It is important to know the physiological variability characteristics of the single sweep record of evoked potential (EP). The problem is how to know whether the measured variability is caused by physiological factors (physiological variability) or only by the noise in the background EEG (noise variability). We proposed a method for judging whether the physiological variability exists or not in a few samples of the single sweep EP, and estimating the physiological variability interval, if it exists. Hypothesis test as to whether the physiological variability exists or not, and its interval estimate were achieved based on the probability density functions these were derived in this study. The proposed algorithm and the derived probability density functions were checked based on the simulation data whose variability characteristics were already known, and showed satisfactory results.
We propose a method for acquisition of generalized action-decision rules for mobile robot which is based on the assumption that selection of adequate focal point is important for the skill acquisition. The method makes it effective to acquire the generalized knowledge which are robust against environmental changes, especially unknown environment. The skill acquisition starts from discovery of rules to drive a robot to a goal point under a certain environment using genetic algorithm. Then, for the generalization of acquired rules, we select an adequate sensing range of the robot for detecting obstacles, which is described in the if-part of rules. Results of computer simulation show that the extracted skills work well to accelerate the learning of robot under new environment.
The electrical stimulation of the auditory nerve can elicit auditory sensations in the subjects with sensorineural deafness. Each stimulating electrode of an electrode array of the multi-channel cochlear implants may stimulate a distinct neural population. However, a great deal of current spreads from each electrode throughout a lymph because of the high electrical conductivity of the lymph liquid. This phenomenon causes the transmitted information to be reduce due to channel interactions. Even if the number of ele-ctrodes is increased, the transmitted information will be limited because of current spread. We have proposed the Tripolar Electrode Stimulation Method (TESM) which may succeed in narrowing the stimulation region and continuously moving the stimulation site for the cochlear implants. We evaluate whether or not TESM works according to a theory which is based on the numerical analysis using the auditory nerve fiber model consisted of unmyelinated and myelinated segments. In this simulation, the neural site and the sum of the excited fibers are compared with the compound action potentials which we obtained through animal experiments. As a result, based on the numerical analysis using this model, it is also proved that the anodal/cathodal threshold stimulation current ratio increases by decreasing unmyelinated segment between the inner hair cell and the habenula perforata. Also by comparing the result of the numerical analysis with that of the animal experiment, It is suggested that an effect of the unmyelinated segment is not significant. Based on their results mentioned above, we succeed in narrowing a stimulation region by controlling the sum of the currents emitted from the electrodes on both sides. Also we succeed in continuously moving a stimulation site by changing the ratio of the currents emitted from the electrodes on both sides.
In this paper we propose a new method using the mathematical induction method to investigate the unique solution of the Hopfield and the T-Model neural networks. The new method, unlike the traditional energy function method, treats the Hopfield and the T-Model neural networks as the nonlinear equations of F-1 (V)=WV+_??_ and investigated their unique solution. The nonlinear equations are similar to the dc equations of nonlinear transistor networks for which many important theorems on the unique solution have been presented. We develop the techniques to invistigate the unique solution of the Hopfield and the T-Model neural network and expect to derive many other useful theorems on the necessary and sufficient conditions for the networks (i. e. the equations) to have a unique solution. In order to verify these results, We give several simulation results on both models.
Morphological filtering based on the theory of mathematical morphology is nonlinear signal and image transformation that locally modifies their geometric features. This method is applied into several image processing techniques such as smoothing, feature extraction, shape representation, coding and so on. However, a large amount of the calculations of morphological filtering has been an obstacle to the real-time processing of its application. In this paper, the parallel algorithm of multi-level morphological filtering is proposed, and it is inspected by the execution results using MIMD parallel machine. Then, the limitation of this method and the performance of the processor node for the real-time processing is estimated.
The effectiveness of artificial life is investigated from an engineering point of view. A system (named S-system) of function-discovery using a bug type of artificial life is proposed in this study. Some functions are extracted by the system. The chromosome of a bug consists of functions, constants and variables. A tree structure is used for the expression of the chromosome. Some observation data are provided for the bugs. After obtaining the data, they reproduce. The concept of sexual/asexual reproduction is introduced in this study. The number of homogeneous bugs is limited for a variety of species. These ideas are very effective for a function-search. A part of the chromosome changes by mutation. As the generation proceeds, the bugs with the function in agreement with the observation data survive selectively, and finally determine the true function. For the validity of this system, some data which obey the known laws have been given for the bugs. The bugs have evolved and discovered some functions in agreement with the laws. As for an unknown function, observation data on glossiness have been provided. They have also discovered the function. In addition, they have determined the multiple curves included in the image data. The S-system has the characteristics that the solution tends to converge and stabilizes in comparison with Genetic Programming. Moreover, the form of the function is relatively simple.
In motion stereo, using a series of image is advantegeous to solve the correspondence problem. Aperture problem and the occlusion of feature points cause the correspondence accuracy to be low. In this paper, a method is proposed to improve the correspondence accuracy. The base theory is that, in the case that an interval of camera movement is enoughly shorter than the distance from a camera to a target object, the trajectories found in a group of true feature points are highly correlative. Even at a long interval of camera movement case, that can be solved using the base theory repeatly. By the correlation of feature point trajectories using Dendrogram, several false feature points are eliminated and true feature points are solved the correspondence. The base theory was led theoriticaly, and experimented on simulated situation. In experimental results on real images, the correspondence is solved with 88% accuracy in such case that a camera movement is rather long.
The final purpose of our study is the non-invasive measurement of the brain temperature change by head capacitance measurements. Establishment of this technique might be useful for the diagnosis of brain death, since it is reported that in the brain death the brain temperature is influenced by the surrounding temperature change. It is necessary for the establishment of this technique to acquire knowledge of the specific dielectric constant - temperature characteristics of the human head tissue. For this purpose, we examined the specific dielectric constant-temperature characteristics of the hog brain and skull which were obtained immediately after the death. As the experiment result, we estimated the specific dielectric constant to be about 1300 for the brain and around 3 for the skull in living condition, and its temperature coefficient about 1.84%/°C for the brain and around 0.23%/°C for the skull. Applying those estimated values to the human head, it was demonstrated to be possible to detect the change in human brain temperature from 37°C to 36°C by measuring the capacitance of the human head model consisting of only the brain and the skull across a pair of measurement electrodes of 1 cm2 area, under the condition of voltage in 20V with 500 kHz.
Chart analysis is one of market analysis techniques in the financial field. In chart analysis, patterns characteristic of price ups and downs are searched in the chart. When incorporating the chart analysis functionality into computer systems, pattern matching algorithm is required to reflect the human experts' know-how about chart interpretation. In this paper, we propose a method of supporting chart analysis using flexible template matching. On a chart called ‘point and figure (P&F), ’ characteristic patterns gathered by experts are registered as templates. Each template is described so that its shape can vary within the allowance reflecting the know-how. By using these templates, the pattern matching algorithm can find the characteristic patterns and their variations from the P & F pattern. Using the matching result, buying and selling timing signals are generated. Result of an experimental simulation using practical price data shows effectiveness of the proposed method.
A design method of nonlinear model following control system, which guaranteed bounded internal states, was proposed by Okubo(1), (2) for a family of plants with separable linear and nonlinear parts, but the requirement of positive real (PR) and inner product conditions for nonlinear part limited the method's application. This paper presents a new approach to Okubo's method by extending state feedback for linear system to nonlinear system, and relaxes the requirement of PR and inner product conditions for nonlinear part greatly. Using nonlinear state feedback, many plants which do not satisfy the inner product condition superficially can be modified to satisfy the inner product condition, thus can be dealt with by Okubo's method. Further, a nonsingular matrix R is brought in successfully to eliminate the requirement of PR condition. Genetic Algorithm is used to create nonlinear state feedback vector, and results the approach suits a family of nonlinear plants which can be expressed in Kronecker vector with any highest degree definitely. A compacted vector is used to benefit the design of control system. This paper presents the related proofs and simulations.
This paper proposes a robust object recognition system based on camera control. The objective of our system is to seek the optimum camera position where the unknown object can be recognized clearly. We define a degree of recognition-ambiguity based on basic probabilities that is calculated by using an input image and model images generated from object model data. Our active vision system makes an act plan iteratively so as to decrease the degree of recognition-ambiguity and controls the camera to move to the optimum position. Our proposed active method is able to recognize the object more accurately than conventional passive methods which analyze only a given input image. Experimental results show effectiveness of our approach.
The plant model is able to be constructed by the neural networks, i.e., by identifying the plant model by the neural networks. But it is somehow difficult to obtain a control law from this neural network based plant model. The iterative inverse method has been proposed for this problem, but for this method it is necessary to determine two parameters (convergence coefficient, iterative number) and calculate the equations iteratively for obtaining a control law. The fact mentioned above is a big drawback for on-line control. This paper is related with improvement of the iterative inverse method. It is shown that the proposed method is effective to control the thermal power plant by simulations.
Universal Learning Network (ULN) has been reported, which is a framework for the modelling and control of the nonlinear large-scale complexed systems such as physical, social and economical phenomena. And a generalized learning algorithm has been proposed for ULN, which can be used in a unified manner for almost all kinds of networks such as static/dynamic networks, layered/recurrent type networks, time delay neural networks and the networks with multi-branches. But, as the signals transmitted through the ULN should be deteministic, the stochastic signals which are comtaminated with noise can not be propagated through the ULN. In this paper, Probabilistic Universal Learning Network(PrULN) is presented, where a new learning algorithm to optimize the criterion function is defined on the stochastic dynamic systems. By using PrULN, the following are expected; (1) the generalization capability of the learning networks will be improved, (2) more sophisticated stochastic control will be obtained than the conventional stochastic control, (3) designing problems for the complex systems such as chaotic systems are expected to develop, whereas now the main research topics for the chaotic systems are only the analysis of the systems.
Concerning the development of firmware, important problems include reducing the number of processes and the periods for testing and validation while preserving quality and reducing the overall time taken for product development. To solve these problems, one effective approach is to reduce the number of design iterations by ensuring adequate drawing up, review and validation of product specifications at an early stage of the design. One method of doing this is to write formal descriptions of the specifications for review and validation. An approach frequently adopted is to use Harel Statecharts as suggested in the OMT (Object Modeling Technique) method. However, design methods that address directly the conditions in Harel diagrams have yet to be established. The article proposes a design method that addresses the control conditions affecting basic product operations. It consists of describing the conditions that affect the entire product as a combination of basic operations, and to describe controls with reference to these basic operations. Furthermore, we propose that by using simulations of the Statecharts so created, practical actuators should be directly controlled to create a validation environment for the specifications. We have applied the proposed method to the development of control software to be built into prototype air-conditioning equipment. The preparation of the Statecharts for the prototype equipment demonstrated the effectiveness of this approach.
An expert system which supports analysis of customers' data for sales forecast with multiple regression analysis is proposed and is evaluated. Traditionally, a stepwise method by computer which improves an intermediate regression equation step by step, or an interactive method by human experts which omits some intermediate calculations just to suit the occasion has been used. In case features of the data are changed by market trends and commercial media, it is difficult to get an enough result in practical time by these methods. In order to solve the above problem, an analysis method is proposed. This method uses the following two steps in conjunction. In the first step, a rule-base of an expert system which is constructed by eliciting expert analysts' knowledge gets an intermediate regression equation by checking statistical data calculated by a multiple regression analysis tool. And in the second step, expert analysts improve the equation interactively. Finally, effects of the proposed method is clarified in application to a deriving process of sales forecast equation for apparel goods.
In recent years, many methods of model reference adaptive control system (MRACS) for a linear time-varying (LTV) plant are proposed. These methods are assumed that the structure of plant parameters is known in advance. However it is difficult to get priori information of plant parameters. In this paper, a MRACS design for a LTV system based on high order estimator (HOE) is proposed. By applying dynamic certainty equivalence (DyCE) to LTV plants, a new MRACS law of LTV system is derived without knowing the structure of the plant parameters. The MRACS law is generated by using high order derivatives of a estimated parameter, so that robust HOE with a normalization signal and a modification for the system introduced. Our proposed method can attain better performance than conventional methods, such that the estimation with variable forgetting factor (VF) and the gradient projection method (GPM). The robust HOE establishes the boundedness of all the estimated parameters under the condition that the estimated parameter and the first derivative of the parameter are bounded. It is shown that all signals in the adaptive loop is bounded and the output error converges to a closed set. The proposed method is compared to the familiar schemes, the gradient projection method and the estimation based on forgetting factor through numerical simulations, and the effectiveness of our proposed method is shown consequently.
Duplication and output comparison of the duplicated modules are widely used for error detection in high-reliable systems such as fault-tolerant, self-checking, and fail-safe systems. The duplication scheme can not detect errors if the duplicated modules produce coincident erroneous output. We propose implementation of time-diversity and present experimental results which show effect of the proposed implementation. The experiment proved that occurrence of coincident errors is decreased by time-diversity.
Effective utilization of information has become more important in information systems of inter/intranet era, such as real time decision support systems or mass-customization systems. In this paper, a concept of information potential is proposed for considering the effectiveness of information utilization. This concept is analogous to the potential field theory, where the information potential shows the degree of necessity of information in the decision and the information power shows the effect of information in the decision. Based on the proposed concept, a design method of decision support systems is proposed, where the intelligent agent technology is used for evaluating information potential. This method is applied to a real time decision support system, especially a securities dealing decision support system to confirm its effectiveness.
This paper is concerned with torque disturbance attenuation for manipulators based on SP-D (Saturated Proportional and Differential) position/force control method. The proposed SP-D position/force control method is derived from quasi-natural potential energy. It is applied to a system which consists of a manipulator and a compliant environment. Without the torque disturbance, the proposed control method can ensure asymptotic stability of the closed-loop system. Even though the torque disturbance exists, stability of the system can be guaranteed in the sense of L2. Furthermore, performance of attenuation against the torque disturbance is analyzed by using L2 gain. The proposed SP-D position/force control method is implemented on a two-link planer manipulator system. Effectiveness of the proposed control method is confirmed experimentally.
The full length inspection of tubes in SG (Steam Generator) to detect defects using the signal measured by ECT (Eddy Current Testing) is performed in PWR (Pressurized Water Reactor) plant. According to technological advances of NDE (Non-Destructive Examination), the estimation of defect shape is required. In this study, by characterization of ECT signals which using imitative defects, we express relation with defect shape in approximate functions, and estimate defect shape using this function. As the result that we estimated depth and volume of defects, relative error of 80% data was restrained less than 19.7% in depth estimation and 27.2% in volume estimation.