The estimation method for human cognitive state at man-machine interface was investigated by means of psycho-physiological method. The estimation method is firstly to reduce data patterns seen in the short-time responses of multiple physiological measures, and then to construct discriminate functions which correlate the obtained data patterns with the different kinds of cognitive state. The used physiological measures are eye saccade fraction, eye pupil size, eye blinking rate, skin potential response, heart rate and respiration rate. By using protocol data taken, from a subject during a, cognitive task experiment, three different kinds of discrimination functions were constructed, based on either fuzzy logic method or statistical analysis (quantification method class ?2). The intercomparison of the three functions thus obtained was made with respect to the accuracy for the cognitive state discrimination, along with the general discussion concerning the limitation of the proposed psycho-physiological method and its applicability to personal differences.
It has been already proposed to provide a linear induction motor system acting as a driving source for a linear fork mechanism, which is effective as a next loading mechanism. Although its structure is simple, it has some problems of low efficiency, electromagnetic noise, high attracting forces acting between a moving part and a fixed stator, and high frictional force. In this paper, it is proposed to study a basic structure, of linear synchronous motor which can solve these problems. We have made an experimental equipment on a trial basis of linear fork mechanism by combining the loading mechanism. Its mechanism has generated a thrust of 588 N. Its equipment has realized a high acceleration and a high deceleration of about 1.0 m/s2 with its mouting mass of 100 kg. Last, we have confirmed that a high speed loading in about 2.2 second could be carried out, when the stroke was 0.8m.
We propose a method of object recognition using the knowledge of the object world described in a form of a frame structure. The hierarchical frame structure consists of two levels, an outline image level and an aspect graph level. Using this database, we can recognize input as an object that has the largest degree of similarity between input edge image and outline images and/or aspect graphs in the database. Firstly, we calculate-the degree of similarity between the input image and images in the database on the outline level in order to narrow down the candidates. Next, we calculate the degree of similarity between the input image and the candidates in the database on the aspect level. In this point, it is available to recognize objects on high recognition rates and high speed. In the case that the object world consists of five kinds of chairs composed by polygonal surfaces, we show some experimental results.
This paper proposes a classification method of fuzzy vectors by neural networks. The proposed method can be applied to classification problems where attribute values of each sample are given as fuzzy numbers. We first propose an architecture of neural networks that can process input vectors of fuzzy numbers. The proposed neural networks map a fuzzy input vector to a fuzzy output. Next we derive a learning algorithm from the cost function defined as the maximum squared error between the target output and the level set of the fuzzy output. The derived algorithm can be regarded as an extension of the BP algorithm to the case of fuzzy input vectors. Last we show that the proposed method can utilize not only attribute values of each sample but also experts' knowledge represented by fuzzy if-then rules in learning of neural networks.