JPH09101822A - Posture stabilization device - Google Patents

Abstract

(57) 【Abstract】 PROBLEM TO BE SOLVED: To realize general-purpose and adaptive disturbance compensation even when there is no knowledge or information about the disturbance that causes the disturbance of the posture of the moving object, and to perform highly accurate posture control of the moving object. Achieve stabilization. SOLUTION: At least one secondary signal obtained by processing a control target value from a control target signal generator 2 is input, and the input signal is subjected to a signal combining operation using a learning function algorithm. A learning control unit 6 that applies the obtained signal as a supplemental control signal to the control signal for the main actuator 5 from the main control logic unit 4 in a feedforward manner is provided in parallel with the main control logic unit 4. .

Description

DETAILED DESCRIPTION OF THE INVENTION

(Technical Field of the Invention) [Technical Field of the Invention] [Prior art] [Problems to be solved by the invention] Means for solving the problems (FIGS. 1 to 3) Embodiments of the invention (a) Description of the first embodiment ( (FIGS. 4 and 5) (b) Description of second embodiment (FIG. 6) (c) Description of third embodiment (FIG. 7) (d) Description of specific application example and simulation result thereof (FIGS. 8 to 8) (Fig. 13) (e) Other effects of the invention

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for stabilizing the posture of a moving object, and more particularly to the characteristics (parameter, size, The present invention relates to a device for controlling or stabilizing the posture of a moving object with high accuracy when the temporal / frequency characteristics are unknown. Here, the moving object refers to, for example, an artificial satellite, a spacecraft, a ship, a submarine, an aircraft, etc., whose posture is an important control purpose.

[0003]

2. Description of the Related Art Conventionally, some methods have been adopted for controlling disturbances acting on a moving object. When the mechanical characteristics of the disturbance have been determined, the influence of the disturbance can be canceled by performing calculation using a mathematical model of the disturbance and performing feedforward compensation. In the case of artificial satellites and spacecraft, this method is used to drive the onboard antenna and manipulator arm. However, in order to adopt such a method, accurate prior knowledge such as the mass of the movable body, the moment of inertia, the position of the center of gravity, and the moment arm length is required, and when these inertial loads change, It is necessary to change the mathematical model of the disturbance each time.

On the other hand, the mechanical characteristics of many natural environmental disturbances are uncertain. Therefore, in order to realize highly accurate attitude control / stabilization, it is necessary to estimate and compensate for these disturbances onboard. As a method for this, there are a disturbance estimation observer and a Kalman filter. In both methods, a mathematical model of the dynamics of the entire control system including disturbance is prepared in the on-board computer, and the state variables of the mathematical model are updated so that the observed values of the attitude sensor and the predicted values of the same mathematical model match. I will do it.

In this case, since the disturbance is also included in the state variables, it is necessary to construct the dynamic state equation of the disturbance.
For example, if the magnitude of the step disturbance is unknown, the first derivative with respect to time of the disturbance state variable is zero, if the initial value and the gradient of the ramp disturbance are unknown, the second derivative of time is zero, and the sinusoidal disturbance is In the case of, the state equation of the disturbance is constructed in such a way that the frequency is known and follows the equation of harmonic vibration. The output of the filter using such state variables converges to the true value of each physical quantity if the mathematical model correctly represents the actual dynamics, and thus complete real-time compensation of disturbance is also possible.

However, in particular, in the case of natural environment disturbance, it is general that the dynamic model of each disturbance factor can be known only inaccurately, and it is quite difficult to accurately construct the dynamic model of total disturbance. Have difficulty. Therefore, if the modeling error is large, the compensation method by such an approach becomes less effective. Moreover, since a dynamic mathematical model simulating the entire actual control system must be configured in the on-board computer, there is a concern that the scale of the installed software and the processing time will increase.

Next, as another example of the artificial disturbance, a disturbance that adversely affects the posture may occur during the operation of the propulsion engine for accelerating the moving object. As an example, there may be a case where there is misalignment or offset with respect to the system center of gravity of the thruster for changing the orbit when the orbit of the artificial satellite or spacecraft is changed. In this case, it is impossible to know the thruster thrust, the above misalignment and offset, the position of the center of gravity, etc. accurately, and it may fluctuate over time. It is desirable to compensate in real time. The same method as described above can be applied to this as well, but the problems described above cannot be avoided.

Another approach to disturbance suppression is to use a so-called robust control system such as H∞ control.
This is to seek a controller that stabilizes the closed loop system by simultaneously weighting the sensitivity function of the closed loop including the controlled object to reduce the weight. This method has the problems that it requires prior knowledge about the frequency characteristics of the disturbance, that the control law design algorithm is complicated, and that the order of the controller is generally high.

[0009]

As described above, in the prior art, when attempting to compensate for a disturbance, it was necessary to incorporate a mathematical model of the disturbance or prior knowledge about the disturbance into the design of the controller. Therefore, when the assumed mechanical characteristics of the disturbance and the actual characteristics of the disturbance are different, the disturbance compensation effect may be significantly reduced. In addition, since the controller is designed for a specific disturbance model, there is a problem that it cannot be used for general purposes.

The present invention was devised in view of such problems, and realizes general and adaptive disturbance compensation even when there is no knowledge or information about the disturbance that causes the disturbance of the posture of the moving object. An object of the present invention is to provide a posture stabilization device capable of achieving highly accurate posture control / stabilization of a moving object.

[0011]

FIG. 1 is a block diagram of the principle of the first invention. In FIG. 1, 1 is a moving object whose posture is to be stabilized, and 2 is a control target value of the posture of the moving object 1. A generated control target signal generator 3 is a posture sensor that detects the posture of the moving object 1, and 4 is a main control logic that generates a control signal based on the deviation of the fed back output value of the posture sensor 3 from the control target value. And 5 are the main control logic unit 4
The posture stabilizing device of the present invention is a main actuator which generates a predetermined torque based on a control signal from the moving object 1 and acts on the moving object 1.
The posture of the moving object 1 is stabilized by controlling so that the posture becomes the control target value from the control target signal generator 2.

In the first invention, the learning control section 6
Are provided in parallel to the main control logic unit 4. The learning control unit 6 receives at least one secondary signal obtained by processing the control target value from the control target signal generation unit 2 and performs a signal combining operation using a learning function algorithm on the input signal. The signal obtained by the application is applied as a supplementary control signal from the main control logic unit 4 to the control signal for the main actuator 5 in a feedforward manner (claim 1). At this time, the learning control unit 6 may be configured by a neural network, and in this case, the adjustment of the synaptic connection weight of the neural network is performed based on the control signal from the main control logic unit 4 (claim 2).

FIG. 2 is a block diagram of the principle of the second invention. As shown in FIG. 2, the second invention has substantially the same structure as that of the first invention shown in FIG. In the second invention, the learning control section 6A and the auxiliary actuator 7 are connected to the main control logic 4 by the same reference numerals.
And the main actuator 5 in parallel.

Here, the learning control unit 6A receives at least one secondary signal obtained by processing the control target value from the control target signal generation unit 2, and applies a learning function algorithm to the input signal. The signal obtained by performing the above-described signal combining operation is output to the auxiliary actuator 7 as an auxiliary control signal. Further, the auxiliary actuator 7 generates a predetermined torque based on the auxiliary control signal from the learning control section 6A, and applies this torque to the torque generated by the main actuator 5 as feedforward torque to act on the moving object 1. It is a thing.

At this time also, the learning control section 6A may be constructed by a neural network. In this case,
Adjust the synaptic connection weight of the neural network,
Control target value from control target signal generator 2 and attitude sensor 3
It is performed based on the attitude control error signal which is the difference from the output value from (Claim 4). FIG. 3 is a block diagram of the principle of the third invention. As shown in FIG. 3, the second invention is also configured almost similarly to the first invention shown in FIG. In the third invention also, the learning control section 6B is provided in parallel with the main control logic section 4, but the learning control section 6 of the third invention is also shown.
B uses the control target value from the control target signal generator 2 and the output value of the attitude sensor 3 as input signals, and supplements the signals obtained by subjecting the input signals to signal combining operation using a learning function algorithm. As a control signal, the control signal from the main control logic unit 4 to the main actuator 5 is applied in a feedforward manner (claim 5).

At this time also, the learning control section 6B may be constructed by a neural network. In this case,
Adjust the synaptic connection weight of the neural network,
Control target value from control target signal generator 2 and attitude sensor 3
It is performed based on the attitude control error signal, which is the difference from the output value from (6). In the posture stabilizing device of the first to third inventions described above, the moving object 1 under the environment in which the disturbance acts
In order to stabilize the posture of the robot, it is possible to use the control target value from the control target signal generator 2 and the posture sensor 3 without using the prior knowledge or information about the structure of the disturbance such as the parameters of the disturbance and the temporal / frequency characteristics. Using the obtained attitude angle and attitude angular velocity information of the moving object 1 and the learning function algorithm of the learning control units 6, 6A, 6B, the disturbance actually acting on the moving object 1 is adaptively estimated in real time. To do.

Then, the result is added to the control signal output from the main control logic unit 4 to the main actuator 5 as an auxiliary control signal in a feedforward manner (claims 1, 2, 5, 6). The torque generated by the main actuator 5 by the auxiliary actuator 7 is applied as a feedforward torque to act on the moving object 1 (Claims 3 and 4), whereby the posture disturbance effect on the moving object 1 due to an unknown disturbance is generated in real time. Can be offset and compensated with.

That is, according to the present invention, the disturbance acting on the main actuator 5, the disturbance acting on the moving object 1, and the disturbance acting on the posture sensor 3 can be compensated without using any prior information. is there.

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. (A) Description of First Embodiment FIG. 4 is a block diagram showing the overall configuration of a control system of a posture stabilizing device as a first embodiment of the present invention. In the figure, the same symbols as those already described are the same parts. , The detailed description thereof will be omitted.

This first embodiment corresponds to the first invention described above with reference to FIG. 1, and as shown in FIG. 4, a portion surrounded by a dotted line is a first addition newly added by the present invention. It is a characteristic part of the embodiment. As shown in the dotted line area, the learning control unit 6 is provided in parallel with the main control logic unit 4 in the first embodiment. In the first embodiment, the control target value R (t) from the control target signal generator 2 is
Input signals to the learning control unit 6 (neuro-algorithm) are synthesized by performing operations of proportional, differential and integral. Here, the disturbance compensation performance can be adjusted by changing the gain constants of proportional, differential and integral (proportional gain K _P, differential gain K _D, integral gain K _I ). Further, the locations of the disturbances targeted by the present invention are shown as disturbance 1, disturbance 2, and disturbance 3 in FIG. Disturbance 1 is the main actuator 5
The disturbance 2 is a disturbance acting on the moving object 1, and the disturbance 3 is a disturbance acting on the attitude sensor 3.

Then, the learning control section 6 of the first embodiment is
The secondary signals R _1, R _{2 and} R ₃ that have been subjected to the proportional, differential and integral operations as described above are input, and these input signals R _1, R _{2 and} R ₃ are input _.
A signal s obtained by subjecting R _{2 and} R ₃ to a signal combining operation using a neuro algorithm as a learning function algorithm is fed from the main control logic unit 4 to a control signal for the main actuator 5 as an auxiliary control signal. It is applied in a forward manner.

At this time, the learning control section 6 is composed of, for example, a neural network as shown in FIG. 5, and adjusts the synapse coupling weight of the neural network based on the control signal ε from the main control logic section 4. . Further, as the teacher signal t, originally, a time function of disturbance should be given, but since it is assumed that it is unknown, zero is given here.

The learning function algorithm is an algorithm that changes the parameters of its constituent elements according to a predetermined law so that a desired motion is achieved with the passage of time or as the number of motions increases. In this embodiment, a neuro algorithm (neural network) having high versatility is adopted as the learning function algorithm.

As shown in FIG. 5, the neural network is a combination of a large number of multi-layered neurons by synaptic connection. A plurality of input functions in the first layer are weighted and summed, and the result is Nonlinear function f in neuron
It has an input non-linear mapping function that obtains an output s (control signal) by repeating the operation of converting in each layer.

Now, in order to make the neural network learn a certain time signal as the teacher signal t, the difference between the output signal s of the neural network and the teacher signal t is always reduced in time so that each synapse connection is made. When the weights are adjusted, a signal close to the teacher signal will always be output for the given input function. This effect is a so-called learning function.

It is known that the neuro algorithm can approximate an arbitrary teacher signal t by increasing the number of neurons in the hidden layer. Therefore, this can be applied to general-purpose online learning of its temporal structure for a wide range of disturbances. Since the posture disturbance acting on the moving object 1 generally depends on the posture and the angular velocity, in the first embodiment, a plurality of secondary signals R _{1 obtained} by linearly processing the control target value R (t). _, R _{2 and} R ₃ are combined and input to the neural network (learning control unit 6). In addition, the synapse coupling load is controlled by the moving object 1
The time is automatically adjusted so as to always suppress the occurrence of the posture angle θ (t), the posture angular velocity, and the like.

In the first embodiment of the present invention constructed as described above, the moving object 1 under the environment in which the disturbances 1 to 3 act.
In order to stabilize the posture of the control target, the control target from the control target signal generator 2 is not used without using prior knowledge or information about the structure of the disturbances 1 to 3 such as parameters of the disturbances 1 to 3 and temporal / frequency characteristics. The disturbances 1 to 3 actually acting on the moving object 1 are adaptively estimated in real time using the value R (t) and the neuro algorithm of the learning control unit 6.

Then, the result is added to the control signal ε output from the main control logic unit 4 to the main actuator 5 as an auxiliary control signal s in a feedforward manner,
The posture disturbance effect on the moving object 1 due to an unknown disturbance can be offset and compensated in real time without using any prior information. As described above, according to the first embodiment of the present invention, the natural environment disturbance or the artificial disturbance acting on the moving object 1 is learned and estimated in real time by using the learning function algorithm of the learning control unit 6. By adding the result as a feedforward to the actuator control signal ε generated by the main control logic unit 4 as an auxiliary control signal s, the adverse effect of the disturbance acting on the moving object 1 on the posture is offset, and the posture is accurately adjusted. Control / stabilization can be performed.

In particular, since the learning function algorithm is used, it is possible to adaptively respond to a wide range of disturbances without using any prior knowledge about the mechanical characteristics of the disturbances 1 to 3. Further, when the disturbance compensation function is incorporated into the design of the control law by the conventional technology, the main control logic unit 4 itself undergoes a large change as compared with the case where it is not.
In the present embodiment, since the disturbance compensation function is carried out by the learning function algorithm of the dedicated learning control unit 6, there is also an advantage that the existing main control logic unit 4 can be supported without any change.

(B) Description of Second Embodiment FIG. 6 is a block diagram showing the overall configuration of a control system of a posture stabilizing device according to a second embodiment of the present invention. In the figure, the same reference numerals as those mentioned above are used. Indicate the same part, and detailed description thereof will be omitted. This second embodiment corresponds to the second invention described above with reference to FIG. 2, and as shown in FIG. 6, a portion surrounded by a dotted line is a second addition newly added by the present invention.
It is a characteristic part of the embodiment. As shown in the dotted line area, in the second embodiment, the learning control unit 6A and the auxiliary actuator 7 are provided in parallel with the main control logic unit 4 and the main actuator 5.

Then, the learning control section 6A of the second embodiment.
In the same manner as in the first embodiment, the secondary signals R _1, R _{2 and} R ₃ subjected to the operations of proportional, differential and integral are input, and these input signals R _1, R _{2 and} R are input. _A signal s obtained by subjecting ₃ to a signal combining operation using a neuro algorithm as a learning function algorithm is output to the auxiliary actuator 7 as an auxiliary control signal. Further, the auxiliary actuator 7 generates a predetermined torque based on the auxiliary control signal s from the learning control unit 6A, applies this torque to the torque generated by the main actuator 5 as feedforward torque, and acts on the moving object 1. It is what makes me.

At this time, also in the second embodiment, the learning control section 6A is constituted by a neural network as shown in FIG. 5, for example, and the control target signal generating section 2 adjusts the synaptic coupling weight of the neural network.
A linear filter G is applied to the attitude control error signal which is the difference between the control target value R (t) from the attitude sensor and the output value from the attitude sensor 3.
It is performed based on the operation (s) (for example, proportional and derivative operations).

According to the second embodiment of the present invention constructed as described above, it is possible to obtain the same effect as that of the first embodiment. (C) Description of Third Embodiment FIG. 7 is a block diagram showing the overall configuration of a control system of a posture stabilizing device as a third embodiment of the present invention, in which the same reference numerals as those used above denote the same parts. , The detailed description thereof will be omitted.

This third embodiment corresponds to the third invention described above with reference to FIG. 3, and as shown in FIG. 7, a portion surrounded by a dotted line is a third addition newly added by the present invention. It is a characteristic part of the embodiment. As shown in the dotted region, in the third embodiment, the learning control unit 6B is provided in parallel with the main control logic unit 4 as in the first embodiment.

Then, the learning control section 6B of the third embodiment.
Is the control target value R (t) from the control target signal generator 2,
An input signal is a secondary signal obtained by performing proportional and derivative operations on the output value of the posture sensor 3, and a learning function algorithm is applied to these input signals R _1, R _{2 and} R ₃ . The signal s obtained by performing the signal synthesizing operation using the neuro-algorithm is applied as a supplemental control signal from the main control logic unit 4 to the control signal for the main actuator 5 in a feedforward manner.

At this time, also in the third embodiment, the learning control unit 6B is constituted by a neural network as shown in FIG. 5, for example, and the control target signal generation unit 2 adjusts the synaptic coupling weight of the neural network. A predetermined operation [G] for the attitude control error signal which is the difference between the control target value R (t) from the attitude sensor 3 and the output value from the attitude sensor 3.
(S)] is applied.

According to the third embodiment of the present invention constructed as described above, it is possible to obtain the same effect as that of the first embodiment. (D) Description of Specific Application Examples and Simulation Results Thereof In the following, the results of computer simulation will be described in order to more specifically show the effects of the present invention, particularly the first embodiment.

FIG. 8 is a block diagram showing an example in which the first embodiment is applied to a wheel control system of an artificial satellite. In this FIG. 8, the apparatus of the first embodiment is applied to the steady attitude control by the wheel of the artificial satellite. Things are shown. The attitude stabilizing device shown in FIG. 8 is for stabilizing the attitude of an artificial satellite (satellite dynamics) 11 as a moving object, and includes a control target signal generator 12, an attitude sensor 13, an attitude control logic unit 14, It is composed of a reaction wheel 15 and a learning control unit 16. These codes 12 to 16
The components indicated by are corresponding to the control target signal generation unit 2, the posture sensor 3, the posture control logic unit 4, the main actuator 5, and the learning control unit 6 of the first embodiment (FIG. 4), respectively.

9A to 9C are graphs showing simulation results of the application example shown in FIG.
(A) shows the disturbance torque of an orbital period which acts on the artificial satellite 11, FIG.9 (b) shows the attitude angle response of the artificial satellite 11 when a disturbance compensation measure is not taken, FIG.9 (c)
8 shows the attitude angle response of the artificial satellite 11 when the disturbance compensation is performed by the device shown in FIG.

As is clear from a comparison between FIGS. 9B and 9C, when the disturbance torque as shown in FIG. 9A acts, the disturbance compensation by the device shown in FIG. 8 is performed. It can be seen that steady posture stabilization is achieved by performing. 10 is a block diagram showing an example in which the first embodiment is applied to a thruster control system of an artificial satellite. The attitude stabilizing device shown in FIG. 10 has a structure similar to that shown in FIG. Artificial satellite (satellite dynamics)
It is for stabilizing the posture of 11, and is composed of a control target signal generating unit 12, a posture sensor 13, a posture control logic unit 14, a thruster 15A and a learning control unit 16.

The components indicated by the reference numerals 12 to 16 are also respectively included in the control target signal generator 2, the attitude sensor 3, the attitude control logic unit 4, the main actuator 5 and the learning controller of the first embodiment (FIG. 4). It corresponds to 6. The thruster 15A is controlled according to a predetermined thruster modulation rule. 11A to 11C are graphs showing simulation results of the application example shown in FIG. Here, as shown in FIG. 11A, it is assumed that a stepwise disturbance torque acts on the artificial satellite 11 due to, for example, the center of gravity offset of the orbit control thrust.

FIG. 11B shows the attitude angle response of the artificial satellite 11 when no disturbance compensation measure is taken, and FIG.
10 shows the attitude angle response of the artificial satellite 11 when the disturbance compensation is performed by the device shown in FIG. This FIG.
As is clear from a comparison between (b) and FIG. 11 (c),
When a disturbance torque as shown in FIG. 11 (a) acts,
By performing the disturbance compensation by the device shown in FIG.
It can be seen that quick posture stabilization has been achieved.

12 (a) and 12 (b) are diagrams showing an example of load movement by the manipulator arm mounted on the artificial satellite. In the example shown here, the two-link type manipulator arm 20 mounted on the artificial satellite 11 is used. Moves each joint while gripping the load 21, and the state shown in FIG.
It is assumed that the state shown in (b) is alternately repeated periodically. The same thruster control system as that shown in FIG. 10 is used.

Normally, in such a case, the disturbance torque or disturbance angle acting on the satellite body is calculated based on the information such as the angle and angular velocity of each joint, the load 21, the mass of each link of the manipulator arm 20, the moment of inertia, the position of the center of gravity, and the like. We calculate the momentum and use it to perform feedforward compensation. However, here, in order to emphasize the effectiveness and versatility of the present invention, the learning control unit 1 is not used at all.
The unknown disturbance was compensated by the neuro algorithm of No. 6.

13 (a) to 13 (c) are graphs showing the simulation results, FIG. 13 (a) shows the disturbance angular momentum generated by the movement of the manipulator arm 20, and FIG. 13 (b) shows the disturbance. FIG. 13C shows the response of the satellite body when no compensation measure is taken, and FIG. 13C shows the response of the satellite body when the disturbance compensation is performed by the device shown in FIG.

As is apparent from a comparison between FIGS. 13B and 13C, even when the disturbance angular momentum as shown in FIG. 13A occurs, the apparatus shown in FIG. It can be seen that quick and steady posture stabilization is achieved by performing the disturbance compensation. In this case, even if a disturbance estimation observer or a Kalman filter is used with the cause of the disturbance unknown and the disturbance cannot be represented by a simple dynamic model, it is expected to be difficult to estimate / compensate it well. On the other hand, it is clear from the result of FIG. 13C that adaptive compensation can be performed by using the learning function algorithm (neuro algorithm) of the learning control unit 16.

(E) Others In each of the above-described embodiments, the case where the neuro algorithm is used as the learning function algorithm has been described, but the present invention is not limited to this. Further, in each of the embodiments and application examples described above, the case where the present invention is applied to the attitude control / stabilization of an artificial satellite is described, but the present invention is not limited to this, and the spacecraft is not limited thereto. The present invention can be applied to attitude control / stabilization of various moving objects such as ships, submarines, and aircrafts, and in any case, the same operational effects as those of the above-described embodiments and application examples can be obtained.

Further, in each of the application examples described above, only the first embodiment is described, but the second embodiment and the third embodiment.
It goes without saying that the same applies to the embodiments and similar results can be obtained.

[0049]

As described in detail above, according to the posture stabilizing device of the present invention, the influence of an unknown posture disturbance acting on a moving object can be learned in real time and compensated for,
It excels in adaptability and versatility as compared with the conventional technology, and can achieve highly accurate posture control / stabilization of a moving object even if there is no knowledge or information about the disturbance that causes the disturbance of the posture of the moving object. effective.

Further, when the disturbance compensation function is incorporated into the design of the control law in the prior art, the main control logic unit itself undergoes a large change as compared with the case where it is not, whereas in the present invention, the disturbance compensation function is dedicated. Since it is assigned to the learning function algorithm of the learning control unit, there is an effect that the existing main control logic unit can be dealt with without any change.

[Brief description of the drawings]

FIG. 1 is a principle block diagram of the first invention.

FIG. 2 is a principle block diagram of the second invention.

FIG. 3 is a principle block diagram of the third invention.

FIG. 4 is a block diagram showing an overall configuration of a control system of the posture stabilizing device as the first embodiment of the present invention.

FIG. 5 is a diagram for explaining a neuro algorithm.

FIG. 6 is a block diagram showing an overall configuration of a control system of a posture stabilizing device as a second embodiment of the present invention.

FIG. 7 is a block diagram showing the overall configuration of a control system of a posture stabilizing device as a third embodiment of the present invention.

FIG. 8 is a block diagram showing an example in which the first embodiment is applied to a wheel control system of an artificial satellite.

9A to 9C are graphs showing simulation results of the application example shown in FIG.

FIG. 10 is a block diagram showing an example in which the first embodiment is applied to a thruster control system of an artificial satellite.

11A to 11C are graphs showing simulation results of the application example shown in FIG.

12 (a) and 12 (b) are diagrams showing an example of load movement by a manipulator arm mounted on an artificial satellite.

13A to 13C are graphs showing simulation results when the first embodiment is applied to the application example shown in FIG.

[Explanation of symbols]

 1 Moving Object 2 Control Target Signal Generator 3 Attitude Sensor 4 Main Control Logic 5 Main Actuator 6, 6A, 6B Learning Controller 7 Auxiliary Actuator 11 Artificial Satellite (Satellite Dynamics, Moving Object) 12 Control Target Signal Generator 13 Attitude Sensor 14 Attitude control logic unit 15 Reaction wheel (main actuator) 15A Thruster (main actuator) 16 Learning control unit (learning function algorithm) 20 Manipulator arm 21 Load

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location G05D 1/08 G05D 1/08 A

Claims (6)

[Claims] 1. A control target signal generator for generating a control target value of the posture of a moving object, a posture sensor for detecting the posture of the moving object, and a feedback target output value of the posture sensor from the control target value. A main control logic unit that generates a control signal based on the deviation, and a main actuator that generates a predetermined torque based on the control signal from the main control logic unit and acts on the moving object By controlling so as to be the control target value, in a posture stabilizing device for stabilizing the posture of the moving object, at least one secondary signal obtained by processing the control target value is input, A signal obtained by subjecting the input signal to a signal combining operation using a learning function algorithm is used as an auxiliary control signal from the main control logic unit to the main actuator. Learning control unit for applying a feedforward manner to a control signal to, characterized in that are provided in parallel with respect to the main control logic unit, posture stabilization system. 2. The learning control unit is configured by a neural network, and the adjustment of the synaptic connection weight of the neural network is performed based on a control signal from the main control logic unit. Posture stabilization device. 3. A control target signal generator for generating a control target value of the posture of a moving object, a posture sensor for detecting the posture of the moving object, and a feedback target output value of the posture sensor from the control target value. A main control logic unit that generates a control signal based on the deviation, and a main actuator that generates a predetermined torque based on the control signal from the main control logic unit and acts on the moving object By controlling so as to be the control target value, in a posture stabilizing device for stabilizing the posture of the moving object, at least one secondary signal obtained by processing the control target value is input, A learning control unit that outputs a signal obtained by performing a signal combining operation using a learning function algorithm on the input signal, and a learning control unit from the learning control unit. An auxiliary actuator that generates a predetermined torque based on the auxiliary control signal and applies this torque as a feedforward torque to the torque generated by the main actuator to act on the moving object is the main control logic unit and the main control unit. An attitude stabilizing device, which is provided in parallel with an actuator. 4. The learning control unit is configured by a neural network, and the adjustment of the synapse coupling load of the neural network is adjusted based on a posture control error signal which is a difference between the control target value and an output value from the posture sensor. The posture stabilizing device according to claim 3, wherein the posture stabilizing device is performed as follows. 5. A control target signal generation unit for generating a control target value of the posture of a moving object, a posture sensor for detecting the posture of the moving object, and a feedback target output value of the posture sensor from the control target value. A main control logic unit that generates a control signal based on the deviation, and a main actuator that generates a predetermined torque based on the control signal from the main control logic unit and acts on the moving object In the posture stabilizing device for stabilizing the posture of the moving object by controlling so that the control target value becomes the control target value, the control target value and the output value of the posture sensor are input signals, and A signal obtained by performing a signal synthesizing operation using a learning function algorithm is used as an auxiliary control signal from the main control logic unit to the main actuator. Learning control unit for applying a feed-forward manner, characterized in that are provided in parallel with respect to the main control logic unit, posture stabilization system. 6. The learning control unit is configured by a neural network, and the adjustment of the synapse coupling load of the neural network is adjusted based on a posture control error signal which is a difference between the control target value and an output value from the posture sensor. The posture stabilizing device according to claim 5, wherein the posture stabilizing device is performed as follows.