JP2023139729A - Servo control device, servo system and servo control method - Google Patents

Abstract

To obtain a servo control device that enables highly accurate and stable control of an actuator.SOLUTION: A servo control device 2 comprises: a command generation unit 5 that generates command information 11 for controlling an actuator; a state quantity prediction unit 7 that simultaneously calculates a predicted value of a state quantity of the actuator or a mechanical system 4 driven by the actuator, and an index for representing a reliability of the predicted value; a correction amount calculation unit 8 that calculates a correction amount for correcting command information 11 based on the predicted value and the index; and a servo control unit 6 that controls the actuator based on the command information 11 corrected using the correction amount.SELECTED DRAWING: Figure 1

Description

The present disclosure relates to a servo control device, a servo system, and a servo control method that control actuators.

A servo control device that controls a motor of a machine tool controls the current supplied to the motor so that the position or speed of a driven object, such as a tool or a table that fixes a workpiece, follows a command. When moving a driven object along a movement trajectory instructed by a machining program, the servo control device controls the motor while precisely managing the position of the driven object. In motor control, a trajectory error, which is an error between a commanded position and an actual position, may occur due to the effects of various disturbances. Disturbances that cause trajectory errors include lost motion caused by friction when the motor reverses the direction of rotation, and vibrations in the mechanical system that occur when the motor is accelerated or decelerated. Conventionally, various methods have been proposed to reduce trajectory errors caused by disturbances.

One method for reducing trajectory errors caused by disturbances is to acquire data on the rotational direction position or rotational speed of the motor shaft using a sensor, etc., and then adjust the amount of correction of the command value based on the position or speed data. A method of correcting a command value by using a model for estimation is known. However, if sufficient consideration is not given to disturbances that may occur during motor control, even if correction is performed using a model, the effect of correction may not be apparent, or the command value may be changed excessively due to correction. On the contrary, it may worsen the trajectory error. Furthermore, if the data input to the model is disturbed by a disturbance or the like, or if there is an unknown input, there is a possibility that unexpected corrections will be made. As described above, even if the command value is corrected using the conventional technology, highly accurate and stable control of the motor may become difficult due to the influence of disturbances. Therefore, in order to improve the accuracy of correction, it has been considered to evaluate the reliability of the estimated value output from the model and utilize the evaluation results for control.

Patent Document 1 discloses that by inputting information representing the physical state of the machine tool into a neural network, the amount of variation in the position of the machine tool components or the amount of variation in the distance between the components is estimated, and the amount of variation is corrected. It is disclosed that in an apparatus for calculating a quantity, the reliability of an estimated variation quantity is evaluated. The device disclosed in Patent Document 1 calculates the adjusted correction amount based on the reliability evaluation result.

JP2020-179429A

The device disclosed in Patent Document 1 determines the amount of variation at the time of performing estimation, and then evaluates the reliability of the determined amount of variation. According to this procedure, the reliability evaluation result is obtained after the estimation of the amount of variation. Therefore, when correcting the command value, corrections adjusted based on the results of evaluating the reliability of the amount of variation in the past are necessary. amount will be used. Since reliability is evaluated after the process of estimating the amount of variation is repeated multiple times, a certain amount of time is required for calculations to obtain reliability evaluation results. In the technique of Patent Document 1, since a correction amount adjusted based on the results of evaluating reliability of past fluctuation amounts is used, the trajectory error may not be reduced by correction in some cases. For this reason, according to the technique disclosed in Patent Document 1, it may not be possible to reduce the trajectory error through correction, so there is a problem in that highly accurate and stable control of an actuator such as a motor is difficult.

The present disclosure has been made in view of the above, and an object thereof is to obtain a servo control device that enables highly accurate and stable control of an actuator.

In order to solve the above problems and achieve the objectives, a servo control device according to the present disclosure includes a command generation unit that generates command information for controlling an actuator, and a state quantity of the actuator or a mechanical system driven by the actuator. a state quantity prediction unit that simultaneously calculates a predicted value and an index representing the reliability of the predicted value; a correction amount calculation unit that calculates a correction amount for correcting command information based on the predicted value and the index; and a servo control section that controls the actuator based on command information corrected using the amount.

The servo control device according to the present disclosure has the effect of enabling highly accurate and stable control of the actuator.

A diagram showing a configuration example of a servo system according to Embodiment 1. A diagram showing a configuration example of a motor and mechanical system of a servo system according to Embodiment 1. An explanatory diagram conceptually showing predicted values of state quantities and variations in predicted values in Embodiment 1 A diagram showing a modification of the servo system according to the first embodiment Flowchart showing the procedure of processing executed by the servo control device of the servo system according to the first embodiment A diagram showing a configuration example of a correction amount calculation unit included in the servo system according to Embodiment 2. A diagram showing an example of an evaluation function used to calculate an evaluation value in the correction amount calculation unit of Embodiment 2. A diagram showing a configuration example of a servo system according to Embodiment 3 Diagram for explaining response characteristics identified in the servo system according to Embodiment 3 A diagram showing a configuration example of a servo system according to a fourth embodiment A diagram showing an example of a hardware configuration for realizing a servo control device of a servo system according to Embodiments 1 to 4.

Below, a servo control device, a servo system, and a servo control method according to embodiments will be described in detail based on the drawings.

Embodiment 1.
FIG. 1 is a diagram showing a configuration example of a servo system 1 according to the first embodiment. The servo system 1 includes a servo control device 2, a motor 3 as an actuator, a mechanical system 4 driven by the motor 3, and a learning section 9. Servo control device 2 controls motor 3 .

The servo control device 2 includes a command generation unit 5 that generates command information 11 for controlling the motor 3, a servo control unit 6 that controls the motor 3 based on the command information 11, and a state of the motor 3 or the mechanical system 4. It has a state quantity prediction section 7 that predicts the amount, and a correction amount calculation section 8 that calculates the correction amount for correcting the command information 11. The state quantities include the position, speed, acceleration, or torque of the motor 3, the current flowing through the motor 3, the position or speed of a driven body included in the mechanical system 4, and the like.

The command generation unit 5 outputs the generated command information 11. The command information 11 includes at least one command value among the position, speed, acceleration, or torque of the motor 3 and the current flowing through the motor 3. The command generation unit 5 is realized by a device such as a numerical control device or a motion controller. The servo control unit 6 controls the position, speed, and acceleration of the motor 3 at every servo control period. The servo control unit 6 monitors the actual operating state of the motor 3 and adjusts the power supplied to the motor 3 as needed so that the operating state of the motor 3 matches the command value. The servo control device 2 drives the motor 3 by applying a voltage to the motor 3. The motor 3 is connected to a mechanical system 4 to be driven. The mechanical system 4 operates as the motor 3 rotates.

FIG. 2 is a diagram showing a configuration example of the motor 3 and mechanical system 4 of the servo system 1 according to the first embodiment. The motor 3 includes a servo motor 21 and a motor end position detector 22 attached to the servo motor 21. A specific example of the motor end position detector 22 is a rotary encoder. The motor end position detector 22 detects the motor end position, which is the rotation angle of the servo motor 21 . The motor end position detector 22 outputs the detection result of the motor end position.

The mechanical system 4 includes a coupling 23 that transmits the power of the servo motor 21, a ball screw 24 that is rotationally driven by the servo motor 21, a ball screw nut 25 that converts the rotation of the ball screw 24 into linear motion, and a ball screw. It has a table 26 fixed to a nut 25. The ball screw 24 and the ball screw nut 25 constitute a mechanism for transmitting the power of the motor 3 to the table 26, which is a driven body. Ball screw 24 is connected to servo motor 21 via coupling 23. The ball screw nut 25 is fitted onto the ball screw 24, and the table 26 and the ball screw nut 25 are driven together.

Assuming that the detection unit of the rotation angle detected by the motor end position detector 22 is "degrees", the rotation angle can be calculated from the table 26 by multiplying the motor end position value by the ball screw lead and dividing by 360 degrees. It can be converted to the length in the moving direction. The ball screw lead is the amount of table movement per rotation of the servo motor 21. The motor end position detector 22 determines the motor end position converted to the length of the table 26 in the moving direction, and outputs a value indicating the converted motor end position.

The control by the servo control unit 6 is not limited to semi-closed control that uses the motor end position for feedback control, but may be full closed control that uses the machine end position, which is the position of the table 26, for feedback control. When the feedback control by the servo control unit 6 is full closed loop control, a machine end position detector 27 and a machine end position detector head 28 are attached to the table 26. The machine end position detector 27 is, for example, a linear encoder. The machine end position detector head 28 is, for example, a linear encoder head. The machine end position detector 27 outputs the detection result of the machine end position.

Information on the motor end position detected by the motor end position detector 22 or information on the machine end position detected by the machine end position detector 27 is input to the servo control unit 6. The position detected for feedback control is called a feedback position. In the following description, a case where the feedback control by the servo control unit 6 is semi-closed control will be exemplified.

The servo control unit 6 acquires information on the motor end position from the motor 3. The servo control unit 6 outputs feedback information 15 which is information on the motor end position. The learning unit 9 generates a prediction model 16 and outputs the generated prediction model 16. Command information 11 , feedback information 15 , and prediction model 16 are input to state quantity prediction unit 7 . The state quantity prediction unit 7 calculates a predicted value of the state quantity of the motor 3 or the mechanical system 4 and a reliability index. The reliability index is an index representing the reliability of the calculated predicted value. The state quantity prediction unit 7 outputs predicted state quantity information 12 which is a calculation result of a predicted value, and index information 13 which is a calculation result of a reliability index.

Predicted state quantity information 12 and index information 13 are input to the correction amount calculation unit 8 . The correction amount calculation unit 8 calculates the correction amount for correcting the command information 11 based on the predicted value of the state quantity and the reliability index. The correction amount calculation unit 8 outputs correction information 14 which is the calculation result of the correction amount. Command information 11 and correction information 14 are input to the servo control unit 6 . The servo control unit 6 corrects the command value shown in the command information 11 by calculation using the correction amount. The servo control unit 6 controls the motor 3 based on the command information 11 corrected using the correction amount.

Next, details of the state quantity prediction unit 7 will be explained. The state quantity prediction unit 7 receives the command information 11 and the measured value of the state quantity shown in the feedback information 15 as input, and uses a prediction model 16 to predict the current or future state quantity. The prediction model 16 is a trained model for predicting the current or future state quantity from the actual measured value of the state quantity and the command information 11. One example of the state quantity of the motor 3 is the motor end position. One example of the state quantity of the mechanical system 4 is the machine end position. The current time is defined as the time point when the state quantity prediction unit 7 predicts the state quantity. The future is defined as a time point after the time when the state quantity prediction unit 7 predicts the state quantity.

Here, let e _N be the state quantity when the servo control cycle is the Nth cycle, and μ _N be the predicted state quantity that is the state quantity predicted in the Nth cycle. For example, when the servo control cycle is the Nth cycle, the state quantity prediction unit 7 calculates the value of the predicted state quantity μ _N+i at the N+ith cycle. i is a natural number greater than or equal to 0. The state quantity prediction unit 7 outputs the value of the predicted state quantity μ _N+i , that is, the predicted value of the state quantity, to the correction amount calculation unit 8 at the timing of the N+i cycle.

In this way, the state quantity prediction unit 7 predicts the state quantity for the state at the time when the predicted value of the state quantity is calculated, or the state quantity for the state at a time later than the time when the predicted value for the state quantity is calculated. Then, the predicted value of the state quantity is output at a timing that coincides with the time point of the predicted state quantity. By matching the timing of outputting the predicted value from the state quantity prediction unit 7 with the time of the predicted state quantity, the servo control device 2 can maintain the command information 11 even when there is a delay in processing or communication. Appropriate corrections can be made.

The state quantity prediction unit 7 simultaneously calculates the predicted value of the state quantity and the reliability index. The state quantity prediction unit 7 uses the prediction model 16 to obtain a distribution of predicted state quantities, that is, a probability distribution of predicted state quantities. The state quantity prediction unit 7 calculates the predicted value and reliability index of the state quantity based on the actual measured value of the state quantity and the probability distribution.

Here, an example of the prediction model 16 and an example of calculation of the predicted state quantity and reliability index using the prediction model 16 will be described. In the first embodiment, the state quantity prediction unit 7 calculates a probability distribution of predicted state quantities from the prediction model 16, and calculates a predicted state quantity and a reliability index from the probability distribution of the predicted state quantities.

Gaussian process regression can be used for the prediction model 16. Gaussian process regression is an example of a probabilistic model that assumes that the predicted state quantity is a random variable that follows a specific distribution. When calculating predicted state quantities and reliability indicators using Gaussian process regression, the following calculations are performed, for example.

Assume that sampling is performed to obtain input data x and output data y of the prediction model 16 while the motor 3 is operating. Input data x is a command value. The output data y is a predicted state quantity. Let x _j be one of the input data x obtained by sampling, and y _j be one of the output data y obtained by sampling. j is a natural number. Assuming _{that input} data x _{1 , .} , is determined by the following equation (1). The variance σ ² (x _M+1 ) is determined by the following equation (2). Let C _M be a Gram matrix. k is a vector whose elements are the values of the kernel function when each of the input data x ₁ , . . . , x _M and new input data x _M+1 are used as arguments. k is expressed by the following equation (3). Let c be a scalar value obtained by adding a precision parameter to the value of each kernel function whose argument includes new input data x _M+1 . The accuracy parameter represents the accuracy of the prediction model 16. The standard deviation σ(x _M+1 ), which is the basis of the reliability index, is found by calculating the square root of the variance σ ² (x _M+1 ). Note that in the following explanation, m(x _M+1 ), which is a predicted value of output data y _M+1 , may be expressed as predicted value y ^* _M+1 .

m(x _M+1 )=k ^T・(C _M ^-1 )・y...(1)
σ ² (x _M+1 )=c−k ^T・(C _M ⁻¹ )・k (2)

Since the probability distribution p ⁽ y* _{M+1) of the predicted value y*M+1} ^for new input data xM ₊₁ follows a Gaussian distribution, the probability distribution p(y ^* _{M +1} ) is calculated using the following formula ( 4).

Here, the predicted value of the state quantity and the dispersion of the predicted value will be explained. FIG. 3 is an explanatory diagram conceptually showing predicted values of state quantities and variations in predicted values in the first embodiment. FIG. 3 shows an example in which the predicted value and the range of variation in the predicted value are calculated using Gaussian process regression. In FIG. 3, the horizontal axis of the graph represents input data x. The vertical axis of the graph represents output data y. The black circles in FIG. 3 represent data acquired by sampling.

In prediction using Gaussian process regression, a predicted value of output data y is calculated on the assumption that output data y follows a Gaussian distribution. Therefore, if the predicted value is the mean m(x) of the Gaussian distribution, and the index indicating the uncertainty of the prediction is the standard deviation σ(x) of the Gaussian distribution, then the actual output data y will have a probability of approximately 95%. , m(x)-2σ(x) or more and m(x)+2σ(x) or less. In FIG. 3, the solid curve represents the average m(x) that is the predicted value of the output data y. The dashed curves represent m(x)-2σ(x) and m(x)+2σ(x). As shown in FIG. 3, the variation in predicted values tends to be small at locations close to the data obtained by sampling, and the variation in predicted values tends to increase at locations far from the data obtained by sampling.

From such a statistical viewpoint, in the first embodiment, the reliability index is defined based on the standard deviation. For example, when standard deviation is used as a reliability index, the smaller the reliability index, the smaller the variation in predicted values, and the more likely the predicted values are.

The state quantity prediction unit 7 calculates the probability distribution of the predicted state quantities, and simultaneously calculates the average m(x), which is the predicted state quantity, and the standard deviation σ(x), which is the reliability index, from the probability distribution. . The state quantity prediction unit 7 calculates the average m(x) in a certain control period and the standard deviation σ(x) indicating the reliability of the average m(x) at the same timing.

Note that calculating the predicted state amount and the reliability index at the same time does not necessarily mean that the timing at which the predicted state amount is calculated and the timing at which the reliability index is calculated are the same. Simultaneously calculating the predicted state quantity and the reliability index may mean calculating the predicted state quantity and the reliability index based on the operation of the motor 3 or the mechanical system 4 at the same timing. Alternatively, calculating the predicted state quantity and the reliability index simultaneously means that the predicted state quantity and the reliability index are calculated by a single estimation process rather than multiple estimation processes for the operation of the motor 3 or the mechanical system 4 at the same timing. It may be something you do. The estimation process refers to calculating the prediction model 16. The state quantity prediction unit 7 can calculate the predicted state quantity and the reliability index by calculating the predicted state quantity and the reliability index in a short time by calculating the predicted state quantity and the reliability index simultaneously. Simultaneously calculating the predicted state quantity and the reliability index is suitable for the servo control device 2, which is an embedded device that requires high-speed calculation.

Note that the variance σ ² (x) of a Gaussian distribution may be used as the reliability index. Alternatively, the reliability index may be the result of an operation using one or more of the variance σ ² (x), the standard deviation σ(x), and the average m(x). As the predicted state quantity, a result value of an operation using one or more of the average m(x), the standard deviation σ(x), and the variance σ ² (x) may be used. In the following explanation, it is assumed that the predicted state quantity is the average m(x), and the reliability index is the standard deviation σ(x).

Although an example of calculating the predicted state quantity and reliability index when Gaussian process regression is used in the prediction model 16 has been described, the method for calculating the predicted state quantity and reliability index is not limited to using Gaussian process regression. For example, a machine learning method such as a decision tree, linear regression, boosting, or neural network may be used to calculate the predicted state quantity. For example, methods such as density estimation, mixed density network, etc. may be used to calculate the reliability index. The probability distribution of the predicted state quantity is not limited to a Gaussian distribution, but may be an exponential distribution, a t-distribution, or a combination of multiple types of distributions.

Next, details of the learning section 9 will be explained. The learning unit 9 learns the prediction model 16 used by the state quantity prediction unit 7. Specifically, the learning unit 9 performs learning to optimize model parameters or hyperparameters that are parameters applied to the prediction model 16.

Any learning algorithm may be used by the learning section 9. As an example, a case will be described in which Gaussian process regression, which is one of the supervised learning algorithms, is applied. When Gaussian process regression is used in the prediction model 16, the parameters used in the kernel function and the like are estimated from training data, and the prediction model 16 is updated. The training data is data that is a combination of the actual measured values of the state quantities shown in the feedback information 15 and the command information 11. The actual measured values of the state quantities are teacher data. A more reliable prediction model 16 can be constructed by using maximum likelihood estimation or the like as a parameter estimation method, but the parameter estimation method is not limited to this method.

The learning unit 9 outputs a prediction model 16 as a learning result when the learning converges. The learning unit 9 may output learned model information as a learning result. The learned model information is a parameter applied to the prediction model 16, that is, a learned model parameter or hyperparameter. A known determination method can be used to determine whether learning has converged.

The learning section 9 shown in FIG. 1 is realized by an information processing device that is an external device to the servo control device 2. A device functioning as the learning section 9 is connected to the servo control device 2. The device functioning as the learning section 9 may be a device connectable to the servo control device 2 via a network. The device functioning as the learning unit 9 may be a device existing on a cloud server. The learning section 9 is not limited to being implemented by a device external to the servo control device 2, and may be provided inside the servo control device 2.

FIG. 4 is a diagram showing a modification of the servo system 1 according to the first embodiment. A servo system 1 that is a modified example shown in FIG. 4 includes a learning section 9 inside a servo control device 2. As shown in FIG. The servo system 1 which is a modified example is the same as the servo system 1 shown in FIG. 1 except that the learning section 9 is provided inside the servo control device 2.

Next, details of the correction amount calculation section 8 will be explained. In the following description, the command information 11 is assumed to be a command value indicating the command position of the motor 3. The predicted state quantity calculated by the state quantity prediction unit 7 is a position error that is an error in the position of the motor 3 with respect to the commanded position. Note that the predicted state amount may be the feedback position of the motor 3. In this case, the correction amount calculation unit 8 can obtain the predicted state quantity, which is the position error, by calculating the difference between the command position and the predicted value of the feedback position. The predicted state quantity is not limited to the command position, but may be a physical quantity such as speed, torque, or amount of current.

Here, the position error which is the predicted state quantity calculated by the state quantity prediction unit 7 is μ, the reliability index calculated by the state quantity prediction unit 7 is σ, the actually occurring position error is e, and the correction amount is Co. do. The servo control unit 6 corrects the command position by subtracting the predicted position error μ from the command position value. The servo control unit 6 corrects the command position using a correction amount Co corresponding to the position error μ, thereby canceling out the error caused by the operation of the motor 3 or the mechanical system 4, and accurately adjusting the actual position to the command position. It is possible to match. However, if the predicted positional error μ is different from the actually generated positional error e, appropriate correction may not be possible and the positional accuracy may deteriorate. Therefore, the correction amount calculation unit 8 uses the reliability index to calculate an effective correction amount Co that can perform appropriate correction. Thereby, the servo system 1 can prevent deterioration of position accuracy due to correction.

Here, first and second examples of the method in which the correction amount calculation unit 8 calculates the correction amount based on the predicted value of the state quantity and the reliability index will be described. The larger the reliability index σ, which is the standard deviation, represents the greater the uncertainty in the value of the position error μ, which is the predicted value. In the first example, the servo system 1 sets a threshold value for the reliability index σ in advance, and stops correcting the command value when the reliability index σ exceeds the threshold value. The servo system 1 corrects the command value when the reliability index σ is less than or equal to the threshold value.

In the first example, the correction amount calculation unit 8 calculates the correction amount Co according to the following equation (5), with the threshold value being t.

In this way, in the first example, the threshold value of the reliability index σ for determining whether the predicted value is an appropriate value is set, and the correction amount calculation unit 8 calculates the calculated reliability index σ. When the value of is outside the range determined by the threshold value, the correction amount Co is set to zero. Thereby, the servo system 1 can prevent deterioration of position accuracy due to correction.

Next, a second example of how the correction amount calculation section 8 calculates the correction amount will be described. The probability distribution p(e) of the actual position error e follows a Gaussian distribution with position error μ and variance σ ² . The probability distribution p(e) can be calculated from the equation p(e)=N(μ, σ ² ) using the predicted position error μ and the reliability index σ. When the servo control unit 6 performs correction by subtracting the correction amount Co from the command position value, if the actual position error e and the correction amount Co satisfy the following equation (6), the position accuracy due to correction will deteriorate. occurs.

|e|<|e-Co| ...(6)

By integrating the probability distribution p(e) within the range of position error e that satisfies Equation (6), the probability that the position accuracy deteriorates can be calculated before correction. In the second example, by setting the target probability in advance, the correction amount calculation unit 8 determines the correction amount Co such that the probability that the position accuracy deteriorates is less than or equal to the target probability. becomes possible.

In this way, the correction amount calculation unit 8 can obtain an effective correction amount that allows appropriate correction by calculating the correction amount based on the predicted value of the state quantity and the reliability index. The servo system 1 is capable of correction that reduces trajectory errors while reducing deterioration in positional accuracy due to correction. The correction amount calculation unit 8 calculates the correction amount by calculation at every servo control period. The servo system 1 can correct the movement locus with high precision by continuously calculating the correction amount by the correction amount calculating section 8.

FIG. 5 is a flowchart showing the procedure of processing executed by the servo control device 2 of the servo system 1 according to the first embodiment. FIG. 5 shows a processing procedure in a servo control cycle.

In step S1, the command generation unit 5 of the servo control device 2 generates command information 11. The command generation unit 5 outputs the generated command information 11. In step S2, the state quantity prediction unit 7 of the servo control device 2 inputs the actual measured value of the state quantity and the command information 11, and simultaneously calculates the predicted value of the state quantity and the reliability index by using the prediction model 16. . The state quantity prediction unit 7 acquires the actual measured value of the state quantity from the feedback information 15. The state quantity prediction unit 7 outputs predicted state quantity information 12 which is a calculation result of a predicted value, and index information 13 which is a calculation result of a reliability index.

In step S3, the correction amount calculation unit 8 of the servo control device 2 calculates the correction amount based on the predicted value and the reliability index. The correction amount calculation unit 8 outputs correction information 14 which is the calculation result of the correction amount. In step S4, the servo control unit 6 of the servo control device 2 controls the motor 3 based on the command information 11 corrected using the correction amount. The servo control device 2 repeats the processing according to the procedure shown in FIG. 5 during a period from starting driving the motor 3 to ending driving the motor 3.

According to the first embodiment, the servo control device 2 simultaneously calculates the predicted value of the state quantity and the reliability index based on the prediction model 16, and calculates the correction amount based on the predicted value of the state quantity and the reliability index. calculate. By simultaneously calculating the predicted value of the state quantity and the reliability index, the servo control device 2 can move more easily than when the correction amount is calculated based on the reliability index of the state quantity estimated in the past. It is possible to calculate a correction amount that allows the trajectory to be corrected with high accuracy. As described above, the servo system 1 and the servo control device 2 have the effect of enabling highly accurate and stable control of the actuator.

Embodiment 2.
In the second embodiment, a modification of the correction amount calculation unit 8 will be described. FIG. 6 is a diagram illustrating a configuration example of a correction amount calculation unit 8A included in the servo system 1 according to the second embodiment. The servo system 1 according to the second embodiment includes a correction amount calculation section 8A that is a modification of the correction amount calculation section 8. The configuration of the servo system 1 according to the second embodiment other than the correction amount calculation section 8A is the same as that of the servo system 1 according to the first embodiment. In Embodiment 2, the same components as in Embodiment 1 described above are given the same reference numerals, and configurations that are different from Embodiment 1 will be mainly explained.

The correction amount calculation unit 8A includes a correction amount candidate value calculation unit 31 that calculates a correction amount candidate value 34, a correction amount evaluation unit 32 that calculates a correction amount evaluation value 35, and a correction amount evaluation unit 32 that calculates a correction amount evaluation value 35. and a correction amount determining section 33 that determines the value. The correction amount calculation unit 8A calculates the correction amount based on the evaluation value 35 by the correction amount determination unit 33 determining the value of the correction amount based on the evaluation value 35 calculated by the correction amount evaluation unit 32.

In the following description, the command information 11 is assumed to be a command value indicating the command position of the motor 3. The predicted state quantity calculated by the state quantity prediction unit 7 is a position error that is an error in the position of the motor 3 with respect to the commanded position. Note that the predicted state amount may be the feedback position of the motor 3. In this case, the correction amount calculation unit 8A can obtain the predicted state quantity, which is the position error, by calculating the difference between the command position and the predicted value of the feedback position. The predicted state quantity is not limited to the command position, but may be a physical quantity such as speed, torque, or amount of current.

The predicted state quantity information 12 and the index information 13 are input to the correction amount candidate value calculation unit 31 . The correction amount candidate value calculation unit 31 determines one or more candidate values 34 for the correction amount. For example, the correction amount candidate value calculation unit 31 determines the candidate value 34 by calculation using the predicted value of the state quantity shown in the predicted state quantity information 12 and the reliability index shown in the index information 13. The correction amount candidate value calculation unit 31 outputs the determined candidate value 34. Instead of outputting the candidate value 34 determined based on the predicted value of the state quantity and the reliability index, the correction amount candidate value calculation unit 31 may output the candidate value 34 set in advance in the table. good.

The predicted state quantity information 12 , the index information 13 , and the candidate value 34 determined by the correction amount candidate value calculating section 31 are input to the correction amount evaluation section 32 . The correction amount evaluation unit 32 calculates an evaluation value 35 for each of the plurality of candidate values 34 based on a preset evaluation function. The correction amount evaluation unit 32 outputs an evaluation value 35 calculated for each candidate value 34.

The correction amount determination section 33 receives the candidate value 34 determined by the correction amount candidate value calculation section 31 and the evaluation value 35 calculated by the correction amount evaluation section 32. The correction amount determination unit 33 determines a value to be the correction amount from among the candidate values 34 based on the evaluation value 35 for each candidate value 34 . In this way, the correction amount calculation unit 8A determines the value of the correction amount from among the plurality of candidate values 34 based on the evaluation value 35 calculated for each of the plurality of candidate values 34.

Here, an example of a method for calculating the evaluation value 35 by the correction amount evaluation section 32 will be described. The correction amount evaluation unit 32 calculates the evaluation value 35 based on the predicted state amount shown in the predicted state amount information 12 and the reliability index shown in the index information 13. As in Embodiment 1, the position error which is the predicted state quantity calculated by the state quantity prediction unit 7 is μ, the reliability index calculated by the state quantity prediction unit 7 is σ, the actually occurring position error is e, Let Co be the correction amount. It is assumed that the servo control unit 6 corrects the command position by subtracting the correction amount Co from the command position value.

The probability distribution p(e) of the actual position error e is defined by the predicted position error μ and the reliability index σ. The probability distribution p(e) may be a Gaussian distribution or a distribution other than a Gaussian distribution. Here, it is assumed that the probability distribution p(e) follows a Gaussian distribution with a position error μ and a variance σ ² . The probability distribution p(e) can be calculated from the equation p(e)=N(μ, σ ² ) using the predicted position error μ and the reliability index σ.

Each of the correction amount candidate values, which are the candidate values 34 input from the correction amount candidate value calculating section 31 to the correction amount evaluation section 32, is expressed as _Con (n=1, 2, 3, . . . ). The correction amount evaluation unit 32 is provided with an evaluation function for calculating the evaluation value of the correction amount candidate value _Con . The evaluation function is determined such that the lower the accuracy of the correction using the correction amount Co, the larger the evaluation value 35 becomes.

Since the commanded position is corrected by subtracting the correction amount Co from the value of the commanded position, when the correction amount Co and the actual position error e match, the positional accuracy after correction becomes the highest. The larger the deviation between the correction amount Co and the actual positional error e, the lower the positional accuracy after correction. Therefore, the evaluation function is set such that the evaluation value becomes larger as the absolute value of the difference between the correction amount candidate value _Con and the position error e becomes larger.

Further, the control of the motor 3 by the servo control device 2 may become more unstable as the value of the correction amount Co becomes larger. In other words, when the value of the correction amount Co deviates positively from the actual positional error e, the positional accuracy after correction tends to be lower than when the value of the correction amount Co deviates negatively from the actual positional error e. . Therefore, the evaluation function is such that when the correction amount candidate value _Con is larger than the position error e, the evaluation value is larger than when the correction amount candidate value _Con is smaller than the position error e. It is set so that

FIG. 7 is a diagram showing an example of an evaluation function used to calculate the evaluation value 35 in the correction amount calculation unit 8A of the second embodiment. FIG. 7 shows a graph of L( _Con , e), which is an example of the evaluation function. In FIG. 7, the vertical axis of the graph is L( _Con , e), and the horizontal axis of the graph is the actual position error e.

L( _Con , e) is expressed by the following equation (7).

L( _Con , e) gives a penalty or loss depending on the deviation of the correction amount candidate value _Con from the position error e, and is also referred to as a loss function.

The evaluation value 35 based on the evaluation function is defined by the value obtained by integrating the product of the evaluation function and the probability distribution p(e). E( _Con ), which is an example of the evaluation value 35, is found by integrating the product of L( _Con , e) and the probability distribution p(e), as shown in the following equation (8).

The correction amount determining unit 33 compares E(Con ₎ calculated for each correction amount candidate value _Con , and selects the correction amount candidate value _Con with the smallest E(Con ₎ . The correction amount determining unit 33 determines the correction amount candidate value _Con selected from among the correction amount candidate values _Con as the value of the correction amount Co.

In this way, the correction amount determination unit 33 determines the candidate value 34 with the smallest evaluation value 35 among the plurality of candidate values 34 as the value of the correction amount. Note that the correction amount calculation unit 8A may determine, in the correction amount determination unit 33, a candidate value 34 that is less than or equal to a preset threshold value as the value of the correction amount. The correction amount calculation unit 8A outputs correction information 14 that is the value of the correction amount determined by the correction amount determination unit 33.

The correction amount calculating section 8A can calculate an effective correction amount that can perform appropriate correction by determining the value of the correction amount from among the candidate values 34 based on the evaluation value 35. Thereby, the servo system 1 can prevent deterioration of position accuracy due to correction.

Note that the evaluation function may be determined such that the lower the accuracy of correction using the correction amount, the smaller the evaluation value 35. In this case, the correction amount determination unit 33 determines the candidate value 34 with the largest evaluation value 35 among the plurality of candidate values 34 as the value of the correction amount. Furthermore, the evaluation function is not limited to a linear function, but may be a higher-order function. The evaluation function may be set in consideration of the mechanical characteristics of the motor 3 or the mechanical system 4, or the control characteristics of the servo control section 6.

According to the second embodiment, the correction amount calculation unit 8A calculates the evaluation value 35 of the candidate value 34 based on the evaluation function, and determines the value of the correction amount from among the candidate values 34 based on the evaluation value 35. The correction amount calculation unit 8A can calculate a correction amount that can correct the movement trajectory with high precision. As described above, the servo system 1 and the servo control device 2 have the effect of enabling highly accurate and stable control of the actuator.

Embodiment 3.
FIG. 8 is a diagram showing a configuration example of a servo system 1A according to the third embodiment. The servo system 1A corrects the amount of correction based on the response characteristics of the motor 3 or mechanical system 4 to the amount of correction. In Embodiment 3, the same components as in Embodiment 1 or 2 described above are given the same reference numerals, and configurations that are different from Embodiment 1 or 2 will be mainly explained.

The servo system 1A includes a servo control device 2A, a motor 3, a mechanical system 4, a learning section 9, and a response characteristic identification section 40. The servo control device 2A includes a correction amount correction section 41. The configuration of the servo control device 2A other than the correction amount correction unit 41 is the same as that of the servo control device 2 of the servo system 1 according to the first embodiment.

In the following description, the command information 11 is assumed to be a command value indicating the command position of the motor 3. The predicted state quantity calculated by the state quantity prediction unit 7 is a position error that is an error in the position of the motor 3 with respect to the commanded position. Note that the predicted state amount may be the feedback position of the motor 3. In this case, the correction amount calculation unit 8 can obtain the predicted state quantity, which is the position error, by calculating the difference between the command position and the predicted value of the feedback position. The predicted state quantity is not limited to the command position, but may be a physical quantity such as speed, torque, or amount of current.

When the command value is corrected using the correction amount in the servo control unit 6, due to the influence of the mechanical characteristics of the motor 3 or the mechanical system 4, or the control characteristics of the servo control unit 6, the expected correction may not be made and the correction may not be completed. The effect may be reduced. The mechanical properties are due to the structure of the motor 3 or mechanical system 4. Therefore, in the third embodiment, the servo control device 2A corrects the correction amount by taking into account the influence of the mechanical characteristics or control characteristics.

The response characteristic identification unit 40 identifies the response characteristic of the motor 3 or mechanical system 4 with respect to the correction amount. The response characteristic identification unit 40 outputs response characteristic information 42 that is information indicating the identified response characteristic.

Here, an example in which the response characteristic identifying section 40 identifies a response characteristic using a transfer function will be described. The feedback position is measured in a state where the command value is set to "0" and white noise corresponding to the correction amount is applied instead of the correction amount calculated by the correction amount calculating section 8. The feedback position measured at this time represents the amount of correction that actually acts in correcting the position of the motor 3 or the position of the driven body of the mechanical system 4. Hereinafter, the amount of correction that actually acts in correcting the position of the motor 3 or the position of the driven body will be referred to as an actual correction amount. Due to the influence of mechanical characteristics or control characteristics, the actual correction amount deviates from the calculated correction amount. Note that when identifying the response characteristics, a sine sweep wave may be used instead of white noise. The feedback position may be measured by actually operating the motor 3 and the mechanical system 4, or by using simulation.

FIG. 9 is a diagram for explaining response characteristics identified in the servo system 1A according to the third embodiment. The response characteristic identification unit 40 acquires correction amount data and actual correction amount data. The response characteristic identification unit 40 calculates a frequency transfer function, which is a response characteristic, by performing numerical processing such as Fourier transform on the correction amount and the actual correction amount. The response characteristic identification unit 40 identifies a transfer function represented by each of gain and phase in the frequency domain.

As shown in FIG. 9, the transfer function is represented by a gain diagram representing gain characteristics and a phase diagram representing phase characteristics. In the gain diagram, the horizontal axis is a logarithmic axis representing frequency, and the vertical axis represents gain, which is a decibel value. In the phase diagram, the horizontal axis is a logarithmic axis representing frequency, and the vertical axis represents phase. In the gain diagram, the resonance point is represented as a positive peak, and the anti-resonance point is represented as a negative peak. As the diagram representing the transfer function, a co-quad diagram in which the real part and imaginary part after Fourier transform are plotted separately on the frequency axis may be used. In the Koquad diagram, the horizontal axis is a logarithmic axis representing frequency, and the vertical axis is the real part and imaginary part.

The response characteristic identification unit 40 outputs response characteristic information 42 that is information on a transfer function that is the identified response characteristic. Note that the response characteristic identifying section 40 may identify the response characteristic using a means other than the transfer function, such as a polynomial or a neural network.

The response characteristic identification section 40 is realized by an information processing device that is a device external to the servo control device 2A. A device functioning as the response characteristic identification section 40 is connected to the servo control device 2A. The device functioning as the response characteristic identification section 40 may be a device connectable to the servo control device 2A via a network. The device functioning as the response characteristic identification section 40 may be a device existing on a cloud server. The response characteristic identification section 40 is not limited to being implemented by a device external to the servo control device 2A, and may be provided inside the servo control device 2A.

The correction information 14 and the response characteristic information 42 are input to the correction amount correction unit 41 . The correction amount modification unit 41 performs a calculation to modify the correction amount calculated by the correction amount calculation unit 8 based on the response characteristic information 42. The correction amount modification unit 41 modifies the correction amount based on the response characteristic identified by the response characteristic identification unit 40. The correction amount modification unit 41 calculates the corrected correction amount using the correction amount calculated by the correction amount calculation unit 8 and the inverse model of the transfer function.

Here, an example of how the correction amount modification unit 41 modifies the correction amount will be described. It is assumed that transfer function information is input to the correction amount modification unit 41 as the response characteristic information 42. It is assumed that the correction amount calculated by the correction amount calculating section 8 is Co, and the transfer function is G(s).

The correction amount correction unit 41 calculates the corrected correction amount Co' using the following equation (9). G ^-1 (s) is an inverse transfer function of G(s) and satisfies G(s)G ^-1 (s)=1.

Co'=G ^-1 (s)Co...(9)

The correction amount correction unit 41 outputs correction correction information 43 indicating the corrected correction amount. Command information 11 and modification correction information 43 are input to the servo control unit 6 . The servo control unit 6 corrects the command value shown in the command information 11 by calculation using the corrected correction amount. The servo control unit 6 controls the motor 3 based on the command information 11 corrected using the corrected correction amount.

The correction amount modification unit 41 adjusts the correction amount so as to reduce the influence of the mechanical characteristics of the motor 3 or the mechanical system 4, or the control characteristics of the servo control unit 6, by modifying the correction amount based on the identified response characteristics. can be corrected. That is, the correction amount modification unit 41 can modify the correction amount so that it matches the actual correction amount. Note that the servo control device 2A may include the correction amount calculation section 8A of the second embodiment instead of the correction amount calculation section 8.

According to the third embodiment, the correction amount correction unit 41 corrects the correction amount to match the actual correction amount by correcting the correction amount based on the response characteristics of the motor 3 or the mechanical system 4 to the correction amount. be able to. The servo control device 2A can correct the movement trajectory with high precision by correcting the command information 11 based on the corrected correction amount. As described above, the servo system 1A and the servo control device 2A have the effect of enabling highly accurate and stable control of the motor 3.

Embodiment 4.
FIG. 10 is a diagram showing a configuration example of a servo system 1B according to the fourth embodiment. The servo system 1B corrects the measured value of the state quantity used to create training data based on the response characteristics of the motor 3 or mechanical system 4 to the correction amount. The training data is learning data used by the learning unit 9 to learn the prediction model 16. In Embodiment 4, the same components as in Embodiments 1 to 3 described above are given the same reference numerals, and configurations that are different from Embodiments 1 to 3 will be mainly described.

The servo system 1B includes a servo control device 2B, a motor 3, a mechanical system 4, a learning section 9, a response characteristic identification section 40, a state quantity correction section 50, and an information storage section 51. The configuration of the servo control device 2B is similar to the servo control device 2 of the servo system 1 according to the first embodiment. Servo control device 2B differs from servo control device 2 in that it outputs command information 11 and feedback information 15 to information storage section 51.

In the following description, the command information 11 is assumed to be a command value indicating the command position of the motor 3. The predicted state quantity calculated by the state quantity prediction unit 7 is a position error that is an error in the position of the motor 3 with respect to the commanded position. Note that the predicted state amount may be the feedback position of the motor 3. In this case, the correction amount calculation unit 8 can obtain the predicted state quantity, which is the position error, by calculating the difference between the command position and the predicted value of the feedback position. The predicted state quantity is not limited to the command position, but may be a physical quantity such as speed, torque, or amount of current.

The response characteristic identification unit 40 identifies the response characteristic of the motor 3 or mechanical system 4 with respect to the correction amount. The response characteristic identification unit 40 identifies a transfer function represented by each of gain and phase in the frequency domain. The response characteristic identification unit 40 outputs response characteristic information 42 that is information on a transfer function that is the identified response characteristic. Note that the response characteristic identifying section 40 may identify the response characteristic using a means other than the transfer function, such as a polynomial or a neural network.

The information storage unit 51 stores the command information 11 and the feedback information 15 in association with each other. Feedback information 15 includes actual measured values of state quantities. The state quantity correction unit 50 reads out the command information 11 and the feedback information 15 from the information storage unit 51. Further, response characteristic information 42 is input to the state quantity correction unit 50. The state quantity correction unit 50 performs a calculation to correct the measured value of the state quantity included in the feedback information 15 based on the response characteristic information 42. The state quantity correction unit 50 corrects the actual measured value of the state quantity based on the response characteristic identified by the response characteristic identification unit 40. The state quantity correction unit 50 calculates a corrected actual measurement value using the actual measurement value of the state quantity included in the feedback information 15 and the inverse model of the transfer function.

Here, an example of a method by which the state quantity correction unit 50 corrects the actual measured value of the state quantity will be described. Hereinafter, the actual measured value of the state quantity corrected by the state quantity correction unit 50 will be referred to as a corrected state quantity. It is assumed that transfer function information is input to the state quantity correction unit 50 as the response characteristic information 42. It is assumed that the transfer function is G(s). Further, it is assumed that the actual measured value of the state quantity is the position error e.

The state quantity correction unit 50 calculates the corrected position error e' using the following equation (10). G ^-1 (s) is an inverse transfer function of G(s) and satisfies G(s)G ^-1 (s)=1.

e'=G ^-1 (s)Co...(10)

The state quantity correction unit 50 outputs command information 11 and modified state quantity information 52 indicating the corrected state quantity. Command information 11 and modified state quantity information 52 are input to the learning section 9 . The learning unit 9 creates training data including the command value shown in the command information 11 and the modified state quantity, and learns the prediction model 16 using the training data. The modified state quantity is the teacher data.

The state quantity correction unit 50 reduces the influence of the mechanical characteristics of the motor 3 or the mechanical system 4, or the control characteristics of the servo control unit 6, by correcting the measured value of the state quantity based on the identified response characteristic. The actual measured value of the state quantity can be corrected. The learning unit 9 can generate a prediction model 16 in which the influence of the mechanical characteristics of the motor 3 or the mechanical system 4 or the control characteristics of the servo control unit 6 is reduced. The servo control device 2B calculates the correction amount using the predicted value of the state quantity calculated based on the prediction model 16 and the reliability index, thereby adjusting the mechanical characteristics of the motor 3 or the mechanical system 4, or the servo control unit. Corrections can be made that can reduce the influence of the control characteristics of No. 6.

The learning section 9, the response characteristic identification section 40, the state quantity correction section 50, and the information storage section 51 are realized by an information processing device that is a device external to the servo control device 2B. The devices functioning as the respective parts are connected to the servo control device 2B. The devices functioning as the respective parts may be devices connectable to the servo control device 2B via a network. The devices that function as the respective units may be devices that exist on a cloud server. The respective parts are not limited to those implemented by devices external to the servo control device 2B, and may be provided inside the servo control device 2B. The learning section 9, the response characteristic identification section 40, the state quantity modification section 50, and the information storage section 51 are not limited to being integrated into one device, but may be distributed among two or more devices.

Note that the servo control device 2B may include the correction amount calculation section 8A of the second embodiment instead of the correction amount calculation section 8. The servo control device 2B may perform the processing described in the fourth embodiment and the processing described in the third embodiment.

According to the fourth embodiment, the learning unit 9 learns the prediction model 16 based on training data including the state quantities modified by the state quantity modification unit 50. The servo control device 2B calculates a predicted value of the state quantity and a reliability index based on the prediction model 16, and calculates a correction amount based on the predicted value of the state quantity and the reliability index. The servo control device 2B can correct the movement trajectory with high precision by correcting the command information 11 based on the correction amount. As described above, the servo system 1B and the servo control device 2B have the effect of enabling highly accurate and stable control of the actuator.

Next, hardware that implements the servo control devices 2, 2A, and 2B will be explained. FIG. 11 is a diagram showing an example of the hardware configuration for realizing the servo control devices 2, 2A, 2B of the servo systems 1, 1A, 1B according to the first to fourth embodiments. FIG. 11 shows a configuration example in which the servo control devices 2, 2A, and 2B are implemented by a processing circuit 60. Processing circuit 60 includes a processor 62 and memory 63.

The processor 62 is a CPU (Central Processing Unit, also referred to as a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, or a DSP (Digital Signal Processor)), a system LSI (Large Scale Integration), or the like. The memory 63 is a non-volatile or These include volatile semiconductor memories, magnetic disks, flexible disks, optical disks, compact disks, mini disks, and DVDs (Digital Versatile Discs).

The memory 63 includes a command generation section 5, a servo control section 6, a state quantity prediction section 7, a correction amount calculation section 8, 8A, a learning section 9, and a correction amount correction section, which are the main parts of the servo control devices 2, 2A, and 2B. A program for operating as 41 is stored. By having the processor 62 read and execute this program, it is possible to realize the main parts of the servo control devices 2, 2A, and 2B. Note that the programs stored in the memory 63 and operating as the main parts of the servo control devices 2, 2A, and 2B may be written on a storage medium such as a CD (Compact Disc)-ROM or DVD-ROM, for example. The information may be provided to users etc. via a network, or may be provided via a network. Further, the processor 62 outputs data such as calculation results to the volatile memory of the memory 63. Alternatively, the processor 62 stores data such as calculation results by outputting the data to an auxiliary storage device via the volatile memory of the memory 63.

The input unit 61 is a circuit that receives input signals to the servo control devices 2, 2A, and 2B from the outside. The output unit 64 is a circuit that outputs signals generated by the servo control devices 2, 2A, and 2B to the outside.

FIG. 11 shows an example of hardware in which the servo control devices 2, 2A, 2B are implemented using a general-purpose processor 62 and memory 63. , 2A, 2B may be realized. That is, the servo control devices 2, 2A, and 2B may be implemented using dedicated processing circuits. Here, the dedicated processing circuit is a single circuit, a composite circuit, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. Note that a part of the servo control devices 2, 2A, and 2B may be realized by the processor 62 and the memory 63, and the rest may be realized by a dedicated processing circuit.

In the servo systems 1, 1A, 1B, each of the learning section 9, the response characteristic identification section 40, the state quantity correction section 50, and the information storage section 51 provided outside the servo control devices 2, 2A, 2B is constructed using the hardware shown in FIG. This can be realized using hardware similar to hardware.

In the first to fourth embodiments, the motor 3 is an example of an actuator. The servo control devices 2, 2A, and 2B may control actuators other than the motor 3. That is, the servo systems 1, 1A, and 1B may include actuators other than the motor 3.

The configurations shown in each of the embodiments above are examples of the contents of the present disclosure. The configuration of each embodiment can be combined with other known techniques. The configurations of each embodiment may be combined as appropriate. It is possible to omit or change a part of the configuration of each embodiment without departing from the gist of the present disclosure.

1, 1A, 1B servo system, 2, 2A, 2B servo control device, 3 motor, 4 mechanical system, 5 command generation section, 6 servo control section, 7 state quantity prediction section, 8, 8A correction amount calculation section, 9 learning part, 11 command information, 12 predicted state quantity information, 13 index information, 14 correction information, 15 feedback information, 16 prediction model, 21 servo motor, 22 motor end position detector, 23 coupling, 24 ball screw, 25 ball screw Nut, 26 Table, 27 Machine end position detector, 28 Machine end position detector head, 31 Correction amount candidate value calculation section, 32 Correction amount evaluation section, 33 Correction amount determination section, 34 Candidate value, 35 Evaluation value, 40 Response Characteristic identification section, 41 Correction amount correction section, 42 Response characteristic information, 43 Correction correction information, 50 State amount correction section, 51 Information storage section, 52 Correction state amount information, 60 Processing circuit, 61 Input section, 62 Processor, 63 Memory , 64 output section.

Claims (16)

a command generation unit that generates command information for controlling the actuator;
a state quantity prediction unit that simultaneously calculates a predicted value of a state quantity of the actuator or a mechanical system driven by the actuator and an index representing reliability of the predicted value;
a correction amount calculation unit that calculates a correction amount for correcting the command information based on the predicted value and the index;
a servo control unit that controls the actuator based on the command information corrected using the correction amount;
A servo control device comprising: The state quantity prediction unit predicts the state quantity for the state at the time of calculating the predicted value or the state quantity for the state at a time after the time of calculating the predicted value, and The servo control device according to claim 1, wherein the predicted value is output at a timing that coincides with a time point of the state quantity. 3. The state quantity prediction unit predicts the current or future state quantity by using a prediction model, using the actual measured value of the state quantity and the command information as input. Servo control device. 3. The state quantity prediction unit uses the prediction model to obtain a probability distribution of the predicted state quantity, and calculates the predicted value and the index based on the probability distribution. Servo control device described in. The servo control device according to claim 3 or 4, further comprising a learning section that learns the prediction model. The command information includes a command value of at least one of the position, speed, acceleration, or torque of the actuator, and a current flowing through the actuator,
The actual measured value includes at least one of the position, speed, acceleration, or torque of the actuator, the current flowing through the actuator, and the position or speed of a driven body of the mechanical system. The servo control device according to any one of claims 3 to 5. A threshold value of the index is set for determining whether the predicted value is an appropriate value,
Any one of claims 1 to 6, wherein the correction amount calculation unit sets the correction amount to zero when the calculated value of the index is outside a range determined by the threshold value. Servo control device described in. further comprising a correction amount evaluation unit that calculates an evaluation value of the correction amount,
The servo control device according to any one of claims 1 to 7, wherein the correction amount calculation section calculates the correction amount based on the evaluation value. The correction amount evaluation unit calculates the evaluation value for each of the plurality of candidate values for the correction amount based on a preset evaluation function,
9. The correction amount calculation unit determines the value of the correction amount from among the plurality of candidate values based on the evaluation value calculated for each of the plurality of candidate values. The servo control device described. The servo control device according to any one of claims 1 to 9, further comprising a correction amount correction unit that corrects the correction amount based on response characteristics of the actuator or the mechanical system with respect to the correction amount. . an actuator;
a command generation unit that generates command information for controlling the actuator;
a state quantity prediction unit that simultaneously calculates a predicted value of a state quantity of the actuator or a mechanical system driven by the actuator and an index representing reliability of the predicted value based on a prediction model;
a correction amount calculation unit that calculates a correction amount for correcting the command information based on the predicted value and the index;
a servo control unit that controls the actuator based on the command information corrected using the correction amount;
a learning unit that learns the predictive model using training data including actual measured values of the state quantities;
A servo system comprising: a response characteristic identification unit that identifies a response characteristic of the actuator or the mechanical system with respect to the correction amount;
The servo system according to claim 11, further comprising a correction amount modification unit that modifies the correction amount based on the identified response characteristic. The response characteristic identification unit identifies a transfer function represented by each of gain and phase in the frequency domain,
The servo system according to claim 12, wherein the correction amount modification unit calculates the corrected correction amount using the correction amount and an inverse model of the transfer function. a response characteristic identification unit that identifies a response characteristic of the actuator or the mechanical system with respect to the correction amount;
The servo system according to claim 11, further comprising: a state amount modification unit that modifies the measured value used to create the training data based on the identified response characteristic. The response characteristic identification unit identifies a transfer function represented by each of gain and phase in the frequency domain,
The servo system according to claim 14, wherein the state quantity correction unit calculates the corrected actual measurement value using the actual measurement value and an inverse model of the transfer function. A servo control method in which a servo control device controls an actuator,
generating command information for controlling the actuator;
simultaneously calculating a predicted value of a state quantity of the actuator or a mechanical system driven by the actuator and an index representing reliability of the predicted value;
calculating a correction amount for correcting the command information based on the predicted value and the index;
controlling the actuator based on the command information corrected using the correction amount;
A servo control method comprising:

Images