GUI (Graphical User Interface) has prevailed as a human interface of small computers and such AP (Application Program) as runs on “window system” is coming to standard.
In this kind of AP, dialogue box or anything like this is used to carry on a dialogue with user. A user's specification is passed on to the AP by accessing to the various dialogue components (text fields, list boxes, icons, buttons, etc.) laid out within the dialogue box. This is considered easy to use bacause of direct manipulation oriented UI (User Interface) in public estimation.
However, these components are laid out by hand works dependent upon the experience and the intuition of UI designer. It seems that this is undesirable from the viewpoints of shortening development term of AP and so on.
In this paper, we try to automate the layout works described above. In a concrete form, we define this as a problem of obtaining the layout which satisfies users best under the constraint of allocation order among these components. And we propose a new method of automatic layout using fuzzy inference. Finally, we verify the usefulness of the method by several experiments.

View full abstract

ID 3 algorithm can automatically acquire a decision tree from a set of training examples. However, ID 3 can only deal with the distinct data values. This paper presents a fuzzy decision tree which expresses some fuzzy classification rules, and an algorithm to induce a fuzzy decision tree from the training examples including numerical or fuzzy data values. This new algorithm, called IDF, has a labeling procedure in each decision tree expanding step to make effective fuzzy classification items. These items are used to do fuzzy decisions in the branch nodes. Fuzzy decision tree can also give some classification results with certainty ratios. The authors examined and exemplified the efficiency of this algorithm by some numerical experiments.

View full abstract

This paper deals with an automatic generation algorithm of a fuzzy model using the genetic algorithm and the delta rule, The fuzzy inference is applied to the various problems. However, the determination of the membership functions is a difficult problem, because the determination depends on human experts. The auto-tuning methods of the fuzzy model have been proposed to develop the time-consuming operation by human experts. Nevertheless, the auto-tuning methods have a weak point, such that it is difficult for human experts to set the initial conditions of the system. The convergence of tuning depends on the initial conditions, which are determined by the scale and parameters of system. So, we propose an AFUGA system (Auto Fuzzy Tuning Method using Genetic Algorithm). This method brings a minimal and optimal structure of the fuzzy model. This general system can be applied to the robotic motion control, sensing and recognition problems and so on. In this paper, we show the validity of the AFUGA system by simulation.

View full abstract

For desired control response, gains of I-PD controller must be adjusted to proper values according to the dynamic characteristic. However, it is difficult to adjust control gains by using conventional method which repeatedly observes the responses of the control system. In proposed system, the control response is improved by auto-tuning using fuzzy reasoning. Two methods are newly proposed as this reasoning method and a conventional method is compared. In this system, a triangular membership function is used for each fuzzy subset. Three basic fuzzy subsets PB (positive big), ZO (zero) and NB (negative big) are used. The parameters of these functions used in the auto-tuning are determined by the simulation and trial and error by considering the simulation results. The fuzzy auto-tuning is obtained by a product-sum compositional rule of inference, and then defuzzified by the height procedure to generate a nonfuzzy value. The original fuzzy rule was determined by the root locus for suppressing the vibration of the controlled value. These root loci are obtained from I-PD controller and the pulse transfer function of the controlled system. The simulation results by using the three methods are presented and the advancement of the proposed controller is shown. Additionally, the improvement of the control response in the case of overshout tuning target 0% is also shown. One of these methods is applied to DC chopper current control system with DSP (TMS 320 C 30) and the effectiveness of this method is shown.

View full abstract

A multilayer neural network having predetermined structure and multiplier and divider neurons is proposed in reference to structure, learning method, generalization and knowledge extraction applied to fuzzy reasoning. This neural network is named the Fuzzy Reasoning Neural Network (FRNN). It can determine membership functions and rules in learning, modify rules, add and delete rules from the set of rules, and extract knowledge in fuzzy reasoning applications. I also propose a new method that is applied to FRNN that extract rules automatically by the improved Backpropagation method for learning.

View full abstract

Fuzzy inference method is introduced to classify landcover using LANDSAT MSS data. The proposed algorithm is intended for use to estimate area components of landcover types which are present in a spacial segment of the order of 0.5 to 1 km². Divided into two hundreds 750m×750m square segments is the study area whose half portion is utilized as the training area and the other half as the test area. Average of the each of the four band MSS data on a segment in the training area is assumed as fuzzy number which makes up an assumption of a fuzzy production rule while the area of each of the landcover types in the segment forms a conclusion of that rule which could be crisp or fuzzy. Widely conducted are the experiments which reveal that; (1) the input MSS data should also be made fuzzy number, (2) the area used in the conclusion should be crisp, (3) 60% rules selected randomly from the provided production rules can produce satisfying estimates, (4) the developed algorithm outperforms the discriminant function algorithm by 33% based on root mean squared error index and is robust against noises lurking in the MSS data.

View full abstract

To control arousal level using a physiological index has been attempted for some time. However, none of the attempts have enough effects. Because they have two problems. One is a problem in measurement: measuring a physiological index in itself is complicated and creates mental blow to subjects. The other is a problem in contol method: though a relationship between physiological index and arousal level can be detected qualitatively, the detected relationship can not apply to a control. This paper describes a control system for drowsiness which has been developed in order to overcome the problems. Electrodermal activities, which has the superior characteristics of detecting low-arousal-level sensitively, non-restriction and easy operation, was selected out of many physiological indices to overcome the problem in measurment. For the problem in control method, fuzzy inference was introduced in order to apply characteristics, that had been detected qualitatively, of the electrodermal activities in the low-arousal-level to this system. A strict mathemetical model is thereby needless and a manipulated variable can be decided easily with 9 fuzzy rules. It was then confirmed that this system could control drowsiness reasonably from the results of the experiment.

View full abstract

Fuzzy control has a good ability for systems of which mathematical modeling is difficult. It also has flexible ability for control. However, when the number of the parameters are large, the number of rules increases in proportion to the square of the number of parameters.
In this paper, we propose a way of decreasing the number of fuzzy rules by classifying and combining them for a mobile robot. Firstly, steering control input and velocity control input are decided separately. Secondly, those control inputs are composed with keeping the performance of control good. Finally, we show the effectiveness of the proposed method by simulations.

View full abstract

This paper presents a new neural network structure from which features of input data can be extracted easily. The basic structure is a two-layer perceptron and each unit in the input layer has a fuzzy membership function. The input space is divided into fuzzy sub-spaces by the membership functions and the new type of neural network has an ability of nonlinear mapping. The new neural network can handle some deformations of input data with the fuzzy divisions. The structure of the new network is very simple and the features of input data are easily known from the connection weights of the learned network. The authors call the neural network Fuzzy Net. A basic experiment of signature recognition is done using the fuzzy net. The capability of the new network in feature extraction is verified.

View full abstract

The Lateral diffusion length of Zn in Si-doped GaAs under a SiO₂ or SiNx diffusion mask was evaluated using (1) the TLM resistor pattern method, (2) the p-n capacitance method, (3) the EBIC method, and (4) the cleaved stain-etching method. The SiO₂ was deposited at 250°C using SiH₄ and N₂O gases, and SiN_x was deposited by plasma-chemical-vapor-deposition at 300°C using SiH₄ and N₂ gases. Employing the open-tube method, Zn was diffused into GaAs through windows opened in the dielectric films at 600_??_650°C for 2.5_??_22.5 minutes in a nitrogen gas ambient. ZnAs₂ was used as a diffusion source. The anomalous fast lateral diffusion of Zn was observed only for the SiO₂-deposited GaAs. The ratio of the lateral to vertical diffusion length was about 4 for 0.3μm-thick SiO₂ and was strongly dependent on SiO₂ film thickness. For SiN_x on the other hand, the ratio was about 0.9 regardless of film thickness. Similar results were obtained by each evaluation method. The anomalous diffusion mechanism for SiO₂ is not clear, however, the results of our systematic experiment suggests that the defects at GaAs surface, which were produced during SiO₂ deposition, correlate with the anomalous diffusion.

View full abstract

Systems are becoming more complicated, and large-scale software systems are required to control them these days. This paper presents a new approach for productivity improvement in case of developing such large-scale software systems. Due to advancement of network technologies and engineering workstations (EWS), distributed and simultaneous development environment is going a current paradaigm for developing such large-scale software systems. However, we have few effective methodologies for this development environment so far.
In this paper, we propose a method to optimally divide a software system in a upper phase of system development process. This method tells us information of how to allocate software modules to development organizations. It gives productivity improvement in the distributed and simultaneous development environment. Moreover, this method is applieded to a project management.
The proposed method has been applied to a practical software system for power systems control to show its ability. The result shows that the target system is optimally decomposed into several smaller subsystems according to productive ability of development organizations. This method is quite promising with systems growing up.

View full abstract