WO2025142115A1 - 学習データ取得装置、及び学習データ取得方法 - Google Patents
学習データ取得装置、及び学習データ取得方法 Download PDFInfo
- Publication number
- WO2025142115A1 WO2025142115A1 PCT/JP2024/038846 JP2024038846W WO2025142115A1 WO 2025142115 A1 WO2025142115 A1 WO 2025142115A1 JP 2024038846 W JP2024038846 W JP 2024038846W WO 2025142115 A1 WO2025142115 A1 WO 2025142115A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- defect
- synthesis
- image
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
- B23K31/00—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by any single one of main groups B23K1/00 - B23K28/00
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
Definitions
- This disclosure relates to a learning data acquisition device and a learning data acquisition method for acquiring image data to be used in machine learning.
- the training data acquisition device disclosed in Patent Document 1 acquires training data to be used for machine learning in a machine learning system that automatically generates an output image from an input image.
- This training data acquisition device acquires, as training data, pseudo sample images used as input images for the machine learning system during machine learning, and pseudo label images used as output images for the machine learning system during machine learning.
- This training data acquisition device first generates a large number of pseudo label images by inputting random values to an image generation engine that has performed machine learning on multiple original label images. Then, based on these pseudo label images, the training data acquisition device generates the pseudo sample images in accordance with the conversion characteristics from the original label images to the original sample images. This makes it possible to acquire a large amount of training data in a short period of time.
- 3D data showing a 3D image It is possible to use machine learning to create a model that outputs defective welding information, including the location of defective welding, based on 3D data showing a 3D image.
- learning data consisting of multiple data sets is required, including 3D data showing an image of the area around the weld mark captured by a 3D sensor, and defective welding information to be output based on the 3D data.
- One possible method for creating the 3D data that constitutes the learning data is to obtain 3D images of weld marks that do not include defective welding and 3D images of defective welding by measuring with a 3D sensor, and then process and synthesize the images.
- this method requires the preparation of a large number of 3D images of defective welding, which is time-consuming and requires a long working time.
- the present disclosure provides a learning data acquisition device for acquiring learning data used in machine learning to create a model, the model outputting welding defect information including the position of a welding defect based on 3D data representing a 3D image, the learning data acquisition device including a first synthesis step of generating synthesized 3D data representing a 3D image in which the bottom of the welding defect is formed at a certain depth in an area on the XY coordinate plane where the bottom of the welding defect is located in the first synthesis 3D data and in an area on the XY coordinate plane where the bottom of the welding defect is located in the second synthesis 3D data, based on first and second synthesis 3D data representing a point cloud including a welding defect recessed in the Z-axis direction in a predetermined XYZ orthogonal coordinate system and having common image sizes in the X-axis direction and the Y-axis direction, and
- the present invention is characterized in that it has a learning data acquisition unit that executes at least one of the synthesis steps of
- a bottom of a certain depth is formed in the region on the XY coordinate plane where the bottom of the weld defect is located in the first synthesis 3D data, and in the region on the XY coordinate plane where the bottom of the weld defect is located in the second synthesis 3D data. That is, no step is formed on the bottom surface of the weld defect shown by the post-synthesis 3D data due to the difference in depth of the bottom of the weld defect shown by the first and second synthesis 3D data. Therefore, it is possible to suppress deterioration of the prediction accuracy of the model due to the formation of the step on the bottom surface of the weld defect shown by the post-synthesis 3D data.
- This disclosure makes it possible to reduce the effort and work time required to obtain learning data.
- FIG. 7 illustrates two 3D images shown by the final 3D data, where the 3D image on the left shows a 3D image of a welding defect in which an unintended step is formed, and the 3D image on the right shows a 3D image of a welding defect in which an unintended step is not formed.
- FIG. 1 shows a welding system 1.
- This welding system 1 has 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.
- the AI model generation device 2 creates a defect detection model through machine learning (deep learning). In addition, the AI model generation device 2 acquires learning data to be used for machine learning to create the defect detection model.
- the learning data generated by the AI model generation device 2 is a set of multiple data sets consisting of 3D data showing 3D images and welding defect information.
- the defective image data storage unit 211 stores image data of a 3D image obtained by capturing an image of the area around the welding point PW, including the welding defect, using the 3D sensor 36 described below as defective image data.
- the segmentation data storage unit 214 stores segmentation data of the image data with defects stored in the image data with defects storage unit 211. This segmentation data specifies the boundary between the areas where welding defects have been formed and areas where they have not been formed in the image with defects based on the image data with defects. This segmentation data is identified by a user's boundary identification input to the first input device 24 while the image with defects is being output to the first output device 23.
- the segmentation data storage unit 214 also stores segmentation data of the defect-free image data stored in the defect-free image data storage unit 212.
- This segmentation data is acquired, for example, by polygon segmentation (polygon annotation).
- This segmentation data specifies the boundary between areas where welding defects have been formed and areas where they have not been formed in the defect-free image based on the defect-free image data.
- This segmentation data is created by the defect-free data processing unit 222.
- the parameter storage unit 217 stores parameters that identify the defect detection model generated by the processor 22.
- the defect data processing unit 221 also outputs to the first output device 23 an image with defects based on the defect image data stored in the defect image data storage unit 211.
- the first input device 24 accepts a boundary identification input from the user that identifies the boundary between the formation area and non-formation area of the weld defect in the image with defects.
- the defect data processing unit 221 stores the segmentation data identified based on the user's boundary identification input to the first input device 24 in the segmentation data storage unit 214 in association with the defect image data of each image.
- the defect-free data processing unit 222 accepts a base material designation input to the first input device 24 from the user, which designates the size of the base material on which the weld bead is to be formed, and a weld mark identification input to the first input device 24 from the user, which designates the shape of the weld bead.
- the defect-free data processing unit 222 then creates defect-free image data representing a defect-free image, which is a 3D image including the weld bead and the base material W, in response to the base material designation input and the weld mark identification input, and stores the defect-free image data in the defect-free image data storage unit 212.
- FIG. 2 illustrates an example of a 3D image represented by the defect-free image data.
- the defect-free data processing unit 222 also automatically creates segmentation data identifying the boundary between the weld bead formation area and non-formation area, and stores the segmentation data storage unit 214.
- the data expansion unit 223 acquires multiple pieces of captured 3D data showing 3D images showing defective welds based on the defective image data stored in the defective image data storage unit 211 and the annotation data of the defective image data stored in the annotation data storage unit 213.
- the images shown by the captured 3D data are, for example, image D11 (corresponding to the first selected 3D data (before synthesis) described later) and image D21 (corresponding to the second selected 3D data (before synthesis) described later) in Figures 5 and 6 showing holes as defective welds.
- the data expansion unit 223 performs data expansion on the multiple captured 3D data thus acquired, thereby acquiring a larger number of defective 3D data than the captured 3D data, and stores the defective 3D data in the defective 3D data storage unit 215.
- the learning data creation unit 224 creates learning image data based on the defect 3D data stored in the defect 3D data storage unit 215 and the defect-free image data stored in the defect-free image data storage unit 212. Specifically, the learning data creation unit 224 creates multiple learning 3D data by combining multiple patterns of one or more of the multiple welding defects represented by the multiple defect 3D data, and synthesizing them with the 3D image represented by the defect-free image data.
- the image represented by each learning 3D data is an image in which one or more of the multiple welding defects represented by the multiple defect 3D data are pasted onto the 3D image represented by the defect-free image data.
- the learning data creation unit 224 then stores the learning data, which is made up of multiple data sets of the created learning 3D data and annotation data for that learning 3D data, in the learning data storage unit 216.
- the AI model generation unit 225 creates a fault detection model by performing machine learning using the learning data stored in the learning data storage unit 216. Specifically, the AI model generation unit 225 identifies the weights and biases of each node constituting a CNN (convolutional neural network) that represents the fault detection model as parameters that identify the fault detection model. Then, the AI model generation unit 225 stores the identified parameters in the parameter storage unit 217.
- a CNN convolutional neural network
- the first output device 23 outputs a defect-containing image based on the defect-containing image data stored in the defect-containing image data storage unit 211.
- the first output device 23 is, for example, configured with a liquid crystal monitor.
- the first output device 23 also outputs an image that prompts the user to perform various inputs.
- the first input device 24 accepts the defect information input, the boundary identification input, the base material designation input, and the weld mark identification input from the user, and transmits a signal corresponding to the accepted input to the processor 22.
- 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.
- 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.
- 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 ease of explanation, these are not shown or described.
- the output control unit 33 is connected to the welding power source 32 and the wire feeder (not shown) and controls the welding output of the welding torch 31, in other words, the power supplied to the welding wire WI and the power supply time, according to specified welding conditions.
- the output control unit 33 also controls the feed speed and feed amount of the welding wire WI fed from the wire feeder (not shown) to the welding torch 31.
- the welding conditions may be input directly to the output control unit 33 via an input unit (not shown), or may be selected from a welding program read out separately from a recording medium or the like.
- the robot arm 34 is a known multi-jointed robot that holds the welding torch 31 at its tip and is connected to the robot control unit 35.
- the robot control unit 35 controls the operation of the robot arm 34 so that the tip of the welding torch 31, in other words, the tip of the welding wire WI held by the welding torch 31, traces a predetermined welding trajectory and moves to the desired position.
- the 3D sensor 36 is attached to the welding torch 31 and measures the shape of the welded portion PW of the base material W, specifically, the area around the weld bead.
- the 3D sensor 36 is a three-dimensional shape measurement sensor that is composed of, for example, a laser light source (not shown) that is configured to be able to scan the surface of the base material W, and a camera (not shown) that captures the reflection trajectory of the laser light (hereinafter, sometimes referred to as a shape line) projected on the surface of the base material W.
- the 3D sensor 36 scans the entire welded portion PW of the base material W with a laser beam, and captures the laser beam reflected by the welded portion PW with the camera, thereby measuring the shape of the welded portion PW.
- the 3D sensor 36 is configured to perform shape measurement not only of the welded portion PW, but also of the surrounding area within a predetermined range. This is to evaluate the presence or absence of spatters, etc.
- the camera has a CCD or CMOS image sensor as an image sensor.
- the configuration of the 3D sensor 36 is not limited to the above, and other configurations may be adopted. For example, an optical interferometer may be used instead of the camera.
- Computer 37 realizes the functions of multiple functional blocks within computer 37 by executing software implemented on a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit).
- Computer 37 has an image processing unit 371, a welding defect information calculation unit 372, and a welding defect display image generation unit 373.
- the image processing unit 371 receives the shape data acquired by the 3D sensor 36, and acquires measurement 3D data of a 3D image including the welding point PW based on this shape data. In detail, 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 base material W, by statistically processing the point cloud data, and acquires measurement 3D data including the welding point PW.
- a predetermined reference plane for example, the installation surface of the base material W
- the welding defect information calculation unit 372 uses the defect detection model generated by the AI model generation device 2 to calculate (acquire) welding defect information based on the measurement 3D data generated by the image processing unit 371.
- the calculation of welding defect information is performed when a predetermined input operation is performed on the second input device 39.
- the poor welding display image generating unit 373 calculates the pixel value of each pixel of the poor welding display image that displays the poor welding information based on the poor welding information calculated by the poor welding information calculating unit 372.
- the information that identifies the poor welding display image may be compressed using a JPEG format or the like.
- 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 with a liquid crystal monitor or the like.
- the second input device 39 accepts a specific input operation by the user to cause the computer 37 to start calculating the welding defect information.
- the user starts the power supply to the welding power source 32 of the welding device 3 and performs arc welding to form a weld bead along with a weld defect on the base material W. Then, the user causes the 3D sensor 36 to capture an image of the welded portion PW of the base material W. After that, the image processing unit 371 acquires defect image data showing an image including a weld defect based on the shape data acquired by the 3D sensor 36.
- the defect data processing unit 221 of the AI model generation device 2 receives the defect image data acquired in (S101) from the welding device 3, converts the data format from CSV data to NPY data, and stores it in the defect image data storage unit 211.
- the defect data processing unit 221 causes the first output device 23 to output a defect image based on the defect image data stored in the defect image data storage unit 211.
- the first input device 24 accepts the user's input of defect information specifying the type and position of the welding defect in the defect image.
- the defect data processing unit 221 causes the annotation data identified based on the user's input of the defect information to the first input device 24 to be associated with the defect image data of each image and stored in the annotation data storage unit 213.
- the defect data processing unit 221 outputs to the first output device 23 an image with defects based on the defect image data stored in the defect image data storage unit 211.
- a boundary identification input is accepted from the user that identifies the boundary between the formation area and non-formation area of the weld defect in the image with defects.
- the defect data processing unit 221 stores segmentation data indicating the boundary between the formation area and non-formation area of the weld defect in the segmentation data storage unit 214 in association with the defect image data of each image.
- the defect-free data processing unit 222 appropriately displays an image prompting the user to input information, and accepts the base material designation input and the weld mark identification input. Note that an input specifying the position of the weld bead in the base material may also be made as the weld mark identification input.
- the defect-free data processing unit 222 then creates defect-free image data that shows a 3D image including the weld bead and the base material. In the 3D image shown by the defect-free image data, for example, as shown in FIG. 2, a weld bead of a shape specified by the weld mark identification input is formed on a base material of a size specified by the base material designation input.
- the defect-free data processing unit 222 automatically creates segmentation data that identifies the boundary between the areas where the weld marks are formed and the areas where they are not formed, based on the defect-free image data, and stores the segmentation data in the segmentation data storage unit 214.
- the data expansion unit 223 acquires multiple captured 3D data showing 3D images showing welding defects based on the defect image data stored in the defect image data storage unit 211 and the annotation data of the defect image data stored in the annotation data storage unit 213.
- the data expansion unit 223 also acquires segmentation data specifying the boundary between the formation area and non-formation area of the welding defect in the 3D image shown by each captured 3D data based on the segmentation data of the defect image data stored in the segmentation data storage unit 214 as segmentation data of the captured 3D data.
- the data expansion unit 223 repeatedly performs data expansion, described below, on the multiple captured 3D data acquired in (S107) to acquire a larger number of defective 3D data than the captured 3D data, and stores the defective 3D data in the defective 3D data storage unit 215.
- the detailed process of data expansion will be described later.
- the learning data creation unit 224 creates learning image data based on the defect 3D data stored in the defect 3D data storage unit 215 and the defect-free image data stored in the defect-free image data storage unit 212. Specifically, the learning data creation unit 224 creates multiple learning 3D data by combining multiple patterns of one or more of the multiple welding defects respectively indicated by the multiple defect 3D data stored in the defect 3D data storage unit 215 and synthesizing them with the 3D image indicated by the defect-free image data. Then, the learning data creation unit 224 stores the learning data consisting of multiple data sets of the created learning 3D data and annotation data for that learning 3D data in the learning data storage unit 216.
- the AI model generation unit 225 creates a defect detection model by performing machine learning using the learning data stored in the learning data storage unit 216. Then, the AI model generation unit 225 stores parameters that specify the generated defect detection model in the parameter storage unit 217. The parameters stored in the parameter storage unit 217 are sent to the welding defect information calculation unit 372 of the welding device 3. The welding defect information calculation unit 372 stores the parameters.
- the user starts the power supply to the welding power source 32 of the welding device 3 and performs arc welding to form a weld bead on the base material W together with the weld defect.
- the image processing unit 371 receives the shape data acquired by the 3D sensor 36.
- the image processing unit 371 then converts this shape data into image data of a welding image including the welded portion PW.
- the welding defect information calculation unit 372 then calculates (predicts) welding defect information based on the image data generated by the image processing unit 371 using a defect detection model specified by the parameters sent from the AI model generation device 2.
- the welding defect display image generation unit 373 calculates the pixel value of each pixel of the welding defect display image that displays the welding defect information based on the welding defect information calculated by the welding defect information calculation unit 372.
- the second output device 38 then displays the welding defect display image based on the pixel value of each pixel calculated by the welding defect display image generation unit 373.
- the data expansion unit 223 first selects first and second selected 3D data from the multiple captured 3D data acquired in step S107 by random numbers.
- the first and second selected 3D data are data representing a point cloud including a welding defect that is recessed or protruded in the Z-axis direction in a predetermined XYZ orthogonal coordinate system.
- Figures 5 and 6 exemplify a set of the first and second selected 3D data and the final 3D data.
- Figures 5 and 6 are examples of 3D images of a hole as a welding defect represented by 3D data.
- Images D11 and D12 are 3D images before synthesis represented by the first selected 3D data
- image D21 is a 3D image before synthesis represented by the second selected 3D data.
- Images D11, D12, and image D21 in Figure 5 are images of the welding defect viewed from the Z-axis direction.
- the symbol B indicates the bottom of the hole.
- the data expansion unit 223 uses the first selected 3D data as it is as the first 3D data to be combined (images D11 and D12).
- the first 3D data to be combined is based on the first selected 3D data.
- image D12 is an image of the same sample as image D11, photographed under different lighting conditions.
- Images D22 to D28 which are images obtained by applying multiple types of conversion processing to image D21, and images D31 to D38, which are obtained by combining the two images, will be described in detail later.
- the data expansion unit 223 randomly selects one mode from among a plurality of modes that each perform a plurality of types of conversion processing, including inversion of the welding defect in the X-axis direction and Y-axis direction, transposition that swaps the X value and the Y value, and a combination thereof, and a mode that does not perform any of the plurality of types of conversion processing.
- the data expansion unit 223 performs one of the above conversion processing on the second selected 3D data (pre-conversion 3D data) representing, for example, image D21, and outputs it as pre-size-adjustment 3D data (post-conversion 3D data), or outputs the second selected 3D data as it is as pre-size-adjustment 3D data without performing the above conversion processing.
- pre-conversion 3D data representing, for example, image D21
- post-conversion 3D data post-conversion 3D data
- the above-mentioned multiple types of conversion processes include horizontal inversion (inversion in the X-axis direction), vertical inversion (inversion in the Y-axis direction), horizontal/vertical inversion (inversion in the X-axis and Y-axis directions), transposition in which the X value and the Y value are swapped, horizontal inversion and transposition, vertical inversion and transposition, and horizontal/vertical inversion and transposition.
- image D21 (corresponding to the second selected 3D data (before synthesis)) is converted to an image of the 3D data before size adjustment.
- image D21 is converted to image D22 by horizontal inversion, to image D23 by vertical inversion, to image D24 by horizontal/vertical inversion, to image D25 by transposition, to image D26 by horizontal inversion and transposition, to image D27 by vertical inversion and transposition, and to image D28 by horizontal/vertical inversion and transposition.
- the data expansion unit 223 acquires second compositing 3D data (not shown) by enlarging or reducing the size of the pre-adjustment 3D data (e.g., images D21 to D28) in at least one of the X-axis direction and the Y-axis direction so as to match the size of the pre-adjustment 3D data to the size of the first selected 3D data (e.g., image D11 or image D12).
- the image sizes in the X-axis direction and the Y-axis direction of the 3D images represented by the first compositing 3D data (e.g., images D11, D12) and the second compositing 3D data are common to each other.
- the pre-adjustment 3D data is used as it is as the second compositing 3D data.
- the second compositing 3D data is based on the converted 3D data and the second selected 3D data.
- the data extension unit 223 normalizes the first and second compositing 3D data by multiplying the Z value of each point in the point cloud by a common value 1/ ⁇ to set the maximum Z value of the point cloud to 1 and the minimum to 0, thereby acquiring the first and second normalized 3D data.
- the data extension unit 223 normalizes the maximum value to 1 and the minimum to 0, but normalization may be performed to set the maximum value to a common first value other than 1 and the minimum value to a common second value other than 0. Normalization makes it possible to align the height or depth of the welding defects before compositing. If normalization is not performed, the change in the height difference of the welding defects before compositing will have a large effect on the shape of the welding defects after compositing, such as the formation of unintended steps in the welding defects after compositing.
- the data expansion unit 223 determines whether the welding defect of the first 3D data to be synthesized is concave (e.g., a hole) or convex (e.g., a spatter, a protrusion) based on the first 3D data to be synthesized and the segmentation data of the first 3D data to be synthesized (obtained in (S107)). If the average Z value of the welding defect formation area in the 3D image shown by the first 3D data to be synthesized is equal to or less than the average Z value of the welding defect non-formation area, the data expansion unit 223 determines that the welding defect is concave, and proceeds to processing of (S206).
- concave e.g., a hole
- convex e.g., a spatter, a protrusion
- the data expansion unit 223 determines that the welding defect is convex, and proceeds to processing of (S207). That is, the data extension unit 223 selects which step to execute, between (S206) and (S207), based on the first compositing 3D data and the segmentation data of the first compositing 3D data (obtained in (S107)).
- the selection of which step to execute, between (S206) and (S207), is based on the first compositing 3D data and its segmentation data, but the selection may also be based on the second compositing 3D data, or the first and second compositing 3D data and their segmentation data.
- the data extension unit 223 generates the composite 3D data by selecting, as the Z value of each point, the value (the minimum Z value) indicating the deeper position among the Z values of points having the same XY coordinates as the point shown in the first and second normalized 3D data.
- the Z value of each point is set to the smaller of the Z values of points having the same XY coordinates as the point shown in the first and second normalized 3D data.
- the XYZ coordinates of one point included in the point cloud shown by the first normalized 3D data are (3, 3, -4) and the XYZ coordinates of one point included in the point cloud shown by the second normalized 3D data are (3, 3, -2), the XYZ coordinates (3, 3, -4) of one point included in the point cloud shown by the composite 3D data can be determined based on these coordinates.
- the bottom of the weld defect is formed at a certain depth in the area on the XY coordinate plane where the bottom of the weld defect is located in the first 3D data for combination, and in the area on the XY coordinate plane where the bottom of the weld defect is located in the second 3D data for combination.
- the data expansion unit 223 proceeds to processing (S208).
- the bottom of the weld defect in the 3D data is, in other words, formed by 3D data indicating a point cloud including a weld defect that is recessed in the Z-axis direction in a predetermined XYZ orthogonal coordinate system.
- the data expansion unit 223 generates the synthesized 3D data by selecting, as the Z value of each point, a value (maximum Z value) that indicates a higher Z value among the Z values of points that have common XY coordinates with the point shown in the first and second normalized 3D data.
- a value maximum Z value
- the Z value of each point is set to the higher Z value among the Z values of points that have common XY coordinates with the point shown in the first and second normalized 3D data.
- the top of the weld is formed with a certain protruding height in the area on the XY coordinate plane where the top of the weld is located in the first synthesis 3D data and in the area on the XY coordinate plane where the top of the weld is located in the second synthesis 3D data.
- the data expansion unit 223 proceeds to the process of (S208).
- the top of the weld defect in the 3D data is, in other words, formed by 3D data that shows a group of points including a weld defect that protrudes in the Z-axis direction in a specified XYZ orthogonal coordinate system.
- the data expansion unit 223 performs a process on the post-combination 3D data by multiplying the Z coordinate of each point by a value ⁇ (the reciprocal of the value 1/ ⁇ multiplied during normalization in (S204)) to obtain final 3D data (corresponding, for example, to images D31 to D38 in Figures 5 and 6).
- the final 3D data is based on the post-combination 3D data.
- the data expansion unit 223 then stores the final 3D data in the defective 3D data storage unit 215.
- the data expansion unit 223 repeatedly executes the processes (S201) to (S208), thereby obtaining a large amount of final 3D data.
- final 3D data indicating images D31 to D34 can be obtained from the first and second selected 3D data indicating images D11 and D21.
- Image D31 is an image obtained by combining images D11 and D21
- image D32 is an image obtained by combining images D11 and D22
- image D33 is an image obtained by combining images D11 and D23
- image D34 is an image obtained by combining images D11 and D24.
- final 3D data indicating images D35 to D38 can be obtained from the first and second selected 3D data indicating images D12 and D21.
- Image D35 is an image obtained by combining images D12 and D25
- image D36 is an image obtained by combining images D12 and D26
- image D37 is an image obtained by combining images D12 and D27
- image D38 is an image obtained by combining images D12 and D28.
- the composite 3D data can be generated by selecting the maximum Z value of each point among the Z values of points that share the same XY coordinates as the point shown in the three or more pieces of normalized 3D data.
- the welding device 3 used to form the weld bead to obtain image data with defects in (S101) and the welding device 3 used to form the weld bead to obtain image data to be input to the defect detection model are a common welding device, but they may be different welding devices.
- the functions of the defect-containing data processing unit 221, the defect-free data processing unit 222, the data extension unit 223, and the learning data creation unit 224, and the functions of the AI model generation unit 225 are realized by one processor 22, but they 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.
- the learning data acquisition device and learning data acquisition method disclosed herein can reduce the effort and work time required to acquire learning data, and are useful as a learning data acquisition device and learning data acquisition method for acquiring image data to be used in machine learning.
- AI model generation device (learning data acquisition device) 22 Processor (learning data acquisition unit) D11, D12, D21-D28, D31-D38 3D images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Mechanical Engineering (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Processing (AREA)
Abstract
Description
22 プロセッサ(学習データ取得部)
D11、D12、D21~D28、D31~D38 3D画像
Claims (8)
- モデルを作成するための機械学習に用いる学習データを取得する学習データ取得装置であって、
前記モデルは、3D画像を示す3Dデータに基づいて溶接不良の位置を含む溶接不良情報を出力するものであり、
前記学習データ取得装置は、
所定のXYZ直交座標系においてZ軸方向に凹む溶接不良を含む点群を示し、X軸方向及びY軸方向の画像サイズが互いに共通の第1及び第2の合成用3Dデータに基づいて、前記第1の合成用3Dデータにおいて前記溶接不良の底部が位置するXY座標平面上の領域、及び前記第2の合成用3Dデータにおいて前記溶接不良の底部が位置するXY座標平面上の領域に、前記溶接不良の底部が一定の深さで形成された3D画像を示す合成後3Dデータを生成する第1の合成ステップ、及び
所定のXYZ直交座標系においてZ軸方向に突出する溶接不良を含む点群を示し、X軸方向及びY軸方向の画像サイズが互いに共通の第1及び第2の合成用3Dデータに基づいて、前記第1の合成用3Dデータにおいて前記溶接不良の頂部が位置するXY座標平面上の領域、及び前記第2の合成用3Dデータにおいて前記溶接不良の頂部が位置するXY座標平面上の領域に、前記溶接不良の頂部が一定の突出高さで形成された3D画像を示す合成後3Dデータを生成する第2の合成ステップのうちの少なくとも一方の合成ステップと、
前記合成後3Dデータに基づく最終3Dデータを含む複数の不良3Dデータによって示される複数の溶接不良の3D画像を複数パターン組み合わせて、溶接痕を含む3D画像と合成することにより、複数の学習用3Dデータを作成する学習データ作成ステップとを実行する学習データ取得部を備えることを特徴とする学習データ取得装置。 - 請求項1に記載の学習データ取得装置において、
前記第1の合成ステップは、
前記第1及び第2の合成用3Dデータに対し、点群の各点のZの値に共通の所定値1/αを掛けて点群のZの値の最高値を共通の第1の値、最低値を共通の第2の値にする正規化を行うことにより、第1及び第2の正規化後3Dデータを取得する正規化ステップと、
各点のZの値として、前記第1及び第2の正規化後3Dデータで示される当該点とXY座標が共通の点のZの値のうち、より深いことを示す値を選択することにより、前記合成後3Dデータを生成する第1のZ値選択ステップとを実行するものであり、
前記第2の合成ステップは、
前記正規化ステップと、
各点のZの値として、前記第1及び第2の正規化後3Dデータで示される当該点とXY座標が共通の点のZの値のうち、より高いことを示す値を選択することにより、前記合成後3Dデータを生成する第2のZ値選択ステップとを実行するものであり、
前記学習データ取得部は、前記合成後3Dデータに対し、各点のZ座標に前記所定値1/αの逆数αを掛ける処理を行うことにより、前記最終3Dデータを取得することを特徴とする学習データ取得装置。 - 請求項2に記載の学習データ取得装置において、
前記学習データ取得部は、
前記第1及び第2の合成用3Dデータの少なくとも一方の合成用3Dデータと、当該少なくとも一方の合成用3Dデータによって示される3D画像における前記溶接不良の形成領域と非形成領域との境界を指定するセグメンテーションデータとに基づいて、前記第1及び第2の合成ステップのうち、いずれの合成ステップを実行するかを選択することを特徴とする学習データ取得装置。 - 請求項1に記載の学習データ取得装置において、
前記学習データ取得部は、
前記第2の合成用3Dデータを、サイズ調整前3DデータをX軸方向及びY軸方向の少なくとも一方に拡大又は縮小することにより取得するサイズ調整ステップをさらに実行することを特徴とする学習データ取得装置。 - 請求項1に記載の学習データ取得装置において、
前記学習データ取得部は、
前記溶接不良を示す複数の3Dデータから、第1及び第2の選択3Dデータを乱数により選択する3Dデータ選択ステップをさらに実行し、
前記第1の合成用3Dデータは、前記第1の選択3Dデータに基づき、かつ前記第2の合成用3Dデータは、前記第2の選択3Dデータに基づくものであることを特徴とする学習データ取得装置。 - 請求項1に記載の学習データ取得装置において、
前記学習データ取得部は、
溶接不良の反転、転置、及びそれらの組み合わせを含む複数種類の変換処理をそれぞれ行う複数のモードと、前記複数種類の変換処理のいずれも行わないモードとから、1つのモードを乱数により選択するモード選択ステップと、
前記モード選択ステップで選択したモードで、溶接不良を示す変換前3Dデータに対していずれか1つの前記変換処理を行って変換後3Dデータとして出力するか、又は前記変換処理を行わずに前記変換前3Dデータをそのまま前記変換後3Dデータとして出力する変換処理ステップとをさらに実行し、
前記第2の合成用3Dデータは、前記変換後3Dデータに基づくものであることを特徴とする学習データ取得装置。 - モデルを作成するための機械学習に用いる学習データを取得する学習データ取得方法であって、
前記モデルは、3D画像を示す3Dデータに基づいて溶接不良の位置を含む溶接不良情報を出力するものであり、
前記学習データ取得方法は、
所定のXYZ直交座標系においてZ軸方向に凹む溶接不良を含む点群を示し、X軸方向及びY軸方向の画像サイズが互いに共通の第1及び第2の合成用3Dデータに基づいて、前記第1の合成用3Dデータにおいて前記溶接不良の底部が位置するXY座標平面上の領域、及び前記第2の合成用3Dデータにおいて前記溶接不良の底部が位置するXY座標平面上の領域に、前記溶接不良の底部が一定の深さで形成された3D画像を示す合成後3Dデータを生成する第1の合成ステップ、及び
所定のXYZ直交座標系においてZ軸方向に突出する溶接不良を含む点群を示し、X軸方向及びY軸方向の画像サイズが互いに共通の第1及び第2の合成用3Dデータに基づいて、前記第1の合成用3Dデータにおいて前記溶接不良の頂部が位置するXY座標平面上の領域、及び前記第2の合成用3Dデータにおいて前記溶接不良の頂部が位置するXY座標平面上の領域に、前記溶接不良の頂部が一定の突出高さで形成された3D画像を示す合成後3Dデータを生成する第2の合成ステップのうちの少なくとも一方の合成ステップと、
前記合成後3Dデータに基づく最終3Dデータを含む複数の不良3Dデータによって示される複数の溶接不良の3D画像を複数パターン組み合わせて、溶接痕を含む3D画像と合成することにより、複数の学習用3Dデータを作成する学習データ作成ステップとを実行するものであり、
前記第1の合成ステップは、
前記第1及び第2の合成用3Dデータに対し、点群の各点のZの値に共通の所定値1/αを掛けて点群のZの値の最高値を共通の第1の値、最低値を共通の第2の値にする正規化を行うことにより、第1及び第2の正規化後3Dデータを取得する正規化ステップと、
各点のZの値として、前記第1及び第2の正規化後3Dデータで示される当該点とXY座標が共通の点のZの値のうち、より深いことを示す値を選択することにより、前記合成後3Dデータを生成する第1のZ値選択ステップとを実行するものであり、
前記第2の合成ステップは、
前記正規化ステップと、
各点のZの値として、前記第1及び第2の正規化後3Dデータで示される当該点とXY座標が共通の点のZの値のうち、より高いことを示す値を選択することにより、前記合成後3Dデータを生成する第2のZ値選択ステップとを実行するものであり、
前記学習データ取得方法は、前記合成後3Dデータに対し、各点のZ座標に前記所定値1/αの逆数αを掛ける処理を行うことにより、前記最終3Dデータを取得することを特徴とする学習データ取得方法。 - 請求項7に記載の学習データ取得方法において、
溶接不良の反転、転置、及びそれらの組み合わせを含む複数種類の変換処理をそれぞれ行う複数のモードと、前記複数種類の変換処理のいずれも行わないモードとから、1つのモードを乱数により選択するモード選択ステップと、
前記モード選択ステップで選択したモードで、溶接不良を示す変換前3Dデータに対していずれか1つの前記変換処理を行って変換後3Dデータとして出力するか、又は前記変換処理を行わずに前記変換前3Dデータをそのまま前記変換後3Dデータとして出力する変換処理ステップとをさらに実行し、
前記第2の合成用3Dデータは、前記変換後3Dデータに基づくものであることを特徴とする学習データ取得方法。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202480004387.XA CN120584349A (zh) | 2023-12-26 | 2024-10-31 | 学习数据获取装置及学习数据获取方法 |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2023-219737 | 2023-12-26 | ||
| JP2023219737 | 2023-12-26 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2025142115A1 true WO2025142115A1 (ja) | 2025-07-03 |
Family
ID=96218924
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2024/038846 Pending WO2025142115A1 (ja) | 2023-12-26 | 2024-10-31 | 学習データ取得装置、及び学習データ取得方法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN120584349A (ja) |
| WO (1) | WO2025142115A1 (ja) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210142467A1 (en) * | 2019-11-13 | 2021-05-13 | Software Ag | Systems and methods of generating datasets for training neural networks |
| WO2022102523A1 (ja) * | 2020-11-10 | 2022-05-19 | パナソニックIpマネジメント株式会社 | データ作成システム、学習システム、推定システム、処理装置、評価システム、データ作成方法、及びプログラム |
-
2024
- 2024-10-31 CN CN202480004387.XA patent/CN120584349A/zh active Pending
- 2024-10-31 WO PCT/JP2024/038846 patent/WO2025142115A1/ja active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210142467A1 (en) * | 2019-11-13 | 2021-05-13 | Software Ag | Systems and methods of generating datasets for training neural networks |
| WO2022102523A1 (ja) * | 2020-11-10 | 2022-05-19 | パナソニックIpマネジメント株式会社 | データ作成システム、学習システム、推定システム、処理装置、評価システム、データ作成方法、及びプログラム |
Non-Patent Citations (1)
| Title |
|---|
| ZHANG, HAODONG ET AL.: "Weld Defect Detection Based on Deep Learning Method", 2019 IEEE 15TH INTERNATIONAL CONFERENCE ON AUTOMATION SCIENCE AND ENGINEERING, 2019, pages 1574 - 1579, XP033618366, Retrieved from the Internet <URL:https://doi.org/10.1109/COASE.2019.8842998> [retrieved on 20241202], DOI: 10.1109/COASE.2019.8842998 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN120584349A (zh) | 2025-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP4492654B2 (ja) | 3次元計測方法および3次元計測装置 | |
| JP5429872B2 (ja) | 加工対象物を溶接するロボットを制御する方法および装置 | |
| JP5090121B2 (ja) | 調整装置、レーザ加工装置、調整方法、および調整プログラム | |
| JP3930482B2 (ja) | 3次元視覚センサ | |
| JP2015044212A (ja) | レーザ加工装置 | |
| CN102067172B (zh) | 用于检流计和激光系统的多模型登记(mmr) | |
| JP2019063954A (ja) | ロボットシステム、キャリブレーション方法及びキャリブレーションプログラム | |
| US20190287258A1 (en) | Control Apparatus, Robot System, And Method Of Detecting Object | |
| JPWO2018168757A1 (ja) | 画像処理装置、システム、画像処理方法、物品の製造方法、プログラム | |
| WO2022163580A1 (ja) | 視覚センサにて取得される3次元の位置情報から断面画像を生成する処理装置および処理方法 | |
| CN121360917A (zh) | 外观检查装置、焊接系统、形状数据的修正方法以及焊接部位的外观检查方法 | |
| WO2025142115A1 (ja) | 学習データ取得装置、及び学習データ取得方法 | |
| JP5371928B2 (ja) | 欠陥検査方法及びその装置 | |
| WO2023074183A1 (ja) | 学習支援システム、外観検査装置、外観検査用ソフトの更新装置及び外観検査用モデルの更新方法 | |
| JP4932202B2 (ja) | 画像測定装置用パートプログラム生成装置、画像測定装置用パートプログラム生成方法、及び画像測定装置用パートプログラム生成用プログラム | |
| WO2023166940A1 (ja) | 注視領域モデル生成システム及び推論装置 | |
| WO2024262271A1 (ja) | 学習データ取得装置、溶接システム、及び学習データ取得方法 | |
| Iakovou et al. | Integrated sensors for robotic laser welding | |
| WO2024262196A1 (ja) | 学習データ取得装置、及び学習データ取得方法 | |
| JP2020051762A (ja) | 三次元形状検査装置、三次元形状検査方法、三次元形状検査プログラム、コンピュータ | |
| WO2022270578A1 (ja) | オフライン教示装置およびオフライン教示方法 | |
| JP7331618B2 (ja) | ガラスの検査装置、ガラスの検査方法 | |
| JP6078356B2 (ja) | テンプレートマッチング条件設定装置、及び荷電粒子線装置 | |
| JP2025002521A (ja) | 学習データ取得装置、及び学習データ取得方法 | |
| JP2026037765A (ja) | 学習データ生成装置、及び学習データ生成方法 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| ENP | Entry into the national phase |
Ref document number: 2025507432 Country of ref document: JP Kind code of ref document: A |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2025507432 Country of ref document: JP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 202480004387.X Country of ref document: CN |
|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 24912008 Country of ref document: EP Kind code of ref document: A1 |
|
| WWP | Wipo information: published in national office |
Ref document number: 202480004387.X Country of ref document: CN |