Control device, control system, and control method

Abstract

To improve the accuracy of feedforward control for an object to be controlled.SOLUTION: A feedback operation amount rb to a controlled object 200 is determined by feedback control means 110 based on an error eq between a target value qr and a control amount q. Feedforward compensation means 120 predicts a feedforward compensation value rf of the feedback operation amount rb from a disturbance d using a prediction model Mp. Learning means 130 performs machine learning on the prediction model Mp using teacher data Ds. The learning means 130 adds to the teacher data Ds a combination including the disturbance d and an operation amount r when an absolute value of the error eq is smaller than a reference value α.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a control device, a control system, and a control method that perform feedforward control on a controlled object. [Background technology]

[0002] Conventionally, control devices that perform feedforward control on a controlled object have been known. For example, Japanese Patent Laid-Open Publication No. 10-222207 (Patent Document 1) discloses a feedforward control device that uses a feedforward signal generator to control a predetermined process that is subject to disturbances. The feedforward control device includes a parameter learner that learns parameters required for calculations in the feedforward signal generator from parameters calculated in the feedforward signal generator. With this feedforward control device, stable control performance can be obtained even when disturbances fluctuate, thanks to the parameter learner's dynamic learning function for the parameters of the feedforward signal generator, and high-precision control performance can be maintained even when the characteristics of the process change over time. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 10-222207 Summary of the Invention [Problem to be solved by the invention]

[0004] The configuration of the feedforward compensator included in the feedforward signal generator disclosed in Patent Document 1 is predetermined. Generally, a configuration for calculating a feedforward compensation value for feedback control, such as a feedforward compensator, is designed as an inverse model (inverse system) of the process to be controlled. Designing an inverse model of a process requires analysis of the process, making it difficult to quickly update the structure of the inverse model in response to fluctuations in disturbances and changes in process characteristics. Therefore, with the feedforward control device disclosed in Patent Document 1, it may be difficult to suppress the influence of disturbances on the control of the controlled object, depending on the degree of fluctuation in the disturbance or the degree of change in the process characteristics.

[0005] The present disclosure has been made to solve the above-mentioned problems, and its purpose is to improve the accuracy of feedforward control of a control target. [Means for solving the problem]

[0006] A control device according to one aspect of the present disclosure determines a feedforward compensation value of a feedback manipulated variable to a control target such that a controlled variable of the control target affected by a disturbance approaches a target value. The feedback manipulated variable is determined by a feedback control means based on an error between the target value and the controlled variable. The sum of the feedback manipulated variable and the feedforward compensation value is output as a manipulated variable to the control target. The control device includes a feedforward compensation means and a learning means. The feedforward compensation means determines the feedforward compensation value from the disturbance using a prediction model. The learning means performs machine learning on the prediction model using training data. The learning means obtains at least one first combination including a target value, a manipulated variable corresponding to the target value, a disturbance, and a controlled variable corresponding to both the disturbance and the manipulated variable. The learning means adds at least one second combination including a disturbance and a manipulated variable for which the absolute value of the error is smaller than a first reference value from the at least one first combination to the training data.

[0007] According to this disclosure, because teacher data is selected from the results of feedforward control, it is possible to adapt a prediction model to the characteristics of disturbances and the controlled object in real time in parallel with the feedforward control, thereby improving the accuracy of feedforward control of the controlled object.

[0008] In the above disclosure, the learning means may calculate an average value of the manipulated variable when the absolute value of the disturbance is smaller than a second reference value in the training data. The learning means may approximate the relationship between the disturbance and the compensation value as a difference between the manipulated variable and the average value, expressed by the prediction model, as a function using the compensation value as a response variable and the disturbance as an explanatory variable. The learning means may terminate the machine learning when a ratio of the number of fourth combinations, among at least one second combination, in which the absolute value of the disturbance is larger than the second reference value to the number of third combinations, among at least one first combination, in which the absolute value of the disturbance is larger than the second reference value, is larger than a third reference value.

[0009] According to this disclosure, machine learning continues until the accuracy of the predictive model becomes sufficiently high, thereby enabling sufficient improvement in the accuracy of feedforward control of the controlled object.

[0010] In the above disclosure, the learning means may calculate an average value of the manipulated variable when the absolute value of the disturbance is smaller than a second reference value in the training data. The learning means may approximate the relationship between the disturbance and the manipulated variable expressed by the prediction model as a function using the manipulated variable as a response variable and the disturbance as an explanatory variable. The learning means may terminate machine learning when a ratio of the number of at least one fourth combination, in which the absolute value of the disturbance is larger than the second reference value, to the number of at least one third combination, in which the absolute value of the disturbance is larger than the second reference value, is larger than a third reference value. The feedforward compensation means may determine, as the feedforward compensation value, a value obtained by subtracting the average value of the manipulated variable from the manipulated variable predicted from the disturbance by the prediction model.

[0011] According to this disclosure, machine learning continues until the accuracy of the predictive model becomes sufficiently high, thereby enabling sufficient improvement in the accuracy of feedforward control of the controlled object.

[0012] In the above disclosure, the learning means may resume the machine learning when the machine learning is completed and the ratio is smaller than a third reference value.

[0013] According to this disclosure, the prediction model is re-adapted to the characteristics of the controlled object in response to changes in the characteristics of the controlled object, thereby making it possible to suppress a decrease in the accuracy of feedforward control due to changes in the characteristics of the controlled object.

[0014] A control system according to another aspect of the present disclosure outputs a sum of a feedback manipulated variable to a controlled object and a feedforward compensation value as a manipulated variable so that a controlled variable of the controlled object affected by a disturbance approaches a target value. The control system includes a feedback control device, a feedforward compensation device, and a learning device. The feedback control device determines a feedback manipulated variable based on an error between a target value and a controlled variable. The feedforward compensation device determines a feedforward compensation value from the disturbance using a prediction model. The learning device performs machine learning on the prediction model using training data. The learning device obtains at least one first combination including a target value, a manipulated variable corresponding to the target value, a disturbance, and a controlled variable corresponding to both the disturbance and the manipulated variable. The learning device adds at least one second combination including a disturbance and a manipulated variable for which the absolute value of the error is smaller than a first reference value, from the at least one first combination, to the training data.

[0015] According to this disclosure, because teacher data is selected from the results of feedforward control, it is possible to adapt a prediction model to the characteristics of disturbances and the controlled object in real time in parallel with the feedforward control, thereby improving the accuracy of feedforward control of the controlled object.

[0016] A control method according to another aspect of the present disclosure determines a feedforward compensation value of a feedback manipulated variable for a controlled object such that a controlled variable of the controlled object affected by a disturbance approaches a target value. The feedback manipulated variable is determined by a feedback control means based on an error between the target value and the controlled variable. The sum of the feedback manipulated variable and the feedforward compensation value is output as a manipulated variable to the controlled object. The control method includes the steps of determining a feedforward compensation value from the disturbance using a prediction model and performing machine learning on the prediction model using training data. The step of performing machine learning includes the steps of acquiring at least one first combination including a target value, a manipulated variable corresponding to the target value, the disturbance, and a controlled variable corresponding to both the disturbance and the manipulated variable, and adding at least one second combination including a disturbance and a manipulated variable from the at least one first combination where the absolute value of the error is smaller than a first reference value to the training data.

[0017] According to this disclosure, because teacher data is selected from the results of feedforward control, it is possible to adapt a prediction model to the characteristics of disturbances and the controlled object in real time in parallel with the feedforward control, thereby improving the accuracy of feedforward control of the controlled object. [Effects of the Invention]

[0018] According to the control device, feedforward compensation device, control system, and control method disclosed herein, it is possible to improve the accuracy of feedforward control of a controlled object. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a block diagram showing a functional configuration of a control device according to a first embodiment. [Figure 2] FIG. 1 is a distribution diagram of disturbances and manipulated variables. [Figure 3] 3 is a diagram showing the relationship between the manipulated variable and the objective variable of the prediction model in FIG. 2. FIG. [Figure 4]FIG. 1 is a diagram illustrating the relationship between a dependent variable and explanatory variables expressed by a trained prediction model. [Figure 5] This is a Venn diagram showing the inclusion relationships of the four combinations. [Figure 6] 10A and 10B are diagrams showing time charts of a controlled variable, a manipulated variable, and a disturbance when a prediction model has not yet been learned. [Figure 7] 10A and 10B are diagrams showing time charts of the controlled variable, the manipulated variable, and the disturbance when the prediction model Mp has already been learned. [Figure 8] 2 is a flowchart showing the flow of processing performed by the feedback control system, the feedforward control system, and the learning means 130 in FIG. 1. FIG. [Figure 9] 9 is a flowchart showing a specific processing flow of the machine learning processing of FIG. 8. [Figure 10] 3 is a diagram showing the flow of additional learning processing performed by the learning means of FIG. 2. FIG. [Figure 11] FIG. 10 is a block diagram showing a functional configuration of a control device according to a first modification of the first embodiment. [Figure 12] 10 is a flowchart showing another example of the specific processing flow of the machine learning processing of FIG. 2. [Figure 13] FIG. 10 is a block diagram showing the functional configuration of a control system according to a second embodiment. [Figure 14] FIG. 11 is a schematic diagram illustrating an example of a network configuration of a control system according to a third embodiment. [Figure 15] FIG. 15 is a block diagram showing an example of the hardware configuration of the control device of FIG. 14. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, embodiments will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference numerals, and their description will not be repeated in principle.

[0021] [Embodiment 1] <Application example> Fig. 1 is a block diagram showing the functional configuration of a control device 100 according to embodiment 1. As shown in Fig. 1, the control device 100 includes feedback control means 110, feedforward compensation means 120, learning means 130, storage means 140, a subtractor 150, and an adder 160.

[0022] The control device 100 outputs the sum of the feedback control value rb and the feedforward compensation value rf to the control object 200 as the control value r to the control object 200 so that the control value q, which is the output value of the control object 200 subjected to a disturbance d, approaches the target value qr. The storage means 140 stores the prediction model Mp and training data Ds. The control device 100 and the control object 200 are connected via a network (for example, the Internet or a cloud system) and may be located remotely from each other. Examples of machine learning algorithms used to construct the prediction model Mp include a deep binary tree or a support vector machine.

[0023] Hereinafter, a configuration including the feedback control means 110 and the subtractor 150 will also be referred to as a feedback control system, and a configuration including the feedforward compensation means 120 and the adder 160 will also be referred to as a feedforward control system. Note that the disturbance d is a quantity that disturbs the state of the control system including the control device 100 and the controlled object 200. The disturbance d includes, for example, the amount of light, voltage, current, and temperature that are accidentally or suddenly input to the controlled object.

[0024] The subtractor 150 outputs the error eq (=qr-q) between the target value qr and the controlled variable q to the feedback control means 110. The feedback control means 110 determines a feedback manipulated variable rb based on the error eq and outputs it to the adder 160. The feedforward compensation means 120 predicts a feedforward compensation value rf from the disturbance d using the prediction model Mp and outputs it to the adder 160. The adder 160 outputs the sum of the feedback manipulated variable rb and the feedforward compensation value rf as the manipulated variable r to the controlled object 200 and the learning means 130.

[0025] The adder 160 outputs the sum of the feedback manipulated variable rb and the feedforward compensation value rf as the manipulated variable r to the controlled object 200 and the learning means 130 .

[0026] The learning means 130 performs machine learning on the prediction model Mp using the training data Ds. The learning means 130 acquires at least one combination Cm1 (first combination) including a target value qr, a manipulated variable r corresponding to the target value qr, a disturbance d, and a controlled variable q corresponding to both the disturbance d and the manipulated variable r. Of the at least one combination Cm1, the learning means 130 adds at least one combination Cm2 (second combination) including the disturbance d and the manipulated variable r when the absolute value of the error eq is smaller than a reference value α (first reference value) (when high-precision control is performed on the control object 200) to the training data Ds as correct answer data for machine learning. The reference value α can be determined appropriately based on, for example, an actual machine experiment, a simulation, a product specification value, or a control value of a certain manufacturing process.

[0027] The learning means 130 calculates the average value rb0 of the manipulated variable r in the training data Ds when the absolute value of the disturbance d is smaller than a reference value β (second reference value) (when the control system is in a steady state). The reference value β can be determined appropriately based on, for example, actual machine experiments, simulations, product specifications, or manufacturing process control values. The learning means 130 approximates the relationship between the disturbance d and the feedforward compensation value rf, which is the difference between the manipulated variable r and the average value rb0 (=r-rb0), as a function (regression curve) with the feedforward compensation value rf as the objective variable and the disturbance d as the explanatory variable. The prediction model Mp includes this function. The learning means 130 terminates machine learning for the prediction model Mp when the ratio Cr of the number of combinations Cm4 (fourth combinations) in which the absolute value of the disturbance d is greater than the reference value β among at least one combination Cm2 to the number of combinations Cm3 (third combinations) in which the absolute value of the disturbance d is greater than the reference value β among at least one combination Cm1 is greater than the reference value β.

[0028] When machine learning (initial learning or additional learning) for the prediction model Mp is completed and the ratio Cr is equal to or less than the reference value δ, the learning means 130 determines that the characteristics of the controlled object 200 have changed and resumes machine learning (additional learning) for the prediction model Mp. The reference value δ can be determined appropriately based on, for example, an actual machine experiment, a simulation, a product specification value, or a control value for a certain manufacturing process. The characteristics of the controlled object 200 include, for example, the correspondence between the disturbance d and the manipulated variable r and the controlled variable q.

[0029] According to the control device 100, the teacher data Ds is selected from the results of the feedforward control, and therefore the prediction model Mp can be adapted to the disturbance d and the characteristics of the control object 200 in real time in parallel with the feedforward control. As a result, the accuracy of the feedforward control for the control object 200 can be improved. Furthermore, machine learning is continued until the accuracy of the prediction model Mp becomes sufficiently high, and therefore the accuracy of the feedforward control for the control object 200 can be sufficiently improved. Furthermore, the prediction model Mp is re-adapted to the characteristics in accordance with changes in the characteristics of the control object 200, and therefore a decrease in the accuracy of the feedforward control due to changes in the characteristics of the control object 200 can be suppressed.

[0030] Fig. 2 is a distribution diagram of the disturbance d and the manipulated variable r. In Fig. 2, square dots represent combinations Cm1 among multiple combinations Cm1 when the absolute value of the error eq is equal to or greater than the reference value α, and circle dots represent combinations Cm2 when the absolute value of the error eq is smaller than the reference value α. The learning means 130 adds the multiple combinations Cm2 represented by the circle dots to the training data Ds.

[0031] Fig. 3 is a diagram showing the relationship between the manipulated variable r and the response variable rf of the prediction model Mp in Fig. 2. As shown in Fig. 3, data obtained by subtracting the average value rb0 from the manipulated variable r of each of the multiple combinations Cm2 corresponds to the response variable rf.

[0032] 4 is a diagram showing the relationship between the objective variable rf and the explanatory variable d expressed by the trained prediction model Mp. This relationship is approximated as a regression curve Rc by the learning means 130. As shown in FIG. 4, the prediction model Mp receives a disturbance d and outputs the objective variable rf corresponding to the disturbance d on the regression curve Rc.

[0033] FIG. 5 is a Venn diagram showing the inclusion relationship of four data sets Scm1, Scm2, Scm3, and Scm4, each including four combinations Cm1 to Cm4. Data set Scm1 is a data set including all of the combinations Cm1, each including a target value qr, a manipulated variable r, a disturbance d, and a controlled variable q. Data set Scm2 is a data set including all of the combinations Cm2, each including a disturbance d and a manipulated variable r, when the absolute value of the error eq is smaller than a reference value α, from data set Scm1. Data set Scm3 is a data set including all of the combinations Cm3, each including a disturbance d whose absolute value is greater than a reference value β, from data set Scm1. Data set Scm4 is a data set including all of the combinations Cm4, each including a disturbance d whose absolute value is greater than the reference value β, from data set Scm2. As shown in FIG. 5, data set Scm1 includes data sets Scm2 and Scm3. Data set Scm4 is the intersection (intersection) of data sets Scm2 and Scm3. The average value rb0 is the average value of the data set obtained by removing the data set Scm4 from the data set Scm2.

[0034] Fig. 6 is a diagram showing time charts of the controlled variable q, the manipulated variable r, and the disturbance d when the prediction model Mp has not yet been learned. As shown in Fig. 6, the disturbance d is input as a step response of a first-order lag system at a time of 500 seconds. In response to the disturbance d, pulse-like noise is superimposed on the controlled variable q in the time interval from 500 seconds to 600 seconds.

[0035] Fig. 7 is a diagram showing time charts of the controlled variable q, the manipulated variable r, and the disturbance d when the predictive model Mp has been trained. As in Fig. 6, a disturbance d is input as a step response of a first-order lag system at 500 seconds in Fig. 7. However, the controlled variable q hardly exhibits the pulse-like noise shown in Fig. 6. In Fig. 7, the feedback manipulated variable rb is corrected by the feedforward compensation value rf predicted by the trained predictive model Mp, thereby maintaining the accuracy of control of the controlled object 200.

[0036] 8 is a diagram showing a flowchart illustrating the flow of processing performed by each of the feedback control system, the feedforward control system, and the learning means 130 in FIG. 1. The routines corresponding to the flowcharts of the feedback control system and the feedforward control system are executed, for example, at each sampling time. The routine corresponding to the flowchart of the learning means 130 is executed, for example, in response to the first execution of the routine corresponding to the flowchart of the feedforward control system. Hereinafter, steps will simply be abbreviated as S.

[0037] 8, in S111, the subtractor 150 calculates the error eq between the target value qr and the controlled variable q, and outputs the error eq to the feedback control means 110. In S312, the feedback control means 110 determines the feedback controlled variable rb based on the error eq, outputs the determined feedback controlled variable rb to the adder 160, and ends the process.

[0038] In S121, the feedforward compensation means 120 determines a feedforward compensation value rf from the disturbance d and outputs it to the adder 160. In S122, the adder 160 outputs the sum of the feedback manipulated variable rb and the feedforward compensation value rf as the manipulated variable r to the control object 200 and the learning means 130, and the process ends.

[0039] In S130, the learning means 130 performs machine learning on the prediction model Mp using the training data Ds, and then ends the process.

[0040] 9 is a flowchart showing a specific process flow of the machine learning process S130 in FIG. 8. In S131, the learning means 130 acquires the target value qr, the manipulated variable r, the disturbance d, and the controlled variable q to calculate the error eq, and proceeds to S132. In S132, the learning means 130 determines whether the absolute value of the error eq is smaller than the reference value α. If the absolute value of the error eq is equal to or greater than the reference value α (NO in S132), the learning means 130 proceeds to S137. If the absolute value of the error eq is smaller than the reference value α (YES in S132), the learning means 130 adds a combination Cm2 including the disturbance d and the manipulated variable r to the training data Ds in S133, and proceeds to S134. In S134, the learning means 130 calculates the average value rb0 of the manipulated variable r when the absolute value of the disturbance d in the training data Ds is smaller than the reference value β, and proceeds to S135. In S135, the learning means 130 approximates the relationship between the disturbance d represented by the prediction model Mp and the feedforward compensation value rf, which is the difference between the manipulated variable r and the average value rb0, as a function (regression curve) with the feedforward compensation value rf as the objective variable and the disturbance d as the explanatory variable, and proceeds to S136. In S136, the learning means 130 calculates the ratio Cr (=M2 / M1) of the number M2 of combinations Cm4, among at least one combination Cm2, whose absolute value of the disturbance d is greater than the reference value β, to the number M1 of combinations Cm3, among at least one combination Cm1, whose absolute value of the disturbance d is greater than the reference value β, and proceeds to S137. In S137, the learning means 130 determines whether the ratio Cr is greater than the reference value δ. If the ratio Cr is equal to or less than the reference value δ (NO in S137), the learning means 130 determines that the accuracy of the feedforward control of the controlled object 200 is insufficient, and returns to S131. If the ratio Cr is greater than the reference value δ (YES in S137), the accuracy of the feedforward control of the controlled object 200 is deemed to have increased sufficiently, and the learning means 130 ends the machine learning.

[0041] FIG. 10 is a diagram showing the flow of additional learning processing performed by the learning means 130 of FIG. 2. The processing shown in FIG. 10 is executed, for example, at each sampling time after the start of the initial machine learning. In S131A, the learning means 130 determines whether the machine learning has ended and whether the ratio Cr is equal to or less than the reference value δ. If the machine learning has not ended or if the ratio Cr is greater than the reference value δ (NO in S131A), the learning means 130 determines that the characteristics of the controlled object 200 have not changed and the existing trained prediction model Mp is adapted to the controlled object 200, and ends the processing. If the machine learning has ended and the ratio Cr is equal to or less than the reference value δ (YES in S131A), the learning means 130 determines that the characteristics of the controlled object 200 have changed since the previous machine learning ended, and resumes machine learning on the prediction model Mp at S130 similar to that of FIG. 8 in order to re-adapt the prediction model Mp to the characteristics of the controlled object 200. According to the control device 100, additional learning is performed on the prediction model Mp in response to changes in the characteristics of the controlled object 200, and therefore, a decrease in the accuracy of the feedforward control due to changes in the characteristics of the controlled object 200 can be suppressed.

[0042] [First Modification of First Embodiment] In the first embodiment, a configuration including both a feedback control system and a feedforward control system has been described. In the first modification of the first embodiment, a configuration not including a feedback control system will be described.

[0043] Fig. 11 is a block diagram showing a functional configuration of a control device 100A according to a first modification of the first embodiment. The configuration of the control device 100A is the same as that of the control device 100 in Fig. 1 except that the subtractor 150 and the feedback control means 110 are removed. Since the rest of the configuration is the same, the description of the same configuration will not be repeated. Note that the adder 160 does not have to be included in the control device 100A.

[0044] The control device 100A determines a feedforward compensation value rf of a feedback manipulated variable rb to the control object 200 such that a controlled variable q of the control object 200 subjected to a disturbance d approaches a target value qr. According to the control device 100A, by adding a control device to an existing feedback control system while leaving the existing feedback control system, the existing feedback control system can be easily expanded into a control system including a feedforward control system and a learning function.

[0045] [Modification 2 of Embodiment 1] In the first embodiment, a case has been described in which the relationship between the disturbance d and the feedforward compensation value rf as the difference between the manipulated variable r and the average value rb0 is expressed by the predictive model. In the second modification of the first embodiment, a case will be described in which the relationship expressed by the predictive model is the relationship between the disturbance d and the manipulated variable r.

[0046] FIG. 12 is a flowchart showing another example of the specific processing flow of the machine learning process S130 of FIG. 2. The flowchart shown in FIG. 12 is a flowchart in which S135 of FIG. 9 is replaced with S135B. As shown in FIG. 12, the learning means 130 performs S131 to S134 as in the first embodiment, and then in S135B, approximates the relationship between the disturbance d and the manipulated variable r expressed by the prediction model Mp as a function (regression curve) with the manipulated variable r as the objective variable and the disturbance d as the predictor variable, and proceeds to S136. The learning means 130 performs S136 and S137 as in the first embodiment, and then terminates the processing. The feedforward compensation means 120 predicts the manipulated variable r from the disturbance d using the prediction model Mp, and outputs the value (=r-rb0) obtained by subtracting the average value rb0 from the manipulated variable r to the adder 160 as the feedforward compensation value rf.

[0047] As described above, the control device and control method according to the first embodiment and the first and second modifications can improve the accuracy of feedforward control of a controlled object.

[0048] [Embodiment 2] In the first embodiment, a case where a feedback control system, a feedforward control system, and a configuration for performing machine learning on a predictive model are included in one control device has been described. In the second embodiment, a configuration will be described in which the feedback control system, the feedforward control system, and a configuration for performing machine learning on a predictive model are separated into separate devices.

[0049] Fig. 13 is a block diagram showing the functional configuration of a control system 2 according to embodiment 2. In Fig. 13, components denoted by the same reference symbols as those in Fig. 1 have the same functions as the components identified by the same reference symbols described in embodiment 1, and therefore description of these similar components will not be repeated.

[0050] As shown in FIG. 13 , the control system 2 includes a feedback control device 11, a feedforward compensation device 12, and a learning device 13. The feedback control device 11 includes a feedback control means 110 and a subtractor 150. The feedforward compensation device 12 includes a feedforward compensation means 120 and an adder 160. The learning device 13 includes a learning means 130 and a storage means 140. The feedback control device 11, the feedforward compensation device 12, the learning device 13, and the controlled object 200 may be connected to each other via a network and located remotely from each other. Note that the adder 160 may be included in the feedback control device 11 instead of the feedforward compensation device 12.

[0051] According to the control system 2, an existing control system can be easily expanded by adding a feedforward compensation device and a learning device to the feedback control device while leaving the existing feedback control device in place.

[0052] As described above, the control system, feedforward compensation device, and control method can improve the accuracy of feedforward control of a controlled object.

[0053] [Embodiment 3] In the third embodiment, as an example of the control device according to the first embodiment, a configuration in which the control device includes a PLC (Programmable Logic Controller) will be described.

[0054] <Example of control system network configuration> 14 is a schematic diagram showing an example of a network configuration of a control system 3 according to the third embodiment. As shown in FIG. 14, the control system 3 includes a device group in which a plurality of devices are configured to be able to communicate with each other. Typically, the devices may include a control device 300 that is a processing entity that executes a control program, and peripheral devices connected to the control device 300. The control device 300 has a functional configuration similar to that of the control device 100 shown in FIG. 1.

[0055] The control device 300 corresponds to an industrial controller that controls control targets such as various facilities or devices. The control device 300 is a type of computer that executes control calculations and typically includes a PLC (Programmable Logic Controller). The control device 300 is connected to field devices 200C via a field network 20. The control device 300 exchanges data with at least one field device 200C via the field network 20.

[0056] The control calculations executed by the control device 300 include a process of collecting data collected or generated in the field device 200C, a process of generating data such as a command value (operation amount) for the field device 200C, and a process of transmitting the generated output data to the target field device 200C. The data collected or generated in the field device 200C includes data related to disturbances input to the field device 200C and a control amount resulting from the actual operation of the field device 200C in accordance with the command value. The command value for the field device 200C is determined by adding a feedforward compensation value predicted from the disturbance using a prediction model to a control amount provisionally calculated based on the error between a control target value (target value) calculated based on a control program executed by the control device 300 and the actual control amount.

[0057] It is preferable that the field network 20 employs a bus or network that performs periodic communication. Known examples of such buses or networks that perform periodic communication include EtherCAT (registered trademark), EtherNet / IP (registered trademark), DeviceNet (registered trademark), and CompoNet (registered trademark). EtherCAT (registered trademark) is preferable because it guarantees the arrival time of data.

[0058] Any field device 200C can be connected to the field network 20. The field device 200C includes an actuator that applies some physical action to a robot or conveyor in the field, and an input / output device that exchanges information with the field.

[0059] In the control system 3, the field device 200C includes a plurality of servo drivers 220_1 and 220_2, and a plurality of servo motors 222_1 and 222_2 connected to the plurality of servo drivers 220_1 and 220_2, respectively. The field device 200C is an example of a "controlled object."

[0060] The servo drivers 220_1 and 220_2 drive the corresponding one of the servo motors 222_1 and 222_2 in accordance with a command value (for example, a position command value or a speed command value) from the control device 300. In this way, the control device 300 can control the field device 200C.

[0061] The control device 300 is also connected to other devices via a higher-level network 32. The higher-level network 32 is connected to the Internet 900, which is an external network, via a gateway 700. The higher-level network 32 may employ Ethernet (registered trademark) or EtherNet / IP (registered trademark), which are common network protocols. More specifically, at least one server device 600 and at least one display device 500 may be connected to the higher-level network 32.

[0062] The server device 600 may be a database system or a manufacturing execution system (MES). The manufacturing execution system acquires information from controlled manufacturing equipment or facilities to monitor and manage the entire production, and may also handle order information, quality information, shipping information, and the like. In addition to these, a device providing information services may be connected to the upper network 32. Possible information services include processing that acquires information from controlled manufacturing equipment or facilities and performs macro or micro analysis. For example, possible information services include data mining that extracts some characteristic trend contained in information from controlled manufacturing equipment or facilities, or machine learning tools that perform machine learning based on information from controlled facilities or machines.

[0063] The display device 500 receives operations from the user, outputs commands to the control device 300 in response to the user operations, and graphically displays the results of calculations performed by the control device 300, etc.

[0064] A support device 400 can be connected to the control device 300. The support device 400 may be connected to the control device 300 via the higher-level network 32 or the Internet 900. The support device 400 is a device that supports the control device 300 in making the preparations necessary for controlling the control target. Specifically, the support device 400 provides a development environment (program creation and editing tools, parsers, compilers, etc.) for programs executed by the control device 300, a setting environment for setting configuration information (configuration) for the control device 300 and various devices connected to the control device 300, a function for outputting generated programs to the control device 300, and a function for online correction and modification of programs executed on the control device 300.

[0065] In the control system 3, the control device 300, the support device 400, and the display device 500 are configured as separate entities, but a configuration may be adopted in which all or part of these functions are integrated into a single device.

[0066] The control device 300 is not limited to being used only at one production site, but may also be used at other production sites, and may also be used at multiple different lines within one production site.

[0067] <Example of control device hardware configuration> Fig. 15 is a block diagram showing an example of the hardware configuration of the control device 300 of Fig. 14. As shown in Fig. 15, the control device 300 includes a processor 302, a main memory 304, a storage 360, a memory card interface 312, a host network controller 306, a field network controller 308, a local bus controller 316, and a USB controller 370 that provides a USB (Universal Serial Bus) interface. These components are connected via a processor bus 318.

[0068] The processor 302 corresponds to an arithmetic processing unit that executes control calculations, and is configured with a CPU (Central Processing Unit) and / or a GPU (Graphics Processing Unit), etc. Specifically, the processor 302 reads out a program stored in the storage 360, expands it in the main memory 304, and executes it to realize control calculations for a control target.

[0069] The main memory 304 is configured with a volatile storage device such as a dynamic random access memory (DRAM) and / or a static random access memory (SRAM). The storage 360 ​​is configured with a non-volatile storage device such as a solid state drive (SSD) and / or a hard disk drive (HDD).

[0070] The storage 360 ​​stores a control program Pc, training data Ds, and a prediction model Mp. The storage 360 ​​corresponds to the storage means 140 in Fig. 1. The control program Pc includes a program for comprehensively controlling the control device 300 and realizing each function of the control device 300. In other words, the processor 302 that executes the control program Pc corresponds to the feedback control system (feedback control means 110 and subtractor 150), the feedforward control system (feedforward compensation means 120 and adder 160), and the learning means 130 in Fig. 1.

[0071] The memory card interface 312 accepts a memory card 314, which is an example of a removable storage medium. The memory card interface 312 is capable of reading and writing any data from and to the memory card 314.

[0072] The host network controller 306 exchanges data with any information processing device connected to the host network 32 (for example, a local area network) via the host network 32 .

[0073] The field network controller 308 exchanges data with any devices such as the servo motors 222_1 and 222_2 via the field network 20.

[0074] The local bus controller 316 exchanges data with any of the functional units 380 constituting the control device 300 via the local bus 122. The functional units 380 may include, for example, an analog I / O unit responsible for inputting and / or outputting analog signals, a digital I / O unit responsible for inputting and / or outputting digital signals, and a counter unit that receives pulses from an encoder or the like.

[0075] The USB controller 370 exchanges data with any information processing device via a USB connection. The USB controller 370 is connected to a support device 400, for example.

[0076] As described above, the control device and control method can improve the accuracy of feedforward control of a controlled object.

[0077] <Additional Notes> The present embodiment as described above includes the following technical idea.

[0078] [Configuration 1] A control device (100, 300) that determines a feedforward compensation value (rf) of a feedback manipulated variable (rb) to a controlled object (200, 200C) that is subjected to a disturbance (d) so that a controlled variable (q) of the controlled object (200, 200C) approaches a target value (qr), The feedback manipulated variable (rb) is determined by a feedback control means (110) based on an error (eq) between the target value (qr) and the controlled variable (q), The sum of the feedback manipulated variable (rb) and the feedforward compensation value (rf) is output as a manipulated variable (r) to the controlled object (200, 200C), The control device (100, 300) a feedforward compensation means (120) for determining the feedforward compensation value (rf) from the disturbance (d) using a prediction model (Mp); A learning means (130) for performing machine learning on the prediction model (Mp) using training data (Ds), The learning means (130) obtaining at least one first combination (Cm1) including the target value (qr), the manipulated variable (r) corresponding to the target value (qr), the disturbance (d), and the controlled variable (q) corresponding to both the disturbance (d) and the manipulated variable (r); A control device (100, 300) that adds, to the teacher data (Ds), at least one second combination (Cm2) including the disturbance (d) and the manipulated variable (r) when the absolute value of the error (eq) is smaller than a first reference value (α) among the at least one first combination (Cm1).

[0079] [Configuration 2] The learning means (130) Calculating an average value (rb0) of the manipulated variable (r) when the absolute value of the disturbance (d) is smaller than a second reference value (β) in the teacher data (Ds); a relationship between the disturbance (d) and a compensation value (rf) as a difference between the manipulated variable (r) and the average value (rb0), which is expressed by the prediction model (Mp), is approximated as a function having the compensation value (rf) as a response variable and the disturbance (d) as an explanatory variable; The control device (100, 300) according to configuration 1, wherein the machine learning is terminated when the ratio of the number of fourth combinations (Cm4) among the at least one second combination (Cm2) in which the absolute value of the disturbance (d) is greater than the second reference value (β) to the number of third combinations (Cm3) among the at least one first combination (Cm1) in which the absolute value of the disturbance (d) is greater than the second reference value (β) is greater than a third reference value (δ).

[0080] [Configuration 3] The learning means (130) Calculating an average value (rb0) of the manipulated variable (r) when the absolute value of the disturbance (d) is smaller than a second reference value (β) in the teacher data (Ds); approximating the relationship between the disturbance (p) and the manipulated variable (r) expressed by the prediction model (Mp) as a function having the manipulated variable (r) as a response variable and the disturbance (d) as an explanatory variable; terminate the machine learning when a ratio (Cr) of the number of fourth combinations (Cm4) in the at least one second combination (Cm2) in which the absolute value of the disturbance (d) is greater than the second reference value (β) to the number of third combinations (Cm3) in the at least one first combination (Cm1) in which the absolute value of the disturbance (d) is greater than the second reference value (β) is greater than a third reference value (δ); The control device (100, 300) according to configuration 1, wherein the feedforward compensation means (120) determines, as a feedforward compensation value, a value obtained by subtracting the average value (rb0) from the manipulated variable (r) predicted from the disturbance (d) by the predictive model (Mp).

[0081] [Configuration 4] The control device (100, 300) according to configuration 2 or 3, wherein the learning means (130) resumes the machine learning when the machine learning is completed and the ratio (Cr) is smaller than the third reference value (δ).

[0082] [Configuration 5] A control system (2) that outputs a sum of a feedback manipulated variable (rb) and a feedforward compensation value (rf) to a controlled object (200, 200C) as a manipulated variable (r) to the controlled object (200, 200C) so that a controlled variable (q) of the controlled object (200, 200C) subjected to a disturbance (d) approaches a target value (qr), a feedback control device (11) that determines the feedback manipulated variable (rb) based on an error (eq) between the target value (qr) and the controlled variable (q); a feedforward compensation device (12) that determines the feedforward compensation value (rf) from the disturbance (d) using a prediction model (Mp); a learning device (13) that performs machine learning on the prediction model (Mp) using teacher data (Ds), The learning device (13) obtaining at least one first combination (Cm1) including the target value (qr), the manipulated variable (r) corresponding to the target value (qr), the disturbance (d), and the controlled variable (q) corresponding to both the disturbance (d) and the manipulated variable (r); A control system (2) adds, to the teacher data (Ds), at least one second combination (Cm2) including the disturbance (d) and the manipulated variable (r) when the absolute value of the error (eq) is smaller than a first reference value (α) among the at least one first combination (Cm1).

[0083] [Configuration 6] A control method for determining a feedforward compensation value (rf) of a feedback manipulated variable (rb) to a controlled object (200, 200C) subjected to a disturbance (d) so that a controlled variable (q) of the controlled object (200, 200C) approaches a target value (qr), comprising: The feedback manipulated variable (rb) is determined by a feedback control means (110) based on an error (eq) between the target value (qr) and the controlled variable (q), The sum of the feedback manipulated variable (rb) and the feedforward compensation value (rf) is output as a manipulated variable (r) to the controlled object (200, 200C), The control method includes: a step (S121) of determining the feedforward compensation value (rf) from the disturbance (d) using a prediction model (Mp); and a step (S130) of performing machine learning on the prediction model (Mp) using training data (Ds), The step of performing machine learning (S130) includes: a step (S131) ​​of acquiring at least one first combination (Cm1) including the target value (qr), the manipulated variable (r) corresponding to the target value (qr), the disturbance (d), and the controlled variable (q) corresponding to both the disturbance (d) and the manipulated variable (r); A control method comprising steps (S132, S133) of adding, to the teacher data (Ds), at least one second combination (Cm2) including the disturbance (d) and the manipulated variable (r) when the absolute value of the error (eq) is smaller than a first reference value (α) among the at least one first combination (Cm1).

[0084] The embodiments disclosed herein are intended to be implemented in appropriate combinations within the scope of compatibility. The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is defined by the claims, not the above description, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0085] 2,3 Control system, 11 Feedback control device, 12 Feedforward compensation device, 13 Learning device, 20 Field network, 32 Upper network, 100, 100A, 300 Control device, 110 Feedback control means, 120 Feedforward compensation means, 122 Local bus, 130 Learning means, 140 Storage means, 150 Subtractor, 160 Adder, 200 Control target, 200C Field device, 220 Servo driver, 222 Servo motor, 302 Processor, 304 Main memory, 306 Upper network controller, 308 Field network controller, 312 Memory card interface, 314 Memory card, 316 Local bus controller, 318 Processor bus, 360 Storage, 370 Controller, 380 Functional unit, 400 Support device, 500 Display device, 600 Server device, 700 Gateway, 900 Internet, Ds Training data, Mp prediction model, Pc control program.

Claims

1. A control device that determines a feedforward compensation value of a feedback manipulated variable to a controlled object that is subjected to a disturbance so that a controlled variable of the controlled object approaches a target value, the feedback manipulated variable is determined by a feedback control means based on an error between the target value and the controlled variable; a sum of the feedback manipulated variable and the feedforward compensation value is output as a manipulated variable to the control target; The control device a feedforward compensation means for determining the feedforward compensation value from the disturbance using a predictive model; a learning means for performing machine learning on the prediction model using training data; The learning means obtaining at least one first combination including the target value, the manipulated variable corresponding to the target value, the disturbance, and the controlled variable corresponding to both the disturbance and the manipulated variable; a control device that adds, to the teacher data, at least one second combination, among the at least one first combination, including the disturbance and the manipulated variable when the absolute value of the error is smaller than a first reference value;

2. The learning means calculating an average value of the manipulated variable when the absolute value of the disturbance is smaller than a second reference value in the teacher data; approximating the relationship between the disturbance and the feedforward compensation value as a difference between the manipulated variable and the average value, which is expressed by the prediction model, as a function having the feedforward compensation value as a response variable and the disturbance as an explanatory variable; 2. The control device according to claim 1, wherein the machine learning is terminated when a ratio of the number of fourth combinations, among the at least one second combination, in which the absolute value of the disturbance is greater than the second reference value to the number of third combinations, among the at least one first combination, in which the absolute value of the disturbance is greater than the second reference value, is greater than a third reference value.

3. The learning means calculating an average value of the manipulated variable when the absolute value of the disturbance is smaller than a second reference value in the teacher data; approximating the relationship between the disturbance and the manipulated variable expressed by the prediction model as a function having the manipulated variable as a response variable and the disturbance as an explanatory variable; terminate the machine learning when a ratio of the number of fourth combinations, among the at least one second combination, in which the absolute value of the disturbance is greater than the second reference value to the number of third combinations, among the at least one first combination, in which the absolute value of the disturbance is greater than the second reference value is greater than a third reference value; 2. The control device according to claim 1, wherein the feedforward compensation means determines, as the feedforward compensation value, a value obtained by subtracting the average value from the manipulated variable predicted from the disturbance by the prediction model.

4. The control device according to claim 2 , wherein the learning means resumes the machine learning when the machine learning is completed and the ratio is smaller than the third reference value.

5. A control system that outputs a sum of a feedback manipulated variable and a feedforward compensation value to a controlled object as a manipulated variable so that a controlled variable of the controlled object that is subjected to a disturbance approaches a target value, a feedback control device that determines the feedback manipulated variable based on an error between the target value and the controlled variable; a feedforward compensation device that determines the feedforward compensation value from the disturbance using a predictive model; a learning device that performs machine learning on the prediction model using training data, The learning device obtaining at least one first combination including the target value, the manipulated variable corresponding to the target value, the disturbance, and the controlled variable corresponding to both the disturbance and the manipulated variable; A control system that adds, to the teacher data, at least one second combination, among the at least one first combination, including the disturbance and the manipulated variable when the absolute value of the error is smaller than a first reference value.

6. 1. A control method for determining a feedforward compensation value of a feedback manipulated variable of a controlled object subjected to a disturbance so that a controlled variable of the controlled object approaches a target value, the method comprising: the feedback manipulated variable is determined by a feedback control means based on an error between the target value and the controlled variable; a sum of the feedback manipulated variable and the feedforward compensation value is output as a manipulated variable to the control target; The control method includes: determining the feedforward compensation value from the disturbance using a predictive model; and performing machine learning on the prediction model using training data; The step of performing machine learning includes: obtaining at least one first combination including the target value, the manipulated variable corresponding to the target value, the disturbance, and the controlled variable corresponding to both the disturbance and the manipulated variable; and adding to the teacher data at least one second combination, among the at least one first combination, that includes the disturbance and the manipulated variable when the absolute value of the error is smaller than a first reference value.

Images