Determination device

By designing a determination device for industrial robots, using data acquisition and machine learning to generate evaluation functions, the error judgment problem in robot power control parameters adjustment is solved, and the effect of accurate judgment is achieved that even if unfamiliar staff are able to make employees, improving The accuracy of machine learning.

CN111438687BActive Publication Date: 2024-11-15FANUC LTD
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202010043066.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-01-16
Filing Date
2020-01-15
Publication Date
2024-11-15
Estimated Expiration
2040-01-15

AI Technical Summary

Technical Problem

In robotic force control, parameter adjustment needs to be manually set, especially for unfamiliar staff, it is easy to lead to misjudgment and affect the accuracy of machine learning.

Method used

A determination device is designed to generate a learning model for judging the actions of industrial robots through data acquisition, evaluation function production and machine learning, so as to realize automatic judgment of force and torque states, independent of the experience of the staff.

Benefits of technology

Even unfamiliar staff can judge the operation of industrial robots based on predetermined benchmarks, improving the determination stability and accuracy of machine learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111438687B_ABST
    Figure CN111438687B_ABST
Patent Text Reader

Abstract

The present invention provides a determination device, which acquires the state of force and torque applied to a manipulator during the action of force control of an industrial robot, the state of position and posture as acquired data, and creates an evaluation function for evaluating the quality of the action of the industrial robot based on the acquired data. In addition, using the evaluation function, determination data for the acquired data is created, and in addition, based on the acquired data, state data for machine learning is created. In this way, using these state data and determination data, a learning model for determining the quality of the action of the industrial robot is generated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a determination device, and more particularly to a determination device for automatically determining state data related to force control of a robot. Background Art

[0002] Sometimes, robots with force sensors are used to precisely fit parts together, align the phase of gears, deburr, grind, and so on. Such work is achieved by having the robot hold parts or tools and control the force in a predetermined direction to a desired value (force control). Impedance control, damping control, hybrid control, and the like are known as such force control methods, but in any of these methods, an appropriate control gain must be set manually. Methods for automatically adjusting the control gain are well known (e.g., Japanese Patent Publication No. 2007-237312, etc.).

[0003] As described above, the method for setting the control gain for force control is well known, but other parameters, such as appropriate pressing force, travel speed, travel direction, etc., need to be adjusted by the staff. There will be differences between skilled staff and unfamiliar staff in the judgment of the quality of such parameter adjustment. In particular, if the judgment of parameter adjustment is made by unfamiliar staff, it may take time to make a judgment by trial and error, or a judgment different from the actual result of parameter adjustment may be made, or subsequent work based on the judgment result may be hindered. For example, there is a problem that if machine learning is performed based on such an erroneous judgment result, a high-precision learning result (learning model) cannot be obtained. Summary of the invention

[0004] Therefore, the present invention provides a determination device that assists in adjusting force control parameters so that even an unfamiliar worker can make the same determination as an experienced worker.

[0005] In one embodiment of the present invention, in the adjustment of various parameters related to the control of the manipulator of the industrial robot, an evaluation function is created that outputs a judgment result related to the success / failure of the action for data obtained from the action when the force control of the industrial robot is performed, thereby solving the above-mentioned problem. In the judgment device configured in this way, by using an evaluation function that outputs a judgment result according to a predetermined reference, the action of the data obtained when the industrial robot is in action is judged using the evaluation function, so that stable action judgment can be performed regardless of the experience of the operator.

[0006] The determination device of the present invention determines the action of an industrial robot having a function of detecting the force and torque applied to a manipulator. The determination device comprises: a data acquisition unit that acquires at least the state, position, and posture of the force and torque applied to the manipulator during the action of the industrial robot when performing force control as acquired data; an evaluation function creation unit that creates an evaluation function for evaluating the quality of the action based on the acquired data; a determination data creation unit that creates determination data for the acquired data using the evaluation function created by the evaluation function creation unit; a preprocessing unit that creates state data for machine learning based on the acquired data; and a learning unit that generates a learning model for determining the quality of the action of the industrial robot using the state data and the determination data.

[0007] The data acquisition unit may generate temporary determination data for determining whether the action is good or bad based on a time taken for the action, and the evaluation function generation unit may generate the evaluation function based on the generated temporary determination data.

[0008] Since the determination device of the present invention has the above-mentioned structure, even an unfamiliar worker can determine whether the action of the industrial robot is good or bad based on a predetermined standard during the force control parameter adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a schematic hardware configuration diagram of a determination device according to one embodiment.

[0010] Figure 2 This is a schematic functional block diagram of the control device according to the first embodiment.

[0011] Figure 3 This is a diagram showing an example of the fitting work of components performed by an industrial robot.

[0012] Figure 4 This is a diagram showing an example of gear assembly work performed by an industrial robot.

[0013] Figure 5 This is a diagram showing an example of deburring and grinding work of a workpiece performed by an industrial robot.

[0014] Figure 6 This is a schematic functional block diagram of a control device according to a second embodiment.

[0015] Figure 7 This is a schematic functional block diagram of a system according to the third embodiment.

[0016] Figure 8 This is a schematic hardware configuration diagram of a computer according to one embodiment.

[0017] Fig. 9 This is a schematic functional block diagram of a system according to a fourth embodiment. DETAILED DESCRIPTION

[0018] Figure 1 This is a schematic hardware configuration diagram showing the main parts of a control device having a machine learning device according to one embodiment. For example, the determination device 1 of this embodiment can be implemented as a control device for controlling an industrial robot that performs fitting, assembly, deburring, etc. of parts.

[0019] The CPU 11 included in the determination device 1 of the present embodiment is a processor that controls the determination device 1 as a whole. The CPU 11 reads a system program stored in the ROM 12 via the bus 20, and controls the determination device 1 as a whole according to the system program. The RAM 13 temporarily stores temporary calculation data, various data input by the operator via the teaching operation panel 60 and the input device 71, and the like.

[0020] The nonvolatile memory 14 is composed of, for example, a memory backed up by a battery not shown, a solid state drive (SSD), etc., and maintains the storage state even if the power supply of the determination device 1 is turned off. In the nonvolatile memory 14, teaching data input from the teaching operation panel 60 via the interface 19, data input from the input device 71, a program for controlling the industrial robot input via an interface not shown, data detected by the sensor 3 installed in the industrial robot (force, moment, etc. applied to the manipulator), data detected from the servo motor 50 driving the axis of the industrial robot (current value, position, speed, acceleration, torque, etc. of the servo motor 50), data read via an external storage device not shown, a network, etc. are stored. When executed / used, the program and various data stored in the nonvolatile memory 14 are expanded in the RAM 13. In addition, various system programs for executing processes such as control of the robot and teaching of a teaching position (including a system program for controlling exchanges with the machine learning device 100 described later) are written in advance in the ROM 12 .

[0021] The data read into the memory, the data obtained as a result of executing the program, the data output from the machine learning device 100 described later, etc. are output and displayed to the display device 70 via the interface 17. In addition, the input device 71 composed of a keyboard, a pointing device, etc. transmits instructions, data, etc. based on the operation of the staff to the CPU 11 via the interface 18.

[0022] The teaching operation panel 60 is a manual data input device having a display, a handle, hard keys, etc., receives and displays information from the determination device 1 via the interface 19 , and transmits pulses, commands, and various data input from the handle, hard keys, etc. to the CPU 11 .

[0023] The axis control circuit 30 for controlling the axis of the joints and the like of the industrial robot receives the axis movement instruction from the CPU 11 and outputs the axis instruction to the servo amplifier 40. The servo amplifier 40 receives the instruction and drives the servo motor 50 for moving the axis of the robot. The servo motor 50 of the axis has a built-in position / speed detector, and the position / speed feedback signal from the position / speed detector is fed back to the axis control circuit 30 to perform position / speed feedback control. Figure 1 In the hardware structure diagram of FIG, only one axis control circuit 30, servo amplifier 40, and servo motor 50 are shown, but in fact, the number of axes corresponding to the number of axes of the robot to be controlled is prepared. For example, in the case of a robot with 6 axes, an axis control circuit 30, a servo amplifier 40, and a servo motor 50 are prepared for each axis.

[0024] The interface 21 is an interface for connecting the determination device 1 and the machine learning device 100. The machine learning device 100 includes a processor 101 for controlling the entire machine learning device 100, a ROM 102 for storing system programs, etc., a RAM 103 for temporary storage in each process of machine learning, and a non-volatile memory 104 for storing learning models, etc. The machine learning device 100 can observe various information (force and torque applied to the manipulator, current value, position, speed, acceleration, torque, etc. of the servo motor 50) that can be obtained by the determination device 1 through the interface 21. In addition, the determination device 1 receives a change instruction of a force command or a torque command output from the machine learning device 100, and performs corrections on the control instructions of the robot based on the program and teaching data.

[0025] Figure 2 This is a schematic functional block diagram of the determination device 1 and the machine learning device 100 according to the first embodiment.

[0026] The determination device 1 of this embodiment has a structure (learning mode) required when the machine learning device 100 performs learning. Figure 1 The CPU 11 of the determination device 1 shown in the figure and the processor 101 of the machine learning device 100 execute the respective system programs to control the operations of the respective parts of the determination device 1 and the machine learning device 100, thereby realizing Figure 2 The functional modules shown.

[0027] The determination device 1 of this embodiment includes a control unit 32, a data acquisition unit 34, an evaluation function creation unit 36, a determination data creation unit 37, and a preprocessing unit 38. The machine learning device 100 included in the determination device 1 includes a learning unit 110. In addition, the non-volatile memory 14 is provided with an acquisition data storage unit 54 for storing data acquired from the industrial robot 2, the sensor 3, etc., and an evaluation function storage unit 56 for storing an evaluation function for evaluating the acquired data. The non-volatile memory 104 of the machine learning device 100 is provided with a learning model storage unit 130 for storing a learning model constructed by machine learning of the learning unit 110.

[0028] The control unit 32 has a general control function required for controlling each part of the industrial robot 2, so as to control the movement of the industrial robot 2 according to the control program 52 stored in the non-volatile memory 104, or control the industrial robot 2 according to the control operation of the operator on the operation panel (not shown), the control program stored in the non-volatile memory 14, etc., and outputs the command data as the change amount of the axis angle to the motor driving the axis in each control cycle when the movement of each axis (joint) provided by the industrial robot 2 is instructed by the control program 52. In addition, the control unit 32 obtains the motor state quantity (motor current value, position, speed, acceleration, torque, etc.) of each motor provided by the industrial robot 2, and outputs it to the data acquisition unit 34.

[0029] The data acquisition unit 34 acquires various data from the industrial robot 2, the sensor 3, and the input device 71. For example, the data acquisition unit 34 acquires commands based on a program for controlling the industrial robot, force and torque applied to the manipulator of the industrial robot 2, and current values, positions, speeds, accelerations, torques, etc. of the servo motors 50 driving the axes of the industrial robot 2, and stores them in the acquired data storage unit 54. The data acquisition unit 34 can acquire data detected from the sensor 3 and data applied to the servo motors 50 as time series data, and can also acquire data from other computers via an external storage device (not shown) or a wired / wireless network 7.

[0030] At the stage where the judgment device 1 generates an evaluation function through the evaluation function generating unit 36 ​​(evaluation function generating mode), the data acquisition unit 34 generates temporary judgment data for classifying the acquired data according to a predetermined predetermined criterion for the acquired data acquired from the industrial robot 2, the sensor 3, and the input device 71, and associates the generated temporary judgment data with the acquired data, and stores them in the acquired data storage unit 54 as data for generating the evaluation function.

[0031] As an example of the temporary determination data generated by the data acquisition unit 34, Figure 3 As an example, a case where a cylinder is inserted into a cylindrical hole and a cylinder is inserted into the cylinder hole for the purpose of inserting the cylinder, etc. is used to judge whether the cylinder is inserted into the cylindrical hole. Then, the following case is considered: the threshold value t is predetermined. lim1 <Threshold t lim2 <Threshold t lim3 , the time t required to reach the target depth is less than t lim1 , it is considered successful (high), and the time t taken until the target depth is reached is t lim1 Above and below the threshold t lim2 , the time t required to fit to the target depth is t lim2 Above and below the threshold t lim3 In the case of success (low), at time t lim3 If the target depth has not been reached, the process is considered a failure.

[0032] In this way, the data acquisition unit 34 classifies the acquired data into a plurality of categories such as success, failure, etc. according to a predetermined criterion, and further classifies each category such as success, failure, etc. into a plurality of sub-categories. In the above example, the case where the successful category is further sub-classified according to the time criterion is shown, but for example, the successful category can be classified into a sub-category that emphasizes details (classified corresponding to the maximum value of the force applied to the front end of the robot), and the failed category can also be classified corresponding to the degree of prying (classified corresponding to the maximum value of the force applied to the front end of the robot) and the accuracy (accuracy of the position of the front end of the robot). For example, the criterion can be appropriately determined by a skilled staff member.

[0033] In addition, when the determination device 1 performs machine learning by the machine learning device 100 (learning mode), the data acquisition unit 34 stores the acquired data acquired from the industrial robot 2, sensor 3, input device 71, etc. as acquired data used for machine learning in the acquired data storage unit 54.

[0034] The evaluation function creation unit 36 ​​creates an evaluation function that outputs judgment data for the acquired data based on the evaluation function creation data acquired by the data acquisition unit 34 and stored in the acquired data storage unit 54, and stores the evaluation function in the evaluation function storage unit 56. The evaluation function created by the evaluation function creation unit 36 ​​is composed of, for example, a function that calculates a predetermined evaluation value using each data value of the acquired data as a parameter, and an evaluation formula that derives judgment data from the evaluation value. The judgment data output by the evaluation function created by the evaluation function creation unit 36 ​​is used for machine learning of the machine learning device 100 together with the state data created based on the acquired data.

[0035] An example of the evaluation function generated by the evaluation function generating unit 36 ​​is described. Figure 3 A cylindrical fitting in which a cylinder is inserted into a cylindrical hole, which is used for the purpose of inserting a cylinder, as shown. In the case of such a cylindrical fitting, as acquired data, the data acquisition unit 34 can acquire the time taken from the start to the end of the action, the transition of 6 components of force data at the front end of the robot, the transition of 6 components of position data (which can be calculated based on the transition of position data of each servo motor), etc. from the industrial robot 2, the sensor 3, etc. In addition, the state data S used for machine learning of the machine learning device 100 is determined, for example, as 6 components of time-series force and torque data (F x 、F y 、F z 、M x 、M y 、M z ), and six components of position and posture data (P x , P y , P z , P w , P d , P r ), and as the judgment data D, there are values ​​of success (high), success (medium), success (low), and failure.

[0036] At this time, the evaluation function storage unit 56 stores in advance, for example, a prototype of a function for calculating an evaluation value as exemplified by the following formula (1) and a prototype of an evaluation formula as exemplified by the following formula (2). In the formula (1), Q k represents the evaluation value, t represents the time, Fp represents the maximum value of the pressing force, Po represents the change in posture, Pd represents the change in position, and w k1 、w k2 、w k3 、w k4 In the formula (2), lim0, lim1, lim2, and lim3 represent threshold values ​​for determining the classification of the judgment data.

[0037] Q k =t×w k1 +Fp×w k2 +Po×w k3 +Pd×w k4 ……(1)

[0038] Among them, w k1 +w k2 +w k3 +wk4 =1

[0039] Failure: If Q k >lim0

[0040] Success (Low): If Q k >lim1,

[0041] Success (Medium): If Q k >lim2,

[0042] Success (High): If Q k >lim3……(2)

[0043] Among them, lim3 <lim2<lim1<lim0

[0044] The evaluation function preparation unit 36 ​​prepares the evaluation function by determining the weighting coefficients and thresholds in the above formulas (1) and (2). More specifically, the time t, the maximum value Fp of the pressing force, the amount of change Po of the posture, and the amount of change Pd of the position calculated based on the acquired data included in the acquired data storage unit 54 acquired by the data acquisition unit 34 and included in the evaluation function preparation data are substituted into the formula (1), and the weighting coefficients and thresholds are determined so that the evaluation values ​​Q calculated at this time are k The evaluation function creation unit 36 ​​may obtain the weighting coefficient by a known statistical method such as correlation analysis or regression analysis. In addition, the evaluation function creation unit 36 ​​may obtain the threshold value by using the boundary value as the threshold value in accordance with the distribution of the evaluation values ​​of each acquired data in the adjacent classification.

[0045] The determination data generating unit 37 generates determination data for the acquired data for machine learning stored in the acquired data storage unit 54 using the evaluation function generated by the evaluation function generating unit 36 ​​and stored in the evaluation function storage unit 56. The determination data generated by the determination data generating unit 37 is used in the generation of learning data by the preprocessing unit 38.

[0046] The preprocessing unit 38 generates learning data used for learning by the machine learning device 100 based on the acquired data for machine learning acquired by the data acquisition unit 34 and the determination data generated by the determination data generation unit 37 for the acquired data. The preprocessing unit 38 generates state data obtained by converting the data acquired by the data acquisition unit 34 into a unified format (digitalization, sampling, etc.) processed in the machine learning device 100.

[0047] The state data S generated by the preprocessing unit 38 may include, for example, force state data S1 including information on the force applied to the manipulator of the industrial robot 2, and position state data S2 including information on the position of the manipulator. In this case, the force state data S1 may include at least the force applied to the manipulator of the industrial robot 2 in the feeding direction, and may also include the force and torque of the force applied in directions other than the feeding direction. The force state data S1 may be detected by the sensor 3 provided in the manipulator of the industrial robot 2, or may be obtained comprehensively based on the current flowing through the servo motor 50 driving each axis of the industrial robot 2. The position state data S2 may also include at least the control instructions such as the speed instruction, force instruction, and torque instruction in the feeding direction of the manipulator as the adjustment behavior of the control instructions (speed instruction, force instruction, torque instruction, etc.) in the feeding direction of the manipulator performed under the force state of the force state data S1, and may also include the control instructions in directions other than the feeding direction.

[0048] Note that the determination data D used as learning data by the preprocessing unit 38 may be the determination data created by the determination data creating unit 37 as it is.

[0049] The learning unit 110 performs machine learning using the data generated by the preprocessing unit 38. The learning unit 110 generates a learning model corresponding to the action state of the industrial robot 2 adjusted by the worker and obtained by learning the judgment of the quality of the action by using a known machine learning method such as supervised learning, and stores the generated learning model in the learning model storage unit 130. Examples of supervised learning methods performed by the learning unit 110 include a multilayer perceptron method, a recurrent neural network method, a long short-term memory method, and a convolutional neural network method.

[0050] The learning unit 110 can perform supervised learning on the state data S generated based on the acquired data using the judgment data D generated by the judgment data generating unit 37 using the evaluation function based on the acquired data as the learning data of the label data, and generate the judgment boundary of the success / failure of the action of the industrial robot 2 (that is, the success / failure of the adjustment action performed by the worker), the judgment boundary of the degree of success or the degree of failure as the learning model. The judgment unit 120 described later can use the learning model generated in this way to judge the success / failure, the degree of success, and the degree of failure of the action based on the data acquired when the industrial robot 2 is in action, and calculate the judgment result.

[0051] In the determination device 1 having the above-mentioned structure, it is possible to create an evaluation function used when determining the motion of the industrial robot 2 based on the data obtained when the industrial robot 2 is in motion. By using the evaluation function created by the determination device 1 to create determination data for the data obtained when the industrial robot 2 is in motion, it is possible to output a determination result of the motion corresponding to the data obtained when the industrial robot 2 is in motion based on a predetermined evaluation standard (based on an evaluation standard of a skilled person), and thus it is possible to create stable determination data of the motion of the industrial robot 2 regardless of the experience of the worker.

[0052] Apart from Figure 3 In addition to the illustrated fitting operation, the determination device 1 of this embodiment can also be applied to, for example, Figure 4 As shown in the example, the robot arm mounted on the industrial robot 2 holds the gear and assembles the gear while matching the phase of the gear with other gears. Figure 5 As shown in the example, deburring and grinding operations of the workpiece are performed by a grinding tool mounted on the industrial robot 2 instead of the robot arm.

[0053] As a variation example of the present embodiment, in addition to the force state data and position state data obtained from the industrial robot 2 and the sensor 3, various data such as the mass of the tool installed at the front end of the robot, the gap distance between the workpiece and the tool, the friction coefficient, the temperature, and the force sensor characteristics may be added to the state data S for learning by the learning unit 110 to perform machine learning.

[0054] Figure 6 This is a schematic functional block diagram of the determination device 1 and the machine learning device 100 of the second embodiment. The determination device 1 of this embodiment has a structure (estimation mode) required when the machine learning device 100 performs estimation. Figure 1 The CPU 11 of the determination device 1 shown in the figure and the processor 101 of the machine learning device 100 execute the respective system programs to control the operations of the respective parts of the determination device 1 and the machine learning device 100 to realize Figure 6 The functional modules shown.

[0055] The determination device 1 of this embodiment includes a control unit 32, a data acquisition unit 34, and a preprocessing unit 38. The machine learning device 100 included in the determination device 1 includes a determination unit 120. In addition, the non-volatile memory 14 is provided with an acquisition data storage unit 54 for storing acquisition data used by the machine learning device 100 to determine the action of the industrial robot 2. The non-volatile memory 104 of the machine learning device 100 is provided with a learning model storage unit 130 for storing a learning model constructed by machine learning of the learning unit 110.

[0056] The data acquisition unit 34 of this embodiment stores the acquired data acquired from the industrial robot 2, the sensor 3, and the input device 71 in the acquired data storage unit 54 as the acquired data used for determining the action of the industrial robot 2 at the stage (determination mode) when the determination device 1 performs machine learning of the machine learning device 100.

[0057] In addition, the control unit 32 of this embodiment has the same function as the control unit 32 of the first embodiment.

[0058] The preprocessing unit 38 of the present embodiment creates state data S used by the machine learning device 100 to determine the motion of the industrial robot 2 based on the acquired data acquired by the data acquisition unit 34 and stored in the acquired data storage unit 54. The preprocessing unit 38 creates data by, for example, converting the acquired data acquired by the data acquisition unit 34 into a unified format (digitalization, sampling, etc.) processed in the machine learning device 100.

[0059] The determination unit 120 determines the motion of the industrial robot 2 using the learning model stored in the learning model storage unit 130 based on the state data S generated by the preprocessing unit 38. In the determination unit 120 of this embodiment, the state data S input from the preprocessing unit 38 is input to the learning model generated (parameters determined) by the learning unit 110, thereby determining the success / failure, degree of success, degree of failure, etc. of the motion of the industrial robot 2 and outputting them. The result determined by the determination unit 120 can be displayed and output on the display device 70, or sent and output to a host computer, a cloud computer, etc. via a wired / wireless network (not shown) and used.

[0060] In the determination device 1 having the above-described structure, the determination unit 120 determines the motion of the industrial robot 2 using the acquired data acquired from the industrial robot 2. The learning model stored in the learning model storage unit 130 is generated by learning the determination data generated according to a predetermined criterion based on the evaluation function generated by the evaluation function generating unit 36, so that the determination unit 120 using the learning model can determine the motion of the industrial robot 2 with high accuracy based on the predetermined criterion.

[0061] Figure 7 A system 170 of a third embodiment including a determination device 1 is shown. The system 170 includes at least one machine learning device 100 installed on a computer 5 such as a cell computer, a fog computer, a host computer, or a cloud server, at least one determination device 1 implemented as a control device (edge ​​computer) for controlling an industrial robot 2, and a wired / wireless network 172 that connects the computer 5 and the determination device 1 to each other.

[0062] The determination device 1 of this embodiment does not include the preprocessing unit 38 and the machine learning device 100, but is configured on a computer 5 connected via a network, which is different from the determination device 1 of the first and second embodiments.

[0063] In the system 170 having the above-mentioned structure, the computer 5 equipped with the machine learning device 100 collects the acquisition data obtained when the industrial robot 2 is in motion and the judgment data (produced by the judgment data production unit 37) from each judgment device 1 via the network 172, performs machine learning using the collected acquisition data and judgment data, and generates a learning model (equivalent to the action mode of the first embodiment).

[0064] In addition, the computer 5 equipped with the machine learning device 100 receives the data obtained when each industrial robot 2 is in motion from each judgment device 1 via the network 172, judges the action of each industrial robot 2, and returns the judgment result (equivalent to the action mode of the second embodiment) to the judgment device 1.

[0065] In the system of the present embodiment, for example, the machine learning device 100 is configured on a computer 5 serving as a fog computer that is provided for a plurality of industrial robots 2 (determination devices 1) serving as edge computers, so that the actions of each industrial robot 2 (determination device 1) can be determined centrally on the computer 5 without the need to prepare a machine learning device 100 for each determination device 1, thereby reducing the operating cost of the machine learning device 100.

[0066] Figure 8 yes Figure 7 The schematic hardware structure diagram of the computer 5 is shown.

[0067] The CPU 511 included in the computer 5 is a processor that controls the entire computer 5. The CPU 511 reads a system program stored in the ROM 512 via the bus 520 and controls the entire computer 5 according to the system program. The RAM 513 temporarily stores temporary calculation data, various data input by the operator via the input device 531, and the like.

[0068] The nonvolatile memory 514 is composed of, for example, a memory backed up by a battery (not shown), a solid-state memory (SSD), etc., and maintains the storage state even when the power supply of the computer 5 is turned off. In the nonvolatile memory 514, a setting area for storing setting information related to the operation of the computer 5, data input from the input device 531, data obtained from each determination device 1, data read via an external storage device (not shown) or a network, etc. are stored. When executed / used, the program and various data stored in the nonvolatile memory 514 are expanded in the RAM 513. In addition, in the ROM 512, system programs including well-known analysis programs for analyzing various data are pre-written.

[0069] The computer 5 is connected to the network 172 via the interface 516. The network 172 connects at least one determination device 1, other computers, etc., and exchanges data with the computer 5.

[0070] The data read into the memory, the data obtained as a result of executing the program, etc. are output and displayed to the display device 530 via the interface 517. In addition, the input device 531 composed of a keyboard, a pointing device, etc. transmits instructions and data based on the operation of the staff to the CPU 511 via the interface 518.

[0071] In addition, the machine learning device 100 has the same Figure 1 The same hardware structure as described in .

[0072] Fig. 9 A system 170 of a fourth embodiment is shown, which includes a determination device 1. The system 170 includes a determination device 1 installed on a computer such as a cell computer, a fog computer, a host computer, or a cloud server, a control device 6 (edge ​​computer) for controlling an industrial robot 2, and a wired / wireless network 172 that connects the determination device 1 and the control device 6 to each other.

[0073] In the system 170 having the above structure, the determination device 1 collects the acquired data obtained when the industrial robot 2 is in motion from each control device 6 via the network 172, and uses the collected acquired data to create an evaluation function common to the motions of the plurality of industrial robots 2. In addition, the determination device 1 collects the acquired data obtained when the industrial robot 2 is in motion from each control device 6 via the network 172, performs machine learning using the collected acquired data, and generates a learning model (equivalent to the motion pattern of the first embodiment).

[0074] The determination device 1 receives data obtained when each industrial robot 2 operates from each control device 6 via the network 172 , determines the operation of each industrial robot 2 , and returns the determination result (equivalent to the operation mode of the second embodiment) to the control device 6 .

[0075] In the system of the present embodiment, for example, a determination device 1 serving as a fog computer is configured for a plurality of industrial robots 2 (control devices 6) serving as edge computers, so that the actions of each industrial robot 2 (control device 6) can be determined centrally on the determination device 1 without the need to prepare a machine learning device 100 for each industrial robot 2, thereby reducing the operating cost of the machine learning device 100.

[0076] Although the embodiment of the present invention has been described above, the present invention is not limited to the examples of the above-described embodiment, and can be implemented in various forms by adding appropriate changes.

[0077] For example, in the above embodiment, the determination device 1 and the machine learning device 100 are described as devices having different CPUs (processors), but the machine learning device 100 can also be implemented by the CPU 11 of the determination device 1 and the system program stored in the ROM 12.

Claims

1. A determination device for determining the motion of an industrial robot having a function of detecting a force and a torque applied to a robot arm, It is characterized in that The determination device comprises: a data acquisition unit that acquires at least the state of force and torque applied to the robot hand during the operation of force control of the industrial robot, and the state of position and posture as acquired data; An evaluation function generating unit that generates an evaluation function for evaluating the quality of the action based on the acquired data; a determination data generating unit that generates determination data for the acquired data using the evaluation function generated by the evaluation function generating unit; A preprocessing unit that generates state data for machine learning based on the acquired data; and a learning unit that generates a learning model for determining whether the action of the industrial robot is good or bad by using the state data and the determination data; The evaluation function creation unit creates the evaluation function by determining a weighting coefficient and a threshold value.

2. The determination device according to claim 1, It is characterized in that The data acquisition unit generates temporary judgment data for judging the quality of the action based on the time taken for the action. The evaluation function creation unit creates the evaluation function based on the provisional determination data.

Citation Information

Patent Citations

  • Control device

    JP2007237312A

  • Mechanical arm movement planning method based on multi-population particle swarm algorithm

    CN108748160A

  • Mirror image vision based system for achieving robot dance posture autonomous judgment

    CN109048930A

  • Part assembling device

    JP1999188680A

  • Mobile robot, learning system for the same, and method of learning action of the same

    JP2012139798A