The purpose of this paper is to derive the general limiting form of the continuous-time Kalman filter in the case that the input covariance matrix tends to infinity. The only assumption for the system to be imposed is left invertible. The Silverman's inversion algorithm can be used to obtain the general limiting form. A lower order filtering equation is also derived to obtain the stable filter. It is shown here that the reduced order filter is equivalent to the partial state estimator derived previously by the present authors for a linear stochastic system with completely unknown disturbances. From this observation, a sufficient condition for the filter to be asymptotic stable is presented.

抄録全体を表示

This paper deals with the problem of designing a model following control system for a multivariable linear, time-invariant, continuous system, in which the state variables of the plant are not available, but the parameters are assumed to be known, and only the inputs and outputs can be measured.
In order to solve the problem, all the zeros of the plant are assumed to be in the left-half plane, and it is also assumed that the relative degree in each row of the transfer matrix of the plant is not higher than that of the reference model.
First, based on the assumption that all the state variables of the plant can be measured, the controller is synthesized such that the output errors between the plant and the reference model, or the output errors between augmented system with respect to the plant and that of the reference model become to zero asymptotically. Then it is shown that the transfer matrix from the reference inputs to the outputs of the plant is matched to that of the reference model.
Second, in the case where the state variables of the plant are not available, a scheme which estimates the state variables of the plant from only inputs and outputs is employed. But the control inputs to the plant can be given without carrying out the explicit state estimation.
Finally, simulation results of the different cases are shown to justify the proposed scheme.

抄録全体を表示

The purpose of this paper is to present a fault location method for a large scale plant.
Operators in a control room should take measures to meet the situation for maintenance of the plant function when a detector informs them of something wrong. On the other hand, the matter of plant is often detected at failure propagated devices rather than at the failed device (failure origin). In addition, the number of detectors which inform them of something wrong increases exponentially. Therefore, because they can not locate the failure origin in a short time, they can not take corrective counter operations.
Fault location we propose is executed in the following procedure. First, a digraph which shows the failure propagation network of the plant is constructed from devices of the plant or failure modes of devices (nodes) and failure propagation relations between nodes (branches). Second, some candidates for the failure origin are selected in the digraph by back-tracing on branches starting from the failure propagated nodes. Third, the candidates are screened by back-tracing starting from normal state nodes, failure propagation probabilities and failure propagation time between adjacent nodes. Fourth, the failure propagation probabilities and the failure rates of the devices are used to evaluate the priority ranking among the screened candidates.
The method makes it easy to find out the failure origin and to take proper counter operations when the detector informs operators of something wrong.

抄録全体を表示

In many modeling problems, the system matrices have a number of fixed zero entries determined by the physical structure of the system while the remaining entries are a priori known functions of the, free (physical) parameters. Then, from input-output data (e.g. transfer function), can we uniquely determine the unknown free, parameters? This problem introduced the concept of structural identifiability (abbreviated to SI).
This paper discusses SI of the system whose parametrization is affine. Firstly we remark the relation between SI and the structural controllability (and observability), and give new algebraic equivalent conditions for SI. Based upon these results, secondly, we obtain all the SI structured systems the indeterminate entries of that are unrelated, and it is made clear how SI of compartmental systems is influenced by the way of connection among its subsystems.

抄録全体を表示

未知パラメータ,変動パラメータを含む制御対象に対し,要求される性能を規範モデルの形式で表現し,規範モデルの出力に制御対象の出力が追従するように制御装置のパラメータを自動調整するモデル規範形適応制御系は,従来より活発に研究されてきた.
連続時間系においては,Monopoli¹⁾以後,実用的な見地から制御対象の出力の微分値を含むことなく制御系を構成する手法について研究が進み,相対次数が2以上の場合の拡張誤差を導入した設計法²⁾を初めとして,現在では大域的に安定な制御系が構成できるまでに至っている^3),4).しかしその設計法の適用にあたっては,制御対象の相対次数が既知であることが前提であり,実用上の障害となっている.
これに対し,相対次数に依存しない適応制御系の構成に関する研究^5),6)も進められているが,その多くは制御入力の決定にあたってその時点の入力の値を必要とする形式をとっており,実用的ではなく,さらに制御対象の出力の微分を間接的に含むといった欠点がある.
本論文では以上の点を踏まえ,従来とは異なる入力合成則と拡張誤差を導入したモデル規範形適応制御系の構成法について述べる.これにより従来の手法では不可能であった,制御対象の出力の微分値を含まず,相対次数に依存しない適応制御系が実現される.なお,本手法においては,拡張誤差の零収束性から直ちには出力誤差の零収束性が保証されない.これについては,サーボ系の構成理論を利用して状態変数フィルタを設計することにより,ある条件のもとで,出力誤差が零に収束することを示した.合せて,本手法の有効性を,数値実験により確認した.

抄録全体を表示

In a recent paper, we have proposed an adaptive observer whose parameter adjusting law is determined by equations forcing the errors into decreasing exponentially. The convergence rate of the observer is specified by a constant gain which is included in the parameter adjusting law. However, use of constant gains occasionally causes excessive feedback in the adaptive loop of the observer so that the identification processes are sensitively affected by the noise which is mixed in the measurement. Hence, it is difficult to make identification values fast convergent and free to measurement noise simultaneously.
In the present paper, we introduce variable gains for the improvement of the abovementioned defect. A sufficient condition for the convergence of errors is obtained.
The effect of introducing variable gains is confirmed experimentally by applying the parameter identification problem of a secondorder thermal system. The results of the experiment show that the effect of observation noise on identification processes is extremely reduced by the use of variable gains.

抄録全体を表示

This paper presents a new design method of discrete adaptive control system for nonminimum phase plants with unknown dead time.
In the proposed method an adaptive control is carried out using a controller designed by a certain decomposed representation of the unknown plant.
In this paper the two cases are considered regarding the dead time of the plant. First, we consider the case where is no a priori knowledge regarding the dead time of the plant. Next, the case where the approximate value of the dead time is known is described.
Moreover, the connection between the proposed method and ordinary model reference adaptive control or adaptive pole assignment is clarified.
Finally, the results of computer simulation of adaptive control applied to stable nonminimum phase plants with and without dead time are included to illustrate the effectiveness of the proposed method.

抄録全体を表示

This paper deals with an optimization problem for a discrete-time system composed of a continuous subsystem and a discrete subsystem. The continuous subsystem has a continuous state and a continuous control. On the other hand, both the state and the control variables of the discrete subsystem are integers of the value 0 or 1. In addition, a constraint is introduced in order to represent the interactions between continuous and integer variables. In this problem, the performance index, the state equation and the constraint are assumed linear. Therefore, the problem treated here can be formulated as a mixed-integer linear program with staircase structure.
The purpose of this paper is to develop a decomposition method for solving mixed-integer linear programs with such special structure. The basic idea of the proposed algorithm is to decompose the problem in the same way as the Glassey's nested decomposition method in linear programs and to solve restricted master programs of mixed-integer type iteratively. A sufficient condition for optimality is obtained. Even if the procedure terminates without satisfying the optimality condition, at least a feasible solution is always obtained. Then, the proposed algorithm is expected to be efficient for finding a good suboptimal solution quickly. With an application to a dynamical planning problem of blending raw materials, it is observed that the algorithm requires less computing time than other methods.

抄録全体を表示

Scheduling rules for asynchronous, concurrent systems are constructed based on a systematic analysis of tasks given to the systems. First, it is assumed that the tasks can be represented as propositions such as “to attain the goal state” or “to maintain the system in normal states” etc., which inevitably include the notion of time (tense) such as “in the future”, “at the next time” etc. Then, based on the general framework called “tense logic” dealing with the qualitative properties of time such as transitivity, discreteness and linearity, each proposition representing the task is transformed into a kind of transition diagram which indicates the action necessary to be done at each time instant. Incorporating it with the Petri net representation of the systems, the scheduling rules are constructed in a systematic way, i.e., the firing of transitions in the Petri nets necessary to be prohibited is clarified.

抄録全体を表示

To determine the statistical response of a pulse rate meter, the Campbell's theorems are noted as usefull method. However, in these theorems, the input pulses are assumed to be the delta function. In the practical circuit, the input voltage has a rectangular wave form, then the output voltage is limited at the input pulse height.
In case of high pulse rate, the output reaches to the saturation voltage, and the output calcurated by using the Campbell's theorems is different from the experimental output.
This paper shows a new method of approach to determine the statistical response of the pulse rate meter when the input pulse rate arbitrarily and saturation of the output changes can not be negligible.
The equations obtained by this method have several features such as applicability to wide range of the input pulse rate, facility for the numerical calculation, and others.
Some examples are presented which illustrate the essential features of the response. These examples clarify the effect of the parameters of the input on the response, and also show the statistical fluctuation of the transient response at high pulse rate.

抄録全体を表示

This study proposes a new method for estimating the epicardial potential distribution from the measured body surface potential mapping. The method uses the boundary integral equations to describe the relation between body surface and applies the mode matching method to solve these equations nummerically. In order to suppress the instability of higher mode estimation of epicardial potentials, the expansion series is truncated optimally, and Phillips-Twomay constrained least square method is applied to determine the expansion coefficients.
Animal experiment was done by a canine in which electrodes were attached both on body surface and epicardium. Epicardial potentials estimated from the measured body surface potentials were compared with the measured epicardial ones. The correlation coefficients between measured and estimated epicardial potentials were ranged from 0.5 to 0.8 with average 0.669, but the main features of the measured epicardial potentials were well recognized in estimated ones.

抄録全体を表示

This paper proposes procedures for statistical design of linear-phase ARMA filters. The key idea of our procedures is that low-order ARMA filters are used to approximate high-order linear-phase FIR filters instead of high-order AR filters. Firstly consistent unit-pulse and covariance sequences are generated by the use of a high-order FIR filter. Secondly a low-order ARMA filters of linear-phase characteristic is obtained by solving normal equations constructed from the unit-pulse and covariance sequences. Computer simulations for low-pass filters show that the resulting low-order ARMA filters enjoy sharp cutoff and linear-phase characteristics with only 20-30th order.

抄録全体を表示