Evaluation system and ai model generation device

Abstract

To improve detection accuracy of a failure detection model.SOLUTION: An evaluation system 30 is provided with: a second display device 38; and a computer 37 for generating position specification information for specifying a position of a welding failure and an unspecifiable portion in an image by using a failure detection model based on image data showing the image, and making the second display device 38 perform display based on the generated position specification information.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an evaluation system that generates information that identifies the positions of characteristic parts in an image using a detection model based on image data that represents the image, and an AI model generation device that generates the detection model through machine learning. [Background technology]

[0002] Patent Document 1 discloses an evaluation system equipped with an inference unit that uses a detection model to generate information for identifying the positions of characteristic features in an image acquired by photographing a welding location. This evaluation system generates a detection model through deep learning using annotation data acquired by manually identifying characteristic features in the image. The evaluation system then uses this detection model to assign annotation data by inference. An operator visually checks the inference results and determines whether the annotation data needs to be corrected, and if necessary, manually corrects the annotation data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 2020-035095 Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, a defect detection model generated by machine learning may be used to detect welding defects such as holes, pits, spatter, undercut, and underfill. Learning data for machine learning is generated by annotation work in which a worker visually identifies welding defects in an image. However, an image may contain unidentifiable parts that cannot be visually identified as corresponding to welding defects. If such unidentifiable parts are uniformly identified as not corresponding to welding defects or as being welding defects, the detection accuracy of the generated defect detection model may be degraded.

[0005] The present disclosure has been made in consideration of the above points, and its purpose is to improve the detection accuracy of a defect detection model. [Means for solving the problem]

[0006] In order to achieve the above object, a first aspect of the present disclosure is characterized by being an evaluation system including a display device and a computer that generates position identification information that identifies the positions of welding defects and unidentifiable areas in an image using a defect detection model based on image data representing the image, and causes the display device to display based on the generated position identification information.

[0007] As a result, by using annotation data that identifies the locations of poor welds and unidentifiable parts in multiple images to generate a defect detection model, the detection accuracy of the defect detection model can be improved compared to when the annotation data uniformly determines that unidentifiable parts are not poor welds or that they are poor welds.

[0008] A second aspect of the present disclosure is characterized by being an AI model generation device that takes image data representing an image as input, and generates a defect detection model that outputs position identification information that identifies the positions of welding defects and unidentifiable areas in the image through machine learning using annotation data for multiple image data that identifies the positions of the welding defects and unidentifiable areas.

[0009] As a result, annotation data that identifies the locations of poor welds and unidentifiable parts in multiple images is used to generate the defect detection model, thereby improving the accuracy of the defect detection model compared to when the annotation data uniformly determines that all unidentifiable parts are not poor welds or uniformly determines that all parts are poor welds. [Effects of the Invention]

[0010] According to the present disclosure, the accuracy of the defect detection model can be improved. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a schematic configuration diagram of a welding system including an evaluation system according to an embodiment of the present disclosure. [Figure 2] FIG. 2 shows an input screen for information required to determine whether or not to display information identifying the positions of the welding defect and the unidentifiable portion on the second display device. [Figure 3] FIG. 3 is a flowchart illustrating the operation of the AI ​​model generation device for acquiring learning data. [Figure 4] FIG. 4 is a diagram illustrating an example of a convex unidentifiable part in a 3D image. [Figure 5] FIG. 5 is a diagram illustrating an example of a concave, unidentifiable area in a 3D image. [Figure 6] FIG. 6 is a diagram illustrating an example of an unidentifiable area in a data-deficient system in a 3D image. [Figure 7] FIG. 7 is a diagram illustrating an example of an unidentifiable portion of a reflection system in a 3D image. [Figure 8] FIG. 8 is a diagram illustrating an example of an image output by the first display device in a state where the types and positions of the welding defect and the unidentifiable portion are designated by the user's input operation. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of the present disclosure will be described in detail 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 its uses.

[0013] 1 shows a welding system 1. The welding system 1 includes an AI model generation device 2 according to an embodiment of the present disclosure and a welding device 3 that performs welding.

[0014] The AI ​​model generation device 2 acquires image data that constitutes learning data used in machine learning to create a defect detection model.

[0015] The defect detection model is an object detection model that predicts and generates position identification information based on input image data. In other words, the input of the defect detection model is image data representing an image, and the output of the defect detection model is position identification information. The defect detection model is represented by a convolutional neural network (CNN). The position identification information includes category information indicating the type of weld defect and unidentifiable portion in the 3D image represented by the image data, and position information indicating the position of the weld defect and unidentifiable portion. Specifically, the category information indicates whether the identified portion is a hole (perforation), pit, spatter, undercut, convex unidentifiable portion, concave unidentifiable portion, data missing unidentifiable portion, or reflective unidentifiable portion. In this way, the position identification information identifies the positions of multiple types of unidentifiable portions with different characteristics as the position of the unidentifiable portion. Holes (perforations), pits, spatter, and undercut are weld defects. A pit is an opening on the surface of a weld bead. The position information indicates the position of a bounding box including the weld defect or unidentifiable portion. The bounding box is a rectangular boundary area that surrounds the periphery of the object (defective weld, unidentifiable part).

[0016] In addition, the AI ​​model generation device 2 acquires annotation data for multiple image data that identifies the locations of welding defects and unidentifiable areas, and generates a defect detection model through machine learning using multiple data sets of image data and the annotation data as learning data.

[0017] Specifically, the AI ​​model generation device 2 includes a storage device 21, a processor 22, a first display device 23, and a first input device 24.

[0018] The storage device 21 includes an image data storage unit 211 , an annotation data storage unit 212 , and a parameter storage unit 213 .

[0019] The image data storage unit 211 stores image data of a 3D image acquired by photographing a welding point with a 3D sensor 36, which will be described later.

[0020] The annotation data storage unit 212 stores position specifying information specified by a user's input to the first input device 24 in a state in which an image based on the image data is output to the first display device 23.

[0021] The parameter storage unit 213 stores parameters that specify a fault detection model generated by the processor 22, which will be described later.

[0022] The processor 22 includes an image data receiving unit 221, an annotation data receiving unit 222, and an AI model generating unit 223.

[0023] The image data receiving unit 221 receives from the welding device 3 image data of a 3D image acquired by photographing the welding point with the 3D sensor 36 (described later), and stores the image data in the image data storage unit 211.

[0024] The annotation data receiving unit 222 outputs an image based on the image data stored in the image data storage unit 211 to the first display device 23. In this state, an input signal corresponding to a user's input operation is received from the first input device 24. Here, the first input device 24 receives the user's input operation specifying the type and position of the poor weld and the unidentifiable portion in the image. When the user's input operation is completed, the annotation data receiving unit 222 stores the position specifying information specified based on the user's input operation in the annotation data storage unit 212 as annotation data of the position specifying information.

[0025] The AI ​​model generation unit 223 creates a fault detection model by performing machine learning using learning data consisting of multiple sets of data pairs, each set consisting of image data stored in the image data storage unit 211 and annotation data stored in the annotation data storage unit 212. Specifically, the AI ​​model generation unit 223 specifies the weights and biases of each node constituting the CNN that expresses the fault detection model as parameters that specify the fault detection model. Then, the AI ​​model generation unit 223 stores the specified parameters in the parameter storage unit 213.

[0026] The first display device 23 outputs an image based on the image data stored in the image data storage unit 211. The first display device 23 is configured by, for example, a liquid crystal monitor.

[0027] The first input device 24 receives an input operation from the user specifying the type and position of the welding defect and the unidentifiable portion in the image, and outputs an input signal according to the received input operation.

[0028] The welding apparatus 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 display device 38, and a second input device 39. The computer 37, the second display device 38, and the second input device 39 constitute an evaluation system 30 according to an embodiment of the present disclosure. When power is supplied from the welding power source 32 to the welding wire WI held in 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 apparatus 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.

[0029] 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 a wire feeder (not shown) to welding torch 31. 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. The welding conditions are changed by welding condition change unit 375, which will be described later.

[0030] 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, traces a predetermined welding trajectory and moves to a desired position. Robot control unit 35 controls the operation of robot arm 34 by executing a predetermined program.

[0031] The 3D sensor 36 is attached to the welding torch 31 and measures the shape of the welded portion PW of the workpiece W. The 3D sensor 36 is a three-dimensional shape measurement sensor that includes, 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.

[0032] Computer 37 executes software implemented on a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) to realize the functions of multiple functional blocks within computer 37. Computer 37 has an image processing unit 371, an inference unit 372, a welding defect display image generation unit 373, a display determination unit 374, and a welding condition change unit 375.

[0033] 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.

[0034] The inference unit 372 generates (predicts) position identification 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 position identification information generated here identifies the type and position of the welding defect and the unidentifiable portion in the image shown by the image data generated by the image processing unit 371.

[0035] The poor welding display image generating unit 373 calculates the pixel value of each pixel of the display image based on the position specifying information generated by the inference unit 372. Then, the poor welding display image generating unit 373 outputs the calculated pixel values, thereby causing the second display device 38 to display an image based on the position specifying information generated by the inference unit 372. The information specifying the display image may be compressed using a JPEG format or the like.

[0036] The display determination unit 374 determines whether to display, on the second display device 38, information identifying the positions of the poor welds and unidentifiable portions whose positions are identified by the position identification information generated by the inference unit 372. This determination is made based on an input to the second input device 39 while an image such as that shown in FIG. 2 is displayed on the second display device 38, and on at least one of the width, length, depth, height, and sensitivity of the poor welds and unidentifiable portions. The information identifying the positions of the poor welds and unidentifiable portions is a rectangular enclosure surrounding the poor welds and unidentifiable portions in a welding image including the welded portion PW. The display determination unit 374 determines whether to display, on the second display device 38, information identifying the positions of each poor weld and each unidentifiable portion based on an input to the second input device 39 while an image such as that shown in FIG. 2 is displayed on the second display device 38. The user sets a numerical range for at least one of the width, length, depth, height, and sensitivity for each poor weld and each unidentifiable portion by inputting to the second input device 39 while an image such as that shown in FIG. 2 is displayed on the second display device 38. Then, the display determination unit 374 determines whether or not to display the enclosure based on whether or not the dimensions of each defective weld and each unidentifiable portion, the position of which is identified by the position identification information, are within a numerical range set by input to the second input device 39. The determination method will be described in detail later.

[0037] Welding condition change unit 375 sets future welding conditions based on the types of poor welds and unidentifiable portions whose positions are indicated by the position identification information generated by inference unit 372. Welding condition change unit 375 then controls output control unit 33, robot control unit 35, etc. so that the set welding conditions are achieved.

[0038] For example, when a predetermined number or more spatter positions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 performs at least one of a process of increasing the welding voltage (voltage between the welding wire WI and the workpiece W) and a process of decreasing the welding current (current flowing between the welding wire WI and the workpiece W). Furthermore, when a predetermined number or more pit positions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 performs at least one of a process of increasing the flow rate of the shielding gas, a process of shortening the extension length of the welding wire WI, a process of narrowing the weaving width, and a process of decreasing the welding speed. Furthermore, when a predetermined number or more hole positions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 performs at least one of a process of decreasing the welding voltage, a process of decreasing the welding current, a process of decreasing the welding speed, or other heat input reduction process, and a program modification of the robot control unit 35. Furthermore, when a predetermined number or more undercut positions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 performs at least one of the following processes: reducing the welding voltage, increasing the welding current, and reducing the welding speed. Furthermore, when a predetermined number or more convex unidentifiable portions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 performs at least one of the following processes: adjusting the waveform of the welding current at the start and end of welding, and increasing the lift-up amount of the welding torch 31 at the start of welding. Furthermore, when a predetermined number or more concave unidentifiable portions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 performs at least one of the following processes: reducing the welding voltage, increasing the welding current, and reducing the welding speed. Furthermore, when a predetermined number or more data-missing unidentifiable portions are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 adjusts at least one of the angle and sensitivity of the 3D sensor 36.In addition, when the positions of a predetermined number or more of unidentifiable parts of the reflection system are identified by the position identification information generated by the inference unit 372, the welding condition change unit 375 adjusts at least one of the angle and sensitivity of the 3D sensor 36.

[0039] The second display device 38 displays a display image based on the pixel value of each pixel calculated by the welding defect display image generating unit 373. The second display device 38 is configured by a liquid crystal monitor or the like.

[0040] The second input device 39 accepts a predetermined input operation by the user. The second input device 39 accepts an input operation for specifying whether or not to display poor welding and unidentifiable portions, and an input operation for specifying a numerical range for each dimension when poor welding and unidentifiable portions are to be displayed.

[0041] Next, the operation of the AI ​​model generation device 2 to acquire learning data will be described with reference to the flowchart of FIG.

[0042] 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.

[0043] Next, in (S102), the image data receiving unit 221 of the AI ​​model generation device 2 receives the image data acquired in (S101) from the welding device 3 and stores it in the image data storage unit 211.

[0044] Next, in (S103), the annotation data receiving unit 224 outputs an image based on the image data stored in the image data storage unit 211 to the first display device 23. In this state, an input signal corresponding to a user's input operation is received from the first input device 24. At this time, the user performs an input operation to specify the type and location of a welding defect in the image output by the first display device 23. Furthermore, the user also performs an input operation to specify the type and location of an unidentifiable portion, which cannot be determined as a hole (perforation), a pit, a sputter, or an undercut. For example, as shown in FIG. 4, if the unidentifiable portion (see the circled portion) has a gently sloping convex shape, the user identifies the unidentifiable portion as a convex unidentifiable portion. Furthermore, as shown in FIG. 5, if the unidentifiable portion (see the circled portion) has a concave shape, the user identifies the unidentifiable portion as a concave unidentifiable portion. Furthermore, for example, as shown in FIG. 6, if at least some data is missing in an unidentifiable portion (see the circled portion), the user identifies the unidentifiable portion as a data-missing unidentifiable portion. For example, as shown in FIG. 7, if the unidentifiable portion (see the circled portion) has a thin, pointed convex shape, the user identifies the unidentifiable portion as a reflective unidentifiable portion. FIG. 8 illustrates an example of an image output by the first display device 23. In this image, the type and location of the poor weld and the unidentifiable portion are specified by the user's input operation, and the poor weld and the unidentifiable portion are enclosed in a rectangular box. In FIG. 8, "Other 1" indicates a convex unidentifiable portion, "Other 2" indicates a concave unidentifiable portion, "Other 3" indicates a data-missing unidentifiable portion, and "Other 4" indicates a reflective unidentifiable portion.

[0045] When the user's input operation in (S103) is completed, in (S104), the annotation data receiving unit 222 stores information indicating the type and position of the poor welds and unidentifiable areas identified based on the user's input operation in the annotation data storage unit 212 as annotation data of position identification information.

[0046] The AI ​​model generation device 2 repeatedly executes the operations (S101) to (S104) in the above flowchart to acquire a plurality of data sets each including image data and annotation data of position identification information for the image data. Thereafter, the AI ​​model generation unit 223 performs machine learning using learning data consisting of the plurality of data sets to generate a defect detection model. The AI ​​model generation unit 223 then stores parameters that specify the generated defect detection model in the parameter storage unit 213. The parameters stored in the parameter storage unit 213 are sent to the inference unit 372 of the welding device 3. The inference unit 372 stores the parameters.

[0047] In the welding apparatus 3, during initial setup, the display determination unit 374 displays an image such as that shown in FIG. 2 on the second display device 38. In this state, the display determination unit 374 sets conditions for determining whether to display enclosures identifying the locations of poor welds and unidentifiable portions, based on a user's input to the second input device 39. In the example of FIG. 2, the "Determination" row indicates whether to display the enclosures. If "Yes" is selected in the "Determination" row by input to the second input device 39 and the conditions indicated in the rows below the "Determination" row are satisfied, the display determination unit 374 determines to display the enclosures. If "No" is selected in the "Determination" row by input to the second input device 39, the display determination unit 374 determines not to display the enclosures. In the example of FIG. 2, "Yes" is selected in the "Determination" row by input to the second input device 39 for all poor welds and unidentifiable portions.

[0048] In Figure 2, in the "Hole" column, the "Length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI Sensitivity" row has a minimum value of 90, and the "Quantity" row has a minimum value of 1. In the "Pit" column, the "Depth" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI Sensitivity" row has a minimum value of 90, and the "Quantity" row has a minimum value of 1. In the "Spatter" column, the "Height" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI Sensitivity" row has a minimum value of 90, and the "Amount" row has a minimum value of 1. In the "Undercut" column, the "Depth" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Length" row has a minimum value of 0.5 mm, the "AI Sensitivity" row has a minimum value of 90, and the "Amount" row has a minimum value of 1. In the "Other 1 (Convex)" column, the "Height" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI Sensitivity" row has a minimum value of 90, and the "Amount" row has a minimum value of 1. In the "Other 2 (concave)" column, the "depth" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI sensitivity" row has a minimum value of 90, and the "quantity" row has a minimum value of 1. In the "Other 3 (defective)" column, the "length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI sensitivity" row has a minimum value of 90, and the "quantity" row has a minimum value of 1. In addition, in the "Other 4 (Reflective)" column, the "Height" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Length" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "Width" row has a minimum value of 0.5 mm and a maximum value of 10,000 mm, the "AI Sensitivity" row has a minimum value of 90, and the "Amount" row has a minimum value of 1.

[0049] 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. Then, the inference unit 372 generates (predicts) position identification 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 the pixel value of each pixel of the defective welding display image based on the position identification information calculated by the inference unit 372. The defective welding display image is an image represented by the image data, to which a fence determined by the display determination unit 374 to be displayed is added, from among fences surrounding the defective welding and unidentifiable positions identified by the position identification information. The defective welding display image may be an image as shown in FIG. 8. The display determination unit 374 determines whether to display a fence surrounding each defective welding and each unidentifiable position identified by the position identification information.

[0050] 2, the display determination unit 374 determines to display enclosures only for holes whose positions are specified in the position identification information and whose lengths are 0.5 mm to 10,000 mm, whose widths are 0.5 mm to 10,000 mm, and whose AI sensitivity is 90 or higher. Similarly, the display determination unit 374 determines to display enclosures only for pits whose positions are specified in the position identification information and whose depths are 0.5 mm to 10,000 mm, whose lengths are 0.5 mm to 10,000 mm, whose widths are 0.5 mm to 10,000 mm, and whose AI sensitivity is 90 or higher. The display determination unit 374 also determines to display an enclosure that only includes spatters, of which the positions are specified in the position identification information, that have a height of 0.5 mm to 10,000 mm, a length of 0.5 mm to 10,000 mm, a width of 0.5 mm to 10,000 mm, and an AI sensitivity of 90 or higher. The display determination unit 374 also determines to display an enclosure that only includes undercuts, of which the positions are specified in the position identification information, that have a depth of 0.5 mm to 10,000 mm, a length of 0.5 mm or higher, and an AI sensitivity of 90 or higher. The display determination unit 374 also determines to display an enclosure that only includes unidentifiable convex portions, of which the positions are specified in the position identification information, that have a height of 0.5 mm to 10,000 mm, a length of 0.5 mm to 10,000 mm, a width of 0.5 mm to 10,000 mm, and an AI sensitivity of 90 or higher. Furthermore, the display determination unit 374 determines to display an enclosure that only surrounds unidentifiable concave regions whose positions are identified in the position identification information, that have a depth of 0.5 mm to 10,000 mm, a length of 0.5 mm to 10,000 mm, a width of 0.5 mm to 10,000 mm, and an AI sensitivity of 90 or higher. Furthermore, the display determination unit 374 determines to display an enclosure that only surrounds unidentifiable defect regions whose positions are identified in the position identification information, that have a length of 0.5 mm to 10,000 mm, a width of 0.5 mm to 10,000 mm, and an AI sensitivity of 90 or higher.Furthermore, the display determination unit 374 determines to display an enclosure for only the unidentifiable parts of the reflective system whose positions are identified in the position identification information, that are 0.5 mm or more and 10,000 mm or less in height, 0.5 mm or more and 10,000 mm or less in length, 0.5 mm or more and 10,000 mm or less in width, and have an AI sensitivity of 90 or more.

[0051] Then, the second display device 38 displays a poor welding display image based on the pixel value of each pixel calculated by the poor welding display image generating unit 373.

[0052] When the welding defect display image includes a predetermined number or more of convex enclosures of unidentifiable portions, the user viewing the welding defect display image checks the tip of the welding wire WI. When the welding defect display image includes enclosures of welding defects, the user determines whether the welding conditions are appropriate based only on the number and type of the welding defects. Only when the welding defect display image does not include enclosures of welding defects, the user can refer to the number and type of unidentifiable portions to determine whether the welding conditions are appropriate.

[0053] Therefore, according to this embodiment, the defect detection model is generated using annotation data that identifies the positions of welding defects and unidentifiable portions in multiple images, so the accuracy of defect detection by the defect detection model can be improved compared to when unidentifiable portions are uniformly determined not to be welding defects or when all unidentifiable portions are uniformly determined to be welding defects in the annotation data.In particular, the overdetection rate (the rate at which portions that are not welding defects are detected) can be reduced.

[0054] In this embodiment, the welding defects are holes, pits, spatters, and undercuts, but the welding defects may be at least one of holes, pits, spatters, undercuts, and underfills.

[0055] Furthermore, although the present invention is applied to arc welding in this embodiment, the present invention may also be applied to laser welding. When the present invention is applied to laser welding and underfill is included in the welding defects, the welding condition changing unit 375 may perform at least one of a process of reducing the laser output, a process of increasing the processing speed, and a process of eliminating the gap between the workpieces when a predetermined number or more underfill positions are identified by the position identification information generated by the inference unit 372.

[0056] Furthermore, the setting of the welding conditions performed by welding condition changing unit 375 in this embodiment may be manually performed by a user who views the defective welding display image.

[0057] Furthermore, when the defective welding display image includes a defective welding enclosure, the welding condition change unit 375 may set the welding conditions based on the number and type of the defective welding only, without referring to the number and type of the unidentifiable portions.Only when the defective welding display image does not include a defective welding enclosure, the welding condition change unit 375 may refer to the number and type of the unidentifiable portions in setting the welding conditions. [Industrial Applicability]

[0058] The evaluation system of the present disclosure can improve the detection accuracy of defective detection models, and is therefore useful as an evaluation system that generates information identifying the positions of characteristic parts in an image using a detection model based on image data representing the image, and as an AI model generation device that generates the detection model through machine learning. [Explanation of symbols]

[0059] 2 AI model generation device 30 Rating System 37 Computer 38 2nd display device 39 Second input device

Claims

1. A display device; and a computer that generates position identification information that identifies the positions of welding defects and unidentifiable parts in an image based on image data representing the image using a defect detection model, and causes the display device to display a display based on the generated position identification information.

2. 2. The evaluation system according to claim 1, The evaluation system is characterized in that the computer further determines whether or not to display information specifying the position of an unidentifiable portion whose position is specified by the position specifying information on the display device.

3. 3. The evaluation system according to claim 2, The evaluation system is characterized in that the computer makes the determination based on at least one of the width, length, depth, height, and sensitivity of the unidentifiable area.

4. 4. The evaluation system according to claim 3, further comprising an input device; An evaluation system characterized in that the computer makes the decision based on whether at least one of the width, length, depth, height, and sensitivity of the unidentifiable area is within a predetermined numerical range set by input to the input device.

5. 3. The evaluation system according to claim 2, The evaluation system is characterized in that the computer sets future welding conditions based on the type of unidentifiable portion whose position is identified by the position identification information.

6. The evaluation system according to any one of claims 1 to 5, The evaluation system is characterized in that the welding defects include at least one of holes, pits, spatters, undercuts, and underfills.

7. The AI ​​model generation device according to any one of claims 1 to 5, The AI ​​model generation device is characterized in that the position identification information identifies the positions of multiple types of unidentifiable parts having different characteristics as the positions of the unidentifiable parts.

8. An AI model generation device that takes image data representing an image as input and generates a defect detection model that outputs position identification information that identifies the positions of welding defects and unidentifiable areas in the image through machine learning using annotation data for multiple image data that identifies the positions of the welding defects and unidentifiable areas.

9. 9. The AI ​​model generation device according to claim 8, The AI ​​model generation device is characterized in that the welding defects include at least one of holes, pits, spatters, undercuts, and underfills.

10. 9. The AI ​​model generation device according to claim 8, The AI ​​model generation device is characterized in that the position identification information identifies the positions of multiple types of unidentifiable parts having different characteristics as the positions of the unidentifiable parts.

Images