This paper describes some properties of randomly sampled M-sequence, and compares those properties for two methods of sampling. The first method, called method I, is to sample ([k·w_i]+1)'st bit from a k-tuple of an M-sequence, where w_i is a random number having any distribution function between 0 and 1. The second method, called method II, is to sample ([k·X_i]+1)'st bit from a ([k·X_i]+1)-tuple of M-sequence, where X_i is a uniform random number between 0 and 1.
It is shown that the maximum value of the ensemble averaged autocorrelation function in case of method I becomes minimum when w_i is a uniform random number. It is also shown that the maximum value of the ensemble averaged autocorrelation function of randomly sampled sequence generated by method II is smaller than that of the sequence generated by method I.
From these results, we see that method II is better than method I in order to get a binary random sequence having good random properties.

抄録全体を表示

This paper describes a geometrical method of discriminating and quantifying gases using an electronic system composed of an integrated sensor and a microprocessor.
The integrated sensor with three different elements on an alumina substrate was fabricated by using thick film techniques. The elements are kept at around 400°C by a Pt heater mounted on the back of the substrate. Since these elements were made from different semiconductor oxides, they possess different sensitivities to methane as the principal component of the city gas, carbon monoxide generated by imperfect combustion, and ethanol and hydrogen as interference gases. Then the integrated sensor can develop specific patterns corresponding to each gas in terms of histograms of conductance ratios for each element.
The microprocessor discriminates target gases for identification on the basis of similarities calculated by comparing standard patterns stored in the memory and a sample pattern developed by the integrated sensor. The gases are then quantified by using the sensor element with the highest sensitivity to the identified gas.
The experimental results show that methane and carbon monoxide in mixture with ethanol can be identified and quantified without any trouble according by the proposed method.

抄録全体を表示

A lot of fruitful applications of model-based methods to measurement of unsteady fluid flow in a pipe have been reported recently. Those applications, however, are limited to the measurement of the fully-developed flow although the application to the entry flow is strongly needed. The limitation comes from the difficulty in making a mathematical model of the entry flow: a nonlinear problem has to be solved to make the model.
This paper presents a unified method of making a linear model for the entry flow as well as for the fully-developed flow to derive a linear estimator of velocity profiles of the flows in both the outside and the inside of the entrance region of a circular pipe. The linear model is derived by, first, modifying the Navier-Stokes' equation to a linear partial differential equation having an unknown function as an input and, then, lumping its parameters by the finite element method. The Kalman filter theory is applied to this linear lumped parameter model with unknown inputs to yield a linear estimator of an inverse Kalman filter form.
The effectiveness of the estimator in flow measurement is shown by simulation and experiment done for many cases. The experiment is done by using a beam scanning laser Doppler velocimeter which can measure velocity profile in a pipe by scanning the laser beam. This velocimeter was developed originally in the authors' laboratory.

抄録全体を表示

Ascertaining fire phenomena moment by momentt in the course of their progression is essential to detect a fire in the early stage, to issue a fire alarm, to provide adequate instructions for evacuation and to take early action to extinguish the fire. Since a fire is generally accompanied by both physical and chemical phenomena, a multi-dimensional measurement using various kinds of sensors is needed to obtain precise information about the fire behavior. In a multi-dimensional coordinate space where each coordinate axis represents the output of each sensor, a point in the coordinate space (or a position vector) represents the state of the fire behavior at a given time. If we could set up a border in this coordinate space that distinguishes between a fire and a non-fire, we can make a fire/non-fire judgment by observing the motion of the position vector.
This paper describes (1) a filtering method to determine the primary direction of the position vector motion, (2) the determination of the fire/non-fire border, and (3) the experimental results of the fire judgment.
Three kinds of sensor, temperature sensor, smoke sensor and gas sensor, were used in our experiments. To determine the primary direction of the position vector motion in the three-dimensional coordinate space, unwanted fluctuations of the measurements need to be removed. To do this, we used a method that calculates the center of gravity among three points in the coordinate space. We chose this method because it gives equal filtering effect to each of the three axes, it provides a stable point among position vectors, and it requires less computation time.
We set up a non-fire area in the three-dimensional coordinate space based on the position vector motions obtained from a number of non-fire experiments. The non-fire area is expressed as a combination of three rectangular solids. Also, we discuss, using fire experiment data, how the position vector moves within the non-fire area and finally goes into the fire area. This discussion includes the determination of the current state of fire/non-fire behavior and a forecast for the fire, by observing the position vector motion.
We used Newton's backward interpolation formula for the fire progression forecast, and we found that we can forecast the position vector motion of one minute later using up to the second term of the formula. Also, observation of the position vector motion allows us to know the details of fire and non-fire behaviors: for instance, we can distinguish between cigarette smoke and cooking, and we are able to know that when the CO gas concentration is increased in the smouldering fire stage, there is, in the current state, a high degree of danger of ignition.

抄録全体を表示

The controllability Gramian in linear time-invariant systems is closely related to control performances such as disturbance decoupling and perfect regulation. Skelton and his colleague, and Ohara and Kitamori have formulated the controllability Gramian assignment problem and have individually derived necessary and sufficient conditions for a given positive definite matrix to be the controllability Gramian of the state feedback system.
In this paper, all positive difinite matrices which can be assigned as the controllability Gramian of the closed-loop system by state feedback are parametrized. First, such a parametrization for the c-controllability Gramian which is from control input to state variable, is provided; especially for the systems of controllable standard form, the class of assignable Gramians is indicated in the explicit matrix representation. It is also shown that its magnitude in the sense of matrix norm can be unconditionally realized. Second, the assignable d-controllability Gramian being from disturbance is characterized as the sum of a c-controllability Gramian and the negative semidifinite matrix satisfying a Lyapunov equation. Because of the negative semi-difinite matrix, the d-controllability Gramian cannot generally be made arbitrarily small.
Finally, a numerical example is illustrated to demonstrate the availability of the controllability Gramian assignment technique in control system design taking account of the disturbance decoupling property.

抄録全体を表示

In this paper, we will proceed to solve an order ρ input-output linearization problem. We will define an order ρ input-output (I/O) linear system whose Volterra series expansion (V.S.E) around the equilibrium point x₀ is
where W(k) is the k-th order term of the V.S.E. This system represents approximately a linear input-output response when the terms of order larger than ρ are negligible. We will define an order ρ input-output linearization problem as finding nonlinear feedback which make the system order ρ I/O linear. We will propose an order ρ structure algorithm to determine the feedback and identify the class of systems which can be transformed into order ρ I/O linear systems. We will also show that, under suitable conditions, an order ρ I/O linear system can be represented in appropriate coordinates as
ξ=Fξ+Gv+O(ξ, η, v)^ρ+1
η=f(ξ, η)+g(ξ, η)v
y=Hξ
where F, G and H are matrices of real numbers. This implies that the state ξ which dominates the input-output response can be represented in the form of an order ρ state space linear system.

抄録全体を表示

A number of synthesis methods of optimal robust servosystems have been proposed. In most of those, however, the optimality is not necessarily well-defined. For example, the initial state of the servocompensator to be decided is not naturally derived from the viewpoint of optimality criterion. For the case of stepwise reference, Ikeda and Suda clarified that the optimal servosystems cannot be obtained till the performance index is minimized in terms of not only control input but undefined initial state of the servocompensator. Although their results are reasonable and interesting, some of those are available only to the stepwise reference case.
This paper considers the problem of synthesising the optimal robust servosystems of which reference is generated by a linear dynamical system, and generalizes their results. We first introduce the ideal trajectories for state and input variables of the augmented system, such that the resulting output completely coincides with the reference. Using the deviation variables from those ideal trajectories, we define the augmented error system and the corresponding quadratic performance index. The optimal control law can be then obtained by minimizing the performance index in terms of both control input and undefined initial state of the servocompensator, which consists of feedback control to stabilize the augmented system, feedforward control from the reference generator, and some outer signals including the same mode with the servocompensator. The outer signals are not needed if the state variables of servocompensator can be suitably adjusted at the moment of changing in reference input. It is also shown that when reference changes only in a steady state, in the case that the reference is of single-mode such as stepwise signal the optimal control law can be constituted of only state feedback and feedforward control, and in the multi-modes case it cannot generally be constituted without the outer signals.

抄録全体を表示

The problem of estimating both states and parameters of continuous-time linear stochastic systems with time-varying parameters are considered in the framework of nonlinear filtering. On the assumption that the parameters vary slowly, the asymptotic behavior of the conditional probability density function (pdf) of parameters with respect to observations is analyzed using the large deviation theory. It is shown that -lim_ε→0εlog (conditional pdf of parameters) follows a Hamilton-Jacobi equation, where ε is a smallness parameter representing the slowness of variation of unknown parameters.
By rescaling conditional moments based on the asymptotic analysis, and applying a singular perturbation method to equations of the conditional moments, an adaptive estimation algorithm is obtained. Due to its construction scheme, the estimates by the algorithm have the same asymptotic properties as those of the optimal ones. Numerical examples are given to demonstrate the asymptotic optimality.
In addition, it is pointed out that a part of the third-order conditional moments plays an important role in the adaptive estimation problem. This means that the approximation methods of nonlinear filter equations which neglect third-order conditional moments from the outset such as the extended Kalman filter are unsuitable for the adaptive estimation problem.

抄録全体を表示

This paper proposes sensitivity analysis and synthesis of minimum sensitivity structures in continuous time linear state-space systems. A sensitivity function called frequency sensitivity is defined using the partial derivatives of a given transfer function with respect to the elements of coefficient matrices. The frequency sensitivity shows a degree of degradation of the frequency response due to small perturbations of the elements of coefficient matrices. A frequency measure called frequency sensitivity norm is defined using the norm of the frequency sensitivity. It is represented by controllability and observability Gramians. The minimum sensitivity structures, that minimize the frequency sensitivity norm, are synthesized via equivalent transformations. The minimum sensitivity structures are equal to balanced realizations in the wide sense. A numerical example shows that a minimum sensitivity structure has much lower sensitivity than a controllability canonical realization.

抄録全体を表示

In order to control a robotic manipulator, the following three computational problems at different levels must be solved: (1) the determination of a desired trajectory in the task-oriented coordinates, (2) the transformation of its coordinates to the joint-angle coordinates (Inverse kinematics) and (3) the generation of motor command (Inverse dynamics). The above first problem (trajectory determination) is mainly discussed in this paper.
In our previous work, we proposed a mathematical model for formation of trajectories in human arm movements, and then confirmed that hand trajectories predicted by this model were in good agreement with the experimental data. Our model was based on minimization of rate of change of motor command.
In the work presented here, we aim to generate the smooth trajectory of a multi-joint manipulator like that of human arm. Therefore, in order to determine a desired trajectory of the multi-joint manipulator, we define the following objective function: the square of the magnitude of rate of change of torque integrated over the entire movement.
Since this objective function critically depends on the dynamics of the manipulator system, it is very difficult to determine the unique trajectory which yields the best performance. We overcome this difficulty by developping an iterative learning control in which the optimal trajectory and the associated motor command (torque) are simultaneously computed. Mathematically, a multipoint boundary-value problem for nonlinear differential equations provided by dynamic optimization theory is solved by using a Newton-like method in a function space.
In our method, the optimal trajectory and the associated torque can be simultaneously obtained without solving inverse kinematics and inverse dynamics, by only repetitively moving the manipulator and measuring its trajectories.

抄録全体を表示

In the injection moulding, it is important to control precisely injection speed in the injection process and polymer pressure in the packing process. These controlled variables mainly determine mechanical or optical properties of mouldings. Each process is essentially a distributed parameter system, and there is some variation of dynamical characteristics depending on the shape of the mould cavity, temperature of the polymer and that of the mould.
Each process is regarded as a multistage process which is a sequential series of subprocesses having different dynamical characteristics. The subprocess is modelled by a lumped parameter system. Then control system is designed as follows.
(1) state feedback control for each subprocess to make the dynamical characteristics homogeneous through the injection process or the packing process.
(2) servo compensation to improve transient characteristics of the controlled variables.
(3) repetitive control to improve steadystate characteristics and to compensate the variation of the polymer temperature and of the mould temperature.
Finally, the experimental results illustrate that this control method is effective for the improvement of the injection speed response and of the polymer pressure response.

抄録全体を表示

Control of discrete system such as FA system can be regarded as an event-driven type control. In such a system, a controller detects changes of system conditions as events occurred in system, and decides necessary control actions according to system states. In order to make programming of such system control logics flexible, the rule based control method has been adopted, where control logics are represented as ‘if (conditions) then (actions)’ format. A rule based controller selects adaptable rules according to system states using a production system inference engine.
By this method, control logics can be input into computers easily and modified flexibly. However, logical contradictions or incompletions are often embeded in control logics because interractions of rules are implicitly described and behaviors of control rules is difficult to comprehend. Therefore, practically speaking, change of rule contents is not so easy as being expected. For these reasons, control rule verification technique is strongly required to construct flexible rule based control system with guarantee and maintenance of control logics correctness.
In this paper, a rule verification method is proposed, in a case where controlled system can be modelled as a finite state machine and a controller is implemented as a rule based system. In proposed method, interactions between system state transitions and control rule outputs are modelled using Petri net, and the set of reachability states of a controlled system is automatically generated using a Petri net reachability tree analysis technique. Rule verification is performed using a generated reachability tree.
Combinatorial explosion problem in reachability tree generation can be avoided by suppressing not-existent states using control rules outputs. An effectiveness of proposed method is also demonstrated with an example.

抄録全体を表示

An active sensor system using parallel processing circuits is proposed and its characteristics are discussed. From the viewpoint of a model of sensory processing mechanism of active touch, the system uses information of efferent copy and internal model of actuator in order to generate active motions from the local pattern information sensed by local pattern sensors such as tactile sensors.
In this paper, an experimental system and its basic experimental results are also shown. The experimental system is a sensor system for active perception of the shape of two-dimensional objects by tracing the edge of the objects. The system realizes both high speed processing of the local pattern and real time control of the actuator.

抄録全体を表示

Independency between replicated modules is very important in fault-tolerant systems. To achieve the independency, the input signals are sampled at different sampling points and processed separately, so that simultaneous errors caused by transient effects can be tolerated. By the use of a new method called “intelligent error detection”, the errors in a single module can be detected within itself, and the erroneous states can be recovered from the other module.
On the other hand, high-speed error detection is significant in digital control systems, to keep the real-time processing and to reduce the system risk. From this point of view, a high-speed detection method called “priority error detection” is proposed. First, the priorities for error detection rules are evaluated on the basis of their effects on the system risk. Then, the number of error detection rules is compressed according to their priorities, so that the inference time can be greatly reduced.
The relation between the system risk and the number of error detection rules is studied, and it is demonstrated that the proposed method is very useful to the highly safe digital control systems.

抄録全体を表示

The dead time of the flow control system is estimated from measured impulse response. This method is based on the fact that the phase characteristics of the minimum phase shift system can be calculated from the gain curve.

抄録全体を表示

This paper describes the three bean mode laser Doppler velocimeter. This system can measure velocity and moving angle of a solid surface by using the specularly reflected beam from the surface.

抄録全体を表示

A design method for constructing hybrid adaptive control systems is considered. The proposed method is different from the usual one in the point that integral operations are introduced instead of sampling operations.

抄録全体を表示

A method to construct an LQ regulator via eigenvectors of the hamiltonian for system with delay is proposed; By this method, each selected pole of the open-loop system is shifted to the mirror image position w.r.t. each specified line parallel to the imaginary axis.

抄録全体を表示

A synthesis method for digital optimal control system with preview feedforward compensation has been already proposed. This paper shows the closed loop transfer function matrix of this preview control system, and investigates its asymptotic properties with respect to the weight on the control effort in the quadratic performance index.

抄録全体を表示