JP2019185742A - Controller and control method - Google Patents

Abstract

To provide a controller and a control method capable of identifying a coefficient of a friction model.SOLUTION: A controller 1 which performs position control in view of friction for one or more axes of a machine includes: a data acquisition part 70 which acquires at least a position command and a position feedback; and a correction torque estimation part 80 which estimates, based on position deviation as a difference between the position command and the position feedback, a coefficient of a friction model upon performing the position control.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a control device and a control method, and more particularly to a control device and a control method capable of identifying a coefficient of a friction model.

  In the control of industrial machines (hereinafter simply referred to as “machines”) including machine tools, injection molding machines, laser processing machines, electrical discharge machines, industrial robots, etc., precise compensation is achieved by compensating for frictional forces acting on the drive mechanism. Control performance can be obtained.

  FIG. 11 is an example of a drive mechanism of a machine tool. When the servo motor is driven and the ball screw supported by the bearing rotates, the stage moves. At this time, for example, a frictional force acts between the bearing and the ball screw and between the ball screw and the stage. In other words, the stage behavior is affected by the frictional force.

  FIG. 12 is a graph showing a typical relationship between the frictional force and the behavior of the drive mechanism. At the time of transition from the stationary state (speed = 0) to the moving state or from the moving state to the stationary state, the change in the frictional force is nonlinear. This is called the Stribeck effect. Due to the Stribeck effect, the machine has a long positioning time and a trajectory error (quadrant projection) during reversal occurs.

  The Lugre model is known as an effective friction model when considering such nonlinear friction compensation. With the Lugre model, a correction value (correction torque) for suppressing a non-linear friction effect can be obtained. As shown in FIG. 13, by adding the correction value to the current command, the nonlinear frictional force is compensated, and the controlled object can be precisely controlled. This correction processing can be performed in known feedback control. The machine control device determines the current command based on the deviation between the position command and the position feedback and the deviation between the speed command and the speed feedback. At this time, the control device adds the correction torque obtained from the Lugre model to the current command.

  The Lugre model is shown in Equation 1. F is a correction torque that is an output of the Lugre model. v and z are variables related to speed and position. Fc, Fs, v0, σ0, σ1, and σ2 are coefficients specific to the drive mechanism.

  As a related technique, Patent Document 1 discloses that correction data can be obtained from a friction model.

JP 2004-234327 A

  However, since the coefficient of the friction model including the Lugre model differs depending on the machine, the usage environment, etc., the coefficient must be individually identified for each control target. In addition, since there are a large number of coefficients to be identified, much work is required for the coefficient identification work. Therefore, there is a demand for means that can identify the coefficient of the friction model without trouble.

  Therefore, a control device and a control method that can identify the coefficient of the friction model are desired.

  One aspect of the present invention is a control device that performs position control in consideration of friction with respect to one or more axes of a machine, and includes a data acquisition unit that acquires at least a position command and position feedback, the position command, and the position feedback. And a correction torque estimation unit that estimates a coefficient of a friction model when performing the position control based on a position deviation that is a difference between the two.

  Another aspect of the present invention is a control method for performing position control in consideration of friction with respect to one or more axes of a machine, a data acquisition step for acquiring at least a position command and position feedback, the position command and the position And a correction torque estimating step for estimating a coefficient of a friction model when performing the position control based on a position deviation which is a feedback difference.

  According to the present invention, it is possible to provide a control device and a control method capable of identifying a coefficient of a friction model.

It is a schematic hardware block diagram of the control apparatus 1 of 1st Embodiment. It is a schematic functional block diagram of the control apparatus 1 of 1st Embodiment. It is a schematic hardware block diagram of the control apparatus 1 of 2nd, 3rd embodiment. It is a schematic functional block diagram of the control apparatus 1 of 2nd Embodiment. It is a functional block diagram of the learning part 83 in 2nd Embodiment. It is a flowchart which shows one form of reinforcement learning. It is a figure explaining a neuron. It is a figure explaining a neural network. It is a schematic functional block diagram of the control apparatus 1 and the machine learning apparatus 100 of 3rd Embodiment. 1 is a schematic functional block diagram showing one form of a system in which a control device 1 is incorporated. It is a schematic functional block diagram which shows the other form of the system incorporating the machine learning apparatus 120 (or 100). It is a figure which shows an example of the drive mechanism of a machine tool. It is a graph which shows the relationship between a frictional force and the behavior of a drive mechanism. It is a figure which shows an example of the correction method of the nonlinear frictional force using a friction model. It is a figure which shows the other example of the correction method of the nonlinear frictional force using a friction model. It is a figure which shows the other example of the correction method of the nonlinear frictional force using a friction model.

  FIG. 1 is a schematic hardware configuration diagram showing a control device 1 according to the first embodiment of the present invention and main parts of an industrial machine controlled by the control device 1. The control device 1 is a control device that controls machines such as machine tools. The control device 1 includes a CPU 11, ROM 12, RAM 13, nonvolatile memory 14, interface 18, bus 20, axis control circuit 30, and servo amplifier 40. A servo motor 50 and an operation panel 60 are connected to the control device 1.

  The CPU 11 is a processor that controls the control device 1 as a whole. The CPU 11 reads out the system program stored in the ROM 12 via the bus 20 and controls the entire control device 1 according to the system program.

  The ROM 12 stores in advance a system program (including a communication program for controlling communication with the machine learning device 100 described later) for executing various machine controls.

  The RAM 13 temporarily stores temporary calculation data and display data, data input by an operator via an operation panel 60 described later, and the like.

  The non-volatile memory 14 is backed up by a battery (not shown), for example, and retains the storage state even when the control device 1 is powered off. The nonvolatile memory 14 stores data input from the operation panel 60, machine control programs and data input via an interface (not shown), and the like. The program and data stored in the nonvolatile memory 14 may be expanded in the RAM 13 at the time of execution and use.

  The axis control circuit 30 controls an operation axis included in the machine. The axis control circuit 30 receives an axis movement command amount output from the CPU 11 and outputs an axis movement command to the servo amplifier 40. At this time, the axis control circuit 30 performs the feedback control described later, and corrects the nonlinear frictional force by the correction torque output from the CPU 11 based on the Lugre model or the like. Alternatively, the nonlinear frictional force may be corrected by the correction torque calculated by the axis control circuit 30 based on the Lugre model or the like. The correction in the axis control circuit 30 is generally faster than the correction by the CPU 11.

  The servo amplifier 40 drives the servo motor 50 in response to an axis movement command output from the axis control circuit 30.

  The servo motor 50 is driven by the servo amplifier 40 to move a shaft included in the machine. The servo motor 50 typically includes a position / speed detector. Alternatively, a position detector may be provided on the machine side without being incorporated. The position / velocity detector outputs a position / velocity feedback signal, and this signal is fed back to the axis control circuit 30 to perform position / velocity feedback control.

  In FIG. 1, only one axis control circuit 30, one servo amplifier 40, and one servo motor 50 are shown, but in reality, as many axes as the number of axes provided in the machine to be controlled are prepared. For example, when a machine having six axes is controlled, a total of six sets of the axis control circuit 30, servo amplifier 40, and servo motor 50 corresponding to each axis are prepared.

  The operation panel 60 is a data input device provided with hardware keys and the like. Among them, there is a manual data input device equipped with a display, hardware keys, etc., called a teaching operation panel. The teaching operation panel displays information received from the CPU 11 via the interface 18 on the display. The operation panel 60 passes pulses, commands, data, and the like input from hardware keys and the like to the CPU 11 via the interface 18.

  FIG. 2 is a schematic functional block diagram of the control device 1 according to the first embodiment. Each functional block illustrated in FIG. 2 is realized by the CPU 11 included in the control device 1 illustrated in FIG. 1 executing a system program and controlling the operation of each unit of the control device 1.

  The control device 1 of the present embodiment includes a data acquisition unit 70 and a correction torque estimation unit 80. The correction torque estimation unit 80 includes an optimization unit 81 and a correction torque calculation unit 82. In addition, an acquisition data storage unit 71 in which data acquired by the data acquisition unit 70 is stored is provided on the nonvolatile memory 14.

  The data acquisition unit 70 is a functional unit that acquires various data from the CPU 11, the servo motor 50, the machine, and the like. The data acquisition unit 70 acquires, for example, a position command, a position feedback, a speed command, and a speed feedback and stores them in the acquired data storage unit 71.

  Based on the data stored in the acquired data storage unit 71, the correction torque estimation unit 80 uses Fc, Fs, v 0, σ 0, Fc, Fs, v 0, σ 0, friction coefficient (typically the Lugre model). This is a functional means for estimating σ1, σ2). In the present embodiment, the optimization unit 81 estimates the coefficient of the friction model by solving an optimization problem that minimizes the deviation between the position command and the position feedback, for example. Typically, grid search that exhaustively searches for coefficient combinations, random search that tries coefficient combinations randomly, and Bayesian optimization that searches for optimal coefficient combinations based on probability distribution and acquisition function Thus, a combination of coefficients that minimizes the deviation between the position command and the position feedback can be estimated. In other words, the optimization unit 81 operates the machine while changing the combination of coefficients in various ways, and repeats a cycle for evaluating the deviation between the position command and the position feedback, thereby finding a combination of coefficients that minimizes the deviation. .

  The correction torque calculation unit 82 calculates and outputs a correction torque based on the friction model using the result estimated by the optimization unit 81 (an optimal combination of the coefficients of the friction model). The control device 1 adds the correction torque output from the correction torque calculation unit 82 to the current command.

  According to the present embodiment, since the optimization unit 81 identifies the coefficient of the friction model by solving the optimization problem, it is possible to easily determine the optimal coefficient according to various machines, usage environments, and the like. become.

  FIG. 3 is a schematic hardware block diagram of the control device 1 including the machine learning device 100 in the second and third embodiments. The control device 1 of the present embodiment has the same configuration as that of the first embodiment except that the configuration of the machine learning device 100 is provided. In the ROM 12 provided in the control device 1 of the present embodiment, a system program including a communication program for controlling the exchange with the machine learning device 100 is written in advance.

  The interface 21 is an interface for connecting the control device 1 and the machine learning device 100. The machine learning apparatus 100 includes a processor 101, a ROM 102, a RAM 103, and a nonvolatile memory 104.

  The processor 101 controls the entire machine learning device 100. The ROM 102 stores system programs and the like. The RAM 103 performs temporary storage in each process related to machine learning. The nonvolatile memory 104 stores a learning model and the like.

  The machine learning device 100 observes various information (position command, speed command, position feedback, etc.) that can be acquired by the control device 1 via the interface 21. The machine learning device 100 learns and estimates a coefficient of a friction model (typically a Lugre model) for precise control of the servo motor 50 by machine learning, and outputs a correction torque to the control device 1 via the interface 21. .

  FIG. 4 is a schematic functional block diagram of the control device 1 and the machine learning device 100 according to the second embodiment. The control device 1 shown in FIG. 4 has a configuration required when the machine learning device 100 performs learning (learning mode). In each functional block shown in FIG. 4, the CPU 11 included in the control device 1 shown in FIG. 3 and the processor 101 of the machine learning device 100 execute the respective system programs, and the control device 1 and the machine learning device 100 This is realized by controlling the operation of each part.

  The control device 1 of this embodiment includes a data acquisition unit 70 and a correction torque estimation unit 80 configured on the machine learning device 100. The correction torque estimation unit 80 includes a learning unit 83. Further, an acquisition data storage unit 71 for storing data acquired by the data acquisition unit 70 is provided on the nonvolatile memory 14, and a learning unit 83 is provided on the nonvolatile memory 104 of the machine learning device 100. A learning model storage unit 84 that stores a learning model constructed by machine learning is provided.

  The operation of the data acquisition unit 70 in this embodiment is the same as that in the first embodiment. The data acquisition unit 70 acquires, for example, a position command, a position feedback, a speed command, and a speed feedback and stores them in the acquired data storage unit 71. Further, the data acquisition unit 70 acquires a set of coefficients (Fc, Fs, v0, σ0, σ1, σ2) of the Lugre model that the control device 1 is currently using for correcting nonlinear friction, and acquires the data storage unit 71. To remember.

  The preprocessing unit 90 creates learning data used for machine learning by the machine learning device 100 based on the data acquired by the data acquisition unit 70. The pre-processing unit 90 creates learning data obtained by converting each data into a unified format handled by the machine learning device 100 (numericalization, sampling, etc.). When the machine learning device 100 performs unsupervised learning, the preprocessing unit 90 creates state data S in a predetermined format in the learning as learning data, and when the machine learning device 100 performs supervised learning. When a set of state data S and label data L in a predetermined format in the learning is created as learning data, and the machine learning device 100 performs reinforcement learning, the state data S and determination data in the predetermined format in the learning A set of D is created as learning data.

  The learning unit 83 performs machine learning using the learning data created by the preprocessing unit 90. The learning unit 83 generates a learning model by a known machine learning method such as unsupervised learning, supervised learning, and reinforcement learning, and stores the generated learning model in the learning model storage unit 84. As an unsupervised learning method performed by the learning unit 83, for example, the autoencoder method, the k-means method, and the like are used. Examples of the reinforcement learning method such as the neural network method include Q learning.

  FIG. 5 shows an internal functional configuration of the learning unit 83 that executes reinforcement learning as an example of the learning method. Reinforcement learning is a trial-and-error cycle that observes the current state (ie, input) of the environment where the learning target exists, executes a predetermined action (ie, output) in the current state, and gives some reward to that action. This is a method of learning, as an optimal solution, a policy (in this embodiment, setting of a coefficient of the Lugre model) such that the total amount of reward is maximized by repetition.

  The learning unit 83 includes a state observation unit 831, a determination data acquisition unit 832, and a reinforcement learning unit 833. In each functional block shown in FIG. 5, the CPU 11 included in the control device 1 shown in FIG. 3 and the processor 101 of the machine learning device 100 execute the respective system programs, and the control device 1 and the machine learning device 100 This is realized by controlling the operation of each part.

  The state observation unit 831 observes a state variable S that represents the current state of the environment. The state variable S includes, for example, a current Slag model coefficient S1, a current position command S2, a current speed command S3, and a position feedback S4 in the previous cycle.

  The state observation unit 831 acquires a set of coefficients (Fc, Fs, v0, σ0, σ1, σ2) of the Lugre model currently used by the control device 1 for correcting the nonlinear friction as the coefficient S1 of the Lugre model.

  The state observing unit 831 acquires the position command and speed command currently output by the control device 1 as the current position command S2 and speed command S3.

  The state observing unit 831 acquires, as the position feedback S4, the position feedback (used in the feedback control for generating the current position command and speed command) acquired by the control device 1 one cycle before.

  The determination data acquisition unit 832 acquires determination data D that is an index indicating a result when the machine is controlled under the state variable S. The determination data D includes position feedback D1.

  The determination data acquisition unit 832 acquires, as the position feedback D1, position feedback obtained as a result of controlling the machine based on the Lugre model coefficient S1, the position command S2, and the speed command S3.

  The reinforcement learning unit 833 uses the state variable S and the determination data D to learn the correlation between the Lugre model coefficient S1, the position command S2, the speed command S3, and the position feedback S4. That is, the reinforcement learning unit 833 generates a model structure indicating the correlation between the constituent elements S1, S2, S3, and S4 of the state variable S. The reinforcement learning unit 833 includes a reward calculation unit 834 and a value function update unit 835.

  The reward calculation unit 834 relates to a result of position control when the coefficient of the Lugre model is set based on the state variable S (corresponding to the determination data D used in the next learning cycle in which the state variable S is acquired). Reward R is calculated.

  The value function updating unit 835 updates the function Q representing the value of the coefficient of the Lugre model using the reward R. When the value function update unit 835 repeats the update of the function Q, the reinforcement learning unit 833 learns the correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4.

  An example of an algorithm for reinforcement learning executed by the reinforcement learning unit 833 will be described. The algorithm according to this example is known as Q-learning (Q-learning), and the behavior s in the state s is defined as an independent variable with the behavior s state s and the behavior a that the behavior subject can select in the state s. This is a method for learning a function Q (s, a) representing the value of an action when a is selected. The optimal solution is to select the action a that has the highest value function Q in the state s. The value function Q is iteratively updated by repeating trial and error by starting Q learning in a state where the correlation between the state s and the action a is unknown, and selecting various actions a in an arbitrary state s. Approach the solution. Here, when the environment (that is, the state s) changes as a result of selecting the action a in the state s, a reward (that is, the weight of the action a) r corresponding to the change is obtained, and a higher reward By inducing learning to select an action a that gives r, the value function Q can be brought close to the optimal solution in a relatively short time.

The updating formula of the value function Q can be generally expressed as the following formula 2. In Equation 2, s _t and a _t is a state and behavior at each time t, the state by action a _t is changed to s _{t + 1.} r t _{+ 1} is a reward obtained by the state changes from s _t in s _{t + 1.} The term maxQ means Q when the action a having the maximum value Q at time t + 1 (and considered at time t) is performed. α and γ are a learning coefficient and a discount rate, respectively, and are arbitrarily set such that 0 <α ≦ 1 and 0 <γ ≦ 1.

  When the reinforcement learning unit 833 executes Q-learning, the state variable S observed by the state observation unit 831 and the determination data D acquired by the determination data acquisition unit 832 correspond to the update state s, and the current state, that is, the position command The behavior of how to determine the coefficient S1 of the Lugre model for S2, the speed command S3 and the position feedback S4 corresponds to the behavior a of the update formula, and the reward R calculated by the reward calculation unit 834 is an update formula Corresponds to the reward r. Therefore, the value function updating unit 835 repeatedly updates the function Q representing the value of the coefficient of the Lugre model with respect to the current state by Q learning using the reward R.

  The reward calculation unit 834 performs, for example, machine control based on the determined Lugre model coefficient S1, and when the result of the position control is determined to be “suitable”, the reward R may be set to a positive (plus) value. it can. On the other hand, when the result of the position control is determined as “No”, the reward R can be set to a negative (minus) value. The absolute values of the positive and negative rewards R may be the same or different.

  The case where the result of the position control is “appropriate” is, for example, a case where the difference between the position feedback D1 and the position command S2 is within a predetermined threshold. The case where the result of the position control is “No” is, for example, a case where the difference between the position feedback D1 and the position command S2 exceeds a predetermined threshold. In other words, the position command S2 is “appropriate” if the position control is accurately realized more than a predetermined standard, and “no” otherwise.

  The result of the position control can be set not only in two ways, “appropriate” and “not”, but in a plurality of stages. For example, the reward calculation unit 834 can set a stepwise reward so that the reward increases as the difference between the position feedback D1 and the position command S2 decreases.

  The value function updating unit 835 can have an action value table in which the state variable S, the determination data D, and the reward R are associated with the action value (for example, a numerical value) represented by the function Q. In this case, the act of the value function updating unit 835 updating the function Q is synonymous with the act of the value function updating unit 835 updating the behavior value table. At the start of Q-learning, the correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4 is unknown, so in the action value table, various state variables S and determination data D A reward R is prepared in a form associated with a value (function Q) determined at random. The reward calculator 834 can immediately calculate the reward R corresponding to the determination data D, and the calculated value R is written in the behavior value table.

  When Q learning is advanced using reward R according to the result of position control, learning is guided in a direction to select an action that can obtain higher reward R, and the environment changes as a result of executing the selected action in the current state In accordance with the state (that is, the state variable S and the determination data D), the action value value (function Q) for the action performed in the current state is rewritten, and the action value table is updated. By repeating this update, the value of the action value (function Q) displayed in the action value table is rewritten so that the more appropriate the action, the larger the value. In this way, the correlation between the current state of the unknown environment, that is, the position command S2, the speed command S3, and the position feedback S4, and the action corresponding thereto, that is, the coefficient S1 of the set Lugre model is gradually clarified. . That is, by updating the behavior value table, the correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4 is gradually brought closer to the optimal solution.

With reference to FIG. 6, the Q learning flow (that is, one form of the machine learning method) executed by the reinforcement learning unit 833 will be further described.
Step SA01: The value function updating unit 835 randomly selects the coefficient S1 of the Lugre model as an action to be performed in the current state indicated by the state variable S observed by the state observation unit 831 while referring to the action value table at that time. select.
Step SA02: The value function updating unit 835 takes in the state variable S of the current state that is being observed by the state observation unit 831.
Step SA03: The value function updating unit 835 takes in the determination data D in the current state acquired by the determination data acquisition unit 832.
Step SA04: The value function updating unit 835 determines whether the coefficient S1 of the Lugre model is appropriate based on the determination data D. If appropriate, the process proceeds to step SA05. If not, the process proceeds to step SA07.
Step SA05: The value function updating unit 835 applies the positive reward R obtained by the reward calculating unit 834 to the function Q update formula.
Step SA06: The value function updating unit 835 updates the action value table using the state variable S, the determination data D, the reward R, and the action value (updated function Q) in the current state.
Step SA07: The value function update unit 835 applies the negative reward R obtained by the reward calculation unit 834 to the update formula of the function Q.

  Reinforcement learning unit 833 repeats steps SA01 to SA07 to repeatedly update the behavior value table, and advances learning. It should be noted that the processing for obtaining the reward R and the value function updating processing from step SA04 to step SA07 are executed for each data included in the determination data D.

  When proceeding with reinforcement learning, for example, a neural network can be used instead of Q learning. FIG. 7A schematically shows a model of a neuron. FIG. 7B schematically shows a model of a three-layer neural network configured by combining the neurons shown in FIG. 7A. The neural network can be configured by, for example, an arithmetic device or a storage device imitating a neuron model.

The neuron shown in FIG. 7A outputs a result y for a plurality of inputs x (here, as an example, inputs x ₁ to x ₃ ). Each input x ₁ ~x _3, the weight w corresponding to the input x (w ₁ ~w ₃₎ is multiplied. As a result, the neuron outputs an output y expressed by the following equation (3). In Equation 3, the input x, the output y, and the weight w are all vectors. Further, θ is a bias, and f _k is an activation function.

  In the three-layer neural network shown in FIG. 7B, a plurality of inputs x (in this example, inputs x1 to x3) are input from the left side, and a result y (in this case, results y1 to y3 as an example) are input from the right side. Is output. In the illustrated example, each of the inputs x1, x2, and x3 is multiplied by a corresponding weight (generically expressed as W1), and each of the inputs x1, x2, and x3 is assigned to three neurons N11, N12, and N13. Have been entered.

  In FIG. 7B, the outputs of the neurons N11 to N13 are collectively represented by z1. z1 can be regarded as a feature vector obtained by extracting the feature amount of the input vector. In the illustrated example, each feature vector z1 is multiplied by a corresponding weight (generically represented by W2), and each feature vector z1 is input to two neurons N21 and N22. The feature vector z1 represents a feature between the weight W1 and the weight W2.

In FIG. 7B, the outputs of the neurons N21 to N22 are collectively represented by z2. z2 can be regarded as a feature vector obtained by extracting the feature amount of the feature vector z1. In the illustrated example, each feature vector z2 is multiplied by a corresponding weight (generically represented by W3), and each feature vector z2 is input to three neurons N31, N32, and N33. The feature vector z2 represents a feature between the weight W2 and the weight W3. Finally, the neurons N31 to N33 output the results y1 to y3, respectively.
It is also possible to use a so-called deep learning method using a neural network having three or more layers.

  By repeating such a learning cycle, the reinforcement learning unit 833 can automatically identify a feature that implies a correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4. become able to. At the start of the learning algorithm, the correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4 is substantially unknown, but the reinforcement learning unit 833 gradually increases the characteristics as the learning proceeds. Identify and interpret correlations. When the correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4 is interpreted to a certain level of reliability, the learning result that the reinforcement learning unit 833 repeatedly outputs is the current state, that is, the position command. It can be used to select an action (decision decision) as to what kind of Lugre model coefficient S1 should be set for S2, speed command S3 and position feedback S4. In this way, the reinforcement learning unit 833 generates a learning model that can output an optimal solution of behavior corresponding to the current state.

  FIG. 8 is a schematic functional block diagram of the control device 1 and the machine learning device 100 according to the third embodiment. The control device 1 of the present embodiment has a configuration required when the machine learning device 100 performs estimation (estimation mode). In each functional block shown in FIG. 8, the CPU 11 included in the control device 1 shown in FIG. 3 and the processor 101 of the machine learning device 100 execute the respective system programs, and the control device 1 and the machine learning device 100 This is realized by controlling the operation of each part.

  Similar to the second embodiment, the control device 1 of the present embodiment includes a data acquisition unit 70 and a correction torque estimation unit 80 configured on the machine learning device 100. The correction torque estimation unit 80 includes an estimation unit 85 and a correction torque calculation unit 82. Further, an acquisition data storage unit 71 for storing data acquired by the data acquisition unit 70 is provided on the nonvolatile memory 14, and a learning unit 83 is provided on the nonvolatile memory 104 of the machine learning device 100. A learning model storage unit 84 that stores a learning model constructed by machine learning is provided.

  The operations of the data acquisition unit 70 and the preprocessing unit 90 according to the present embodiment are the same as those of the second embodiment. The data acquired by the data acquisition unit 70 is converted (numerized, sampled, etc.) into a unified format handled by the machine learning device 100 by the preprocessing unit 90, and the state data S is generated. The state data S created by the preprocessing unit 90 is used for estimation by the machine learning device 100.

  Based on the state data S created by the preprocessing unit 90, the estimation unit 85 uses the learning model stored in the learning model storage unit 84 to estimate the coefficient S1 of the Lugre model. The estimation unit 85 of the present embodiment inputs the state data S input from the preprocessing unit 90 to the learning model generated by the learning unit 83 (parameters are determined), so that the coefficient of the Lugre model S1 is estimated and output.

  The correction torque calculation unit 82 calculates and outputs a correction torque based on the friction model using the result estimated by the estimation unit 85 (the combination S1 of friction model coefficients). The control device 1 adds the correction torque output from the correction torque calculation unit 82 to the current command.

  According to the second and third embodiments, the machine learning device 100 generates a learning model indicating the correlation between the coefficient S1 of the Lugre model and the position command S2, the speed command S3, and the position feedback S4, and the learning model is generated. Since the coefficient of the friction model is estimated by use, it is possible to easily obtain the optimum coefficient corresponding to various machines, usage environments, and the like.

  Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and can be implemented in various modes by making appropriate changes.

  For example, in the above-described embodiment, the control device 1 and the machine learning device 100 are described as devices having different CPUs (processors). However, the machine learning device 100 is a system stored in the ROM 11 and the CPU 11 provided in the control device 1. -It may be realized by a program.

  As a modification of the machine learning device 100, the learning unit 83 uses the state variable S and the determination data D obtained for each of a plurality of machines of the same type, and calculates an appropriate Lugre model coefficient common to those machines. Can learn. According to this configuration, the amount of the data set including the state variable S and the determination data D obtained in a certain time can be increased and more diverse data sets can be input, so that the learning speed and reliability can be improved. it can. Further, the Lugre model can be further optimized for each machine by using the learning model thus obtained as an initial value and performing additional learning for each machine.

  FIG. 9 shows a system 170 in which a plurality of machines are added to the control device 1. The system 170 includes a plurality of machines 160 and a machine 160 '. All machines 160 and 160 'are connected to each other by a wired or wireless network 172.

  Machine 160 and machine 160 'have the same type of mechanism. On the other hand, the machine 160 includes the control device 1, but the machine 160 ′ does not include the control device 1.

  In the machine 160 including the control device 1, the estimation unit 85 uses the learning model that is the learning result of the learning unit 83 to calculate the coefficient S1 of the Lugre model corresponding to the position command S2, the speed command S3, and the position feedback S4. Can be estimated. Further, the control device 1 of at least one machine 160 uses the state variable S and the determination data D obtained for each of the other machines 160 and 160 ′, and is common to all the machines 160 and 160 ′. The position control to be learned is learned, and the learning result can be shared by all the machines 160 and 160 ′. According to the system 170, it is possible to improve the speed and reliability of position control learning using a more diverse data set (including the state variable S and the determination data D) as an input.

  FIG. 10 shows a system 170 'comprising a plurality of machines 160'. The system 170 ′ includes a plurality of machines 160 ′ having the same machine configuration, and a machine learning device 120 independent of the control device 1 (or the machine learning device 100 included in the control device 1). The plurality of machines 160 ′ and the machine learning device 120 (or the machine learning device 100) are connected to each other by a wired or wireless network 172.

  The machine learning device 120 (or the machine learning device 100) learns the Lugre model coefficient S1 common to all the machines 160 ′ based on the state variable S and the determination data D obtained for each of the plurality of machines 160 ′. . The machine learning device 120 (or machine learning device 100) can estimate the coefficient S1 of the Lugre model corresponding to the position command S2, the speed command S3, and the position feedback S4 using the learning result.

  According to this configuration, the necessary number of machines 160 ′ are connected to the machine learning device 120 (or the machine learning device 100) when necessary regardless of the location and timing of each of the plurality of machines 160 ′. Can do.

  In the above-described embodiment, it is assumed that the control device 1 and the machine learning device 100 (or the machine learning device 120) are one information processing device installed locally, but the present invention is limited to this. For example, the control device 1 or the machine learning device 100 (or the machine learning device 120) may be implemented in an information processing environment called cloud computing, fog computing, edge computing, or the like.

  In the above-described embodiment, a method for determining a coefficient in a typical Lugre model as a friction model has been described. However, the present invention is not limited to this, and the Seven parameter model, the State variable model, the Karnopp model, The present invention can be applied to determination of coefficients of various friction models such as LuGre model, Modified Dahl model, and M2 model.

  In the above-described embodiment, the processing machine is mainly exemplified as the machine. However, the present invention is not limited to this, and has a drive mechanism, typically a positioning mechanism, in which friction becomes a problem. The present invention can be applied to various machines (for example, medical robots, disaster robots, construction robots, etc.).

  In the above-described embodiment, the coefficient of the friction model is obtained based on the control system shown in FIG. 13, but the present invention is not limited to this, and can be applied to various control systems obtained by modifying the coefficient. Is possible. For example, as shown in FIG. 14, instead of the speed command, a control system in which s = speed command equivalent that is a derivative of the position command is input to the friction model may be used. In this case, the state observation unit 831 of the machine learning device 100 observes s corresponding to the speed command instead of the speed command. According to this configuration, since the correction torque can be calculated only by the position command, there is an advantage that the calculation of the correction torque can be completed only by the control device 1 side.

  Or as shown in FIG. 15, you may use the control system which inputs a position feedback and a speed feedback into a friction model instead of a position command and a speed command. In this case, the state observation unit 831 of the machine learning device 100 observes position feedback instead of position command and speed feedback instead of speed command. This configuration is easy to realize on the axis control circuit 30 side. High-speed processing is possible, and since the feedback is used, the actual friction force is easy to estimate.

1 control device 11 CPU
12 ROM
13 RAM
DESCRIPTION OF SYMBOLS 14 Nonvolatile memory 18, 19, 21, 22 Interface 20 Bus 30 Axis control circuit 40 Servo amplifier 50 Servo motor 60 Operation panel 70 Data acquisition part 71 Acquisition data storage part 80 Correction torque estimation part 81 Optimization part 82 Correction torque calculation part 83 learning unit 831 state observation unit 832 determination data acquisition unit 833 reinforcement learning unit 834 reward calculation unit 835 value function update unit 84 learning model storage unit 85 estimation unit 100 machine learning device (included in the control device 1)
101 processor 102 ROM
103 RAM
104 Non-volatile memory 120 Machine learning device (independent of control device 1)
160, 160 'machine 170, 170' system 172 network

Claims (8)

A control device that performs position control in consideration of friction with respect to one or more shafts of a machine,
A data acquisition unit for acquiring at least a position command and position feedback;
And a correction torque estimating unit that estimates a coefficient of a friction model when performing the position control based on a position deviation that is a difference between the position command and the position feedback. The control device according to claim 1, wherein the correction torque estimation unit includes an optimization unit that estimates a coefficient of a friction model by solving an optimization problem that minimizes the position deviation. The correction torque estimation unit includes a learning unit that performs machine learning using a state variable including a coefficient of the friction model, a position command and position feedback, and a speed command or speed feedback to generate a learning model. The control device described. The control device according to claim 3, wherein the learning unit performs reinforcement learning based on determination data indicating a result of the position control. The correction torque estimating unit
A coefficient of the friction model, a position command and position feedback, and a learning model storage unit that stores a learning model that has been machine-learned using a speed command or speed feedback;
The control device according to claim 1, further comprising: an estimation unit configured to estimate a coefficient of the friction model using the learning model based on a position command and position feedback, and a speed command or speed feedback. The control device according to claim 1, wherein the data acquisition unit acquires data from a plurality of the machines. The friction model is a Lugre Model, a Seven parameter model, a State variable model, a Karnopp model, a LuGre model, or a Modified Dahl model, or an M2 model. A control method for performing position control in consideration of friction with respect to one or more shafts of a machine,
A data acquisition step for acquiring at least a position command and position feedback;
And a correction torque estimation step of estimating a coefficient of a friction model when performing the position control based on a position deviation that is a difference between the position command and the position feedback.

Images