In this paper, a two-degree-of freedom control system with fuzzy controllers for the progressive wave type ultrasonic motor (PUM) is proposed. The system consists of a fuzzy feedforward controller (FFC) and a fuzzy disturbance observer (FDO). The FFC adjusts both the driving frequency and the phase, of which objective is to improve the tracking performances. The frequency and the phase control rules of the FFC are tuned so that the tracking error converges to zero. After tuning process is completed, the fuzzy rules which represent the equivalent dynamic inverse are obtained in the FFC. The tuned fuzzy rules of the FFC are copied and used to estimate the applied disturbance in the FDO. The responses of the proposed positioning system for the desired output and the disturbance are measured. Experimental results demonstrate the effectiveness of the proposed system for improvement of the tracking performances and suppression of the disturbance input.

View full abstract

In recent years, many methods which assist WWW retrieval based on user's interests have been proposed. However, it is generally difficult to estimate user's interests directly from HTML files, since they often contain multiple topics some of which may not interest a user. In this paper, we propose a method of estimating user's interests and WWW retrieval both of which are based on topics extracted from HTML files. The characteristics of the method are as follows: (1) Topics in a HTML file are extracted by identifying repetitive sequence of its HTML tags, (2) User's interests are estimated by clustering topics extracted from HTML files which contain user's interesting portions. (3) The accuracy of estimation of user's interests as well as WWW retrieval is improved by incorporating retrieved topics into the case-base as positive or negative cases which are specified by a user. Experimental results for 151 HTML files show that the method improves the precision in both of estimating user's interests and WWW retrieval, compared with a method without the extraction of topics.

View full abstract

This paper deals with estimations of aspirated air flow in a three-dimensional (3-D) vocal tract during fricative consonant phonation using the Finite Element Method (FEM). The shapes of the 3-D vocal tracts during phonation of fricative consonant /s/, /sh/, and /shu/ are reconstructed from 32 coronal Magnetic Resonance (MR) images. MR images of the dental crown that contains a small amount of water were obtained using a dental crown plate. The 3-D vocal tract shapes are divided with a tetrahedral element (/s/: 28, 686 elements and 7, 010 nodes, /sh/: 31, 681 elements and 7920 nodes, /shu/: 31, 360 elements and 7619 nodes). A rigid wall constitutes the vocal tract wall. It was shown that the flow rate was high in the narrow space made between the upper central incisors and the tongue surface. An electrical equivalent circuit model for fricative consonant phonation was proposed based on the location of the noise source. The computed spectra from the equivalent circuit is good agreement with the spectra of the subject's original voice.

View full abstract

This paper proposes a method where the mobile host itself watches both a wired-LAN and a wireless-LAN interfaces. If it watched that it is connected to a wired-LAN, it activates the internal dynamic-bridge-function between the wired and the wireless-LAN, and transits in an accesspoint(AP) and provides a wireless-LAN service-cell for each station(ST). If it is disconnected from a wired LAN, it suspends the internal bridge, and transits into a ST. If it is composed as an ad-hoc network component, it provides gateway functions to some wired-LAN, and is effective as a flexible mobile environment for the can't-stay-still office environment. Moreover, compensation of the frame at the time of disconnection and delay were measured supposing voice adaptive application.

View full abstract

In a sales promotion task, suppliers prepare and present the sales promotion proposal plans for negotiating with retailers' buyers what commodities they should sell. For the electronic sales promotion tasks, a B-to-B EC system using mass customization technology has been developed. In this system, a standard sales promotion proposal plan is modified with customization rules for each retailer. The customization rules vary by the retailer's conditions. Furthermore, the effective lifetime of the customization rules is very short because of the appearance of new commodities and seasonal changes. Therefore, the rules should be generated automatically.
This paper proposes a customization rule generation method from sales log data for mass customization technology. Customization rules determine the related commodities from a target retailers' conditions and an arbitrary commodities on the standard plan. In proposal method, these rules are generated based on correlative probability, existence probability, and reverse correlative probability according to the related commodities from order transactions. The related commodities for target customization rules become the addition commodities for personal sales promotion proposal plan.
The proposed method has been applied to real log data in wholesale industry. It is confirmed that the proposal plans from customization rules are efficient for sales promotion task.

View full abstract

In order to make a decision for new system investment, it is important for a decision-maker to forecast what benefits can be realized after investment. However, it is very difficult to predict the benefits because a new system affects several factors, such as business style, product cost and marketability. To forecast benefits of new system clearly, we propose a scenario generation method based on the qualitative simulation with a structural model that derives the scenario documents of new system benefits by determination of an effective functional item.
As there are many system functions in the new system, an effective function item is the most suitable system function to acquire the benefits of a desired item. The scenario document is generated when the selected effective function item and desired item are put into a sentence template.
This scenario generation method is applied to business e-commerce system. By using the generated scenario document, the system engineer can know the stress points so he can explain the desired benefits of new system to the company which is willing to introduce a new system.

View full abstract

In this paper, we propose a method of extracting and tracking regions of human face and hands appeared on sequential images considering fusion of color information and tracking region. Conventionally, it has been hard to extract the skin regions of human in the case when a background reflects skin color objects like corrugated cardboards, or including moving picture such as TV screen. Coping with such kind of difficult situation to extract skin regions, we firstly obtain three kinds of information based on background color, skin color and its tracking skin regions, then apply Dempster-Shafer's combination rule to the three kinds of information. This approach using the fused information becomes a robust extraction method, because other information can extract the regions even if one of the information lacks. The effectiveness of our approach is examined by the experiments.

View full abstract

Automatic classification of bill money has been well developed and it is important that the classifier must have high accuracy. Generally, accuracy of classification is represented as a recognition rate of sample data. To evaluate the accuracy more strictly, we will introduce a reliability criterion.
In the pattern recognition, Neural Networks (NNs) have been adopted. Among them a competitive NN has a simple structure and can explain the relation between the inputs and the outputs more easily than a layered NN based on the back-propagation method. Thus, we use a competitive NN for the bill money classification and use the Learning Vector Quantization (LVQ) method for training the NN.
After introducing a reliability criterion based on a probability distribution for the classification by the LVQ method, we classify the US dollar by the LVQ method and show the effectiveness of the proposed method.

View full abstract

With the increasing number of dispersed power sources interconnected to the power supply system, power systems can take the form similar to a longitudinal system. Also, they can become looped, with the both ends of the logitudinal system interconnected each other. In the present paper, the steadystate stability of longitudinal power systems and looped power systems is analyzed by means of the mode analysis method via eigenvalue and eigenvector calculation, and the power system stabilization by SMES (Superconducting Magnetic Energy Storage) for looped systems is examined compared with that for longitudinal systems.

View full abstract

Service restoration in distribution systems can be formulated as a combinatorial optimization problem, and it is very difficult to obtain a global optimal solution in practical applications. Numerous works have been reported for this problem. In these works, only a restricted part of network associated with the out-of-service area is reconfigured. Thus, spare capacity can not be transferred between adjacent feeders, and there may be a case where some de-energized load is not restored even if sufficient spare capacity exists. We propose a new genetic algorithm for service restoration problem on the basis of the coexistence of heterogeneous populations. In the proposed method, the whole network associated with all systems is taken into account, and reconfigured. Consequently, spare capacity can be restored as much as possible. We have also applied the proposed method to several service restoration problems corresponding to various faults in order to demonstrate the effectiveness of the proposed method.

View full abstract

Railway crew scheduling is a problem in which the crew service patterns are generated so that they cover the entire timetable while considering various constraints. We propose a crew operation scheduling method which consists of two parts, an initial solution scheduling part and a modification part. Initial scheduling allocates all trains and satisfies constraint of rest for meal. In order to accomplish this, we apply state selection method to the initial scheduling. At the modification part, subset of crew scheduling about constraint of the meal break is fixed and other constraint violation is dissolved. This approach enables to find feasible solution by only searching around an initial solution. In this part, the relaxation search method is adopted.
The method was applied to an actual timetable. From the result of the schedule by the proposed method, the effciency of this method is clarified.

View full abstract

In this paper, we propose a novel dynamic filter using a multilayer neural network, called the neural filter with spatiotemporal inputs (NFST), for improving image quality of medical low-dose X-ray image sequences. We model the noise in the medical X-ray image sequences to make simulated low-dose X-ray images for training the NFST, and have trained the NFST using them. We have performed experiments on the NFST with not only the simulated low-dose X-ray image sequences but also real low-dose X-ray image sequences. Experimental results have demonstrated that the NFST can effectively remove the actual noise in the medical X-ray image sequences, and the ability of NFST is superior to that of the conventional dynamic filter by the quantitative evaluation of image quality. We expect that medical X-ray systems reduce dosage to the patients effectively using it.

View full abstract

This paper describes a study on individual prediction of reliability of electric parts by using neuro-fuzzy network. In general, mean time to failure(MTTF) and mean time between failure (MTBF) in reliability test indicate the average property of tested parts or materials. In this situation, only average information on their reliability can be revealed. However, if we able to know when the each part will be broken, more efficient maintenance can be performed. Thus, we aim to establish a method for failure prediction of individual parts by using soft-computing paradigm. In this paper, carbon filmed registors are adopted and their failures are individually predicted by using neuro-fuzzy network. According to the time series of the measured value of the registors in the reliability test, we attempt to make prediction for the failures of each registor in order to establish a convinient method for quality control and maintenance. This paper describes some experimental results of reliability prediction using neuro-fuzzy network and comparative study on accuracy between conventional auto regressive model and the neuro-fuzzy network. We first discuss some essential issues to be considered in reliability problems and soft-computing techniques. Then, it is shown that the proposed method is better than conventional methods about performance for estimating the failure of individual registors.

View full abstract

This paper treats with 3-D vision system for cylindrical targets using highlight patterns from the target surface. This system consists of two optical sets. Each of them includes a TV camera and a point light source irradiating light beams radially. The position of the principal point of the camera lens coincides with that of the point source. The optical sets are placed at constant intervals. Each camera observes specular reflections from the target surface produced by illuminations by both light sources. By selection of pairs of the cameras and the light sources, the system can detects together the coordinates of the center and the radius of the circular cross section of the target This paper shows the performance of the proposed system by basic experiments, and examines the restriction of measurement region and accuracy theoretically. The correspondences of the experimental results with the theoretical analyses are also shown in this paper.

View full abstract

In recent years there has been renewal of interest in the concept of symbiosis. We propose a new modeling method of the complex systems named modified Lotka-Volterra systems using the concept of symbiosis. The symbiosis parameters are built in modified Lotka-Volterra systems and theses parameters are changed by fuzzy inference whose fuzzy rules are composed of Predator-Prey model. The purpose of this paper is to analyze the various phenomena of the proposed model and to train parameters in the fuzzy membership functions using RasID learning algorithm. The result of our experiment shows that we can more easily obtain the similar phenomena to real chaotic systems than the conventional methods.

View full abstract

Impedance control is one of the most effective control method for a manipulator in contact with its environment. The method, however, cannot regulate the end-effector impedance accurately without the exact model of the manipulator. The present paper proposes a new impedance control method in which an identification model based on a neural network (NN) is connected in parallel with a model of the manipulator. The NN in this method can identify the uncertainties of the manipulator model and modify a control signal to the manipulator according to the identified results. Computer simulations are performed to illustrate the effectiveness of the proposed method.

View full abstract

In order to improve the response property of the closed-loop control system constructed by generalized minimum variance control(GMVC), in general, the control-weighting of cost function is given a positive value. But this may cause the offset in the output of the system. Also, the offset may occur because of input disturbance. For eliminating these offsets, some methods have been proposed. In this paper, we design the self-tuning control system by adding a simple identification scheme. There exist parameter errors between the real plant and the estimation model. Therefore, we study robust asymptotic tracking performance for GMVC.

View full abstract

This paper describes a method for the design of linear function observer for a multivariable plant with measurement noise. This method uses the free parameters of the controller. Simulation results are presented which confirm the validity of the design method.

View full abstract

Genetic Algorithm is governed by its system parameters which are usually set in a heuristic manner at present. These parameters; the reproduction rate, the crossover rate, the mutation rate, and the population size are set to appropriate values depending on the characteristics of the target problem. This paper studies the behavior of Meta-GA system which searches to the best parameter values for problem oriented 2nd-GA. Here the “number of generation cycle” is utilized to evaluate the performance of the 2nd-GA.

View full abstract