In Japan, about 84 percent of real plants adopts PID controllers. However, when we use the PID controllers, it generally needs much effort and time to tune PID gains. In this paper, we propose a method to tune the PID gains by using three-layered neural networks. Taking into consideration that the PID gains are non-negative real number, we select functions whose derivatives are sigmoid as output functions in the output layer. To find system's Jacobian, we identify the unknown plant by using a neural network as an emulator. Finally, numerical results are illustrated to show effectiveness of the present method through simulations.

抄録全体を表示

In technical documents, diagrams are generally utilized with expository text aiming at concise presentation of information. Since the information is conveyed by both text and diagrams, it is necessary to understand both of them to extract information from documents. In this paper, we propose a method of understanding of expository text with diagrams. One of the central issues of such understanding is how to cope with ambiguities in the results of processing of text and diagrams. To solve this problem, the following steps are employed: (1) Information extracted from text and a diagram is described based on the common representation called relational representation which enables us to describe the ambiguities explicitly. (2) The extracted relational representations are unified by hypothesizing the references from text to a diagram. (3) The maximum and consistent set of hypothesized references is obtained to resolve ambiguities. Based on the experimental results for some pairs of text and a diagram, we discuss the advantages and the limitations of our method.

抄録全体を表示

In the usual vector control method for induction motors, the only electrical nonlinear terms are canceled by a nonlinear state feedback, and assuming the rotor flux is constant, the mechanical nonlinear term is approximately linearized. However, the speed response becomes worse in case that the rotor flux is not constant as the field weakening control. Therefore, in this paper, a novel noninteracting control method in consideration of the mechanical nonlinearity is proposed. In our proposal, the induction motor model is exactly linearized and decoupled by not only the nonlinear state feedback but also the state coordinate transformation, and the high performance speed control is realized. The validity of our proposal is verified in some simulations.

抄録全体を表示

In order to plan the road design, it is necessary to measure the traffic flow. Though the main factors of these parameters are how the speed of the vehicles are and what kind of the vehicle size it is, the basic one is the vehicle volume. This paper deals with the method to recognize automobiles by analyzing the brightness distribution on TV image of vehicle flow.
Though the vehicle flow on the expressways is measured by 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 methods are not useful for automobile traffic flow survey. For the practical survey in recent years, the standard video systems are employed because setting of them on arbitrary point is easy.
In this paper, a road and a vehicle are distinguishable by analyzing the brightness distribution of TV image. The brightness distribution of the road is generally uniform instead of illuminance change, and this distribution changes by over lap of vehicle on the road. It is possible to analyze it by the simple algorithm. Therefore, the brightness distribution is rated by the standard deviation. And the real time measurement become realizable.

抄録全体を表示

This paper deals with distinguishing vehicle types in the automobile traffic flow by image processing. In the traffic flow, a traffic jam always occurs. The traffic jam gives a driver's burden and it results in time-delay, a traffic accident, and air pollution. In that case, it is important to offer the traffic information such as vehicle volume and the distinction of vehicle types. In this paper, we attempt to do them using a value of vehicle width.
The general measurement method of traffic flow is done using an ultra sonic sensor. But the system is expensive, and the method is not useful for automobile traffic flow. Thus, we have developed the simple system using a portable camera, a personal computer, and DSP. This system makes the width measurement of vehicles in real time. The measurement method uses Fourier transform, and takes an advantage of robustness for noise. Taking into consideration that frequency analysis for width measurement on vehicle image differs from it on road image, we will measure vehicle volume and distinguish vehicle types in real time.

抄録全体を表示

The timing belts are widely used for torque transmission in many inductrial applications including the industrial robots. In such system, the vibration is often generated because of the elasticity of the belts. This vibration makes it difficultt to achieve quick responses and smooth motion control. In this paper, the position control system of a horisontal robot arm with flexible timing belts is proposed. The mechanical system of the robot is modeled as a two-mass system, which consists of the motor side mass, the arm side mass and the spring element due to elasticity of the belts. The position control without vibration is based on the state feedback of the motor position, the motor speed and the torsional torque caused by the flexible belts, where the motor speed and the torsional torque are estimated by the least-order observer because the detectable information is only the motor position from the rotory-encorder. The arm side inertia varies according to the arm position, as a result the mechanical system is nonlinear. Therefore, the feedback gains are adjusted according to the arm position in the proposed system. The effectiveness of the proposed system is verified by the experimental results.

抄録全体を表示

One of the most serious problems in medical science is an increment of the Alzheimer's disease. It is known that the patients have the brain atrophy because of the neural cell loss. It is useful for the diagnosis of the Alzheimer's disease to measure the volumes of and to display the brain portions. We can obtain anatomical information from 2D slice images produced by MRI system. We construct the 3D voxel data from these images. We propose a 3D computer aided system for the diagnosis of the Alzheimer's disease. The system consist of three parts: (1) extraction of the portions from MRI data based on region growing method, (2) measure the volumes of and display the portions, and (3) user interface for a easy manipulation of a medical doctor. In this paper we describe procedures to carry out them. The comparison of the obtained volumes with manually obtained volumes by a doctor shows that the error ratio is about 2% for 48 MRI data.

抄録全体を表示

The purpose of this study is to develop a sensory feedback system for the biomimetic myoelectrically-controlled prosthetic hand. This system transmits an information about the force, which is applied to the fingers of the prosthetic hand, to its user with vibrotactile stimulus. When the larger force is applied to the fingers, the stronger vibrotactile stimulus is given to the user. The developed sensory feedback system consists of a signal processor and a vibrator made of a small-sized voice-coil speaker. Usefulness of this sensory feedback system was shown by the experiment, where a healthy subject was grasping soft objects with the prosthetic hand.

抄録全体を表示

This paper proposes an extraction method of charcters string from scene image using HSL transform. In this method, an input color scene image is divied into 15 or 28 binary images by HSL transform. In case of 28 images, 13 images are obtained by dividing on H-S distribution map and 15 are obtained by dividing lightness histgram. Label allotment is carried out in each binary image and rectangle regions are extracted from label images by label allotment. These rectangles are merged by using 3 decision criteria. Candidate rectangles which surround characters are decided by the connectivity in each image and characters string area is extracted by integrating these candidates in all binary images. The experimental result shows that 92.4% of charcters included in 100 types of different color scene images are correctly extracted by this method.

抄録全体を表示

It is required to establish the precise method of diagnosing separation or void in building walls. We have considered a diagnosis method which discriminate whether separation is in the wall or not. Test specimens with various separations have been used in this investigation. Power spectral densities of tapped sounds on the specimen are used for diagnosing a separation. The spectral densities for normal wall does not have any peaks, but for separation or void existing in the wall have sharp peaks at some frequencies. Namely, the features of abnormal wall can be recognized by intensity of spectral density and peak frequency. The distributions of those features may be approximated by appropriate triangular membership functions. Then diagnosis of the separation is performed by using the fuzzy reasoning about those features. The investigations have been performed about various kinds of separations such as size of separation area and depth of separation. The results of diagnosis for those separations show that the proposed procedure in this paper is available for diagnosing separation or void in building walls.

抄録全体を表示

The inverter-fed resonant DC-DC converters which make the most of parasitic reactive components of high-voltage transformer have attracted special interest for medical equipments as X-ray tube voltage generator in order to realize high-power density, high-performance in steady-state and transient state, and high-efficiency. Though desired output voltage performances for X-ray tube are specified, it is not easy to find out its optimum control parameters for wide load setting ranges. To overcome this difficulty, the repetitive learning control scheme is introduced to the resonant DC-DC converter for medical-use X-ray tube voltage generator. Its adequate DC voltage response in steady-state and transient state can be automatically achieved by using the repetitive learning control process.
This paper presents a digital learning control scheme to suppress overshoot and realize high speed response of X-ray tube voltage during the first transition in spite of wide load setting ranges. In addition, since the X-ray tube voltage control system is easy to be instability due to the periodic input voltage fluctuation in commercial utility-grid system and abruptly-change DC filter voltage disturbances, it is fully stabilized by using the feedforward control loop. Its effectiveness is confirmed by several simulating and experimental results.

抄録全体を表示

T&T olfactometer, standard odors for measuring olfactory sense, seems to be suitable for the standard odorants to compare the sensor response with the human feelings. The responses of the quartz resonator sensors with synthetic lipid membrane to five odors of T&T Olfactometer were evaluated. The five odorants were not recognized separately from the response patterns, because the sensors had low sensitivity and selectivity for five odorants. Five odorants were discriminated after applying the principal component analysis to the response data. Comfort or discomfort feelings of five odors was successfully distinguished with the responses of three sensors and the principal component analysis. New sensor coated with urethane resins were demonstrated and we found that they were candidates for the sensors with high sensitivity and selectivity for the T&T olfactometer.

抄録全体を表示

In the heat source system, the circulating water transmits heat to provide warm or cool air with every room according to the environment. In order to confirm that the system is controlled satisfactorily, it is necessary to detect its fault from the viewpoint of the environment (ex. hot, cold) and energy conservation (by equipment and control system) and so on. However, the cost of the system is limited so strictly that the system can not have enough sensors to obtain crucial data of fault detection. Collecting data specially will be an additional burden for daily operation. And each building air conditioning system has various, vague and changeable structure.
The stochastic qualitative reasoning is one of the methods for analyzing behaviors of a plant. In this technique, behaviors of a target can be pursued on the basis of its unsubtle models. The qualitative reasoning is effective for fault detection by using simple qualitative models. Therefore, the technique is applied to fault detection of a heat source system.
To construct a comprehensible qualitative model of the heat source system, the feedback function, which can determine the current state reflecting the previous state, is introduced to the models. Using field data, sensor faults, refrigerator movement faults, and three-way valve faults, were successfully detected.

抄録全体を表示

Today, the technique of virtual reality has become popular in various fields because of its technical progress. We focused on ‘presence’ - one of the three elements of virtual reality, and constructed a simple supporting system for piano education. The virtual world and the real world are overlapped by showing the visual information in this system. The purpose of this system is to use the realistic virtual image as an interface (or tool) of education.
The unity of the virtual image and the reality has been done with standing pictures by using, for example, ‘See-through HMD’. In this research, however, we overlapped the virtual image and the real world with moving pictures. The overlapping is accomplished not by using HMD, but by using a half mirror. The virtual hands can automatically play a tune from a score so that learners can choose what they like without instructors. This is one example of the system which has a proven effectiveness.

抄録全体を表示

VLSI-oriented dynamic data-driven processors Qv-x is constructed by a self-timed elastic superpipeline, which has inherent buffering capability of instantaneous fluctuations in packet-flow. Thus the pipeline can smooth the packet-flow in the processor, so that it is recognized as a very promising hardware implementation method for dynamic data-driven processors. Due to the structure of programs and/or the characteristics of input packet streams, however, excessive fluctuations in packet-flow may happen, which lead to the performance degradation. To eliminate ill effects of excess fluctuations, elastic queue-buffers are also incorporated in the Qv-x. This paper proposes an allocation method of queue buffers in the Qv-x to cope with different sources of the fluctuation and provides the quantitative evaluation of buffering capability of fluctuations through extensive simulation experiments. The simulation results show that, by an allocation method, large fluctuations can be absorbed and full performance of the processor can be achieved.

抄録全体を表示

To predict electromagnetic (EM) noise emission from a signal line on a digital printed circuit board (PCB), we propose an equivalent circuit model of CMOS logic IC which requires less calculation time and simple procedures in modeling. In this paper, we described characteristics of the model and the procedure to determine parameters. Calculated voltage waveform of IC output was first compared with measured one. Calculated result of the fall time as a function of load capacitance was very similar to the measured one. Spectra of output voltage are also close to each other between them. Spectral profile is -20dB/dec in the low frequencies, which becomes to be -40dB/dec in the higher frequency region. EM spectra were then calculated, resulting in the peak height 6dB larger than the measured one when the load capacitance was large. Calculated spectral profiles, however, agree approximately with measured ones.

抄録全体を表示

Data-driven scheme is a natural paradigm for describing system specifications and for designing massively parallel processing architectures. This paper discusses an advanced software production environment by which diagrammatic specifications are directly transformed into executable programs by utilizing inherent virtues of the data-driven paradigm covering software development and maintenance processes. In the paper, a direct transformation of block-diagram-like specifications, such as signal flow graph, into the data-driven programs in an interactive environment is proposed. The transformation system verifies specifications under the data-driven interpretation and interactively inquires about diagrammatic definitions related to behavior or data-structure which is necessary information to generate the executable programs. A generic implementation of the system as a kernel of our environment is then discussed and its practicability is illustrated through a specification process of a MUSE signal decoder.

抄録全体を表示