DE112019007404T5 - PARAMETER CALCULATION DEVICE, ROBOT CONTROL SYSTEM AND ROBOT SYSTEM - Google Patents

Abstract

A parameter calculation device (100) includes: a parameter storage unit (110) that stores parameters for controlling movements of a robot for each control section with a speed of zero as a limit; a state variable acquisition unit (120) that acquires state variables indicative of a state of the robot in a series of movements, made up of a plurality of control sections, based on the parameters; an evaluation unit (130) that calculates an evaluation value for evaluating the series of movements and a cycle time of the series of movements based on the state variable; and a machine learning unit (140) that calculates parameters that cause the evaluation value to satisfy a first condition and the cycle time to satisfy a second condition through learning using multi-criteria optimization, and causes the parameter storage unit (110) to store the calculated parameters to save.

Description

area

The present invention relates to a parameter calculation device that calculates a parameter for controlling a movement of a robot, and a robot control system and robot system involved therein.

background

In recent years, industrial robots have been required to increase operation speed according to their intended work in order to improve productivity. However, when an actuator of a propeller shaft of a robot is repeatedly operated at its highest possible speed, components installed in the robot may be consumed so excessively that the design life cannot be maintained depending on the size of the load. For this reason, it is desirable for a controller designed to control the robot to monitor a load torque or the like applied to the joint shaft of the robot and a speed or acceleration parameter to make one to be output to the actuator set motion command.

The life estimation simulation device described in Patent Literature 1 decreases a speed command value when the value of the load on a reduction gear obtained by simulating robot motions is larger than a load reference value required to meet the target life.

List of citations

patent literature

Patent Literature 1: Japanese Patent Application Laid-Open No. 2013-144349

abstract

Technical problem

However, the above technique of Patent Literature 1 lengthens a lifetime by reducing the speed command value at a certain point in time, and does not consider a constraint condition such as a lifetime with respect to the entire series of movements of the speed reducer. Therefore, with the above technique of Patent Literature 1, it is not possible to set a movement command considering both a constraint condition such as a lifetime and a cycle time with respect to the entire series of movements. That is, with the technique of Patent Literature 1 listed above, it is not possible to make both a constraint condition for a series of motions and a condition for the cycle time of the series of motions to be satisfied with their respective specific conditions.

The present invention has been developed in view of the foregoing circumstances, and an object thereof is to provide a parameter calculation device capable of achieving overall optimization in which both a constraint condition for a series of movements and a condition for the cycle time of the series of movements are satisfied with their respective specific conditions.

the solution of the problem

In order to solve the above problems and achieve the object, the present invention provides a parameter calculation device comprising: a parameter storage unit to store one or more parameters for controlling one or more movements of a robot for each control section with a speed of zero as a limit save; a state variable acquiring unit for acquiring, based on the parameters, one or more state variables indicative of a state of the robot in a series of movements, made up of a plurality of control sections; an evaluation unit to calculate an evaluation value for evaluating the series of movements and a cycle time of the series of movements based on the state variable; and a machine learning unit for calculating one or more parameters with which the evaluation value can satisfy a first condition and the cycle time can satisfy a second condition through learning using multi-criteria optimization, and causing the parameter storage unit to store the calculated parameters .

Advantageous Effects of the Invention

The parameter calculation device according to the present invention can achieve an advantageous effect of achieving an overall optimization in which both a constraint condition for a series of movements and a condition for the cycle time of the series of movements are satisfied with their respective specific conditions.

character list

• 1 14 is a block diagram illustrating a configuration of a parameter calculation device according to a first embodiment.
• 2 14 is a flowchart illustrating a parameter calculation procedure of the parameter calculation device according to the first embodiment.
• 3 14 is a block diagram illustrating a first other exemplary configuration of the parameter calculation device according to the first embodiment.
• 4 12 is an illustration for explaining a series of movements of a robot according to the first embodiment.
• 5 FIG. 14 is a diagram illustrating an example of a velocity pattern prepared by the parameter calculation device according to the first embodiment.
• 6 is a diagram illustrating an example of a velocity pattern when a maximum velocity and an acceleration with respect to the in 5 illustrated speed patterns are lowered.
• 7 12 is a diagram for explaining differences between speed patterns prepared by the parameter calculation device according to the first embodiment and speed patterns of a comparative example.
• 8th 14 is a block diagram illustrating a second other example configuration of the parameter calculation device according to the first embodiment.
• 9 14 is a block diagram illustrating a third other example configuration of the parameter calculation device according to the first embodiment.
• 10 14 is a block diagram illustrating a fourth other example configuration of the parameter calculation device according to the first embodiment.
• 11 Fig. 14 is an illustration for explaining information displayed on an interface screen by the parameter calculation device 10 are to be displayed.
• 12 14 is a block diagram illustrating a fifth other example configuration of the parameter calculation device according to the first embodiment.
• 13 14 is a block diagram illustrating a first example configuration of a parameter calculation device according to a second embodiment.
• 14 14 is a block diagram illustrating a second example configuration of the parameter calculation device according to the second embodiment.
• 15 14 is a block diagram illustrating an example configuration of a parameter calculation device according to a third embodiment.
• 16 14 is a block diagram illustrating an example configuration of a robot system including a parameter calculation device according to a fifth embodiment.

• 17 12 is a diagram illustrating an example hardware configuration of the parameter calculation device according to each of the first to fifth embodiments.
• 18 14 is a diagram illustrating a control section before the layout of the work environment is adjusted by the parameter calculation device according to a sixth embodiment.
• 19 14 is a diagram illustrating a control section after the layout of the work environment is adjusted by the parameter calculation device according to the sixth embodiment.

Description of Embodiments

Hereinafter, a parameter calculation device, a robot control system, and a robot system according to embodiments of the present invention will be described in detail with reference to the drawings. The present invention is not necessarily limited by these embodiments.

First embodiment.

<device configuration>

1 14 is a block diagram illustrating a configuration of a parameter calculation device according to the first embodiment. The parameter calculation device 100, which is a robot movement parameter calculation device, functions as a device that calculates parameters (movement parameters) for use in operating a robot. Motion parameters include parameters that control the speed of the robot, such as maximum speed, acceleration the robot's movement and standby time, and parameters that determine the robot's movement route. The first to fifth embodiments are aimed at describing a case where movement parameters are parameters that determine the speed of the robot, and the sixth embodiment is aimed at describing a case where movement parameters are parameters that determine the movement route of the robot . The parameter calculation device 100 includes a parameter storage unit 110, a state variable acquisition unit 120, an evaluation unit 130, a machine learning unit 140, and a termination condition determination unit 150.

The present embodiment is directed to the description of a case where the robot is an arm robot, but the robot may be any type of robot. The user moves an arm portion of the robot using a remote control device called a handheld programmer, thereby teaching specific positions (teaching points) through which the arm portion should pass during a manufacturing process of a product. Alternatively, the user can teach the learning points to the arm portion of the robot by directly manipulating the arm portion of the robot or the like. In the robot, an actuator designed to operate the arm portion directly is attached to the arm portion, and an operator directly holds the actuator with his hand to move the arm portion, thereby teaching specific positions (teaching points) through which the arm section should run in the product manufacturing process. As a result, the robot moves the arm portion to the learning points one by one during the manufacturing process of the product.

The parameter storage unit 110 stores one or more parameters for use in creating a motion command for the robot for each session. The parameter storage unit 110 stores a combination of parameters obtained through learning and calculation of the machine learning unit 140 for each learning session. A learning segment is a movement segment from a learning point, i. H. a starting point of a movement of the robot, to another learning point, d. H. an end point of it, meant. The robot performs a series of movements by moving through a variety of learning stages. Note that the start point and end point of a movement can be set based on a control program (robot language) designed to control the movement of the robot. In this case, the parameter storage unit 110 stores one or more parameters for use in creating a movement command for the robot for each control section. A control section corresponds to a movement section from the start point to the end point of movement of the robot.

The state variable acquisition unit 120 reads out the parameter(s) for each session, the parameter(s) having been stored in the parameter storage unit 110 . Based on the parameter(s) set for each session, the state variable acquisition unit 120 calculates one or more state variables of each component (e.g., each joint) of the robot in a trial run of a series made up of a plurality of sessions Movements, whereby state variables are detected in the series of movements. A state variable is information that indicates the state of the robot, and varies depending on the movement of the robot. Examples of state variables include displacement, velocity, acceleration, electrical current consumption, and electrical power consumption of each joint in a series of movements.

The evaluation unit 130 calculates an evaluation value for evaluating the series of movements and the cycle time of the series of movements using the state variables detected by the state variable detection unit 120 . In the present embodiment, the evaluation value is a predicted lifetime of one or more components installed in the robot in the case of repetitions of the series of movements, an effective current (power consumption) of one or more components consumed in the series of movements, or an electric power (power consumption) of one or more components consumed in the series of movements.

Using the one or more state variables detected by the state variable detection unit 120, the evaluation unit 130 calculates the time required for the trial run of the series of movements as a cycle time. The evaluation unit 130 further calculates, as an evaluation value, at least one of the predicted lifetime associated with the series of movements, the power consumption due to the consumption of electric power in the series of movements, and the power consumption due to the consumption of electric power in the series of movements.

The evaluation unit 130 evaluates the evaluation value by determining whether or not the evaluation value satisfies a constraint condition. The constraint condition means a condition that is set based on at least one of design life, an upper limit of power consumption in the series of movements, and an upper limit of power consumption in the series of movements of one or more components of the robot. Design life is a life that can be freely set by the user and differs from a nominal life (described below). In a case where the evaluation value is the predicted lifetime associated with the series of movements, the evaluation unit 130 uses the constraint condition set based on the design lifetime. In a case where the evaluation value is the power consumption in the series of movements, the evaluation unit 130 uses the restriction condition that is set based on the upper limit of the power consumption in the series of movements. In a case where the evaluation value is the power consumption in the series of movements, the evaluation unit 130 uses the constraint condition that is set based on the upper limit of the power consumption in the series of movements.

For example, the evaluation unit 130 uses the design life, the power consumption upper limit, or the power consumption upper limit as the constraint condition to judge whether or not the evaluation value satisfies the constraint condition. Alternatively, the evaluation unit 130 may use a specific percentage (e.g. 90%) of the design lifetime, a specific percentage of the power consumption cap, or a specific percentage of the power consumption cap to evaluate whether the evaluation value satisfies the constraint condition or not. In short, the constraint condition may be a specific percentage of the design lifetime, a specific percentage of the power consumption cap, or a specific percentage of the power consumption cap. The evaluation unit 130 sends the result of determination of whether or not the constraint condition is satisfied, the cycle time, the constraint condition, and the parameters used for the evaluation to the machine learning unit 140 .

The machine learning unit 140 stores the result of determination of whether or not the constraint condition is satisfied, the cycle time, and the constraint condition in advance. The machine learning unit 140 searches for a combination of parameters such that the cycle time calculated by the evaluation unit 130 is minimized to the extent that the constraint condition is satisfied on the entire series of movements. In the present embodiment, making the evaluation value satisfy the constraint condition corresponds to making the evaluation value satisfy a first condition, and shortening the cycle time corresponds to making the cycle time satisfy a second condition. The machine learning unit 140 updates the values of parameters stored in the parameter storage unit 110 to the values of parameters (combination of parameters) such that the cycle time is the smallest to be found in the search.

The completion condition determination unit 150 determines whether or not the completion condition is satisfied based on the completion condition of the parameter setting (parameter calculation). The termination condition of the parameter setting is e.g. B. that the number of searches for a combination of parameters in the machine learning unit 140 should reach a specified maximum number of searches. Alternatively, the termination condition of the parameter adjustment may be that the cycle time to be minimized has a sufficiently small reduction, such as less than 10 ^-6 seconds, from the minimum value of cycle time obtained up to the previous machine learning. In the present embodiment, parameters in the case where the termination condition is satisfied are referred to as optimal parameters, and the cycle time in the case where the termination condition is satisfied is referred to as minimum cycle time. That is, parameters are referred to as optimal parameters in the case where the termination condition is satisfied during the process of improving the evaluation value. In addition, the cycle time in the case where the termination condition is satisfied during the process of improving the evaluation value is referred to as the minimum cycle time.

In the case where the ending condition is satisfied, the ending condition determining unit 150 outputs the combination of parameters such that the cycle time obtained so far is minimized to an external device as a combination of the optimal parameters. An example of the external device is a controller designed to control the robot.

In the case where the termination condition is not satisfied, the termination condition determination unit 150 notifies the state variable detection unit 120 that the termination condition has not been satisfied. In this case, the parameter calculation device 100 recalculates parameters such that the cycle time is minimized.

<Repetitive processing of machine learning unit>

2 14 is a flowchart illustrating a parameter calculation procedure performed by the parameter calculation device according to the first embodiment. The user inputs learning point information (positions of one or more learning points) into the parameter calculation device 100 to determine the number of parameters to be set. The parameter calculation device 100 receives the learning point information (step S101). Thus, the parameter calculation device 100 receives the learning point information and stores the learning point information in the parameter storage unit 110.

The parameter storage unit 110 stores one or more initial values of a combination of parameters to be optimized in the series of movements (step S102). The initial values of a combination of parameters to be stored in the parameter storage unit 110 may be input to the parameter calculation device 100 by the user or may be received by the parameter calculation device 100 from another device. Note that either step S101 or step S102 may be executed first.

Next, the state variable acquisition unit 120 reads out the combination of parameters stored in the parameter storage unit 110 and acquires one or more state variables of the series of movements (step S103). That is, based on the parameter stored in the parameter storage unit 110, the state variable acquisition unit 120 acquires the state variable of each joint of the robot in a trial run of the series of movements. The present embodiment is directed to the description of a case where the state variables are the displacement, velocity, and acceleration of each joint.

Next, the evaluation unit 130 calculates a cycle time and an evaluation value (step S104). Specifically, the evaluation unit 130 calculates at least one of the predicted lifetime, the power consumption, and the power consumption as an evaluation value, and calculates a cycle time. In the present embodiment, since the state variables are the displacement, velocity, and acceleration of each joint, the evaluation unit 130 calculates the predicted life as an evaluation value.

The evaluation unit 130 evaluates the evaluation value by determining whether or not the evaluation value satisfies the constraint condition. Next, the evaluation unit 130 determines whether or not the evaluation value satisfies the constraint condition (step S105). The evaluation unit 130 also sends the machine learning unit 140 the result of determination of whether or not the constraint condition is satisfied, the cycle time, the constraint condition, and the parameters used for evaluation.

In response to determining that the constraint condition is satisfied (step S105: Yes), the machine learning unit 140 sets, as a candidate for a combination of the optimal parameters, a combination by which the cycle time is reduced among the combinations of parameters that the have met the constraint condition so far will be the shortest. Hereinafter, a candidate for the optimal parameters at the time of setting, which is set by the machine learning unit 140, is referred to as an optimal candidate parameter. An optimal candidate parameter is obtained by improving a parameter such that the cycle time is shortened. The cycle time obtained with the optimal candidate parameter is a minimum candidate cycle time. The machine learning unit 140 updates the values of parameters stored in the parameter storage unit 110 to the set combination of the optimal candidate parameters (step S106).

In this way, the machine learning unit 140 through learning calculates parameters that allow the evaluation value as a first evaluation item and the cycle time as a second evaluation item to satisfy their respective specific conditions. In the first embodiment, the specific condition to be satisfied by the evaluation value is that the constraint condition is to be satisfied, and the specific condition to be satisfied by the cycle time is that the cycle time is to be improved (shortened).

In response to determining that the constraint condition is not satisfied (step S105: No), the unit 140 creates the machine ll learn a new combination of parameters according to an optimization algorithm (multi-criteria optimization). Examples of the optimization algorithm are particle swarm optimization, Bayesian optimization or the like (described further below). The machine learning unit 140 sets the created new combination of parameters in the parameter storage unit 110 (step S107).

After the machine learning unit 140 updates the values of parameters stored in the parameter storage unit 110 to the set combination of optimal candidate parameters, the termination condition determination unit 150 determines whether or not the termination condition is satisfied (step S108). In response to the determination that the ending condition is satisfied (step S108: Yes), the parameter calculation device 100 ends the optimal parameter search processing. In response to the determination that the termination condition is not satisfied (step S108: No), the machine learning unit 140 executes a process of step S107. That is, the machine learning unit 140 sets in the parameter storage unit 110 a new combination of parameters made according to the optimization algorithm.

After the process of step S107, the parameter calculation device 100 returns to the process of step S103 and repeats the process of steps S103 to S108 until the termination condition is satisfied.

<Device configuration for using a simulator>

3 14 is a block diagram illustrating a first other exemplary configuration of the parameter calculation device according to the first embodiment. 3 12 shows the configuration of a parameter calculation device 100A in the case where detection of a state variable is implemented using a simulator. components in 3 , which has the same functions as those of in 1 parameter calculation apparatus 100 illustrated are denoted by the same reference numerals and redundant descriptions thereof are omitted.

The parameter calculation device 100A, which corresponds to the first other example configuration of the parameter calculation device 100, includes a state variable acquisition unit 120A instead of the state variable acquisition unit 120. The parameter calculation device 100A acquires a state variable using a robot controller and a simulator that simulates a kinetic model of the robot. That is, the state variable acquisition unit 120A acquires a state variable by trial running a series of movements using a robot controller and a simulator that simulate a kinetic model of the robot.

The state variable detection unit 120</b>A includes a command creating unit 121 and a movement probing unit 122 . The movement testing unit 122 is connected to the command generation unit 121 and the evaluation unit 130 .

The command creation unit 121 reads one or more parameters stored in the parameter storage unit 110 and creates a movement command value for the robot.

The motion probing unit 122 performs simulation based on the motion command values for the robot using a simulator that simulates a kinetic model of the robot, and detects state variables of joints of the robot in the series of motions from the simulation result.

In the simulation based on the movement command values, the movement probing unit 122 may be configured to directly input the movement command values created by the command creating unit 121 into the kinetic model of the robot, or may be configured to perform specific processing on the ones created by the command creating unit 121 execute motion command values and then input results of execution therein. In this case, the motion testing unit 122 returns e.g. B. passes the movement command values prepared by the command generation unit 121 through at least one of a filter that shapes movement command values, a filter that simulates a feedforward control system, and a filter that simulates a feedback control system, and inputs the resultant values of the filtering into the model.

Although himself 3 directed to a description of an example in which values output from a kinetic model are detected as state variables of the joints, the motion probing unit 122 does not necessarily need to include a kinetic model of the robot. For example, in a case where the movement command value created by the command creating unit 121 is a speed of each joint of the robot or an electric Power consumed when each joint is driven, the motion probing unit 122 acquires the motion command values generated during the trial run of the series of motions as they are as state variables of the robot's speed or power consumption.

In addition, the motion probing unit 122 may perform specific processing on the motion command values created by the command creating unit 121 during the trial run of the series of motions, and capture the resultant values of the execution as state variables of the robot. In this case, the motion testing unit 122 returns e.g. B. passes the movement command values prepared by the command generation unit 121 through at least one of a filter that shapes movement command values, a filter that simulates a feedforward control system, and a filter that simulates a feedback control system, and acquires the resultant values of the filtering as state variables of the robot.

<definition of the series of movements>

4 12 is an illustration for explaining a series of movements of a robot according to the first embodiment. The robot 300 operates according to movement command values output from a robot controller 200 . The user designing the movement of the robot 300 grips the robot 300 to move a distal end 310 of the robot 300, thereby setting learning points P1, P2, and P3. The distal end 310 of the robot 300 in 4 includes a gripping mechanism for gripping a product that is an object to be conveyed.

The term "series of movements" in the present embodiment means a series or sequence of speed patterns formed from a plurality of movement sections (control sections) from a starting point to an ending point, with a zero speed being regarded as a limit. That is, “a control section” in the present embodiment is not a section obtained by dividing a speed pattern in which the speed increases from zero and returns to zero, but refers to a section in which the speed changes from zero increases and returns to zero. In a case where movements performed by the robot 300 are composed of a movement from the learning point P1 to the learning point P2, a movement from the learning point P2 to the learning point P3, and a movement from the learning point P3 to the learning point P1, these three movements are used in the present Embodiment collectively referred to as a series of movements. In this series of movements, the command generation unit 121 sets the movement section from the learning point P1 to the learning point P2 as a learning section P1-P2, sets the movement section from the learning point P2 to the learning point P3 as a learning section P2-P3, and sets the movement section from the learning point P3 to the learning point P1 as a session P3-P1.

For the learning period P1-P2, the command creating unit 121 creates a movement command for each control cycle in which the distal end 310 moves away from the learning point P1, i.e., based on the parameter(s) set for the learning period P1-P2. H. the starting point, to the learning point P2, i. H. the end point, moved.

Also, the command creation unit 121 for the learning portion P2-P3 creates a movement command for each control cycle from the learning point P2 to the learning point P3 based on the parameters set for the learning portion P2-P3. The command creating unit 121 also creates a movement command for the learning portion P3-P1 for each control cycle from the learning point P3 to the learning point P1 based on the parameters set for the learning portion P3-P1.

<Parameters to be set>

When the number of parameters to be set in one session is "m" (m is a natural number) and the number of sessions included in a series of movements is "n" (n is a natural number ), the dimension of combinations of parameters to be optimized by the machine learning unit 140 is m×n.

In the present embodiment, at least one of the maximum speed of the distal end 310, the acceleration of the distal end 310, and the standby time until the next movement of the distal end 310 is used as a parameter to be set to determine a movement command value that for each control cycle in each session. In the following description, the standby time until the next movement of the distal end 310 is referred to as "delay time".

5 FIG. 14 is a diagram illustrating an example of a velocity pattern prepared by the parameter calculation device according to the first embodiment. 5 shows a speed pattern PT1 in a learning section of the robot made by the parameter calculation device 100A based on the parameters of the maximum speed and acceleration. The horizontal axis a in 5 The curve shown is time and the vertical axis thereof is the velocity of the distal end 310.

The command creation unit 121 creates a movement command value for each joint to move the in 5 illustrated velocity pattern PT1. The speed pattern PT1 is composed of an acceleration section from a time t0 to a time t1, a constant speed section from a time t1 to a time t2, and a deceleration section from a time t2 to a time t3. A period Tt from time t3 to time t4 is a delay time. In 5 a duration from time t0 to time t4 corresponds to a learning section indicated by section TA. Therefore, the speed v is zero during a period from time t3 to time t4, and the command making unit 121 can set a control section with the speed of zero taken as a limit.

The slope of the speed pattern PT1 in the acceleration section is the acceleration in the acceleration section, and the slope of the speed pattern PT1 in the deceleration section is the deceleration in the deceleration section. The command generation unit 121 determines the speed v1 in the constant speed section based on the maximum speed parameter, and determines the acceleration=v1/(t1-t0) in the acceleration section and deceleration=-v1/(t3-t2) in the deceleration section using of the acceleration parameter.

In some cases, when the actuator of each joint of the robot is operating at its maximum possible speed and acceleration, an excessively large load may be applied to the robot. In such cases, the command generation unit 121 can reduce the load in exchange for increasing the cycle time by adjusting adjustable parameters for each session to reduce the maximum speed and acceleration or to increase the deceleration time. As a result, the parameter calculation device 100A can satisfy the constraint condition of the predicted lifetime in the entire series of movements.

6 is a diagram illustrating an example of a speed pattern when the maximum speed and an acceleration with respect to the in 5 illustrated speed patterns are reduced. 6 FIG. 12 shows a speed pattern PT2 created by the parameter calculation device 100A, which reduces the maximum speed and acceleration with respect to the speed pattern PT1. The horizontal axis a in 6 The curve shown is time and the vertical axis of which is the velocity of the distal end 310.

Here it is meant that the speed pattern PT1 and the speed pattern PT2 represent their respective different speed patterns for a control section with the speed of zero as a boundary.

The speed pattern PT2 is composed of an acceleration section from time t0 to time t1a, a constant speed section from time t1a to time t2a, and a deceleration section from time t2a to time t3a. The slope of the speed pattern PT2 in the acceleration section corresponds to the acceleration in the acceleration section, and the slope of the speed pattern PT2 in the deceleration section corresponds to the deceleration in the deceleration section.

The command generation unit 121 decreases the speed in the constant speed section from v1 to v2, decreases the acceleration in the acceleration section, and decreases the deceleration in the deceleration section. Assuming that the moving route from the starting point to the ending point of the session is fixed, the amount of movement of each shaft of the robot is fixed. Therefore, the command generation unit 121 prepares the speed pattern PT2 such that the area of a trapezoid drawn by the speed pattern PT2 from time t0 to time t3a is equal to the area of a trapezoid drawn by the speed pattern PT1 from time t0 to time t3 .

Therefore, the machine learning unit 140 can calculate the difference amount (t3a - t3) > 0 in the cycle time resulting from reducing the maximum speed and acceleration from the parameters of the maximum speed (speed v2), acceleration, deceleration and deceleration time, the were set for the speed pattern PT2. Note that in a case where the absolute value of acceleration and the absolute value of deceleration are the same, the machine learning unit 140 calculates the difference amount in the cycle time using a Calculate acceleration instead of deceleration or calculate the difference in cycle time using deceleration instead of acceleration.

Note that the command generation unit 121 may set the speed patterns PT1 and PT2 using parameters other than the parameters for setting the maximum speed and acceleration values. For example, the command generation unit 121 may set the speed patterns PT1 and PT2 using a parameter for setting a relaxation rate with respect to an upper limit of the maximum speed and a parameter for setting a relaxation rate with respect to an upper limit of the acceleration.

For example, the command generation unit 121 may be configured to set the relaxation rate d for the maximum speed of a joint having a maximum speed v_max within the range from zero to one and set the speed patterns PT1 and PT2 such that the set maximum speed (v_max) × ( 1 - d) is. At this time, with respect to v_max, the maximum speed upper limit designed in advance for each joint can be applied as v_max. The upper limit of the maximum speed designed in advance for each joint is a value determined depending on the capability of the actuator of the motor operating the joint. Alternatively, with respect to v_max, the maximum speed achievable in each session can be applied as v_max.

The evaluation unit 130 in the present embodiment evaluates the evaluation values for the entire series of movements as a whole, the evaluation values being calculated based on state variables detected by the state variable detection unit 120A. In addition, lessons included in the series of moves where the parameters are adjusted to minimize the cycle time under the constraint condition can be specified by the user using a robotic language. The user can specify some portions of the movements that make up the work given to the robot and cause the parameter calculation device 100A to optimize the parameters. In this case, the user specifies in a program generated in a robot language the start and end point of the section in which parameter optimization is performed.

7 12 is a diagram for explaining differences between speed patterns prepared by the parameter calculation device according to the first embodiment and speed patterns of a comparative example. 7 is provided to explain that under the constraint that the design lifetime should be met, an evaluation for a series of moves can yield a solution with a lower cycle time than an evaluation for one move in each session.

The horizontal axis a in 7 The curve shown is time and the vertical axis of which is the velocity of the distal end 310. A curve in FIG 7 The upper part illustrated represents the speed patterns of the comparative example and another curve in FIG 7 The lower part illustrated represents the speed patterns prepared by the parameter calculation device 100A. Also in this case, learning sections (control sections) are set with the speed of zero as a boundary.

In the case of an evaluation for a movement in each session, velocity patterns are generated such that the design lifetime is satisfied in each session. For example, in a case where the design life is 10,000 hours, the speed patterns are generated such that each of the learning periods P1-P2, P2-P3, and P3-P1 achieves a predicted life longer than 10,000 hours. 7 FIG. 12 shows a case where the speed patterns of the comparative example are made such that each of the learning periods P1-P2, P2-P3, and P3-P1 achieves a predicted lifetime of 11,000 hours.

In the case where this evaluation unit 130 is evaluated for a series of movements, even if the predicted lifetime associated with repetitions of a partial movement (movement in some of the learning sections) is less than the design lifetime, a combination of parameters , with which the design lifetime can be satisfied by repetitions of the entire series of movements, can be recognized as a candidate for an optimal solution. This is because even if a specific learning segment has a predicted lifetime that is less than the design lifetime, the predicted lifetime associated with the entire series of movements can meet the design lifetime when other segments have predicted lifetimes that are equal to or greater than than the design lifetime. Therefore, the evaluation unit 130 can make a solution obtained with a low cycle time to the extent that the design life is satisfied by balance sections (learning sections) that contribute to cycle time reduction in high-speed movement and sections (learning sections) that contribute to life improvement in low-speed movement.

7 12 shows a case where the evaluation unit 130 prepares speed patterns such that the learning periods P1-P2, P2-P3, and P3-P1 reach the predicted lives of 11,000 hours, 15,000 hours, and 7,000 hours, respectively. In this case, the predicted lifetime associated with the series of movements is 11,000 hours, which is equal to the predicted lifetime in the speed patterns of the comparative example, but since the predicted lifetime is evaluated by the series of movements, the cycle time may be less than that of the comparative example .

<Device configuration in case of using a predicted lifetime as a constraint condition>

8th 14 is a block diagram illustrating a second other example configuration of the parameter calculation device according to the first embodiment. 8th 12 shows the configuration of the parameter calculation device 100Bb configured to set a constraint condition based on a design lifetime. Here, the parameter calculation device 100B is used to set a constraint condition based on a design lifetime with respect to the FIG 3 Illustrated parameter calculation device 100A will be described. components in 8th , which has the same functions as those of in 3 The parameter calculation device 100A illustrated are denoted by the same reference numerals, and redundant descriptions thereof are omitted.

The parameter calculation device 100B as the second other example configuration of the parameter calculation device 100 includes an evaluation unit 130B instead of the evaluation unit 130. The evaluation unit 130B includes a kinetics calculation unit 131, a damage amount calculation unit 132, and a restriction determination unit 133a. The kinetic calculation unit 131 is connected to the movement probing unit 122 and the damage amount calculation unit 132 . The damage amount calculation unit 132 is connected to the restriction determination unit 133a. The constraint determination unit 133a is connected to the machine learning unit 140 .

The kinetic calculation unit 131 uses the values of the displacement, velocity, and acceleration of each joint detected by the state variable detection unit 120A to calculate a load torque to be applied to the joint. The damage amount calculation unit 132 calculates a damage amount of the robot's joint based on the load torque and the movement degree of each joint, to thereby calculate the predicted life of the robot associated with repetitions of the series of movements. The degree of movement of each joint is a rotational speed or a linear movement distance obtained by converting a speed based on the structure of the joint. The degree of damage of a component (part) of the robot is the degree of damage of the component of the robot. The degree of damage to a component of the robot increases with increasing load torque and degree of movement of each joint. A detailed definition of the degree of damage is described below.

The constraint determination unit 133a determines whether or not a combination of parameters read from the parameter storage unit 110 satisfies the constraint condition for the series of movements using a determination expression based on the predicted life calculated by the damage amount calculation unit 132 and the design life. An example of the determination expression is an expression (12) described below.

<Kinetic Calculation>

Each joint of the robot is made up of two or more components. A joint whose component is rotatable is called a rotary shaft, and a joint whose component is linearly movable is called a linear motion shaft. The kinetic calculation unit 131 included in the evaluation unit 130B calculates the magnitude T_i[k] of load torque applied to each joint and the input speed n_i[k] of a component of the robot in each control cycle Δt (unit: second) of the robot. Specifically, the kinetic calculation unit 131 converts the displacement, velocity, and acceleration of each joint detected by the motion sample unit 122 into displacement, velocity, and acceleration on the output side of a motive power transmission mechanism of each joint, which is a subset tion gear, a ball screw and the like includes. Then, using the values obtained by the conversion, the kinetic calculation unit 131 calculates the magnitude T_i[k] of load torque applied to each joint and the input rotational speed n_i[k] of a component of the robot. The subscript "i" of T_i and n_i indicates that the value in question refers to the ith (i is a natural number) axis of the robot. In addition, [k] indicates that the value in question is calculated in the k-th (k is a natural number) control cycle Δt, counting the number from the time the counter in a damage measure calculation cycle (described below) is reset.

<Definition of and calculation method for degree of damage>

In the present embodiment, the degree of damage of the robot is defined as follows to use the predicted lifetime associated with repetitions of a series of movements as an evaluation value. A damage measure is used as an index of how much fatigue accumulates in the current operating condition, assuming that reaching “end of life” means that a component of the robot experiences a repetitive motion fatigue failure. In the present embodiment, the damage amount calculation unit 132 calculates the damage amount of a component q using an expression of (past cumulative value of driving time or cumulative value of rotational speed of component q) / (rated life or rated life-equivalent rotational speed of component q, under the current operating conditions is required). The rated life-equivalent rotational speed is a rotational speed equivalent to the rated life.

Since a calculation expression for the rated life or rated life-equivalent rotational speed varies depending on a kind of the component, a calculation expression for a damage amount is set for each kind of the components in the damage amount calculation unit 132 . By defining the degree of damage in the parameter calculator 100B, it is possible to use the same criteria to evaluate the predicted lives of two or more components that have different life ratings from manufacturer to manufacturer.

In the present embodiment, the damage amount calculation unit 132 calculates a damage amount for each session and sets the sum of the damage amounts in all the sessions as a damage amount associated with the series of movements. Moreover, the damage amount calculation unit 132 may calculate the damage amount associated with the series of movements using the point in time when the entire series of movements is completed as a damage amount calculation cycle, which is a damage amount calculation cycle.

For example, it is considered that the damage amount calculated in the damage amount calculation cycle is cumulated by repetitions of the same operation, and the point in time when the cumulative value reaches one is set as the predicted lifetime. Assuming that the damage amount of the robot is represented as Rlc and the cycle time is represented as tc (unit: seconds) at the time when a trial run of the series of movements ends, the damage amount calculation unit 132 can calculate the life predicted value Lc (unit : time) using the following expression (1). $$\text{LC}=\left(\text{etc}/3600\right)/\text{Rlc}$$

Assuming that the damage amount of the i-th axis component q for one execution of the series of movements is Rlc_i_q, the damage amount calculation unit 132 can calculate the value of the damage amount Rlc using the following expression (2), where the value in the calculation of the predicted lifetime of the entire robot should be used. $$\text{Rlc}=\text{Max}\_\text{i},\text{q}\left(\text{Rlc}\_\text{i}\_\text{q}\right)$$

That is, of all the axes i and components q, the damage amount calculation unit 132 uses the axis having the largest damage amount and a damage amount of the corresponding component to evaluate the damage amount of the entire robot.

The damage amount calculation unit 132 calculates the damage amount Rl_i[k]_q of the i-th axis component q based on the average load torque Tav_i[k] and the average input rotational speed Nav_i[k] calculated in each specified damage amount calculation cycle t1[k]=k×Δt ].

$$\begin{array}{l}\text{Nacc}\_\text{i}\left[\text{k}\right]=\text{Nacc}\_\text{i}\left[\text{k}-1\right]+\text{n}\_\text{i}\left[\text{k}\right]\times \Delta \text{t},\\ \text{Nacc}\_\text{i}\left[0\right]=0\end{array}$$

$$\text{tb}\_\text{i}\left[\text{k}\right]=\text{tb}\_\text{i}\left[\text{k}-1\right]+\Delta \text{t},\text{tb}\_\text{i}\left[0\right]=0$$

The damage amount calculation unit 132 may calculate the r-th (r is a multiplier used in varies depending on the type of component q) cumulative value of output torque Tacc_i[k], cumulative value of input rotational speed Nacc_i[k], and cumulative value of driving time tb_i[k] (unit: second) up to the k-th Δt, respectively Calculate using the following expressions (3) to (5). The rth cumulative value of output torque Tacc_i[k] is a cumulative value of torque used to drive component q, with its torque multiplier being r. The input rotational speed cumulative value Nacc_i[k] is a cumulative value of the input rotational speed n_i[k] of the q component. The cumulative value of the driving time tb_i[k] is a cumulative value of the driving time of the component q. $$\begin{array}{l}\text{tacc}\_\text{i}\left[\text{k}\right]=\text{tacc}\_\text{i}\left[\text{k}-1\right]+\text{T}\_\text{i}{\left[\text{k}\right]}^{\text{right}}\times \text{n}\_\text{i}\left[\text{k}\right]\times \Delta \text{t},\\ \text{tacc}\_\text{i}\left[0\right]=0\end{array}$$

$$\begin{array}{l}\text{Nacc}\_\text{i}\left[\text{k}\right]=\text{Nacc}\_\text{i}\left[\text{k}-1\right]+\text{n}\_\text{i}\left[\text{k}\right]\times \Delta \text{t},\\ \text{Nacc}\_\text{i}\left[0\right]=0\end{array}$$

$$\text{tb}\_\text{i}\left[\text{k}\right]=\text{tb}\_\text{i}\left[\text{k}-1\right]+\Delta \text{t},\text{tb}\_\text{i}\left[0\right]=0$$

$$\text{Nav}\_\text{i}\left[\text{k}\right]=\text{Nacc}\_\text{i}\left[\text{k}\right]/\text{tb}\_\text{i}\left[\text{k}\right]$$

Using the r-th output torque cumulative value Tacc_i[k], the input rotational speed cumulative value Nacc_i[k], and the driving time cumulative value tb_i[k] represented by the following expressions (3) to (5), the damage amount calculation unit 132 calculates the average Calculate load torque Tav_i[k] and mean input rotational speed Nav_i[k] using the following expressions (6) and (7), respectively. The average load torque Tav_i[k] is an average of the load torque, and the average input rotation speed Nav_i[k] is an average of the input rotation speed n_i[k]. $$\text{Tav}\_\text{i}\left[\text{k}\right]={\left(\text{tacc}\_\text{i}\left[\text{k}\right]/\text{Nacc}\_\text{i}\left[\text{k}\right]\right)}^{\left(1/\text{right}\right)}$$

$$\text{Nav}\_\text{i}\left[\text{k}\right]=\text{Nacc}\_\text{i}\left[\text{k}\right]/\text{tb}\_\text{i}\left[\text{k}\right]$$

$$\begin{array}{l}\text{Rl}\_\text{i}\left[\text{k}\right]\_\left(\text{Untersetzungsgetriebe}\right)=\left(\text{tb\_i}\left[\text{k}\right]/3600\right)/\text{L}\\ \text{h}\_\text{i}\left[\text{k}\right]\end{array}$$

In a case where the component q is a reduction gear, the damage amount calculation unit 132 can calculate the rated life Lh_i[k] (unit: time) under actual operating conditions using the following expression (8), where the rated torque Tr_i, the rated rotational speed Nr i and the rated life L10 i (unit: time) of the reduction gear are set as constants. Moreover, the damage amount calculation unit 132 can calculate the damage amount Rl_i[k]_(reduction gear) using the following expression (9). $$\text{Lh}\_\text{i}\left[\text{k}\right]=\text{L}10\_\text{i}\times {\left(\text{Tr}\_\text{i}/\text{Tav}\_\text{i}\left[\text{k}\right]\right)}^{\text{right}}\times \left(\text{No}\_\text{i}/\text{Nav}\_\text{i}\left[\text{k}\right]\right)$$

$$\begin{array}{l}\text{Rl}\_\text{i}\left[\text{k}\right]\_\left(\text{reduction gear}\right)=\left(\text{tb\_i}\left[\text{k}\right]/3600\right)/\text{L}\\ \text{H}\_\text{i}\left[\text{k}\right]\end{array}$$

$$\text{Rl\_i}\left[\text{k}\right]\_\left(\text{Larger}\right)=\text{Nacc\_i}\left[\text{k}\right]/\text{Lr\_i}\left[\text{k}\right]$$

In a case where the component to be evaluated is a bearing, the damage amount calculation unit 132 calculates the rated life-equivalent rotational speed Lr_i under the current operating conditions using the following expression (10) with the conversion factor u_i from the average load torque Tav _i[k] to the mean load exerted at the location of the bearing, the basic basic dynamic load C i of the bearing, the load factor fw_i to take into account the magnitude of the vibrations of the machine during its operation and the multiplier r, which depends on the type of rolling element(s) (s ) varies in the camp. Moreover, the damage amount calculation unit 132 can calculate the damage amount Rl_i[k]_(bearing) using the following expression (11). The basic dynamic rated load C_i is a constant load at which the rated life associated with a specific rotational speed is obtained. $$\text{Lr\_i}\left[\text{k}\right]={\left(\text{C}\_\text{i}/\left(\text{fw}\_\text{i}\times \text{Tav\_i}\left[\text{k}\right]\times \text{and}\_\text{i}\right)\right)}^{\text{right}}\times {10}^6$$

$$\text{Rl\_i}\left[\text{k}\right]\_\left(\text{larger}\right)=\text{Nacc\_i}\left[\text{k}\right]/\text{Lr\_i}\left[\text{k}\right]$$

<fulfillment of a life constraint>

The constraint condition that the predicted lifetime Lc associated with repetitions of the series of movements is not smaller than the design lifetime Ls can be represented by the following expression (12). $$\text{LC}\ge \text{Ls}$$

The parameter calculation device 100B can use the following expression (13) as the constraint condition. In the expression (13), a value (a × Ls) obtained by multiplying the specified design life Ls by a specified constant a is used. $$\text{LC}\ge \text{a}\times \text{Ls}$$

If the predicted life calculated for the specified combination of parameters does not meet the constraint condition based on the design life, the evaluation unit 130B may impose a severe penalty on the evaluation value such that the combination of parameters that does not satisfy the constraint is not adopted in the unit 140 for machine learning.

For example, in a case where the predicted lifetime is less than the design lifetime, the evaluation unit 130B may output an evaluation value of about 100 times the cycle time to be minimized so that the combination of parameters that does not meet the constraint condition does not is output as an optimal solution at the time of completion of learning.

<Operation of the machine learning unit>

The machine learning unit 140 uses the score values calculated by the score unit 130B to search for a combination of parameters such that the cycle time is minimized to the extent that the constraint condition based on the design lifetime is satisfied. As described above, when the number of parameters to be set in one session is m and the number of sessions included in a series of movements is n, the dimension of combinations of parameters after which by the machine learning unit 140 are to be searched for, m×n. As the search method for this, the machine learning unit 140 may adopt a publicly known multi-objective optimization method, in which a continuous value or discrete values can be processed. The machine learning unit 140 may apply particle swarm optimization (PSO), Bayesian optimization, or the like.

In addition, as a publicly known reinforcement learning method, the machine learning unit 140 may also apply the actor-critic method which involves selecting a continuous value or discrete values as an action to be taken by a learning subject and calculating a rating value in response to the selected action. The machine learning unit 140 performs optimization processing until the learning completion condition in the completion condition determination unit 150 is satisfied, and updates the combination of parameters stored in the parameter storage unit 110 .

<optimization sample>

$$\text{v}\leftarrow \text{w}\times \text{v}+\text{c}1\times \text{r}1\times \left(\text{pj}-\text{p}\right)+\text{c}2\times \text{r}2\times \left(\text{pg}-\text{p}\right)$$

As an optimization example, a PSO-based method is described that works with combinations of continuous values. In a general particle swarm optimization method, a set of N (N is a natural number of two or more) particles is defined in a multidimensional space, and in each parameter search process, the position p and velocity v of each particle are determined using the following expressions (14) and (15) updated. $$\text{p}\leftarrow \text{p}+\text{v}$$

$$\text{v}\leftarrow \text{w}\times \text{v}+\text{c}1\times \text{right}1\times \left(\text{pj}-\text{p}\right)+\text{c}2\times \text{right}2\times \left(\text{pg}-\text{p}\right)$$

Where w is an inertia constant, c1 and c2 are the fraction of particles directed to positions that have higher scores within the group, and r1 and r2 are random numbers ranging from zero to one. Furthermore, pj represents the best position ever found by the particle and pg represents the best position ever found in the whole group.

In the present embodiment, a combination of parameters to be set corresponds to particle position p. The parameter calculation device 100B outputs, as an optimal solution, the combination of parameters that has reached the minimum cycle time at the time a specified number of searches T has occurred.

In a case where the PSO is applied to the machine learning unit 140, the parameter calculation device 100B performs the detection and evaluation of state variables of a series of movements for each of the N particles. The evaluation unit 130B determines whether or not the constraint condition is satisfied for each particle and re-determines the positions of the N particles using the expressions (14) and (15) such that the particle with the best evaluation value of the cycle time among the particles which satisfy the constraint condition is approximated by other particles.

<Sophistication of a setting method for a parameter search area>

[Setting a search seed using an evaluation value for each session]

The machine learning unit 140 may limit the search range for a combination of parameters to be set using the damage degree calculated by the damage degree calculation unit 132 for each learning session. In case of use the PSO, the machine learning unit 140 does not randomly set the initial value of each particle, but may set the initial values such that a lower speed and a lower acceleration are implemented in learning sections with higher damage levels according to the magnitude of the damage levels in a trial run of the series of movements which has the maximum velocity and the maximum acceleration in all sessions. As a result, the parameter calculation device 100B can easily find a combination of parameters that can effectively improve the predicted life with respect to the difference in cycle time and accordingly reduce the cycle time with a smaller number of searches.

[Selecting a session to set using a rating value for each session]

In order to reduce the search time by reducing the number of parameters to be adjusted, the machine learning unit 140 may set as an adjustment target only the movements in higher-level learning sections with higher damage degrees within the series of movements. The machine learning unit 140 may determine a target movement based on the relationship between the amount of damage of the movement and a period of time of the movement in each learning section. For example, a session with a higher damage level is a session with a larger load. Therefore, it is considered that a session with a higher degree of damage can obtain a higher benefit of durability improvement in the series of movements due to a decrease in speed or acceleration than a session with a lower degree of damage for the same period of time in motion. Therefore, the machine learning unit 140 can set as an adjustment target only a higher-rank movement having a larger ratio of the damage amount to the movement period. It should be noted that the machine learning unit 140 may determine a learning period to be set based on the relationship between the evaluation value (displacement, speed, acceleration, or the like) and the movement time length in each learning period.

Moreover, even if all the lessons are to be set, the machine learning unit 140 can set as the initial values of a combination of parameters parameters in the case where only the parameters of a higher-level lesson having a higher degree of damage in the series of movements are set . Moreover, the machine learning unit 140 can select a learning period to be set with initial values based on the relationship between the damage amount of each movement and the movement time period of each movement.

As described above, the machine learning unit 140 may set the movements in all the learning sections included in the series of movements as adjustment targets, or may set only the movement in the learning section learned by the machine learning unit 140 based on evaluation values by the evaluation unit 130B as a setting target. In addition, the machine learning unit 140 can set only the movement in the learning section specified by the user as the setting target.

9 14 is a block diagram illustrating a third other example configuration of the parameter calculation device according to the first embodiment. 9 12 shows the configuration of the parameter calculation device 100C for setting only the movement in the learning section set by the user as the setting target. components in 9 , which has the same functions as those of in 1 parameter calculation apparatus 100 illustrated are denoted by the same reference numerals and redundant descriptions thereof are omitted.

The parameter calculation device 100C as the third other exemplary configuration of the parameter calculation device 100 includes a setting section storage unit 160 among the components of the parameter calculation device 100. The setting section storage unit 160 stores the learning section to be set, which is input by the user using a programming tool. The setting section storage unit 160 stores a value corresponding to the learning section to be set.

The machine learning unit 140 reads the value stored in the setting section storage unit 160 and sets only the parameters of the learning section corresponding to the read value. This causes z. B., when it is not desired to decrease the maximum speed and acceleration in a certain session, since the parameter calculation device 100C can perform adjustment only in the given session to find a combination of parameters that the constraint condition to the series of movements fulfilled. In addition, the parameter calculation The device 100C reduces the search time by reducing the number of parameters to be set.

As described above, the parameter calculation device 100C can create a combination of parameters that can be used to prevent an increase in the cycle time while satisfying the constraint condition of the design life in the series of movements of the robot with a small number of trial runs compared to the case is met with extensive trial runs of combinations of parameters based on all parameters set for each session.

Moreover, the parameter calculation device 100</b>C may include a tool that visualizes information related to parameter setting while parameter setting is performed by the machine learning unit 140 . Information related to the parameter setting is at least one of the values of parameters newly stored in the parameter storage unit 110 , the cycle time calculated by the evaluation unit 130 , and the evaluation value of the predicted lifetime calculated by the evaluation unit 130 .

10 14 is a block diagram illustrating a fourth other example configuration of the parameter calculation device according to the first embodiment. 10 12 shows the configuration of the parameter calculation device 100D for visualizing information related to parameter setting while parameter setting is being performed. components in 10 , which has the same functions as those of in 1 parameter calculation apparatus 100 illustrated are denoted by the same reference numerals and redundant descriptions thereof are omitted. 11 Fig. 14 is an illustration for explaining information displayed on an interface screen by the parameter calculation device 10 are displayed.

The parameter calculation device 100D as the fourth other example configuration of the parameter calculation device 100 includes a learning progress visualization unit 170 among the components of the parameter calculation device 100. The learning progress visualization unit 170 is connected to the evaluation unit 130 and the machine learning unit 140. The learning progress visualization unit 170 acquires the cycle time and the predicted lifetime from the evaluation unit 130 and acquires a value of a parameter from the machine learning unit 140 .

The learning progress visualization unit 170 displays, on an interface screen 500, the information acquired by the evaluation unit 130 and the machine learning unit 140, i. H. Information related to the status of a learning progress of parameters. The interface screen 500 is a display screen for information (results to be visualized) output by the learning progress visualization unit 170 .

1. (A) Bewertungswert (wie etwa einen Bewertungswert der vorhergesagten Lebensdauer jedes Gelenks), der in Schritt S104 aus 2 berechnet wird
2. (B) Ergebnis der Bestimmung in Schritt S105 aus 2 dahingehend, ob der Bewertungswert der vorhergesagten Lebensdauer die Beschränkungsbedingung erfüllt oder nicht
3. (C) Beste Kombination von Parametern zum Zeitpunkt der Aktualisierung, die in Schritt S106 aus 2 aktualisiert wurde
4. (D) Änderung der Zykluszeit mit optimierten Parametern

For example, the learning progress visualization unit 170 outputs the following types of information (A) to (D) to the interface screen 500 as the status of the learning progress.

1. (A) Evaluation value (such as an evaluation value of the predicted life of each joint) obtained in step S104 2 is calculated
2. (B) Result of determination in step S105 2 whether or not the predicted lifetime evaluation value satisfies the constraint condition
3. (C) Best combination of parameters at the time of update selected in step S106 2 has been updated
4. (D) Change in cycle time with optimized parameters

The interface screen 500 displays the evaluation value of the series of movements of each joint, the best combination of parameters (optimal parameters), cycle time, and the like. This allows the user to monitor the learning progress and capture a parameter that greatly contributes to the evaluation value of the series of movements and the number of searches required for the machine learning unit 140 to merge them into a solution.

In addition, the parameter calculation device 100 may have a function of outputting a program for operating the robot based on the parameters that are results set by the machine learning unit 140 . 12 14 is a block diagram illustrating a fifth other example configuration of the parameter calculation device according to the first embodiment. 12 12 shows the configuration of the parameter calculation device 100E for creating a program for operating the robot based on set parameters. components in 12 , which have the same funk ations like those of in 1 parameter calculation apparatus 100 illustrated are denoted by the same reference numerals and redundant descriptions thereof are omitted.

The parameter calculation device 100</b>E as the fifth other exemplary configuration of the parameter calculation device 100 includes a program creation unit 180 among the components of the parameter calculation device 100 .

The program creating unit 180 acquires an optimal combination of parameters from the completion condition determination unit 150 at the time when the completion condition determination unit 150 determines that the completion condition of the parameter setting has been satisfied. The optimal combination of parameters acquired from the ending condition determination unit 150 by the program creating unit 180 is the combination of parameters stored in the parameter storage unit 110 .

The program creation unit 180 creates a program to cause the robot to perform the series of movements with the movement command values based on the detected parameters. The program created by the program creation unit 180 is described in a programming language used by the user to operate the robot. The program creation unit 180 outputs the created program to an external device (e.g., a controller configured to control the robot). As a result, the user can get the optimal program for the series of movements of the robot.

Incidentally, a partial optimization based on a parameter adjustment on a velocity pattern cannot produce a solution that satisfies the constraint condition required for the entire series of motions involving a plurality of velocity patterns. In contrast, in the present embodiment, overall optimization based on parameter adjustment is performed on a series of motions that includes a variety of velocity patterns. Therefore, in the present embodiment, it is possible to obtain a solution that satisfies the constraint condition required for the entire series of movements and shortens the cycle time.

There is also a method for creating an operating program that satisfies evaluation criteria by reducing the target speed and the target acceleration by a predetermined amount at a specific point in time. In addition, with this method, the entire series of motions is not an object to be evaluated, and thus it is impossible to generate an optimal solution that satisfies the constraint condition required for the entire series of motions.

In addition, there is a method of evaluating the durability of a machine element by calculating the torque and moving speed acting on the machine element from the joint displacement, speed and acceleration of the robot, and uniformly reducing the speed and acceleration of the series of movements using the ratio between the target lifetime and the evaluated lifetime. With this method, it is not possible to perform an overall optimization in which the moving speed and the acceleration are adjusted at a varying rate from one control section to the next. In contrast, in the present embodiment, since the contribution to the damage amount and the like is taken into account in the series of movements for each control section, overall optimization can be achieved using the predicted lifetime calculated from the damage amount and the like as a constraint condition.

As described above, in the first embodiment, the machine learning unit 140 searches for a combination of parameters such that the cycle time of the series of movements is minimized to the extent that the constraint condition of a lifetime associated with repetitions of the series of movements is satisfied will. Therefore, the parameter calculation devices 100 and 100A to 100E can prevent an increase in the cycle time while satisfying the constraint condition on a lifetime for a whole series of movements of the robot. As a result, the robot can easily cope with variety and variable-quantity production. In addition, the robot can increase the operation speed according to its target work, thus improving productivity. As a limit the speed of zero is used so that a movement walk is completed in a control section, and in this situation if an optimization can be implemented for a series of movements formed from the cumulation of two or more from such a control section are, an overall optimization through the entire work accomplished. Accordingly, it is possible to achieve an overall optimization in which both a constraint condition for a series of moves and the cycle time for the series of moves satisfy their respective specific conditions.

Second embodiment.

Next, the second embodiment of the present invention will be described with reference to FIG 13 and 14 described. In the second embodiment, the state variable acquisition unit 120A acquires an electric current value consumed in driving each joint as a state variable, and the evaluation unit calculates, as an evaluation value, electric current consumption or electric power consumption in driving for a series of movements of the joints. The learning unit learns parameters so that the cycle time is improved while the evaluation value of current consumption or power consumption satisfies the constraint condition in the series of movements, and searches for optimal parameters so that the cycle time is minimized.

<Device configuration using power consumption as a constraint condition>

13 14 is a block diagram illustrating a first example configuration of the parameter calculation device according to the second embodiment. components in 13 , which has the same functions as those of in 8th Illustrated parameter calculation device 100B according to the first embodiment are denoted by the same reference numerals and redundant descriptions thereof are omitted.

The parameter calculation device 100F calculates electric power consumption of each component in a series of movements of the robot. An example of a component of the robot is a joint equipped with an actuator. The parameter calculation device 100F calculates the power consumption of an actuator in the series of movements of the robot as the power consumption of a component of the robot, and prepares parameters such that the restriction condition set based on the power consumption upper limit is satisfied. It should be noted that the parameter calculation device 100F may be configured to calculate the power consumption of a component or components other than an actuator that a robot is equipped with and to prepare parameters such that the constraint condition imposed on based on the upper limit for electricity consumption is met.

Compared to the parameter calculation device 100B, the parameter calculation device 100F includes an evaluation unit 130F instead of the evaluation unit 130B. The evaluation unit 130F includes a power consumption calculation unit 134 and a restriction determination unit 135a.

The power consumption calculation unit 134 is connected to the motion probing unit 122 and the constraint determination unit 135 a , and the constraint determination unit 135 a is connected to the machine learning unit 140 . The power consumption calculation unit 134 calculates the power consumption in the series of movements using the power consumption value of the actuator of each joint detected by the state variable detection unit 120A.

The power consumption calculation unit 134 calculates a power consumption for each session and sets the sum of the power consumption in the sessions as a power consumption associated with the series of movements. Furthermore, the power consumption calculation unit 134 can calculate the power consumption associated with the series of movements using the point in time at which the entire series of movements ends for one power consumption calculation cycle, i. H. a calculation cycle for a power consumption.

The power consumption calculation unit 134 calculates a value Ii[k], which is an effective value obtained by converting the current value detected by the state variable detection unit 120A in each control cycle Δt (unit: second) of the robot. The power consumption calculation unit 134 calculates the cumulative power consumption value Iacc_i[k] of the i-th axis in the power consumption calculation cycle t2[k]=k×Δt using the following expression (16). The cumulative power consumption value Iacc_i[k] is a cumulative value of power consumption when the actuator is driven. $$\begin{array}{l}\text{Iacc\_i}\left[\text{k}\right]=\text{Iacc\_i}\left[\text{k}-1\right]+\text{I\_i}\left[\text{k}\right]\times \Delta \text{t},\\ \text{Iacc\_i}\left[0\right]=0\end{array}$$

The restriction determination unit 135a determines whether a combination of parameters read out from the parameter storage unit 110 satisfies the restriction condition for the Series of motions satisfied or not, using a determination expression based on the power consumption in the series of motions calculated by the power consumption calculation unit 134 and the power consumption upper limit. Assuming that the cumulative power consumption value Iacc_i[k] of the i-th axis in the entire series of movements is Ic_i and the upper limit of the power consumption of the ith axis is Is_i, the constraint condition that the cumulative power consumption value Iacc_i[k ] of the i-th axis should not exceed the upper limit current consumption of the i-th axis can be expressed by the following expression (17). In this case, as the power consumption upper limit Is_i, the value of the power consumption upper limit required for using the robot in the actual production process can be set. Alternatively, the power consumption upper limit set by the user via an interface when the user designs the series of movements of the robot may be applied as Is_i. $$\text{IC}\_\text{i}\le \text{is}\_\text{i}$$

<Device configuration in case of using power consumption as a constraint condition>

14 14 is a block diagram illustrating a second example configuration of the parameter calculation device according to the second embodiment. components in 14 , which has the same functions as those of in 13 The parameter calculation device 100F illustrated are denoted by the same reference numerals, and redundant descriptions thereof are omitted.

The parameter calculation device 100G calculates the power consumption of the actuator of each joint in the series of movements of the robot, and sets parameters such that the constraint condition set based on the upper limit of the power consumption is satisfied. It should be noted that the parameter calculation device 100G may be configured to calculate electric power consumption of a component or components other than an actuator that a robot is equipped with and to prepare parameters such that the constraint condition that based on the power consumption upper limit is satisfied.

Compared to the parameter calculation device 100G, the parameter calculation device 100F includes an evaluation unit 130G instead of the evaluation unit 130F. The evaluation unit 130G includes a power consumption calculation unit 136 and a restriction determination unit 137a.

The power consumption calculation unit 136 is connected to the motion probing unit 122 and the constraint determination unit 137a, and the constraint determination unit 137a is connected to the machine learning unit 140. The power consumption calculation unit 136 calculates the power consumption in the series of movements using the power consumption value of the actuator of each joint detected by the state variable detection unit 120A.

The power consumption calculation unit 136 calculates the cumulative power consumption value Iacc_i[k] in the same manner as the power consumption calculation unit 134. Further, the power consumption calculation unit 136 calculates the cumulative power consumption value Wacc_i[k] using the cumulative power consumption value Iacc_i[k] calculated with Expression (16) and a Nominal voltage V i. That is, the power consumption calculation unit 136 calculates the cumulative power consumption value Wacc_i[k] using the following expression (18). The cumulative power consumption value Wacc_i[k] is a cumulative value of power consumption when the actuator is driven. $$\text{Wacc\_i}\left[\text{k}\right]=\text{Iacc\_i}\left[\text{k}\right]\times \text{V}\_\text{i}$$

The restriction determination unit 137a determines whether or not a combination of parameters read from the parameter storage unit 110 satisfies the restriction condition for the series of movements using a determination expression based on the power consumption in the series of movements calculated by the power consumption calculation unit 136 and the upper limit for power consumption is based. Assuming that the cumulative power consumption value Wacc_i[k] of the i-th axis in the entire series of movements is Wc_i and the upper limit of the power consumption of the ith axis is Ws_i, the constraint condition that the cumulative power consumption value Wacc_i[k ] of the i-th axis should not exceed the upper limit of the power consumption of the i-th axis can be expressed by the following expression (19). In this case, as the upper limit power consumption Ws_i, the value of an upper limit power consumption required for using the robot in the actual production process can be set. Alternatively, can the upper limit of power consumption set by the user via an interface when the user designs the series of movements of the robot can be applied as Ws_i. $$\text{WC}\_\text{i}\le \text{Ws}\_\text{i}$$

As described above, in the case where electric power consumption is used as the constraint condition, the parameter calculation device 100G can determine whether the constraint condition is satisfied using Expression (19) in the constraint determination unit 137a instead of Expression (17) in the constraint determination unit 135a is or not.

The machine learning unit 140 according to the present embodiment searches for a combination of parameters such that the cycle time is minimized to the extent that the current constraint condition represented by expression (17) or the power constraint condition represented by expression (19) is satisfied. That is, the operation to be performed by the machine learning unit 140 is to search for a combination of parameters that minimizes the cycle time to the extent that the current constraint condition of expression (17) or the power constraint condition of expression (19) is met, rather than search for a combination of parameters that minimizes the cycle time enough to satisfy the lifetime constraint of expression (12). The machine learning unit 140 performs the same processing as in the first embodiment except for this search processing.

The machine learning unit 140 can set a session in which to set parameters based on the cumulative power consumption value Iacc_i[k] calculated for each session by the power consumption calculation unit 134, or the cumulative value of the power consumption amount calculated for each session by the power consumption calculation unit 136 is calculated for the combination of parameters to be set. In this case, the machine learning unit 140 selects a learning portion that has a large cumulative power consumption value Iacc_i[k] or a large cumulative value of power consumption amount as a parameter setting target. Consequently, the machine learning unit 140 can limit the parameter search range.

Moreover, in order to reduce the search time by reducing the number of parameters to be set, the machine learning unit 140 can set only the movement in a higher-level learning section with higher power consumption in the series of movements as a setting target. The machine learning unit 140 may determine a target movement based on the relationship in each learning period between the current consumption or power consumption of the movement and the movement time period.

In addition, even if all the sessions are to be set, the machine learning unit 140 can set as the initial values of a combination of parameters parameters obtained in the case where only the parameters of a higher-level session with a higher current consumption or power consumption are included the series of movements. Moreover, the machine learning unit 140 can select a learning period to be set with initial values based on the relationship between the current consumption or power consumption of each movement and the movement time period of each movement.

As described above, in the second embodiment, the machine learning unit 140 searches for a combination of parameters such that the cycle time of the series of movements is minimized to the extent that the constraint condition on current consumption or power consumption in the series of movements is satisfied. Therefore, the parameter calculation device 100F can prevent an increase in the cycle time while satisfying the power consumption restriction condition for the entire series of movements of the robot. In addition, the parameter calculation device 100G can prevent an increase in the cycle time while satisfying the power consumption constraint condition for the entire series of movements of the robot.

In addition, the parameter calculation device 100F uses a specific learning section as a parameter setting target. Therefore, it is possible to create a combination of parameters that can prevent an increase in the cycle time while satisfying the constraint condition for the upper limit of power consumption in the series of movements of the robot with a smaller number of trial runs compared to the case of extensive trial runs of combinations of parameters based on all parameters set for each session.

In addition, the parameter calculation device 100G may use a combination of Para meters that can prevent an increase in cycle time while satisfying the upper limit power consumption constraint condition in the series of movements of the robot with a smaller number of trial runs compared to the case with extensive trial runs of combinations of parameters based on all parameters that set for each session is met.

Third embodiment.

Next, the third embodiment of the present invention will be explained with reference to FIG 15 described. In the third embodiment, the evaluation unit calculates, as evaluation values, two or more of the predicted lifetime, the power consumption, and the power consumption in a series of movements. The learning unit learns parameters that improve the cycle time while making the evaluation values of two or more of the predicted lifetime, power consumption, and power consumption meet the constraint conditions in the series of movements, and searches for optimal parameters that minimize the cycle time .

<Device Configuration in Case of Using Two or More of Predicted Life, Current and Power as Restriction Conditions>

15 14 is a block diagram illustrating an example configuration of the parameter calculation device according to the third embodiment. components in 15 , which achieve the same functions as those of the parameter calculation device 100B, 100F, or 100G are denoted by the same reference numerals, and redundant descriptions thereof are omitted.

The parameter calculation device 100H creates parameters such that two or more of a constraint condition based on the design life, a constraint condition based on the upper limit on power consumption, and a constraint condition based on the upper limit on power consumption in a series of movements of the robot are fulfilled.

Compared to the parameter calculation device 100B, the parameter calculation device 100H includes an evaluation unit 130H instead of the evaluation unit 130B. The evaluation unit 130H sets two or more of a design life-based restriction condition, an upper-limit power consumption-based restriction condition, and an upper-limit power consumption-based restriction condition. The evaluation unit 130H includes a restriction selection unit 138, the kinetics calculation unit 131, the damage amount calculation unit 132, the power consumption calculation unit 134, the power consumption calculation unit 136, and a restriction determination unit 139.

The constraint selection unit 138 is connected to the parameter storage unit 110 , the state variable acquisition unit 120</b>A, the kinetic calculation unit 131 , the power consumption calculation unit 134 , and the power consumption calculation unit 136 . The kinetic calculation unit 131 is connected to the damage amount calculation unit 132 . The restriction determination unit 139 is connected to the damage amount calculation unit 132 , the power consumption calculation unit 134 , the power consumption calculation unit 136 and the machine learning unit 140 .

When the constraint condition based on the design life is selected by the user, the state variable acquisition unit 120A acquires state variables (values of displacement, velocity, and acceleration of each joint) related to a predicted life. In addition, when the restriction condition based on a power consumption upper limit or a power consumption upper limit is selected by the user, the state variable acquisition unit 120A acquires state variables (power consumption value of the actuator of each joint) related to power consumption or power consumption.

The restriction selection unit 138 selects the type of restriction condition to be set from restriction conditions based on the design life, the power consumption upper limit, and the power consumption upper limit according to an instruction from the user. When the constraint condition based on the design life is selected by the user, the constraint selection unit 138 sends state variables to the kinetics calculation unit 131. Accordingly, the kinetics calculation unit 131 calculates the load torque applied to the joint and the damage amount calculation unit 132 calculates the predicted life associated with repetitions of the set of movements is associated. The damage amount calculation unit 132 sends the predicted one lifetime to the constraint determination unit 139.

In addition, when the restriction condition based on an upper limit of power consumption is selected by the user, the restriction selection unit 138 sends state variables to the power consumption calculation unit 134. Thus, the power consumption calculation unit 134 calculates the cumulative power consumption value Ic_i in the series of movements using the state variables . The power consumption calculation unit 134 sends the cumulative power consumption value Ic_i to the restriction determination unit 139.

In addition, when the restriction condition based on an upper limit of power consumption is selected by the user, the restriction selection unit 138 sends the state variables to the power consumption calculation unit 136. Accordingly, the power consumption calculation unit 136 calculates the cumulative power consumption value Wc_i in the series of movements using the state variables. The power consumption calculation unit 136 sends the cumulative power consumption value Wc_i to the restriction determination unit 139.

When the constraint condition based on a design lifetime is selected by the user, the constraint determination unit 139 determines whether or not the constraint condition is satisfied using expression (12). When the restriction condition based on an upper limit of power consumption is selected by the user, the restriction determination unit 139 determines whether or not the restriction condition is satisfied using expression (17). When the constraint condition based on an upper limit of power consumption is selected by the user, the constraint determination unit 139 determines whether or not the constraint condition is satisfied using expression (19).

When the evaluation values satisfy all of the restriction conditions selected by the user, the restriction determination unit 139 determines that the restriction conditions on the series of movements are satisfied. The machine learning unit 140 learns the parameters in the same manner as in the first and second embodiments.

As described above, in the third embodiment, the machine learning unit 140 searches for a combination of parameters that minimizes the cycle time of the series of movements to the extent that two or more of a lifetime, a power consumption, and a power consumption satisfy the constraint conditions. Therefore, the parameter calculation device 100H can prevent an increase in the cycle time while satisfying the constraint conditions of two or more of the lifespan, power consumption, and power consumption for the entire series of movements of the robot.

Fourth embodiment.

Next, the fourth embodiment of the present invention will be explained with reference to FIG 15 described. In the fourth embodiment, the parameter calculation device creates a combination of parameters that can improve the evaluation value of the damage amount, power consumption, or power consumption in a series of movements while satisfying the cycle time constraint condition. That is, the evaluation unit calculates the damage amount, power consumption, or power consumption in the series of movements as an evaluation value. The learning unit learns the parameters that improve the damage level, power consumption, or power consumption in the series of moves while making the cycle time meet the constraint condition, and searches for parameters that improve the damage level, power consumption, or power consumption be optimized in the series of movements.

The improvement in the evaluation value of the degree of damage is achieved by decreasing a degree of damage, the improvement in the evaluation value of power consumption is achieved by decreasing power consumption, and the improvement in the evaluation value of power consumption is achieved by decreasing power consumption. In the fourth embodiment, the specific condition to be satisfied by the cycle time is that the constraint condition should be satisfied (within a certain period of time), and the specific condition to be satisfied by the evaluation value is that the evaluation value should be improved. That is, improving the evaluation value corresponds to causing the evaluation value to satisfy a first condition, and making the cycle time satisfy the constraint condition corresponds to causing the cycle time to satisfy a second condition.

<Case of extending a predicted lifetime with the cycle time as a constraint condition>

In the present embodiment, any one of the parameter calculation devices 100B, 100F, and 100G is used. The parameter calculation device 100B, 100F, or 100G finds a combination of parameters that minimizes the damage amount, power consumption, or power consumption in the series of movements while satisfying the constraint condition based on the cycle time.

The constraint determining unit 133a, 135a, or 137a determines whether or not the constraint condition set based on the cycle time is satisfied, and the machine learning unit 140 finds a combination of parameters that minimizes the damage amount, power consumption, or power consumption. Minimizing the damage level is synonymous with maximizing the predicted lifetime.

Operational processing of the parameter calculation device according to the fourth embodiment will be described below. In the case of decreasing a damage amount (predicted life), the parameter calculation device 100B is used. In the case of reducing power consumption, the parameter calculation device 100F is used. In the case of reducing power consumption, the parameter calculation device 100G is used.

Note that the operation processing of the parameter calculation device 100B, 100F, or 100G in the fourth embodiment that is similar to the operation processing of the parameter calculation device 100B, 100F, or 100G in the first to third embodiments will not be described. In the fourth embodiment, the restriction determination units 133a, 135a, and 137a perform substantially the same processing, and thus the processing of the restriction determination unit 133a will be described here. Also, in the fourth embodiment, the judging units 130B, 130F, and 130G perform substantially the same processing, and thus the processing of the judging unit 130B will be described here.

In the restriction determination unit 133a, a cycle time reference value ts is set as an upper limit of the allowable deceleration for the series of movements. The cycle time reference value ts is the constraint condition of a cycle time and is the longest cycle time allowed in executing the series of moves. When the cycle time tc calculated by the evaluation unit 130B satisfies the following expression (20), the constraint determination unit 133a determines that the constraint condition of the cycle time in the series of movements is satisfied. $$\text{etc}\le \text{ts}$$

For example, the cycle time reference value ts is set to 90% of the cycle time implemented in the case where the speed and acceleration are reduced at the same rate until the design life is satisfied in each learning period. Alternatively, the cycle time reference value ts can be set to the value of a cycle time required for designing the movements of the robot in the actual production process.

The machine learning unit 140 searches for a combination of parameters that minimizes the degree of damage, power consumption, or power consumption associated with the series of movements calculated by the evaluation unit 130B to the extent that the constraint condition of expression (20) is satisfied . In other words, the machine learning unit 140 searches for a combination of parameters by which the evaluation value calculated by the evaluation unit 130B is improved to satisfy the constraint condition of the expression (20). The machine learning unit 140 updates the combination of parameters stored in the parameter storage unit 110 with the found combination of parameters.

In a case where the parameter calculation device 100B is used, the machine learning unit 140 searches for a combination of parameters that minimizes the degree of damage associated with the series of movements to the extent that the constraint condition of expression (20) is satisfied is.

In a case where the parameter calculation device 100F is used, the machine learning unit 140 searches for a combination of parameters that minimizes the power consumption associated with the series of movements to the extent that the constraint condition of expression (20) is satisfied is.

In a case where the parameter calculation device 100G is used, the machine learning unit 140 searches for a combination of parameters that reflects the power consumption associated with the series of movements is minimized to the extent that the constraint condition of expression (20) is satisfied.

Note that the processing described in the fourth embodiment can be applied to the parameter calculation device 100H.

Note that the parameter calculation device 100H according to the third embodiment can create a combination of parameters that can improve at least one of the evaluation values of the damage amount, power consumption, and power consumption in the series of movements while satisfying the cycle time constraint condition. In this case, the constraint selection unit 138 selects at least one optimization target from the damage amount, power consumption, and power consumption.

As described above, in the fourth embodiment, the machine learning unit 140 searches for a combination of parameters such that at least one of the lifespan, power consumption, and power consumption in the series of movements is improved such that the constraint condition of the cycle time that associated with repetitions of the series of movements is fulfilled. Therefore, the machine learning unit 140 can create a motion command for the series of motions, the motion command enabling the extension of the predicted lifetime, reduction of power consumption, or reduction of power consumption while meeting the cycle time allowed for the series of motions of the robot is. That is, the machine learning unit 140 can prevent the cycle time from increasing in consideration of the predicted lifetime, power consumption, or power consumption in the series of movements.

Fifth embodiment.

Next, the fifth embodiment of the present invention will be explained with reference to FIG 16 described. In the fifth embodiment, a state variable is detected from an actual robot.

<Device Configuration in Case of Acquiring State Variables from an Actual Robot>

16 14 is a block diagram illustrating an example configuration of a robot system including a parameter calculation device according to the fifth embodiment. components in 16 , which achieve the same functions as those of the parameter calculation device 100 are denoted by the same reference numerals, and redundant descriptions thereof are omitted.

The robot system 700 includes a robot control system 600 and the robot 300. The robot control system 600 is a system that controls the robot 300, and includes a parameter calculation device 1001 and the robot controller 200.

The parameter calculation device 1001 communicates with a sensor 400 connected to the actual robot 300 and acquires state variables from the robot 300. The parameter calculation device 1001 includes the parameter storage unit 110, a state variable acquisition unit 120B, the evaluation unit 130, the machine learning unit 140, and the termination condition determination unit 150

That is, compared to the parameter calculation device 100, the parameter calculation device 1001 includes the state variable acquisition unit 120B instead of the state variable acquisition unit 120.

The state variable acquisition unit 120B includes a parameter transmission unit 123 and a state variable reception unit 124. The parameter transmission unit 123 is connected to the parameter storage unit 110, the termination condition determination unit 150 and the robot controller 200, and the state variable reception unit 124 is connected to the sensor 400 and the evaluation unit 130.

The parameter transmission unit 123 transmits a parameter stored in the parameter storage unit 110 to the robot controller 200. The state variable receiving unit 124 receives from the sensor 400 a state variable obtained when the robot 300 is operated.

The robot controller 200 is a device configured to control the robot 300 using parameters calculated by the parameter calculation device 1001 . The robot controller 200 includes a command generation unit 201.

The command creation unit 201 receives one or more parameters from the parameter transmission unit 123 and creates one or more movement command values for the robot 300 based on the parameters. The command maker Processing unit 201 transmits the created motion command value to robot 300. Robot 300 performs a series of motions based on the motion command values created by command creating unit 201.

The sensor 400 detects one or more state variables of each joint of the robot 300 in the series of movements and transmits the state variables to the state variable receiving unit 124. The sensor 400 may be placed inside the robot 300 or may be placed outside the robot 300. Furthermore, the sensor 400 can be placed inside the robotic system 700 or can be placed outside of the robotic system 700 .

In the robot system 700, in response to the parameter transmission unit 123 transmitting one or more parameters stored in the parameter storage unit 110 to the command generation unit 201, the command creation unit 201 creates movement command values for the robot 300 based on the parameters. The robot 300 performs a series of movements based on the movement command values, and the sensor 400 detects one or more state variables of each joint of the robot 300 in the series of movements and transmits the state variables to the state variable receiving unit 124.

Thus, the state variable receiving unit 124 receives from the sensor 400 the state variables obtained when the robot 300 operates based on the movement command values prepared by the command preparing unit 201 . In this way, the parameter calculation device 1001 acquires one or more state variables based on the result of the operation of the actual robot 300. The state variable receiving unit 124 transmits the received state variables to the evaluating unit 130.

The evaluation unit 130, the machine learning unit 140, and the termination condition determination unit 150 perform processing similar to that described in the first embodiment. Consequently, the parameter calculation device 1001 prepares parameters that can prevent an increase in the cycle time while satisfying the design life constraint condition for the entire series of movements of the robot 300 .

Note that the parameter calculation device 100I can be placed inside the robot controller 200 . In this case, the robot controller 200 includes a learning point input unit that receives one or more learning points and inputs the learning points to the parameter calculation device 100I, and a parameter output unit that outputs one or more parameters prepared by the parameter calculation device 100I. In addition, the state variable acquisition unit 120B can be applied to the parameter calculation devices 100A to 100H.

As described above, the parameter calculation device 100I according to the fifth embodiment can find an optimal combination of parameters under the constraint condition on the series of movements using the state variables detected from the actual robot 300. FIG.

Here, the hardware configuration of the parameter calculation devices 100 and 100A to 100I will be described. 17 12 is a diagram illustrating an example hardware configuration of the parameter calculation device according to each of the first to fifth embodiments. Since the parameter calculation devices 100 and 100A to 100I have the same hardware configuration, the hardware configuration of the parameter calculation device 100 will be described here.

The parameter calculation device 100 can be configured by the in 17 illustrated control circuit, ie a processor 301 and a memory 302, can be implemented. The processor 301 is exemplified by a CPU (also referred to as a central processing unit, central processing device, processing device, computing device, microprocessor, microcomputer, processor, or DSP (digital signal processor)) or a large scale integration (LSI). Memory 302 is represented by random access memory (RAM) or read only memory (ROM).

The parameter calculation device 100 is implemented by the processor 301 reading and executing a program stored in the memory 302 for executing the operation of the parameter calculation device 100 . It can also be said that this program causes a computer to execute the procedure or method for the parameter calculation device 100 . The memory 302 is also used as a temporary storage when the processor 301 performs various types of processes.

It should be noted that some of the functions of the parameter calculation device 100 can be implemented by dedicated hardware n and some others can be implemented by software or firmware.

Sixth embodiment.

Next, an embodiment for setting parameters for determining the layout of the working environment of the robot 300 performing a series of movements will be described. Parameters for determining the layout of the work environment of the robot 300 are parameters for determining the movement route (control sections) of the robot 300. That is, the present embodiment is directed to a description of a case where movement parameters serve as parameters intended to to determine the movement route of the robot.

Any of the parameter calculation devices 100 and 100A to 100I described in the first to fifth embodiments can be used as the parameter calculation device according to the present embodiment. That is, any one of the parameter calculation devices 100 and 100A to 100I can set the parameters for determining the layout of the work environment of the robot 300.

The present embodiment is directed to the description of a case where the parameter calculation device 100 sets the parameters for determining the layout of the work environment. In particular, the parameter calculation device 100 sets the parameters for determining the layout of the working environment by using the function of any one of the parameter calculation devices 100 and 100A to 100I using the parameter storage unit 110, the state variable acquisition unit 120, the evaluation unit 130, the machine learning unit 140 and of the termination condition determination unit 150 executes.

Referring to the 18 and 19 the process of adjusting the layout of the work environment with the robot 300 and the effects of the adjustment will be described. 18 14 is a diagram illustrating control sections before the parameter calculation device according to the sixth embodiment adjusts the layout of the working environment. 19 14 is a diagram illustrating control sections after the parameter calculation device according to the sixth embodiment sets the layout of the working environment. In this example we consider the series of movements shown in the schematic representation of a layout A0 in 18 is illustrated in which the robot 300 installed at a robot base origin O1 first moves from the learning point P1, that is, the home position of the arm tip, to the learning point P2 on a parts feeder 800, then moves from the learning point P2 to the learning point P3 on a assembling table 801 and finally returns from the learning point P3 to the learning point P1.

The schematic representation of a layout A1 in 19 shows that changes from the robot base origin O1 and the learning points P1, P2 and P3 to a robot base origin O1' and the learning points P1', P2' and P3' lead to changes in the posture of the robot 300 and the degree of movement of each joint of the robot in the Lead sessions (control sessions) P1'-P2', P2'-P3' and P3'-P1' which make up the series of movements.

Changes in the installation position of the robot 300 and the arrangement of the parts feeder 800 and the assembly table 801, which are elements for determining the layout of the work environment, lead to changes in the posture of the robot 300 and the degree of movement of each joint of the robot in each of the control sections that make up the row form of movements. Therefore, this results in changing the evaluation value of the series of movements calculated in the evaluation unit 130 included in the parameter calculation device 100 according to the manner described in any one of the first to fifth embodiments. Based on this evaluation value, the machine learning unit 140 calculates parameters for determining the layout of the work environment using the manner described in any one of the first to fifth embodiments.

Examples of parameters for determining the layout of the work environment are the three-dimensional coordinates (x, y, z) of the robot base origin O1 and the learning points P1, P2, and P3. The parameter storage unit 110 stores these three-dimensional coordinates as parameters to be set. The machine learning unit 140 calculates a combination of three-dimensional coordinates using the cycle time and the evaluation value of the series of movements calculated by the evaluation unit 130, e.g. B. using PSO, in which the three-dimensional coordinate (x, y, z) of each learning point is defined as a position vector of a particle.

In order to make the robot 300 perform work on the layout of the optimal parameters output by the parameter calculation device 100, it can be used the user moves the positions of installations such as the parts feeder 800 and the mounting table 801. Alternatively, the user can move a point where the robot 300 picks up an object on an installation, such as a conveyor belt, whose position has been fixed.

The machine learning unit 140 can set the parameter search range according to the manner described in any one of the first to fifth embodiments. For example, with layout A0 in 18 the machine learning unit 140 fixes the arrangement positions of the parts feeder 800 and the assembly table 801 to set only the coordinates of the robot base origin O1 and the learning point P1 without setting the coordinates of the learning points P2 and P3.

Further, the parameter calculation device 100 according to the present embodiment can calculate parameters for each control portion of the series of movements described in any one of the first to fifth embodiments together with the parameters for determining the layout of the working environment. For example, the parameter calculation device 100 may be configured to calculate the three-dimensional coordinates of the three-dimensional coordinates included in the series of movements as parameters, and to calculate, as parameters, the speed and acceleration in the control section going to the next learning point. Alternatively, the parameter calculation device 100 may only calculate parameters for determining the layout of the work environment, without calculating parameters for each control portion of the series of movements described in any one of the first to fifth embodiments.

As described above, in the present embodiment, parameters for controlling the movements of the robot 300 inevitably include parameters for determining the layout of the work environment in which the robot 300 performs the series of movements. On the other hand, in the present embodiment, parameters for controlling the operation of the robot 300 described in any one of the first to fifth embodiments are not required to serve as parameters stored in the parameter storage unit 110 or parameters learned by the machine learning unit 140 .

As described above, according to the sixth embodiment, it is possible to determine the layout of the working environment that contributes to preventing the cycle time from increasing while satisfying the restriction condition of the series of movements.

The configurations described in the above embodiments are examples of the contents of the present invention, and can be combined with other publicly known techniques and partially omitted and/or modified without departing from the scope of the present invention.

reference list

100, 100A to 100I

parameter calculation device;

110

parameter storage unit;

120, 120A, 120B

state variable detection unit;

121, 201

command generation unit;

122

movement test unit;

123

parameter transmission unit;

124

state variable receiving unit;

130, 130B, 130F, 130G, 130H

evaluation unit;

131

kinetics calculation unit;

132

damage measure calculation unit;

133a, 135a, 137a, 139

restriction determination unit;

134

power consumption calculation unit;

136

power consumption calculation unit;

138

constraint selection unit;

140

machine learning unit;

150

termination condition determination unit;

160

setting section storage unit;

170

learning progress visualization unit;

180

programming unit;

200

robot controller;

300

Robot;

301

Processor;

302

Storage;

310

distal end;

400

Sensor;

500

interface screen;

600

robot control system;

700

robotic system;

800

parts feeder;

801

assembly table,

O1, O1'

robot base origin;

P1, P1', P2, P2', P3, P3'

learning point.

QUOTES INCLUDED IN DESCRIPTION

This list of the documents cited by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent Literature Cited

• JP 2013144349 [0004]

Claims (19)

Parameter calculation device, comprising: a parameter storage unit to store one or more parameters for controlling one or more movements of a robot for each control section with a speed of zero as a limit; a state variable acquiring unit for acquiring, based on the parameters, one or more state variables indicative of a state of the robot in a series of movements, made up of a plurality of control sections; an evaluation unit to calculate an evaluation value for evaluating the series of movements and a cycle time of the series of movements based on the state variable; and a machine learning unit to calculate one or more parameters that cause the evaluation value to meet a first condition and the cycle time to meet a second condition through learning using multi-criteria optimization, and to cause the parameter storage unit to store the calculated parameters . Parameter calculation device according to claim 1 , where the parameters include a parameter that determines a speed of the robot. Parameter calculation device according to claim 1 , wherein the parameters include a parameter that determines a layout of a work environment in which the robot performs the series of motions. Parameter calculation device according to one of Claims 1 until 3 , wherein the rating value is at least one of a predicted lifetime of a component of the robot associated with repetitions of the series of movements, a power consumption of the component in the series of movements, and a power consumption of the component in the series of movements. Parameter calculation device according to one of Claims 1 until 4 , wherein the machine learning unit calculates, through learning, one or more parameters such that the cycle time is shortened while the evaluation value satisfies a constraint condition. Parameter calculation device according to one of Claims 1 until 4 , wherein the machine learning unit calculates, through learning, one or more parameters such that the score is improved while the cycle time satisfies a constraint condition. Parameter calculation device according to claim 4 , wherein when one of the evaluation values is the predicted lifetime, the state variables include a displacement of a joint of the robot, a speed of the joint and an acceleration of the joint in the series of movements, and the evaluation unit includes a load torque applied to the joint calculated using the displacement, the velocity and the acceleration, calculated a damage amount of the component in the series of movements based on the load torque and a movement degree of the joint, and calculated the predicted life based on the damage amount. Parameter calculation device according to claim 4 , wherein when one of the evaluation values is power consumption, the state variables include a speed of a joint of the robot, an acceleration of the joint, and a current value for driving the joint in the series of movements, and the evaluation unit calculates power consumption using the current value. Parameter calculation device according to claim 4 , wherein when one of the evaluation values is power consumption, the state variables include a speed of a joint of the robot, an acceleration of the joint, and a current value for driving the joint in the series of movements, and the evaluation unit calculates the power consumption using the current value. Parameter calculation device according to one of Claims 1 until 9 , wherein the machine learning unit sets an initial value of a parameter to be learned based on an evaluation value calculated for each of the control sections. Parameter calculation device according to one of Claims 1 until 10 , wherein the machine learning unit selects a control section in which learning is to be performed based on an evaluation value calculated for each of the control sections. Parameter calculation device according to one of Claims 1 until 10 , where the unit for machine learning, calculates a ratio between a moving time and the evaluation value in the control section for each of the control sections, and selects a control section in which the learning is to be performed based on the ratio. Parameter calculation device according to one of Claims 1 until 12 , wherein the state variable acquisition unit acquires the state variables from a simulation result provided by a simulator that simulates movements of the robot. Parameter calculation device according to one of Claims 1 until 12 , wherein the state variable acquisition unit acquires the state variable based on an operational result obtained through actual operation of the robot. Parameter calculation device according to one of Claims 1 until 10 , further comprising a section storage unit for storing a control section in which the machine learning unit learns the parameter, the section storage unit storing a control section specified by a user, and the machine learning unit learning the parameter for the control section stored in the section storage unit . Parameter calculation device according to one of Claims 1 until 15 , further comprising a learning progress visualization unit to create information that visualizes at least one of a value of the parameter stored in the parameter storage unit, a cycle time calculated by the evaluation unit, and an evaluation value calculated by the evaluation unit while the machine learning unit is doing the parameters learned. Parameter calculation device according to one of Claims 1 until 16 , further comprising a program creation unit to create a program configured to operate the robot based on a parameter calculated by the machine learning unit. A robot control system comprising: the parameter calculation device according to any one of Claims 1 until 17 ; and a robot controller for controlling a movement of the robot based on a parameter calculated by the parameter calculation device. A robot system comprising: the robot control system according to Claim 18 ; and the robot controlled by the robot control system.

Images