Robot system, braking determination method and braking determination program

Abstract

To provide a robot system, a braking determination method and a braking determination program which can determine an abnormality of an electromagnetic brake accurately.SOLUTION: A robot system 100 is provided with an electromagnetic brake 60 that switches between a braked state where a brake is turned on to brake rotation of a motor as a driving source for a robot 10 and a released state where the brake is turned off, a brake current detecting part that detects currents flowing into the electromagnetic brake 60, and a control part. The control part obtains an estimated brake operation signal, as output of a predicted value from a learnt model 80 with respect a current signal, by using the learnt model 80 generated by machine learning, where the current signal based on a detected result by the brake current detecting part is set as an explanatory variable in teacher data and a braking operation signal showing switching timing for the switching by the electromagnetic brake is set as an objective variable in the teacher data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a robot system, a braking judgment method, and a braking judgment program. [Background technology]

[0002] Conventionally, motor drive devices equipped with a brake device are known. For example, Patent Document 1 describes a motor drive device equipped with a friction type brake device. In this brake device, the armature is pulled away from the friction plate by an electromagnetic force generated by applying a brake coil voltage to the brake coil, and the brake of the motor is released. In addition, in the motor drive device described in Patent Document 1, it is determined whether the brake is normally released in the brake device. Specifically, in a state in which a brake coil voltage is applied, the motor current when the motor is operated at a predetermined load is observed, and it is determined whether the brake is normally released or not based on the magnitude relationship with a preset current at the time of normal brake release. In the motor drive device described in Patent Document 1, when the motor current is larger than the current at the time of normal brake release, it is determined that the brake is not normally released, and the brake coil voltage is changed to a larger value. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6227612 Summary of the Invention [Problem to be solved by the invention]

[0004] However, when determining whether an abnormality in an electromagnetic brake is detected by determining whether a detected value such as a current value is greater than a predetermined threshold value, as in the motor drive device of Patent Document 1, the accuracy of the determination decreases due to disturbances such as noise. For this reason, it is desirable to be able to accurately determine an abnormality in an electromagnetic brake.

[0005] This disclosure has been made to solve the problems described above, and one objective of this disclosure is to provide a robot system, a brake judgment method, and a brake judgment program that are capable of accurately determining abnormalities in an electromagnetic brake. [Means for solving the problem]

[0006] A robot system according to a first aspect of this disclosure includes an electromagnetic brake that switches between a braking state in which the brake is on for the rotation of a motor that is the driving source of the robot, and a released state in which the brake is off, a brake current detection unit that detects a current flowing through the electromagnetic brake, and a control unit that uses a trained model generated by machine learning to obtain an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal, using a current signal based on the detection result by the brake current detection unit as an explanatory variable of teacher data and a brake operation signal that indicates the switching timing of the switching operation that switches the electromagnetic brake between the braking state and the released state as an objective variable of the teacher data.

[0007] As described above, the robot system according to the first aspect of the disclosure includes a control unit that uses a trained model generated by machine learning to acquire an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal, by using a trained model generated by machine learning with a current signal based on the detection result by the brake current detection unit as an explanatory variable of the teacher data and a brake operation signal indicating a switching timing of a switching operation for switching between an engaged state and a released state of the electromagnetic brake as an objective variable of the teacher data. As a result, since an estimated brake operation signal is acquired as an output of a predicted value from the trained model generated by machine learning, it is possible to accurately distinguish between a change in the current signal due to the switching operation of the electromagnetic brake and a change in the current signal due to a disturbance such as noise. Therefore, it is possible to accurately predict the switching timing of the switching operation of the electromagnetic brake compared to a case where the switching operation of the brake is predicted by comparing the amount of change in the current signal or the differential value of the amount of change in the current signal with a preset threshold value. As a result, it is possible to determine whether the electromagnetic brake is operating normally based on the predicted switching timing, and therefore it is possible to accurately determine an abnormality in the electromagnetic brake. It should be noted that "determining an abnormality in the electromagnetic brake" as used here is described as a broad concept that not only refers to determining whether or not an abnormality such as deterioration of each part of the electromagnetic brake or a malfunction has occurred, but also includes determining the degree to which deterioration has progressed even if it has not yet reached the point where an abnormality has occurred, and predicting the lifespan at which an abnormality will occur in the future.

[0008] A brake judgment method according to a second aspect of this disclosure obtains a current signal based on the detection result of a current flowing through an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of a motor that is the driving source of the robot, and a released state in which the brake is applied, and obtains an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the learned model for the obtained current signal, by using the current signal as an explanatory variable of teacher data and a brake operation signal that indicates the switching timing of the switching operation that switches the electromagnetic brake between the braking state and the released state as an objective variable of the teacher data, and using a trained model generated by machine learning.

[0009] As described above, the brake determination method according to the second aspect of the disclosure uses a trained model generated by machine learning with the current signal as an explanatory variable of the teacher data, and a brake operation signal indicating the switching timing of the switching operation for switching the electromagnetic brake between the braking state and the released state as the objective variable of the teacher data, thereby obtaining an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the obtained current signal. As a result, since the estimated brake operation signal is obtained as an output of a predicted value from the trained model generated by machine learning, it is possible to accurately distinguish between a change in the current signal due to the switching operation of the electromagnetic brake and a change in the current signal due to a disturbance such as noise. Therefore, it is possible to accurately predict the switching timing of the switching operation of the electromagnetic brake compared to a case where the switching operation of the brake is predicted by comparing the amount of change in the current signal or the differential value of the amount of change in the current signal with a preset threshold value. As a result, it is possible to determine whether the electromagnetic brake is operating normally based on the predicted switching timing, and therefore it is possible to provide a brake determination method capable of accurately determining an abnormality in the electromagnetic brake.

[0010] A braking judgment program according to a third aspect of the present disclosure causes a computer to execute the braking judgment method according to the second aspect.

[0011] As described above, the brake judgment program according to the third aspect of the present disclosure causes a computer to execute the brake judgment method according to the second aspect, thereby making it possible to accurately judge an abnormality in the electromagnetic brake.

[0012] A robot system according to a fourth aspect of this disclosure includes an electromagnetic brake that switches between a braking state in which the brake is on against the rotation of a motor that is the driving source of the robot, and a released state in which the brake is off, a brake current detection unit that detects a current flowing through the electromagnetic brake, and a control unit that uses a trained model generated by machine learning to obtain an estimated brake operation signal that is an estimated brake operation signal as an output of a predicted value from the trained model for the current signal, with a current signal based on the detection result by the brake current detection unit as an explanatory variable of the teacher data, and a brake operation signal that is a brake drive voltage applied to the electromagnetic brake in a switching operation in which the electromagnetic brake is switched between a braking state and a released state as an objective variable of the teacher data.

[0013] As described above, the robot system according to a fourth aspect of the disclosure includes a control unit that uses a trained model generated by machine learning to acquire an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal, by using a trained model generated by machine learning with a current signal based on the detection result by the brake current detection unit as an explanatory variable of the teacher data and a brake operation signal, which is a brake drive voltage applied to the electromagnetic brake in a switching operation in which the electromagnetic brake is switched between an engaged state and a released state, as an objective variable of the teacher data. As a result, since an estimated brake operation signal is acquired as an output of a predicted value from the trained model generated by machine learning, it is possible to accurately distinguish between a change in the current signal due to the switching operation of the electromagnetic brake and a change in the current signal due to a disturbance such as noise. Therefore, it is possible to accurately predict the brake drive voltage in the switching operation of the electromagnetic brake compared to a case in which the amount of change in the current signal or a differential value of the amount of change in the current signal is compared with a preset threshold value to predict the brake drive voltage in the switching operation of the brake. As a result, it is possible to determine whether the electromagnetic brake is operating normally based on the predicted brake drive voltage, and therefore it is possible to accurately determine an abnormality in the electromagnetic brake.

[0014] A brake judgment method according to a fifth aspect of this disclosure obtains a current signal based on the detection result of a current flowing through an electromagnetic brake that switches between a braking state in which the brake is on against the rotation of a motor that is the driving source of the robot, and a released state in which the brake is off, and obtains an estimated brake operation signal that is an estimated brake operation signal as an output of a predicted value from the learned model for the obtained current signal, by using the current signal as an explanatory variable of teacher data and a brake operation signal that is a brake drive voltage applied to the electromagnetic brake in a switching operation in which the electromagnetic brake is switched between a braking state and a released state as a target variable of the teacher data, and using a trained model generated by machine learning.

[0015] As described above, the brake determination method according to the fifth aspect of the disclosure uses a trained model generated by machine learning, with the current signal as an explanatory variable of the teacher data, and the brake operation signal, which is a brake drive voltage applied to the electromagnetic brake in a switching operation in which the electromagnetic brake is switched between an engaged state and a released state, as a target variable of the teacher data, to obtain an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the acquired current signal. As a result, the estimated brake operation signal is obtained as an output of a predicted value from the trained model generated by machine learning, so that it is possible to accurately distinguish between a change in the current signal due to the switching operation of the electromagnetic brake and a change in the current signal due to a disturbance such as noise. Therefore, it is possible to accurately predict the brake drive voltage in the switching operation of the electromagnetic brake compared to a case in which the amount of change in the current signal or the differential value of the amount of change in the current signal is compared with a preset threshold value to predict the brake drive voltage in the switching operation of the brake. As a result, it is possible to determine whether the electromagnetic brake is operating normally based on the predicted brake drive voltage, and therefore it is possible to provide a brake determination method capable of accurately determining an abnormality in the electromagnetic brake.

[0016] A braking judgment program according to a sixth aspect of the present disclosure causes a computer to execute the braking judgment method according to the fifth aspect.

[0017] As described above, the brake judgment program according to the sixth aspect of the present disclosure enables accurate judgment of an abnormality in the electromagnetic brake by causing a computer to execute the brake judgment method according to the fifth aspect. Effect of the Invention

[0018] According to the present disclosure, an abnormality in an electromagnetic brake can be determined with high accuracy. [Brief description of the drawings]

[0019] [Figure 1] 1 is a schematic diagram showing an overall configuration of a robot system according to a first embodiment. [Diagram 2] FIG. 1 is a block diagram showing a configuration of a robot system according to a first embodiment. [Diagram 3] FIG. 2 is a cross-sectional view showing the configuration of a motor and an electromagnetic brake. [Figure 4] FIG. 1 is a flowchart illustrating a method for generating a trained model according to the first embodiment. [Diagram 5] FIG. 4 is a diagram showing an example of a gradually changing voltage applied to an electromagnetic brake. [Figure 6] FIG. 13 is a diagram showing an example of a current signal in generating a trained model. [Figure 7] 11 is a diagram showing an example of acquisition of a brake operation signal based on a detection signal of an operation sound. FIG. [Figure 8] FIG. 1 is a diagram for explaining generation of a trained model by a learning device. [Figure 9] FIG. 4 is a flowchart illustrating a brake determination method performed by the robot system according to the first embodiment. [Figure 10] FIG. 4 is a diagram showing an example of a current signal acquired in the brake determination method. [Figure 11] FIG. 13 is a diagram for explaining acquisition of an estimated brake operation signal using a trained model. [Figure 12] 10 is a diagram for explaining acquisition of a brake driving voltage based on an estimated brake operation signal. FIG. [Figure 13] FIG. 11 is a block diagram showing a configuration of a robot system according to a second embodiment. [Figure 14] FIG. 11 is a diagram for explaining acquisition of a brake operation signal in the second embodiment. [Figure 15] FIG. 11 is a block diagram showing a configuration of a robot system according to a third embodiment. [Figure 16] FIG. 11 is a flowchart illustrating a control process for acquiring a brake operation signal in the third embodiment. [Figure 17] FIG. 13 is a block diagram showing a configuration of a robot system according to a fourth embodiment. [Figure 18] FIG. 13 is a flowchart illustrating a control process for acquiring a brake operation signal in the fourth embodiment. [Figure 19] FIG. 13 is a block diagram showing a configuration of a robot system according to a fifth embodiment. [Figure 20] FIG. 13 is a diagram for explaining acquisition of a current signal in the fifth embodiment. [Figure 21] FIG. 13 is a block diagram showing a configuration of a robot system according to a sixth embodiment. [Figure 22] FIG. 23 is a diagram for explaining acquisition of a brake driving voltage as a response variable in the sixth embodiment. [Figure 23] FIG. 13 is a flowchart illustrating a brake determination method performed by the robot system according to the sixth embodiment. [Figure 24] FIG. 11 is a diagram showing a first example of a brake actuation signal according to the modifications of the first to sixth embodiments. [Diagram 25] FIG. 13 is a diagram showing a second example of a brake actuation signal according to the modifications of the first to sixth embodiments. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] [First embodiment] Hereinafter, a first embodiment of the present disclosure will be described with reference to the drawings.

[0021] The configuration of a robot system 100 according to a first embodiment will be described with reference to FIGS. 1 to 3.

[0022] (Robot system configuration) As shown in Fig. 1, a robot system 100 according to the first embodiment includes a robot 10, a learning device 30, and an inspection device 40. The robot 10 includes a robot controller 20. In the robot system 100, a trained model 80 is generated in advance by the learning device 30 through machine learning. Then, in the robot system 100, an abnormality in an electromagnetic brake 60 arranged on the robot 10 is determined by the inspection device 40 using the generated trained model 80. The learning device 30 is an example of a learning unit.

[0023] As shown in FIG. 2, the robot 10 includes a robot arm 11. The robot 10 is, for example, an industrial robot that performs handling tasks such as transportation, assembly, and machining by operating the robot arm 11 and an end effector attached to the tip of the robot arm 11. The robot arm 11 is, for example, a six-axis vertical articulated robot arm having a plurality of link members and joints that connect the plurality of link members to each other. That is, six joints are arranged in the robot arm 11. A motor 50 that serves as a drive source for the robot 10 is arranged in each of the six joints of the robot arm 11. The robot arm 11 operates as each of the six joints rotates due to the driving force of the motor 50. The motor 50 is a servo motor that is servo-controlled by the robot controller 20. The motor 50 is, for example, a brushless DC servo motor that operates by a current supplied from a motor driver including an inverter circuit. In addition, an electromagnetic brake 60 is arranged in the motor 50. The electromagnetic brake 60 switches between a braking state in which the brake is applied to the rotation of the motor 50 and a released state in which the brake is turned off.

[0024] As shown in FIG. 3, the motor 50 is provided with an encoder 51 that detects the rotation of the motor 50. The encoder 51 outputs a signal related to the detected rotation of the motor 50 to the robot controller 20. For example, the encoder 51 outputs a signal indicating the position of the rotation angle of the motor 50. The encoder 51 is provided on a shaft 55 that is a rotation axis of the motor 50. A disk-shaped bracket 52 is provided on the motor 50 as a part of the housing. In the motor 50, a rotor 54 on which a plurality of permanent magnets are arranged is rotated by a motor driver passing a current through three-phase coils of U-phase, V-phase and W-phase arranged on a stator 53. The rotor 54 rotates integrally with the shaft 55. In the motor 50, the rotation of the shaft 55 is controlled by controlling the current passed through the coils of the stator 53.

[0025] The electromagnetic brake 60 is disposed on the bracket 52 inside the motor 50. The electromagnetic brake 60 stops or brakes the rotation of the shaft 55 of the motor 50. The electromagnetic brake 60 includes a yoke 61, a coil 62, an elastic member 63, an armature 64, and a friction member 65. In the electromagnetic brake 60, when no voltage is applied to the coil 62 disposed on the yoke 61, the armature 64 is pressed against the friction surface of the friction member 65 by the elastic force of the elastic member 63. The elastic member 63 is, for example, a coil spring. The friction member 65 rotates integrally with the shaft 55, and the armature 64 does not rotate, so that the rotation of the friction member 65 is braked by the friction with the armature 64, thereby braking the rotation of the shaft 55. In other words, the electromagnetic brake 60 is of a non-excitation operation type in which the brake is in an on-braking state when no voltage is applied to the coil 62. In the electromagnetic brake 60, when a predetermined brake drive voltage is applied to the coil 62, a magnetic force is generated, causing the armature 64 to move in a direction away from the friction member 65 against the elastic force of the elastic member 63. That is, the electromagnetic brake 60 is switched from the braked state to the released state when a brake drive voltage of a predetermined magnitude is applied. In the electromagnetic brake 60, the magnitude of the brake drive voltage changes according to the size of the gap between the yoke 61 and the armature 64 when no voltage is applied to the coil 62. When the size of the gap increases due to wear of the friction member 65 caused by repeated use of the electromagnetic brake 60, the brake drive voltage in the switching operation for switching between the released state and the braked state increases.

[0026] For example, when the electromagnetic brake 60 is a non-excitation type brake, if the voltage applied to the electromagnetic brake 60 is gradually increased from 0V to 24V, the brake state is switched from the braked state to the released state before the voltage increases to 24V. The brake drive voltage in this switching operation from the braked state to the released state is the attraction voltage. Also, when the electromagnetic brake 60 is a non-excitation type brake, if the voltage applied is gradually decreased from 24V to 0V, the brake state is switched from the released state to the braked state before the voltage decreases to 0V. The brake drive voltage in this switching operation from the released state to the braked state is the release voltage. By detecting the values ​​of the attraction voltage and the release voltage, or the change over time and the trend of the change in the attraction voltage and the release voltage, it is possible to diagnose deterioration of the electromagnetic brake 60, such as wear of the friction member 65 in the electromagnetic brake 60, and to predict the lifespan. For example, when the degree of wear of the friction member 65 increases, the attraction voltage increases.

[0027] As shown in FIG. 2, the robot controller 20 includes a control unit 21, a brake power supply unit 22, a brake current detection unit 23, and a motor current detection unit 24.

[0028] The control unit 21 is a computer including an arithmetic device such as a central processing unit (CPU), memories such as a random access memory (RAM) and a read only memory (ROM), and a storage device such as a hard disk. The control unit 21 executes control processing by the arithmetic device based on programs and parameters stored in the storage device. The control unit 21 also includes a communication module. The control unit 21 communicates with the learning device 30 and the inspection device 40 via the communication module. The control unit 21 controls the operation of each part of the robot 10. The control unit 21 has, for example, a main CPU that performs overall control of the operation of the robot 10 and a servo CPU that controls the current supplied to the motor 50. The servo CPU of the control unit 21 outputs an operation signal that operates a motor driver that generates the current supplied to the motor 50. The control unit 21 may perform control processing using one CPU.

[0029] The brake power supply unit 22 applies a predetermined voltage to the coil 62 of the electromagnetic brake 60. The brake power supply unit 22 changes the magnitude of the voltage applied to the coil 62 under the control of the control unit 21. The brake current detection unit 23 detects the current flowing through the coil 62 of the electromagnetic brake 60. The brake current detection unit 23 includes, for example, a shunt resistor. The motor current detection unit 24 detects the motor drive current supplied to the coil of the stator 53 of the motor 50. The motor current detection unit 24 includes, for example, a shunt resistor. The detection results of the brake current detection unit 23 and the motor current detection unit 24 are output to the control unit 21. The control unit 21 controls the rotation angle position of the motor 50 by feedback control using, for example, the difference between a detection value of the rotation angle position of the motor 50 from the encoder 51 and a position command value for the motor 50.

[0030] Conventionally, a change in the current flowing through the coil 62 of the electromagnetic brake 60 due to a change in inductance of the coil 62 during switching of the electromagnetic brake 60 is detected by an electric circuit or the like, and the detected amount of change in the current flowing through the coil 62, a derivative of the amount of change in the current, or a second derivative of the amount of change in the current, is compared with a preset threshold value to estimate the timing of the switching operation. In the robot system 100 according to the first embodiment, an abnormality in the electromagnetic brake 60 is determined using the trained model 80 generated by the learning device 30 as described above.

[0031] The learning device 30 is a computer including one or more arithmetic units such as a CPU, and a storage unit including a storage device such as a ROM, a RAM, and a hard disk. The learning device 30 also includes a communication module. The learning device 30 communicates with the robot controller 20 and the inspection device 40 via the communication module. The learning device 30 generates a trained model 80 for determining an abnormality in the electromagnetic brake 60 by machine learning as a result of a predetermined program stored in the storage unit being executed. A method for generating the trained model 80 will be described in detail later.

[0032] The robot system 100 also includes an operation sound detection unit 70. In the first embodiment, the operation sound detection unit 70 detects the switching timing of a switching operation in which the electromagnetic brake 60 is switched between an engaged state and an released state. Specifically, the operation sound detection unit 70 detects operation sounds in the switching operation in order to detect the switching timing of the switching operation of the electromagnetic brake 60. The operation sound detection unit 70 is an example of a operation detection unit.

[0033] 3, the operation sound detection unit 70 includes a microphone attached to the electromagnetic brake 60 disposed on the motor 50. For example, the operation sound detection unit 70 is disposed on the bracket 52. The operation sound detection unit 70 outputs a detected audio signal to the control unit 21 of the robot controller 20. The operation sound detection unit 70 detects operation sounds generated by the movement of the armature 64 during the switching operation of the electromagnetic brake 60.

[0034] As shown in FIG. 2, the inspection device 40 is a computer including a control unit 41 having one or more processors such as a CPU, and a storage unit 42 including a storage device such as a ROM, a RAM, and a hard disk. The inspection device 40 is, for example, a personal computer or a tablet PC used by an operator. The inspection device 40 includes a display unit 43 such as a liquid crystal display device, and an operation unit 44 including input devices such as a keyboard and a mouse. The inspection device 40 executes a control process of a brake judgment method for judging an abnormality of the electromagnetic brake 60. The storage unit 42 stores a trained model 80 generated by the learning device 30, a program 42a, and various parameters for executing the brake judgment method. The program 42a is an example of a brake judgment program.

[0035] The brake judgment method according to the first embodiment is implemented by having the inspection device 40, which is a computer, execute the program 42a. The inspection device 40 judges an abnormality in the electromagnetic brake 60 by having the control unit 41 execute the program 42a stored in the storage unit 42. The program 42a may be read from a computer-readable recording medium such as an optical disk, a magnetic disk, or a non-volatile semiconductor memory, or may be provided from an external server or the like via a transmission path such as a network such as the Internet or a LAN (Local Area Network).

[0036] In the robot system 100 according to the first embodiment, for example, the brake determination method by the inspection device 40 is executed when the robot 10 is installed or during periodic maintenance of the installed robot 10. Then, in the robot system 100, before the brake determination method by the inspection device 40 is executed, the learning device 30 generates the trained model 80 in advance. The trained model 80 is generated, for example, before the robot 10 is shipped. Then, the generated trained model 80 is transmitted from the learning device 30 to the inspection device 40 via a transmission path such as serial communication, a network such as the Internet, or a LAN.

[0037] (How to generate a trained model) 4 to 8, a method for generating a trained model 80 by the learning device 30 will be described. The control processes from step S1 to step S3 in the method for generating a trained model 80 are executed by the learning device 30.

[0038] First, in step S1, a current signal is acquired based on the detection result of the current flowing through the electromagnetic brake 60 by the brake current detection unit 23. That is, based on the detection signal detected by the brake current detection unit 23, a current signal is acquired as an explanatory variable of teacher data in machine learning.

[0039] As shown in FIG. 5, in order to obtain a current signal, the brake power supply unit 22 of the robot controller 20 applies a voltage whose magnitude gradually changes to the coil 62 of the electromagnetic brake 60. For example, the brake power supply unit 22 applies a voltage to the electromagnetic brake 60 in a braking state while gradually increasing the value at a constant rate from 0 V to 24 V for 20 seconds. In the example shown in FIG. 5, the change in the voltage applied by the brake power supply unit 22 is started at timing t1, and the change in the voltage applied by the brake power supply unit 22 is ended at timing t2. At timing t1, the voltage applied to the electromagnetic brake 60 is 0 V, and at timing t2, the voltage applied to the electromagnetic brake 60 is 24 V. The electromagnetic brake 60 is switched from the braking state to the release state before the applied voltage reaches 24 V. For example, the brake power supply unit 22 changes the voltage applied at a constant rate for each predetermined control period. As an example, the brake power supply unit 22 gradually increases the voltage applied from 0 V to 24 V over 20 seconds by increasing the voltage by 12 mV every 10 milliseconds as a predetermined control period.

[0040] As shown in FIG. 6, the brake current detection unit 23 of the robot controller 20 detects the current flowing through the electromagnetic brake 60 by applying a voltage whose magnitude gradually changes from the brake power supply unit 22, and outputs the detection result to the control unit 21. That is, the detection result of the current flowing through the coil 62 of the electromagnetic brake 60 detected from timing t1 to timing t2 is acquired by the control unit 21. In the electromagnetic brake 60, at the switching timing when the switching operation is performed, the inductance of the coil 62 changes depending on the positional relationship between the armature 64 and the yoke 61. Therefore, as shown at timing t3 in FIG. 6, the current detected when the switching operation is performed changes. For example, in the example shown in FIG. 6, a current whose value increases linearly with time is detected from timing t1 to timing t3, but at timing t3, the value changes so as to oscillate slightly. The control unit 21 of the robot controller 20 acquires the detection result of the current from timing t1 to timing t2, and outputs the acquired detection result to the learning device 30. The learning device 30 obtains the detection result of the current detected by the voltage whose magnitude gradually changes as a current signal, which is time-series data representing the waveform of the current.

[0041] Next, in step S2, a brake operation signal is obtained that indicates the switching timing of the switching operation of the electromagnetic brake 60. That is, the brake operation signal based on the detection signal detected by the operation sound detection unit 70 is obtained as the objective variable of the teacher data in the machine learning.

[0042] As shown in FIG. 7, the operation sound detection unit 70 detects the operation sound in the switching operation of the electromagnetic brake 60 from timing t1 when the brake power supply unit 22 starts applying a voltage whose magnitude gradually changes to timing t2 when the change in the voltage applied by the brake power supply unit 22 ends. Then, a detection signal indicating the detection result detected by the operation sound detection unit 70 is output to the control unit 21 of the robot controller 20. For example, the control unit 21 acquires a detection signal indicating the detection result of the operation sound, rectifies the acquired detection signal, and smoothes it using an integral filter or the like. The detection signal is smoothed in order to suppress malfunction due to noise. Then, the control unit 21 compares the rectified and smoothed detection signal with a predetermined threshold value using a comparator, thereby acquiring a signal indicating the timing at which the operation sound in the switching operation was detected as a brake operation signal indicating the switching timing in the switching operation of the electromagnetic brake 60. The control unit 21 acquires a brake operation signal that indicates the time from timing t1, which is the time when the change in voltage applied by the brake power supply unit 22 in the switching operation starts, to timing t4, which is the switching timing when the electromagnetic brake 60 switches between the braking state and the switching state. Specifically, the control unit 21 acquires a brake operation signal that indicates the time from timing t1 to timing t4 as a ratio to the time from timing t1 to timing t2, which is the time when the change in voltage applied by the brake power supply unit 22 ends. For example, if the time from timing t1 to timing t2 is 20 seconds and the time from timing t1 to timing t4 is 17 seconds, the control unit 21 acquires a signal that indicates a numerical value of "85%" as the brake operation signal. Then, the learning device 30 acquires the brake operation signal output from the control unit 21. The timings t1, t2, and t4 are acquired based on the control period of the control unit 21, for example.

[0043] Next, in step S3, a trained model 80 is generated by machine learning using the current signal acquired in step S1 as an explanatory variable and the brake operation signal acquired in step S2 as a response variable.

[0044] 8, the learning device 30 has, as its functional configuration, a teacher data acquisition unit 31, an evaluation unit 32, and an update unit 33. The learning device 30 acquires a current signal as an explanatory variable of teacher data in machine learning, and acquires a brake operation signal as a target variable of teacher data in machine learning, and generates a trained model 80 by repeating machine learning using the explanatory variables and the target variable.

[0045] Specifically, the teacher data acquisition unit 31 of the learning device 30 acquires the current signal, which is an explanatory variable, and the brake operation signal, which is an objective variable, from the control unit 21 of the robot controller 20 as teacher data in machine learning. The teacher data acquisition unit 31 acquires the current signal, which is time-series data, as the explanatory variable of the teacher data, and acquires the brake operation signal indicating a predetermined numerical value as the objective variable of the teacher data. Then, the learning device 30 causes the machine learning model to output learning output data by providing the current signal, which is the explanatory variable, to the machine learning model. The learning output data is data similar to the brake operation signal, which is the objective variable. The evaluation unit 32 acquires the learning output data output from the machine learning model. Then, the evaluation unit 32 evaluates the accuracy of prediction of the machine learning model by comparing the acquired learning output data with the brake operation signal, which is the objective variable. That is, the evaluation unit 32 evaluates the learning output data using the brake operation signal, which is the objective variable, as correct answer data. The evaluation unit 32 evaluates the accuracy of prediction of the machine learning model by using, for example, a loss function. The loss function is a function that represents the degree of difference between the learning output data and the correct answer data that is the learning objective variable. The closer the calculated value of the loss function is to 0, the higher the accuracy of the machine learning model is. The evaluation unit 32 outputs an evaluation result based on the calculated value of the loss function to the update unit 33. The update unit 33 updates the machine learning model by changing the values ​​of various parameters of the machine learning model according to the evaluation result from the evaluation unit 32. The learning device 30 repeatedly inputs a current signal that is learning input data to the machine learning model, outputs learning output data to the evaluation unit 32, evaluates the accuracy of the prediction based on the brake operation signal by the evaluation unit 32, and updates the machine learning model by the update unit 33 until the accuracy of the prediction by the machine learning model becomes a preset value. The learning device 30 stores the machine learning model whose prediction accuracy is greater than the preset value in the storage unit as a trained model 80 in which learning by machine learning has been completed.

[0046] The machine learning by the learning device 30 corresponds to a supervised regression learning using the brake operation signal as the correct answer data. The machine learning by the learning device 30 may be performed by, for example, deep learning. In deep learning, a process of automatically selecting appropriate features is performed, so that the trained model 80 is constructed by the automatically selected features even if the operator does not select the features. By the above-mentioned machine learning, the learning device 30 generates a trained model 80 in which the relationship between the current signal and the brake operation signal has been learned. In the machine learning by the learning device 30, for example, the partial least squares method or the principal component analysis method is used.

[0047] In addition, the learning device 30 may perform machine learning using a plurality of robots 10. The learning device 30 acquires current signals in the electromagnetic brakes 60 of the motors 50 disposed at the joints of each of the plurality of robots 10 as explanatory variables, and acquires brake operation signals corresponding to the acquired current signals as objective variables. For example, the wear levels of the electromagnetic brakes 60 are different in each of the plurality of robots 10. That is, the learning device 30 generates a trained model 80 by machine learning using a plurality of types of teacher data indicating different wear levels of the electromagnetic brakes 60.

[0048] (Control process of brake determination method according to the first embodiment) Next, the control process of the braking judgment method using the trained model 80 by the robot system 100 in the first embodiment will be described based on a flowchart with reference to Fig. 9 to Fig. 12. The control processes from step S11 to step S14 in the braking judgment method are performed by the control unit 41 of the inspection device 40 executing the program 42a. For example, an instruction to start the control process of the braking judgment method is received by an input operation by an operator to the operation unit 44 of the inspection device 40. The inspection device 40 outputs a signal to the control unit 21 of the robot controller 20 indicating that the control process of the braking judgment method has been started.

[0049] First, in step S11, a current signal is acquired in the same manner as in the method of generating the trained model 80. That is, a voltage whose magnitude gradually changes is applied by the brake power supply unit 22, and a current signal is acquired based on the detection result of the current flowing through the electromagnetic brake 60.

[0050] As shown in FIG. 10, the brake power supply unit 22 of the robot controller 20 applies a voltage whose magnitude gradually changes to the coil 62 of the electromagnetic brake 60 in order to acquire a current signal, similar to step S1 in the method for generating the trained model 80. The brake power supply unit 22 applies a voltage whose value gradually changes from 0 V to 24 V for 20 seconds, similar to step S1, from timing t11 to timing t12. Then, the brake current detection unit 23 of the robot controller 20 detects the current flowing through the electromagnetic brake 60 by applying the voltage whose magnitude gradually changes from timing t11 to timing t12 by the brake power supply unit 22, and outputs the detection result to the control unit 21. As shown at timing t13 in FIG. 10, similar to the generation of the trained model 80, when a switching operation is performed, the inductance of the coil 62 of the electromagnetic brake 60 changes, and thus a change occurs in the current flowing through the coil 62. In addition, as shown at timing t15 in FIG. 10, the current signal acquired by the inspection device 40 may contain noise caused by spot welding machines or arc welding machines arranged around the location where the robot 10 is installed, or noise caused by the magnetic field leaking out to the coil 62 from the relatively large rotating magnetic field generated by the stator 53 of the motor 50.

[0051] The inspection device 40 acquires the detection result of the current detected by the voltage whose magnitude gradually changes like the explanatory variable of the teacher data in machine learning as a current signal as time series data representing the waveform of the current. That is, the current signal acquired in the control process of the brake judgment method is detected by a voltage having a similar aspect to the voltage applied to the electromagnetic brake 60 in the generation of the trained model 80. In other words, the inspection device 40 acquires the detection result of the current flowing through the electromagnetic brake 60 as a current signal by applying to the electromagnetic brake 60 a voltage having the same voltage value and change amount as in the generation of the trained model 80. Therefore, the length from timing t1 to timing t2 in FIG. 5 is equal to the length from timing t11 to timing t12 in FIG. 10. Moreover, the voltage value at timing t1 and the voltage value at timing t11 are equal to each other, and the voltage value at timing t2 and the voltage value at timing t12 are also equal to each other. Moreover, the rate of change of the voltage from timing t1 to timing t2 is equal to the rate of change of the voltage from timing t11 to timing t12.

[0052] Next, in step S12, by using the trained model 80 generated by machine learning in the learning device 30, an estimated brake operation signal is obtained as an output of a predicted value from the trained model 80 for the current signal acquired in step S11.

[0053] As shown in FIG. 11, the control unit 41 of the inspection device 40 inputs the acquired current signal to the trained model 80, thereby acquiring an estimated brake operation signal as an output of an estimation result that is a predicted value by the trained model 80. The estimated brake operation signal is a signal that represents the time from timing t11, which is the time when the change in voltage applied by the brake power supply unit 22 in the switching operation starts, to timing t14, which is the time of the switching timing in the switching operation, similar to the brake operation signal used as the objective variable in the machine learning in the learning device 30. Timing t14 is the switching timing estimated in the switching operation by the estimated brake operation signal. FIG. 11 shows an example in which a value of "85%" is output as an estimated value for the estimated brake operation signal. In the estimation of the brake operation signal using the trained model 80 generated by machine learning, even when a change due to noise is included in the current signal as shown at timing t15 in FIG. 10, the brake operation signal is estimated in a state in which the change in current due to noise and the change in current due to the switching operation are distinguished from each other.

[0054] Next, in step S13, an estimated brake drive voltage is obtained. Specifically, the brake drive voltage applied to the electromagnetic brake 60 at the switching timing of the switching operation of the electromagnetic brake 60 is obtained based on the estimated brake operation signal obtained in step S12.

[0055] As shown in FIG. 12, the control unit 41 of the inspection device 40 acquires, as time-series data, from the control unit 21 of the robot controller 20, the voltage applied to the electromagnetic brake 60 by the brake power supply unit 22 during the period from timing t11 to timing t12 when the current signal is acquired. The voltage applied by the brake power supply unit 22 is acquired, for example, based on a control amount for operating the brake power supply unit 22 of the robot controller 20. Then, the inspection device 40 acquires the value of the brake drive voltage, which is the voltage applied to the electromagnetic brake 60 at the switching timing of the switching operation, based on the voltage applied to acquire the current signal and the estimated brake operation signal acquired as an output from the trained model 80. For example, when the brake operation signal is 85%, the control unit 41 of the inspection device 40 acquires, as the brake drive voltage, the voltage value at the time point corresponding to 85% of the period from timing t11 to timing t12.

[0056] Next, in step S14, an abnormality in the electromagnetic brake 60 is determined. The inspection device 40 stores a threshold value for determination in advance in the storage unit 42. The control unit 41 of the inspection device 40 then determines an abnormality in the electromagnetic brake 60 by comparing the brake drive voltage acquired in step S13 with the threshold value for determination stored in the storage unit 42. For example, when the control unit 41 determines that the brake drive voltage for switching the electromagnetic brake 60 from the braking state to the released state is greater than the threshold value for determination, it determines that the wear of the friction member 65 of the electromagnetic brake 60 is large and an abnormality has occurred in the electromagnetic brake 60. For example, when an abnormality in the electromagnetic brake 60 is determined, the control unit 41 displays information indicating that an abnormality has been determined on the display unit 43.

[0057] In the robot system 100, during the period in which the inspection device 40 is performing the control process of the brake determination method from step S11 to step S14, the control unit 21 of the robot controller 20 executes servo control of the motor 50. That is, even when the electromagnetic brake 60 is switched from the braking state to the released state in the determination of the electromagnetic brake 60, the drive of the motor 50 is controlled so that the robot 10 does not move and maintains the current posture.

[0058] [Effects of the first embodiment] In the first embodiment, the following effects can be obtained.

[0059] As described above, the robot system 100 includes a control unit 41 that uses a trained model 80 generated by machine learning to acquire an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model 80 for the current signal, with the current signal based on the detection result by the brake current detection unit 23 as an explanatory variable and a brake operation signal indicating the switching timing of the switching operation for switching the electromagnetic brake 60 between the braking state and the released state as an objective variable. In this way, the estimated brake operation signal is acquired as an output of a predicted value from the trained model 80 generated by machine learning, so that it is possible to accurately distinguish between a change in the current signal due to the switching operation of the electromagnetic brake 60 and a change in the current signal due to a disturbance such as noise. Therefore, it is possible to accurately predict the switching timing of the switching operation of the electromagnetic brake 60 compared to a case where the switching operation of the brake is predicted by comparing the amount of change in the current signal or the differential value of the amount of change in the current signal with a preset threshold value. As a result, it is possible to determine whether the electromagnetic brake 60 is operating normally based on the predicted switching timing, so that it is possible to accurately determine an abnormality in the electromagnetic brake 60.

[0060] The robot system 100 includes a brake power supply unit 22 that applies a voltage whose magnitude gradually changes to the electromagnetic brake 60. The control unit 41 acquires a current signal based on the detection result of the current flowing through the electromagnetic brake 60 as a result of the brake power supply unit 22 applying a voltage whose magnitude gradually changes. The control unit 41 also acquires an estimated brake operation signal as an output of a predicted value from the trained model 80 for the acquired current signal. As a result, a voltage whose magnitude gradually changes is applied, and thus a current signal whose magnitude gradually changes can be acquired. Therefore, the switching timing of the switching operation for switching the electromagnetic brake 60 between the braking state and the released state can be acquired more accurately than when a voltage whose value changes discretely is applied. Therefore, an abnormality in the electromagnetic brake 60 can be determined more accurately.

[0061] The control unit 41 acquires the brake drive voltage applied to the electromagnetic brake 60 at the switching timing of the switching operation based on the acquired estimated brake operation signal. Here, in the electromagnetic brake 60, the size of the gap between the armature 64, which moves with the switching of the brake between on and off, and the yoke 61 in which the coil 62 is arranged, changes due to wear of the electromagnetic brake 60. The brake drive voltage at the switching timing changes according to the size of the gap caused by wear of the electromagnetic brake 60. Therefore, by acquiring the brake drive voltage applied to the electromagnetic brake 60 at the switching timing, it is possible to grasp the degree of wear of the electromagnetic brake 60. Therefore, by acquiring the brake drive voltage, it is possible to accurately determine an abnormality in the electromagnetic brake 60.

[0062] The robot system 100 includes an operation sound detection unit 70 as an operation detection unit that detects the switching timing of the switching operation, and a learning device 30 as a learning unit that acquires a current signal as an explanatory variable and acquires a brake operation signal indicating the state of the switching operation detected by the operation detection unit as an objective variable, thereby generating a trained model 80 by machine learning using the explanatory variables and the objective variable. As a result, the operation sound detection unit 70 can detect the switching timing of the switching operation, so that the brake operation signal indicating the switching timing of the switching operation can be easily acquired. Therefore, the learning device 30 can easily generate the trained model 80.

[0063] The robot system 100 includes an operation sound detection unit 70 as an operation detection unit that detects an operation sound in a switching operation to detect the switching timing of the switching operation. The learning device 30 as a learning unit acquires a brake operation signal as a target variable based on a detection signal detected by the operation sound detection unit 70. Here, in the electromagnetic brake 60, an operation sound is generated accompanying the operation of the armature 64 in the switching operation. Therefore, by detecting the operation sound by the operation sound detection unit 70, a brake operation signal indicating the switching timing of the switching operation can be easily and accurately acquired. As a result, the accuracy of prediction by the generated trained model 80 can be easily improved. In addition, since the switching timing in the switching operation of the electromagnetic brake 60 can be detected by detecting the operation sound, it is possible to detect whether the braking state and the release state of the electromagnetic brake 60 are switched without operating the motor 50. Therefore, it is not necessary to operate the robot 10 in order to detect the switching timing, and therefore the brake operation signal can be easily acquired without considering physical interference in the operating range of the robot 10.

[0064] The control unit 41 acquires, as a predicted value output from the trained model 80, an estimated brake operation signal representing the time from the point when the change in voltage applied by the brake power supply unit 22 starts in the switching operation to the switching timing when the electromagnetic brake 60 is switched between the engaged state and the released state. As a result, since the trained model 80 is generated by machine learning so as to output an estimated brake operation signal representing time as a predicted value, the trained model 80 can be generated so as to output the estimated brake operation signal as data indicating one value, rather than time-series data which is continuous data. Therefore, data processing in learning the trained model 80 is easier than time-series data, and the processing burden on the learning device 30 as a learning unit in learning the trained model 80 can be reduced.

[0065] [Second embodiment] Next, the configuration of a robot system 200 according to a second embodiment will be described with reference to Fig. 13 and Fig. 14. In the second embodiment, the robot system 200 includes a vibration detection unit 270 that detects vibrations during switching operations, unlike the robot system 100 of the first embodiment that includes an operation sound detection unit 70 that detects operation sounds during switching operations. Note that the same components as those in the first embodiment are denoted by the same reference numerals and will not be described.

[0066] As shown in Fig. 13, a robot system 200 according to the second embodiment includes a robot 210 including a robot controller 220, and a vibration detection unit 270. The robot controller 220 has a control unit 221. The control unit 221 controls the operation of each unit of the robot 210, similar to the control unit 21 of the first embodiment. In the second embodiment, the control unit 221 obtains a brake operation signal to be output to the learning device 30 based on the detection result of the vibration detection unit 270. The vibration detection unit 270 is an example of a operation detection unit.

[0067] The vibration detection unit 270 detects vibrations during switching operation to detect switching timing of the switching operation of the electromagnetic brake 60. The vibration detection unit 270 includes, for example, an AE sensor (Acoustic Emission sensor). The vibration detection unit 270 detects elastic waves generated by movement of the armature 64 during switching operation of the electromagnetic brake 60. The vibration detection unit 270 is disposed on the bracket 52 of the motor 50, for example, similar to the operation sound detection unit 70 of the first embodiment. The vibration detection unit 270 outputs a detection signal indicating the detection result of the detected vibration to the control unit 221 of the robot controller 220.

[0068] 14, the control unit 221, like the control unit 21 of the first embodiment, rectifies and smooths the acquired detection signal, and compares the smoothed detection signal using a comparator to acquire a brake operation signal indicating the switching timing in the switching operation of the electromagnetic brake 60. The acquired brake operation signal is, for example, a signal in which the switching timing is indicated by a numerical value representing a ratio, like the first embodiment. The brake operation signal acquired by the control unit 221 is output to the learning device 30, like the first embodiment.

[0069] As shown in FIG. 13, the learning device 30 acquires the brake operation signal acquired by the control unit 221 as a target variable, similar to the first embodiment. Then, the learning device 30 generates a trained model 280 by machine learning using the current signal acquired by the control unit 221 as an explanatory variable and the brake operation signal as a target variable, similar to the first embodiment. The method of generating the trained model 280 by the learning device 30 is similar to the method of generating the trained model 80 in the first embodiment. Then, the generated trained model 280 is stored in the storage unit 42 of the inspection device 40. The control unit 41 of the inspection device 40 acquires an estimated brake operation signal using the trained model 280 by executing a control process of a brake judgment method similar to the first embodiment. The other configurations of the second embodiment are similar to those of the first embodiment.

[0070] [Effects of the second embodiment] As described above, the robot system 200 according to the second embodiment includes the vibration detection unit 270 as an operation detection unit that detects vibrations in a switching operation in order to detect the switching timing of the switching operation. The learning device 30 as a learning unit acquires a brake operation signal as a target variable based on a detection signal detected by the vibration detection unit 270. Here, in the electromagnetic brake 60, vibrations occur with the operation of the armature 64 in the switching operation. Therefore, by detecting vibrations accompanying the switching operation by the vibration detection unit 270, a brake operation signal indicating the switching timing of the switching operation can be easily and accurately acquired. As a result, the accuracy of prediction by the generated trained model 280 can be easily improved. In addition, by detecting vibrations accompanying the switching operation, the switching timing in the switching operation of the electromagnetic brake 60 can be detected, so that it is possible to detect whether the braking state and the release state of the electromagnetic brake 60 are switched without operating the motor 50. Therefore, it is not necessary to operate the robot 10 in order to detect the switching timing, so that the brake operation signal can be easily acquired without considering physical interference in the operating range of the robot 10. Note that other effects of the second embodiment are similar to those of the first embodiment.

[0071] [Third embodiment] Next, the configuration of a robot system 300 according to a third embodiment will be described with reference to Fig. 15. In the third embodiment, the robot system 300 detects the motor drive current in the switching operation, unlike the robot system 100 of the first embodiment that includes an operation sound detection unit 70 that detects the operation sound of the switching operation. Note that the same components as those in the first embodiment are denoted by the same reference numerals and will not be described.

[0072] As shown in FIG. 15, a robot system 300 according to the third embodiment includes a robot 310 including a robot controller 320. The robot controller 320 has a control unit 321. The control unit 321 controls the operation of each unit of the robot 310, similar to the control unit 21 of the first embodiment. In the third embodiment, the control unit 321 acquires a brake operation signal to be output to the learning device 30 based on the motor drive current, which is the detection result of the motor current detection unit 24. That is, in the third embodiment, the motor current detection unit 24 is an example of an operation detection unit. As in the first embodiment, the motor current detection unit 24 detects the motor drive current supplied to the coil of the stator 53 of the motor 50, and outputs the detection result of the motor drive current to the control unit 321. In the third embodiment, the motor current detection unit 24 detects the motor drive current to detect the switching timing of the switching operation of the electromagnetic brake 60.

[0073] Next, the control process of acquiring the brake actuation signal by the robot controller 320 will be described with reference to Fig. 16. The control processes from step S331 to step S335 are executed by the control unit 321 of the robot controller 320. Fig. 16 shows an example of acquiring the brake actuation signal in a switching operation in which the electromagnetic brake 60 is switched from the braking state to the released state.

[0074] First, in step S331, a control process for acquiring a brake operation signal is started. Specifically, as an initial setting, the voltage applied to the electromagnetic brake 60 is set to 0, and n, which is a counter value indicating timing, is set to 1. Then, a current is passed through the motor 50, thereby causing the motor 50 to generate a driving force. In step S331, a driving force is generated in the motor 50, but because the electromagnetic brake 60 is in a braking state, the shaft 55, which is the output shaft of the motor 50, does not rotate.

[0075] Next, in step S332, the applied voltage, which is the voltage applied to the electromagnetic brake 60, is updated and the voltage is increased by a preset amount ΔV, and the counter value n is incremented by 1. Then, in step S333, it is determined whether the motor drive current has changed. That is, in step S332, when the electromagnetic brake 60 is switched from the braking state to the released state by updating the applied voltage, the motor 50 starts rotating, and the motor drive current flowing through the motor 50 becomes smaller. On the other hand, when the electromagnetic brake 60 cannot be switched to the released state even when the applied voltage is increased by ΔV, the value of the motor drive current does not change. In step S333, when it is determined that the motor drive current has changed, the process proceeds to step S334. In the case where it is not determined that the motor drive current has changed, the process returns to step S332, and again in step S332, the applied voltage is updated and the counter value n is incremented.

[0076] In step S334, the value of the counter value n is acquired. That is, the value of the counter value n is acquired at the time of the switching timing when the switching operation of the electromagnetic brake 60 is performed. The counter value n acquired in step S334 indicates the number of times the increment process in step S332 has been performed.

[0077] Then, in step S335, a brake operation signal indicating the switching timing in the switching operation of the electromagnetic brake 60 is acquired. For example, the brake operation signal is a signal indicating the switching timing by a ratio, as in the first embodiment. For example, steps S332 and S333 are repeatedly executed at every 10 millisecond control period. When the voltage change by the brake power supply unit 22 is executed for 20 seconds, the change in the applied voltage is 2000 counts in total. Therefore, the counter value n at the time of the switching timing acquired in step S334 is divided by the total count number of 2000 to acquire a brake operation signal indicating the switching timing by a ratio. The brake operation signal acquired by the control unit 321 is output to the learning device 30, as in the first embodiment.

[0078] As shown in FIG. 15, the learning device 30 acquires the brake operation signal acquired by the control unit 321 as a target variable, similar to the first embodiment. Then, the learning device 30 generates a trained model 380 by machine learning using the current signal acquired by the control unit 321 as an explanatory variable and the brake operation signal as a target variable, similar to the first embodiment. The method of generating the trained model 380 by the learning device 30 is similar to the method of generating the trained model 80 in the first embodiment. Then, the generated trained model 380 is stored in the storage unit 42 of the inspection device 40. The control unit 41 of the inspection device 40 acquires an estimated brake operation signal using the trained model 380 by executing a control process of a brake determination method similar to the first embodiment. The other configurations of the third embodiment are similar to those of the first embodiment.

[0079] [Effects of the third embodiment] In the robot system 300 according to the third embodiment, as described above, the motor current detection unit 24 as an operation detection unit detects the motor drive current supplied to the motor 50 in order to detect the switching timing of the switching operation. The learning device 30 as a learning unit acquires a brake operation signal as a target variable based on the motor drive current. Here, the magnitude of the motor drive current supplied to the motor 50 is different between the case where the motor 50 is rotated in the braking state and the case where the motor 50 is rotated in the released state. Therefore, it is possible to detect that the electromagnetic brake 60 has switched between the braking state and the released state based on the change in the motor drive current. As a result, even when the surrounding noise or vibration is large or the operating sound or vibration accompanying the switching operation is small, it is possible to generate a trained model 380 capable of accurately predicting the switching operation of the electromagnetic brake 60 by acquiring a brake operation signal indicating the switching timing of the switching operation based on the motor drive current. Note that other effects of the third embodiment are similar to those of the first embodiment.

[0080] [Fourth embodiment] Next, the configuration of a robot system 400 according to a fourth embodiment will be described with reference to Fig. 17. In the fourth embodiment, unlike the robot system 300 according to the third embodiment in which the switching operation of the electromagnetic brake 60 is detected by a change in the motor drive current, the robot system 400 detects the switching operation of the electromagnetic brake 60 based on the detection value of the encoder 51 that detects the rotation of the motor 50. Note that the same components as those in the third embodiment are denoted by the same reference numerals and will not be described.

[0081] As shown in FIG. 17, a robot system 400 according to the fourth embodiment includes a robot 410 including a robot controller 420. The robot controller 420 has a control unit 421. The control unit 421 controls the operation of each unit of the robot 410, similar to the control unit 321 of the third embodiment. In the fourth embodiment, the control unit 421 acquires a brake operation signal to be output to the learning device 30 based on the output from the encoder 51. That is, in the fourth embodiment, the encoder 51 is an example of an operation detection unit. As in the first embodiment, the encoder 51 detects the rotation of the motor 50 and outputs the detection result of the rotation of the motor 50 to the control unit 321. In the fourth embodiment, the encoder 51 detects the rotation of the motor 50 to detect the switching timing of the switching operation of the electromagnetic brake 60.

[0082] Next, the control process of acquiring the brake operation signal by the robot controller 420 will be described with reference to Fig. 18. Note that in steps S431, S432, S434, and S435, the same control processes as those in steps S331, S332, S334, and S335 of the third embodiment are performed, respectively. The control processes from step S431 to step S435 are executed by the control unit 421 of the robot controller 420. Note that Fig. 18 shows an example of the case where the brake operation signal is acquired in a switching operation in which the electromagnetic brake 60 is switched from the braking state to the released state.

[0083] First, in step S431, similar to step S331 in the third embodiment, the voltage applied to the electromagnetic brake 60 is set to 0, and the counter value n indicating the timing is set to 1. Next, in step S432, similar to step S332 in the third embodiment, the applied voltage, which is the voltage applied to the electromagnetic brake 60, is updated and the voltage is increased by a preset amount ΔV, and the counter value n is incremented by 1.

[0084] Next, in step S433, it is determined whether the encoder value, which is the output of the encoder 51, has changed. That is, in step S432, if the electromagnetic brake 60 is switched from the braking state to the released state by updating the applied voltage, the motor 50 starts rotating, and the encoder value, which is the output of the encoder 51, changes. On the other hand, if the electromagnetic brake 60 cannot be switched to the released state even when the applied voltage is increased by ΔV, the encoder value does not change. If it is determined in step S433 that the encoder value has changed, the process proceeds to step S434. If it is not determined that the encoder value has changed, the process returns to step S432, where the applied voltage is updated again and the counter value n is incremented.

[0085] In step S434, the value of the counter numerical value n is acquired, similar to step S334 in the third embodiment. Then, in step S435, a brake operation signal indicating the switching timing in the switching operation of the electromagnetic brake 60 is acquired, similar to step S335 in the third embodiment. For example, the brake operation signal is a signal indicating the switching timing by a ratio, similar to the third embodiment. The brake operation signal acquired by the control unit 421 is output to the learning device 30, similar to the third embodiment.

[0086] As shown in FIG. 17, the learning device 30 acquires the brake operation signal acquired by the control unit 421 as a target variable, similar to the third embodiment. Then, the learning device 30 generates a trained model 480 by machine learning using the current signal acquired by the control unit 421 as an explanatory variable and the brake operation signal as a target variable, similar to the third embodiment. The method of generating the trained model 480 by the learning device 30 is similar to the method of generating the trained model 80 of the first embodiment. Then, the generated trained model 480 is stored in the storage unit 42 of the inspection device 40. The control unit 41 of the inspection device 40 acquires an estimated brake operation signal using the trained model 480 by executing a control process of the brake determination method similar to the first embodiment. The other configurations of the fourth embodiment are similar to those of the third embodiment.

[0087] [Effects of the fourth embodiment] As described above, the robot system 400 according to the fourth embodiment includes the encoder 51 as an operation detection unit that detects the rotation of the motor 50 to detect the switching timing of the switching operation. The learning device 30 as a learning unit acquires a brake operation signal as a target variable based on the output from the encoder 51. As a result, it is possible to accurately detect whether the motor 50 has rotated or not based on the output from the encoder 51, and therefore it is possible to accurately detect whether the switching operation of the electromagnetic brake 60 has caused the braking state and the released state to be switched over. As a result, it is possible to generate a trained model 480 that can accurately predict the switching operation of the electromagnetic brake 60. Other effects of the fourth embodiment are similar to those of the third embodiment.

[0088] [Fifth embodiment] Next, a configuration of a robot system 500 according to a fifth embodiment will be described with reference to Fig. 19 and Fig. 20. In the fifth embodiment, the robot system 500 uses a trained model 580 trained using a current signal from which an AC component has been extracted as an explanatory variable. Note that the same components as those in the first embodiment are denoted by the same reference numerals and will not be described.

[0089] 19, a robot system 500 according to the fifth embodiment includes a robot 510 including a robot controller 520. The robot controller 520 has a control unit 521. The control unit 521 controls the operation of each unit of the robot 510, similar to the control unit 21 of the first embodiment.

[0090] As shown in FIG. 20, the control unit 521 outputs a current signal as an explanatory variable of teacher data in machine learning to the learning device 30 based on the detection signal detected by the brake current detection unit 23, similar to the control unit 21 in the first embodiment. In the fifth embodiment, the control unit 521 outputs a current signal in which an AC component of the current detected by the brake current detection unit 23 is extracted, as an explanatory variable, to the learning device 30. Specifically, similar to the first embodiment, the control unit 521 acquires the detection result by the brake current detection unit 23 of the current flowing through the electromagnetic brake 60 when a voltage whose magnitude gradually changes is applied by the brake power supply unit 22. Then, the control unit 521 performs a process of extracting the AC component from the acquired detection result by the brake current detection unit 23. When a voltage whose magnitude gradually changes at a constant level is applied by the brake power supply unit 22, the current flowing through the electromagnetic brake 60 changes at a constant rate except for the switching timing of the switching operation of the electromagnetic brake 60. Therefore, the control unit 521 detects a change in current from, for example, timing t1 when the change in voltage applied by the brake power supply unit 22 starts to timing t6, which is a predetermined time point before the switching operation is performed. Timing t6 is, for example, a time point three seconds after timing t1. The control unit 521 obtains a gradient of the change in current detected in the period from timing t1 to timing t6, and obtains a difference of the amount of change corresponding to the obtained gradient of the change from the detection result of the entire current from timing t1 to timing t2, thereby extracting an AC component of the detected current. Note that the control unit 521 may extract the AC component using a high-pass filter.

[0091] As shown in FIG. 19, the learning device 30 acquires the current signal from which the AC component is extracted by the control unit 521 as an explanatory variable. Then, the learning device 30 generates a trained model 580 by machine learning using the current signal acquired by the control unit 521 as an explanatory variable and the brake operation signal as a target variable, similar to the first embodiment. The method of generating the trained model 580 by the learning device 30 is similar to the method of generating the trained model 80 in the first embodiment. Then, the generated trained model 580 is stored in the storage unit 42 of the inspection device 40. Note that, when performing the control process of the brake judgment method, the control unit 521 of the robot controller 520 outputs the current signal from which the AC component is extracted to the inspection device 40, similar to the learning device 30. The control unit 41 of the inspection device 40 acquires an estimated brake operation signal as an output of a predicted value from the trained model 580 for the current signal from which the AC component is extracted, by executing the control process of the brake judgment method similar to the first embodiment using the trained model 580. In addition, other configurations of the fifth embodiment are similar to those of the first embodiment.

[0092] [Effects of the fifth embodiment] In the robot system 500 according to the fifth embodiment, as described above, the control unit 41 acquires an estimated brake operation signal as an output of a predicted value from the trained model 580 for the current signal by using the trained model 580 generated by machine learning with the current signal from which the AC component of the current detected by the brake current detection unit 23 is extracted as an explanatory variable. Here, in the current signal of the current detected by the brake current detection unit 23, a change in the current signal due to the switching operation of the electromagnetic brake 60 appears as an AC current. Therefore, by performing machine learning with the current signal from which the AC component is extracted as an explanatory variable, learning can be performed using the explanatory variable in which the change due to the switching operation is extracted. Other effects of the fifth embodiment are similar to those of the first embodiment.

[0093] [Sixth embodiment] Next, the configuration of a robot system 600 according to the sixth embodiment will be described with reference to Fig. 21 and Fig. 22. In the sixth embodiment, unlike the robot system 100 of the first embodiment which uses a trained model 80 trained with a brake operation signal indicating the switching timing of the switching operation as a response variable, the robot system 600 uses a trained model 680 trained with a brake operation signal, which is a brake drive voltage in the switching operation, as a response variable. Note that the same components as those in the first embodiment are denoted by the same reference numerals and will not be described.

[0094] As shown in FIG. 21, a robot system 600 according to the sixth embodiment includes a robot 610 including a robot controller 620, a learning device 630, and an inspection device 640. The robot controller 620 has a control unit 621. The control unit 621 controls the operation of each unit of the robot 610, similar to the control unit 21 of the first embodiment. The control unit 621 outputs a current signal as an explanatory variable of teacher data in machine learning to the learning device 630, based on a detection signal detected by the brake current detection unit 23, similar to the control unit 21 of the first embodiment. The learning device 630 is an example of a learning unit.

[0095] As shown in FIG. 22, in the sixth embodiment, the control unit 621 outputs the brake drive voltage applied to the electromagnetic brake 60 in the switching operation of the electromagnetic brake 60 to the learning device 630 as a brake operation signal, which is a response variable. Specifically, like the control unit 21 in the first embodiment, the control unit 621 acquires a ratio representing the time from timing t1, which is the time when the change in the voltage applied by the brake power supply unit 22 starts, to timing t4, which is the switching timing when the braking state and the switching state of the electromagnetic brake 60 are switched, based on the operation sound detected by the operation sound detection unit 70. In addition, the control unit 621 acquires the voltage applied to the electromagnetic brake 60 from timing t1 to timing t2, when the current signal is acquired. The applied voltage is acquired, for example, based on a control amount for operating the brake power supply unit 22 of the robot controller 20. Then, the control unit 621 acquires the value of the brake drive voltage at the switching timing of the switching operation based on the voltage applied to acquire the current signal and the ratio indicating the switching timing acquired based on the operation sound detected by the operation sound detection unit 70. That is, the control unit 621 acquires the voltage applied to the electromagnetic brake 60 at the timing when the operating sound is detected as the brake drive voltage. The control unit 621 outputs the acquired brake drive voltage to the learning device 630 as a brake operation signal.

[0096] As shown in FIG. 21 , the learning device 630 acquires a brake operation signal, which is a brake driving voltage, as a response variable. Then, the learning device 630 generates a trained model 380 by performing machine learning using a current signal as an explanatory variable and a brake operation signal, which is a brake driving voltage, as a response variable. A method for generating the trained model 680 by the learning device 630 is similar to a method for generating the trained model 80 by the learning device 30 of the first embodiment. The learning device 630 generates a trained model 680 in which a relationship between the current signal and the brake driving voltage has been learned by performing machine learning using the brake driving voltage as correct answer data. The generated trained model 680 is stored in the memory unit 42 of the inspection device 640.

[0097] The inspection device 640 includes a control unit 641. The inspection device 640 judges whether or not there is an abnormality in the electromagnetic brake 60, similarly to the inspection device 40 of the first embodiment, by causing the control unit 641 to execute a program 642a stored in the storage unit 42. The program 642a is an example of a brake judgment program.

[0098] (Control process of brake determination method according to the sixth embodiment) Next, the control process of the brake determination method by the robot system 600 in the sixth embodiment will be described based on a flowchart with reference to Fig. 23. Note that the control processes from step S611 to step S613 in the brake determination method are performed by the control unit 641 of the inspection device 640 executing the program 642a.

[0099] First, in step S611, a current signal based on the detection result of the current flowing through the electromagnetic brake 60 is acquired, similar to step S11 of the first embodiment. Then, in step S612, by using the trained model 680 generated by machine learning in the learning device 630, an estimated brake operation signal is acquired as an output of a predicted value from the trained model 680 for the current signal acquired in step S611. Note that in the sixth embodiment, the estimated brake operation signal output from the trained model 680 that has learned the relationship between the current signal and the brake drive voltage is the brake drive voltage estimated by the trained model 680. For example, when a current signal is input to the trained model 680 in the inspection device 640, an estimated value that estimates the value of the brake drive voltage, such as "17.3V", is output. Then, in step S613, an abnormality in the electromagnetic brake 60 is determined, similar to step S14 of the first embodiment. Other configurations of the sixth embodiment are the same as those of the first embodiment.

[0100] [Effects of the sixth embodiment] In the sixth embodiment, the following effects can be obtained.

[0101] As described above, the robot system 600 according to the sixth embodiment includes a control unit 641 that uses a trained model 680 generated by machine learning to acquire an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model 680 for the current signal, using a current signal based on the detection result by the brake current detection unit 23 as an explanatory variable and a brake operation signal, which is a brake drive voltage applied to the electromagnetic brake 60 in a switching operation in which the electromagnetic brake 60 is switched between an engaged state and a released state, as an objective variable. In this way, since an estimated brake operation signal is acquired as an output of a predicted value from the trained model 680 generated by machine learning, it is possible to accurately distinguish between a change in the current signal due to the switching operation of the electromagnetic brake 60 and a change in the current signal due to a disturbance such as noise. Therefore, it is possible to accurately predict the brake drive voltage in the switching operation of the electromagnetic brake 60, compared to a case in which the amount of change in the current signal or a differential value of the amount of change in the current signal is compared with a preset threshold value to predict the brake drive voltage in the switching operation of the brake. As a result, it is possible to determine whether the electromagnetic brake 60 is operating normally or not based on the predicted brake drive voltage, and therefore it is possible to accurately determine an abnormality in the electromagnetic brake 60. Other effects of the sixth embodiment are similar to those of the first embodiment.

[0102] [Variations] It should be noted that the embodiments disclosed herein are illustrative and not restrictive in all respects. The scope of the present disclosure is indicated by the claims, not by the description of the embodiments above, and further includes all modifications (variations) within the meaning and scope equivalent to the claims.

[0103] For example, in the above first to sixth embodiments, an example has been shown in which an abnormality in the electromagnetic brake 60 is determined by acquiring a current signal in a switching operation in which the electromagnetic brake 60 is switched from a braking state to a release state by applying a voltage whose value gradually increases by the brake power supply unit 22, but the present disclosure is not limited to this. In the present disclosure, a current signal may be acquired by applying a voltage whose value gradually decreases. That is, in a switching operation in which the electromagnetic brake is switched from a release state to a braking state, a learned model may be generated by machine learning by acquiring a current signal and a brake operation signal, and an estimated brake operation signal in a switching operation in which the released state is switched to the braking state may be acquired by using the generated learned model. In addition, the switching operation in which the braking state is switched to the release state and the switching operation in which the released state is switched to the braking state may be determined together by gradually increasing and then gradually decreasing the value of the voltage applied by the brake power supply unit. In addition, a plurality of values ​​of voltage may be applied to the electromagnetic brake by a voltage whose value changes discretely, rather than a voltage whose value changes gradually. In addition, the change amount and magnitude of the voltage applied by the brake power supply unit may be changed according to the type of the electromagnetic brake to be targeted. That is, the applied voltage may be gradually increased from 0 V to a value other than 24 V, and the period during which the voltage changes may be other than 20 seconds. If a brake with a rated voltage of 24 V is used, the voltage may be applied from 0 V to 30 V with some margin.

[0104] In the above first to sixth embodiments, the brake drive voltage is acquired based on the output from the trained model, and an abnormality of the electromagnetic brake 60 is determined based on the acquired brake drive voltage. However, the present disclosure is not limited to this. In the present disclosure, an abnormality of the electromagnetic brake may be determined without acquiring the brake drive voltage. For example, when an estimated brake operation signal indicating the switching timing of the switching operation as a ratio is acquired, the acquired estimated brake operation signal itself may be compared with a threshold value to determine an abnormality of the electromagnetic brake. Also, the acquired brake drive voltage or the estimated brake operation signal may be output without determining an abnormality. For example, the acquired brake drive voltage or the estimated brake operation signal may be displayed on the display unit. Also, when determining an abnormality, not only whether an abnormality has occurred but also whether there is a possibility that an abnormality will occur in the future, or a life prediction that is a prediction of the period until an abnormality occurs may be performed as the abnormality determination. Also, when determining an abnormality, the degree of deterioration of the electromagnetic brake may be determined.

[0105] In the above first to sixth embodiments, the robot controller 20, 220, 320, 420, 520, 620 that controls the operation of the robot 10, 210, 310, 410, 510, 610, the learning device 30, 630 as a learning unit that generates the trained model 80, 280, 380, 480, 580, 680, and the inspection device 40, 640 that executes the control process of the brake judgment method are arranged separately from each other, but the present disclosure is not limited to this. In the present disclosure, at least two of the control of the robot operation, the generation of the trained model, and the control process of the brake judgment method may be executed by a common control device. In addition, the control process of acquiring the brake operation signal based on the detection result of the operation detection unit may be executed by the learning device, not by the robot controller that controls the robot operation.

[0106] In the above-mentioned first, second, fifth and sixth embodiments, the operation sound detection unit 70 as the operation detection unit or the vibration detection unit 270 as the operation detection unit is disposed on the bracket 52 of the motor 50, but the present disclosure is not limited to this. In the present disclosure, the operation detection unit may be disposed at a position separated from the motor. Also, the operation detection unit may be disposed on the motor at a position different from the bracket on the encoder side. Also, as the operation detection unit, a mechanical switch that detects the movement of the armature of the electromagnetic brake may be disposed on the motor. For example, the movable iron core may be disposed so as to move integrally with the armature, and the electromagnetic brake switching operation may be detected by switching the switch due to the movement of the movable iron core. Also, a capacitance sensor, a photoelectric sensor, a piezoelectric sensor, or the like may be used as the operation detection unit. Also, the operation detection unit may be disposed only when the brake operation signal as the objective variable is acquired, and the operation detection unit may be removed after the trained model is generated. That is, the operation detection unit may not be disposed at the time of executing the control process of the brake determination method.

[0107] In the above first to sixth embodiments, the brake operation signal of the electromagnetic brake 60 is estimated by using the trained models 80, 180, 280, 380, 480, 580, and 680 generated by performing machine learning using the multiple robots 10, 210, 310, 410, 510, and 610, but the present disclosure is not limited to this. In the present disclosure, the robot used to acquire the predicted value of the brake operation signal estimated using the trained model generated and the robot used to acquire the explanatory variables and the objective variables of the teacher data for machine learning may be different robots. In that case, the robot used to acquire the predicted value of the brake operation signal estimated may not be provided with a motion detection unit. In addition, machine learning may be performed using multiple types of robots. For example, machine learning may be performed using multiple types of robots having motors and electromagnetic brakes of a common model, or machine learning may be performed using multiple types of motors or electromagnetic brakes. Also, a current signal as an explanatory variable for generating a trained model and a brake operation signal as a target variable may be obtained by applying a voltage from a brake power supply unit to the motor and electromagnetic brake removed from the robot, or to only the electromagnetic brake. In this case, since it is not necessary to operate the robot to obtain the teacher data, the teacher data can be obtained without considering the motion range of the robot. In addition, when the electromagnetic brake is disposed on the output shaft side of the motor for transmitting rotation to the outside, the switching operation can be more easily detected by the motion sound detection unit or vibration detection unit as the motion detection unit when the motor and electromagnetic brake are removed from the robot.

[0108] In the second embodiment, the vibration detection unit 270 serving as the motion detection unit is an AE sensor, but the present disclosure is not limited to this. In the present disclosure, the vibration detection unit may be a vibration meter.

[0109] In the above first to fifth embodiments, an example is shown in which a brake operation signal is acquired that indicates the time from timing t1, which is the time when the change in voltage applied by the brake power supply unit 22 in the switching operation, to timing t4, which is the switching timing when the braking state and the switching state of the electromagnetic brake 60 are switched, as a ratio, but the present disclosure is not limited to this. In the present disclosure, the time of the switching timing may be represented by a brake operation signal expressed by a numerical value such as the number of seconds, the count number of control periods, or the number of clocks. Also, as in the brake operation signal according to the modified example shown in FIG. 24, a brake operation signal of time series data indicating the switching timing by switching a signal between high and low may be acquired. In this case, instead of a signal in which a switched state is maintained as in the example of FIG. 24, a brake operation signal of time series data indicating the switching timing may be acquired by a signal that switches between high and low in a pulsed manner as in the brake operation signal according to the modified example shown in FIG. 25.

[0110] In the above first to sixth embodiments, the robots 10, 210, 310, 410, 510, and 610 are industrial robots that include six-axis vertical joints and perform handling tasks such as transportation, assembly, and machining, but the present disclosure is not limited to this. In the present disclosure, the robot may include joints with five or fewer axes or seven or more axes. A horizontal multi-joint robot may be used as the robot. The robot may have a drive mechanism other than a robot arm with a parallel link structure or a serial link structure such as a linear motion mechanism. The robot may be a medical robot rather than an industrial robot.

[0111] In addition, in the above first to fifth embodiments, when the brake driving voltage is obtained based on the output from the trained models 80, 280, 380, 480, and 580, an example is shown in which the voltage applied by the brake power supply unit 22 is obtained based on the controlled amount, and in the above sixth embodiment, when the brake operation, which is the brake driving voltage as the objective variable for generating the trained model 680, is obtained, an example is shown in which the voltage applied by the brake power supply unit 22 is obtained based on the controlled amount, but the present disclosure is not limited to this. In the present disclosure, the voltage applied by the brake power supply unit may be obtained as an actual measured value using a voltmeter, rather than being obtained based on the controlled amount. Also, a value set in advance as the voltage applied by the brake power supply unit may be stored.

[0112] In the above first and second embodiments, an example has been shown in which the brake operation signal is acquired by rectifying and smoothing the detection signal indicating the detection result from the operation sound detection unit 70 as a motion detection unit or the vibration detection unit 270 as a motion detection unit, but the present disclosure is not limited to this. In the present disclosure, delay correction may be performed on the smoothed detection signal in consideration of the timing delay that occurs when the detection signal from the motion detection unit is rectified and smoothed.

[0113] The functions of the elements disclosed herein can be performed using circuits or processing circuits, including general purpose processors, special purpose processors, integrated circuits, Application Specific Integrated Circuits (ASICs), conventional circuits, and / or combinations thereof, configured or programmed to perform the disclosed functions. Processors are considered processing circuits or circuits because they include transistors and other circuits. In this disclosure, a circuit, unit, or means is hardware that performs the recited functions or hardware that is programmed to perform the recited functions. The hardware may be hardware disclosed herein or other known hardware that is programmed or configured to perform the recited functions. Where the hardware is a processor, which is considered a type of circuit, the circuit, means, or unit is a combination of hardware and software, and the software is used to configure the hardware and / or the processor.

[0114] [Aspects] It will be appreciated by those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0115] (Aspect 1) an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of the motor that is the driving source of the robot and a released state in which the brake is turned off; a brake current detection unit that detects a current flowing through the electromagnetic brake; a control unit that uses a trained model generated by machine learning, with a current signal based on the detection result by the brake current detection unit as an explanatory variable, and a brake operation signal indicating the switching timing of a switching operation for switching the electromagnetic brake between the braking state and the released state as an objective variable, and acquires an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal.

[0116] (Aspect 2) A brake power supply unit that applies a voltage whose magnitude gradually changes to the electromagnetic brake is further provided. The control unit is a current signal based on a detection result of a current flowing through the electromagnetic brake by applying a voltage whose magnitude gradually changes from the brake power supply unit; The robot system of aspect 1, wherein the estimated brake operation signal is obtained as an output of a predicted value from the trained model for the acquired current signal.

[0117] (Aspect 3) The robot system according to aspect 1 or aspect 2, wherein the control unit acquires a brake drive voltage applied to the electromagnetic brake at a switching timing of the switching operation based on the acquired estimated brake operation signal.

[0118] (Aspect 4) An operation detection unit that detects a switching timing of the switching operation; The robot system according to any one of aspects 1 to 3, further comprising a learning unit that acquires the current signal as the explanatory variable and acquires the brake operation signal indicating a state of the switching operation detected by the operation detection unit as the objective variable, thereby generating the trained model by machine learning using the explanatory variables and the objective variable.

[0119] (Aspect 5) the operation detection unit includes an operation sound detection unit that detects an operation sound in the switching operation in order to detect a switching timing of the switching operation, 5. The robot system according to claim 4, wherein the learning unit acquires the brake operation signal as the objective variable based on a detection signal detected by the operation sound detection unit.

[0120] (Aspect 6) the motion detection unit includes a vibration detection unit that detects vibrations caused by the switching motion in order to detect a switching timing of the switching motion, 5. The robot system according to claim 4, wherein the learning unit acquires the brake operation signal as the objective variable based on a detection signal detected by the vibration detection unit.

[0121] (Aspect 7) the operation detection unit detects a motor drive current supplied to the motor in order to detect a switching timing of the switching operation; 5. The robot system of claim 4, wherein the learning unit acquires the brake operation signal as the objective variable based on the motor drive current.

[0122] (Aspect 8) the operation detection unit includes an encoder that detects rotation of the motor in order to detect a switching timing of the switching operation, 5. The robot system of claim 4, wherein the learning unit acquires the brake operation signal as the objective variable based on an output from the encoder.

[0123] (Aspect 9) The control unit acquires, as an output of a predicted value from the learned model, the estimated brake operation signal representing the time from the point at which the change in voltage applied by the brake power supply unit begins in the switching operation to the switching timing at which the electromagnetic brake is switched between the applied state and the released state.

[0124] (Aspect 10) A robot system described in any one of aspects 1 to 9, wherein the control unit uses the trained model generated by machine learning with the current signal, from which the AC component of the current detected by the brake current detection unit is extracted, as the explanatory variable, to obtain the estimated brake operation signal as an output of a predicted value from the trained model for the current signal.

[0125] (Aspect 11) A current signal is obtained based on the detection result of a current flowing through an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of a motor that is the driving source of the robot and a released state in which the brake is turned off; A brake judgment method in which the current signal is used as an explanatory variable, a brake operation signal indicating the switching timing of the switching operation for switching the electromagnetic brake between the braking state and the released state is used as an objective variable, and an estimated brake operation signal, which is an estimated brake operation signal, is obtained as an output of a predicted value from the learned model for the acquired current signal by using a trained model generated by machine learning.

[0126] (Aspect 12) A brake judgment program that causes a computer to execute the brake judgment method according to aspect 11.

[0127] (Aspect 13) an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of the motor that is the driving source of the robot and a released state in which the brake is turned off; a brake current detection unit that detects a current flowing through the electromagnetic brake; a control unit that uses a trained model generated by machine learning to obtain an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal, using a current signal based on the detection result by the brake current detection unit as an explanatory variable and a brake operation signal that is a brake drive voltage applied to the electromagnetic brake in a switching operation in which the electromagnetic brake is switched between the braking state and the released state as an objective variable.

[0128] (Aspect 14) A current signal is obtained based on the detection result of a current flowing through an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of a motor that is the driving source of the robot and a released state in which the brake is turned off; A brake judgment method in which the current signal is an explanatory variable, a brake operation signal which is a brake driving voltage applied to the electromagnetic brake during a switching operation in which the electromagnetic brake is switched between the braking state and the released state is a target variable, and an estimated brake operation signal which is an estimated brake operation signal is obtained as an output of a predicted value from the trained model for the acquired current signal by using a trained model generated by machine learning.

[0129] (Aspect 15) A brake judgment program that causes a computer to execute the brake judgment method according to aspect 14. [Explanation of symbols]

[0130] 10, 210, 310, 410, 510, 610 Robot 22 Brake power supply unit 23 Brake current detection section 24 Motor current detection section (operation detection section) 30, 630 Learning device (learning unit) 40,640 Inspection equipment 41, 641 Control section 42a, 642a program (braking judgment program) 50 Motor 51 Encoder (motion detection unit) 60 Electromagnetic brake 70 Operation sound detection unit (operation detection unit) 80, 280, 380, 480, 580, 680 trained models 100, 200, 300, 400, 500, 600 Robot Systems 270 Vibration detection unit (motion detection unit)

Claims

1. an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of the motor that is the driving source of the robot and a released state in which the brake is turned off; a brake current detection unit that detects a current flowing through the electromagnetic brake; a control unit that uses a trained model generated by machine learning to acquire an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal, using a current signal based on the detection result by the brake current detection unit as an explanatory variable of teacher data, and a brake operation signal that indicates a switching timing of a switching operation for switching the electromagnetic brake between the braking state and the released state as an objective variable of the teacher data.

2. A brake power supply unit that applies a voltage whose magnitude gradually changes to the electromagnetic brake is further provided. The control unit is a current signal based on a detection result of a current flowing through the electromagnetic brake by applying a voltage whose magnitude gradually changes from the brake power supply unit; The robot system of claim 1 , wherein the estimated brake operation signal is obtained as an output of a predicted value from the trained model for the acquired current signal.

3. The robot system according to claim 1 , wherein the control unit acquires a brake drive voltage applied to the electromagnetic brake at a switching timing of the switching operation based on the acquired estimated brake operation signal.

4. An operation detection unit that detects a switching timing of the switching operation; 3. The robot system according to claim 1, further comprising a learning unit that acquires the current signal as the explanatory variable and acquires the brake operation signal indicating a state of the switching operation detected by the operation detection unit as the objective variable, thereby generating the trained model by machine learning using the explanatory variables and the objective variable.

5. the operation detection unit includes an operation sound detection unit that detects an operation sound in the switching operation in order to detect a switching timing of the switching operation, The robot system according to claim 4 , wherein the learning unit acquires the brake operation signal as the objective variable based on a detection signal detected by the operation sound detection unit.

6. the motion detection unit includes a vibration detection unit that detects vibrations caused by the switching motion in order to detect a switching timing of the switching motion, The robot system according to claim 4 , wherein the learning unit acquires the brake operation signal as the objective variable based on a detection signal detected by the vibration detection unit.

7. the operation detection unit detects a motor drive current supplied to the motor in order to detect a switching timing of the switching operation; The robot system according to claim 4 , wherein the learning unit acquires the brake operation signal as the objective variable based on the motor drive current.

8. the operation detection unit includes an encoder that detects rotation of the motor in order to detect a switching timing of the switching operation, The robot system according to claim 4 , wherein the learning unit acquires the brake operation signal as the objective variable based on an output from the encoder.

9. 3. The robot system according to claim 2, wherein the control unit obtains, as a predicted value output from the learned model, the estimated brake operation signal representing the time from a point at which a change in the voltage applied by the brake power supply unit begins in the switching operation to a switching timing at which the electromagnetic brake is switched between the applied state and the released state.

10. The robot system of claim 1 or 2, wherein the control unit uses the trained model generated by machine learning with the current signal from which the AC component of the current detected by the brake current detection unit is extracted as the explanatory variable, thereby acquiring the estimated brake operation signal as an output of a predicted value from the trained model for the current signal.

11. A current signal is obtained based on the detection result of a current flowing through an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of a motor that is the driving source of the robot and a released state in which the brake is turned off; A brake judgment method in which the current signal is used as an explanatory variable of teacher data, a brake operation signal indicating the switching timing of the switching operation for switching the electromagnetic brake between the braking state and the released state is used as the objective variable of the teacher data, and an estimated brake operation signal, which is an estimated brake operation signal, is obtained as an output of a predicted value from the learned model for the acquired current signal by using a learned model generated by machine learning.

12. A braking judgment program that causes a computer to execute the braking judgment method according to claim 11.

13. an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of the motor that is the driving source of the robot and a released state in which the brake is turned off; a brake current detection unit that detects a current flowing through the electromagnetic brake; a control unit that uses a trained model generated by machine learning to acquire an estimated brake operation signal, which is an estimated brake operation signal, as an output of a predicted value from the trained model for the current signal, using a current signal based on the detection result by the brake current detection unit as an explanatory variable of teacher data, and a brake operation signal that is a brake drive voltage applied to the electromagnetic brake in a switching operation in which the electromagnetic brake is switched between the braking state and the released state as a target variable of the teacher data.

14. A current signal is obtained based on the detection result of a current flowing through an electromagnetic brake that switches between a braking state in which the brake is applied to the rotation of a motor that is the driving source of the robot and a released state in which the brake is turned off; A brake judgment method in which the current signal is an explanatory variable of teacher data, a brake operation signal which is a brake driving voltage applied to the electromagnetic brake during a switching operation in which the electromagnetic brake is switched between the braking state and the released state is an objective variable of the teacher data, and an estimated brake operation signal which is an estimated brake operation signal is obtained as an output of a predicted value from the trained model for the acquired current signal by using a trained model generated by machine learning.

15. A braking judgment program that causes a computer to execute the braking judgment method according to claim 14.

Images