This paper studies approximate realizations of pseudo-rational impulse responses and discusses the related convergence problems Introducing a subclass S of pseudo-rational impulse responses, we prove that various delay-differential systems are included in this class: systems of retarded type, neutral type and infinitedelay systems. Since pseudo-rational impulse responses can be represented in terms of distributions, one can discretize them by replacing some distributions with suitable linear combinations of Dirac δ-distributions and derive the approximate impulse responses. The resulting approximate impulse responses are shown to admit finitedimensional realizations and these realizations are given in the observable canonical form by the aid of the dual form of Fuhrmann realizations, in relation to the topologically observable realizations of pseudo-rational impulse responses. Estimation of approximation rate, as well as the proof of convergence, of the approximate models is given. A neutral delay-differential system is presented as an example along with some numerical results in the final section.
In the controller design of a linear timeinvariant system, it is important to improve feedback properties such as robust stability and sensitivity. In the multi-input multi-output case, these properties can be estimated by using the singular values of return difference matrix. The design method to obtain better singular-value-plots is desired. In this paper, we study how to tune the weight matrix of performance index and/or covariance matrix of noise in the LQG (Linear Quadratic Gaussien) theory to get a desired singular-value-plot. First, the property of singular-value plots of return difference matrix of a system designed by LQG theory is examined from the view-point of tuning weight. Second, a distance between real singular-value-plots and a desired plot is defined, and the weigh of performance index is numerically determined by quasi-Newton method so that the distance is minimized.
This paper considers a fault diagnosis problem of linear dynamical systems. Specifically, we derive conditions for fault detectability and fault distinguishability, and also we derive a fault diagnosis algorithm which decides the faulty element from the system observation. As the main result of this paper, graphical necessary and sufficient conditions for fault distinguishability are derived. The conditions are given in terms of the system representation graphs which depict the system structures. The conditions are essential in carrying out the fault diagnosis algorithm at two stages; to check the distinguishability assumed in the algorithm; to find the covering set defined in the algorithm. An example is given to see how the algorithm works.
The optimal dynamic logic is developed for safety monitoring systems, which have two types of contradictory failures: a fail-dangerous and a fail-safe failure. The term “dynamic” means that the logic is reconstructed every time when new monitored data are obtained. The optimal dynamic logic is determined by a switching function of a history of monitored data, and it minimizes an expected damage caused by the two failures at the next monitoring time. Illustrative examples show various characteristics of the dynamic logic as compared with ordinary static logic. A calculation method of the switching function is demonstrated in detail.
This paper presents the derivation of a dynamic model for a robot mechanism including a closed kinematic chain. The model can be used to calculate joint torques and forces from the joint displacements, velocities, and acceleration. The expression in this paper, which is derived using the motor algebra, can be applied without any modification to mechanisms composed of any lower kinematic pairs with single degree of freedom. A comparison of computational complexity with the conventional Newton-Euler formuation is also discussed and it is shown that the formulation derived here offers the better method of dynamic computation of robot mechanisms.
In this paper, the characteristics during human gait stopping are analyzed by two methods, experiment and simulation. In the experiment, two force plates were used to measure the force to each leg separately. A simple model consisting of a rigid body and two massless legs was used in the simulation. Parameter values characterizing the process such as the motion of the point of application supporting times, and step lengths were measured and analyzed by the regression method. The input data for the simulation were determined based on this experimental analysis. It was assumed that the stopping process required two steps after the steady state. The simulation results are quantitatively compared with the characteristics of the point of application measured experimentally. The comparison shows that the simulation method is appropriate.
Step and impulse response tests have been performed for cavity-mounted pressure transducers subjected to pneumatic signals of 0.1 msec or shorter risetimes and 0.25 to 2kPa amplitudes. The system is modeled as a second order nondimensional differential equation with the static resistance obtained experimentally and the dynamic one proportional to square root of the natural frequency. Theoretical response curves have a good agreement with experimental ones. The overshoots of responses are lower and later for larger input pressures, and for larger volume ratios of cavity to hole. The oscillatory frequency after the overshoot is, unless the cavity is extremely flat, nearly equal to the natural frequency based on the orifice inertance with the double end-correction and the adiabatic capacitance of cavity. The response within an early few oscillations may be reasonably predicted by only a simple static resistance of constant discharge coefficient.
Two-dimensional humidity distribution can be measured based on the infrared absorption principle, when an infrared beam is introduced into the humidity field from its one side through an optical fiber and the damped beam is caught from the other side of the field by another optical fiber. This paper proposes a measuring system and presents its calibration results. Moist air of known dew-point as 20, 25, 30, 35 and 40°C of 16cm and 30cm optical path length and infrared light of 1.86μm wave length are used in the calibration test. Apparent absorption coefficents obtained based on Beer's law are 0.043 (kPa·m)-1 and 0.034 (kPa·m)-1 and the accuracy of the present measurement is estimated to be ±3g/m3 and ±2g/m3 for 16cm and 30cm optical path, respectively. These results suggest the necessity of case by case calibration at present. Some proposals for improving the elements of the measuring system are also discussed, particularly for measuring low humidity.
A radiometric correction method for linear array sensor image is proposed. This method removes stripe noise pattern caused by the errors of detector characteristics of the sensor. In this method, mean and variance of output image are calculated for each detector. Highpass filters are applied to the statistics to estimate gain and offset errors of the detectors. In this paper, power spectra of mean and variance of each detector are studied by assuming a 2-dimensional auto-correlation model for the original image. Based on the power spectrum models of the statistics, Wiener filters (least square filters) are derived. In the method the filter parameters are determined adaptively by the sensor image. An experiment was made by simulating image of linear array sensor using a LAN-DSAT image. The stripes were removed in the corrected image and the decrease of the rms error of the image intensity showed the effectiveness of the method.
In this paper we present a stereo matching method based on a network algorithm. The network to transport the characteristic features from one image to anothor is constructed as an electrical network connected by variable resistors. By this analogy the characteristic features can be considered as electrical current. Using this network, matching problem can be transformed as a computation of current distribution satisfying the constraints for the physical surface. In the step to find the distribution, we utilize the lateral inhibition theory and change iteratively the values of resistors. Several algorithms were devised for stereo matching, such as a method using correlation coefficient, dynamic programming or the relaxation method. But these methods were not satisfactory because they need vast amount of calculation or they were not capable to obtain enough resolution. In this paper, the method for detection of corresponding points by means of the network algorithm resolves the deficits of conventional methods. The method is able to use the images or those differences directly, furthermore, by the method a pixel of one image can be matched to any point between the pixels of the other image. Because of these facts, fine resolution can be obtained. Experimental results are shown to confirm effectiveness of this method to navigate the locomotive vehicle. Although this method is defined on one dimensional, extension to two dimensional is easily performed by use of three dimensional network.
Given a desired output for a class of dynamical systems, certain kinds of iterative learning control are shown to be effective in the sense that the output converges to the desired one with repeating operations. One of those methods can be applied to a robot manipulator with high nonlinearities even though its physical parameters are not known. In this case, the velocity vector of the robot motion is regarded as the output and the input at the present operation is modified by the derivative signal of the error at the previous operation, which is the difference between the output and the desired one. Through those iterative operations, the desired motion given to the robot is obtained if some conditions for the modification of the input are satisfied. However, when the output is contaminated by noise, the derivative error is extremely different from the real derivative error. In such a case, it is difficult to realize the desired motion with high accuracy since the modified input itself becomes noisy. To overcome this difficulty, we propose an alternative learning control method for robot manipulators, which uses the error directly. To show the effectiveness of this method, firstly the robot dynamics is linearlized around the desired motion. Next, it is proved that by this method the robot motion converges to the prescribed motion trajectory with repeating operations. Moreover, from a practical point of view, it is discussed how the parameters concerned with this method should be determined. Finally, this learning control method is practically applied to a robot manipulator with three degrees of freedom and its effectiveness is shown experimentally.
There exist many kinds of biped locomotion as locomotive means. A good example of natural and dynamic biped locomotion is human walking. In fact, human walking utilizes the gravity effect very skillfully and does not depend on the ankle torque at the supporting leg so much except for the kick-phase. The robot in this study realizes such a dynamic biped locomotion with kick action only during double-legs-supporting-phase. Its step length is about 44 cm and the walking cycle is 0.9∼1.0 sec. In its control, a hierarchical control structure is adopted in which each joint is locally controlled. Moreover, it is shown by the consideration of angular momentum that the kick action makes a biped locomotion more stable.
One of authors has shown in the previous paper that hard self-excited oscillations can occur in a hydraulic sevomechanism with an asymmetrically underlapped spool valve by delivering a step input beyond a certain critical value, even if the neutral position of the valve is stable. In this paper, it is shown that there exists an absolutely stable region in the system, where no self-excited oscillation is induced by any input disturbance. To study the global stability of the system, the fundamental system equations containing marked nonlinearities are analyzed by a digital simulation procedure by using Runge-Kutta-Gill method. Conclusions obtained are as follows: (1) Stability of the system is divided into three parts, that is, (a) a soft self-excited region, (b) a hard self-excited region, and (c) an absolutely stable region. (2) The local stability limit of the valve neutral position actually coincides with the absolute stability limit of the system for Λ<0.7, but they separate into two curves for Λ>0.7 and the hard self-excited region appears between them. (3) To asymmetrize the valve underlap is effective not only to stabilize the valve neutral position, but also to make the system absolutely stable.