Method for acquiring image data used to determine whether scrap materials have a drop defect, and method for creating a trained model using the image data acquired by the acquisition method.

Abstract

To provide a method for acquiring high-quality image data that indicates poor scrap material dropping, and a method for creating a trained model using the image data obtained by this acquisition method. [Solution] A method for acquiring image data used to determine whether scrap 72 generated from a workpiece 70 processed by a press die 52 is falling properly, comprising the steps of: a computer 11 performing a simulation of the operation in which a plurality of scraps 72 are discharged to the outside of the press die via a scrap chute 58, and arranging a camera 76 to photograph the corresponding scrap 72 for each scrap 72; and before the scrap is discharged to the outside of the scrap chute, when the speed of the scrap becomes zero, the camera corresponding to the scrap photographs the scrap, and the computer acquires image data of the scrap.

Description

[Technical Field]

[0001] The present invention relates to a method for acquiring image data used to determine whether scrap is falling properly, a method for creating a trained model using the image data acquired by the acquisition method, and a trained model created by the creation method. More particularly, the present invention relates to a method for acquiring image data and a method for creating a trained model using a simulation of dropping scrap into a scrap chute. [Background technology]

[0002] In the manufacturing process of pressed products, after the sheet metal workpiece is drawn and formed using a press die installed in the press machine, a trimming process is performed to cut off and discard the scrap portion of the workpiece. The scrap cut off in the trimming process is discharged to the outside of the press die through the scrap chute of the press machine. At this time, if a scrap disposal defect occurs where the scrap remains inside the press machine due to an unintended falling motion, it can lead to the production of defective products in subsequent processing of pressed products or cause damage to the press die.

[0003] To solve these problems, a technique has been developed to simulate the motion of scrap falling into a scrap chute when designing a press machine.

[0004] For example, Patent Document 1 describes a simulation method that virtually creates a press device and scrap generated from a press-processed workpiece within a computer, and virtually reproduces the operation of this scrap falling into a scrap chute and being discharged to the outside.

[0005] This simulation method involves repeatedly simulating the dropping motion while varying the force applied to the scrap during the fall. The probability that the scrap is discharged to the outside via the scrap chute, i.e., the scrap discharge rate, is calculated. Furthermore, the quality of the press machine's design is judged based on this discharge rate. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2022-146544 [Overview of the project] [Problems that the invention aims to solve]

[0007] In such simulations, methods are employed that reduce the frame rate (number of images per second) or that do not consider the movement of the upper die in a press die, in order to prevent a decrease in the computer's information processing speed. As a result, the simulation may show scrap dropping defects that would not occur in an actual press die. For example, due to the low number of images, situations may occur where scrap gets stuck in the scrap chute, or because the upper die is not considered, situations may occur where scrap rides up on the upper surface of the lower die.

[0008] When calculating the scrap discharge rate, if the calculation includes not only scrap dropping defects that can actually occur (hereinafter also referred to as "correct dropping defects") but also scrap dropping defects that cannot actually occur (hereinafter also referred to as "incorrect dropping defects"), the accuracy of the calculated discharge rate will decrease. Therefore, in order to improve the accuracy of the scrap discharge rate, it is necessary for the computer to automatically detect correct and incorrect dropping defects, that is, to accurately determine whether a dropping defect is "appropriate" or "inappropriate". Hereinafter, the process of determining whether a scrap dropping defect is appropriate or inappropriate will also be referred to as the determination of appropriateness of dropping defects.

[0009] One possible method for a computer to determine whether a fall defect is valid or not is to use image recognition based on artificial intelligence (hereinafter also simply referred to as "AI"). However, AI-based image recognition requires high-quality image data to be used for machine learning to automatically determine whether a fall defect is valid or false.

[0010] However, in actual press dies, multiple pieces of scrap are generated from a single workpiece, and each piece of scrap is discharged from the scrap chute to the outside of the press machine in a short time. Therefore, it was not easy to understand the dropping motion of each piece of scrap and acquire image data suitable for AI machine learning to determine whether the scrap dropping was improper or not.

[0011] The present invention has been made in view of the above problems, and aims to provide a method for acquiring image data that can obtain high-quality image data indicating poor scrap dropping, and a method for creating a trained model using the image data obtained by the acquisition method. [Means for solving the problem]

[0012] To achieve the above objective, one embodiment of the present invention is A method for acquiring image data used to determine whether scrap generated from a workpiece processed by a press die is falling properly, The computer performs a simulation of the operation in which multiple pieces of scrap generated from the workpiece are discharged to the outside of the press die via a scrap chute, and in the simulation, The process involves placing a camera to photograph the corresponding scrap for each piece of scrap, Before the scrap is discharged outside the scrap chute, when the speed of the scrap becomes zero, the camera corresponding to the scrap photographs the scrap, and the computer acquires image data of the scrap. It is characterized by including.

[0013] Furthermore, one embodiment of the present invention is a method for creating a trained model using image data acquired by the above-described method for acquiring image data, The method is characterized by creating a trained model based on training data obtained by annotating the aforementioned image data with whether the fall failure is actually possible or not. [Effect of the Invention]

[0014] According to the method for acquiring image data used for determining the suitability of scrap dropping failure according to the present invention and the method for creating a learned model using the image data acquired by the acquisition method, high-quality image data indicating scrap dropping failure can be acquired. [Brief Description of the Drawings]

[0015] [Figure 1] It is a schematic diagram of a press device. [Figure 2] It is a block diagram showing the hardware configuration of a simulation device. [Figure 3] It is a block diagram showing the functional parts of a simulation device. [Figure 4] It is a perspective view showing a part of a press device displayed in simulation. [Figure 5] It is a diagram for explaining the random force applied to the scrap. [Figure 6A] It is a flowchart showing the procedure of simulation processing [Figure 6B] It is a flowchart showing the procedure of simulation processing [Figure 7A] It is an explanatory diagram showing the dropping of scrap and the movement of the camera. [Figure 7B] It is an explanatory diagram showing the dropping of scrap and the movement of the camera. [Figure 7C] It is an explanatory diagram showing the dropping of scrap and the movement of the camera. [Figure 8A] It is an explanatory diagram showing a scrap dropping failure that cannot occur. [Figure 8B] It is an explanatory diagram showing a scrap dropping failure that cannot occur. [Figure 8C] It is an explanatory diagram showing a scrap dropping failure that cannot occur. [Modes for Carrying Out the Invention]

[0016] One embodiment of the present invention will be described in detail with reference to the drawings. Figure 1 is a schematic diagram of the press device 50, Figure 2 is a block diagram showing the hardware configuration of the simulation device 10, and Figure 3 is a block diagram showing the functional parts of the simulation device 10. In the method for acquiring image data according to one embodiment of the present invention, the image data is acquired using a simulation performed by the simulation device 10. This simulation simulates the operation in which a plurality of scraps 72 generated in the press die 52 are discharged to the outside of the press die 58 via a scrap chute 58. The image data used to determine whether the scraps 72 have failed to fall properly is acquired by a plurality of cameras 76 placed on the simulation, as shown in Figure 4. The cameras 76 are virtual cameras. Among the image data from the cameras 76, the image data showing the failure to fall properly of the scraps 72 is used to determine whether the scraps 72 have failed to fall properly. The press device 50 and the simulation device 10 will be described in detail below.

[0017] The press device 50 presses a sheet metal workpiece 70 and also performs trimming to cut off the scrap portion of the workpiece 70. The press device 50 includes a press die 52 and a scrap chute 58 provided in the press die 52 that guides the scrap 72 cut from the workpiece 70 to the outside of the press die 52.

[0018] The press die 52 is the part that performs press working and trimming on the workpiece 70, and comprises a lower die 54 fixed to the installation site and an upper die 56 that can move up and down relative to the lower die 54. The lower die 54 is equipped with a lower cutting blade 55 supported by the lower die body. The upper die 56 is equipped with a pad 56a that holds down the workpiece 70 placed on the lower die 54, a cam slider 56b, and an upper cutting blade 57. The cam slider 56b moves forward toward the workpiece 70 by contacting the cam driver 54a provided on the lower die body when the upper die 56 moves downward, and moves backward toward the workpiece 70 when the upper die 56 moves upward. The upper cutting blade 57 is provided on the cam slider 56b so as to face the lower cutting blade 55. The workpiece 70 is cut by the upper cutting blade 57 and lower cutting blade 55 of the press die 54, and a portion of the cut workpiece 70 falls as scrap 72 into the scrap chute 58 of the press device 50.

[0019] The scrap chute 58 is positioned below the lower cutting edge 55 of the press die 52 and extends at an inclination with respect to the horizontal direction so that the scrap discharge port is downward. In the illustrated example, the scrap chute 58 and the lower die body are integrally formed. The scrap chute 58 has a pair of side walls arranged at a predetermined distance in the width direction and a bottom wall connecting the lower ends of each side wall, and is open at the top. The scrap 72 slides downward along the upper surface of the bottom wall in the inclined direction and is discharged to the outside of the press device 50. In the simulation of the scrap 72 dropping motion by the simulation device 10, the operation of the upper die 56 can be excluded from consideration in order to increase the processing speed of the computer 11.

[0020] Next, a simulation device 10 for implementing an image data acquisition method according to one embodiment of the present invention will be described. As shown in Figure 2, the simulation device 10 includes a computer 11 having a known hardware configuration. The computer 11 includes a processor 12, RAM 13, ROM 14, storage 15, communication device 16, input device 17, and output device 18.

[0021] The processor 12 executes the operating system and application programs. The storage 15 can be a known storage device for storing information, such as a hard disk drive (HDD) or a solid-state drive (SSD). The communication device 16 is a transmitting and receiving device for communication between computers 11 via at least one of a wired network and a wireless network, and consists of, for example, a network device, a network controller, a network card, a wireless communication module, etc. The input device 17 consists of a keyboard, mouse, touch panel, microphone, etc. The output device 18 consists of a display, speaker, etc.

[0022] As shown in Figure 3, the simulation device 10 includes, as functional units, a storage unit 20, a communication unit 21, an input unit 22, an output unit 23, and an information processing unit 24.

[0023] The memory unit 20 includes storage 15 and stores information for performing simulations, image data obtained from the simulations, etc. The communication unit 21 includes a communication device 16 and enables the transmission and reception of data with external devices of the simulation device 10. The information stored in the memory unit 20 includes programs for performing simulations. These programs may be provided on tangible recording media such as CD-ROMs, DVD-ROMs, or semiconductor memory, or they may be provided as data signals via the communication unit 21. The input unit 22 includes an input device 17 and acquires information input from the user. The output unit 23 includes an output device 18 and outputs information to the user. The output unit 23 outputs information such as images, text, and audio so that the user can recognize the information using their sight and hearing.

[0024] The information processing unit 24 has a simulation execution unit 26. The information processing unit 24 includes a processor 12, and the simulation execution unit 26 is composed of software executed by the processor 12. The simulation execution unit 26 performs a simulation on the computer 11 in which scrap 72 is dropped into a scrap chute 58. The simulation execution unit 26 includes a three-dimensional model construction unit 31, a random force setting unit 32, a force application unit 33, a deceleration processing unit 34, a camera control unit 35, a color setting unit 36, an image data acquisition unit 37, and a learning unit 38.

[0025] The three-dimensional model construction unit 31 reads morphological data of the press die 52 including the scrap chute 58, the workpiece 70, and the scrap 72, and constructs three-dimensional models of them. Morphological data can be obtained via the input unit 22. In this embodiment, the three-dimensional model is created using CAD data used when designing the press machine 50 and the workpiece 70. The three-dimensional model can be, for example, a three-dimensional mesh model in which the shapes of the press die 52 and the workpiece 70 are represented by a polygon mesh.

[0026] The random force setting unit 32 sets a random force F of a predetermined magnitude to act on the scrap 72 during its initial fall, within a predetermined range. This random force F (hereinafter also referred to as "random force F") can be expressed by the random function shown in the following equation (1). force = (upper limit, lower limit) ......Equation (1)

[0027] In this embodiment, random forces F are set for each of the mutually orthogonal X, Y, and Z directions, as shown in the following equations (2), (3), and (4). X-direction force = random(upper limit x) up Lower limit x low )...Equation (2) Y-direction force = random(upper limit y) up , lower limit y low )...Equation (3) Z - direction force = random(upper limit value of z up , lower limit value of z low ) ····· Equation (4)

[0028] In this embodiment, as an example, the positive direction of the Y - direction is set as the gravitational direction, and the lower limit value of y low is set to 0. Also, the upper limit value of x up , the upper limit value of y up and the upper limit value of z up are set to positive values, and the lower limit value of x low [[ID=I8]]and the lower limit value of z low are set to negative values. The random force F acting on the scrap 72 is the resultant force of the forces in the X - direction, Y - direction, and Z - direction set by Equations (2) to (4). As shown in FIG. 5, the range of the magnitude of the random force F is set so that it is smaller than the magnitude of the gravitational force G acting on the scrap 72.

[0029] The force application unit 33 applies the gravitational force G and the random force F set by the random force setting unit 32 to the scrap 72. The random force F is applied at the initial time of the fall of the scrap 72 generated from the work 70.

[0030] The deceleration processing unit 34 performs deceleration processing on the scrap 72 when the falling speed v of the scrap 72 exceeds a predetermined threshold value v th (hereinafter, also referred to as "speed threshold value v th "). The deceleration processing is performed according to a preset rule. In this embodiment, when the speed threshold value v th is exceeded, as shown in the following Equation (5), a process of halving the magnitude of the falling speed v of the scrap 72 is performed, and the falling speed v' after deceleration processing is set as the current falling speed v. The falling speed v' after deceleration processing = v / 2 ····· Equation (5)

[0031] The camera control unit 35 controls the camera 76 that is placed in the simulation. Setting information and control information regarding the camera 76 can be input by the user to the computer 11 via the input unit 22. In this embodiment, in the simulation, a camera 76 is set to be placed to photograph each piece of scrap 72 generated from the workpiece 70, corresponding to the scrap 72.

[0032] Figure 4 is a perspective view showing a part of the press device 50 as displayed in the simulation. Figure 4 shows a part of the lower die 54 of the press device 50 and a scrap chute 58 integrally formed with the lower die 54, and further shows how the scrap 72 falls into the scrap chute 58. For example, if the scrap portion to be separated from the workpiece 70 is large, the scrap portion is divided into multiple pieces and removed, as shown in Figure 4. The scrap portion is supported from below by support parts 54b provided on the lower die 54 until it is cut from the workpiece 70 by the upper cutting blade 57 shown in Figure 1. As an example, Figure 4 shows two support parts 54b protruding from the bottom surface of the scrap chute 58.

[0033] In the example simulation shown in Figure 4, three cameras 76-1, 76-2, and 76-3 are positioned for each of the three scraps 72-1, 72-2, and 72-3. Each camera 76 is a virtual camera positioned in the simulation and is positioned by the camera control unit 35 to correspond to each scrap before the multiple scraps 72 are dropped. As an example, in this embodiment, the position of each camera 76 is set so that image data is acquired from a vertically upward position at a certain distance from each scrap 72. Specifically, in the world coordinate system defined in the simulation, which has X, Y, and Z directions, the position coordinates of the camera 76 are set so that the X and Z coordinates coincide with the centroid coordinate of the scrap 72, and the Y coordinate, which is vertical, is positioned vertically upward by a certain distance (e.g., 2m). Each camera 76 is set to move in accordance with the falling motion of the corresponding scrap 72. In this embodiment, the position coordinates of the camera 76, which is positioned in relation to the scrap 72, are set to always remain constant in a local coordinate system based on the center of gravity of the scrap 72. Note that two or more cameras 76 may be positioned for a single piece of scrap 72, each with a different field of view. In this case, each of the multiple cameras positioned for a single piece of scrap 72 is set to move in accordance with that piece of scrap 72. Although this description assumes three pieces of scrap 72, in reality, there can be more; for example, when processing a car side panel, approximately 20 pieces of scrap may be involved.

[0034] The camera control unit 35 controls the camera 76 to switch between an active state where it can take pictures and an inactive state where it cannot take pictures. The camera 76 can also record video of the scrap 72 falling, but video data is larger in size than image data, and processing speed is slower. Therefore, in this embodiment, the camera control unit 35 sets the camera 76 to an inactive state at the beginning of the scrap 72's fall. Then, before the scrap 72 falls and is discharged to the outside, when the velocity of the scrap 72 becomes zero, the camera control unit 35 controls the camera 76 corresponding to the scrap 72 to switch from the inactive state to the active state and take pictures.

[0035] Camera 76 can be configured to display only one corresponding scrap 72 out of several scraps 72. For example, in the example shown in Figure 4, the image captured by camera 76-1 may include both scrap 72-1, which corresponds to camera 76-1, and an adjacent scrap 72-2, which does not correspond to camera 76-1. In such a situation, the user can configure camera 76-1 to display only the corresponding scrap 72-1 in the image it captures. In this embodiment, such configuration can be performed in a simulation, for example, using a game development engine with a built-in IDE (Integrated Development Environment). The game development engine can be configured to display only one corresponding scrap out of several scraps 76 by using a function that specifies which objects to display to camera 76 for each layer. Specifically, two layers, layer L1 and layer L2, are prepared, with layer L1 being the layer displayed by camera 76 and layer L2 being the layer not displayed. Layer L2 is set as the default layer for each scrap 72. Then, after the scrap 72 begins to fall, the layer of the scrap 72, whose velocity has become zero, is changed from layer L2 to layer L1 so that it is visible to camera 76, and a picture is taken with camera 76 corresponding to this scrap 72. After the picture is taken, the layer of the scrap 72 is changed from layer L1 to layer L2 so that it is no longer visible to camera 76. This makes it possible to capture the scrap 72, whose velocity has become zero, with camera 76. When camera 76 acquires image data, it is set so that only the one scrap 72 corresponding to this camera 76 is captured. As a result, the image captured by camera 76-1 will show the press mold 52 including the scrap chute 58 and one piece of scrap 72-1.

[0036] The color setting unit 36 ​​changes the color of the scrap 72 to a predetermined color different from that of the scrap chute 58 when the scrap 72's velocity becomes zero and it is photographed by the camera 76 before it is discharged to the outside via the scrap chute 58. In the simulation, the scrap 72 and the press device 50 can be appropriately colored by the color setting unit 36. For example, if the scrap chute 58 is set to orange and the scrap 72 is set to green at the beginning of its fall, and the scrap 72's velocity becomes zero before it is discharged to the outside, i.e., a scrap jam occurs, the color setting unit 36 ​​changes the color of the scrap 72 to a predetermined color (for example, blue). At the beginning of its fall, multiple pieces of scrap 72 can be set to different colors (for example, green or yellow-green), but the color setting unit 36 ​​sets the colors so that when the camera 76 photographs scrap 72 that is in a falling malfunction state, they are changed to the same color (in this case, blue).

[0037] The image data acquisition unit 37 acquires image data indicating a faulty fall of the scrap 72 captured by each camera 76 and stores it in the storage unit 20. In this embodiment, the camera control unit 35 controls the camera 76 to switch to an active state only when the fall speed of the scrap 72 becomes zero and a faulty fall occurs, and the image data captured at this time is stored in the storage unit 20 as image data indicating a faulty fall. When the camera 76 captures a video of the scrap 72 falling, for example, an image when the color of the scrap 72 is changed to a predetermined color by the color setting unit 36 ​​can be acquired as image data indicating a faulty fall of the scrap 72. The image data indicating a faulty fall of the scrap 72 is stored in a predetermined storage area of ​​the storage unit 20. The user can output the image data stored in the storage unit 20 to the output unit 23 by inputting an output instruction for image data via the input unit 22.

[0038] The learning unit 38 creates a trained model by performing machine learning based on training data obtained by annotating the image data acquired by the image data acquisition unit 37 with the trained model's creator. The learning unit 38 can be configured to include, for example, a neural network, which is an information processing model that mimics the structure of the neural network in the human brain.

[0039] The simulation device 10 may consist of one computer 11 or multiple computers 11. When multiple computers 11 are used, each computer 11 is connected via a communication network such as the Internet or an intranet.

[0040] In the simulation device 10 described above, various morphological data and calculation condition data are input to the computer 11 via the input unit 22. The morphological data includes the morphological data of the press device 50 including the scrap chute 58, the workpiece 70, and the scrap 72. The calculation condition data includes the pressing force of the press die 52, the time Te for dropping the scrap 72 (for example, 4 seconds), the number of frames N (N is an integer of 2 or more) of the scene obtained by equally dividing the dropping time Te, the upper and lower limits of the random function described above, the magnitude of gravity G, and the conditions of the reaction force that the scrap 72 receives from the scrap chute 58 when it comes into contact with the scrap chute 58.

[0041] Next, the simulation process of the scrap 72 dropping operation performed on the computer 11 by the simulation execution unit 26 of the simulation device 10 will be described. Figures 6A and 6B are flowcharts of the simulation process procedure. First, as shown in Figure 6A, in step S10, the three-dimensional model construction unit 31 creates a three-dimensional model of the press device 50 including the scrap chute 58, the workpiece 70, and the scrap 726 based on the morphological data input to the computer 11. In the next step S11, the number of times M to drop the scrap 72 in the simulation is set. The number of times M is an integer of 1 or more.

[0042] In the next step S12, before dropping the multiple pieces of scrap 72 generated from the workpiece 70, a camera 76 is placed for each piece of scrap 72 to photograph the corresponding piece of scrap 72.

[0043] In the next step, S13, the local coordinates K of camera 76 are set. As described above, in this embodiment, the local coordinates K are relative coordinates for each camera 76 with respect to the corresponding scrap 72.

[0044] In the next step S14, the rotation amount L of the camera 76 is set. In this embodiment, the orientation of each camera 76 is set to be downward in the vertical direction.

[0045] In the next step S15, the number of times the scrap 72 falls, m=1, is set. In the next step S16, scene n=1 is set for the falling motion of the scrap 72. Here, n is the number of each scene in the frame count N read in the calculation condition data, and is an integer from 1 to N. In this embodiment, scene 1 is set to the scene in which scrap 72 is generated by the cutting of the workpiece 70, that is, the scene at the beginning of the falling of the scrap 72.

[0046] In the next step, S17, the forces acting on the scrap 72 during its initial fall are set. In this embodiment, the random force setting unit 32 sets a random force F based on the random functions of equations (2) to (4) described above, so that gravity G and the random force F act on the scrap 72 during its initial fall.

[0047] In the next step S18, the falling velocity v of the scrap 72 is set to a predetermined velocity threshold v. th It is determined whether or not it exceeds the speed threshold v. th If it exceeds the velocity threshold v, in the following step S19, the deceleration processing unit 26 performs a deceleration process on the falling velocity v. In step S18, when the falling velocity v is exceeded the velocity threshold v th In the following cases, the process will proceed to step S20 without performing the deceleration process.

[0048] In the next step, S20, the position of camera 76 is set. The position of camera 76, that is, the position coordinates of camera 76 in world coordinates, is set by the camera control unit 35 so that the local coordinates of camera 76 are equal to the local coordinates K set in step S13 with respect to the scrap 72 that has moved due to the falling motion. This maintains a constant positional relationship between the scrap 72 and the corresponding camera 76.

[0049] In the next step, S21, the rotation amount L of the camera 76 is set. The camera control unit 35 sets the camera 76 to a value equal to the rotation amount L set in step S14, so that it always faces a constant direction, in this case, vertically downward. In this embodiment, steps S20 and S21 set the position and rotation amount of the camera 76 relative to the corresponding scrap 72 for each scene n, thereby maintaining a constant position and orientation of the camera 76 relative to the scrap 72 from the start to the end of the scrap 76's fall.

[0050] In the next step, S22, it is determined whether the velocity of scrap 72 is zero or not. If the velocity of scrap 72 is not zero in step S22 (step S22: No), the process proceeds to step S23. In step S23, it is determined whether the fall time of scrap 72 exceeds the set fall time Te or not.

[0051] In step S23, if time Te has not elapsed (step S23: No), the process proceeds to step S24, where scene n is countered, and then processing resumes from step S18. In step S23, if time Te has elapsed (step S23: Yes), the process proceeds to step S25, where it is determined whether the number of falls m is equal to or greater than the number M set in step S11. If the number is less than M (step S25: No), the process proceeds to step S26, where the number of falls m is countered, and then processing resumes from step S16. In step S25, if the number is M or greater (step S25: Yes), the simulation ends.

[0052] Figures 7A, 7B, and 7C are explanatory diagrams illustrating the fall of scrap 72 and the movement of camera 76, showing the process of scrap 72 falling into the scrap chute 58 and being discharged to the outside. In the simulation process, scene n is counter-processed, and each time the fall position of the scrap 72 changes, the position of camera 76 also changes. This maintains a constant relative positional relationship between the scrap 72 and camera 76.

[0053] In step S22 above, if the velocity of the scrap 72 is zero (step S22: Yes), the process proceeds to step S27 shown in Figure 6B. In step S27, the scrap 72 that is captured by the camera 76 is set. In this embodiment, the camera control unit 35 sets the scrap 72 that is captured by the camera 76 to be only one scrap 72 corresponding to this camera 76. Before the setting in step S27 is performed, the camera 76 may be set so that no scrap 72 is captured, or it may be set so that all scrap 72 within the shooting range of the camera 76 are captured. Alternatively, in the initial setting of the fall, it may be set so that only one scrap 72 corresponding to the camera 76 is captured. In such cases, the processing of step S27 can be omitted.

[0054] In the next step, S28, the color setting unit 36 ​​changes the color of the scrap 72 to a predetermined color. In this embodiment, the color is changed so that the scrap 72 whose velocity becomes zero due to a falling failure turns blue.

[0055] In the next step S29, the camera control unit 35 switches the camera 76, which is provided for the scrap 72 whose velocity has become zero, from an inactive state to an active state. In the following step S30, the active camera 76 photographs the corresponding scrap 72, and the computer 11 acquires image data. The image data showing the faulty fall condition of the photographed scrap 72 is stored in the storage unit 20 by the image data acquisition unit 37.

[0056] In the next step S31, the camera control unit 35 switches the camera 76 from an active state to an inactive state after shooting. In the following step S32, the color setting unit 36 ​​returns the color of the scrap 72 from a predetermined color back to its original color, and the settings of the scrap 72 as seen by the camera 76, which were changed in step S27, are returned to their original settings. After executing the process in step S32, the simulation execution unit 26 moves on to step S33. In step S33, it is determined whether the fall time of the scrap 72 exceeds the set fall time Te, and after time Te has elapsed (step S33: Yes), the process moves on to step S25 shown in Figure 6A, and the subsequent processing continues.

[0057] In the simulation process described above, the processing of scrap 72 and camera 76 is performed individually for each piece of scrap 72 and each camera 76.

[0058] The aforementioned simulation of the scrap 72 falling motion is performed as a preliminary simulation prior to the actual simulation of scrap falling motion performed for the design of the press die. In this preliminary simulation, as described above, a dedicated camera 76 is positioned to photograph one corresponding piece of scrap 72 for each of the multiple pieces of scrap 72 generated from the workpiece 70. Each camera 76 moves in accordance with each piece of scrap 72 and photographs the scrap 72 individually when its velocity becomes zero, before the scrap 72 is discharged to the outside. Individual photography in this preliminary simulation provides image data based on individual photographs of the final falling state of the cut scrap.

[0059] The image data obtained in this way makes it easy to determine whether the final falling state is a real-world failure, such as the scrap 72 getting stuck in the scrap chute 58 and stopping, or a non-real-world failure, such as the scrap getting lodged in the scrap chute 58.

[0060] Figures 8A, 8B, and 8C are explanatory diagrams illustrating scrap 72 drop defects that occur in simulations but do not occur in actual press dies 52. For example, when the scrap 72 falls, it may hit the support part 54b that supports the scrap portion of the workpiece 70 from below, causing it to bounce upwards. In actual press dies 52, the bounced scrap 72 hits the upper die 56 and is returned to the scrap chute 58. However, if the presence and operation of the upper die 56 are not considered in order to reduce the computational load for the computer 11 simulation, the scrap 72 will ride up onto the lower die 54, as shown in Figure 8A. Also, if the time interval for calculating the drop speed of the scrap 72 is lengthened (i.e., the value of the number of frames N is reduced) in order to reduce the computational load for the simulation, the scrap 72 may get stuck in the scrap chute 58, as shown in Figure 8B. Furthermore, if the value of frame number N is reduced, as shown in Figure 8C, a phenomenon may occur where scrap 72, which is located at the position indicated by the dashed line in scene (n-1), slips through the scrap chute 58 in the subsequent scene n, as shown by the solid line.

[0061] In the image data acquisition method of this embodiment, a dedicated camera 76 provided for each scrap 72 can capture images from a certain angle of view of not only the actual possible drop defects but also impossible drop defects as shown in Figures 8A, 8B, and 8C, thereby acquiring image data indicating drop defects. In this way, the images of the individual state of each scrap 72 after dropping provide excellent information for determining whether or not a drop defect is present. As a result, after preliminary simulations, high-quality image data can be used for AI machine learning in the drop simulation for press die design that is actually performed.

[0062] Furthermore, in the image acquisition method of this embodiment, since the scrap 72 in a faulty fall state can be photographed each time from an upper position at a predetermined distance, the accuracy of determining whether the scrap 72 is in a faulty fall state can be improved. For example, when training an AI with machine learning, the information about the scrap 72 included in the image data is provided as unified upper position field of view information, thus improving the learning effect on the AI. Moreover, since the acquired image data contains only one piece of scrap, when training an AI with machine learning using this image data, the accuracy of the AI ​​in recognizing the scrap 72 in a faulty fall state included in the image data can be improved.

[0063] Furthermore, in the image data acquisition method of this embodiment, the acquired image data can be stored as historical information, thus reducing the work required for press die designers to constantly monitor the simulation and check the condition of the scrap 72 clogging. Also, since the amount of image data used for machine learning increases with each simulation, the accuracy of the trained model can be improved by performing AI machine learning using the large amount of acquired image data.

[0064] Next, we will explain how to create a trained model using the image data acquisition method described above. First, the computer 11 acquires image data showing the fall defects of the scrap 72 using the image acquisition method described above. Next, the creator of the trained model annotates the acquired image data, classifying whether the fall defects are actually possible or not. The annotation process can be performed by the creator by adding annotation data for classification to the image data in the computer 11 via the input unit 23. Next, the learning unit 38 of the computer 11 performs machine learning based on the annotated image data, i.e., the training data.

[0065] By using the trained model created in this way, the computer 11, which is equipped with an artificial intelligence learning unit 38, can determine whether a fall defect is one that could actually occur or one that could not actually occur.

[0066] When performing a simulation for press die design using a trained model, the information processing unit 24 of the simulation device 10 calculates the probability that the scrap 72 was properly ejected from the press die 52 based on the results of performing a simulation of dropping the scrap 72 M times (M is an integer of 2 or more). For example, if the number of simulations M = 200, and the number of times the scrap 72 was properly ejected Q = 180 (i.e., the number of times the scrap 72 failed to drop properly was 20), then the probability R = 180 / 200 × 100 (%). Also, if the number of times the scrap 72 failed to drop properly was I = 1, then this number is excluded from the probability calculation, and the probability R = 180 / (200-1) × 100 (%). The information processing unit 24 of the simulation device 10 then checks if the calculated probability R is equal to a predetermined threshold R th If it is less than the threshold R, the design quality is judged to be low, and the threshold R th In the above cases, the design quality is judged to be high. Threshold R th This can be set as appropriate by the input unit 22. The judgment result is output to the output unit 23 of the simulation device 10.

[0067] It should be noted that the present invention is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the invention. [Explanation of Symbols]

[0068] 10 Simulation device 11 Computer 20 Memory section 21 Communications Department 22 Input section 23 Output section 24 Information Processing Unit 26 Simulation Implementation Department 50 Pressing device 52 Press dies 58 Scrap chute 70 Work 72 Scrap 76 Cameras

Claims

1. A method for acquiring image data used to determine whether scrap generated from a workpiece processed by a press die is falling properly, The computer performs a simulation of the operation in which multiple pieces of scrap generated from the workpiece are discharged to the outside of the press die via a scrap chute, and in the simulation, The process involves placing a camera to photograph the corresponding scrap for each piece of scrap, Before the scrap is discharged outside the scrap chute, when the speed of the scrap becomes zero, the camera corresponding to the scrap photographs the scrap, and the computer acquires image data of the scrap. A method for acquiring image data, characterized by including [a specific element].

2. The method for acquiring image data according to claim 1, characterized in that the arrangement of each camera is set so that image data is acquired from an elevated position at a certain distance away from each piece of scrap.

3. The method for acquiring image data according to claim 1 or 2, characterized in that, in the step of acquiring the image data, the camera is set to show only one corresponding scrap among the plurality of scraps.

4. The method for acquiring image data according to claim 1 or 2, characterized in that each camera is set to an inactive state in which it is not possible to take pictures at the initial stage of each piece of scrap falling, and is switched from the inactive state to an active state in which it is possible to take pictures when the velocity of the corresponding piece of scrap becomes zero.

5. A method for creating a trained model using image data obtained by the acquisition method described in claim 1 or 2, A method for creating a trained model, characterized in that the creator annotates the aforementioned image data to determine whether the fall failure is actually possible or not, and then creates the trained model based on the resulting training data.

Images