Abrasive plane state estimation device, estimation program, and abrasive plane state estimation method

Abstract

To finely manage the state of an abrasive plane of a rotary grinding wheel.SOLUTION: An abrasive plane state estimation device 30 comprises an input part, a storage part 56, and an output part 57. The input part stores an input image 31 composed of a photographic image obtained by photographing an abrasive plane of a grinding wheel to be estimated in a first period. The storage part 56 stores a trained learning unit 32 subjected to machine learning by using training data 33 containing example data 331 and correct answer data 332. The example data 331 is composed of photographic images obtained by photographing the abrasive plane of the grinding wheel for training data acquisition in the first period. The correct answer data 332 is composed of photographic images obtained by photographing the abrasive plane of the grinding wheel for training data acquisition in a second period. The output part 57 outputs an output image 34 that shows an abrasive plane corresponding to the input image 31, the output image 34 showing the abrasive plane of the grinding wheel in the second period, from the trained learning unit 32, by using the input image 31 as input data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a grinding surface condition estimating device, an estimation program, and a grinding surface condition estimating method. [Background technology]

[0002] In grinding processing using a grinding device, the finish quality of the machined surface of a workpiece depends on the condition of the machining surface (so-called grinding surface) of the grinding wheel used as a grinding tool. Conventionally, in order to grasp the condition of the grinding surface of a grinding wheel, a measuring device for measuring the condition of the grinding surface has been proposed (for example, Patent Document 1). In the device described in Patent Document 1, the grinding surface is imaged using a metallurgical microscope equipped with a camera, and the condition of the grinding surface is measured based on the image data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-45078 Summary of the Invention [Problem to be solved by the invention]

[0004] In order to meticulously manage the condition of the grinding surface, there is a need for a technology that can retroactively check the condition of the grinding surface and a technology that can grasp the condition of the grinding surface in the near future. In this regard, the technology described in Patent Document 1 is a technology for grasping the current condition of the grinding surface, and therefore it is difficult to meet the above demands. [Means for solving the problem]

[0005] A grinding surface condition estimation device for solving the above problem is a grinding surface condition estimation device that estimates the condition of the grinding surface of a rotary grinding wheel, where a first period is a period when the amount of use of the rotary grinding wheel in a grinding process is a first predetermined amount, and a second period is a period when the amount of use is a second predetermined amount different from the first predetermined amount.The grinding surface condition estimation device includes: an input unit that stores input images consisting of captured images of the grinding surface of the rotary grinding wheel to be estimated at the first period; a memory unit that stores trained learning machines that have been machine-learned using teacher data that includes example data consisting of captured images of the grinding surface of the rotary grinding wheel for obtaining teacher data at the first period, and ground truth data consisting of captured images of the grinding surface of the rotary grinding wheel for obtaining teacher data at the second period; and an output unit that uses the input image as input data and outputs from the trained learning machine an image of the grinding surface that corresponds to the input image and that shows the grinding surface of the rotary grinding wheel at the second period.

[0006] The estimation program for solving the above problem is an estimation program that causes an electronic control device provided in the abrasive surface condition estimation device to execute the processing of each of the above parts provided in the abrasive surface condition estimation device.

[0007] The grinding surface condition estimation method for solving the above problem is a grinding surface condition estimation method for estimating the condition of the grinding surface of a rotary grinding wheel, and includes: an input process for storing, in an input unit of a grinding surface condition estimation device, an input image composed of an image of the grinding surface of the rotary grinding wheel to be estimated at the first time, where the first time is defined as a time when a first predetermined amount of the grinding wheel has been used in a grinding process, and the second time is a time when the amount of use is a second predetermined amount different from the first predetermined amount; a storage process for training a learning machine using teacher data including example data composed of an image of the grinding surface of the rotary grinding wheel for obtaining teacher data at the first time, and ground truth data composed of an image of the grinding surface of the rotary grinding wheel for obtaining teacher data at the second time, and then storing the trained learning machine in a memory unit of the grinding surface condition estimation device; and an output process for using the input image as input data and outputting, from the trained learning machine, an image of the grinding surface corresponding to the input image and showing the grinding surface of the rotary grinding wheel at the second time,

[0008] The above-described grinding surface condition estimation device, estimation program, and grinding surface condition estimation method can use a trained learning machine to acquire an image corresponding to the grinding surface of the grinding wheel to be estimated at a second time point, based on an input image of the grinding surface of the grinding wheel to be estimated at a first time point. This makes it possible to estimate and understand the condition of the grinding surface of the grinding wheel at a second time point, i.e., the past grinding surface condition or the future grinding surface condition, based on the condition of the grinding wheel at the first time point. This makes it possible to check the grinding surface condition of the grinding wheel retroactively and predict and understand the near future, thereby enabling detailed management of the grinding surface condition of the grinding wheel. [Effects of the Invention]

[0009] According to the present invention, the condition of the grinding surface of the grindstone can be meticulously controlled. [Brief explanation of the drawings]

[0010] [Figure 1]1 is a schematic diagram showing a schematic configuration of an automatic grinding device to which an embodiment of a grinding surface condition estimation device is applied; [Figure 2] FIG. 2 is a block diagram showing a schematic configuration of the grinding surface condition estimating device. [Figure 3] 1 is a schematic diagram showing the structure of a camera unit. [Figure 4] 1 is a schematic diagram showing an input image. [Figure 5] 1 is a schematic diagram showing training data. [Figure 6] FIG. 10 is an explanatory diagram for explaining first to fourth acquisition periods. [Figure 7] 10 is a table showing the relationship between a learning device and training data. [Figure 8] 10 is a flowchart showing the execution procedure of an image output process. [Figure 9] 1 is a schematic diagram showing the relationship between the grinding surface of a grindstone and the imaging range of a camera. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the grinding surface condition estimating device, the estimation program, and the grinding surface condition estimating method will be described with reference to FIGS. <Automatic grinding device 20> As shown in FIG. 1, the grinding surface condition estimation device 30 of this embodiment is applied to an automatic grinding machine 20. This automatic grinding machine 20 is a numerically controlled (NC) surface grinding machine that uses a rotary grinding wheel (hereinafter referred to as grinding wheel 21) to grind the surface of a workpiece (not shown). The grinding wheel 21 for grinding has an outer peripheral surface that forms a grinding surface 211. Specifically, the grinding wheel 21 has a structure in which a large number of abrasive grains are fixed to the outer peripheral surface of a cylindrical base material with a binder. The grinding wheel 21 is rotatably supported by a grinding wheel support part 22. The grinding wheel 21 and the grinding wheel support part 22 are movable relative to a base 23 in the vertical direction (hereinafter referred to as the Y direction) and the horizontal direction, specifically the depth direction in FIG. 1 (hereinafter referred to as the Z direction). In the automatic grinding machine 20, the grinding wheel 21 is brought into contact with the workpiece while being rotationally driven, thereby grinding the surface of the workpiece.

[0012] The automatic grinding device 20 has a table 24 as a support base for supporting a workpiece. The table 24 is movable in a horizontal direction perpendicular to both the Z direction and the Y direction, specifically in the left-right direction in FIG. 1 (hereinafter referred to as the X direction). In the automatic grinding device 20, the workpiece on the table 24 can be moved by moving the table 24 in the X direction. This allows the relative positions of the grinding wheel 21 and the workpiece in the X direction to be changed.

[0013] 1 and 2, the automatic grinding device 20 has an operation control unit 25 that controls the operation of the device. The operation control unit 25 has an X-axis drive unit 26, a Y-axis drive unit 27, a Z-axis drive unit 28, and a grindstone drive unit 29.

[0014] The automatic grinding device 20 can change the relative positions of the grinding wheel 21 and the workpiece in the X direction, the Y direction, and the Z direction, respectively, by moving the grinding wheel 21 in the Y direction and the Z direction and by moving the table 24 in the X direction. The X-axis drive unit 26 is a device for moving the table 24 in the X direction relative to the base 23. The X-axis drive unit 26 includes a ball screw mechanism, a servo motor, and the like (not shown). The X-direction position of the table 24 relative to the base 23 is controlled by controlling the operation of the X-axis drive unit 26. The Y-axis drive unit 27 is a device for moving the grinding wheel 21 and the grinding wheel support unit 22 in the Y direction relative to the base 23. The Y-axis drive unit 27 includes a ball screw mechanism, a servo motor, and the like (not shown). The Y-direction position of the grinding wheel 21 relative to the base 23 is controlled by controlling the operation of the Y-axis drive unit 27. The Z-axis drive unit 28 is a device for moving the grinding wheel 21 and the grinding wheel support unit 22 in the Z direction relative to the base 23. The Y-axis driving unit 27 has a ball screw mechanism, a servo motor, etc. (not shown). Through the operation control of the Y-axis driving unit 27, the position of the grinding wheel 21 in the Z direction relative to the base 23 is controlled.

[0015] The grinding wheel driving unit 29 includes an electric motor connected to the rotation shaft of the grinding wheel 21, a rotation phase sensor that detects the rotation phase of the grinding wheel 21, and the like. In the operation control of the grinding wheel driving unit 29 (more specifically, the electric motor), the grinding wheel 21 is basically rotated at a constant speed. The rotation speed of the grinding wheel 21 is set based on the processing conditions input to the automatic grinding device 20 and is a speed that matches the processing conditions.

[0016] <Grinding surface condition estimation device 30> The grinding surface condition estimating device 30 of this embodiment will be described in detail below. The grinding surface condition estimating device 30 is a device for estimating the condition of the grinding surface 211 of the grinding wheel 21. The grinding surface condition estimating device 30 outputs images (hereinafter, output images 34) showing the grinding surface 211 of the grinding wheel 21 at predetermined times, more specifically, at a first acquisition time T1, a second acquisition time T2, a third acquisition time T3, and a fourth acquisition time T4, all of which will be described later. Based on these output images 34, it is possible to confirm, for example, the condition of the grinding surface 211 of the grinding wheel 21. The grinding surface condition estimating device 30 includes a camera unit 40 and an electronic control device 50.

[0017] <Camera Unit 40> 1 to 3, the camera unit 40 captures an image of the grinding surface 211 of the grindstone 21. The camera unit 40 includes a camera 41 as an imaging unit, and an illumination unit 43.

[0018] As shown in FIGS. 1 and 3, the camera 41 is disposed below the grinding wheel 21. The camera 41 captures an image of the grinding surface 211 of the grinding wheel 21 from a direction perpendicular to the grinding surface 211 (more specifically, the tangent plane of the grinding surface 211 indicated by the dashed line in FIG. 3). In this embodiment, the camera 41 captures an image of the grinding surface 211 while the grinding wheel 21 is rotating at a predetermined speed VW. The captured image captured by the camera 41 is stored as an input image 31 in the electronic control device 50, which will be described later. In this embodiment, the grinding wheel 21 attached to the automatic grinding machine 20 at this time corresponds to the rotating grinding wheel to be estimated.

[0019] As shown in FIG. 3, the illumination unit 43 illuminates the lower end portion of the grinding surface 211, i.e., the area to be imaged by the camera 41 (hereinafter, referred to as the image-capturing area RE). The illumination unit 43 includes a first illumination unit 45 and a second illumination unit 46. The first illumination unit 45 irradiates the image-capturing area RE of the grinding surface 211 with first illumination light 45L having a first emission color (e.g., blue) from a direction perpendicular to the grinding surface 211 (more specifically, the tangent plane of the grinding surface 211 shown by the dashed-dotted line in FIG. 3). The second illumination unit 46 irradiates the image-capturing area RE of the grinding surface 211 with second illumination light 46L having a second emission color (e.g., white) different from the first emission color from a direction oblique to the grinding surface 211 (more specifically, the tangent plane of the grinding surface 211 shown by the dashed-dotted line in FIG. 3).

[0020] In this embodiment, the camera unit 40 is detachable from the automatic grinding apparatus 20. The camera unit 40 is attached to the automatic grinding apparatus 20 when imaging the grinding surface 211. In the camera unit 40, when imaging the grinding surface 211 using the camera 41, the illumination unit 43 illuminates the imaging target area RE of the grinding surface 211. Note that in this embodiment, when grinding is performed by the automatic grinding apparatus 20, the camera unit 40 is detached from the automatic grinding apparatus 20.

[0021] <Electronic control device 50> The electronic control device 50 executes various controls for acquiring an image showing the grinding surface 211 of the grinding wheel 21 .

[0022] The electronic control device 50 includes a PU 51, a ROM 52, a RAM 53, an operation input unit 54, a display unit 55, and a memory unit 56. The PU 51 is a processing unit such as a CPU, a GPU, and a TPU. The operation input unit 54 is an input device such as a mouse, a keyboard, a touch panel, and buttons. The operation input unit 54 is operated when configuring various settings for the grinding surface condition estimation device 30 or when operating the grinding surface condition estimation device 30. The display unit 55 includes a display device such as an LCD display or an organic EL display. The display unit 55 displays various information related to the output of the output image 34, such as the output image 34 showing the grinding surface 211 output through the image output process described below. The memory unit 56 is composed of a non-volatile memory such as a hard disk drive or a solid-state drive that can be written to and read from as needed. The memory unit 56 includes a memory area for storing various execution programs 35, a memory area for storing training data 33 used in machine learning by the learner 32, and a memory area for storing the learned learner 32. The storage unit 56 also has a storage area as an input unit that stores the input image 31 configured from a captured image of the grinding surface 211 of the grinding wheel 21 that is the estimation target. The execution program 35 includes a program for causing the learning device 32 to execute machine learning, and an estimation program for causing the learning device 32 to execute various processes related to the state estimation of the grinding surface 211 of the grinding wheel 21.

[0023] The electronic control device 50 includes an output unit 57 as a functional unit. The output unit 57 receives the input image 31 from the first period as input data, and outputs, from the trained learning device 32 stored in the memory unit 56, an image showing the grinding surface 211 corresponding to the input image 31, and also showing the grinding surface 211 from the second period (more specifically, an output image 34).

[0024] <Image capture processing> In this embodiment, in order to acquire the input image 31, an imaging process is executed in which the camera 41 captures an image of the grinding surface 211 of the grindstone 21.

[0025] In the imaging process, first, the grinding wheel 21 to be estimated is attached to the automatic grinding device 20, and the grinding wheel 21 is rotated at a predetermined speed VW. Then, in this state, an image of the grinding surface 211 is captured by the camera 41. In this embodiment, the rotational drive pattern of the grinding wheel 21, the position control pattern of the camera 41, and the imaging pattern of the camera 41, which enable the camera 41 to efficiently capture the entire surface of the grinding surface 211, are determined in advance and stored in the electronic control device 50.

[0026] Possible methods for storing the above patterns in the memory unit 56 include the following (Method A) and (Method B). (Method A) After the user of the automatic grinding device 20 installs the grinding wheel 21 that will actually be used, the user operates the automatic grinding device 20 to determine the above patterns suitable for the grinding wheel 21, and stores these patterns in the electronic control device 50. (Method B) Based on the results of various experiments and simulations, the manufacturer of the automatic grinding device 20 determines in advance the above patterns that will enable efficient imaging of the entire grinding surface 211 of the grinding wheel 21. Data related to these patterns is then stored in the automatic grinding device 20 before shipment, or in an automatic grinding device 20 installed in the user's factory.

[0027] When capturing an image of the grinding surface 211 of the grinding wheel 21, the operation of the grinding wheel driver 29 is controlled based on the rotational drive pattern of the grinding wheel 21, and the operation of the X-axis driver 26, the Y-axis driver 27, and the Z-axis driver 28 is controlled based on the position control pattern of the camera 41. Furthermore, the camera 41 captures an image of the grinding surface 211 based on the image capturing pattern of the camera 41. The electronic control device 50 then captures image data of the grinding surface 211 captured by the camera 41 and stores the image data in the memory 56. Specifically, the camera 41 captures images of the grinding surface 211 multiple times. The captured images are then combined in a predetermined order, vertically and horizontally, to form a single image (input image 31) showing the entire grinding surface 211, as shown in FIG. 4 . The electronic control device 50 stores the input image 31 thus generated in the memory 56.

[0028] In the grinding surface condition estimation device 30 of this embodiment, it is possible to estimate the past condition of the grinding surface 211 and the future condition of the grinding surface 211 using a trained learning device 32 based on the input image 31, which is composed of an image of the grinding surface 211.

[0029] <Learning Module 32> In the grinding surface condition estimation device 30, a plurality of trained learning modules 32 are stored in the memory unit 56 of the electronic control device 50. Each of the plurality of learning modules 32 employs a generative adversarial network (GAN), specifically an image generation model known as "pix2pix." The learning modules 32 undergo machine learning using training data 33. Specifically, the plurality of learning modules 32 undergo machine learning using different training data 33. The training data 33 is data used for performing machine learning on the learning modules 32 so that the learning modules 32 acquire the ability to generate, based on an input image 31, an image showing the grinding surface 211 at a time different from the time the input image 31 was acquired.

[0030] <Teacher Data 33> As shown in FIG. 5, the teacher data 33 is composed of image data sets in which example data 331 and supervised answer data 332 are paired. In this embodiment, a large number of image data sets in which example data 331 and supervised answer data 332 are paired are created. Then, as shown in FIG. 2, these image data sets are used as teacher data 33 for the learner 32. More specifically, the electronic control device 50 extracts one of the multiple image data sets that make up the teacher data 33 and uses this as input data to cause the learner 32 to perform machine learning. This learning process is repeatedly performed for each image data set that makes up the teacher data 33. Then, the machine learning model constructed through the execution of machine learning is stored in the storage unit 56 as the trained learner 32. Note that in this embodiment, the process of storing the trained learner 32 in the storage unit 56 corresponds to the storage process.

[0031] In this embodiment, both the example data 331 and the correct answer data 332 are image data composed of captured images of the grinding surface 211. In this embodiment, the example data 331 and the correct answer data 332 that constitute the same teacher data 33 are composed of captured images of the same grinding surface 211 captured at different times.

[0032] As shown in FIG. 6 , in this embodiment, a first acquisition time T1, a second acquisition time T2, a third acquisition time T3, and a fourth acquisition time T4 are defined as the times for acquiring the captured images constituting the sample data 331 and the captured images constituting the answer data 332. In this embodiment, the amount of grinding of the workpiece by the grinding wheel 21 after dressing of the grinding wheel 21 (hereinafter, the grinding amount) is defined as the “usage amount of the grinding wheel 21 used in grinding.” The first acquisition time T1 is defined as, for example, a time when the grinding amount is “0,” i.e., a time immediately after dressing of the grinding wheel 21. The second acquisition time T2 is defined as, for example, a time when the grinding amount becomes “V (e.g., 8000 cubic millimeters).” The third acquisition time T3 is defined as, for example, a time when the grinding amount becomes “V × 2.” The fourth acquisition time T4 is defined as, for example, a time when the grinding amount becomes “V × 3.”

[0033] <Example data 331> The example data 331 can be generated, for example, by the following procedure. First, the grinding wheel 21 for creating the training data is attached to the automatic grinding device 20 (FIG. 1). Then, the imaging process is executed in this state. The captured image acquired through this imaging process is used as the example data 331. Alternatively, it is also possible to use a separate device different from the automatic grinding device 20 to capture an image of the grinding surface 211 of the grinding wheel 21 while rotating the grinding wheel 21 at a predetermined speed VW, and use this captured image as the example data 331. As the example data 331, a single image showing the entire grinding surface 211 is generated.

[0034] In this embodiment, example images EX1 to EX4 composed of captured images of the grinding surface 211 are acquired at each of acquisition times T1 to T4 as image data used for the example data 331. Specifically, the first example image EX1 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the first acquisition time T1. The second example image EX2 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the second acquisition time T2. The third example image EX3 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the third acquisition time T3. The fourth example image EX4 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the fourth acquisition time T4.

[0035] <Correct answer data 332> The correct answer data 332 can be generated, for example, by the following procedure. First, the grinding wheel 21 for creating the training data is attached to the automatic grinding device 20. Then, with the grinding wheel 21 stopped or rotating at a relatively slow speed (slower than the predetermined speed VW), the camera 41 captures an image of the grinding surface 211. Specifically, the operation of the grinding wheel driver 29 is controlled based on a predetermined rotational drive pattern of the grinding wheel 21, and the operation of the X-axis driver 26, the Y-axis driver 27, and the Z-axis driver 28 is controlled based on a predetermined position control pattern of the camera 41. Furthermore, the camera 41 captures an image of the grinding surface 211 based on a predetermined imaging pattern of the camera 41. Then, the electronic control device 50 captures image data of the grinding surface 211 captured by the camera 41 and stores the image data in the memory 56. Specifically, the camera 41 captures images of the grinding surface 211 multiple times so that the entire grinding surface 211 is in focus. Then, by combining the captured images in a state where they are arranged lengthwise and widthwise in a predetermined order, a single image (correct data 332) showing the entire grinding surface 211 is formed.

[0036] Alternatively, a separate device different from the automatic grinding device 20 may be used to capture images of the grinding surface 211 with the camera 41 while the grinding wheel 21 is stopped from rotating or while the grinding wheel 21 is rotating at a relatively slow speed. In this case, the camera 41 captures images of the grinding surface 211 multiple times so that the entire grinding surface 211 is in focus. The captured images are then combined in a state where they are arranged vertically and horizontally in a predetermined order, thereby forming a single image (correct answer data 332) showing the entire grinding surface 211.

[0037] In this embodiment, as image data used for the supervised data 332, supervised images CA1 to CA4 composed of captured images of the grinding surface 211 are acquired for each of the acquisition times T1 to T4. Specifically, the first supervised image CA1 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the first acquisition time T1. The second supervised image CA2 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the second acquisition time T2. The third supervised image CA3 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the third acquisition time T3. The fourth supervised image CA4 is composed of a captured image of the grinding surface 211 of the grinding wheel 21 used to create the teacher data at the fourth acquisition time T4.

[0038] In this embodiment, for each of the multiple learning devices 32, an image data set is set as teacher data 33, with one of the example images EX1 to EX4 as example data 331 and one of the correct images CA1 to CA4 as correct data 332. However, in this embodiment, as described above, the example data 331 and the correct data 332 that make up the same teacher data 33 are composed of captured images of the grinding surface 211 taken at different times. Each of the multiple learning devices 32 will be described in detail below.

[0039] <Learning Unit 32[1-2]> 7, the learning device 32[1-2] performs machine learning using training data 33[1-2] in which the first example image EX1 is used as example data 331 and the second correct image CA2 is used as correct data 332[1-2]. The learned learning device 32[1-2] is stored in the memory unit 56 of the electronic control device 50.

[0040] The trained learning device 32[1-2] uses the input image 31 at the first acquisition time T1, i.e., the input image 31 showing the grinding surface 211 at the first acquisition time T1, as input data to generate and output an output image 34 showing the grinding surface 211 at the second acquisition time T2. By using the trained learning device 32[1-2], it is possible to estimate the state of the grinding surface 211 at the second acquisition time T2, when the grinding amount has reached "V," based on the input image 31 showing the grinding surface 211 at the first acquisition time T1 immediately after dressing the grinding wheel 21. In this case, the first acquisition time T1 corresponds to a first time when the amount of use of the grinding wheel 21 in the grinding process is a first predetermined amount. The second acquisition time T2 corresponds to a second time when the amount of use is a second predetermined amount different from the first predetermined amount.

[0041] <Learning Unit 32[1-3]> The learning device 32[1-3] performs machine learning using the training data 33[1-3], which uses the first example image EX1 as example data 331 and the third correct image CA3 as correct data 332[1-3]. The trained learning device 32[1-3] is stored in the memory unit 56 of the electronic control device 50.

[0042] The trained learning device 32[1-3] uses the input image 31 at the first acquisition time T1, i.e., the input image 31 showing the grinding surface 211 at the first acquisition time T1, as input data to generate and output an output image 34 showing the grinding surface 211 at the third acquisition time T3. By using the trained learning device 32[1-3], it is possible to estimate the state of the grinding surface 211 at the third acquisition time T3, when the grinding amount has become "V x 2", based on the input image 31 showing the grinding surface 211 at the first acquisition time T1 immediately after dressing.

[0043] In this case, the first acquisition time T1 corresponds to the first time period, and the third acquisition time T3 corresponds to the second time period. Furthermore, when the correct answer data 332[1-2] is the first correct answer data, the teacher data 33[1-2] is the first teacher data, and the learning device 32[1-2] is the first learning device, the following occurs. In this case, the third acquisition time T3 corresponds to the third time period when the usage amount is a third predetermined amount different from the first predetermined amount and the second predetermined amount. In this case, the correct answer data 332[1-3] corresponds to the second correct answer data, the teacher data 33[1-3] corresponds to the second teacher data, and the learning device 32[1-3] corresponds to the second learning device.

[0044] <Learning Unit 32[1-4]> The learning devices 32[1-4] are machine-learned using the training data 33[1-4], which uses the first example image EX1 as example data 331 and the fourth correct answer image CA4 as correct answer data 332[1-4]. The trained learning devices 32[1-4] are stored in the memory unit 56 of the electronic control device 50.

[0045] The trained learning device 32[1-4] uses the input image 31 at the first acquisition time T1, i.e., the input image 31 showing the grinding surface 211 at the first acquisition time T1, as input data to generate and output an output image 34 showing the grinding surface 211 at the fourth acquisition time T4. By using the trained learning device 32[1-4], it is possible to estimate the state of the grinding surface 211 at the fourth acquisition time T4, when the grinding amount has become "V x 3", based on the input image 31 showing the grinding surface 211 at the first acquisition time T1 immediately after dressing.

[0046] In this case, the first acquisition time T1 corresponds to the first period, and the fourth acquisition time T4 corresponds to the second period. Furthermore, when the correct answer data 332[1-2] is the first correct answer data, the teacher data 33[1-2] is the first teacher data, and the learning device 32[1-2] is the first learning device, the following occurs: In this case, the fourth acquisition time T4 corresponds to the third period, the correct answer data 332[1-4] corresponds to the second correct answer data, the teacher data 33[1-4] corresponds to the second teacher data, and the learning device 32[1-4] corresponds to the second learning device.

[0047] <Learning Unit 32[2-1]> The learning device 32[2-1] is machine-learned using teacher data 33[2-1], which uses the second example image EX2 as example data 331 and the first correct image CA1 as correct data 332[2-1]. The trained learning device 32[2-1] is stored in the memory unit 56 of the electronic control device 50.

[0048] The trained learning device 32[2-1] uses the input image 31 at the second acquisition time T2, i.e., the input image 31 showing the grinding surface 211 at the second acquisition time T2, as input data to generate and output an output image 34 showing the grinding surface 211 at the first acquisition time T1. By using the trained learning device 32[2-1], it is possible to estimate the state of the grinding surface 211 at the first acquisition time T1 immediately after dressing, based on the input image 31 showing the grinding surface 211 at the second acquisition time T2 when the grinding amount has reached "V." In this case, the second acquisition time T2 corresponds to the first time period, and the first acquisition time T1 corresponds to the second time period.

[0049] <Learning Unit 32[2-3]> The learning device 32[2-3] is machine-learned using the training data 33[2-3], which uses the second example image EX2 as example data 331 and the third correct image CA3 as correct data 332[2-3]. The trained learning device 32[2-3] is stored in the memory unit 56 of the electronic control device 50.

[0050] The trained learning device 32[2-3] uses the input image 31 at the second acquisition time T2, i.e., the input image 31 showing the grinding surface 211 at the second acquisition time T2, as input data to generate and output an output image 34 showing the grinding surface 211 at the third acquisition time T3. By using the trained learning device 32[2-3], it is possible to estimate the state of the grinding surface 211 at the third acquisition time T3 when the grinding amount has become "V x 2" based on the input image 31 showing the grinding surface 211 at the second acquisition time T2 when the grinding amount has become "V".

[0051] In this case, the second acquisition time T2 corresponds to the first time period, and the third acquisition time T3 corresponds to the second time period. Furthermore, when the correct answer data 332[2-1] is the first correct answer data, the teacher data 33[2-1] is the first teacher data, and the learning device 32[2-1] is the first learning device, the following occurs: In this case, the third acquisition time T3 corresponds to the third time period, the correct answer data 332[2-3] corresponds to the second correct answer data, the teacher data 33[2-3] corresponds to the second teacher data, and the learning device 32[2-3] corresponds to the second learning device.

[0052] <Learning Unit 32[2-4]> The learning device 32[2-4] is machine-learned using the training data 33[2-4], which uses the second example image EX2 as example data 331 and the fourth correct answer image CA4 as correct answer data 332[2-4]. The trained learning device 32[2-4] is stored in the memory unit 56 of the electronic control device 50.

[0053] The trained learning device 32[2-4] uses the input image 31 at the second acquisition time T2, i.e., the input image 31 showing the grinding surface 211 at the second acquisition time T2, as input data to generate and output an output image 34 showing the grinding surface 211 at the fourth acquisition time T4. By using the trained learning device 32[2-4], it is possible to estimate the state of the grinding surface 211 at the fourth acquisition time T4 when the grinding amount has become "V x 3" based on the input image 31 showing the grinding surface 211 at the second acquisition time T2 when the grinding amount has become "V".

[0054] In this case, the second acquisition time T2 corresponds to the first period, and the fourth acquisition time T4 corresponds to the second period. Furthermore, when the correct answer data 332[2-1] is the first correct answer data, the teacher data 33[2-1] is the first teacher data, and the learning device 32[2-1] is the first learning device, the following occurs: In this case, the fourth acquisition time T4 corresponds to the third period, the correct answer data 332[2-4] corresponds to the second correct answer data, the teacher data 33[2-4] corresponds to the second teacher data, and the learning device 32[2-4] corresponds to the second learning device.

[0055] <Learning Unit 32[3-1]> The learning device 32[3-1] is machine-learned using training data 33[3-1], which uses the third example image EX3 as example data 331 and the first correct image CA1 as correct data 332[3-1]. The trained learning device 32[3-1] is stored in the memory unit 56 of the electronic control device 50.

[0056] The trained learning device 32[3-1] uses the input image 31 at the third acquisition time T3, i.e., the input image 31 showing the grinding surface 211 at the third acquisition time T3, as input data to generate and output an output image 34 showing the grinding surface 211 at the first acquisition time T1. By using the trained learning device 32[3-1], it is possible to estimate the state of the grinding surface 211 at the first acquisition time T1 immediately after dressing, based on the input image 31 showing the grinding surface 211 at the third acquisition time T3 when the grinding amount has reached "V x 2." In this case, the third acquisition time T3 corresponds to the first period, and the first acquisition time T1 corresponds to the second period.

[0057] <Learning Unit 32[3-2]> The learning device 32[3-2] is machine-learned using the training data 33[3-2], which uses the third example image EX3 as example data 331 and the second correct image CA2 as correct data 332[3-2]. The trained learning device 32[3-2] is stored in the memory unit 56 of the electronic control device 50.

[0058] The trained learning device 32[3-2] uses the input image 31 at the third acquisition time T3, i.e., the input image 31 showing the grinding surface 211 at the third acquisition time T3, as input data to generate and output an output image 34 showing the grinding surface 211 at the second acquisition time T2. By using the trained learning device 32[3-2], it is possible to estimate the state of the grinding surface 211 at the second acquisition time T2 when the grinding amount became "V" based on the input image 31 showing the grinding surface 211 at the third acquisition time T3 when the grinding amount became "V x 2".

[0059] In this case, the third acquisition time T3 corresponds to the first time period, and the second acquisition time T2 corresponds to the second time period. Furthermore, when the correct answer data 332[3-1] is the first correct answer data, the teacher data 33[3-1] is the first teacher data, and the learning device 32[3-1] is the first learning device, the following occurs: In this case, the second acquisition time T2 corresponds to the third time period, the correct answer data 332[3-2] corresponds to the second correct answer data, the teacher data 33[3-2] corresponds to the second teacher data, and the learning device 32[3-2] corresponds to the second learning device.

[0060] <Learning Unit 32[3-4]> The learning device 32[3-4] performs machine learning using the training data 33[3-4], which uses the third example image EX3 as example data 331 and the fourth correct answer image CA4 as correct answer data 332[3-4]. The trained learning device 32[3-4] is stored in the memory unit 56 of the electronic control device 50.

[0061] The trained learning device 32[3-4] uses the input image 31 at the third acquisition time T3, i.e., the input image 31 showing the grinding surface 211 at the third acquisition time T3, as input data to generate and output an output image 34 showing the grinding surface 211 at the fourth acquisition time T4. By using the trained learning device 32[3-4], it is possible to estimate the state of the grinding surface 211 at the fourth acquisition time T4, when the grinding amount has become "V × 3", based on the input image 31 showing the grinding surface 211 at the third acquisition time T3, when the grinding amount has become "V × 2".

[0062] In this case, the third acquisition time T3 corresponds to the first period, and the fourth acquisition time T4 corresponds to the second period. Furthermore, when the correct answer data 332[3-1] is the first correct answer data, the teacher data 33[3-1] is the first teacher data, and the learning device 32[3-1] is the first learning device, the following occurs: In this case, the fourth acquisition time T4 corresponds to the third period, the correct answer data 332[3-4] corresponds to the second correct answer data, the teacher data 33[3-4] corresponds to the second teacher data, and the learning device 32[3-4] corresponds to the second learning device.

[0063] <Learning Unit 32[4-1]> The learning device 32[4-1] performs machine learning using the training data 33[4-1], which uses the fourth example image EX4 as example data 331 and the first correct image CA1 as correct data 332[4-1]. The trained learning device 32[4-1] is stored in the memory unit 56 of the electronic control device 50.

[0064] The trained learning device 32[4-1] uses the input image 31 at the fourth acquisition time T4, i.e., the input image 31 showing the grinding surface 211 at the fourth acquisition time T4, as input data to generate and output an output image 34 showing the grinding surface 211 at the first acquisition time T1. By using the trained learning device 32[4-1], it is possible to estimate the state of the grinding surface 211 at the first acquisition time T1 immediately after dressing, based on the input image 31 showing the grinding surface 211 at the fourth acquisition time T4, when the grinding amount has reached "V x 3." In this case, the fourth acquisition time T4 corresponds to the first period, and the first acquisition time T1 corresponds to the second period.

[0065] <Learning Unit 32[4-2]> The learning device 32[4-2] is machine-learned using the training data 33[4-2], which uses the fourth example image EX4 as example data 331 and the second correct answer image CA2 as correct answer data 332[4-2]. The trained learning device 32[4-2] is stored in the memory unit 56 of the electronic control device 50.

[0066] The trained learning device 32[4-2] uses the input image 31 at the fourth acquisition time T4, i.e., the input image 31 showing the grinding surface 211 at the fourth acquisition time T4, as input data to generate and output an output image 34 showing the grinding surface 211 at the second acquisition time T2. By using the trained learning device 32[4-2], it is possible to estimate the state of the grinding surface 211 at the second acquisition time T2 when the grinding amount became "V" based on the input image 31 showing the grinding surface 211 at the fourth acquisition time T4 when the grinding amount became "V x 3".

[0067] In this case, the fourth acquisition time T4 corresponds to the first time period, and the second acquisition time T2 corresponds to the second time period. Furthermore, when the correct answer data 332[4-1] is the first correct answer data, the teacher data 33[4-1] is the first teacher data, and the learning device 32[4-1] is the first learning device, the following occurs: In this case, the second acquisition time T2 corresponds to the third time period, the correct answer data 332[4-2] corresponds to the second correct answer data, the teacher data 33[4-2] corresponds to the second teacher data, and the learning device 32[4-2] corresponds to the second learning device.

[0068] <Learning Unit 32[4-3]> The learning device 32[4-3] is machine-learned using the training data 33[4-3], which uses the fourth example image EX4 as example data 331 and the third correct image CA3 as correct data 332[4-3]. The trained learning device 32[4-3] is stored in the memory unit 56 of the electronic control device 50.

[0069] The trained learning device 32[4-3] uses the input image 31 at the fourth acquisition time T4, i.e., the input image 31 showing the grinding surface 211 at the fourth acquisition time T4, as input data to generate and output an output image 34 showing the grinding surface 211 at the third acquisition time T3. By using the trained learning device 32[4-3], it is possible to estimate the state of the grinding surface 211 at the third acquisition time T3, when the grinding amount became "V x 2", based on the input image 31 showing the grinding surface 211 at the fourth acquisition time T4, when the grinding amount became "V x 3".

[0070] In this case, the fourth acquisition time T4 corresponds to the first time period, and the third acquisition time T3 corresponds to the second time period. Furthermore, when the correct answer data 332[4-1] is the first correct answer data, the teacher data 33[4-1] is the first teacher data, and the learning device 32[4-1] is the first learning device, the following occurs: In this case, the third acquisition time T3 corresponds to the third time period, the correct answer data 332[4-3] corresponds to the second correct answer data, the teacher data 33[4-3] corresponds to the second teacher data, and the learning device 32[4-3] corresponds to the second learning device.

[0071] <Image output processing> The process of outputting the output image 34 showing the grinding surface 211 of the grinding wheel 21 in the second period (image output process) will be specifically described below. Fig. 8 shows the execution procedure of the image output process. A series of processes shown in the flowchart of Fig. 8 is executed by the electronic control device 50 as processes at predetermined intervals.

[0072] 8, in this process, first, an input image 31 formed by capturing an image of the grinding surface 211 of the grinding wheel 21 attached to the automatic grinding device 20 is stored in the storage unit 56 (step S11). Specifically, the imaging process is executed. Then, the input image 31 is formed based on the image data acquired by the imaging process, and the input image 31 is stored in the storage unit 56. In this embodiment, the process of step S11 corresponds to an input step.

[0073] Thereafter, a learning device 32 to be used for outputting the output image 34 is selected (step S12). Specifically, through operation of the operation input unit 54, a learning device 32 is selected that corresponds to the time when the input image 31 was acquired and the time of the subject of estimation for estimating the state of the grinding surface 211. For example, if the input image 31 was acquired at the first acquisition time T1 and the time of the subject of estimation is the second acquisition time T2, the learning device 32[1-2] is selected.

[0074] Thereafter, the input image 31 stored in the process of step S11 is used as input data, and the trained learning device 32 selected in the process of step S12 generates and outputs an output image 34 showing the grinding surface 211 corresponding to the input image 31 (step S13). In this embodiment, the process of step S13 corresponds to the output step.

[0075] An example of an execution mode of the image output process will be described in detail below, in which the input image 31 is acquired and the output image 34 is output at the second acquisition time T2.

[0076] In this example, when the second acquisition time T2 arrives, the grinding process by the automatic grinding device 20 is stopped. Then, the camera unit 40 is attached to the automatic grinding device 20. Thereafter, when the operation input unit 54 is operated to acquire the input image 31, the grinding surface condition estimation device 30 uses the automatic grinding device 20 to form and acquire the input image 31. Then, a learning device 32 is selected through operation of the operation input unit 54. In this example, three learning devices 32[2-1], 32[2-3], and 32[2-4] are selected.

[0077] As a result, using the input image 31 as input data, the learning device 32[2-1] outputs an output image 34 that shows the grinding surface 211 corresponding to the input image 31 and that shows the grinding surface 211 of the grinding wheel 21 at the first acquisition time T1. Based on this output image 34, it is possible to estimate and understand the state of the grinding surface 211 of the grinding wheel 21 at the first acquisition time T1, i.e., the state of the grinding surface 211 in the past. Furthermore, based on the understood state of the grinding surface 211, it is also possible to investigate the machining state of the workpiece, etc.

[0078] Furthermore, using the input image 31 as input data, the learning device 32 [2-3] outputs an output image 34 that shows the grinding surface 211 corresponding to the input image 31 and that shows the grinding surface 211 of the grinding wheel 21 at the third acquisition time T3. Based on this output image 34, it is possible to estimate and grasp the state of the grinding surface 211 of the grinding wheel 21 at the third acquisition time T3, i.e., the future state of the grinding surface 211. Furthermore, because the machining state of the workpiece can be estimated based on the state of the grinding surface 211 estimated in this way, subsequent workpiece inspection can be easily performed.

[0079] Furthermore, using the input image 31 as input data, the learning device 32[2-4] outputs an output image 34 that shows the grinding surface 211 corresponding to the input image 31 and that shows the grinding surface 211 of the grinding wheel 21 at the fourth acquisition time T4. Based on this output image 34, it is possible to estimate and grasp the state of the grinding surface 211 of the grinding wheel 21 at the fourth acquisition time T4, i.e., the future state of the grinding surface 211.

[0080] <Actions and Effects of This Embodiment> The operation and effects of this embodiment will be described. (1) The grinding surface condition estimation device 30 includes an input unit, a memory unit 56, and an output unit 57. The input unit stores an input image 31 composed of a captured image of the grinding surface 211 of the grinding wheel 21 to be estimated at the first time point. The memory unit 56 stores a trained learner 32 that has undergone machine learning using teacher data 33 including example data 331 and ground truth data 332. The example data 331 consists of captured images of the grinding surface 211 of the grinding wheel 21 used to acquire the teacher data at the first time point. The ground truth data 332 consists of captured images of the grinding surface 211 of the grinding wheel 21 used to acquire the teacher data at the second time point. The output unit 57 uses the input image 31 as input data and outputs, from the trained learner 32, an output image 34 that shows the grinding surface 211 corresponding to the input image 31 and that also shows the grinding surface 211 of the grinding wheel 21 at the second time point.

[0081] According to the above configuration, based on an input image 31, which is an image of the grinding surface 211 of the grinding wheel 21 to be estimated at a first time period, an output image 34 corresponding to the grinding surface 211 of the grinding wheel 21 to be estimated at a second time period can be obtained using a trained learning device 32.

[0082] As a result, based on the state of the grinding surface 211 of the grinding wheel 21 at the first time period, it is possible to estimate and grasp the state of the grinding surface 211 of the grinding wheel 21 at the second time period, i.e., the state of the grinding surface 211 in the past or the state of the grinding surface 211 in the future. In this way, with the above configuration, it is possible to check the state of the grinding surface 211 of the grinding wheel 21 retroactively and predict and grasp the state of the grinding surface 211 in the near future, so it is possible to meticulously manage the state of the grinding surface 211 of the grinding wheel 21. Moreover, it is also possible to meticulously manage the processing state of the workpiece ground by this grinding wheel 21.

[0083] Here, the deterioration of the grinding wheel 21 progresses such that the proportion of the surface constituting the outermost periphery of the grinding surface 211 (hereinafter referred to as the outermost periphery surface) in the entire grinding surface 211 gradually increases. Therefore, by checking the proportion based on the output image 34, it is possible to estimate and grasp the degree of deterioration of the grinding wheel 21 in the second period.

[0084] In this embodiment, the output image 34 is generated using a trained learning device 32, so the grinding surface 211 shown in the output image 34 will not be exactly the same as the actual grinding surface 211 of the grinding wheel 21 being estimated at the second time point. However, in this embodiment, the inventors have confirmed the following: The grinding surface 211 shown in the output image 34 is a grinding surface 211 from which the proportion of the outermost peripheral surface can be determined with high accuracy. Therefore, according to this embodiment, the proportion can be determined with high accuracy based on the output image 34, and the degree of deterioration of the grinding wheel 21 at the second time point can be determined with high accuracy.

[0085] (2) Each of the input image 31, the example data 331, and the correct answer data 332 is a single image showing the entire grinding surface 211. According to the above configuration, the state of the entire grinding surface 211 can be estimated all at once with a simple configuration in which a single input image 31 capturing the entire grinding surface 211 is used as input data.

[0086] Here, it is also conceivable that the input image 31, the example data 331, and the correct answer data 332 are each composed of multiple images. In this case, the output image 34 is also composed of multiple images. Therefore, misalignment is likely to occur at the boundaries between the multiple images that make up the output image 34, which can reduce the accuracy of estimating the condition of the grinding surface 211. In this regard, the above configuration can suppress the decrease in accuracy that results from being composed of multiple images, thereby enabling the condition of the grinding surface 211 to be estimated with high accuracy.

[0087] (3) The storage unit 56 stores a trained first learning device that has been machine-learned using first teacher data including the example data 331 and first supervised data configured from captured images of the grinding surface 211 of the grinding wheel 21 used to obtain teacher data at a second time. The output unit 57 uses the input image 31 as input data and outputs, from the trained first learning device, an output image 34 that shows the grinding surface 211 corresponding to the input image 31 and that shows the grinding surface 211 of the grinding wheel 21 at the second time. The storage unit 56 also stores a trained second learning device that has been machine-learned using second teacher data including the example data 331 and second supervised data configured from captured images of the grinding surface 211 of the grinding wheel 21 used to obtain teacher data at a third time. The output unit 57 uses the input image 31 as input data and outputs, from the trained second learning device, an output image 34 showing the grinding surface 211 corresponding to the input image 31, which is also an output image 34 showing the grinding surface 211 of the grinding wheel 21 at the third period.

[0088] According to the above configuration, based on the state of the grinding surface 211 of the grinding wheel 21 at the first time point, it is possible to estimate and grasp the state of the grinding surface 211 of the grinding wheel 21 at the second time point, and also to estimate and grasp the state of the grinding surface 211 of the grinding wheel 21 at the third time point. Therefore, it is possible to manage the state of the grinding surface 211 of the grinding wheel 21 more meticulously.

[0089] (4) The grinding wheel 21 has an outer peripheral surface that forms a grinding surface 211 . 9, in this embodiment, the grinding surface 211 of the grinding wheel 21 is curved in one direction, so if the imaging range P1 in one imaging is made too wide, both end portions P2 of the imaging range P1 in the curvature direction will fall outside the range AR in focus. For this reason, it can be said that the imaging range P1 in one imaging of the grinding wheel 21 is likely to be narrow, and it can also be said that it is likely to take a long time to image the grinding surface 211.

[0090] According to this embodiment, even if the input image 31 captured by the camera 41 is an out-of-focus image, the learned learner 32 can be used to acquire an output image 34 showing the grinding surface 211 in focus based on the input image 31. Therefore, when capturing an image of the grinding surface 211 of the grinding wheel 21 to acquire the input image 31, it is acceptable for both end portions P2 in the curvature direction of the imaging range P1 to be out of focus. This allows the imaging range P1 to be widened in one imaging session, thereby shortening the time required to acquire the input image 31. Therefore, with the above configuration, the state of the grinding surface 211 of the grinding wheel 21 at the second time point can be estimated and understood in a short time based on the state of the grinding surface 211 of the grinding wheel 21 at the first time point.

[0091] <Example of change> The above embodiment can be modified as follows: The above embodiment and the following modifications can be combined with each other within the scope of technical compatibility.

[0092] In order to obtain the captured image that serves as the basis for the input image 31 and the example data 331, the grinding surface 211 of the grinding wheel 21 may be captured while the grinding wheel 21 is stopped from rotating or while the grinding wheel 21 is rotating at a relatively slow speed (a speed slower than the predetermined speed VW).

[0093] The grinding wheel 21 used to form the correct answer data 332 may be a grinding wheel 21 with debris such as chips and grinding fluid still attached to the grinding surface 211, or a grinding wheel 21 with the debris removed from the grinding surface 211. When a grinding wheel 21 with debris still attached is used, an output image 34 showing the grinding surface 211 with the debris attached can be obtained as an output image 34 from the trained learning device 32. The state of the grinding surface 211 can be confirmed based on this output image 34. On the other hand, when a grinding wheel 21 with debris removed from the grinding surface 211 is used, an output image 34 showing the grinding surface 211 with no debris attached can be obtained as an output image 34 from the trained learning device 32. The state of the grinding surface 211 can be confirmed based on this output image 34.

[0094] The camera unit 40 may be configured integrally with the automatic grinding device 20. In this configuration, output images 34 showing the grinding surface 211 at the second and third periods can be acquired while the grinding wheel 21 is rotating between grinding processes of the workpiece by the automatic grinding device 20. Then, the state of the grinding surface 211 of the grinding wheel 21 can be confirmed based on these output images 34. With the above configuration, the automatic grinding device 20 can be operated efficiently while the state of the grinding wheel 21 is confirmed at appropriate times.

[0095] The camera unit 40 may be disposed to the side of the grindstone 21 or above the grindstone 21. The structure of the illumination unit 43 can be changed as desired. For example, one of the first illumination unit 45 and the second illumination unit 46 may be omitted. The first emission color of the first illumination light 45L from the first illumination unit 45 and the second emission color of the second illumination light 46L from the second illumination unit 46 can be changed as desired.

[0096] A separate device other than the automatic grinding device 20 and a camera included in the device may be used to capture an image of the grinding surface 211 of the grinding wheel 21 to be estimated, and the input image 31 may be formed based on the captured image. In this case, in step S11 of the image output process (FIG. 8), the input image 31 formed as described above may be stored in the memory unit 56 of the electronic control device 50. When this configuration is adopted, the camera unit 40 may be omitted.

[0097] The work of storing the learned learning device 32 in the electronic control device 50 can be performed in any manner, such as the following (Work A) to (Work C). (Task A) Using the automatic grinding device 20 and grinding surface condition estimation device 30 used by the user, teacher data 33 is acquired, and the learner 32 is trained by machine learning using the teacher data 33.

[0098] (Work B) The manufacturer of the automatic grinding device 20 or the manufacturer of the grinding wheel 21 provides the training data 33 corresponding to the grinding wheel 21 that is actually used. The training data 33 is then stored in the electronic control device 50 of the grinding surface condition estimation device 30 used by the user, thereby allowing the learning device 32 to perform machine learning.

[0099] (Operation C) When manufacturing the grinding surface condition estimating device 30, the learned learning unit 32 is stored in the electronic control device 50. The grindstone 21 may be a general grindstone having general abrasive grains or a superabrasive grindstone having superabrasive grains.

[0100] The input image 31 is not limited to a single image showing the entire grinding surface 211, but may also be a single image showing a portion (e.g., half) of the grinding surface 211, or multiple images obtained by dividing an image showing the entire grinding surface 211 into multiple parts. When multiple images are used as the input image 31, the image output process (see FIG. 8) may be executed for each of the multiple images.

[0101] The input image 31 captured by the camera unit 40 can be used as input data either at the size at which it was captured or after enlarging or reducing it. This configuration allows for a high degree of freedom in the imaging process for acquiring the input image 31. For example, the following configurations (Configuration A) and (Configuration B) are conceivable.

[0102] (Configuration A) The example data 331 and the correct answer data 332 are constructed from captured images of the grinding surface 211 taken at a high magnification (M1). The input image 31 is constructed from captured images of the grinding surface 211 taken at a relatively low magnification (M2). The input image 31 is enlarged by a predetermined ratio R1 (R1=M1 / M2) and used as input data.

[0103] Here, when the abrasive grains of the grinding wheel 21 are small, it is preferable to image the grinding surface 211 at a high magnification in order to capture the characteristics of the grinding surface 211. However, when the grinding surface 211 is imaged at a high magnification, the imaging range in one image capture becomes narrower, which creates a trade-off in that it takes a long time to image the entire grinding surface 211.

[0104] According to the above configuration A, an input image 31, which is an image of the grinding surface 211 captured at a relatively low magnification M2, is used as input data, and an output image 34 equivalent to an image of the grinding surface 211 captured at a high magnification M1 is generated and output from the trained learning device 32. Therefore, when the abrasive grains of the grinding wheel 21 are small, an image equivalent to an image of the grinding surface 211 captured at a high magnification can be obtained as the output image 34, even if the grinding surface 211 is not captured at a high magnification to obtain the input image 31.

[0105] (Configuration B) The example data 331 and the correct answer data 332 are constructed from images of the grinding surface 211 captured at a low magnification (M3). The input image 31 is constructed from images of the grinding surface 211 captured at a relatively high magnification (M4). The input image 31 is reduced by a predetermined ratio R2 (R2 = M4 / M3) and used as input data.

[0106] According to the above configuration B, an input image 31, which is an image of the grinding surface 211 captured at a relatively high magnification M4, is used as input data, and an output image 34 equivalent to an image of the grinding surface 211 captured at a low magnification M3 is generated and output from the trained learning device 32. Therefore, when the abrasive grains of the grinding wheel 21 are large, an image equivalent to an image of the grinding surface 211 captured at a low magnification can be obtained as the output image 34, even if the grinding surface 211 is not captured at a low magnification to obtain the input image 31.

[0107] The execution procedure of the image output process (FIG. 8) can be changed as desired. For example, the process of selecting the learning device 32 (step S12) in the image output process may be omitted. In this case, the image output process may be executed so that, when an input image 31 at a first time point is stored in the memory unit 56, an output image 34 showing the grinding surface 211 at a second time point is automatically output. For example, when an input image 31 at a first acquisition time point T1 is stored in the memory unit 56, the image output process may be executed so that an output image 34 showing the grinding surface 211 at each of the second to fourth acquisition times T2 to T4, which are acquisition times other than the first acquisition time point T1, is automatically output.

[0108] The "amount of use of the grinding wheel 21 in the grinding process" can be set to the grinding amount described above, or can be changed as desired. For example, the time during which the grinding wheel 21 is used in the grinding process after dressing the grinding wheel 21 (hereinafter referred to as the grinding time) can be set to the "amount of use of the grinding wheel 21 in the grinding process."

[0109] The first acquisition time T1, the second acquisition time T2, the third acquisition time T3, and the fourth acquisition time T4 can be changed to any time. As the times for acquiring the captured images that make up the example data 331 and the captured images that make up the answer data 332, four acquisition times T1 to T4 can be set, or two acquisition times, three acquisition times, or five or more acquisition times can be set. In any configuration, based on the state of the grinding surface 211 at one of the multiple acquisition times (first time), it is possible to estimate and understand the state of the grinding surface 211 at the other one (second time), that is, the state of the grinding surface 211 in the past or the state of the grinding surface 211 in the future.

[0110] As the learning device 32, any image generation model other than "pix2pix" can be used as long as it can generate an output image 34 showing the grinding surface 211 at the second time period from an input image 31 obtained by capturing the grinding surface 211 at the first time period. It is also possible to use an image generation model other than a generative adversarial network as the learning device 32.

[0111] The grinding surface condition estimation device, estimation program, and grinding surface condition estimation method according to the above-described embodiments can be applied to a rotating grinding wheel for circumferential grinding, in which the outer peripheral surface of a cylindrical substrate serves as the grinding surface, and can also be applied to a rotating grinding wheel for face grinding, in which the bottom surface of a cylindrical substrate serves as the grinding surface.

[0112] <Additional Notes> The above embodiment includes the configurations described in the following supplementary notes. [Appendix 1] A grinding surface condition estimation device for estimating the condition of the grinding surface of a rotary grinding wheel, wherein a first period is defined as a period when a first predetermined amount of the grinding wheel has been used in a grinding process, and a second period is defined as a period when the amount of use is a second predetermined amount different from the first predetermined amount. The grinding surface condition estimation device comprises: an input unit that stores input images composed of captured images of the grinding surface of the rotary grinding wheel to be estimated at the first period; a memory unit that stores trained learning devices that have been machine-learned using teacher data including example data composed of captured images of the grinding surface of the rotary grinding wheel for obtaining teacher data at the first period, and ground truth data composed of captured images of the grinding surface of the rotary grinding wheel for obtaining teacher data at the second period; and an output unit that uses the input images as input data and outputs from the trained learning device an image of the grinding surface that corresponds to the input image and that shows the grinding surface of the rotary grinding wheel at the second period.

[0113] [Appendix 2] The grinding surface condition estimation device according to [Appendix 1], wherein each of the input image, the example data, and the correct answer data is a single image showing the entire grinding surface. [Appendix 3] The grinding surface condition estimation device described in [Appendix 1] or [Appendix 2], wherein the correct data is first correct data, the teacher data is first teacher data, the learning device is a first learning device, and a third period is a period when the usage amount is a third predetermined amount different from the first predetermined amount and the second predetermined amount, the memory unit stores a trained second learning device that has been machine-learned using second teacher data that includes the example data and second correct data composed of captured images of the grinding surface of the grinding wheel used to obtain the teacher data at the third period, and the output unit uses the input image as input data and outputs from the trained second learning device an image showing the grinding surface corresponding to the input image and an image showing the grinding surface of the grinding wheel at the third period.

[0114] [Appendix 4] The grinding surface condition estimating device according to any one of [Appendix 1] to [Appendix 3], wherein the rotating grindstone has an outer peripheral surface that serves as a grinding surface. [Appendix 5] [Appendix 1] to [Appendix 4]. An estimation program that causes an electronic control device provided in the grinding surface condition estimation device to execute the processing of each of the parts provided in the grinding surface condition estimation device.

[0115] [Appendix 6] A grinding surface condition estimation method for estimating the condition of the grinding surface of a rotary grinding wheel, the method comprising: an input step of storing, in an input unit of a grinding surface condition estimation device, an input image composed of an image of the grinding surface of the rotary grinding wheel to be estimated at the first time period, where the first time period is a time period when the amount of use of the rotary grinding wheel in a grinding process is a first predetermined amount, and the second time period is a time period when the amount of use is a second predetermined amount different from the first predetermined amount; a storage step of training a learning machine using teacher data including example data composed of an image of the grinding surface of the rotary grinding wheel for obtaining teacher data at the first time period and ground truth data composed of an image of the grinding surface of the rotary grinding wheel for obtaining teacher data at the second time period, and then storing the trained learning machine in a storage unit of the grinding surface condition estimation device; and an output step of using the input image as input data and outputting, from the trained learning machine, an image of the grinding surface corresponding to the input image and showing the grinding surface of the rotary grinding wheel at the second time period. [Explanation of symbols]

[0116] T1…first acquisition period T2…Second acquisition period T3…Third acquisition period T4…4th acquisition period EX1...First example image EX2...Second example image EX3...Third example image EX4...4th example image CA1: First correct image CA2: Second correct image CA3...Third correct image CA4...4th correct image 20...Automatic grinding device 21...Grinding stone 211...Abrasive surface 22...Grinding stone support part 23...Foundation 24...Table 25...Operation control unit 29...Grinding wheel drive unit 30... Grinding surface condition estimation device 31...Input image 32...Learning unit 33…Teacher data 331...Example data 332...Correct data 34...Output image 35...Execution program 40...Camera unit 41...Camera 50...Electronic control device 51...PU 52...ROM 53...RAM 54...Operation input section 55...Display section 56...Storage section 57...Output section

Claims

1. A grinding surface condition estimation device for estimating the condition of a grinding surface of a rotating grinding wheel, comprising: When a time when the amount of the grinding wheel used in a grinding process is a first predetermined amount is defined as a first period, and a time when the amount of the grinding wheel used in a grinding process is a second predetermined amount different from the first predetermined amount is defined as a second period, The grinding surface condition estimating device includes: an input unit that stores an input image formed by capturing an image of the grinding surface of the grinding wheel that is the estimation target at the first time period; a storage unit that stores a trained learning machine that has been machine-learned using teacher data including example data constituted by captured images of the grinding surface of the rotary grindstone for acquiring teacher data at the first time period, and correct answer data constituted by captured images of the grinding surface of the rotary grindstone for acquiring teacher data at the second time period; an output unit that uses the input image as input data and outputs, from the trained learning device, an image showing the grinding surface corresponding to the input image and showing the grinding surface of the grinding wheel at the second time point. Grinding surface condition estimation device.

2. Each of the input image, the example data, and the correct answer data is a single image showing the entire grinding surface. The grinding surface condition estimating device according to claim 1 .

3. When the correct data is defined as first correct data, the teacher data is defined as first teacher data, the learning device is defined as a first learning device, and a time when the usage amount is a third predetermined amount different from the first predetermined amount and the second predetermined amount is defined as a third time period, the storage unit stores a trained second learning machine that has been machine-learned using second teacher data including the example data and second supervised answer data configured from an image of the grinding surface of the grinding wheel used to acquire the teacher data at the third time period; the output unit uses the input image as input data and outputs, from the trained second learning device, an image showing the grinding surface corresponding to the input image and showing the grinding surface of the grinding wheel at the third time point. The grinding surface condition estimating device according to claim 1 or 2.

4. The rotating grindstone has an outer circumferential surface that serves as a grinding surface. The grinding surface condition estimating device according to claim 1 or 2.

5. 3. An estimation program that causes an electronic control device provided in the abrasive surface condition estimation device to execute the processing of each of the units provided in the abrasive surface condition estimation device according to claim 1 or 2.

6. A grinding surface condition estimation method for estimating the condition of a grinding surface of a rotating grinding wheel, comprising: When a time when the amount of the grinding wheel used in a grinding process is a first predetermined amount is defined as a first period, and a time when the amount of the grinding wheel used in a grinding process is a second predetermined amount different from the first predetermined amount is defined as a second period, an input step of storing an input image, which is an image of the grinding surface of the grinding wheel to be estimated at the first time point, in an input unit of the grinding surface condition estimation device; a storage step of causing a learning machine to perform machine learning using teacher data including example data constituted by captured images of the grinding surface of the rotary grinding wheel for acquiring teacher data at the first time period and ground truth data constituted by captured images of the grinding surface of the rotary grinding wheel for acquiring teacher data at the second time period, and then storing the trained learning machine in a storage unit of the grinding surface condition estimation device; an output step of outputting, from the learned learning device, an image showing the grinding surface corresponding to the input image and showing the grinding surface of the grinding wheel at the second time period, using the input image as input data; A grinding surface condition estimation method comprising:

Images