change bordering at a low and high applied magnetic flux strength. Also, the border characteristics show clearly experimental method.
    It needs to make clear between an influence for dispersion with the initial resistance and magnetic bias to the magnetoresistance elements, connecting to the series with a divided potential circuit. A magnet give the most suitable magnetic bias to the magnetoresistance elements included in a magnetic sensor is discussed.
    Download PDF (1725K)
  • Itaru Nagayama
    1996 Volume 116 Issue 7 Pages 843-848
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    A new approach for segmentation in characters overlapped a border is proposed. OCR processing for handwritten characters in account sheets or postal cards requires segmentation and elimination of a border to avoid the difficulty of character recognition. In order to avoid the difficulty, drop-out colored border for special scanner device or multi-step processing under the document analysis approach are used. When an input image of handwritten character overlapped a border is given, the border is segmented at first step, then contour is extracted as feature pattern. This paper presents a new method which simultaneously execute the eliminating a border overlapping on handwritten character and the extraction of character contuor. The method uses double threshold binarization and neuro- fuzzy approach. Binarization levels are decided by using neuro-fuzzy network according to the quality of images. It is shown that the proposed method is better than conventional methods about performance of eliminating the border and extracting of character contour. Availability of this method is indicated by applying it to handwritten postal character images obtained from postal matters.
    Download PDF (1266K)
  • Hirro Yamaura, Dawei Cai, Takato Zusi, Yasunari Shidama
    1996 Volume 116 Issue 7 Pages 849-857
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper describes a new gradient method for some nonlinear optimal problems with unknown parameters and final time. The algorithm is developed by applying the viewpoint of gradient method. By introducing a false time variable and a new function vector that is defined from the derivative of constraint function with respect to false time variable, a set of differential equations about the partial derivative variables of this function can be derived. Based on the above partial derivatives, a modification rule of control output, a false time parameter and unknown parameters in system is presented. If the condition of the modification rule is satisfied, the value of cost funciton will monotonyly reach a minimum. According to the modification rule, a numerical calculation algorithm is developed. To verify the performance of the algorithm, two simulation examples are provided. By comparing the simulation result with the theoretical resolution, it is cleard that proposed algorithm can provide a correct and useful means for nonlinear optimal control problems.
    Download PDF (1296K)
  • Chul-Hwan Kim, Woo-Gon Chung, Jong-Bum Lee
    1996 Volume 116 Issue 7 Pages 858-864
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper presents a fault identification system with the help of neural networks for underground cable transmission systems (UCTS), In order to get the data for transient phenomena in transmission systems we used the package EMTP which models the transient phenomena which is necessary information for fault type identification purpose. Data for various fault types for underground cable system were created and were used for training back-propagation neural networks. For operation of the proposed system a new data is used for testing for fair assessment of the designed system. Normalization of input data is adopted to expect more reliable learning in neural networks. A proper size of the neural network was found via trial and error method, a brute-force method. The system was tested with various fault distances and fault incidence angles and proved its reliability for various situations
    Download PDF (956K)
  • Masayoshi Kamiya, Hiroaki Ikeda
    1996 Volume 116 Issue 7 Pages 865-872
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper describes a simple optical wavelength-division-multiplexed (WDM) signal transmission system, consisting of a dual-channel-color LED assembly, a plastic optical fiber with a length of 5m, a dual-channel-color sensor assembly, a crosstalk elimination circuit, and other electronic circuits. This system transmits a pair of separate analog signals at frequencies of DC to 10 kHz from the transmitter to the receiver through the optical fiber without the use of expensive optical couplers. The performance was such that the distortion factors was below 2% and the crosstalk was below -40 dB. This system was experimentally applied to transmit data of mechanical vibration from a pair of vibration sensors attached to a printed circuit board under test to the monitoring instrument in order to separately readout the forced vibration and natural vibration. The experiment was successful.
    Download PDF (2101K)
  • Norio SASAKI, Terumitsu MITOMO, Takao SHIBUYA
    1996 Volume 116 Issue 7 Pages 873-874
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (896K)
  • Mitsuru Tsukamoto, Miho Kanemaki, Saori Nozaki, Ryoko Kobayashi
    1996 Volume 116 Issue 7 Pages 875-876
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (233K)
  • Takao Yamaguchi, Fumio Maehara
    1996 Volume 116 Issue 7 Pages 877-878
    Published: June 20, 1996
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (1647K)
feedback
Top