Training data acquisition apparatus and training data acquisition method

Abstract

An AI model generation apparatus 2 is provided with: a photographed image data storage unit 211 that stores image data of a 3D image acquired by photographing a weld part by a 3D sensor 36; and a processor 22 that identifies, on the basis of an input to a first input device 24 by a user, a correction area on the XY coordinate plane of the image data stored in the photographed image data storage unit 211, amends the Z coordinate of the correction area to a value in a state of no weld defect in the image data stored in the photographed image data storage unit 211, and acquires the amended image data as image data to be used for machine learning.

Description

Learning data acquisition device and learning data acquisition method

[0001] The present disclosure relates to a learning data acquisition device and a learning data acquisition method for acquiring image data used in machine learning.

[0002] Patent Document 1 discloses that when creating a model by machine learning using training data, training data determined to be inappropriate is excluded from the training data stored in a training data storage means.

[0003] International Publication No. 2019 / 187594

[0004] It is conceivable to create a model using machine learning to predict welding defect information, including the location of welding defects, based on image data of a 3D image. Creating such a model requires training data consisting of multiple data sets: image data of 3D images of the area around a weld bead captured by a 3D sensor, and annotation data of welding defect information for the image data. One possible method for acquiring image data constituting the training data is to actually create a weld bead through welding work and then capture the image using a 3D sensor. However, it is difficult to perform actual welding work without any welding defects. Furthermore, if weld beads created without welding defects are stored for many years and repeatedly used, there is a problem that the area around the weld bead will deteriorate over time before being captured. Therefore, it is time-consuming to acquire image data of 3D images that do not contain welding defects as training data.

[0005] On the other hand, if image data of a 3D image that does not include welding defects is created by a computer without being based on image data obtained by photography, the created image data will be significantly different from the image data obtained by actual photography, and the predictive accuracy of the created model will be poor.

[0006] The present disclosure has been made in consideration of these points, and its purpose is to reduce the effort required to obtain training data while suppressing deterioration in the prediction accuracy of the model.

[0007] In order to achieve the above object, the present disclosure provides a learning data acquisition device that acquires image data to be used for machine learning to create a model, the model outputting welding defect information including the location of welding defects based on image data of a 3D image, and the learning data acquisition device is characterized by comprising: an image data storage unit that stores image data of a 3D image acquired by photographing a welding location with a 3D sensor; and a processing unit that identifies a correction area on an XY coordinate plane of the image data stored in the image data storage unit based on input by a user to an input device, corrects the Z coordinate of the correction area in the image data stored in the image data storage unit to a value in a state where the welding defect does not exist, and acquires the corrected image data as image data to be used for the machine learning.

[0008] As a result, even if the 3D image of the image data stored in the image data storage unit contains a welding defect, the image data stored in the image data storage unit can be corrected to image data of a 3D image that does not contain a welding defect by specifying the area containing the welding defect through input to the input device. Therefore, even if welding work that does not cause any welding defect is not performed, image data of a 3D image that does not contain a welding defect can be obtained, thereby reducing the effort required to obtain learning data.

[0009] In addition, since image data for machine learning is generated by correcting image data obtained by actual photography using a 3D sensor, the predictive accuracy of the created model can be improved compared to when it is created without being based on image data obtained by actual photography.

[0010] According to the present disclosure, it is possible to reduce the effort required to obtain training data while suppressing deterioration in the prediction accuracy of the model.

[0011] FIG. 1 is a block diagram illustrating a configuration of a welding system including an AI model generation device as a learning data acquisition device according to a first embodiment of the present disclosure. FIG. 2 is an explanatory diagram illustrating a 3D image output to a first output device based on captured image data. FIG. 3 is an explanatory diagram illustrating an image on an XY coordinate plane output to a first output device based on captured image data. FIG. 4 is an explanatory diagram illustrating a method for identifying a correction area. FIG. 5 is a flowchart illustrating an operation of an AI model generation device for acquiring defect-free data. FIG. 6 is a diagram illustrating a 3D image showing a welding defect. FIG. 7 is a plan view illustrating, on an XY coordinate plane, a defect area in a 3D image represented by captured image data. FIG. 8 is a diagram equivalent to FIG. 7 in a first example in which a point at the Y-axis end of the correction area is located at the edge of the XY coordinate plane represented by the captured image data. FIG. 9 is a diagram equivalent to FIG. 7 in a second example in which a point at the Y-axis end of the correction area is located at the edge of the XY coordinate plane represented by the captured image data. FIG. 10 is a side view illustrating, on a YZ coordinate plane, a portion of the correction area after linear interpolation. Fig. 11 is a diagram equivalent to Fig. 10 in which there are two sets of two points that sandwich the correction area in the Y-axis direction and have a common X-coordinate. Fig. 12 is a schematic diagram of the area around the defect area in a 3D image represented by captured image data. Fig. 13 is a diagram equivalent to Fig. 12 in which no defect data is present. Fig. 14 is a diagram illustrating an example of an image output by the first output device in a state in which the type and position of a welding defect have been specified by a user's input operation.

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. The following description of the preferred embodiments is merely exemplary in nature and is not intended to limit the present invention, its applications, or uses in any way.

[0013] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0014] 1 shows a welding system 1. The welding system 1 includes an AI model generation device 2 as a learning data acquisition device according to an embodiment of the present disclosure, and a welding device 3 that performs welding.

[0015] The AI ​​model generation device 2 generates a defect detection model through machine learning. The AI ​​model generation device 2 also acquires image data that constitutes learning data used in the machine learning for generating the defect detection model.

[0016] The defect detection model is an object detection model that predicts and outputs welding defect information based on image data of an input 3D image. The defect detection model is represented by a CNN (Convolutional Neural Network). The welding defect information includes category information indicating the type of welding defect in the 3D image represented by the image data, and position information indicating the position of the welding defect. In more detail, the category information indicates whether the welding defect is a hole, pit, spatter, undercut, protrusion, etc. A pit is an opening on the surface of a weld bead. The position information indicates the position of a bounding box containing the welding defect. The bounding box is a rectangular boundary area that surrounds an object such as a welding defect.

[0017] In addition, the AI ​​model generation device 2 acquires annotation data of welding defect information, and generates a defect detection model through machine learning using multiple data sets of image data and the annotation data as learning data.

[0018] Specifically, the AI ​​model generation device 2 includes a storage device 21, a processor 22 as a processing unit, a first output device 23, and a first input device 24.

[0019] The storage device 21 includes a captured image data storage unit 211 , a defect-free data storage unit 212 , an annotation data storage unit 213 , and a parameter storage unit 214 .

[0020] The photographed image data storage unit 211 stores, as photographed image data, image data of a 3D image acquired by photographing a welding point using a 3D sensor 36 (described later).

[0021] The defect-free data storage unit 212 stores defect-free data acquired by the processor 22 correcting the captured image data.

[0022] The annotation data storage unit 213 stores defective welding information specified by a user's input to the first input device 24 while an image based on the image data stored in the photographed image data storage unit 211 is being output to the first output device 23. The annotation data storage unit 213 also stores defective welding information indicating that there is no defective welding as annotation data in the annotation data storage unit 213, in association with the no-defect data stored in the no-defect data storage unit 212.

[0023] The parameter storage unit 214 stores parameters that specify a fault detection model generated by the processor 22, which will be described later.

[0024] The processor 22 has an image data receiving unit 221 , a correction area identifying unit 222 , a correction unit 223 , an annotation data receiving unit 224 , and an AI model generating unit 225 .

[0025] The image data receiving unit 221 receives image data of a 3D image obtained by photographing the welding point using the 3D sensor 36 described later from the welding device 3, and stores it as photographed image data in the photographed image data storage unit 211.

[0026] The correction area identification unit 222 outputs an image based on the captured image data stored in the captured image data storage unit 211 to the first output device 23, and in this state, based on input by the user to the first input device 24, identifies a correction area CA (see Figures 2 and 3) on the XY coordinate plane of the captured image data.

[0027] As a method of identifying the correction area CA, for example, the correction area identification unit 222 outputs a 3D image such as that shown in Figure 2 based on the captured image data to the first output device 23, and identifies as the correction area CA an area on the XY coordinate plane corresponding to the area where the label LA is superimposed by a predetermined input operation by the user to the first input device 24.

[0028] 3 based on the captured image data to the first output device 23, and identifies a circle CI of a predetermined size centered on a portion designated by a predetermined input operation by the user to the first input device 24. The correction area identification unit 222 then identifies, as the correction area CA, an area surrounded by a square SQ that surrounds the circle CI from both ends of the circle CI in the X-axis direction by a predetermined number of columns (e.g., five columns) and from both ends of the circle CI in the Y-axis direction by a predetermined number of rows (e.g., five rows). The correction area identification unit 222 can identify multiple areas as correction areas CA in a single image.

[0029] The method of specifying the correction area CA is not limited to the above. As shown in FIG. 4, the correction area CA may be an area surrounding the area DA designated by the user from one point outside.

[0030] The correction unit 223 corrects the Z coordinate of the correction area CA identified by the correction area identification unit 222 in the captured image data stored in the captured image data storage unit 211 to a value corresponding to a state in which there is no welding defect.The correction unit 223 then stores the corrected image data as defect-free data in the defect-free data storage unit 212.If there are multiple correction areas CA, correction is performed on all of the correction areas CA.The correction method used by the correction unit 223 will be described in detail below.

[0031] The annotation data receiving unit 224 outputs an image based on the photographed image data stored in the photographed image data storage unit 211 to the first output device 23. In this state, the first input device 24 accepts an input operation by the user specifying the type and position of a welding defect in the image. When the user's input operation is completed, the annotation data receiving unit 224 associates the welding defect information identified based on the user's input operation to the first input device 24 with the photographed image data of each image and stores the information as annotation data in the annotation data storage unit 213.

[0032] The annotation data receiving unit 224 also stores defective welding information indicating that there is no welding defect as annotation data in the annotation data storage unit 213 in association with the no-defect data stored in the no-defect data storage unit 212 .

[0033] In addition, the annotation data receiving unit 224 may also create annotation data for the defect-free data stored in the defect-free data storage unit 212 based on the user's input to the first input device 24, in the same way as for the captured image data stored in the captured image data storage unit 211.

[0034] The AI ​​model generation unit 225 creates a defect detection model by performing machine learning using learning data consisting of multiple sets of data pairs: image data stored in the captured image data storage unit 211 and the defect-free data storage unit 212; and annotation data stored in the annotation data storage unit 213. Specifically, the AI ​​model generation unit 225 specifies the weights and biases of each node constituting the CNN that expresses the defect detection model as parameters that specify the defect detection model. Then, the AI ​​model generation unit 225 stores the specified parameters in the parameter storage unit 214.

[0035] In this way, the processor 22 acquires the image data corrected by the correction unit 223 as image data to be used for machine learning by the AI ​​model generation unit 225.

[0036] The first output device 23 outputs an image based on the captured image data stored in the captured image data storage unit 211. The first output device 23 is configured by, for example, a liquid crystal monitor.

[0037] The first input device 24 accepts a user's input operation for identifying the correction area CA, and transmits a signal corresponding to the accepted input operation to the correction area identifying unit 222. The first input device 24 also accepts a user's input operation for specifying the type and position of a welding defect in the image output to the first output device 23, and transmits a signal corresponding to the accepted input operation to the annotation data receiving unit 224.

[0038] The welding device 3 has a welding torch 31, a wire feeder (not shown), a welding power source 32, an output control unit 33, a robot arm 34, a robot control unit 35, a 3D sensor 36, a computer 37, a second output device 38, and a second input device 39. When power is supplied from the welding power source 32 to the welding wire WI held by the welding torch 31, an arc is generated between the tip of the welding wire WI and the workpiece W, and the workpiece W is heated to perform arc welding. Note that the welding device 3 has other components and equipment such as piping and gas cylinders for supplying shielding gas to the welding torch 31, but for convenience of explanation, these are not shown or described.

[0039] Output control unit 33 is connected to welding power source 32 and a wire feeder (not shown) and controls the welding output of welding torch 31, in other words, the power supplied to welding wire WI and the power supply time, in accordance with predetermined welding conditions. Output control unit 33 also controls the feed speed and feed amount of welding wire WI fed from the wire feeder (not shown) to welding torch 31. Note that the welding conditions may be input directly to output control unit 33 via an input unit (not shown), or may be selected from a welding program separately read from a recording medium or the like.

[0040] Robot arm 34 is a known articulated robot that holds welding torch 31 at its tip and is connected to robot control unit 35. Robot control unit 35 controls the operation of robot arm 34 so that the tip of welding torch 31, in other words, the tip of welding wire WI held by welding torch 31, moves to a desired position while tracing a predetermined welding trajectory.

[0041] The 3D sensor 36 is attached to the welding torch 31 and measures the welded portion PW of the workpiece W, specifically, the shape around the weld bead. The 3D sensor 36 is a three-dimensional shape measurement sensor composed of, for example, a laser light source (not shown) configured to scan the surface of the workpiece W and a camera (not shown) that captures the reflection trajectory of the laser light projected onto the surface of the workpiece W (hereinafter, sometimes referred to as a shape line). The 3D sensor 36 scans the entire welded portion PW of the workpiece W with a laser beam, and the camera captures the laser beam reflected by the welded portion PW, thereby measuring the shape of the welded portion PW. The 3D sensor 36 is configured to measure the shape of not only the welded portion PW but also a predetermined range around it. This is to evaluate the presence or absence of spatter, etc. The camera has a CCD or CMOS image sensor as an imaging element. The configuration of the 3D sensor 36 is not limited to the above, and other configurations are possible. For example, an optical interferometer may be used instead of the camera.

[0042] The computer 37 executes software implemented on a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) to realize the functions of a plurality of functional blocks within the computer 37. The computer 37 has an image processing unit 371, a welding defect information calculation unit 372, and a welding defect display image generation unit 373.

[0043] The image processing unit 371 receives the shape data acquired by the 3D sensor 36 and converts it into image data of a 3D image including the welding point PW. For example, the image processing unit 371 acquires point cloud data of the shape lines captured by the 3D sensor 36. The image processing unit 371 also corrects the inclination and distortion of the base portion of the welding point PW relative to a predetermined reference plane, for example, the installation surface of the workpiece W, by statistically processing the point cloud data, and acquires image data including the welding point PW.

[0044] The defective welding information calculation unit 372 calculates (predicts) defective welding information based on the image data generated by the image processing unit 371, using the defect detection model generated by the AI ​​model generation device 2. The calculation of the defective welding information may be performed when a predetermined input operation is performed on the second input device 39.

[0045] The defective welding display image generating unit 373 calculates a pixel value of each pixel of the defective welding display image displaying the defective welding information, based on the defective welding information calculated by the defective welding information calculating unit 372. The information specifying the defective welding display image may be compressed using a JPEG format or the like.

[0046] The second output device 38 displays the poor welding display image based on the pixel value of each pixel calculated by the poor welding display image generating unit 373. The second output device 38 is configured by a liquid crystal monitor or the like.

[0047] The second input device 39 accepts a predetermined input operation by the user to cause the computer 37 to start calculating the welding defect information.

[0048] Next, the operation of the AI ​​model generation device 2 to acquire defect-free data will be described with reference to the flowchart of FIG.

[0049] First, in (S101), the 3D sensor 36 captures an image of the welding spot PW, and the image processing unit 371 receives the shape data acquired by the 3D sensor 36 and acquires image data of the 3D image. The image data acquired here is 3D point cloud data, but may be data in another format.

[0050] Next, in (S102), the image data receiving unit 221 of the AI ​​model generating device 2 receives the image data acquired in (S101) from the welding device 3 and stores it in the captured image data storage unit 211 as captured image data.

[0051] Next, in (S103), the correction area identifying unit 222 causes the first output device 23 to output an image based on the photographed image data stored in the photographed image data storage unit 211 in (S102), and in this state identifies a correction area CA on the XY coordinate plane of the photographed image data based on a user input to the first input device 24. The 3D image displayed by the photographed image data acquired in (S101) may contain a welding defect, as shown by the curve SL in Fig. 6, for example. In (S103), the user identifies the area containing the welding defect by input to the first input device 24.

[0052] Next, the correction unit 223 corrects the Z coordinate of the correction area CA to a value corresponding to a state in which there is no welding defect, by the following processes (S104) to (S106).

[0053] First, in (S104), the correction unit 223 identifies adjacent points in the Y-axis direction to the correction area CA identified in (S103). Figure 7 shows the area around the correction area CA in the 3D image represented by the captured image data stored in the captured image data storage unit 211 on an XY coordinate plane. In Figure 7, points included in the correction area CA are indicated by diagonal hatching. In the example of Figure 7, the correction unit 223 identifies points P11 to P15 and P21 to P25 as adjacent points.

[0054] As shown in Figures 8 and 9, if the point at the Y-axis end of the correction area CA is located at the edge of the XY coordinate plane represented by the image data, the correction unit 223 identifies the point at the Y-axis end of this correction area as an end point. In the example of Figure 8, the correction unit 223 identifies points P31 to P34 as adjacent points of the correction area and points P41 to P44 as end points of the correction area CA. In the example of Figure 9, the correction unit 223 identifies points P51 to P56 and P61 to P63 as adjacent points of the correction area CA and points P64 to P66 as end points of the correction area CA. Furthermore, for each end point, the correction unit 223 identifies a reference point that is closest in the X-axis direction and shares a common position in the Y-axis direction. If there are two points that are closest in the X-axis direction and share a common position in the Y-axis direction, the point with the smaller X coordinate value is used as the reference point. In the example of Figure 8, point PR is identified as the reference point for end points P41 to P44. In the example of FIG. 9, point P61 is specified as the reference point for end points P64 and P65, and point P62 is specified as the reference point for end point P66.

[0055] Next, in (S105), the correction unit 223 determines the equation of a linear function for calculating the Z coordinate based on the Y coordinate for each two points that share the same X coordinate and sandwich the correction area CA in the Y-axis direction among the adjacent points identified in (S104). That is, it determines the coefficients a and b of the equation of the linear function expressed as Z = aY + b. The equation of the linear function can be calculated based on the Y coordinates and Z coordinates of the two points that share the same X coordinate among the adjacent points identified in (S104).

[0056] Furthermore, for adjacent points that do not have an adjacent point with a common X coordinate, such as points P31 to P34 in Figure 8 and points P52 to P54 in Figure 9, the equation of a linear function for calculating the Z coordinate based on the Y coordinate is determined using the Y and Z coordinates of the adjacent point, the Y coordinate of an end point that shares an X coordinate with the adjacent point, and the Z coordinate of a reference point for the end point. For example, for adjacent point P31 in Figure 8, the equation of the linear function is determined based on the Y and Z coordinates of adjacent point P31, the Y coordinate of end point P41, and the Z coordinate of reference point PR. For adjacent point P53 in Figure 9, the equation of the linear function is determined based on the Y and Z coordinates of adjacent point P53, the Y coordinate of end point P65, and the Z coordinate of reference point P61.

[0057] Next, in (S106), the correction unit 223 calculates the Z coordinate value of each point included in the correction area CA by applying the Y coordinate value of each point included in the correction area CA to the linear function equation identified in (S105). The Z coordinate value of each point included in the correction area CA is calculated using the linear function equation determined for adjacent points whose X coordinates are equal to those of each point. For example, in FIG. 9, the Z coordinate value of a point whose X coordinate is equal to that of adjacent point P53 is calculated using the linear function equation determined based on the Y and Z coordinates of adjacent point P53, the Y coordinate of endpoint P65, and the Z coordinate of reference point P61. Then, the correction unit 223 corrects the Z coordinate of each point (point whose XY coordinates are located in the correction area CA) in the captured image data stored in the captured image data storage unit 211 to the calculated Z coordinate value. The correction unit 223 then proceeds to the process of (S107).

[0058] By the processes of (S104) to (S106), for example, as shown in FIG. 10, the Z coordinate of a point between points P15 and P25 in FIG. 7 is linearly interpolated so that the point is located on the straight line connecting points P15 and P25 on the YZ coordinate plane.

[0059] 11, there are two pairs of points that sandwich the correction area CA in the Y-axis direction and have a common X-coordinate. In this example, the Z coordinate of a point between points PA and PB is linearly interpolated so that the point is located on the line connecting points PA and PB on the YZ coordinate plane, and the Z coordinate of a point between points PC and PD is linearly interpolated so that the point is located on the line connecting points PC and PD on the YZ coordinate plane.

[0060] In (S107), the correction unit 223 stores the corrected image data in the defect-free data storage unit 212 as defect-free data.

[0061] For example, if the 3D image based on the captured image data is an image as shown in Figure 12, and the area indicated by diagonal lines is the correction area CA, the Z coordinate of the black dot within the correction area CA is corrected to a value corresponding to a state in which there is no welding defect, as shown in Figure 13.

[0062] The AI ​​model generation device 2 repeatedly executes the operations (S101) to (S107) in the above flowchart to obtain multiple pieces of defect-free data.

[0063] Thereafter, the annotation data receiving unit 224 sequentially outputs images based on the captured image data stored in the captured image data storage unit 211 to the first output device 23. Then, the annotation data receiving unit 224 receives a signal corresponding to a user's input operation from the first input device 24. The user performs an input operation to specify the type and position of a welding defect for each image output by the first output device 23. In FIG. 14 , the type and position of a welding defect are specified by the user's input operation in the image output by the first output device 23. The annotation data receiving unit 224 stores the welding defect information identified based on the user's input operation as annotation data of the welding defect information in the annotation data storage unit 213 in association with each captured image data.

[0064] The annotation data receiving unit 224 also stores defective welding information indicating that there is no welding defect as annotation data in the annotation data storage unit 213 in association with the no-defect data stored in the no-defect data storage unit 212 .

[0065] Thereafter, the AI ​​model generation unit 225 reads out learning data consisting of multiple sets of data pairs of image data and annotation data from the annotation data storage unit 213, and performs machine learning using the read learning data to generate a defect detection model. Then, the AI ​​model generation unit 225 stores parameters that specify the generated defect detection model in the parameter storage unit 214. The parameters stored in the parameter storage unit 214 are sent to the defective welding information calculation unit 372 of the welding device 3. The defective welding information calculation unit 372 stores the parameters.

[0066] Thereafter, in the welding device 3, when the 3D sensor 36 captures an image of the welded portion PW, the image processing unit 371 receives the shape data acquired by the 3D sensor 36 and converts it into image data of a welding image including the welded portion PW. The defective welding information calculation unit 372 calculates (predicts) defective welding information based on the image data generated by the image processing unit 371, using a defect detection model identified by parameters sent from the AI ​​model generation device 2. The defective welding display image generation unit 373 calculates a pixel value of each pixel of the defective welding display image displaying the defective welding information, based on the defective welding information calculated by the defective welding information calculation unit 372. The second output device 38 displays the defective welding display image based on the pixel value of each pixel calculated by the defective welding display image generation unit 373.

[0067] Therefore, according to this embodiment, even if the 3D image of the photographed image data acquired in (S101) includes a welding defect, the image data stored in the photographed image data storage unit 211 can be corrected to defect-free data of a 3D image that does not include a welding defect by specifying the area including the welding defect through input to the first input device 24. Therefore, even if welding work that does not generate any welding defects is not performed, image data of a 3D image that does not include a welding defect can be acquired, thereby reducing the effort required to acquire learning data.

[0068] Furthermore, since defect-free data for machine learning is generated by correcting the captured image data obtained by photographing using the 3D sensor 36, the predictive accuracy of the defect detection model created can be improved compared to when it is created without being based on image data obtained by actual photographing.

[0069] In the above embodiment, the Z coordinates of the points included in the correction area CA are linearly interpolated based on the Z coordinates of the points located on both sides outside the correction area CA in the Y-axis direction. However, the Z coordinates of the points included in the correction area CA may also be linearly interpolated based on the Z coordinates of the points located on both sides outside the correction area CA in the X-axis direction.

[0070] In the above embodiment, the Z coordinate of the point whose XY coordinate is located in the correction area CA is corrected by linear interpolation. However, if the correction area CA after correction can be made free of welding defects, the Z coordinate of the point whose XY coordinate is located in the correction area CA may be corrected to a predetermined value.

[0071] In addition, in the above embodiment, the functions of the correction area specifying unit 222 and the correction unit 223 and the function of the AI ​​model generating unit 225 are realized by one processor 22, but the functions of the correction area specifying unit 222 and the correction unit 223 and the function of the AI ​​model generating unit 225 may be realized by different processors. In other words, the functions of the processor 22 in the above embodiment may be realized by multiple processors.

[0072] The training data acquisition device and training data acquisition method disclosed herein can reduce the effort required to acquire training data while suppressing deterioration in the predictive accuracy of the model, and are useful as a training data acquisition device and training data acquisition method for acquiring image data to be used in machine learning.

[0073] 2 AI model generation device (learning data acquisition device) 22 Processor (processing unit) 24 First input device 36 3D sensor 211 Captured image data storage unit CA Correction area

Claims

1. A learning data acquisition device that acquires image data to be used for machine learning to create a model, the model outputting welding defect information including the location of welding defects based on image data of a 3D image, the learning data acquisition device comprising: an image data storage unit that stores image data of a 3D image acquired by photographing a welding point with a 3D sensor; and a processing unit that identifies a correction area on an XY coordinate plane of the image data stored in the image data storage unit based on an input by a user to an input device, corrects the Z coordinate of the correction area in the image data stored in the image data storage unit to a value in a state where there is no welding defect, and acquires the corrected image data as image data to be used for the machine learning.

2. A learning data acquisition device as described in claim 1, characterized in that the correction by the processing unit is performed by linearly interpolating the Z coordinate of a point whose XY coordinates are located within the correction area in the image data stored in the image data storage unit, based on the Z coordinate of a point whose XY coordinates are located outside the correction area in both the X and Y directions.

3. A learning data acquisition method for acquiring image data to be used in machine learning to create a model, wherein the model outputs welding defect information including the location of welding defects based on image data of a 3D image, and the learning data acquisition method is characterized by: storing image data of a 3D image acquired by photographing a welding point with a 3D sensor in an image data storage unit; identifying a correction area on an XY coordinate plane of the image data stored in the image data storage unit based on an input by a user to an input device; correcting the Z coordinate of the correction area in the image data stored in the image data storage unit to a value in a state where there is no welding defect; and acquiring the corrected image data as image data to be used for the machine learning.