Surface defect detection method, surface defect detection device, steel manufacturing method, steel quality management method, steel manufacturing system, surface defect judgment model generation method and surface defect judgment device

Abstract

The surface defect detection method of the present invention is a surface defect detection method for optically detecting surface defects of steel, which includes: an irradiation step, using more than two identifiable light sources to irradiate illumination light from different directions to the same inspection object part; and a detection step, based on the degree of overlap of bright parts extracted from more than two images formed by reflected light of each illumination light, detecting surface defects at the inspection object part.

Description

Technical Field

[0001] The present invention relates to a surface defect detection method for optically detecting surface defects of steel, a surface defect detection device, a steel manufacturing method, a steel quality management method, steel manufacturing equipment, a surface defect judgment model generation method and a surface defect judgment model. Background Art

[0002] In recent years, in the manufacturing process of steel products, from the perspective of improving the yield rate by preventing a large number of defects, it is required to detect the surface defects of steel in hot or cold state. The steel described here means steel products such as steel pipes (seamless steel pipes, welded steel pipes, etc.), steel plates (hot-rolled steel plates, cold-rolled steel plates, thick plates, etc.) and steel sections, as well as semi-finished products (slabs, etc.) generated in the process of manufacturing these steel products. When attempting to detect the surface defects of steel by optical surface inspection, it is sometimes difficult to distinguish between harmful uneven surface defects and harmless flat scale textures (caused by uneven cooling, etc.) if the image of the steel is simply taken.

[0003] Therefore, as a method for accurately distinguishing concavo-convex surface defects and flat oxide scale textures, the following method has been proposed: irradiating distinguishable light from different directions, receiving reflected light of each light to obtain an image, and detecting surface defects by obtaining a difference image between the obtained images (see Patent Document 1). In addition, the following method has been proposed: based on the obtained difference image, the light and dark pattern generated by the shape of the surface defect is recognized, and the concave shape or convex shape of the surface defect is determined, thereby further improving the detection accuracy of the surface defect (see Patent Document 2).

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Publication No. 2015-125089

[0007] Patent Document 2: Japanese Patent Application Publication No. 2015-210150 Summary of the invention

[0008] Problems to be solved by the invention

[0009] According to the methods described in Patent Documents 1 and 2, it is possible to distinguish between uneven surface defects and flat oxide scale textures with high precision through differential images. However, the inventors of the present invention have conducted research and found that: in the methods described in Patent Documents 1 and 2, the differential signal of the bottom steel portion with a high reflectivity and almost no uneven surface is similar to the signal of the concave surface defect, so it is sometimes difficult to distinguish. It should be noted that the bottom steel portion is mainly caused by "the peeling of black oxide scale due to friction, collision, etc. at a temperature lower than the temperature at which the black oxide scale is generated on the surface", which is equivalent to cold scratches. Generally, uneven surface defects are caused by pressing and transferring protrusions that are burned and stuck to the roller while the steel is soft in a hot state. Therefore, the uneven surface defects have a clear uneven shape, and are oxidized in the same way as the healthy part during the cooling process and have the same degree of reflectivity. As a result, if Fig. 22 As shown in (a), a clear light and dark pattern is produced by unilateral projection. In contrast, the bottom steel part is often produced by the friction between the steel material and the roller and other structures in the cold state. The black oxide scale peels off and the bottom steel is exposed, and the reflectivity becomes higher than that of the healthy part covered by the black oxide scale. As a result, Fig. 22 As shown in (b), the bottom steel portion has almost no depth but high reflectivity, so it is imaged as a high-brightness signal regardless of the irradiation direction.

[0010] In addition, when the bottom steel portion is slightly concavo-convex, when the differential image is obtained by applying the techniques described in Patent Documents 1 and 2, a light and dark pattern similar to that of a concavo-convex surface defect may be produced. The bottom steel portion is a surface defect that may be harmful or harmless depending on the purpose of the inspection, and is the cause of non-detection or over-detection of surface defects. In particular, in inspections where depth is important, such as wall thickness assurance, there are many cases where concavo-convex surface defects are considered harmful and the bottom steel portion is considered harmless, and it is desired to distinguish concavo-convex surface defects and bottom steel portions with high accuracy.

[0011] The present invention has been made in view of the above problems, and its purpose is to provide a surface defect detection method, a surface defect detection device, a method for generating a surface defect determination model, and a surface defect determination model that can distinguish bottom steel parts and surface defects with high accuracy. In addition, another purpose of the present invention is to provide a steel manufacturing method, a steel quality management method, and a steel manufacturing device that can distinguish bottom steel parts and surface defects with high accuracy and improve the manufacturing yield of steel materials.

[0012] Means for solving problems

[0013] The surface defect detection method of the present invention is a surface defect detection method for optically detecting surface defects of steel, which includes: an irradiation step, using more than two identifiable light sources to irradiate illumination light from different directions to the same inspection object part; and a detection step, based on the degree of overlap of bright parts extracted from more than two images formed by reflected light of each illumination light, detecting surface defects at the inspection object part.

[0014] The degree of overlap of the bright portions is preferably a ratio of the overlapped portion of the bright portions at the inspection target site to the surface defect candidate portion.

[0015] The detection step preferably includes the following steps: calculating a surface defect candidate portion at the inspection object portion based on bright portions extracted from two or more images formed by reflected light of each illumination light; and detecting surface defects at the inspection object portion based on the ratio of the overlapping portion of the bright portions to the surface defect candidate portion.

[0016] The detection step preferably includes the following steps: using a surface defect judgment model implemented with machine learning in a manner that outputs a judgment value of whether a surface defect exists in the inspection object part corresponding to the two or more images when the degree of overlap of the bright parts extracted from the two or more images is input to detect the surface defect at the inspection object part.

[0017] The surface defect detection device of the present invention is a surface defect detection device that optically detects surface defects of steel materials, and comprises: an irradiation unit that uses two or more distinguishable light sources to irradiate illumination light from different directions to the same inspection object part; and a detection unit that detects surface defects at the inspection object part based on the degree of overlap of bright parts extracted from two or more images formed by reflected light of each illumination light.

[0018] The method for manufacturing a steel material of the present invention comprises the following steps: detecting surface defects of a steel material using the surface defect detection method of the present invention and manufacturing the steel material.

[0019] The quality management method of a steel material of the present invention includes the step of managing the quality of the steel material by classifying the steel material based on the presence or absence of surface defects using the surface defect detection method of the present invention.

[0020] The manufacturing facility of the steel product of the present invention comprises: a manufacturing facility for manufacturing the steel product; and the surface defect detection device of the present invention for inspecting the steel product manufactured by the manufacturing facility.

[0021] The method for generating a surface defect judgment model of the present invention uses the degree of overlap of bright parts extracted from two or more images formed by reflected light of illumination light irradiated from different directions to the same inspection object part using two or more distinguishable light sources and the judgment result of whether there is a surface defect in the inspection object part as training data, and generates a learned model as a surface defect judgment model through machine learning, which uses the degree of overlap of bright parts extracted from the two or more images as an input value and the judgment value of whether there is a surface defect in the inspection object part corresponding to the two or more images as an output value.

[0022] The surface defect determination model of the present invention is generated by the surface defect determination model generation method of the present invention.

[0023] Effects of the Invention

[0024] According to the surface defect detection method, surface defect detection device, surface defect judgment model generation method and surface defect judgment model of the present invention, bottom steel parts and surface defects can be distinguished with high accuracy. In addition, according to the steel manufacturing method, steel quality management method and steel manufacturing equipment of the present invention, bottom steel parts and surface defects can be distinguished with high accuracy to improve the manufacturing yield of steel. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram showing the structure of a surface defect inspection device as one embodiment of the present invention.

[0026] Figure 2 This is a diagram showing an example of a difference image of a bottom steel portion.

[0027] Figure 3 This is a diagram showing the result of performing AND processing on a bright portion image of a concave surface defect.

[0028] Figure 4 This is a diagram showing the result of performing AND processing on the bright part image of the bottom steel part.

[0029] Figure 5 This is a flowchart showing the flow of the detection step in the surface defect detection method as the first embodiment of the present invention.

[0030] Figure 6 This is a flowchart showing the flow of the detection step in the surface defect detection method as the second embodiment of the present invention.

[0031] Figure 7 This is a flowchart showing the flow of the detection step in the surface defect detection method as the third embodiment of the present invention.

[0032] Figure 8 It is used to illustrate Figure 7 A diagram showing an example of labeling processing.

[0033] Fig. 9 This is a flowchart showing the flow of the detection step in the surface defect detection method as the fourth embodiment of the present invention.

[0034] Fig.10 The figure shows an example of two two-dimensional images of uneven surface defects, scale texture, and harmless texture, and a difference image thereof.

[0035] Fig.11 The diagram shows shadows when light is irradiated from one side in the case where the surface shape of the inspection target site is a concave shape or a convex shape.

[0036] Fig.12 This is a diagram showing an example of a difference image of a concave surface defect.

[0037] Fig.13 This is a flowchart showing the flow of a method for calculating the positional relationship between bright and dark parts using dilation processing.

[0038] Fig.14 It is a diagram for explaining the expansion and contraction process.

[0039] Fig.15 FIG. 1 is a diagram showing an example of a difference image and a one-dimensional profile of a light-dark pattern.

[0040] Fig.16 It is a diagram showing an example of a two-dimensional image and a one-dimensional profile of a filter.

[0041] Fig.17 It shows that it has been implemented and used Fig.16 FIG. 1 is a diagram showing an example of a difference image and a one-dimensional profile after filtering by the filter shown.

[0042] Fig.18 This is a diagram showing the result of calculating the histogram of the double bright portion occupancy rate with respect to the bottom steel portion and the surface defects of the unevenness.

[0043] Fig.19 This is a flowchart showing the flow of the detection step in the surface defect detection method as the fifth embodiment of the present invention.

[0044] Fig. 20 It is a schematic diagram showing a modified example of the arrangement position of the light source.

[0045] Fig.21 It is shown by Fig. 20 Schematic diagram of the light and dark pattern obtained by the configuration position of the light source shown.

[0046] Fig. 22This is a schematic diagram for explaining the surface defects of the unevenness and the characteristics of the bottom steel part. DETAILED DESCRIPTION

[0047] Hereinafter, the structure and operation of a surface defect inspection device as one embodiment of the present invention will be described with reference to the drawings.

[0048] [Structure of surface defect detection device]

[0049] Figure 1 Schematic diagram showing the structure of a surface defect detection device as one embodiment of the present invention. Figure 1 As shown, a surface defect detection device 1 as one embodiment of the present invention is a device for detecting surface defects of a cylindrical steel pipe P transported in the direction of the arrow in the figure, and has light sources 2a, 2b, a function generator 3, area sensors 4a, 4b, an image processing device 5 and a monitor 6 as main components.

[0050] The light sources 2a and 2b are irradiation units. The light sources 2a and 2b irradiate the same inspection target site on the surface of the steel pipe P with identifiable illumination light L according to the trigger signal from the function generator 3. The light sources 2a and 2b are arranged symmetrically with respect to the inspection target site. That is, the light sources 2a and 2b are arranged so as to be offset at the same incident angle with respect to the normal vector of the surface of the steel pipe P and so that the irradiation direction vector of the illumination light L and the normal vector of the surface of the steel pipe P are on the same plane.

[0051] Here, the purpose of making the incident angles of the illumination light L the same is to make the optical conditions as equal as possible when distinguishing light sources with different incident directions. In experience, if the difference between the incident angles is within a range of 20°, it appears as the same incident angle. On the other hand, regarding the incident angle itself, in experience, it is hoped that the incident angle of one light source is in the range of 25° to 82.5°. Moreover, when it is desired to obtain light quantity, it is preferably selected from the range of 25° to 55°. When the light quantity is sufficient and the S / N ratio is desired to be improved, it is preferably selected from the range of 60° to 82.5°.

[0052] It should be noted that, in this specification, the incident angle represents the angle between the incident direction of the light from the light sources 2a, 2b and the normal to the surface of the inspection object. In addition, the normal to the surface of the inspection object is set to 0°. In addition, in this embodiment, the number of light sources is set to 2, but as long as it can be distinguished, the number of light sources can also be set to 3 or more. The distinguishable light source described here means a light source for which the amount of reflected light obtained from the inspection object can be obtained for each light source.

[0053] The area sensors 4a and 4b are shooting units. The area sensors 4a and 4b shoot a two-dimensional image formed by the reflected light of the illumination light L emitted from the light sources 2a and 2b according to the trigger signal from the function generator 3. The area sensors 4a and 4b input the data of the captured two-dimensional image to the image processing device 5. The area sensors 4a and 4b are set as much as possible on the normal vector of the inspection object part while ensuring their respective shooting fields of view. It should be noted that in this embodiment, an area sensor is used, but a line sensor can also be used. In this case, although the captured image becomes a one-dimensional image, the surface defect detection method described later can be directly applied.

[0054] The image processing device 5 is a detection unit. The image processing device 5 is a device that uses two two-dimensional images input from the area sensors 4a and 4b to perform a determination process of a harmless bottom steel portion at the inspection object portion. The determination process of the bottom steel portion is the core technology of the present invention, so it will be described in detail later. The image processing device 5 can further perform defect determination based on machine learning. Moreover, as needed, it is also possible to detect surface defects at the inspection object portion by performing a differential process described later. The image processing device 5 outputs the two-dimensional images input from the area sensors 4a and 4b and information related to the detection results of surface defects to the monitor 6.

[0055] The surface defect detection device 1 having such a structure performs the determination processing of the bottom steel part at the inspection object part by executing the surface defect detection method shown below. Moreover, the surface defect detection device 1 distinguishes the oxide scale texture, harmless texture, uneven surface defects and bottom steel part at the inspection object part as needed. It should be noted that the oxide scale texture and harmless texture mean a surface film and a surface property part having optical properties different from the bottom steel part having a thickness of about several μm to tens of μm, which is a part that becomes a noise factor in the surface defect detection method.

[0056] 〔Surface defect detection method〕

[0057] In the surface defect detection methods described in Patent Documents 1 and 2, if the same difference processing is performed on the two-dimensional image of the bottom steel portion such as cold scratches, sometimes the same difference as in the example above is generated. Figure 2 The light and dark pattern of the uneven surface defect shown in (a) is also as Figure 2 (b) to (e) show light and dark patterns. This is because the bottom steel part has high reflectivity and uneven brightness, so the light and dark parts are generated randomly. In particular, in the field of steel materials such as thick plates and steel pipes where the depth of surface defects is important, the bottom steel part is generated much more than the surface defects, so it is difficult to distinguish based on feature quantities, etc.

[0058] However, there are cases where the bottom steel part must be judged as harmful or harmless depending on the inspection requirements. Therefore, if the bottom steel part having the same light and dark pattern as the surface defect cannot be distinguished, it will be over-detected, and the detection performance of the surface defect will be reduced. In addition, in the case where the bottom steel part must be judged as harmful, there is also a case where the bottom steel part must be distinguished from the surface defect. In addition, the bottom steel part has various shapes, and there are many cases where it is necessary to judge it as harmless.

[0059] In order to stably detect such a bottom steel portion, a method using the characteristic that the illumination light is brighter than the stable portion regardless of the direction from which it is irradiated can be considered. The stable portion described here refers to a black oxide scale portion that is neither a surface defect nor a bottom steel portion. That is, the following method can be considered: the bottom steel portion is detected by detecting the overlapping portion of the bright portion in the area of ​​the surface defect candidate portion extracted by threshold processing, labeling, etc. from the two two-dimensional images input from the area sensors 4a and 4b.

[0060] Figure 3 and Figure 4 These are examples of the results of detecting the bright parts of the two two-dimensional images input from the area sensors 4a and 4b by binarization and performing AND processing on the detected bright part images, respectively, with respect to the concave surface defects and the bottom steel part in the surface defect candidate part. Figure 3 As shown in (c) to (e), in the case of concave surface defects, two two-dimensional images ( Figure 3 The positions of the bright parts of (c) and (d) are different, so even if AND processing is performed, no signal can be detected ( Figure 3 (e)). In contrast, Figure 4 As shown in (c) to (e), in the bottom steel part, two two-dimensional images ( Figure 4 Since the positions of the bright parts of (c) and (d) coincide with each other, if AND processing is performed, the signal ( Figure 4 (e)).

[0061] It should be noted that the method for detecting the surface defect candidate portion at the inspection object portion can use a known method. In particular, the method using a differential image described in Patent Documents 1 and 2 can offset the oxide scale texture and the harmless texture by differential processing, and uses a unique light and dark pattern generated by a concave-convex surface defect, so there is an advantage of being able to detect the surface defect candidate portion, especially the concave-convex surface defect candidate portion, with high accuracy, and is suitable.

[0062] It should be noted that, regarding the timing of the detection of the surface defect candidate portion, as long as the detection step described later is earlier than the step of calculating the occupancy rate of the two bright portions (S5, S15, S25, S37, S48 described later), it can be performed at any time. For example, it is possible to pre-detect independently of the detection step described later (refer to the first and second embodiments). In addition, for example, it is also possible to detect in the detection step described later based on the copies of the bright portion binarized image a and the bright portion binarized image b used in the detection step (refer to the third embodiment). Moreover, for example, it is also possible to detect concave and convex surface defects in parallel with the detection step described later using a differential image obtained based on the copies of the original image a, the original image b, the corrected image a, and the corrected image b used in the detection step, and set the detected concave and convex surface defects as the surface defect candidate portion.

[0063] Therefore, by "extracting two bright part images by performing AND processing on the bright part images of two two-dimensional images, and evaluating the surface defect candidate part using the two bright part images", it is possible to determine whether the surface defect candidate part is a surface defect or a bottom steel part. The bottom steel part can be detected by extracting the bright parts of the two two-dimensional images using threshold processing and performing AND processing. In addition, even if it is not a bottom steel part but a surface defect, there is a case where the bright parts of the two two-dimensional images irradiated from other directions overlap slightly occasionally. In contrast, in the case of the bottom steel part, the reflectivity is high in the two two-dimensional images irradiated from other directions, and the bright parts overlap in most areas. Therefore, the degree of overlap of the bright parts can also be calculated as a feature value, and the bottom steel part and the surface defect can be distinguished using this index. In this method, it is possible to distinguish with higher accuracy than the case of judging by the presence or absence of overlapping parts, which is preferred.

[0064] Hereinafter, surface defect detection methods according to first to fifth embodiments of the present invention, which were conceived based on the above-described concept, will be described.

[0065] [First embodiment]

[0066] First, refer to Figure 5 , a surface defect detection method as a first embodiment of the present invention is described.

[0067] The surface defect detection method as the first embodiment of the present invention includes an irradiation step, a photographing step, and a detection step. In the irradiation step, the light sources 2a and 2b irradiate the same inspection target part on the surface of the steel pipe P with identifiable illumination light L according to the trigger signal from the function generator 3. In the photographing step, the area sensors 4a and 4b photograph the two-dimensional image formed by the reflected light of the illumination light L irradiated from the light sources 2a and 2b according to the trigger signal from the function generator 3. In the detection step, the image processing device 5 uses the two two-dimensional images input from the area sensors 4a and 4b to determine the harmless bottom steel part.

[0068] Figure 5 FIG. 1 is a flow chart showing the flow of the detection steps in the surface defect detection method as the first embodiment of the present invention. Figure 5 As shown, in the detection steps of this embodiment, first, the image processing device 5 generates a corrected image a and a corrected image b by performing image correction processing on two two-dimensional images (original image a and original image b) input from the area sensors 4a and 4b (steps S1a and S1b). The image correction processing described here includes processing for cutting out the inspection object area and shadow correction for correcting the overall brightness unevenness caused by the optical system.

[0069] Next, the image processing device 5 performs a bright part binarization process on the corrected image a and the corrected image b, respectively, by setting the value of pixels whose brightness is greater than the threshold to 1 and setting the value of pixels whose brightness is less than the threshold to 0 to detect the bright part. Through this process, the image processing device 5 generates a bright part binarized image a and a bright part binarized image b (steps S2a and S2b). Figure 5 In the example of bright part binary image a, an example is shown on the left side of the label "bright part binary image a", and an example is shown on the right side of the label "bright part binary image b". Next, the image processing device 5 performs AND processing on the bright part binary image a and the bright part binary image b. Through this processing, the image processing device 5 extracts pixels with a value of 1 from both the bright part binary image a and the bright part binary image b to generate two bright part images (step S3). Figure 5 In FIG. 1 , an example of a two-brightness image is shown on the right side of the label “two-brightness image”.

[0070] Next, the image processing device 5 uses the surface defect candidate part to perform masking calculation processing on the two bright part images. Through this processing, the image processing device 5 generates a cut-out two bright part image that cuts out the area that is the object of determining whether it is the bottom steel part (that is, the area that is the surface defect candidate part and the bright part) (step S4). Next, the image processing device 5 calculates the proportion of the cut-out two bright part images in the entire surface defect candidate part as the two bright part occupancy rate (step S5). Moreover, the image processing device 5 determines whether the surface defect candidate part is a bottom steel part or a surface defect by using threshold processing of the calculated two bright part occupancy rate. Here, in the case where it is determined that it is not a bottom steel part, the image processing device 5 determines it as a surface defect (step S6). It should be noted that the present embodiment is an example of calculating the degree of overlap of the two bright part images relative to the surface defect candidate part, but as long as it can calculate the degree to which the two bright part images overlap in the entire surface defect candidate part, other processing methods can also be used. For example, the surface defect candidate portion may be measured by other techniques such as eddy current testing and ultrasonic testing, and the ratio of the two bright portion images therein may be calculated.

[0071] [Second embodiment]

[0072] Next, refer to Figure 6 , a surface defect detection method as a second embodiment of the present invention is described.

[0073] The surface defect detection method as a second embodiment of the present invention includes an irradiation step, a photographing step, and a detection step. In the irradiation step, the light sources 2a and 2b irradiate the same inspection target portion on the surface of the steel pipe P with identifiable illumination light L according to the trigger signal from the function generator 3. In the photographing step, the area sensors 4a and 4b photograph the two-dimensional image formed by the reflected light of the illumination light L irradiated from the light sources 2a and 2b according to the trigger signal from the function generator 3. In the detection step, the image processing device 5 uses the two two-dimensional images input from the area sensors 4a and 4b to determine the harmless bottom steel portion.

[0074] Figure 6 FIG. 1 is a flow chart showing the flow of the detection steps in the surface defect detection method as the second embodiment of the present invention. Figure 6As shown, in the detection steps of this embodiment, first, the image processing device 5 performs mask calculation processing on the two two-dimensional images (original image a, original image b) input from the area sensors 4a and 4b respectively using the surface defect candidate unit, thereby generating a cut-out original image a and a cut-out original image b (steps S11a and S11b) in which the area to be determined as the object of the bottom steel portion is cut out. Next, the image processing device 5 generates a cut-out corrected image a and a cut-out corrected image b by performing image correction processing on the cut-out original image a and the cut-out original image b (steps S12a and S12b). The image correction processing described here includes processing such as shadow correction to correct the overall brightness unevenness caused by the optical system.

[0075] Next, the image processing device 5 performs a bright part binarization process on the cut-out correction image a and the cut-out correction image b, respectively, by setting the value of the pixel whose brightness is above the threshold value to 1 and setting the value of the pixel whose brightness is less than the threshold value to 0 to detect the bright part. Through this process, the image processing device 5 generates a cut-out bright part binarization image a and a cut-out bright part binarization image b (steps S13a, S13b). Next, the image processing device 5 performs an AND process on the cut-out bright part binarization image a and the cut-out bright part binarization image b. Through this process, the image processing device 5 extracts pixels whose value is 1 from both the cut-out bright part binarization image a and the cut-out bright part binarization image b, and generates a cut-out two bright part image (step S14). Next, the image processing device 5 calculates the proportion of the cut-out two bright part images in the entire surface defect candidate part as the two bright part occupancy rate (step S15). Moreover, the image processing device 5 determines whether the surface defect candidate part is a bottom steel part or a surface defect by threshold processing using the calculated two bright part occupancy rate. Here, when it is determined that it is not a bottom steel portion, the image processing device 5 determines that it is a surface defect (step S16).

[0076] [Third Embodiment]

[0077] Next, refer to Figure 7 , a surface defect detection method as a third embodiment of the present invention is described.

[0078] The surface defect detection method as the third embodiment of the present invention includes an irradiation step, a photographing step, and a detection step. In the irradiation step, the light sources 2a and 2b irradiate the same inspection target part on the surface of the steel pipe P with identifiable illumination light L according to the trigger signal from the function generator 3. In the photographing step, the area sensors 4a and 4b photograph the two-dimensional image formed by the reflected light of the illumination light L irradiated from the light sources 2a and 2b according to the trigger signal from the function generator 3. In the detection step, the image processing device 5 uses the two two-dimensional images input from the area sensors 4a and 4b to determine the harmless bottom steel part.

[0079] Figure 7 FIG. 4 is a flow chart showing the flow of the detection steps in the surface defect detection method as the third embodiment of the present invention. Figure 7 As shown, in the detection step of this embodiment, first, the image processing device 5 performs image correction processing such as calibration, shadow correction and noise removal on the two two-dimensional images (original image a and original image b) input from the area sensors 4a and 4b using the camera parameters derived in advance, thereby generating a corrected image a and a corrected image b (steps S21a and S21b). Through this image correction processing, the subsequent bright part extraction processing can be performed with high accuracy.

[0080] Next, the image processing device 5 performs a bright part binarization process on the corrected image a and the corrected image b, respectively, to detect the bright part by setting the value of the pixel whose brightness is greater than the threshold value to 1 and setting the value of the pixel whose brightness is less than the threshold value to 0. Through this process, the image processing device 5 generates a bright part binarization image a and a bright part binarization image b (steps S22a, S22b). Next, the image processing device 5 performs an AND process on the bright part binarization image a and the bright part binarization image b. Through this process, the image processing device 5 extracts pixels with a value of 1 from both the bright part binarization image a and the bright part binarization image b to generate two bright part images (step S23). The set of pixels with a value of 1 in the two bright part images becomes a bright part overlapping portion where the bright parts of the bright part binarization image a and the bright part binarization image b overlap. In this way, it is possible to extract the overlapping portion of the high reflectivity portion in the bright part binarization image a and the bright part binarization image b. Figure 8 (a) shows an example of a bright part binary image a. Figure 8 (b) shows an example of a bright part binary image b. Figure 8 (c) shows an example of two bright area images.

[0081] Next, the image processing device 5 marks the pixels of the bright portion whose value is 1 in the generated two bright portion images, and sets each spot (a collection of marked pixels) as a bottom steel portion candidate portion. Figure 8 The white part (i.e., the bright part) in the two bright part images of (c) is the bottom steel part candidate part, which is equivalent to the overlapping part of the bright part of both the bright part binary image a and the bright part binary image b. In addition, the image processing device 5 performs OR processing (or processing) on ​​the bright part binary image a and the bright part binary image b generated in steps S22a and S22b. Through this processing, the image processing device 5 generates a bright part OR image. Figure 8(d) shows an example of a bright part OR image. Then, the image processing device 5 marks the above-mentioned bottom steel part candidate part on the bright part OR image obtained by the processing. Then, the image processing device 5 extracts the blocks of the bright part including the marked bottom steel part candidate part as spots (blobs). Here, each spot is set as a surface defect candidate part (step S24). Figure 8 (e) shows an example of a surface defect candidate. The white part (i.e., the bright part) is the surface defect candidate. Figure 8 In the two bright image (c), the bright part that is considered not to be a surface defect is deleted by the AND processing in step S23. Figure 8 As is apparent from (d), in the OR process, all bright portions remain regardless of whether they are defects. Therefore, by marking the bottom steel portion candidate portion, the surface defect candidate portion can be reliably detected.

[0082] Next, the image processing device 5 calculates the value obtained by dividing the area of ​​the bottom steel portion candidate portion by the area of ​​the surface defect candidate portion as the two-bright portion occupancy rate for each surface defect candidate portion (step S25). And finally, the image processing device 5 determines whether the surface defect candidate portion is a bottom steel portion based on the calculated two-bright portion occupancy rate, and detects the bottom steel portion. Here, when it is determined that it is not a bottom steel portion, the image processing device 5 determines it as a surface defect (step S26).

[0083] It should be noted that multiple indicators can be considered to indicate the degree of overlap of the bottom steel portion candidate portion relative to the surface defect candidate portion. For example, the ratio of the area of ​​each bottom steel portion candidate portion to the area of ​​the surface defect candidate portion of the two images before AND processing can be considered. In this case, as the area of ​​the surface defect candidate portion, the area of ​​the two surface defect candidate portions after OR processing, the sum of the areas of each surface defect candidate portion, the maximum value, etc. can be exemplified. The higher the occupancy rate of the two bright portions, the higher the possibility of the bottom steel portion. Therefore, a fixed threshold can also be simply used, and if it is above the threshold, it is determined to be a bottom steel portion.

[0084] In addition, the two-bright-part occupancy rate can be used as a feature value, and a well-known machine learning method such as a binary decision tree method can be used to detect the bottom steel part. That is, the two-bright-part occupancy rate and the judgment result of whether there is a surface defect in the inspection object part are used as training data, and a learned model is generated by machine learning, which uses the two-bright-part occupancy rate calculated based on more than two images as an input value and the judgment value of whether there is a surface defect in the inspection object part corresponding to more than two images as an output value. The generated learned model can also be set as a surface defect judgment model, the two-bright-part occupancy rate is input to the surface defect judgment model, and the output value of the surface defect judgment model is used to judge whether it is a bottom steel part, and the bottom steel part is detected. In this way, the bottom steel part and surface defects that are difficult to distinguish by previous methods can be distinguished with high accuracy.

[0085] In addition, it is also possible to determine whether the surface defect candidate portion is the bottom steel portion based on the positional relationship between the bottom steel portion and the surface defect candidate portion calculated in the first to third embodiments. Specifically, if the surface defect candidate portion and the bottom steel portion calculated in the first to third embodiments are approximately at the same position, the possibility that the surface defect candidate portion is the bottom steel portion is high. In addition, as long as the degree of overlap of the two bright portions in the surface defect candidate portion can be calculated in the same manner as in the first to third embodiments, other processing methods may also be used.

[0086] [Fourth Embodiment]

[0087] When the bottom steel portion is slightly uneven, the bottom steel portion and the uneven surface defect may not be distinguished with high accuracy because the same light and dark pattern as the uneven surface defect is generated. Therefore, in this embodiment, the uneven surface defect that can be calculated by the method described in Patent Documents 1 and 2 is used as the surface defect candidate portion, so as to distinguish the bottom steel portion and the uneven surface defect with high accuracy. Fig. 9 , a surface defect detection method as a fourth embodiment of the present invention is described.

[0088] The surface defect detection method as the fourth embodiment of the present invention includes an irradiation step, a photographing step and a detection step. In the irradiation step, the light sources 2a and 2b irradiate the same inspection target part on the surface of the steel pipe P with identifiable illumination light L according to the trigger signal from the function generator 3. In the photographing step, the area sensors 4a and 4b photograph the two-dimensional image formed by the reflected light of the illumination light L irradiated from the light sources 2a and 2b according to the trigger signal from the function generator 3. In the detection step, the image processing device 5 uses the two two-dimensional images input from the area sensors 4a and 4b to distinguish the oxide scale, harmless texture, uneven surface defects and bottom steel part.

[0089] Fig. 9FIG. 4 is a flow chart showing the flow of the detection steps in the surface defect detection method as the fourth embodiment of the present invention. Fig. 9 As shown, in the detection steps of this embodiment, first, the image processing device 5 performs image correction processing such as calibration, shadow correction and noise removal on the two two-dimensional images (original image a, original image b) input from the area sensors 4a and 4b using the pre-derived camera parameters, thereby generating a corrected image a and a corrected image b (steps S31a and S31b). Next, the image processing device 5 generates a differential image by performing a differential process between the corrected image a and the corrected image b (step S32). Next, the image processing device 5 calculates the surface defect candidate portion of the concave-convex property at the inspection object part based on the generated differential image, and outputs information related to the scale texture and harmless texture removed by the differential process (step S33).

[0090] On the other hand, the image processing device 5 performs a bright part binarization process for detecting the bright part by setting the value of the pixel whose brightness is greater than the threshold value to 1 and setting the value of the pixel whose brightness is less than the threshold value to 0 on the corrected image a and the corrected image b generated in steps S31a and S31b, respectively. Through this process, the image processing device 5 generates a bright part binarization image a and a bright part binarization image b (steps S34a and S34b). Next, the image processing device 5 performs an AND process on the bright part binarization image a and the bright part binarization image b. Through this process, the image processing device 5 extracts pixels whose value is 1 from both the bright part binarization image a and the bright part binarization image b, and generates two bright part images (step S35). Next, the image processing device 5 performs a mask calculation process on the two bright part images using the surface defect candidate portion of the concave-convexity calculated in the process of step S33. Through this process, the image processing device 5 generates two cut-out bright part images in which the area to be determined as the object of the bottom steel portion is cut out (step S36). Next, the image processing device 5 calculates the proportion of the two bright part images cut out in the entire concave-convex surface defect candidate part as the two bright part occupancy rate (step S37). Moreover, the image processing device 5 determines whether it is a bottom steel part or a concave-convex surface defect by threshold processing using the calculated two bright part occupancy rate. Here, in the case where it is determined that it is not a bottom steel part, the image processing device 5 determines it as a surface defect (step S38). For the bottom steel part, the entire concave-convex surface defect candidate part becomes a bright signal due to irradiation from two directions, but for the concave-convex surface defect, the shadow appears differently due to irradiation from two directions, so the bright part and the dark part are generated at different positions in the entire concave-convex surface defect candidate part. Therefore, the two bright part occupancy rate of the bottom steel part becomes higher, and the two bright part occupancy rate of the concave-convex surface defect becomes lower.

[0091] It should be noted that, as a method for detecting the surface defect candidate portion of the concavo-convexity at the inspection object portion using the differential image in the processing of step S33, a known method can be used. In particular, the method described in Patent Documents 1 and 2 can offset the oxide scale texture and harmless texture by differential processing, and use the unique light and dark pattern generated by the concavo-convex surface defect, so there is an advantage of being able to detect the surface defect candidate portion of the concavo-convexity with high precision, which is suitable. Below, as an example of the processing of step S33, an example using the technology described in Patent Documents 1 and 2 is described.

[0092] Now, when the luminance value of each pixel constituting the two-dimensional image Ia obtained when the illumination light L is irradiated from the light source 2a is set to Ia(x, y) (wherein the number of pixels is X×Y, the x coordinate is set to 1≤x≤X, and the y coordinate is set to 1≤y≤Y), and the luminance value of each pixel constituting the two-dimensional image Ib obtained when the illumination light L is irradiated from the light source 2b is set to Ib(x, y), the luminance value I_diff(x, y) of each pixel of the differential image I_diff obtained by the differential processing of step S32 is expressed by the mathematical formula (1) shown below.

[0093] I_diff(x,y)=Ia(x,y)-Ib(x,y)···(1)

[0094] Here, examples of two-dimensional images Ia and Ib of surface defects that are not uneven, healthy scale textures that are surface defects, and harmless textures, and their difference images I_diff are shown in FIG. Fig.10 (a), (b), and (c) are shown. Fig.10 As shown in (a), (b) and (c), in the healthy part, regardless of the presence or absence of oxide texture and harmless texture, the angle formed by the surface normal vector and the light source 2a is equal to the angle formed by the surface normal vector and the light source 2b, so the luminance value Ia(x, y) = luminance value Ib(x, y), that is, the luminance value I_diff(x, y) = 0.

[0095] However, in the concavo-convex surface defect portion, since the surface has a concavo-convex shape, there must be a portion where the angle formed by the surface normal vector and the light source 2a and the angle formed by the surface normal vector and the light source 2b are not equal, and the brightness value Ia(x, y) ≠ brightness value Ib(x, y), that is, the brightness value I_diff(x, y) ≠ 0. Therefore, by using the difference processing of the differentiator 11 to generate the difference image I_diff of the two two-dimensional images, it is possible to remove the image of a healthy oxide scale texture or harmless texture that is not a surface defect.

[0096] Next, the logic of the surface defect candidate unit that detects the unevenness based on the difference image I_diff will be described. Fig.11 (a) and (b) are diagrams showing shadows when the inspection target portion is irradiated with illumination light from one light source when the surface shape of the inspection target portion is concave and convex, respectively. Fig.11 As shown in (a), when the surface shape of the inspection object is concave, the front side of the light source becomes darker due to the decrease in the amount of irradiated light per unit area, and the back side of the light source becomes brighter due to the direction of regular reflection. Fig.11 As shown in (b), when the surface shape of the inspection target part is convex, the front side of the light source becomes brighter because it is close to the regular reflection direction, and the back side of the light source becomes the shadow of the convex shape and becomes darker.

[0097] That is, the light and dark pattern of the reflected light of the illumination light is different when the surface shape of the inspection object part is concave and when it is convex. Therefore, by identifying the light and dark pattern of the reflected light, the presence or absence of the concave-convex surface defect candidate part can be detected. Therefore, the method for detecting the concave-convex surface defect candidate part by identifying the light and dark pattern of the reflected light is described below. It should be noted that, below, it is set to detect the concave surface defect candidate part in the concave-convex surface defect candidate part, but the convex surface defect candidate part can also be detected by the same logic. In addition, the bright part described below means a spot with an area greater than a specified value obtained by connecting pixels with a brightness greater than a specified threshold in the differential image I_diff. In addition, the dark part described below refers to a spot with an area greater than a specified value obtained by connecting pixels with a brightness less than a specified threshold in the differential image I_diff. Spot means a collection of marked pixels.

[0098] It should be noted that the threshold value and the area value for determining the detection of the above-mentioned bright and dark parts are adjusted in consideration of the final detection performance. In other words, the larger the threshold value and the value are set, the more defects with stronger signals can be detected, but the over-detection signals caused by surface roughness and noise can be reduced. In other words, from the definition of harmful defects, the use purpose, and the discrimination performance of defect determination based on light and dark patterns, it is preferred to adjust in a manner that satisfies the target detection performance and the allowable over-detection degree.

[0099] In this embodiment, threshold processing is performed to extract adjacent bright and dark areas, thereby identifying light and dark patterns. Figure 1In the surface defect detection device 1 shown, since the light sources 2a and 2b are arranged symmetrically with respect to the normal vector of the inspection object, the light and dark pattern of the reflected light caused by the concave and convex shape of the surface is generated in the left and right directions. The light and dark are opposite to each other according to the order of differential processing. Therefore, here, the case where the right is bright and the left is dark is regarded as a concave shape, and the case where the right is dark and the left is bright is regarded as a convex shape. Therefore, the differential image I_diff of the concave surface defect becomes Fig.12 Then, if the bright and dark images are binarized by the luminance thresholds The and -The, respectively, the binarized images I_bright and I_dark of the bright and dark parts are expressed as shown in the following mathematical formula (2), respectively.

[0100]

[0101] Then, after the images of the bright and dark parts are binarized and connected and isolated points are removed as needed, the presence or absence of surface defects of unevenness is detected by calculating the positional relationship between adjacent bright and dark parts. It should be noted that there are various methods for calculating the positional relationship between adjacent bright and dark parts. The following describes three representative calculation methods, but even other calculation methods can be used as long as the positional relationship between the bright and dark parts can be calculated.

[0102] The first positional relationship calculation method is a method for calculating the positional relationship between adjacent bright and dark parts by performing expansion and contraction processing in a specific direction on adjacent bright and dark parts. Fig.13 As shown in . In this embodiment, the case of identifying a light and dark pattern in which the right is bright and the left is dark in order to detect a concave surface defect is described. The right is bright and the left is dark means that there must be a dark part on the left side of the light part, and there must be a light part on the right side of the dark part. Therefore, in this calculation method, first, the image processing device 5 performs an expansion process on the dark part in the right direction and an expansion process on the light part in the left direction (steps S331a, S331b). Here, if the images of the light part and the dark part to which the expansion process is performed are set to I_bright_extend and I_dark_extend respectively, and the length of the expansion is set to W, the expansion process is expressed as shown in the mathematical formula (3) shown below. Among them, the upper left of the two-dimensional image is set to the origin, the lower direction is set to the positive y-axis direction, and the right direction is set to the positive x-axis direction.

[0103] I_bright_extend(x1, y)=1 xW≤x1≤x (when I_bright(x, y)=1)

[0104] I_dark_extend(x1, y)=1x≤x1≤x+W (when I_dark(x, y)=1)

[0105] ···(3)

[0106] It should be noted that, in this embodiment, the bright part and the dark part are expanded by the same length W, but the expanded length W does not necessarily need to be the same. In extreme terms, the expansion process may be performed on only one of the bright part and the dark part. In addition, the expanded length W also depends on the size of the surface defect to be detected.

[0107] Next, the image processing device 5 performs AND processing on the dilated bright and dark images I_bright_extend and I_dark_extend as shown in the following mathematical formula (4), thereby extracting the overlapping part of the dilated bright and dark images I_bright_extend and I_dark_extend as the concave defect candidate image I_defect (steps S332a, S332b).

[0108] I_defect=I_bright_extend&I_dark_extend···(4)

[0109] Next, the image processing device 5 performs connection and isolated point removal processing as needed on each of the obtained concave defect candidate images I_defect, and then generates a concave defect candidate spot I_defect_blob by performing labeling processing (step S333). Then, the image processing device 5 extracts the feature quantity of each concave defect candidate spot I_defect_blob, and determines whether each concave defect candidate spot I_defect_blob is a concave surface defect based on the extraction result (steps S334a, S334b). It should be noted that in order to investigate the feature quantity of the concave defect candidate spot I_defect_blob, information on the bright part and the dark part is required, so the bright part and the dark part are restored from the concave defect candidate spot I_defect_blob.

[0110] Specifically, since there must be a bright part on the right side of the concave shape defect candidate part and a dark part on the left side, the image processing device 5 uses the centroid of the concave shape defect candidate spot I_defect_blob as the starting point to explore the dark part binary image I_dark to the left, and sets the first found spot as the dark part concave shape defect candidate spot I_dark_blob. Similarly, the image processing device 5 uses the centroid of the concave shape defect candidate spot I_defect_blob as the starting point to explore the bright part binary image I_bright to the right side, and sets the first found spot as the bright part concave shape defect candidate spot I_bright_blob. Then, the image processing device 5 extracts feature values ​​from the bright part concave shape defect candidate spot I_bright_blob and the dark part concave shape defect candidate spot I_dark_blob thus restored, and determines whether each concave shape defect candidate spot I_defect_blob is a concave surface defect based on the extracted feature values. The specific feature amount varies depending on the defect, so it is not described here but an example is given in the embodiment described later.

[0111] Based on the restoration of the bright concave defect candidate spots and the dark concave defect candidate spots, the concave-convex surface defect candidate parts used in the calculation of the two bright part occupancy rates described later can be calculated. Specifically, the areas obtained in the binary images obtained by calculating the bright and dark part spots and obtaining their respective binarized images and performing OR processing, or the areas filled between the bright and dark parts by further performing expansion and contraction processing, are set as the concave-convex surface defect candidate parts. It should be noted that in the present embodiment, in order to facilitate the description, the concave shape defect candidate parts are described, but regarding the convex shape defect candidate parts, such as Fig.11 (a) and (b) show that the light and dark patterns are reversed, so if Fig.13 Even if the positional relationship between the bright portion and the dark portion in the processing shown is reversed, the calculation can be performed by the same processing.

[0112] It should be noted that if Fig.14 As shown, the above expansion and contraction processing means obtaining "first, the original image ( Fig.14 (a)) The dark binarized image generated by Fig.14 (b)) and bright part binarization image ( Fig.14 (c)) and the image obtained by AND processing ( Fig.14 (d)) After dilation processing is performed in a certain shape (the pixels that form the outline of the object spot are defined and replaced with the shape), contraction processing is then performed in a certain shape (the shape that defines the pixels that form the outline of the object spot is deleted)" ( Fig.14(e)). As for the parameter of the expansion and contraction distance, any shape of the spot can be used as long as it is larger than the distance between the bright part and the dark part, but it is preferably set to the same shape in the expansion process and the contraction process. In addition, even if the expansion and contraction process is performed multiple times for each pixel, the same result will be obtained.

[0113] In the second positional relationship calculation method, the above-mentioned threshold processing is performed, and after the connection and isolated point removal processing is performed as needed, the bright part and the dark part are extracted and marked, and the positional relationship between the adjacent bright parts and the dark parts is identified, thereby detecting the concave surface defects. Specifically, first, the image processing device 5 independently identifies the bright part and the dark part by marking, and obtains the center of gravity information of the bright part and the dark part. Next, the image processing device 5 determines whether there is a center of gravity of the dark part within the specified range on the right side of each bright part based on the center of gravity information of the bright part and the dark part. In addition, in the case where there is a center of gravity of the dark part, the image processing device 5 recognizes the combination of the paired bright part and the dark part as a light and dark pattern, and determines whether it is a concave surface defect by analyzing the feature quantity of the light and dark pattern. It should be noted that the center of gravity information is used here to identify the light and dark pattern, but as long as the information that can grasp the position of the adjacent bright and dark parts (such as the upper end position, the lower end position, etc.) can be used, the information used in the identification of the light and dark pattern may not be the center of gravity information. It should be noted that, in the present embodiment, in order to facilitate the description, the concave defect candidate portion is described, but regarding the convex defect candidate portion, as shown in FIG. Fig.11 (a) and (b) show that the light and dark patterns are reversed, so if Fig.13 Even if the positional relationship between the bright portion and the dark portion in the processing shown is reversed, the calculation can be performed by the same processing.

[0114] In the third positional relationship calculation method, the above-mentioned threshold processing is not performed, and a concave surface defect is detected by recognizing a light and dark pattern using a filter. Figure 1 In the surface defect inspection device 1 shown, the light sources 2a and 2b are arranged bilaterally symmetrically with respect to the normal line of the inspection target portion, so that a light and dark pattern caused by the surface irregularities is generated in the horizontal direction. Fig.15 (a) and (b) show an example of a difference image and Fig.15 (a) is a graph of the one-dimensional profile of the light and dark pattern on the line segment L shown.

[0115] like Fig.15As shown in (a) and (b), in a concave surface defect, the right side is bright and the left side is dark, so the one-dimensional profile of the light-dark pattern becomes a characteristic one-dimensional profile with a mountain shape on the right side and a valley shape on the left side. Therefore, in this embodiment, a filter H is prepared in advance to make the right side a mountain shape and the left side a valley shape, and the filter H is applied to the difference image I_diff as shown in the following mathematical formula (5), thereby generating a two-dimensional image I_cont in which high-frequency noise is reduced and only the light-dark pattern is emphasized.

[0116] I_cont=H*I_diff···(5)

[0117] Fig.16 (a) and (b) are diagrams showing examples of a two-dimensional image of a filter H prepared in advance and its one-dimensional profile in the left and right directions, respectively. Fig.17 (a) and (b) respectively show the implementation of the Fig.16 (a) and (b) are diagrams showing the difference image of the filter H and its one-dimensional profile in the left and right directions. Fig.17 As shown in (a) and (b), it can be seen that a two-dimensional image is obtained in which high-frequency noise is reduced and only the light and dark patterns are emphasized. It should be noted that in this embodiment, in order to facilitate the description, the concave defect candidate part is described, but for the convex defect candidate part, as shown in Fig.11 As shown in (a) and (b), the light and dark patterns are reversed, and therefore, if the shape of the filter is reversed, it can be calculated by the same processing.

[0118] It should be noted that, as needed, a plurality of filters with different ranges in the width direction may be prepared so as to cope with a wide range of surface defect sizes. After the image processing device 5 performs connection and isolated point removal processing as needed on the two-dimensional image in which the light and dark pattern is emphasized, it extracts the defect candidate image Idefect by performing threshold processing. Then, the image processing device 5 detects concave surface defects by performing the same processing as the first position relationship calculation method on the extracted defect candidate image Idefect. It should be noted that the oxide scale texture and harmless texture that have the same visual appearance in the two images before the difference and do not form a light and dark pattern after the difference can be extracted by excluding the surface defect candidate part that forms a light and dark pattern after the difference from the surface defect candidate part with concave and convex properties extracted by binarizing and marking the image before the difference.

[0119] [Example]

[0120] exist Fig.18 1 shows an example of a histogram of double bright portion occupancy rates calculated using the surface defect detection method and the surface defect detection device 1 according to the fourth embodiment in an actual machine test. Fig.18The horizontal axis shows the percentage of the bright part, and the vertical axis shows the frequency of the bottom steel part or the uneven surface defects as a percentage. Fig.18 The black diamond symbol in the figure shows how much the occupancy rate is distributed in each of the two bright parts, with respect to 100% of the part confirmed as the "bottom steel part" by visual inspection. Fig.18 The white squares in the figure represent the percentage of each of the two bright parts, with respect to 100% of the part that is visually confirmed as a "concave-convex surface defect". It should be noted that in this embodiment, the bright part is set to a part with a brightness of 1.5 times or more of the sound part. In addition, the concave-convex surface defect candidate part is detected using light and dark pattern detection and expansion and contraction processing.

[0121] like Fig.18 As shown, it can be confirmed that the concave-convex surface defects and the bottom steel part are very well separated with the two-bright part occupancy rate of about 30% as the threshold value. In the case where the bottom steel part becomes an over-detected inspection, the two-bright part occupancy rate of, for example, 30% or more is judged as the bottom steel part relative to the concave-convex surface defect candidate part. By this method, the concave-convex surface defects will not be completely excluded, and more than 95% of the bottom steel part can be removed. It should be noted that the method of directly distinguishing by the two-bright part occupancy rate as a threshold value is also effective, and even if the two-bright part occupancy rate is used as one of the feature quantities and machine learning is used for distinction, the same result is obtained.

[0122] [Fifth Embodiment]

[0123] Finally, refer to Fig.19 , a surface defect detection method as a fifth embodiment of the present invention is described.

[0124] The surface defect detection method as the fifth embodiment of the present invention includes an irradiation step, a photographing step and a detection step. In the irradiation step, the light sources 2a and 2b irradiate the same inspection target part on the surface of the steel pipe P with identifiable illumination light L according to the trigger signal from the function generator 3. In the photographing step, the area sensors 4a and 4b photograph the two-dimensional image formed by the reflected light of the illumination light L irradiated from the light sources 2a and 2b according to the trigger signal from the function generator 3. In the detection step, the image processing device 5 uses the two two-dimensional images input from the area sensors 4a and 4b to distinguish the oxide scale, harmless texture, uneven surface defects and bottom steel part.

[0125] Fig.19 FIG. 5 is a flow chart showing the flow of the detection steps in the surface defect detection method as the fifth embodiment of the present invention. Fig.19As shown, in the detection steps of this embodiment, first, the image processing device 5 performs image correction processing such as calibration, shadow correction and noise removal on the two two-dimensional images (original image a, original image b) input from the area sensors 4a and 4b using the pre-derived camera parameters, thereby generating a corrected image a and a corrected image b (steps S41a and S41b). Next, the image processing device 5 generates a differential image by performing a differential process between the corrected image a and the corrected image b (step S42). The surface defect candidate portion of the concave-convex property at the inspection object part is calculated from the generated differential image (step S43). Next, the image processing device 5 calculates the surface defect candidate portion of the concave-convex property at the inspection object part from the generated differential image, and outputs information related to the scale texture and harmless texture removed by the differential process.

[0126] Next, the image processing device 5 performs mask calculation processing on the two two-dimensional images (original image a, original image b) input from the area sensors 4a and 4b, respectively, using the surface defect candidate parts of the concavo-convexity calculated in the process of step S43. Through this process, the image processing device 5 generates a cut-out original image a and a cut-out original image b in which the area to be determined as the bottom steel part is cut out (steps S44a and S44b). Next, the image processing device 5 generates a cut-out corrected image a and a cut-out corrected image b by performing image correction processing on the cut-out original image a and the cut-out original image b (steps S45a and S45b).

[0127] Next, the image processing device 5 performs a bright part binarization process to detect the bright part by setting the value of the pixel whose brightness is above the threshold value to 1 and setting the value of the pixel whose brightness is less than the threshold value to 0 on the cut-out correction image a and the cut-out correction image b, respectively. Through this process, the image processing device 5 generates a cut-out bright part binarization image a and a cut-out bright part binarization image b (steps S46a, S46b). Next, the image processing device 5 performs an AND process on the cut-out bright part binarization image a and the cut-out bright part binarization image b. Through this process, the image processing device 5 extracts pixels whose value is 1 from both the cut-out bright part binarization image a and the cut-out bright part binarization image b, and generates a cut-out two bright part image (step S47). Next, the image processing device 5 calculates the proportion of the cut-out two bright part images in the entire concavo-convex surface defect candidate part as the two bright part occupancy rate (step S48). Moreover, the image processing device 5 determines whether it is a bottom steel part or a concavo-convex surface defect by threshold processing using the calculated two bright part occupancy rate. Here, when it is determined that it is not a bottom steel portion, the image processing device 5 determines that it is a surface defect (step S49).

[0128] It should be noted that, as a method for detecting the surface defect candidate portion of the concavo-convexity at the inspection object portion using the differential image in the processing of step S43, a known method can be used. In particular, the method described in Patent Documents 1 and 2 can offset the oxide scale texture and harmless texture by differential processing, and use the unique light and dark pattern generated by the concavo-convex surface defect, so there is an advantage of being able to detect the surface defect candidate portion of the concavo-convexity with high precision, which is suitable. The processing of step S33 described above can be set as an example of the processing of step S43.

[0129] Although the above describes the embodiment to which the invention completed by the inventor is applied, the present invention is not limited by the description and drawings of the embodiment which constitute a part of the disclosure of the present invention. For example, in the present embodiment, since the light sources 2a and 2b are arranged symmetrically on the left and right, the light and dark patterns on the left and right are recognized, but even if the arrangement positions of the light sources 2a and 2b are not symmetrical or asymmetrical on the top and bottom, the surface defects of the concave and convex properties can be detected by the same process. Specifically, in the case where the light sources are arranged symmetrically on the top and bottom, only the light and dark pattern is changed from the left and right direction to the top and bottom direction. Therefore, if the light and dark pattern is rotated 90 degrees, the surface defects of the concave and convex properties can be detected by the same process.

[0130] In addition, in Fig. 20 When light sources 2a and 2b are arranged so that the irradiation directions of the illumination light differ by 90 degrees as shown in FIG. 1 , if the surface defect is concave, the front side of the light source becomes darker and the back side becomes brighter, and if the surface defect is convex, the front side of the light source becomes brighter and the back side becomes darker. Specifically, when the surface defect is concave, the two-dimensional image obtained by the illumination light from light source 2a becomes Fig.21 As shown in (a), the two-dimensional image obtained by the illumination light from the light source 2b becomes Fig.21 (b). Therefore, the difference image becomes Fig.21 (c) shows a light and dark pattern with contrast from the lower left to the upper right. Therefore, if the light and dark pattern is rotated 45 degrees, the concave surface defects can be detected by the same method as the light and dark pattern in the left and right directions. In addition, by using more than three light sources, differential images of multiple patterns can be obtained respectively, so the detection accuracy of surface defects can be further improved.

[0131] In addition, in the present embodiment, when the illumination light is irradiated from a direction symmetrical to the normal line of the inspection object, the surface defects of the concave-convex properties are detected, but the irradiation direction of the illumination light does not necessarily need to be symmetrical. In addition, the surface defect detection method of the present embodiment can be applied to the entire steel manufacturing line regardless of the hot state or the cold state. In addition, in the case where a mark with high reflectivity is applied to the surface of the product for product identification, only the mark portion can be detected by the same technique. In many cases, the mark has a unique shape, so it can be identified using general image feature quantities.

[0132] In addition, by manufacturing steel while detecting the surface defects of steel using the surface defect detection device 1 or the surface defect detection method as one embodiment of the present invention, the bottom steel part and harmful surface defects can be distinguished with high precision to improve the manufacturing yield of steel. For example, cold scratches are one of the most representative bottom steel parts in the steel process. Surface defects (especially, concave and convex surface defects) are often generated by pressing protrusions such as rollers during hot rolling, and cold scratches are often generated by surface friction during transportation. Therefore, if these defects can be distinguished and detected, it becomes easy to determine the cause of the defect. As a result, the cause on the manufacturing line can be immediately removed, and further generation can be suppressed. In addition, the surface defect detection device 1 as one embodiment of the present invention can also be used as an inspection device for manufacturing equipment constituting steel. Specifically, the surface defect detection device of the present invention is used to inspect the steel manufactured by the manufacturing equipment to detect the surface defects of the steel. In this case, too, for the above reasons, the bottom steel part and harmful surface defects can be distinguished with high precision to improve the manufacturing yield of steel.

[0133] In addition, the surface defect detection device 1 or the surface defect detection method as an embodiment of the present invention can be used to manage the quality of steel by classifying steel based on the presence or absence of surface defects. In other words, the bottom steel part and harmful surface defects (especially, concave-convex surface defects) can be distinguished with high precision to improve the manufacturing yield of steel. For example, cold scratches are one of the most representative bottom steel parts in steelmaking. Moreover, in the case where the concave-convex surface defects are harmful and the cold scratches are harmless, if the cold scratches are mistakenly detected, the healthy steel will be judged as defective, which sometimes reduces the manufacturing yield. In this way, in the case where the severity of concave-convex surface defects and cold scratches is different, if these defects can be distinguished and detected, it is possible to make a judgment corresponding to the inspection requirements of the steel, and it is possible to prevent the manufacturing yield of the steel from decreasing. In this way, other embodiments, examples, and application technologies completed by technicians in this field based on this embodiment are all included in the scope of the present invention.

[0134] Industrial Applicability

[0135] According to the present invention, a surface defect detection method, a surface defect detection device, a method for generating a surface defect determination model, and a surface defect determination model capable of distinguishing a bottom steel portion and a surface defect with high accuracy can be provided. In addition, according to the present invention, a steel manufacturing method, a steel quality management method, and a steel manufacturing device capable of distinguishing a bottom steel portion and a surface defect with high accuracy and improving the manufacturing yield of the steel can be provided.

[0136] Description of symbols

[0137] 1 Surface defect detection device

[0138] 2a, 2b Light source

[0139] 3 Function Generator

[0140] 4a, 4b Area Sensors

[0141] 5 Image processing device

[0142] 6 Monitor

[0143] L Illumination

[0144] P Steel Pipe

Claims

1. A surface defect detection method for optically detecting surface defects of steel, wherein: include: an irradiation step, irradiating illumination light from different directions to the same inspection object part using two or more distinguishable light sources; and a detection step of detecting surface defects at the inspection object portion based on the degree of overlap of bright portions extracted from two or more images formed by reflected light of each illumination light, The degree of overlap of the bright parts is the ratio of the overlapped portion of the bright parts at the inspection object part to the surface defect candidate part. The surface defects are surface defects in a concavo-convex shape.

2. The surface defect detection method according to claim 1, The detection step comprises the following steps: Calculating a surface defect candidate portion at the inspection target portion based on a bright portion extracted from two or more images formed by reflected light of each illumination light; and The surface defect at the inspection target portion is detected based on the ratio of the overlapping portion of the bright portion to the surface defect candidate portion.

3. The surface defect detection method according to claim 1 or 2, The detection step comprises the following steps: The surface defect at the inspection object part is detected using a surface defect judgment model implemented with machine learning in a manner that outputs a judgment value of whether there is a surface defect at the inspection object part corresponding to the two or more images when the degree of overlap of the bright parts extracted from the two or more images is input.

4. A surface defect detection device for optically detecting surface defects of steel materials, wherein: have: an irradiation unit that irradiates illumination light from different directions to the same inspection object part using two or more distinguishable light sources; and a detection unit for detecting a surface defect at the inspection target portion based on a degree of overlap of bright portions extracted from two or more images formed by reflected light of each illumination light, The degree of overlap of the bright parts is the ratio of the overlapped portion of the bright parts at the inspection object part to the surface defect candidate part. The surface defects are surface defects in a concavo-convex shape.

5. A method for manufacturing a steel material, comprising the following steps: The surface defects of a steel material are detected and the steel material is manufactured using the surface defect detection method according to any one of claims 1 to 3.

6. A method for quality management of steel, comprising the following steps: By using the surface defect detection method according to any one of claims 1 to 3, steel materials are classified based on the presence or absence of surface defects to manage the quality of the steel materials.

7. A steel manufacturing system comprising: Manufacturing equipment for the manufacture of steel products; and The surface defect detection device according to claim 4 inspects the steel material manufactured by the manufacturing equipment.

8. A method for generating a surface defect determination model, The degree of overlap of bright parts extracted from two or more images formed by reflected light of illumination light irradiated from two or more distinguishable light sources from different directions to the same inspection object part and the determination result of whether there is a surface defect in the inspection object part are used as training data, A learned model is generated by machine learning, which takes the degree of overlap of bright parts extracted from the two or more images as an input value and takes the determination value of whether there is a surface defect in the inspection object part corresponding to the two or more images as an output value as a surface defect determination model; The degree of overlap of the bright parts is the ratio of the overlapped portion of the bright parts at the inspection object part to the surface defect candidate part. The surface defects are surface defects in a concavo-convex shape.

9. A surface defect determination device, comprising a surface defect determination model generated by the surface defect determination model generation method according to claim 8.

Images