JP7779272B2 - Inspection System - Google Patents

Description

This disclosure relates to an inspection system.

Technology for detecting abnormalities in weld beads is known (for example, see Patent Document 1). Patent Document 1 discloses detecting incomplete fusion in a weld bead based on variations in brightness values in an image of the weld bead.

Japanese Patent Application Laid-Open No. 2017-148841

Foreign matter can adhere to beads. Until now, there has been no specific consideration of determining whether the adhesion of foreign matter to beads is an abnormality.

This disclosure can be realized in the following forms:

(1) According to one aspect of the present disclosure, there is provided an inspection system for inspecting a workpiece including a columnar portion having a welded portion, the inspection system including: an image acquisition unit that acquires an image of the appearance of a bead in the welded portion as an inspection image; a first inspection unit that determines whether or not a void exists on the bead in the inspection image; a second inspection unit that determines whether or not a foreign matter exists on the bead in the inspection image by detecting an edge included in the inspection image; and a determination unit that determines whether or not an abnormality exists in the workpiece when at least one of the void and the foreign matter is included in the inspection image.
In this configuration, the first inspection unit can determine the presence or absence of voids in the inspection image, and the second inspection unit can determine the presence or absence of foreign matter based on edges contained in the inspection image. The workpiece is determined to be abnormal if either a void or a foreign matter is contained in the learning image. Therefore, in addition to voids, the adhesion of foreign matter to the bead can also be determined to be abnormal.
(2) In the above aspect, the first inspection unit may determine the presence or absence of a void using a machine learning model that has been trained using a plurality of training images of the appearance of the bead, the machine learning model being for determining whether or not the void is included in an image input to the machine learning model. In this aspect, the first inspection unit can determine the presence or absence of a void using the machine learning model.
(3) In the above aspect, a first camera may be provided that captures an image of a portion of the bead's appearance, and a second camera that captures the image of the portion of the bead at an angle different from the angle at which the first camera captures the image of the portion. The machine learning model may have been trained using, as the training images, images of the bead's appearance captured by the first camera and images of the bead's appearance captured by the second camera. With this aspect, it is possible to reduce the shortage of training images compared to, for example, a case where only images captured by the first camera or only images captured by the second camera are used for machine learning.
(4) In the above embodiment, the image acquisition unit may acquire, as the inspection images, a first image of the appearance of the bead of the welded portion to be inspected taken by the first camera and a second image of the appearance of the bead of the welded portion to be inspected taken by the second camera, and the determination unit may determine that the workpiece has an abnormality if at least one of the first image and the second image contains at least one of the void and the foreign matter. According to this embodiment, the workpiece can be inspected more precisely than, for example, when determining the presence or absence of a void or foreign matter based on only one of the image taken by the first camera and the image taken by the second camera.
(5) In the above embodiment, the second inspection unit may determine the presence or absence of the foreign matter in the inspection image based on at least one of the angle of the edge, the coordinates of the edge, the spacing between the edges, and the strength of the edge in the inspection image.

In addition to the inspection system described above, the present disclosure can be realized in various forms, such as a workpiece inspection method.

FIG. 1 is an explanatory diagram showing a schematic configuration of an inspection system. FIG. 2 is an explanatory diagram showing how a workpiece is manufactured. FIG. 10 is a first explanatory diagram showing an example of a captured image. FIG. 10 is a second explanatory diagram showing an example of a captured image. FIG. 10 is a third explanatory diagram showing an example of a captured image. 10 is a flowchart of an inspection process.

A. First embodiment:
1 is an explanatory diagram showing a schematic configuration of an inspection system 100. The inspection system 100 inspects a workpiece W including a columnar portion CP having a welded portion WP. The inspection system 100 of this embodiment includes a photographing unit 110, a storage device 130, a control device 140, an inspection device 150, and an alarm unit 160. In this specification, the term "columnar" includes not only a solid columnar shape but also a hollow columnar shape.

In this embodiment, the workpiece W is made of a resin material and configured as a liner for a tank that stores hydrogen. The liner is a hollow container for storing hydrogen gas or the like. It has a generally cylindrical body P1 and a pair of dome portions P2 connected to both ends of the body P1. In this embodiment, the columnar portion CP is composed of the entire body P1 and a portion of the dome portion P2. Each dome portion P2 has a generally hemispherical shape that convex toward the outside of the workpiece W in the axial direction along the axis AX, which is the central axis of the columnar portion CP. An opening is formed at the apex of each dome portion P2, and a nozzle M is fitted into the opening. The hollow portion within the workpiece W is connected to the outside via the nozzle M. In this embodiment, the central axis of the body P1 and the central axis of each dome portion P2 are generally aligned with the axis AX. Therefore, the central axis of the workpiece W is aligned with the axis AX.

In this embodiment, the workpiece W is connected to a rotation device (not shown) and rotates around axis AX by driving the rotation device. The rotation device has, for example, a motor that generates a rotational driving force that is applied to the workpiece W, and is driven under the control of the control device 140.

Figure 2 is an explanatory diagram showing how the workpiece W is manufactured in this embodiment. As shown in Figure 2, the workpiece W is manufactured by joining the body portion P1 and each dome portion P2 to each other by infrared welding. More specifically, first, a cylindrical member P1a and two dome members P2a are prepared. The cylindrical member P1a is a member that will later constitute the body portion P1. The dome members P2a are a member that will later constitute each dome portion P2. Next, as shown in the upper part of Figure 2, both ends of the body portion P1 and the ends of each dome member P2a facing the cylindrical member P1a are heated by infrared lamps L. Next, as shown in the middle part of Figure 2, the cylindrical member P1a and each dome member P2a are brought into contact so that each end of the cylindrical member P1a and the end of each dome member P2a facing the cylindrical member P1a are in tight contact with each other. Then, as shown in the lower part of Figure 2, each dome member P2a is pressed against the cylindrical member P1a from the outside, thereby joining the cylindrical member P1a and each dome member P2a by welding.

The above welding forms a columnar portion CP having a welded portion WP. More specifically, a welded portion WP is formed between the body portion P1 and each dome portion P2, connecting the body portion P1 and each dome portion P2 and extending along the circumferential direction of the workpiece W. Each welded portion WP has a bead Bd. The bead Bd is a welding mark that extends along the circumferential direction. The bead Bd is raised higher than its surrounding areas, and has an inner portion Bp1 that protrudes toward the inside of the workpiece W, and an outer portion Bp2 that protrudes toward the outside.

The photographing unit 110 includes a photographing unit 111 and a camera controller 116. The photographing unit 111 photographs the welded portion WP to photograph the appearance of the bead Bd of the welded portion WP. More specifically, in this embodiment, the photographing unit 111 photographs the appearance of the inner portion Bp1 of the bead Bd. In other embodiments, the photographing unit 111 may photograph the outer portion Bp2 in addition to or instead of the inner portion Bp1.

The photographing unit 111 has a first camera 112, a second camera 113, and a shaft 114. The shaft 114 is axially shaped and is inserted into the workpiece W via a nozzle M. The first camera 112 and the second camera 113 are each fixed to the shaft 114 and are inserted into the workpiece W together with the shaft 114. The first camera 112 and the second camera 113 may be configured as an endoscopic camera such as a fiberscope or borescope, for example.

The first camera 112 and the second camera 113 capture images of a portion of the exterior of the bead Bd. The second camera 113 captures the portion of the bead Bd captured by the first camera 112 at an angle different from the angle at which the first camera 112 captures that portion.

Figure 1 shows the first camera 112 and the second camera 113 photographing a first portion Pt1 of the bead Bd. The first portion Pt1 is a circumferential portion of the bead Bd. As shown in Figure 1, in this embodiment, the first camera 112 and the second camera 113 are positioned in the axial direction to sandwich the bead Bd, which is the subject of the photograph, between them when photographing the bead Bd. More specifically, the first camera 112 and the second camera 113 are positioned approximately symmetrically in the axial direction across the bead Bd, which is the subject of the photograph. The first camera 112 is positioned to face the first direction d1 and photographs the inner portion Bp1 of the bead Bd from one side in the axial direction. The second camera 113 is positioned to face the second direction d2 and, together with the first camera 112, photographs the inner portion Bp1 of the bead Bd from the other side in the axial direction. The first direction d1 and the second direction d2 are opposite to each other in the axial direction but are substantially the same on a plane perpendicular to the axial direction. More specifically, the first direction d1 is the direction from the first camera 112 to the second camera 113 and from the center of the columnar portion CP to the outer periphery. The second direction d2 is the direction from the second camera 113 to the first camera 112 and from the center of the columnar portion CP to the outer periphery. Hereinafter, an image capturing the appearance of the bead Bd will also be referred to as a captured image Pi. Furthermore, the captured image Pi captured by the first camera 112 will also be referred to as a first captured image, and the captured image Pi captured by the second camera 113 will also be referred to as a second captured image.

In this embodiment, an illumination device 115 that illuminates the bead Bd, which is the subject of imaging, is fixed to the shaft portion 114. In this embodiment, the illumination device 115 is disposed between the first camera 112 and the second camera 113 in the axial direction, and illuminates the bead Bd by irradiating it with light in a direction perpendicular to the axial direction.

Note that Figure 1 shows the image capturing unit 111 capturing an image of one of the two beads Bd. Although not shown, in this embodiment, the image capturing unit 111 is configured to be able to capture an image of the other bead Bd as well. More specifically, in this embodiment, the insertion positions of the first camera 112 and the second camera 113 are configured to be changeable under the control of the control device 140, and when the object to be captured is changed, the insertion positions of the first camera 112 and the second camera 113 are changed to correspond to the position of the bead Bd to be captured.

The camera controller 116 is configured, for example, as a circuit or computer for controlling each camera provided in the imaging unit 111, and is connected to each camera. The camera controller 116, for example, sends an imaging trigger signal to each camera and receives images Pi captured by each camera. The captured images Pi received by the camera controller 116 are sent to the storage device 130 and stored in the storage device 130. In this embodiment, the storage device 130 is a NAS (Network Attached Storage) connected to the inspection device 150 via a network.

Figures 3 to 5 are explanatory diagrams each showing an example of a captured image Pi. Image Pi1 shown in Figure 3 is an example of a captured image Pi including a bead Bd that does not contain voids Vd or foreign matter Fs on its surface. Image Pi2 shown in Figure 4 is an example of a captured image Pi that includes voids Vd but does not contain foreign matter Fs. Image Pi3 shown in Figure 5 is an example of a captured image Pi that includes foreign matter Fs but does not contain voids Vd. In Figures 3 to 5, the bead Bd is hatched. More specifically, portions of the bead Bd where blown-out highlights HL occur are hatched with an upward sloping pattern. Portions of the bead Bd where voids Vd occur are hatched vertically. Portions of the bead Bd where blown-out highlights HL and voids Vd do not occur are hatched with a dotted pattern. In this embodiment, as shown in Figures 3 to 5, the captured image Pi is captured so that the bead Bd crosses the center of the captured image Pi in the vertical direction.

As shown in Figures 4 and 5, voids Vd may form on the bead Bd, or foreign matter Fs may adhere to the bead Bd. The voids Vd shown in Figure 4 are hole-shaped defects on the surface of the bead Bd, and are formed, for example, when moisture contained in the cylindrical member P1a or the dome member P2a evaporates during welding. The foreign matter Fs shown in Figure 5 is linear foreign matter, such as fibers or hair. In this embodiment, such foreign matter Fs adheres to the surface of the bead Bd due to, for example, static electricity. Note that blown-out highlights HL occur when the brightness in the captured image Pi increases due to, for example, reflection of light irradiated onto the bead Bd from the lighting device 115.

The control device 140 is configured as a computer including a CPU 141, a memory unit 142, and an input/output interface. The CPU 141 executes programs stored in the memory unit 142, causing the control device 140 to perform various functions, including the function of the inspection control unit 143. The inspection control unit 143 executes the inspection process described below. In other embodiments, the control device 140 may be configured, for example, by a PLC (Programmable Logic Controller). The functions of the control device 140 may also be realized by a circuit.

The inspection device 150 is configured as a computer equipped with a CPU 151, a memory unit 152, and an input/output interface. In this embodiment, the memory unit 152 stores a machine learning model 200 (described below) and an edge detection program 205.

The inspection device 150 has a first inspection unit 155, a second inspection unit 156, an image acquisition unit 157, and a determination unit 158. In this embodiment, the first inspection unit 155, the second inspection unit 156, the image acquisition unit 157, and the determination unit 158 are each functional units realized by the CPU 151 executing a program stored in the memory unit 152.

The image acquisition unit 157 acquires the captured image Pi as the inspection image Di. In this embodiment, the image acquisition unit 157 acquires the first captured image and the second captured image stored in the storage device 130 as the inspection image Di. More specifically, the image acquisition unit 157 acquires the first image and the second image. The first image refers to the first captured image captured by the first camera 112 of the appearance of the bead Bd of the welded portion WP, which is the object of inspection. The second image refers to the second captured image captured by the second camera 113 of the appearance of the bead Bd of the welded portion WP, which is also the object of inspection.

The first inspection unit 155 determines the presence or absence of voids Vd on the bead Bd based on the inspection image Di acquired by the image acquisition unit 157. In this embodiment, the first inspection unit 155 determines the presence or absence of voids Vd using a machine learning model 200 and the inspection image Di. The machine learning model 200 is a learning model that has been trained using multiple training images of the appearance of the bead Bd and is used to determine whether voids Vd are included in an input image to the machine learning model 200. In this embodiment, the training images include a first captured image and a second captured image that were previously captured. Furthermore, in this embodiment, the machine learning model 200 is a convolutional neural network that has been trained regarding whether voids Vd are included in the bead Bd through supervised learning using training data. This training data is composed of a data set in which each training image is associated with a label indicating the presence or absence of voids Vd.

The shooting angle of the bead Bd in the inspection image Di may be different from the shooting angle of the bead Bd in the learning image. For example, the shooting angle of the bead Bd in the first image may be different from the shooting angle of the bead Bd in the first and second captured images used as learning images. Similarly, the shooting angle of the bead Bd in the second image may be different from the shooting angle of the bead Bd in the first and second captured images used as learning images.

The second inspection unit 156 executes the edge detection program 205 to detect edges contained in the inspection image Di, thereby determining the presence or absence of foreign matter Fs on the bead Bd. The edge detection program 205 is a program for detecting edges contained in the inspection image Di using an edge detection algorithm.

In this embodiment, the second inspection unit 156 determines the presence or absence of a foreign substance Fs based on the edge angle, edge coordinates, edge spacing, and edge strength in the inspection image Di. More specifically, the second inspection unit 156 determines the presence or absence of a foreign substance Fs in the inspection image Di by detecting edges that differ from edges representing beads Bd and from edges representing blown-out highlights HL based on the above. Note that edge strength refers to the magnitude of change in pixel brightness. For example, in this embodiment, as shown in Figures 3 to 5, edges representing beads Bd and blown-out highlights HL are generally aligned along the horizontal axis of the captured image Pi, are detected in the vertical center of the captured image Pi, are relatively widely spaced apart, and have relatively low strength. Therefore, in this embodiment, the second inspection unit 156 detects edges that satisfy the following determination conditions as edges that differ from edges representing beads Bd and blown-out highlights HL: The criteria for determination are that the slope of the edge relative to the horizontal axis in the test image Di is greater than a predetermined slope threshold, the vertical coordinate of the edge is outside the predetermined central coordinate, the distance between adjacent edges is narrower than a predetermined distance threshold, and the edge strength is greater than a predetermined strength threshold.

The determination unit 158 determines that the workpiece W has an abnormality when the inspection image Di contains at least one of a void Vd and a foreign matter Fs. In other words, the determination unit 158 determines that the workpiece W has an abnormality both when the first inspection unit 155 determines that the workpiece W contains a void Vd and when the second inspection unit 156 determines that the workpiece W contains a foreign matter Fs.

Under the control of the control device 140, the notification unit 160 notifies the user of various information, for example, related to the inspection system 100 and the workpiece W. In this embodiment, the notification unit 160 is configured as a display device that displays visual information under the control of the control device 140. For example, in addition to or instead of such a display device, the notification unit 160 may have a speaker for outputting audio information.

Figure 6 is a flowchart of the inspection process. In this embodiment, the inspection process is executed, for example, when the first camera 112 and the second camera 113 are positioned within the workpiece W and a predetermined start operation is performed by the user on the control device 140. In this embodiment, steps S10 and S70 shown in Figure 6 are executed by the control device 140, and steps S20 to S60 are executed by the inspection device 150.

In step S10, the inspection control unit 143 uses the photographing unit 110 to photograph the appearance of the bead Bd of the welded portion WP to be inspected. In this embodiment, in step S10, the control device 140 alternately rotates the workpiece W around the axis AX by a predetermined unit angle, stops the rotation of the workpiece W, and photographs one of the beads Bd using the first camera 112 and the second camera 113, until the workpiece W has rotated once around the axis AX. The inspection control unit 143 then photographs the other bead Bd in a similar manner. In this manner, multiple first images and multiple second images of one of the beads Bd and multiple first images and multiple second images of the other bead Bd are obtained. For example, it is preferable that the unit angle be set to an angle small enough to photograph all of the circumferential portions of the bead Bd during one rotation of the workpiece W around the axis AX in step S10. Each captured image Pi is sent to the storage device 130.

In step S20, the image acquisition unit 157 acquires one of the captured images Pi captured in step S10 from the storage device 130 as the test image Di.

In step S30, the first inspection unit 155 determines whether or not there is a void Vd on the bead Bd in the inspection image Di acquired in step S20.

If it is determined in step S30 that the test image di contains voids Vd, then in step S32 the first inspection unit 155 records void information for the test image Di used in step S30. Void information is information regarding the presence or absence of voids Vd in the test image Di. In step S32, the first inspection unit 155 records information indicating that the test image Di contains voids Vd in the storage device 130 as void information for the test image Di used in step S30. Note that the void information may be recorded in, for example, the storage unit 142 or the storage unit 152. If it is determined in step S30 that the test image Di does not contain voids Vd, then in step S34 the first inspection unit 155 records information indicating that the test image Di does not contain voids Vd as void information for the test image Di, in the same manner as in step S32.

In step S40, the second inspection unit 156 determines whether or not there is a foreign object Fs on the bead Bd in the inspection image Di acquired in step S30.

If it is determined in step S40 that the inspection image di contains a foreign substance Fs, the second inspection unit 156 records foreign substance information for the inspection image Di used in step S40 in step S42. Foreign substance information is information regarding the presence or absence of a foreign substance Fs in the inspection image Di. In step S42, information indicating that a foreign substance Fs is contained is recorded as foreign substance information, similar to step S32. If it is determined in step S40 that a void Vd is not contained, the second inspection unit 156 records information indicating that a foreign substance Fs is not contained as foreign substance information for the inspection image Di in step S44, similar to step S42.

In step S50, the inspection device 150 determines whether all of the captured images Pi captured in step S10 have been acquired as test images Di. If there are any remaining captured images Pi that have not been acquired as test images Di, the inspection device 150 returns to step S20. When step S20 is executed again, any captured images Pi that were previously acquired as test images Di are excluded from the acquisition targets.

If it is determined in step S50 that all captured images Pi have been acquired as test images Di, then in step S60, the determination unit 158 determines whether or not the workpiece W contains an abnormality. More specifically, in this embodiment, the determination unit 158 determines that the workpiece W contains an abnormality if at least one of the first and second images acquired as test images Di contains a void Vd or a foreign substance Fs. Furthermore, the determination unit 158 determines that the workpiece W contains an abnormality even if any bead Bd contains a void Vd or a foreign substance Fs. In step S60 of this embodiment, the determination unit 158 determines whether or not the workpiece W contains an abnormality by referencing the recorded void information and foreign substance information. Note that in other embodiments, for example, test images Di determined to contain voids Vd or foreign substances Fs during the inspection process may be stored in a predetermined storage area, such as a directory or folder, within the storage device 130, storage unit 142, or storage unit 152. In this case, in step S60, the determination unit 158 determines whether or not the workpiece W has an abnormality by determining whether or not the inspection image Di is stored in this storage area.

If it is determined in step S60 that the workpiece W has an abnormality, then in step S70 the inspection control unit 143 uses the notification unit 160 to notify the abnormality. A workpiece W determined to have an abnormality may, for example, undergo further testing to determine whether or not there is an abnormality, may be sent to a process for repairing the welded portion WP, or may be discarded. A workpiece W determined to have no abnormality in step S60 is sent to the next process for manufacturing a tank, for example. Note that in this embodiment, each captured image Pi captured in the inspection process, i.e., each inspection image Di used in the inspection process, can be used, for example, to update the machine learning model 200 by re-learning, or to re-generate the machine learning model 200.

The inspection system 100 of this embodiment described above includes a first inspection unit 155 that determines the presence or absence of voids Vd on the bead Bd in the inspection image Di; a second inspection unit 156 that determines the presence or absence of foreign matter Fs on the bead Bd in the inspection image Di by detecting edges contained in the inspection image Di; and a determination unit 158 that determines that there is an abnormality in the workpiece W if the inspection image Di contains at least one of a void Vd and a foreign matter Fs. In this configuration, the first inspection unit 155 can determine the presence or absence of voids Vd based on the inspection image Di, and the second inspection unit 156 can determine the presence or absence of foreign matter Fs based on edges contained in the inspection image Di. The workpiece W is determined to be abnormal whether the inspection image Di contains either a void Vd or a foreign matter Fs. Therefore, in addition to voids Vd, adhesion of foreign matter Fs to the bead Bd can be determined to be an abnormality. Furthermore, compared to a configuration in which both the presence or absence of voids Vd and the presence or absence of foreign matter Fs are determined using only the first inspection unit 155 and the second inspection unit 156, for example, it is easier to specialize the second inspection unit 156 in determining the presence or absence of foreign matter Fs. This increases the likelihood that foreign matter Fs on the bead Bd can be accurately detected as an abnormality. In particular, if the foreign matter Fs is linear, as in this embodiment, foreign matter Fs can be more effectively detected based on the edge. Furthermore, because it is easier to specialize the first inspection unit 155 in determining the presence or absence of voids Vd, it also increases the likelihood that voids Vd can be accurately detected as an abnormality.

Furthermore, in this embodiment, the first inspection unit 155 determines the presence or absence of voids Vd using a machine learning model 200 that has been trained using learning images of the appearance of the bead Bd. Therefore, the second inspection unit 156 can determine the presence or absence of foreign matter Fs based on edges, and the first inspection unit 155 can determine the presence or absence of voids Vd using the machine learning model 200.

In addition, in this embodiment, the machine learning model 200 has already been trained using the first captured image taken by the first camera 112 and the second captured image taken by the second camera 113 as training images. Therefore, compared to, for example, using only one of the first captured image and the second captured image as a training image, it is possible to reduce the shortage of training images. Furthermore, since the second captured image can be captured by the second camera 113 at the same time as the first captured image is captured by the first camera 112, training images can be prepared more efficiently.

In addition, in this embodiment, the image acquisition unit 157 acquires the first image and the second image as inspection images Di, respectively, and the determination unit 158 determines that the workpiece W has an abnormality if at least one of the first image and the second image contains at least one of a void Vd and a foreign matter Fs. Therefore, the workpiece W can be inspected more precisely than, for example, when determining the presence or absence of a void Vd or a foreign matter Fs based on only one of the first image and the second image. Furthermore, not only can the presence or absence of a void Vd in the first captured image and the second captured image be determined using a single machine learning model 200, but the first captured image and the second captured image used for inspection can be used to update or regenerate the machine learning model 200.

B. Other Embodiments:
(B1) In the above embodiment, the first captured image and the second captured image are used as training images for training the machine learning model 200. However, the first captured image and the second captured image do not have to be used for training the machine learning model 200. For example, only the first captured image may be used for training the machine learning model 200, or only an image of the appearance of the bead Bd captured in a direction perpendicular to the axial direction may be used for training the machine learning model 200. In this case, for example, the photographing unit 111 does not have to have two cameras, and may have only one camera. Furthermore, the image acquiring unit 157 acquires the first captured image and the second captured image as the inspection image Di. Similarly, the first captured image and the second captured image do not have to be acquired as the inspection image Di.

(B2) In the above embodiment, the photographing unit 111 may have three or more cameras. For example, in addition to the first camera 112 and the second camera 113, a third camera and a fourth camera may be provided so that multiple beads Bd can be photographed at the same time. Also, multiple cameras may be arranged so that one bead Bd can be photographed with three or more cameras.

(B3) In the above embodiment, the machine learning model 200 has been trained by supervised learning. In contrast, the machine learning model 200 may be generated by a learning algorithm other than supervised learning. For example, the machine learning model 200 may be generated by unsupervised learning, or may be an autoencoder that has been trained using only images of beads Bd that contain voids Vd or only images of beads Bd that do not contain voids Vd as training images.

(B4) In the above embodiment, the first inspection unit 155 detects voids Vd using the machine learning model 200. In contrast, the first inspection unit 155 does not need to use the machine learning model 200 to detect voids Vd, and may detect voids Vd based on, for example, brightness differences or brightness variations within the inspection image Di.

(B5) In the above embodiment, the second inspection unit 156 determines the presence or absence of a foreign substance Fs based on the edge angle, coordinates, spacing, and strength. However, the second inspection unit 156 may determine the presence or absence of a foreign substance Fs based on, for example, any one, two, or three of these.

(B6) In the above embodiment, in the inspection process, the second inspection unit 156 detects the presence or absence of foreign matter Fs after the first inspection unit 155 detects the presence or absence of voids Vd. In contrast, the detection of the presence or absence of foreign matter Fs may be performed before the detection of the presence or absence of voids Vd, or may be performed in parallel with the detection of the presence or absence of voids Vd.

(B7) In the above embodiment, the inspection system 100 is composed of the photographing unit 110, the storage device 130, the control device 140, the inspection device 150, and the alarm unit 160. In contrast, in the inspection system 100, all of these may be configured as an integrated unit, or any two or more of them may be configured as an integrated unit. For example, the inspection device 150 may be incorporated into the control device 140, or the inspection device 150 may be incorporated into the control device 140.

This disclosure is not limited to the above-described embodiments, and can be realized in various configurations without departing from its spirit. For example, the technical features in the embodiments corresponding to the technical features in each aspect described in the Summary of the Invention section can be replaced or combined as appropriate to solve some or all of the above-described problems or achieve some or all of the above-described effects. Furthermore, if a technical feature is not described as essential in this specification, it can be deleted as appropriate.

100...Inspection system, 110...Photographing unit, 111...Photographing unit, 112...First camera, 113...Second camera, 114...Axis, 115...Illumination device, 116...Camera controller, 130...Storage device, 140...Control device, 141...CPU, 142...Storage unit, 143...Inspection control unit, 150...Inspection device, 151...CPU, 152...Storage unit, 155...First inspection unit, 156...Second inspection unit, 157...Image acquisition unit, 158...Determination unit, 160...Notification unit, 200...Machine learning model, 205...Edge detection program

Claims (5)

An inspection system for inspecting a workpiece including a column having a weld, comprising:
an image acquisition unit that acquires an image of the appearance of the bead in the welded portion as an inspection image;
a first inspection unit that determines whether or not a void exists on the bead in the inspection image;
a second inspection unit that detects edges included in the inspection image to determine whether or not there is a foreign substance on the bead in the inspection image;
an inspection system comprising a determination unit that determines that there is an abnormality in the workpiece when at least one of the void and the foreign matter is included in the inspection image. 10. The inspection system of claim 1,
The first inspection unit is an inspection system that determines the presence or absence of a void using a machine learning model that has been trained using multiple learning images of the appearance of the bead, and is used to determine whether the void is included in the image input to the machine learning model. 3. The inspection system of claim 2,
a first camera that captures an image of a part of the outer appearance of the bead;
a second camera that photographs the portion of the bead at an angle different from the angle at which the first camera photographs the portion,
An inspection system in which the machine learning model has been trained using an image of the bead's appearance photographed by the first camera and an image of the bead's appearance photographed by the second camera as the training images. 4. The inspection system of claim 3,
the image acquisition unit acquires, as the inspection images, a first image obtained by the first camera capturing an appearance of the bead of the welded portion to be inspected, and a second image obtained by the second camera capturing an appearance of the bead of the welded portion to be inspected,
An inspection system, wherein the judgment unit judges that the workpiece has an abnormality when at least one of the first image and the second image contains at least one of the void and the foreign matter. 5. The inspection system according to claim 1,
an inspection system in which the second inspection unit determines the presence or absence of the foreign matter in the inspection image based on at least one of an edge angle, an edge coordinate, an edge spacing, and an edge strength in the inspection image.