Inspection device and inspection method
The inspection device uses striped line light projection and AI models to analyze weld bead images, addressing the challenge of distinguishing quality by employing unsupervised and supervised learning for precise weld quality assessment.
Patent Information
- Application Number
- JP2023083143
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-19
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies face difficulties in distinguishing between good and poor-quality weld beads from two-dimensional external shape images, particularly in identifying irregularities such as undercuts, due to limitations in two-dimensional image analysis.
An inspection device employing striped line light projection and AI models, including unsupervised and supervised learning, to analyze weld bead images for quality assessment, using a striped line light projection unit and a determination unit that classifies weld bead quality based on machine-learned features.
The device effectively distinguishes between good and poor-quality weld beads by analyzing two-dimensional images, enhancing the accuracy of weld quality assessment through machine-learned AI models.
Smart Images

Figure 2024166797000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to an inspection apparatus and an inspection method. [Background technology]
[0002] Patent Document 1 discloses an appearance inspection device for a welded portion that judges the welding quality from a two-dimensional appearance shape image and a judgment model, and corrects the welding conditions to improve the quality. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] International Publication No. 2020 / 129618 Summary of the Invention [Problem to be solved by the invention]
[0004] In the technology disclosed in Patent Document 1, it is difficult to judge the unevenness of the weld shape of an arc bead (also called a weld bead) from a two-dimensional appearance shape image. Therefore, in the technology disclosed in Patent Document 1, it is difficult to distinguish between a normal arc bead of good quality and an arc bead of poor quality such as an undercut from a two-dimensional appearance shape image. In other words, the technology disclosed in Patent Document 1 has a problem in that it is difficult to distinguish between an object of good quality and an object of poor quality from a two-dimensional appearance shape image.
[0005] In view of such problems, the present disclosure has an object to provide an inspection device and an inspection method capable of distinguishing between objects of good quality and objects of poor quality from two-dimensional appearance shape images. [Means for solving the problem]
[0006] The inspection device disclosed herein comprises a striped line light projection unit that projects striped line light onto an object to be inspected; an imaging unit that captures an image of the object to be inspected onto which the striped line light is projected from a direction inclined at a predetermined angle relative to the projection axis of the striped line light; and a judgment unit that inputs the image of the object to be inspected onto which the striped line light is projected into an AI model that has been machine-learned using an image of a good-quality training object onto which the striped line light is projected, and judges whether the object to be inspected is of good quality or poor quality.
[0007] The inspection device of the present disclosure, with the above-mentioned configuration, can distinguish between good-quality objects and poor-quality objects from two-dimensional appearance shape images. Also, with the above-mentioned configuration, the inspection device of the present disclosure can distinguish between good-quality objects and poor-quality objects using an AI model machine-learned by unsupervised learning that learns only good-quality objects.
[0008] The judgment unit of the inspection device disclosed herein inputs an image of the object to be inspected onto which the striped line light is projected into the AI model, calculates the degree of anomaly of the object to be inspected relative to the learning object classified as good quality, and judges whether the object to be inspected is of good quality or poor quality based on the degree of anomaly.
[0009] With the above-described configuration, the inspection device disclosed herein can distinguish between good-quality objects and poor-quality objects in more detail using an AI model trained by unsupervised learning that learns only good-quality objects.
[0010] In addition, in the case where another AI model is used, the judgment unit of the inspection device disclosed herein inputs an image of the object to be inspected onto which the striped line light is projected into an AI model that has been machine-learned using an image of a good-quality training object onto which the striped line light is projected and an image of a poor-quality training object onto which the striped line light is projected, and judges whether the object to be inspected is of good quality or poor quality.
[0011] With the above-described configuration, the inspection device disclosed herein can distinguish between good-quality objects and poor-quality objects using other AI models machine-learned through supervised learning to learn about good-quality objects and poor-quality objects.
[0012] In the inspection method disclosed herein, an inspection device projects stripe line light onto an object, captures an image of the object onto which the stripe line light is projected from a direction inclined at a predetermined angle relative to the projection axis of the stripe line light, inputs the image of the object to be inspected onto which the stripe line light is projected into an AI model that has been machine-learned using an image of a good-quality training object onto which the stripe line light is projected, and determines whether the object to be inspected is of good quality or poor quality.
[0013] The inspection method of the present disclosure can distinguish between good quality objects and poor quality objects from two-dimensional appearance shape images by using the above-mentioned configuration. Also, the inspection method of the present disclosure can distinguish between good quality objects and poor quality objects by using an AI model trained by unsupervised learning that learns only good quality objects by using the above-mentioned configuration. Effect of the Invention
[0014] The present disclosure makes it possible to provide an inspection device and an inspection method that can distinguish between objects to be inspected that are of good quality and objects to be inspected that are of poor quality from a two-dimensional appearance shape image. [Brief description of the drawings]
[0015] [Figure 1] 1 is a diagram illustrating an example of a configuration of an inspection apparatus according to a first embodiment. [Diagram 2] 5 is a flowchart showing an example of the operation of the inspection apparatus according to the first embodiment. [Diagram 3] FIG. 11 is a diagram showing an example of an image captured by an imaging section of the inspection device according to the first embodiment; [Figure 4] 13 is a diagram showing an example of the configuration of a striped line light projection unit of a second modified example of the inspection device according to the first embodiment. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0016] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In each drawing, the same or corresponding elements are denoted by the same reference numerals, and duplicated descriptions will be omitted as necessary for clarity of explanation.
[0017] (First embodiment) First, the configuration of an inspection device 1 according to the first embodiment will be described with reference to FIG. 1 is a diagram showing an example of the configuration of an inspection device 1 according to the first embodiment. As shown in FIG. 1, the inspection device 1 includes a striped line light projection unit 10, an imaging unit 20, and a determination unit 30.
[0018] The stripe line light projection unit 10 projects stripe line light consisting of a plurality of laser line lights in a stripe shape at a predetermined pitch (interval) onto an arc bead (indicated by X in FIG. 1) to be inspected or studied. The stripe line light projected onto the arc bead is indicated by L in FIG. 1. The pitch of the stripe line light is preferably about 0.1 mm to 0.5 mm. This is because there are many types of defects in the arc bead, and the smallest surface defect to be detected is a pit (i.e., a hole on the bead surface). Since it is sufficient for the pit to be detected at 0.5 mm or more in practical use, the pitch of the stripe line light needs to be 0.5 mm or less. Moreover, since it is not necessary to detect the pit at a size smaller than that, the pitch of the stripe line light needs to be about 0.1 mm or more. The pitch of the stripe line light is not limited to about 0.1 mm to 0.5 mm, and may be another length. Moreover, the range in the arc bead where the stripe line light is projected is, for example, 30 mm×30 mm. The striped line light may be a red laser light or a green laser light.
[0019] Specifically, the stripe line light projection unit 10 includes a laser light generation unit 11, a DOE (diffractive optical element) 12, and an Fθ lens (condensing lens) 13. The laser light generation unit 11 generates a laser light. The DOE 12 spatially branches the laser light generated by the laser light generation unit 11 by utilizing the diffraction phenomenon of light. The Fθ lens 13 condenses the laser light branched by the DOE 12 and projects it onto the arc bead. Furthermore, the stripe line light projection unit 10 projects the stripe line light onto the arc bead, for example, from a direction perpendicular to the projection surface of the stripe line light of the arc bead.
[0020] The photographing unit 20 is, for example, a camera. The photographing unit 20 photographs a two-dimensional image of an arc bead to be inspected or studied, onto which the striped line light is projected by the striped line light projection unit 10. In the example shown in Fig. 1, the photographing unit 20 photographs an image of a bright background in a state where illumination is present.
[0021] In the image captured by the imaging unit 20, it is possible to determine whether the arc bead is uneven based on the state of the light beam of the stripe line light projected onto the arc bead. Specifically, in the image, if the light beam of the stripe line light projected onto the arc bead is curved toward a direction perpendicular to the stripe line, it can be determined that the arc bead has unevenness.
[0022] The photographing unit 20 photographs an image of the arc bead on which the stripe line light is projected from a direction tilted at a predetermined angle with respect to the projection axis of the stripe line light projected by the stripe line light projection unit 10. By tilting the photographing direction of the photographing unit 20 at a predetermined angle from the projection axis of the stripe line light projection unit 10, the state of the light beam of the stripe line light projected on the arc bead, that is, the curvature of the light beam, can be confirmed in the photographed image. The predetermined angle is preferably 24 degrees to 60 degrees. By tilting the photographing unit 20, the unevenness of the arc bead can be judged in the image, but if the photographing unit 20 is tilted too much, the photographing unit 20 may not be able to focus.
[0023] When the arc bead is long (for example, when the arc bead is 30 mm or longer), the inspection device 1 may scan the arc bead by having a robot (not shown) hold the striped line light projection unit 10 and the photographing unit 20. At this time, the photographing unit 20 may photograph the image of the arc bead onto which the striped line light is projected multiple times for a single arc bead.
[0024] The determination unit 30 is a terminal such as a PC (Personal Computer), etc. The determination unit 30 communicates with the imaging unit 20 wirelessly or via wires.
[0025] The determination unit 30 acquires an image of the learning target arc bead onto which the striped line light is projected from the photographing unit 20. The determination unit 30 uses the image of one or more learning target arc beads for machine learning of an AI model described below. The image of one or more learning target arc beads includes an image of a learning target arc bead of good quality. The image of one or more learning target arc beads may also include an image of a learning target arc bead of poor quality.
[0026] The determination unit 30 uses both the image of the arc bead of the learning object of good quality and the image of the arc bead of the learning object of poor quality to train the AI model by machine learning. The AI model is a model trained using algorithms such as artificial neurons and deep learning in artificial intelligence technology, and is, for example, a neural network model such as a convolutional neural network (CNN) or a recurrent neural network (RNN). Specifically, the determination unit 30 extracts feature amounts from the image of the arc bead of the learning object of good quality and the feature amount of the arc bead of the learning object of poor quality, respectively. For example, the determination unit 30 extracts the state of the light beam of the striped line light projected onto the arc bead of good quality from the image of the arc bead of the learning object of good quality as the feature amount. In addition, the determination unit 30 extracts the state of the light beam of the striped line light projected onto the arc bead of poor quality from the image of the arc bead of the learning object of poor quality as the feature amount. Then, the determination unit 30 trains the AI model by machine learning based on the extracted feature amount of the arc bead of the learning object of good quality and the feature amount of the learning object of poor quality. The determination unit 30 creates an AI model through machine learning.
[0027] Then, the determination unit 30 acquires an image of the arc bead to be inspected onto which the striped line light is projected from the photographing unit 20. The determination unit 30 inputs the image of the arc bead to be inspected onto which the striped line light is projected into an AI model, and classifies the arc bead to be inspected as being of good quality or bad quality using the AI model. The determination unit 30 determines whether the arc bead to be inspected is of good quality or bad quality based on the classification.
[0028] The determination unit 30 also machine-trains the AI model using only images of the learning object arc bead classified as good quality. Specifically, the determination unit 30 extracts features from the images of the learning object arc bead of good quality. For example, the determination unit 30 extracts the state of the light beam of the striped line light projected onto the good quality arc bead from the images of the learning object arc bead of good quality as features. Then, the determination unit 30 machine-trains the AI model based on the extracted features of the learning object arc bead of good quality. The determination unit 30 creates the AI model through machine learning.
[0029] Then, the determination unit 30 acquires an image of the arc bead of the inspection target on which the stripe line light is projected from the photographing unit 20. The determination unit 30 inputs the image of the arc bead of the inspection target into the AI model and calculates the degree of abnormality of the arc bead of the inspection target with respect to the arc bead of the learning target of good quality. The degree of abnormality is a numerical representation of how much the arc bead of the inspection target differs from the arc bead of the learning target of good quality. For example, the determination unit 30 calculates the degree of abnormality as how much the state of the light beam of the stripe line light projected onto the arc bead of the inspection target differs from the state of the light beam of the stripe line light projected onto the arc bead of the learning target of good quality. Then, the determination unit 30 determines whether the arc bead of the inspection target is of good quality or of poor quality based on the calculated degree of abnormality. For example, the determination unit 30 determines that the arc bead of the inspection target is of poor quality when the calculated degree of abnormality is equal to or greater than a predetermined threshold value.
[0030] In addition, the judgment unit 30 may not only use images captured by the photographing unit 20 for machine learning of the AI model, but may also use images captured by the photographing unit 20 and then stored in the judgment unit 30 or an external device (not shown) for machine learning.
[0031] Next, the operation of the inspection device 1 according to the first embodiment will be described with reference to Fig. 2. Fig. 2 is a flowchart showing an example of the operation of the inspection device 1 according to the first embodiment.
[0032] First, as shown in Fig. 2, the striped line light projection unit 10 of the inspection device 1 projects striped line light onto the arc bead to be learned (step S101). There may be one or more arc beads to be learned, which are classified in advance into arc beads of good quality or arc beads of poor quality. Next, the photographing unit 20 photographs an image of the arc bead to be learned onto which the striped line light is projected (step S102).
[0033] The inspection device 1 may repeat steps S101 to S102 until the number of captured images of the learning target arc bead reaches a required number.
[0034] Next, the determination unit 30 acquires an image of the learning target arc bead onto which the captured striped line light is projected (step S103). The determination unit 30 uses the image of one or more learning target arc beads for machine learning of an AI model described below. The image of one or more learning target arc beads includes an image of a learning target arc bead of good quality. The image of one or more learning target arc beads may also include an image of a learning target arc bead of poor quality.
[0035] After step S103, the decision unit 30 executes a process of pattern 1 (processing of steps S104 to S108 described later) or a process of pattern 2 (processing of steps S109 to S113 described later).
[0036] In the processing of pattern 1, the determination unit 30 performs machine learning on the AI model using both the image of the arc bead of the learning object of good quality and the image of the arc bead of the learning object of poor quality (step S104). The determination unit 30 creates the AI model by machine learning (step S105).
[0037] Next, the striped line light projection unit 10 projects the striped line light onto the arc bead to be inspected. Next, the photographing unit 20 photographs an image of the arc bead to be inspected onto which the striped line light is projected.
[0038] Next, the determination unit 30 obtains an image of the arc bead to be inspected onto which the stripe line light is projected from the photographing unit 20 (step S106). The determination unit 30 inputs the image of the arc bead to be inspected onto which the stripe line light is projected into the AI model, and classifies the arc bead to be inspected as being of good quality or bad quality using the AI model (step S107). The determination unit 30 determines whether the arc bead to be inspected is of good quality or bad quality based on the classification result (step S108).
[0039] In the processing of pattern 2, the determination unit 30 performs machine learning to generate an AI model using only images of the learning target arc beads classified as good quality (step S109). The determination unit 30 creates an AI model by machine learning (step S110).
[0040] Next, the striped line light projection unit 10 projects the striped line light onto the arc bead to be inspected. Next, the photographing unit 20 photographs an image of the arc bead to be inspected onto which the striped line light is projected.
[0041] Next, the determination unit 30 acquires an image of the arc bead to be inspected onto which the striped line light is projected from the photographing unit 20 (step S111). Next, the determination unit 30 inputs the image of the arc bead to be inspected into the AI model, and calculates the degree of abnormality of the arc bead to be inspected relative to the arc bead to be learned that is of good quality (step S112). Next, the determination unit 30 determines whether the arc bead to be inspected is of good quality or of poor quality based on the calculated degree of abnormality (step S113).
[0042] As described above, in the inspection device 1 according to the first embodiment, the stripe line light projection unit 10 projects the stripe line light onto the arc bead. The photographing unit 20 photographs an image of the arc bead onto which the stripe line light is projected from a direction inclined at a predetermined angle with respect to the projection axis of the stripe line light. The judging unit 30 inputs the image of the arc bead to be inspected onto which the stripe line light is projected into an AI model machine-learned using an image of a good-quality arc bead to be learned onto which the stripe line light is projected, and judges whether the arc bead to be inspected is of good quality or of poor quality. By doing so, the inspection device 1 can judge the unevenness of the welding shape of the arc bead from the two-dimensional appearance shape image, and can distinguish between good-quality arc beads and poor-quality arc beads. In addition, the inspection device 1 can distinguish between good-quality arc beads and poor-quality arc beads using an AI model machine-learned by unsupervised learning that learns only good-quality arc beads.
[0043] Furthermore, the AI model of the inspection device 1 may be a model machine-learned using an image of a good-quality arc bead to be learned on which striped line light is projected, and an image of a bad-quality arc bead to be learned on which striped line light is projected. In this way, the inspection device 1 can distinguish between good-quality arc beads and bad-quality arc beads using the AI model machine-learned by supervised learning that learns good-quality arc beads and bad-quality arc beads.
[0044] The present invention is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit and scope of the present invention.
[0045] The inspection device 1 according to the first embodiment judges whether the arc bead to be inspected is of good quality or bad quality. However, the object to be inspected according to the first embodiment is an arc bead, but is not limited to this. It is also possible to inspect other objects that may have unevenness on their surface during the manufacturing process, etc., which cannot be judged only by a two-dimensional image. For example, the object to be inspected according to the present disclosure may be a sealer-coated surface or a pressed surface after press processing, which may have unevenness on the surface. This is because unevenness can occur on the sealer-coated surface due to unevenness in the coating, etc. Also, unevenness can occur on the pressed surface after press processing due to cracks, wrinkles, and warping, etc.
[0046] (Modification 1 of the first embodiment) Fig. 3 is a diagram showing an example of an image captured by the imaging unit 20 of the first modified example of the inspection device 1 according to the first embodiment. The imaging unit 20 of the inspection device 1 according to the first embodiment captures an image of a bright background in a state where illumination is present, as shown in Fig. 1. On the other hand, in the first modified example of the inspection device 1 according to the first embodiment, the imaging unit 20 captures an image of a dark background as shown in Fig. 3 in a state where illumination is absent.
[0047] (Modification 2 of the first embodiment) FIG. 4 is a diagram showing an example of the configuration of the stripe line light projection unit 10 of the inspection device 1 according to the first embodiment. As shown in FIG. 1, the stripe line light projection unit 10 of the inspection device 1 according to the first embodiment includes a laser light generation unit 11, a DOE (diffractive optical element) 12, and an Fθ lens 13. The configuration of the stripe line light projection unit 10 is not limited to these configurations. For example, the stripe line light projection unit 10 according to the second modification of the inspection device 1 of the first embodiment may include a laser light generation unit 41, a cylindrical lens 42, and a slit 43. The stripe line light projection unit 10 according to the second modification may project stripe line light consisting of a plurality of laser line lights in a stripe shape with a predetermined pitch onto the arc bead.
[0048] Each component of the inspection device 1 in the above-described embodiment is configured by hardware or software, or both, and may be configured by one piece of hardware or software, or may be configured by multiple pieces of hardware or software. Each device and each function (processing) may be realized by a computer having a processor such as a CPU (Central Processing Unit) and a memory that is a storage device. For example, a program for performing the method (inspection method) in the embodiment may be stored in the memory, and each function may be realized by executing the program stored in the memory by the processor.
[0049] These programs include instructions (or software codes) that, when loaded into a computer, cause the computer to perform one or more functions described in the embodiments. The programs may be stored in a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disc (DVD), Blu-ray® disk or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device. The programs may be transmitted on a transitory computer-readable medium or a communication medium. By way of example and not limitation, a transitory computer-readable medium or a communication medium includes electrical, optical, acoustic, or other forms of propagated signals. [Explanation of symbols]
[0050] 1. Inspection equipment 10 Stripe line light projection unit 11 Laser light generating unit (laser light generating element) 12 DOE (diffractive optical element) 13 Fθ lens (condenser lens) 20. Photography Department 30 Judgment section 41 Laser light generating unit 42 Cylindrical Lens 43 Slit
Claims
1. a stripe line light projection unit that projects stripe line light onto an object to be inspected; an imaging unit that captures an image of an object to be inspected onto which the stripe line light is projected from a direction inclined at a predetermined angle with respect to a projection axis of the stripe line light; a determination unit that inputs an image of an object to be inspected onto which the striped line light is projected into an AI model that has been machine-learned using an image of a learning object of good quality onto which the striped line light is projected, and determines whether the object to be inspected is of good quality or of poor quality. Inspection equipment.
2. The determination unit is The image of the object to be inspected onto which the striped line light is projected is input to an AI model that has been machine-learned using an image of a learning object of good quality onto which the striped line light is projected and an image of a learning object of poor quality onto which the striped line light is projected, and it is determined whether the object to be inspected is of good quality or poor quality.
2. The inspection device according to claim 1.
3. The determination unit is An image of the object to be inspected onto which the striped line light is projected is input to the AI model, the degree of anomaly of the object to be inspected relative to the object to be trained which is of good quality is calculated, and whether the object to be inspected is of good quality or of poor quality is judged based on the degree of anomaly.
2. The inspection device according to claim 1.
4. The inspection device, A stripe line of light is projected onto the object to be inspected, taking an image of the object to be inspected onto which the stripe line light is projected from a direction inclined at a predetermined angle with respect to the projection axis of the stripe line light; The image of the object to be inspected onto which the stripe line light is projected is input to an AI model that has been machine-learned using an image of a good-quality learning object onto which the stripe line light is projected, and it is determined whether the object to be inspected is of good quality or of poor quality. Testing method.
Citation Information
Patent Citations
Method for selecting an optimized evaluation feature subset for an inspection of free-form surfaces and method for inspecting a free-form surface
EP2177898A1
Linear light projecting device
JP1987063804A
Method and apparatus for measuring profile of three-dimensional curved surface
JP1995260443A
Surface inspection apparatus and surface inspection method
JP2002148195A
Welding method, and apparatus for detecting stepped portion
JP2008260043A
