WO2020026447A1

WO2020026447A1 - Parameter learning method and work system

Info

Publication number: WO2020026447A1
Application number: PCT/JP2018/029287
Authority: WO
Inventors: 内田　剛; 博史大池; 弘健江嵜; アヌスヤナラサンビ
Original assignee: 株式会社Ｆｕｊｉ
Priority date: 2018-08-03
Filing date: 2018-08-03
Publication date: 2020-02-06
Also published as: JP7121127B2; CN112512942B; JPWO2020026447A1; CN112512942A

Abstract

Provided is a parameter learning method for controlling an actuator so as to loosen a bulk state of a plurality of workpieces through a prescribed operation, the method including a mid-operation imaging step for imaging the plurality of workpieces during the execution of a prescribed operation, an evaluation step for processing an image captured in the mid-operation imaging step and evaluating the separation state of the plurality of workpieces, and a learning step for learning the relationship between the evaluation results in the evaluation step and a parameter in the prescribed operation during execution thereof.

Description

Parameter learning method and work system

This specification discloses a parameter learning method and a work system.

Conventionally, in a work system in which works such as electronic parts and mechanical parts are supplied to a flexible support part, and are collected by a robot and a predetermined work is performed, to unravel the bulk state of the work supplied to the support part Japanese Patent Application Laid-Open No. H11-163,087 discloses a technique in which a shock is applied to a supporting portion from below. In this system, parameters such as impact energy and additional points are variable so that the work can be properly unraveled in accordance with the weight, size, shape, material, and the like of the work. In addition, by imaging the workpiece in a stable state before and after the impact is applied, processing the captured image to recognize the separation state of the work, and continuing to learn the relationship between the changed parameter and the separation state of the work, Parameters are determined.

Japanese Patent No. 3172494

However, in the above-described working system, since the imaging of the work is performed in a stable state, it is necessary to perform the imaging in a state where the impact applied to the supporting portion is contained. In other words, it is necessary to temporarily stop the application of the impact each time the separated state of the workpiece is recognized, so that the time during which the application of the impact is stopped may be long in order to continue learning. In that case, it takes time to loosen the bulk state of the work, which leads to a decrease in work efficiency.

開示 The present disclosure has a main object of appropriately performing parameter learning without lowering work efficiency for unraveling a bulk state of works.

This disclosure employs the following means to achieve the above-mentioned main object.

A parameter learning method according to an embodiment of the present disclosure is a parameter learning method for controlling an actuator that unravels a bulk state of a plurality of works by a predetermined operation, and images the plurality of works during execution of the predetermined operation. An imaging step during operation, an evaluation step of processing an image captured in the imaging step during operation to evaluate a separation state of the plurality of workpieces, an evaluation result of the evaluation step, and an evaluation result of the predetermined operation being executed. And a learning step of learning a relationship with the parameter.

The parameter learning method of the present disclosure captures a plurality of workpieces during execution of a predetermined operation of unraveling a bulk state of a plurality of workpieces, processes the captured images to evaluate a separation state of the plurality of workpieces, The relationship between the evaluation result and the parameter in the predetermined operation being executed is learned. Thereby, the learning of the parameter can be performed using the image captured during the execution of the predetermined operation, so that it is not necessary for the actuator to interrupt the predetermined operation for learning. For this reason, the parameter can be appropriately learned without lowering the work efficiency for unraveling the bulk state of the works.

FIG. 1 is a configuration diagram illustrating a schematic configuration of a work system. FIG. 2 is a configuration diagram illustrating an outline of a configuration of a work transfer device. FIG. 3 is a partial external view of the work transfer device 20 as viewed from the back side. Explanatory drawing which shows the electrical connection relationship of the control apparatus 70. Explanatory drawing which shows a mode that the bulk state of the work W is unraveled. FIG. 3 is a block diagram showing functions of a control device 70. 9 is a flowchart illustrating an example of a loosening operation control routine. 9 is a flowchart illustrating an example of a process at the time of starting a loosening operation. Explanatory drawing which shows an example of the calculation method of difference (DELTA) A, (DELTA) G. Explanatory drawing which shows an example of area | region area A and the center of gravity G.

Next, embodiments for implementing the present disclosure will be described with reference to the drawings.

1 is a configuration diagram schematically showing the configuration of the work system 10, FIG. 2 is a configuration diagram schematically showing the configuration of the work transfer device 20, and FIG. 3 is a partial external view of the work transfer device 20 as viewed from the back. FIG. 4 is an explanatory diagram showing an electrical connection relationship of the control device 70. 1 and 2, the left-right direction is the X-axis direction, the front-rear direction is the Y-axis direction, and the up-down direction is the Z-axis direction.

The work system 10 according to the present embodiment is a system in which the work W stored in the supply box 12 is transferred to the mounting table T and aligned. As shown in FIG. 1, the work system 10 includes a mounting table transfer device 16, a work transfer device 20, a supply robot 40, and a sampling robot 50. These are installed on the workbench 11.

(4) The mounting table transport device 16 has a pair of belt conveyors that are stretched in the left-right direction (X-axis direction) with an interval in the front-rear direction (Y-axis direction). The mounting table T is transported from left to right by a belt conveyor.

The supply robot 40 is a robot for taking out the work W as various parts such as mechanical parts and electric parts from the supply box 12 and supplying the work W to the supply area A1 of the work transfer device 20 (see FIG. 2). The supply robot 40 includes a vertical articulated robot arm 41 and an end effector 42. The robot arm 41 includes a plurality of links, a plurality of joints that rotatably or pivotally connect the links, a drive motor 44 that drives each joint (see FIG. 4), and an encoder 45 that detects the angle of each joint. FIG. 4). The plurality of links include a distal link to which the end effector 42 is attached, and a proximal link fixed to the worktable 11. The end effector 42 can hold and release the work W. The end effector 42 can use, for example, a mechanical chuck, a suction nozzle, an electromagnet, or the like, and supplies the work W to the supply area A1 in a bulk state.

(4) The collection robot 50 is a robot for collecting the work W in the collection area A2 (see FIG. 2) of the work transfer device 20, transferring the work W to the mounting table T, and aligning the work W. The sampling robot 50 includes a vertical articulated robot arm 51 and an end effector 52. The robot arm 51 includes a plurality of links, a plurality of joints that rotatably or pivotally connect the links, a drive motor 54 (see FIG. 4) that drives each joint, and an encoder 55 that detects an angle of each joint. FIG. 4). The plurality of links include a distal link to which the end effector 52 is attached, and a proximal link fixed to the worktable 11. The end effector 52 can hold and release the work W. As the end effector 52, for example, a mechanical chuck, a suction nozzle, an electromagnet, or the like can be used. A camera 53 for imaging the work W conveyed by the work conveyance device 20 and the mounting table T conveyed by the mounting table conveyance device 16 to grasp their positions and states is provided at the end link of the robot arm 51. Is also attached.

As shown in FIGS. 1 and 2, the work transfer device 20 has a plurality of transfer lanes 21 that can transfer the work W from the supply area A1 to the collection area A2 in the front-rear direction (Y-axis direction). A plurality of supply boxes 12 for accommodating works W to be supplied to each of the plurality of transfer lanes 21 are arranged behind the work transfer device 20.

The work transfer device 20 includes a conveyor belt 22 and a partition 25. As shown in FIG. 2, the conveyor belt 22 is stretched over a driving roller 23a and a driven roller 23b. The work W is placed on the upper surface portion 22a (placement portion) of the conveyor belt 22, and the work roller W is driven to rotate by the drive motor 38 (see FIG. 4) to convey the work W in the belt feeding direction. On both sides of the conveyor belt 22,

side walls

24a and 24b are provided. The driving roller 23a and the driven roller 23b are rotatably supported by the

side walls

24a and 24b. As shown in FIG. 3, the work transfer device 20 has a support plate 28 on the back side of the upper surface 22 a of the conveyor belt 22. The support plate 28 prevents the conveyor belt 22 from bending due to the weight of the work W placed on the upper surface 22a. In the support plate 28, openings 28a are formed at positions corresponding to the collection areas A2 of the plurality of transport lanes 21, respectively. Below each opening 28a, a vertical movement device 30 for pushing up the upper surface portion 22a from the back surface and moving the upper surface portion 22a up and down is arranged. The vertical movement device 30 includes a contact body 31 and a cylinder 32 for vertically moving the contact body 31 so as to penetrate the opening 28a. The cylinder 32 is supported by a support base 29 fixed to the

side walls

24a, 24b. The partition 25 is a partition plate that partitions one conveyor belt 22 (upper surface portion 22a) into a plurality of transport lanes 21. The partitions 25 extend in parallel to the

side walls

24a and 24b arranged on both sides of the conveyor belt 22, and are arranged at equal intervals so that each of the transport lanes 21 has the same lane width.

Although not shown, the control device 70 is configured as a known computer including a CPU, a ROM, a HDD, a RAM, an input / output interface, a communication interface, and the like. Various signals from the encoder 45 of the supply robot 40, the encoder 55 of the collection robot 50, the camera 53, the input device 80, and the like are input to the control device 70. From the control device 70, the drive motor 38 of the work transfer device 20, the vertical movement device 30 (cylinder 32), the drive motor 44 of the supply robot 40, the drive motor 54 of the collection robot 50, the camera 53, the mounting table transfer device 16, and the like. Are output.

The control device 70 learns parameters for controlling the cylinder 32 of the vertically moving device 30 and can control the cylinder 32 by determining an appropriate parameter based on the learning result. FIG. 5 is an explanatory diagram showing a state where the bulk of the works W is unraveled. As shown in the drawing, a loosening operation in which the cylinder 32 moves the contact body 31 up and down (vibrates) dislodges the lump of the work W in a bulk state and becomes a separated state, so that the work W can be easily collected by the collection robot 50. Becomes The control device 70 controls the cylinder 32 with parameters suitable for unraveling the work W according to the specifications of the work W, such as the weight, size, shape, and material of the work W. In addition, examples of the parameters include an impact force and a vibration frequency when the conveyor belt 22 is pushed up by the vertical movement (vibration) of the contact body 31.

FIG. 6 is a block diagram showing functions of the control device 70. As shown in the figure, the control device 70 includes a parameter learning unit 70A that mainly learns parameters, and a drive control unit 70B that mainly determines appropriate parameters and drives and controls the vertical movement device 30. The parameter learning unit 70A includes a learning model 71, an imaging processing unit 72, an evaluation processing unit 73, and a learning processing unit 74. The imaging processing unit 72 causes the camera 53 to image the work W in a bulk state or the separated state in which the work W is unraveled, and inputs the taken image. The evaluation processing unit 73 processes the captured image, calculates a predetermined evaluation value regarding the separation state of the work W, and evaluates the separation state. The learning processing unit 74 learns the relationship between the parameter in the unraveling operation being executed and the evaluation result with the evaluation processing unit 73 by well-known machine learning, and constructs a learning model 71 including a correlation with the specification of the work W and the like. I do. Note that examples of the learning method include reinforcement learning and a genetic algorithm, and other methods may be used. The drive control unit 70B includes a parameter determination unit 75 and a drive unit 76. The parameter determining unit 75 determines a parameter corresponding to the specification of the work W or the like using the learning model 71, or appropriately determines an arbitrary parameter. The drive unit 76 controls the cylinder 32 of the vertical movement device 30 based on the parameters determined by the parameter determination unit 75.

作業 In the work system 10 configured as described above, each control such as the supply control and the transfer control of the work W, the loosening operation control, and the sampling and placement control is sequentially performed. The supply control is performed by collecting the work W from the supply box 12 in accordance with the supply order and controlling the drive of the supply robot 40 so as to supply the work W to the supply area A1 of the corresponding transfer lane 21. The supply order is, for example, an order specified by an operator by operating the input device 80. The transfer control is performed by controlling the drive of the work transfer device 20 so that the work W supplied to the supply area A1 reaches the collection area A2. The loosening operation control is performed by controlling the driving of the cylinder 32 of the vertical movement device 30 corresponding to the collection area A2 in a state where the work W has reached the collection area A2. Sampling and placement control is performed by sampling the work W that has been unraveled and separated by the unraveling operation control, and drives and controls the sampling robot 50 so as to be aligned and mounted on the mounting table T. . In the collection and placement control, the collection robot 50 is driven and controlled so that the work W in the collection area A2 is captured by the camera 53, and the captured image is processed to collect the selected work W. These controls performed on the work W in each transfer lane 21 may be performed in parallel if they do not affect the control on the work W in the other transfer lanes 21. Hereinafter, the details of the loosening operation control will be described based on the loosening operation control routine shown in FIG.

In the loosening operation control routine of FIG. 7, the control device 70 determines whether or not it is time to start the loosening operation (S100). The control device 70 determines that it is the start timing when the work W reaches the sampling area A2 by the transport control and the vertical movement device 30 is in a state where it can be driven. Note that the control device 70 may determine that the start timing is reached, for example, when it is necessary to perform another loosening operation on the remaining work W from which some work W has been collected by performing the loosening operation. When it is determined that the timing for starting the loosening operation has come, the control device 70 executes the process for starting the loosening operation shown in FIG. 8 (S105).

In the processing at the start of the loosening operation in FIG. 8, the control device 70 first initializes the number n indicating the image capturing order to a value of 1 (S200), and before the start of the loosening operation, the image which is the image of the number 1 with the camera 53 1 is imaged (S205). Next, the control device 70 processes the image 1 to detect the outer edge of the lump area of the workpiece W, and calculates the area area A (1) and the center of gravity G (1) of the area area A (1). (S210). The work W is supplied in a bulk state to the supply area A1 and transported to the collection area A2. Therefore, in the image 1, a plurality of works W are intertwined to form a lump, and the area of the work W and a conveyor belt serving as a background. The brightness value and the like of the upper surface portion 22a of the second 22 differ. For this reason, in S210, for example, the image 1 is converted into a gray scale image, and the boundary between the lump of the work W and the upper surface portion 22a is detected as the outer edge of the area of the work W from the gray scale image, and the area surrounded by the outer edge is detected. Is calculated as a region area A (n) (here, A (1)). Further, the position of the center of gravity of the region surrounded by the outer edge is calculated as the center of gravity G (n) (here, G (1)). Note that the control device 70 is not limited to a device using a grayscale image, and may use a binarized image. Subsequently, the control device 70 sets the parameters of the loosening operation (S215), starts the loosening operation by driving the cylinder 32 of the vertical movement device 30 based on the set parameters (S220), and starts the loosening operation. The time processing ends. In S215, the control device 70 selects and sets parameters suitable for the current work W from the learning model 71. Note that the control device 70 may appropriately set an arbitrary parameter when the selection is difficult, for example, when the work W is new.

After performing the loosening operation start process in S105, the control device 70 determines whether a predetermined timing during the loosening operation has come (S110). Here, the predetermined timing may be a timing each time a predetermined time elapses after starting the loosening operation, or a timing each time the cylinder 32 of the vertical movement device 30 performs the vertical movement a predetermined number of times. It is assumed that the predetermined timing occurs a plurality of times from the start of the loosening operation to the end thereof. If it is determined in S110 that the predetermined timing has come, the control device 70 updates the number n by incrementing the number n by one (S115), and the image n which is the image of the number n by the camera 53 during the execution of the loosening operation. Is imaged (S120). To capture the image n during the execution of the loosening operation means to capture the image n without interrupting the continuous vertical movement of the cylinder 32. For this reason, since the image n is captured in a state where the work W is jumping in various directions due to the vibration, the image is in a state where each work W is shaken.

Subsequently, the control device 70 processes the image n to calculate the area area A (n) and the center of gravity G (n) (S125), and separates the workpiece W sufficiently to be able to be collected by the collection robot 50. It is determined whether (spreading) has been performed (S130). The process of S125 is performed in the same manner as the process at the start of the loosening operation S210, except that the image n in which the workpiece W is blurred is used. Note that, even in the image n in which the work W is shaken, since it is possible to detect the approximate boundary between the lump of the work W and the upper surface 22a of the conveyor belt 22, the outer edge of the area of the work W is detected. To calculate the area A (n) and the center of gravity G (n). Further, the control device 70 determines that the work W has been sufficiently separated in S130 when a state in which the work can be collected by the collection robot 50 because some of the individual works W can be recognized from the processed image n. Alternatively, the control device 70 predicts a region area Ae in which the work W is sufficiently separated based on, for example, the specifications and the number of the work W, and calculates the region area A (n) of the work W calculated in S125. Is greater than or equal to the area Ae, it is determined that the workpiece W has been sufficiently separated.

If the control device 70 determines in S130 that the work W is not sufficiently separated, the control device 70 calculates the difference ΔA and the difference ΔG as evaluation values to evaluate the separation state (S135). Here, the difference ΔA is calculated as an area difference between the area A (n) and the reference area. The difference ΔG is calculated as the distance difference between the center of gravity G (n) and the reference center of gravity.

FIG. 9 is an explanatory diagram showing an example of a method of calculating the differences ΔA and ΔG. In the method shown in FIG. 9A, the difference ΔA (A (n) −A (1) is used by using the area A (1) and the center of gravity G (1) of the image 1 captured in the processing at the start of the loosening operation as a reference. )) And the difference ΔG (G (n) −G (1)) are calculated. According to this method, the difference ΔA, ΔG can be calculated more accurately than the clear image 1 in which the workpiece W is stationary and does not shake because the image is captured before the start of the loosening operation. In addition, the differences ΔA and ΔG may appear more prominently as the loosening operation time becomes longer, so that the evaluation becomes easier and learning can be performed more appropriately. In the method shown in FIG. 9B, the area area A (n-1) and the center of gravity G (n-1) of the image (n-1) captured at a predetermined timing immediately before the predetermined timing at which the image n is captured are calculated. Is used as a reference to calculate the difference ΔA (A (n) −A (n−1)) and the difference ΔG (G (n) −G (n−1)). In this method, evaluation can be performed even when a parameter is changed during a loosening operation in a process described later. That is, since the control device 70 can perform the evaluation while changing the parameter while continuing the loosening operation, the control device 70 can appropriately learn the parameter without interrupting the loosening operation. 9A and 9B may be selected based on, for example, an operator's designation by operating the input device 80, or the method shown in FIG. May be changed to the method shown in FIG.

FIG. 10 is an explanatory view showing an example of the area A and the center of gravity G. In FIG. 10, the area A (1) and the center of gravity G (1) as references used in the method shown in FIG. 9A are indicated by solid lines, and the area A (n) and the center of gravity G (n) of the image n are indicated by dotted lines. Show. FIG. 10A shows a state in which the area A is large without changing the center of gravity G, and the difference ΔG is small and the difference ΔA is large. Since such a state is a state in which the work W is unraveled and separated without greatly deviating from the sampling area A2, the evaluation is high. FIG. 10B shows a state in which the center of gravity G moves with almost no change in the area A of the region, and the difference ΔA is small and the difference ΔG is large. In such a state, the position of the work W is shifted as a whole without sufficiently loosening the lump of the work W, so that the evaluation is low. As described above, even if the image n captured during the unraveling operation and the work W is shaken is used, it is possible to detect the lump area of the work W and evaluate the separation state by simple processing. Although illustration is omitted, if the area A is large even if the center of gravity G is moved, it can be evaluated that the lump of the work W is unraveled, so the evaluation is higher than that of FIG. 10B. It will be.

The control device 70 updates the learning model 71 by learning such an evaluation result for the current parameter (S140), and determines whether or not a change to a more suitable parameter is necessary (S145). The control device 70 makes the determination in S145 based on whether or not a parameter more suitable for the current work W can be selected from the updated learning model 71. When determining that the parameter needs to be changed, the control device 70 changes the parameter to a more suitable one, continues the loosening operation (S150), and returns to S110. On the other hand, when the control device 70 determines that the parameter change is unnecessary because the evaluation of the separation state with respect to the current parameter is relatively high, the control device 70 continues the loosening operation with the current parameter (S155), and returns to S110. . If the control device 70 determines that the work W has been sufficiently separated in S130 while performing such processing, the control device 70 stops driving the cylinder 32 of the vertical movement device 30 to end the loosening operation (S160). , And returns to S100.

Here, the correspondence between the components of the present embodiment and the components of the present disclosure will be clarified. The cylinder 32 of the vertically moving device 30 of the present embodiment corresponds to an actuator, S120 of the loosening operation control routine in FIG. 7 corresponds to an in-operation imaging step, and S125 and S135 of the same processing correspond to evaluation steps. S140 corresponds to a learning step. S205 of the processing at the start of the loosening operation in FIG. 7 corresponds to a pre-operation imaging step, and S210 of the processing corresponds to an acquisition step. The work system 10 corresponds to a work system, the sampling robot 50 corresponds to a robot, the camera 53 corresponds to an imaging device, and the imaging processing unit 72 that executes S120 of the unraveling operation control routine corresponds to an imaging processing unit. The evaluation processing unit 73 executing S125 and S135 of the processing corresponds to an evaluation processing unit, and the learning processing unit 74 executing S140 of the processing corresponds to a learning processing unit.

In the parameter learning method of the present embodiment described above, the separation state of the plurality of works W is evaluated by processing the image n captured during the unraveling operation of the plurality of works W. Learn the relationship with the parameters in the operation. For this reason, since the lifting / lowering device 30 does not need to interrupt the loosening operation for learning, learning can be appropriately performed without lowering the work efficiency of the loosening operation.

In addition, since the image 1 captured before the start of the unraveling operation is processed to obtain a plurality of works W in a bulk state and the separated state of the works W is evaluated based on the bulk state, the evaluation value is accurately obtained. Learning can be performed more appropriately. Further, the separation state of the workpiece W is evaluated based on the immediately preceding image (n-1) with respect to the image n captured at a predetermined timing during the unraveling operation. Therefore, even if the parameter is changed during the unraveling operation. The learning can be performed by acquiring the evaluation value without interrupting the loosening operation.

Further, since the separation state is evaluated using the area A (n) surrounded by the outer edge of the area of the work W, even if a clear image cannot be obtained, the separation state is appropriately evaluated by simple processing. can do. Further, since the separation state is evaluated using the area A (n) and the center of gravity G (n), the separation state can be more appropriately evaluated.

Note that the present disclosure is not limited to the above-described embodiments at all, and it goes without saying that the present disclosure can be implemented in various modes as long as they belong to the technical scope of the present disclosure.

For example, in the above-described embodiment, the evaluation is performed using the area A (n) and the center of gravity G (n). However, the present invention is not limited to this, and only the area A (n) may be used. Alternatively, another evaluation value may be used. For example, since the height of the lump of the work W also decreases while the bulk state is unraveled, the height of the lump of the work W is detected from an image taken from the side, and the height is evaluated as an evaluation value. May be used.

In the above-described embodiment, as a criterion at the time of evaluation, the bulk state of the works W in the image 1 and the separated state of the works W in the immediately preceding image n are used. However, the present invention is not limited to this. Only one of them may be used. Alternatively, the image n captured immediately before the parameter change or the separation state of the workpiece W in the image n captured immediately after the parameter change may be used as a reference.

In the above-described embodiment, the work transfer device 20 includes the vertical movement device 30 for each of the transfer lanes 21. However, the upper and lower portions 22a of the plurality of transfer lanes 21 are moved up and down by one vertical movement device 30. It may be a thing. In this case, the work transfer device 20 may include a support plate having an opening formed to extend over the plurality of transfer lanes 21.

In the above-described embodiment, the work transfer device 20 includes the up-down movement device 30 that moves the collection area A2 of the conveyor belt 22 (upper surface portion 22a) up and down, but moves the supply area A1 and other regions up and down. The vertical movement device 30 may be provided.

In the above-described embodiment, the cylinder 32 of the up-and-down movement device 30 is exemplified as an actuator for unraveling the bulk state of the works W, but the actuator is not limited to this. For example, the workpiece W may be unraveled by an actuator that reciprocates a leveling member such as a brush in the X-axis direction or the Y-axis direction. Even in this case, the separation state of the workpiece W can be evaluated without interrupting the reciprocation of the actuator by capturing the image n when the leveling member moves to the forward end position or the backward end position. It is possible to The parameters include the angle at which the leveling member is brought into contact with the workpiece W, the speed of the reciprocating motion, and the like. Alternatively, for example, a leveling member may be attached as the end effector 52 of the collection robot 50, and the collection robot 50 may perform a loosening operation (leveling operation). In this case, a parameter for controlling the drive motor 54 as an actuator of the sampling robot 50 may be learned. As described above, the actuator that releases the work by the predetermined operation may be included in a robot that collects the work and performs the predetermined work.

Here, the parameter learning method and the work system by the computer of the present disclosure may be configured as follows. For example, in the parameter learning method of the present disclosure, before the start of the predetermined operation, a pre-operation imaging step of imaging the plurality of workpieces, and processing the images captured in the pre-operation imaging step to process the plurality of images. An acquiring step of acquiring a bulk state of the work; and the evaluating step evaluates the separation state using the bulk state acquired in the acquiring step as a reference. By doing so, the separation state can be accurately evaluated based on a relatively clear image taken before the execution of the operation is started, so that learning can be performed more appropriately.

In the parameter learning method according to an embodiment of the present disclosure, in the in-operation imaging step, the plurality of workpieces are imaged at every predetermined timing during execution of the predetermined operation, and in the evaluation step, the image is captured in the in-operation imaging step. Each time is captured, the separation state may be evaluated using the separation state immediately before processing the image captured in the immediately preceding operation-in-progress imaging step as a reference. In this way, even when the parameter is changed during execution of the predetermined operation, the change in the separation state can be properly grasped, so that the learning can be performed more appropriately.

In the parameter learning method according to an embodiment of the present disclosure, in the evaluating step, the image is processed to detect an outer edge of a region where the plurality of workpieces are present, the area of the region is calculated, and the separation state is calculated based on the area. May be evaluated. In this case, even if a clear image cannot be obtained because an image is captured during execution of the predetermined operation, the separation state can be appropriately evaluated by simple processing.

A work system according to an embodiment of the present disclosure is a work system including an actuator that releases a plurality of works in a bulk state by a predetermined operation, a robot that collects the works and performs a predetermined work, and an imaging device that captures an image. An imaging processing unit that causes the imaging device to image the plurality of workpieces during execution of the predetermined operation; an evaluation processing unit that processes a captured image to evaluate a separation state of the plurality of workpieces; The gist of the present invention is to include a learning processing unit that learns a relationship between the evaluation result of the evaluation processing unit and the parameter in the predetermined operation being executed.

The working system according to the present disclosure performs learning of a parameter for controlling an actuator that performs a predetermined operation using an image captured during execution of the predetermined operation, similarly to the above-described parameter learning method. Therefore, the work device does not need to interrupt the predetermined operation for learning. For this reason, the parameter can be appropriately learned without lowering the work efficiency for unraveling the bulk state of the works. In this working system, a function for realizing each step of the parameter learning method may be added.

The present disclosure is applicable to, for example, the manufacturing industry of working systems.

10 work system, 11 work table, 12 supply box, 16 work table transfer device, 20 work transfer device, 21 transfer lane, 22 conveyor belt, 22a top surface, 23a drive roller, 23b driven roller, 24a, 24b side wall, 28 support Plate, 28a opening, 29 support, 30 vertical movement device, 31 contact body, 32 cylinder, 38 drive motor, 40 supply robot, 41 robot arm, 42 end effector, 44 drive motor, 45 encoder, 50 sampling robot, 51 robot arm, 52 end effector, 53 camera, 54 drive motor, 55 encoder, 70 control device, 70A parameter learning unit, 70B drive control unit, 71 learning model, 72 imaging processing unit, 73 evaluation processing unit 74 learning processing unit, 75 parameter determination unit, 76 drive unit, 80 input unit, T table.

Claims

A parameter learning method for controlling an actuator that unravels a bulk state of a plurality of works by a predetermined operation,
In-operation imaging step of imaging the plurality of workpieces during execution of the predetermined operation,
An evaluation step of processing an image captured in the imaging step during operation to evaluate a separation state of the plurality of works,
A learning step of learning a relationship between the evaluation result of the evaluation step and the parameter in the predetermined operation being executed;
Learning method of parameters including.
The parameter learning method according to claim 1, wherein:
Before the start of the predetermined operation, a pre-operation imaging step of imaging the plurality of works,
An acquisition step of processing an image captured in the pre-operation imaging step to obtain a bulk state of the plurality of workpieces,
In the evaluation step, a parameter learning method for evaluating the separation state using the bulk state acquired in the acquisition step as a reference.
The parameter learning method according to claim 1, wherein:
In the in-operation imaging step, at each predetermined timing during execution of the predetermined operation, image the plurality of workpieces,
In the evaluation step, each time an image is captured in the active imaging step, the separation state immediately before processing the image captured in the active imaging step is used as a reference. How to learn parameters to evaluate the state.
A parameter learning method according to any one of claims 1 to 3,
In the evaluation step, a parameter learning method for processing the image, detecting an outer edge of a region where the plurality of works are present, calculating an area of the region, and evaluating the separation state based on the area.
An operation system comprising: an actuator that unravels a bulk state of a plurality of works by a predetermined operation, a robot that performs the predetermined work by collecting the work, and an imaging device that captures an image,
An imaging processing unit that causes the imaging device to image the plurality of workpieces during execution of the predetermined operation;
An evaluation processing unit that processes a captured image to evaluate a separation state of the plurality of works,
A learning processing unit that learns a relationship between the evaluation result of the evaluation processing unit and the parameter in the predetermined operation being executed;
A working system comprising: