Mototaka Sone

An application of neural network for harmonic detection in active filter

This paper describes the method which uses neural networks for the harmonic detection process in an active filter. Originally, Fourier transformation is used to analyze the distorted wave from the power line and a low-pass filter is used to eliminate the fundamental wave before each harmonic component is detected. Due to these complicated processes, the behavior of the active filter is delayed such that it is difficult to compensate the harmonics in real time. In order to improve the processing speed and simplify the harmonic detection process, a neural network is used to detect the harmonics in the distorted wave instead of Fourier transformation and low-pass filter. The authors simulated the distorted wave including 3rd, 5th and 7th harmonics and used a neural network to recognize each harmonic. The results shows that a neural network enable to detect each harmonic effectively.<<ETX>>

The reliability of neural networks on pattern recognition

Researches on pattern recognition using neural networks (NN) have mainly been studied about the speed of training or the recognition rate. However, nobody has so far studied about the reliability whether NN distinguish the pattern by the right judgement or not. Therefore, this paper defines the ideal condition between the hidden layer and the output layer in order to consider the reliability of NN. As the result, the reliability rates of NN are calculated by using 3 types of training data, and the training data to get a high reliability rate is suggested.<<ETX>>

Translation and rotation-invariant pattern recognition method using neural network with back-propagation

The authors present a new translation and rotation invariant pattern recognition method using a neural network. It is clear that the left-right, up-down translation or/and rotation invariance are achieved by simple preprocessing of the original patterns without improvement of the network structure. They use a three layer feed-forward network with back-propagation for learning and recognition. The proposed method has the following merits: the net size is relative small, learning and recognition is easy. Moreover, a 100 percent recognition rate is realized by the proposed method, for the alphabet.<<ETX>>

Relationship between field emission current characteristic and electrical breakdown in vacuum

It has been assumed that an electrical breakdown in vacuum is caused by field emission electrons emitted from a cathode surface. And these field emission electrons emitted from microprotrusions on the cathode surface. It has been recognized that this electrons current is expressed by F-N theory. Therefore, it is thought that a characteristic of a field emission current has influence on the electrical breakdown in vacuum. In this study, the electrical breakdown voltage in vacuum gap was measured in order to investigate whether the correlation exists between this voltage and the characteristic of the field emission current. At this time, the field emission current was measured from appearance of electron to just before electrical breakdown in vacuum, by using a Micro Channel Plate (MCP) and a pico-ammeter. As the result, it is guessed that the characteristic of field emission current has influence on the electrical breakdown. Because the applied voltage to keep the same field emission current correlates to the electrical breakdown voltage in all cathodes when the field emission current is near to the electrical breakdown.

CRITICAL CURRENT INCREASE OF HIGH-TC SUPERCONDUCTOR PIPE WITH COAXIAL CONFIGURATION

The critical current (Ic) of high‐Tc pipe‐like superconductor was found to increase under external magnetic field, when the superconductor pipe and bar are constructed into coaxial cylinder configuration. The superconductor specimens in the Bi‐Sr‐Ca‐Cu‐O system used were fabricated by a glass‐ceramic method. When the currents in the inner bar are made to flow in the opposite direction against the outer pipe, the critical current of the pipe is observed to increase by a measurable amount, although the pipe is exposed to the magnetic field produced by the inner conductor current. This phenomenon was observed not only in direct current but also in alternating current.

Tuning of learning rate and momentum on back-propagation

It is known well that backpropagation is used in recognition and learning on neural networks. The backpropagation, modification of the weight is calculated by learning rate ( eta =0.2) and momentum ( alpha =0.9). The number of training cycles depends on eta and alpha , so that it is necessary to choose the most suitable values for eta and alpha . Then, changing eta and alpha , the authors tried to search for the most suitable values for the learning. As a result, the combination of eta and alpha given the minimum value of the number of training cycles behave under the constant rule. Thus eta =K(1- alpha ). Moreover, the constant K is decided by the ratio between the number of output units and hidden units or the initialized weight.<<ETX>>

Observation of field emission behavior on cathode surface

A field emission from metal surface is known to play a dominant role of a breakdown in vacuum. We observed the field emission electrons and its emission sites on the cathode surface by using a micro channel plate (MCP) with a luminescent screen under the high vacuum condition. A CCD camera observed location of emission sites and its change with time. By using a computer analysis, the change of emission sites with time at each 16 m second could be observed. As a result, it was found that appearance and disappearance of emission sites was repeated in a short time less than 0.1 s, however, the number of emission sites counted in an observation was kept constant. It was seemed that the fluctuation of emission current was related to the disappearance of emission sites. This observation of field emission revealed that almost emission sites could sustain an electron emission for no longer than 0.1 sec. It was estimated that the chemical or geometrical changes were occurred at emission site by electron emission.

The effect of cathode deposition on the behavior of field emission

It is shown that cathode deposition by the ion vapour deposition method makes the number of field emission sites decrease and also makes field emission difficult. The number of electron emitted from individual site are also decreased by cathode deposition. It is predicted that the metallic whisker model and the non-metallic model are intermingled in the field emission of a normal cathode.

Method of deciding ANNs parameters for pattern recognition

A method for tuning artificial neural network (ANN) parameters for pattern recognition is described. A pattern recognition experiment carried out for phoneme recognition of English pure vowels in ANNs is presented. The significant of parameters that affect recognition rate seriously are defined. To determine the influence of these parameters on the recognition rate a tuning method is given. The tuning method is independent of the recognition rate.<<ETX>>

The effect of steric hindrance of alkyl chain on the conduction of water cluster in liquid

The state of water cluster in dielectric liquid was observed by FT-IR and relation between the ratio of cluster and conductivity was discussed. In this experiment, an ester compound with several length of alkyl chain was used. From the results, dissolved water molecules in dielectric liquid form some types of cluster and the composition of each cluster depends on the length of alkyl chain. It appeared that the conductivity of liquid depends on the ratio of clusters. The ratio of dimer and trimer in the liquid especially increases conductivity.