電気学会論文誌C(電子・情報・システム部門誌)
Online ISSN : 1348-8155
Print ISSN : 0385-4221
ISSN-L : 0385-4221
114 巻, 11 号
選択された号の論文の26件中1~26を表示しています
  • 薦田 憲久
    1994 年 114 巻 11 号 p. 1059
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
  • 岩佐 健司, 森住 昇, 大松 繁, 服部 洋文, 田中 明志, 大西 光義
    1994 年 114 巻 11 号 p. 1060-1066
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper, we deal with a pressure control problem of a chemical plant which produces soda and chloride by electrolysis of salt water. In order to reduce the electrical cost, this plant adopts a switching method of electric current twice a day, which causes fluctuations of pressure inside electrolytic tanks. To stabilize the pressure, PID controllers have been devised. The purpose of this paper is to determine suitable PID gains of the controller in order to achieve above object. First, we construct an adaptive controller for the plant based on the generalized minimum-variance control (MVC). Unknown PID gains are estimated by the least squares method such that PID controller can approximate an output of the generalized MVC as much as possible. Finally, pressure control results for the plant generating chloride have been illustrated to show the effectiveness of the present approach.
  • 松村 雅史, 新川 拓也, 清水 公治, 橋本 泰嗣, 森田 龍彌
    1994 年 114 巻 11 号 p. 1067-1072
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    This paper deals with the measurement of 3D shapes of vocal tract, dental crown and nasal cavity using the magnetic resonance imaging (MRI). 3D MR images of the vocal tract and the nasal cavity during Japanese vowels and consonant /s/ productions were obtained. MR images of dental crown that contain a small amount of water were obtained using a dental crown plate. A computer algorithm for boundary tracing of cross sections of the vocal tract and the nasal cavity from the 3D MR image was proposed. 3D shapes of the vocal tract and nasal cavity were obtained using the present algorithm.
  • 松村 雅史, 木村 和生, 吉野 勝美, 舘村 卓, 和田 健
    1994 年 114 巻 11 号 p. 1073-1077
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    This paper deals with the measurement of palato-lingual contact stress and pattern, during consonant productions using a force sensor mounted palatal plate. A force sensor is composed of a strain gauge and a fixed stainless beam, mounted together into a palatal plate, and attached to the hard palate. The palato-lingual contact stress is concentrated in the center of the fixed beam by a protuberance on the fixed beam. When the tongue touches the force sensor, the palato-lingual contact stress is detected from the strain of fixed beam. We have developed a prototype system, which consists of a five-force-sensor mounted palatal plate (thickness: lmm) and a multi-channel strain-meter. The five force sensors are placed approximately at a 6mm interval along the midsagittal line of the hard palate. The most anterior sensor is located 5mm posterior to the alveolar ridge. Output of the force sensor was proportional to the stress applied to the fixed beam and had on hysteresis. Error by the interference among the sensors was less than 5.4%. The palato-lingual contact stress and pattern for an adult male were measured during the productions of consonant/t, /d, /n/ using the present system.
  • 樋笠 博正, 松村 茂憲, 大松 繁
    1994 年 114 巻 11 号 p. 1078-1083
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    We have devised an electric vehicle (EV) testing system, by which EV driving on road can be simulated in laboratory. This system includes some nonlinear elements so that the modern control theory could not be applied directly. Thus, a PID controller has been used in this system, but the PID gains have been tuned by trial and error.
    In this paper, the self-tuning PI-type neuro-control (NC) is applied to speed control of EV testing system. PI controller has been included in the EV testing system. Thus, only proportional band (PB) and/or integral time (TI) can be tuned. First, PB and TI are tuned by a layered neural network from maximum set points of the controller as initial values. In this case, the device is operated on no load, since it can not be operated normally by such values. After that, the tuned PB and TI are applied to operating on load. Moreover, PB and TI are tuned on load operation from various initial values. Experimental results has been illustrated to show effectiveness of the presented approach.
  • 稲葉 裕道, 加藤 和範, 赤澤 堅造
    1994 年 114 巻 11 号 p. 1084-1089
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    A self-producing classification method applicable to a large amount of human faces is proposed. Feature patterns obtained from face images distribute continuously in the pattern space because of various changes of face in size, angle, position, lighting, etc. Categories are self-produced with face images given successively, and the face image is classified through multi-step processing. When an input image is given, it is classified to a large group based on the degree of belonging to it, where the projection on base vectors of KL expantion is used. And then it is classified to a small group based on KL coefficients. Finally identification is executed by human, by comparing the input image with images of the classified small groups. If the input images are classified to none of the large groups, they are temporally stored, and when the number of the stored images becomes sufficiently large, a new set of category including both large and small groups is created self-producingly. Availability of this method is indicated by applying it to 525 face images obtained from 105 persons.
  • 奥野 竜平, 吉田 正樹, 赤澤 堅造
    1994 年 114 巻 11 号 p. 1090-1094
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    The purpose of this study was to develop a new type of myoelectric prosthetic hand, which simulated dynamic properties of the neuromuscular control system of human hand. This prosthetic hand consisted of a processing unit of surface electromyogram (EMG), a digital servo system of DC motor and a one-degree-of-freedom mechanical hand. EMG signal was used as control signal for opening and closing of the prosthetic hand. This prosthetic hand mimicked the property that both muscle viscoelasticity and gain of the stretch reflex vary linearly with the activity of muscle. That is to say, stiffness of the prosthetic hand was designed to change linearly with the amplitude of IEMG (rectified and smoothed surface EMG).
    Usefulness of this prosthetic hand was shown in myoelectric control experiments executed by normal subjects and amputate. Both the finger angle and the stiffness of the prosthetic hand were voluntarily controlled with the surface EMG by these subjects, and soft objects were grasped smoothly.
    Preliminary experiments on sensory feedback was done in such a way that magnitude of the sound varied increased with an increase in force applied to the fingers. Usefulness of the sensory feedback of force was indicated by executing the grasp of soft object.
  • 京兼 純, 吉野 勝美
    1994 年 114 巻 11 号 p. 1095-1100
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    Recently organic thin films have been widely used as electrical and electronic elements, sensors, coatings, protectors, lithographic layers, etc. Among various methods for preparation of thin films, an evaporation technique under vacuum is one of the most simple and convenient methods. In this study we applied a new technique, ion-beam-assisted deposition, utilizing ion beam irradiation during the preparation of the organic thin films by evaporation.
    As an example, conducting organic thin films of charge transfer complex have been prepared on porous oxide Al plates by means of evaporation under ion beam irradiation. A solid electrytic capacitor with a large capacitance per unit area and a good frequency response of impedance can be fabricated using organic thin films obtained with this method.
  • 西村 卓也, 堀口 和巳, 長田 朗, 平城 久雄
    1994 年 114 巻 11 号 p. 1101-1108
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper we present a practical design method for MIMO linear discrete-time systems based on two-degrees-of-freedom compensation scheme. We aim to design the control system satisfying the following specifications: [1] For reference inputs and outputs relations, the taransfer function matirix from refer- rences to outputs is diagonalized. [2] For reference inputs, the outputs are settled in finite steps, and the transient responses of the output errors and the inputs are controlled optimally. [3] For disturbances, the outputs are settled to zero in finite steps, and the transient responses of the output errors and the inputs are also controlled optimally. [4] The outputs are ripple-free after the settling time. The designed system turns out to be a decoupled disturbance suppressed ripple-free dead-beat control system with tran sient re-sponses of inputs and outputs considered. The design problem is reduced to the linear optimization problem and the solution is easily obtained.
  • 森本 茂雄, 大橋 真和, 武田 洋次
    1994 年 114 巻 11 号 p. 1109-1115
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    In the industrial motor drive systems such as industrial plants and industrial robots, a torsional vibration is often generated because of the elastic elements in torque transmission. This vibration makes it difficult to achieve quick responses of speed and may result in damage to the plant. Such system is simply modelled as a two-mass mechanical system.
    In this paper, H control theory is applied to design the speed controller for the two-mass system. This controller determines the closed-loop characteristics such as the suppression of torsional vibration and the rejection of torque disturbance. Moreover, the two-degrees-of-freedom controller, which includes the above H controller, is designed in order to improve the tracking characteristics for the commanded speed. In the TDOF system, the performances of the disturbance rejection and the tracking can be designed independently. The control performances are examined in detail by the computer simulations. As the designed controller is usually high order controller, the order reduction of controller is also discussed for the practical applications. From several examinations, it is clarified that the proposed speed control system is useful for the two-mass system.
  • 新井 重伸, 中村 徳明, 今西 茂
    1994 年 114 巻 11 号 p. 1116-1121
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    Many applicatoins of error-detecting and -correcting have been useful for communications and computer systems. It is similary important that ternary error-detecting and -correcting enhance reliability in the ternary information processing systems.
    We present a probabilistic burst-error-correcting scheme based on interleaving method with ternary cyclic codes. It is also shown that this method is superior to the conventional methods in point of the ability of the burst error correction.
  • 松坂 晃太郎, 隈元 昭
    1994 年 114 巻 11 号 p. 1122-1127
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    Any recent application of an expert system requires an effective way to build a very large-scale knowledge base. An on-line massive document, such as a corpus, has a potential possibility to be utilized in automatically preparing such knowledge base. Provided that a natural language interface is available, a user can inquire to the acquired knowledge for a specific topic he wants to know in an interactive way.
    This paper first makes investigation on semantic structures required in such a corpus oriented knowledge inquiry system focusing on plan-structure and LFG representation. It then proposes a new knowledge representation framework called a PFT-network. This network was introduced to effectively acquire knowledge from corpus with reduced domain knowledge for both the input corpus and user's utterance. Experimental results of extracting knowledge are given and it is demonstrated that the presented approach suggests the feasibility in knowledge acquisition in view of understanding input sentences so that inter-relationship of knowledge is reflected in the obtained knowledge.
  • 曽我 修治, 大川 剛直, 薦田 憲久
    1994 年 114 巻 11 号 p. 1128-1133
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    To obtain appropriate solutions in business information systems planning, it is necessary to perform the problem analysis elaborately. A structural model which represents the causal relations of problems using a directed graph is one of the effective techniques for the problem analysis. However, it is very difficult for novice system planners to make a structural model which has satisfying quality, since advanced skills, such as acquiring and selecting items, relating the items and so forth, are needed for structural modeling.
    In this paper, we propose the support method of structural modeling by providing useful information from knowledge base for user. In this method, to extract serviceable knowledge from cases of structural model, nodes of structural model are divided into four parts. From the knowledge base which is constructed by this dividing techniques, effective information for getting idea of nodes is retrieved and provided timely for user.
    We have developed a support system for structural modeling based on the method mentioned above. Using this system in structural modeling, user can hit on a high quality idea of nodes efficiently and ability which is required in structural modeling can be reduced.
  • 廣田 敦志, 瀬尾 和宏, 大澤 靖治, 荒井 健次
    1994 年 114 巻 11 号 p. 1134-1140
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    Nowadays the anxiety about drying up of fossil fuels and the global environmental problems such as greenhouse effect by CO2 become serious. Therefore the use of renewal energy, which exists inexhaustibly and does not emit waste matters, is paid attention. Of all natural energy, wind power generation and photovoltaic power generation are most practical in terms of technology and cost. Because the output power of natural energy generation, however, varies with change of weather condition, a hybrid generation system of natural energy and diesel power is considered.
    Though it is intended that the hybrid generation system compensates the power variation of natural energy generation by diesel generation, a frequency deviation beyond an allowable limit may be caused by fast power variation of natural power generation. We have focused on the wind/diesel hybrid power system and proposed to install a superconducting magnetic energy storage (SMES) to suppress frequency deviation.
    In this paper, we represent the hybrid power system that uses a PI controller to control the SMES and propose to apply pole assignment method to design PI controller. Furthermore, we examine the effect of proposed method by analyzing system eigenvalues and performing transient calculation.
  • 大川 剛直, 畑 慎也, 薦田 憲久
    1994 年 114 巻 11 号 p. 1141-1147
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper, we propose a new qualitative simulation method for envisioning behavior of a system modeled by a structural model. The structural model consists of qualitative causal relations between several factors in the system. It offers, however, neither useful constraint for pruning unreasonable situations nor characteristic points which correspond to the landmarks, because the causal relations are not substantially quantitative. For the purpose of reducing ambiguities caused by lack of quantitative information, we introduce a notion of time scaling and several typical patterns, which specify primitive behavior in equilibrium. The proposed method consists of the following processes. 1) The target structural model is divided to sub-models based on time scale. 2) The equilibrium in each sub-model is evaluated using typical patterns. 3) The results for all sub-models are integrated.
  • 田口 耕造, 金端 利典, 中西 恒彦, 溝尻 勲
    1994 年 114 巻 11 号 p. 1148-1153
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    This paper deals with extraction method of moving objects using a temporal linear filter and digital subtraction. The temporal linear filter estimates in sequence background image that changes slowly according to environment, and the temporal linear filter needs inter frame operation but dose not need inter pixel operation in the same frame. This makes processing faster and simple. A moving object has the higher velocity, its spectrum band becomes the wider compared with the spectrum of background image. We are able to regard the spectrum of moving object as white noise. So, we may compose the temporal linear filter as the matched lowpass filter. As a result, we are able to extract moving objects by subtracting estimated background image from input image.
    At the end of this paper, we show usefullness of this method to extract moving vehicles.
  • 吉川 仲次, 田口 耕造, 中西 恒彦
    1994 年 114 巻 11 号 p. 1154-1159
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    Though the vehicle flow measurement on the expressways is done using a loop coil sensor or an ultra sonic sensor, the system construction has many difficulties such as shutoff of traffic flow and requirement of many funds, these detecting method are not useful for automobile traffic flow survey. For the practical survey in recent days, the standard video systems are employed because setting of them on arbitrary point is very easy, but detecting accuracy is not high enough. Authors developed the detecting system which function is real time processing and high accuracy using usual video system. The recognition method of the vehicles and the scheme of system construction method are reported in this paper.
    The recognition of a vehicle in this system is achieved on the spacial Fourier spectrum for the finite part of video picture, and the real time processing is done by substitute transformation for Fourier transformation. This system construction requires only simple digital signal processor and small size personal computer.
  • 土井 滋貴, 長井 一郎, 佐久間 崇, 高橋 晴雄
    1994 年 114 巻 11 号 p. 1160-1165
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    The body movement while sleeping is said to be closely related with the sleeping-states, and then the sleeping-states may conversely enable to be judged from the body movement. The sleeping-states in human being are generously divided into three states: awakening state, REM sleeping state and NON-REM sleeping state.
    In this paper, a method to decide the sleeping-states from the body movement while sleeping is proposed by using an infrared sensor and the Neural Network. The infrared sensor detects non-contactly the frequency and amplitude of body movements while sleeping, and the Neural Network with the feedback structure decides the sleeping-states by calculating the movement data patterns.
  • 乾 誠貴, 亀井 且有, 井上 和夫
    1994 年 114 巻 11 号 p. 1166-1171
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    A Clustering is to classify data set into some clusters that are subsets with clear borderlines. Accordingly, data are expressed in binary logic, which means a data clearly belongs to a cluster, or not. In a fuzzy clustering, data are expressed in non-binary logic, which means a data is allowed to belong to some clusters with grade [0, 1]. Fuzzy c-Means (FCM) algorithm is one of the most famous fuzzy clustering methods, and it has two problems: [1] Computational time is extremely long because of computational iteration; [2] The number of clusters should be preset before the computation. The first problem will be solved because the development of computer in recent is quite remarkable. On the other hand, the second one will still remain unsolving.
    In this paper, we propose an advanced FCM algorithm with cluster number estimation, which is named FCM*, and show an example of computational results. Additionally, we also show the validity of the estimated cluster number by FCM* comparing with that by DB index.
  • 今川 剛, 森 隆知, 亀井 且有, 井上 和夫
    1994 年 114 巻 11 号 p. 1172-1178
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    This paper describes a navigation method for an indoor mobile robot using an active vision system. The robot with an active camera and an odometer is navigated based on a map of landmarks. The camera is controlled to take a landmark around it. Mobile robot's location is calculated based on the location of the landmark on the map, the coordinate of the landmark in the image taken by the camera and the yaw angel of the camera. The corridor, which is our navigation environment, is divided into two areas, the corner area and the straight area. In the corner area, the corner point is chosen as the landmark because it exists in the center of the turn when a mobile robot turns. In the straight area, doors are chosen as the landmark because some doors exist in this area and have a simple shape. In this area, the robot chooses one among the doors as a landmark according to the least estimated distance error between itself and the doors. The navigation results using these methods are shown, too.
  • 溝尻 勲, 新野 俊之, 角田 広樹, 杉本 哲哉
    1994 年 114 巻 11 号 p. 1179-1184
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    This paper deals with classifying and discriminating normal from hypertrophic cardiomyopathy (HCM) using echocardiogram.
    Normal myocardial fiber orientation has three-layered appearance, so we can observe the three-layered appearance on echocardiogram as the strength of ultrasonic echo efected by myocardial fiber orientation. But HCM myocardial fiber orientation is disturbed, so we can't observe the three-layered appearance.
    Thus we consider that the three-layered appearance of myocardium effects echocardiogram. We classify and discriminate normal from HCM. Region of interest is setted on the interventricular septum of the echocardiogram apical four chamber view. Discriminant analysis is applied to the normalized echocardiogram using Thebycheff polynomial coefficient as discriminant vectors. Then, we obtain overall determination from local classified and discriminated results, as we use correlation ratio in order to avoid influence of degradation of echocardiogram. Finally, we show the usefullness of this method for some examples.
  • 増田 達也, 太田 俊彦
    1994 年 114 巻 11 号 p. 1185-1193
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    Formerly, we have proposed a method for acquiring fuzzy reasoning rules by a neural network. Since the neural network used in the method has a creative function of reasoning rules, so the only necessary rules to express the characteristics of controlled object are created in the network. The method, however, uses the fixed number of membership functions while learning. Thus we must determine beforehand the suitable number of them.
    In this paper, we propose a new acquiring method that extends the former method. Neural networks used in this method have a creative function of membership functions in addition to fuzzy reasoning rules. The result is that this method obtains the necessary and the minimum number of membership functions and fuzzy reasoning rules. Also we demonstrate the effectiveness of this method by appling it to two problems, i.e., an identification problem of nonlinear functions, and an obstacle avoidance problem of moving robot.
  • 増田 達也, 佐藤 幸史, 池谷 浩彦, 藤井 善行
    1994 年 114 巻 11 号 p. 1194-1200
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    When we apply a hierarchical neural network based on the back-propagation algorithm to a particular problem, we must determine beforehand the suitable size of network for the problem. But it is a very difficult problem. Too small a network will not learn at all, while too large a network will be inefficient and worsen its generalization ability due to overfitting.
    In order to solve this problem, in this paper we propose a compact structuring method based on learning with a large size network and then compacting gradually the network by eliminating extra hidden layers and units. The result is a small and efficient network that performs better than the original. Also we demonstrate the effectiveness of this method by appling it to an identification problem of logic function.
  • 黄瀬 浩一, 白石 忠道, 高松 忍, 福永 邦雄
    1994 年 114 巻 11 号 p. 1201-1207
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    In general, a document includes diagrams as well as sentences for readers' benefit of understanding. To extract information from documents by document understanding, it is required to integrate information extracted by natural language processing and image processing. This paper presents a method of integrated understanding of sentences and diagrams, aiming at automated understanding and solution of geometry problems. The characteristics of our method are as follows: 1) Ambiguity in interpretation of a sentence and a diagram is explicitly represented, and then reduced by identifying coreferece between a sentence and a diagram with the aid of knowledge about geometry. 2) A problem is solved using the knowledge about geometry to verify the extracted information. From the experimental results and comparison with related work, we discuss the effectiveness of our method.
  • 松井 伸之, 石見 憲一
    1994 年 114 巻 11 号 p. 1208-1213
    発行日: 1994/10/20
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper, we propose a multilayered neural network including neurons with fluctuated threshold, which are closer to real neurons, and investigate the performance of this network by solving, for example, the N-bit parity problem. We estimate the learning capabilities such as the success rate and the number of learning times. As numerical results, we show that our simple method of incorporating the fluctuation into the network improves these capabilities noticeably. For example, in the 4-bit parity check problem using 4-4-1 network, the success rate of the network with no fluctuation is very low (25%), but that of the network with fluctuation is much higher (89%). From a practical point of applications, we also discuss the role of the fluctuation and its optimum parameters in detail. Furthermore, we apply this method to function identification problem, and see that it is effective in more general learning problem.
  • 1994 年 114 巻 11 号 p. e1
    発行日: 1994年
    公開日: 2008/12/19
    ジャーナル フリー
feedback
Top