In this paper, we deal with disturbance attenuation property of a controller for the visual servo systems. First, based on Lyapunov stability theory, we propose an asymptotically stable controller without exogenous inputs, e. g. the joint torque disturbances and the motion of a target. Next we show that the proposed controller achieves disturbance attenuation in the L₂-gain sense, that is, the L₂-gain of the closed loop system from the exogenous inputs to the regulated outputs can be made sufficiency small with a suitable selection of the feedback gains. Finally, our experiments on an industrial manipulator are shown to evaluate the effectiveness of the proposed controller and L₂-gain performance analysis for the visual servo system.

View full abstract

The purpose of the present paper is to propose a new design method of low order Two-Degree-of-Freedom (TDF) controllers. In general, the TDF controller design is conducted with the sequential approaches in which a feedback controller and a feedforward controller are designed one by one. As a result, in these approaches, it is not easy to design the low order TDF controllers which take advantages of the freedom in TDF con-troller. In this paper, a simultaneous design method of the feedfoward controller and the feedback controller is proposed. Not only H_∞ norm optimization problem but also H₂/H_∞ mixed norm optimization problem can be utilized to design Two-Degree-of-Freedom control system via Linear Matrix Inequality approach. The advantages of the proposed method over the sequential approaches are as follows: (1) controllers which share common dynamics can naturally be derived in this method, (2) the trade-off between the feedback perfor-mance and the target-following performance and the order of the controller can easily be considered in a one-step design process.

View full abstract

PID controllers have been widely used in many industrial processes. Because they have only three control parameters, and we can easily understand their physical meanings. However, it is difficult to tune their parameters in the case where the control system has uncertainties, for example, modelling errors, system fluctuation, and so on. Therefore, we propose a design scheme of discrete robust PID controllers in this paper. That is, we design a PID controller based on the generalized minimum variance control (GMVC) scheme, and present a method how to design a weighting factor included in the GMVC criterion from the viewpoint of the robust stability. Furthermore, we investigate the behavior of the proposed control scheme by applying it to an air pressure control system.

View full abstract

We propose a new ANN learning algorithm based on hierarchical clustering of training data. The proposed algorithm first constructs a tree of sub-learning problems by hiearchically clustering given learning patterns in a bottom-up manner and decides a corresponding network structure. The proposed algorithm trains the whole network giving teacher signals of the original learning problem to the output units, and trains sub-networks giving teacher signals of the divided sub-learning problems to the hidden units simultaneously. The hidden units which learn sub-learning problems become feature detectors, which promote the learning of the original learning problem. We demonstrate the advantages of our learning algorithm by solving N-bits parity problems, a non-liner function approximation, iris classification problem, and two-spirals problem. Experimen-tal results show that our learning algorithm obtains better solutions than the standard back-propagation algorithms and one of constructive algorithms in terms of the learning speed and the convergence rate.

View full abstract

Learning automata select an action from a finite set of their available actions and update their strategy on the basis of response received from the random environment using what is known as a reinforcement scheme. As an environment changes, the ordering of the actions with the performance criterion may vary. If a learning automaton with a fixed strategy is used in such an environment, it may become less expedient with time and even inexpedient. However, using the learning scheme that has sufficient flexibility to track the better actions, makes the performance improved.
In this paper, a variable structure learning automaton network with periodic random environment is proposed. The results of some numerical simulations shown that our model can be used for tracking some periodic non-stationary environments that an upper bound on the period is known.

View full abstract

In this paper, we deal with the fundamental equation of a optical flow estimation by gradient method. We introduce the fundamental equation keeps its form invariant in any coordinate systems, even if coordinates are time varying.
With the above fact, we can calculate a relative optical flow corresponding to the coordinate system, and/or local optical flow in the coordinate system spanned in a local area on the moving object, using the fundamental equation of the same form.
This kind of optical flow implies movements on the corresponding coordinate system. We call it the Local and Relative Optical Flow (LROF), and confirm it by simulation of some examples.
This method extremely saves calculation of optical flow estimation compared with the conventional method, since LROF needs a small calculation only on the region of interest.
LROF may be widely applicable to many problems for estimating movements in local area on an object moving or form-changing in scene, such as strain analysisses or motion estimations of muscle on some medical images.

View full abstract

A computational method of the formal linearization for nonlinear systems is proposed by using a cubic spline interpolation. We introduce a linearizing function that consists of the state variables, their squares, and the cubes, thus the dimension of the linearizing function is determined finitely by the number of state variables. Then we construct the differential equation with respect to the linearizing function. By approximating the nonlinear terms by the cubic spline interpolation, we obtain the linearized form with respect to the linearizing function. We contrive to determine the coefficients of the cubic spline interpolation, so that we can easily get them by simple matrix calculations. From the features of the cubic spline interpolation, a nonlinear system is converted into the piecewise linear systems, but they are continuously connected on the whole domain. A nonlinear filter is synthesized as an application of the method and is verified through numerical examples.

View full abstract

The biorthonormal expansions of piecewise polynomials are considered to be one of the least squares approximation methods for functions. Based on the locality of the sampling functions and their biorthonormal ones in the signal space of piecewise polynomials, a practical method of a least squares approximation by piecewise polynomials is proposed by truncating the duration of the bases, which has the advantage of obtaining approximation results without solving the simultaneous linear equations for expansion coefficients. The relation between the truncation error in the sense of the squares norm and the truncation width is derived as one of the upper bounds of the error. Based on this relation a practical approximation by piecewise polynomials within a specified tolerance of error is obtained with a small delay, so that the proposed method is useful for not only real-time systems such as audio and communication systems, but image coding and compression.

View full abstract

For the purpose of implementation by digital devices, a continuous-time system, which models a system in the real world. should be replaced by a discrete-time one. To have a correspondence between those two kinds of systems, a continuous-time signal must be limited within a signal space that has a freedom in a discrete sense. In digital control theory, it is commonly assumed that the input signal is a staircase one. The degree of freedom with the signal space of staircase functions is one degree per sampling interval. In this paper, we aim at the design of a family of models which includes and generalizes the conventional sampled-data, systems by employing the piecewise polynomial functions of degree (m-1) with a parameter of (m-2) times continuous differentiability. The conventional model is the special case of the proposed family of models. While the conventional model solely depended on the parameter of sampling interval to vary the accuracy of approximation, the proposed model has, in addition, the adaptability by choosing the parameter m appropriate to the input signal. This feature enables more efficient calculation in simulation as shown in a numerical example.

View full abstract

The continuous deadbeat controller which includes some delay-element has already reported and the finite settling responses can be realized in continuous-time domain. This controller is, however, impractical because it is difficult to construct delay-element.
In this paper, we propose a new design method of an optimal continuous deadbeat controller without using delay-element for the multivariable system. The sequence of the optimal control signal which can minimize the quadratic performance index for the manipulated variable and deviation can be easily obtained from the computation of the matrix described by the step response of the system.
Our new controller is composed of both digital deadbeat compensator which can be designed by using sequence of the control signal and a series smoothing filter of the analog element. Finally, we show a numerical example of the 2 input-2 output system.

View full abstract

In this paper, linear discrete-time systems subject to constraints are considered. We consider the conditions for non-violation of such constraints in the presence of exogenous inputs. For this purpose, we define the constrained positively invariant (CPI) set based on the notion of state constrained set and positively invariant set. If the initial state of linear discrete-time system with exogenous inputs lies in CPI set, it is guaranteed that no constraint violations occur. It is also shown that there exists maximal CPI set which is a CPI set and also contains any CPI sets in it. The necessary and sufficient condition for non-violation of constraints in the presence of exogenous inputs is the initial state of linear discrete-time system lies in the maximal CPI set. We investigate the computational procedure of maximal CPI set and derive the condition for finite determination of maximal CPI set. Numerical computation of maximal CPI set is realized by employing linear programs.

View full abstract

Pigeons have causing lot of problems. As one of the solution of these bird problems, we have carried out to search for the birds by using the robot which is installed the camera. The camera is in outside environment, so the input image is so changeable by time and seine change. To adapt outside environmet, we have seeked the pigeons using the model of environmental form and the method of frame subtraction. We use Hough transform to search for the model of environment. And to adapt the seine change, we have used the method of interpolating background using frame subtraction.

View full abstract

Recently, a design of model matching and adaptive control systems under the presence of measurement noise is reported by selecting a degree of characteristic polynomial of state observer. But this methods can be only used for the case where the plant has an stable inverse. In this paper, we will show that the idea is also applicable to the case of pole placement. This method can be carried out by simple calculation comparing with the design of Kalman filter or H_∞ control. So the proposed method is adequate for an adaptive control in which the controllers must be calculated on line.

View full abstract

The present paper deals with the linear quadratic dynamic game problem for the descriptor systems Ex_k+1=Ax_k+Bu_k+Cv_k+Dw_k, where E is in general singular matrix and the system structure is noncausal. Condition are obtained under which a strategy of a leader controller forces the two other controllers (follower) to use a team-optimal solution while the follower controllers of the second-level are in fact a Nash equilibrium. Two different information structeres of the followers, that is, the open-loop infomation structure and the closed-loop infomation structure, are considered in the paper.

View full abstract

We developed a new method to construct a digital filter with recurrent neural networks by using the back propagation through time (BPTT) learning. We presented a systematic method to create teacher signals by multiplying the Maximum length linear shift register sequence (M sequence) and desired dynamics in the frequency domain and by transforming it to the time domain. Digital filters were constructed from fully cross-coupled and partially cross-coupled recurrent neural networks. Partially cross-coupled recurrent neural networks modeled after Direct Type I and II of IIR digital filter acquired appropriate properties through the learning. Our method is featured to be construct any complicated filter with desired dynamics from a recurrent neural network.

View full abstract

This paper proposes a novel approach for solving a sub-class of combinatorial problem, which allows decomposition into subproblems that are interconnected by sub-problems objective functions. The optimization problem is solved by an interactive search process based on the Immune System response regulatory mechanism. The original problem constraints and objective function are satisfied by solving the sub-problems independently and by modifying dynamically the objective function of each sub-problem. This approach is applied for finding firing sequences of a sub-class of Petri net defined as P1-net, whose firing sequence problem is decomposable into sub-problems. Considering discrete event systems that can be modeled by P1-net, the objective is to find a sequence of events that leads the system from a known initial state to a desired final state. Simulations were conducted and our approach successfully found 100% of feasible and near-optimal solutions for simulated P1 models. The characteristics of our approach indicate that it may be applied in relatively large systems with distributed characteristics.

View full abstract

In a reactor plant, reactor power is the most important parameter for plant protection, control and monitoring. Hence, it is necessary to verify that the nuclear instrumentation system (NIS) satisfies the design criteria of response as well as accuracy. As the response depends on the condition in which the instrumentation is operated, it is desirable to develop an in-service method for the response measurement. Thus, applicability of auto-regressive (AR) model to in-service measurement of the response was examined for the NIS at MONJU, the prototype fast breeder reactor in Japan. In order for AR model to be applied to measurement of sensor response time, it is necessary that the physical quantity to be sensed has sufficiently higher frequency fluctuations than the sensor's frequency response limit.
Response of a NIS can be measured by use of AR model fitting when the noise fluctuations from the neutron detection process inside the detector have higher frequency and magnitude than those inherent in the reactor dynamics. This can be confirmed by checking that there is no coherence between outputs from any two detectors. In case of the wide range monitoring (WRM) system and the source range monitoring (SRM) system of Monju, the fluctuations caused by detection process are indeed greater (both frequency and magnitude) than those from the reactor dynamics. The WRM time constant was measured at 60 to 70ms by AR model fitting process while that of SRM could not be measured precisely because of shortage of the sampling rate.

View full abstract

In recent building air conditioning systems, it is difficult to clarify qualitative physical relationships among components of the system because of diversification and limited sensor points. In the stochastic qualitative reasoning, which we have proposed, fault diagnosis is performed with the stochastic qualitative model which represents the behavior of a target system with probabilities. However, the probability tuning process in the model is the most difficult and important task in the model construction.
This paper proposes an approach to automatic model generation to cope with this problem. Firstly, propagation rules and functions are formalized with several characteristic parameters. And it is confirmed that the agreement rate, which represents the degree of agreement of the simulation with the measured behavior, is smoothly changed according to the change of parameters. Then, based on the tentative model, which is constructed from a target system design, the reasonable qualitative model is generated with measured field data by using the characteristic parameter tuning method.
This method has applied to a real building air conditioning system. As a result, We can generate a stochastic qualitative model which expresses the normal behavior of the target system.

View full abstract

With an increased utilization of electronic systems, we are disturbed by the noises from other systems. In the array-antenna systems, the recived signal is composed of any sources including undesirable noise sources. Finding out their direction of arrival (DOA) is important in order to eliminate those sources, and DOA is a very important problem for EMC. Many techniques have so far been proposed, but DOA has not been well established. For example, most of these techniques are based on the assumption that the noise sources are a point source. Such an assumption is not appropriate in real situations, because the sources are generally distributed in the space and also their shapes are not always Gaussian. This paper presents a new method for finding the DOA of distributed man-made noises using artifitial neural networks. In particular, we use the Hopfield network. This is due to the feature of this network that we can expect rapid convergence of network energy to a minimum point. The sources are characterized by arbitrary parameters (e. g., amplitude, frequency, incident angle, and distribution). Simulation results are presented and discussed for the case of a single source characterized by two parameters, namely, the incident angle and the distribution of source. Additionally, we also present the results for the case of two sources, and we study the influence of S/N ratio on our method.

View full abstract

A repeaterless WDM transmission system having twelve 10Gbits/s-signal wavelengths, four wavelengths in 1.53μm band and eight wavelengths in 1.55μm band, is proposed. Using two wavelength bands lead to advantages of the utilizability of the wide wavelength range to avoid the nonlinear interaction between neighboring channel, and ease of demuxing by wavelength selective couplers. BER's for the transmission experiment through a 100-km dispersion shifted fiber show well balanced characteristics among wavelengths and additional optical power margin of 9dB. The same constitution as the 12-WDM system is applied to the cases of 4-WDM and 8-WDM transmission. The BER properties of these cases are similar to those for 12-WDM case. This confirms that the present WDM system structure is useful for increasing transmission capacity without changing the system construction.

View full abstract