Device for inspecting defect in welded portion based on radiographic testing and method therefor

Abstract

To provide a device for inspecting a defect in a welded portion based on radiographic testing and a method therefor.SOLUTION: A device for inspecting a defect in a welded portion based on radiographic testing comprises a radiographic device RTA, a scanner SC, and a testing device 10. A method for inspecting the defect in the welded portion includes the steps of: evaluating a quality of a radiographic image according to the number of wires of an image quality indicator; performing image processing so as to highlight features of the defect in a welded portion in a reading region when the quality of the radiographic image satisfies a preset reference value; detecting the defect in the welded portion in the reading region; and outputting a defect report including a defect location, a defect area, and a defect type according to the detected defect.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a defect inspection technique for welds, and more particularly to an apparatus and method for inspecting defects in welds based on radiographic testing (RT). [Background technology]

[0002] Radiographic testing (RT) is an inspection method that detects defects inside a test object by selecting radiation such as X-rays or gamma rays according to the conditions of use and the application, and passing it through the test object to form an X-ray film. It is currently the most widely used non-destructive testing method for detecting internal defects. Summary of the Invention [Problem to be solved by the invention]

[0003] SUMMARY OF THE INVENTION It is an object of the present invention to provide an apparatus and method for radiographically based inspection of welds for defects. [Means for solving the problem]

[0004] In order to achieve the above object, a method for inspecting defects in welds according to a preferred embodiment of the present invention includes the steps of: a reading area processing unit performing image processing to make features of a reading area including a welded portion of an object stand out in the radiation image; a reading area processing unit detecting the reading area in the radiation image; an image quality processing unit performing image processing to make features of the penetrometer stand out in the radiation image; a step of the image quality processing unit counting the number of wires of the penetrometer in the radiation image; a step of the image quality processing unit evaluating the quality of the radiation image according to the number of wires of the penetrometer counted; and if the quality of the radiation image satisfies a predetermined reference value, a defect processing unit performing image processing to make features of defects in the welded portion stand out in the reading area; a step of the defect processing unit detecting defects in the welded portion in the reading area; and a step of an output unit outputting a defect report including defect positions, defect areas, and defect types according to the detected defects.

[0005] The step of detecting the reading area is characterized in that the reading area processing unit extracts a signal indicating a pixel level of the radiation image through a pole search based detection model, detects edges indicating weld beads from the signal extracted from maximum and minimum points of the extracted signal, and detects an area occupied by the edges indicating the weld beads in the radiation image using a bounding box as the reading area.

[0006] The step of detecting the reading area is characterized in that the reading area processing unit detects the area occupied by the reading area in the radiation image as a bounding box via a reading area detection model, which is a learning model.

[0007] The step of detecting the reading area includes the steps of: the reading area processing unit extracting a signal indicating a pixel level of the radiation image through a polar search based detection model; detecting edges indicating weld beads from the signal extracted from the signal's maximum and minimum points; and detecting an area occupied by the edges indicating the weld beads in the radiation image as the reading area using a first bounding box; the reading area processing unit detecting an area occupied by the reading area using a second bounding box using a reading area detection model; and the reading area processing unit detecting the intersection area of ​​the first bounding box and the second bounding box as the final reading area.

[0008] The step of detecting the number of penetrometer wires includes the steps of the image quality processing unit identifying a text area in the radiation image, which is an area occupied by the penetrometer text; the image quality processing unit identifying a wire area based on the text area, which is an area occupied by the penetrometer wires; and the image quality processing unit counting the number of penetrometer wires in the wire area.

[0009] The step of counting the number of penetrometer wires in the wire region includes the steps of: the image quality processing unit performing a Hough transform on the wire region to construct an image of the wire region having a plurality of lines; the image quality processing unit clustering the plurality of lines based on the density of the plurality of lines to derive one or more line clusters; and the image quality processing unit counting the number of penetrometer wires according to the number of the derived line clusters.

[0010] The step of detecting the number of wires of the penetrameter includes the steps of: the image quality processing unit detecting wire regions, which are regions occupied by the multiple wires of the penetrameter, in the radiation image by bounding boxes using a penetrameter detection model; and the image quality processing unit detecting count wire regions, which are regions occupied by each of the multiple wires of the penetrameter, in the wire region by bounding boxes using a count model, and counting the number of wires of the penetrameter according to the number of detected count wire regions.

[0011] The step of detecting defects in the welded portion is characterized in that the defect processing unit performs a weighting operation in which learned weights are applied to the reading area of ​​the radiographic image to derive a defect vector, and the defect vector includes a defect bounding box indicating the area occupied by the defect in the reading area of ​​the radiographic image, and a defect class indicating the type of the detected defect.

[0012] The method further includes, before the step of processing the image to make the features of the reading area stand out, the following steps: a learning unit loads learning data including a learning radiation image and a target vector; a learning unit inputs the learning radiation image to a defect detection model having weights that have not yet been trained; the defect detection model performs a weighting operation to which the weights that have not yet been trained are applied to the learning radiation image, thereby deriving a defect vector including a defect bounding box that detects an area occupied by a defect in the learning radiation image and a defect class that indicates the type of the detected defect; a learning unit calculates a loss that indicates the difference between the target vector and the defect vector using a loss function; and a learning unit performs optimization to modify the weights of the defect detection model so that the calculated loss is minimized.

[0013] The training radiological image includes a radiological image of an object having a defect in a welded portion, which is generated by combining an image of the defect with a radiological image of an object having no defect in a welded portion, and the target vector includes a target bounding box indicating an area occupied by the defect in the training radiological image, and a target class indicating a class of the defect present in the training radiological image.

[0014] In order to achieve the above object, according to a preferred embodiment of the present invention, an apparatus for inspecting defects in welds includes: a reading area processing unit that performs image processing on the radiation image to make features of a reading area including a welded portion of an object stand out, and detects the reading area in the radiation image; an image quality processing unit that performs image processing on the radiation image to make features of the penetrometer stand out, counts the number of penetrometer wires in the radiation image, and evaluates the quality of the radiation image in accordance with the counted number of penetrometer wires; a defect processing unit that performs image processing on the reading area to make features of defects in the welded portion stand out, and detects defects in the welded portion in the reading area if the quality of the radiation image satisfies a predetermined reference value; and an output unit that outputs a defect report including defect positions, defect areas, and defect types in accordance with the detected defects.

[0015] The reading area processing unit extracts signals indicating pixel levels of the radiation image through a polar search based detection model, detects edges indicating weld beads from signals extracted from maximum and minimum points of the extracted signals, and detects an area occupied by the edges indicating the weld beads in the radiation image using a bounding box as the reading area.

[0016] The reading area processing unit detects an area occupied by the reading area in the radiation image as a bounding box through a reading area detection model, which is a learning model.

[0017] The reading area processing unit extracts a signal indicating a pixel level of the radiation image through a polar search based detection model, detects edges indicating weld beads from the signal extracted from maximum and minimum points of the signal, detects an area occupied by the edges indicating the weld beads in the radiation image as the reading area using a first bounding box, detects an area occupied by the reading area using a second bounding box using a reading area detection model, and detects an intersection area of ​​the first bounding box and the second bounding box as the final reading area.

[0018] The image quality processing unit is characterized in that it identifies a text area in the radiation image, which is an area occupied by the text of the penetrometer, identifies a wire area, which is an area occupied by the wires of the penetrometer based on the text area, and counts the number of wires of the penetrometer in the wire area.

[0019] The image quality processing unit performs a Hough transform on the wire region to construct an image of the wire region having a plurality of lines, clusters the plurality of lines based on the density of the plurality of lines to derive one or more line clusters, and counts the number of wires of the transmittance meter according to the number of the derived line clusters.

[0020] The image quality processing unit detects wire regions, which are regions occupied by multiple wires of the transmittance meter, in the radiation image by a bounding box using a transmittance meter detection model, detects count wire regions, which are regions occupied by each of the multiple wires of the transmittance meter, in the wire region by a bounding box using a count model, and counts the number of wires of the transmittance meter according to the number of detected count wire regions.

[0021] The defect processing unit performs a weighting operation in which learned weights are applied to a reading area of ​​the radiographic image to derive a defect vector, and the defect vector includes a defect bounding box indicating an area occupied by the defect in the reading area of ​​the radiographic image, and a defect class indicating a type of the detected defect.

[0022] The device further includes a learning unit that loads learning data including a training radiographic image and a target vector, inputs the training radiographic image to a defect detection model having weights for which learning has not been completed, and causes the defect detection model to perform a weighting operation in which the weights for which learning has not been completed are applied to the training radiographic image to derive a defect vector including a defect bounding box that detects an area occupied by a defect in the training radiographic image and a defect class that indicates the type of the detected defect, calculates a loss that indicates the difference between the target vector and the defect vector using a loss function, and performs optimization by correcting the weights of the defect detection model so that the calculated loss is minimized.

[0023] The training radiological image includes a radiological image of an object having a defect in a welded portion, which is generated by combining an image of the defect with a radiological image of an object having no defect in a welded portion, and the target vector includes a target bounding box indicating an area occupied by the defect in the training radiological image, and a target class indicating a class of the defect present in the training radiological image. [Effects of the Invention]

[0024] According to the present invention, it is possible to shorten the inspection time and obtain objective results compared to when an inspector evaluates with the naked eye. [Brief explanation of the drawings]

[0025] [Figure 1] 1 is a diagram illustrating the configuration of a system for inspecting defects in welds based on radiographic inspection according to an embodiment of the present invention. [Figure 2]1 is a diagram illustrating the configuration of an apparatus for inspecting defects in welds based on radiographic inspection according to an embodiment of the present invention. [Figure 3] 1 is a flowchart illustrating a method for generating a defect detection model by learning according to an embodiment of the present invention. [Figure 4] 1 is a flowchart illustrating a method for deriving a digitized radiation image and performing a radiographic test (RT) according to an embodiment of the present invention. [Figure 5] 1 is a flow chart illustrating a method for radiographic inspection-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 6] 10A-10C are example screen shots illustrating a method for radiography-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 7] 10A-10C are example screen shots illustrating a method for radiography-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 8] 10A-10C are example screen shots illustrating a method for radiography-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 9] 10A-10C are example screen shots illustrating a method for radiography-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 10] 10A-10C are example screen shots illustrating a method for radiography-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 11] 10A-10C are example screen shots illustrating a method for radiography-based weld defect inspection in accordance with an embodiment of the present invention. [Figure 12] FIG. 1 illustrates a computing device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0026] Although the present invention can be implemented in various forms by adding various modifications, specific embodiments will be illustrated and described in detail in the detailed description, but it should be understood that this is not intended to limit the present invention to the specific embodiments, but rather to include all modifications, equivalents, or alternatives that fall within the spirit and technical scope of the present invention.

[0027] The terms used in the present invention are merely used to describe specific embodiments and are not intended to limit the present invention. A singular expression includes a plural expression unless the context clearly indicates otherwise. It should be understood that, in the present invention, terms such as "comprise" or "have" are intended to specify the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, and do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0028] In particular, the terms and words used in the following specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, but should be interpreted as meanings and concepts that correspond to the technical idea of ​​the present invention, in accordance with the principle that the inventor can appropriately define the concepts of terms to best describe his / her invention. In particular, in the embodiments of the present invention, estimation means deriving a calculation result based on what a machine learning model (LM: Machine Learning Model / Deep Learning Model) has learned.

[0029] First, an apparatus for inspecting defects in welds based on radiographic inspection according to an embodiment of the present invention will be described. Fig. 1 is a diagram for explaining the configuration of a system for inspecting defects in welds based on radiographic inspection according to an embodiment of the present invention. Fig. 2 is a diagram for explaining the configuration of an apparatus for inspecting defects in welds based on radiographic inspection according to an embodiment of the present invention.

[0030] Referring to FIG. 1, a system for inspecting radiographically based weld defects according to an embodiment of the present invention includes a radiography device RTA, a scanner SC, and an inspection device 10.

[0031] The radiographic imaging apparatus RTA is used to perform radiography on an object to which an Image Quality Indicator (IQI) is attached after welding has been performed on the object, and to extract a radiographic inspection film in which the object is photographed. Here, the object may be a pipe, a tube, or the like.

[0032] The scanner SC scans the radiographic inspection film to generate a digitized radiation image. Once the digitized radiation image is generated, it is input to the inspection device 10. At this time, the radiation image may be input directly from the scanner SC or may be stored in another storage medium and then input from the storage medium.

[0033] The inspection device 10 is for inspecting defects in welds from radiation images. To this end, the inspection device 10 includes a learning unit 100, a reading area processing unit 200, an image quality processing unit 300, a defect processing unit 400, and an output unit 500.

[0034] The learning unit 100 is basically for training a learning model through learning (deep learning or machine learning). The learning unit 100 can also perform continual learning on the learning model. The learning model according to an embodiment of the present invention includes a reading area detection model, a penetrameter detection model, a counting model, and a defect detection model. The reading area detection model is trained to detect a reading area, which is an area including a weld bead, from a radiation image. The penetrameter detection model is trained to detect a wire area WA including multiple penetrameter wires from a radiation image. The counting model is trained to detect each individual wire from the wire area WA. The defect detection model is trained to detect defects in a welded portion in the reading area.

[0035] Examples of learning models according to embodiments of the present invention include CNN, YOLO, RCNN, and Faster RCNN. A learning model including a reading area detection model, a penetrometer detection model, a count model, and a defect detection model includes multiple interconnected layers (or modules), and the multiple layers (or modules) perform multiple calculations. The multiple layers (or modules) are also connected by weights (W). That is, the output of a calculation result from any one layer (or module) is input to the calculation in the next layer after a weight is applied. The detection model DM performs multiple calculations on input data, to which weights are applied between multiple layers (or modules), to derive an output. In other words, it performs multiple calculations connected by weights between multiple layers (or modules). Such multiple calculations connected by weights between multiple layers (or modules) of a learning model are referred to as "weight calculations."

[0036] In particular, the learning unit 100 performs continual learning on learning models including the read area detection model, the penetrometer detection model, the count model, and the defect detection model. Such continual learning involves continuously updating the learning models using new learning data. The learning unit 100 can evaluate the performance of each learning model using the new learning data. For example, the learning unit 100 can measure the degree of performance degradation using an average forgetting index. If the performance evaluation of each learning model indicates that the performance degradation exceeds a performance degradation threshold, the learning unit 100 can update the learning model using new learning data.

[0037] The reading region processing unit 200 detects the reading region (ROI) in the radiation image after performing image processing to highlight the features of the reading region (ROI) including the welded portion of the object in the radiation image.

[0038] The image quality processor 300 processes the image so that the characteristics of the transmittance meter stand out in the radiation image, counts the number of transmittance meter wires in the radiation image, and evaluates the quality of the radiation image based on the number of transmittance meter wires counted.

[0039] The defect processing unit 400 is for performing image processing to highlight the characteristics of defects in the welded portion in the reading region (ROI) and for detecting defects in the welded portion in the reading region (ROI).

[0040] The output unit 500 is for outputting a defect report including the defect location, defect area and defect type when a defect in the weld is detected.

[0041] The specific operation of the inspection device 10, which includes the aforementioned learning unit 100, reading area processing unit 200, image quality processing unit 300, defect processing unit 400, and output unit 500, will be described in more detail below.

[0042] Next, a method for generating a defect detection model by learning according to an embodiment of the present invention will be described. Fig. 3 is a flowchart for explaining a method for generating a defect detection model by learning according to an embodiment of the present invention.

[0043] Referring to FIG. 3, the learning unit 100 loads pre-defined learning data in step S110. The learning data includes a learning radiation image (RI) and a target vector. The learning radiation image (RI) is a radiation image of an object having a welded portion. In particular, the learning radiation image (RI) may be a radiation image of an object having a defect in the welded portion or a radiation image of an object without a defect in the welded portion. In particular, if the number of learning data for a class indicating a specific defect type is insufficient, an image of the defected portion may be synthesized with a radiation image of an object without a defect in the welded portion to generate a radiation image of an object with a defect in the welded portion.

[0044] The target vector includes a target bounding box (e.g., GT) indicating an area occupied by a defect in the training radiological image and a target class indicating a class of a defect present in the training radiological image. Here, the target bounding box (e.g., GT) may be expressed by the center coordinates, width, and height of the bounding box.

[0045] Next, in step S120, the learning unit 100 inputs the training radiographic image to a defect detection model having weights for which training has not been completed. Then, in step S130, the defect detection model performs a weighting operation in which the weights for which training has not been completed are applied to the training radiographic image to derive a defect vector. The defect vector includes a defect bounding box (e.g., BB) and a defect class. The defect bounding box is a detected area occupied by a defect in the training radiographic image. The defect bounding box (e.g., BB) may be expressed by the center coordinates, width, and height of the bounding box. The defect class indicates the class of the detected defect. The defect class is used to classify the type of defect, and examples of defect classes include porosity, slag inclusion, poor penetration, undercut, overlap, and weld crack.

[0046] Next, in step S140, the learning unit 100 calculates a loss indicating the difference between the target vector and the defect vector using a loss function. After that, in step S150, the learning unit 100 performs optimization to correct the weights of the defect detection model so that the loss derived by the loss function is minimized.

[0047] Next, in step S160, the learning unit 100 determines whether a learning completion condition is satisfied. According to one embodiment, the learning completion condition may be when a predetermined learning rate is exceeded and the loss calculated earlier (S140) converges and falls below a preset target value. If the result of the determination in step S160 shows that the learning completion condition is not satisfied, the aforementioned steps S120 to S150 are repeated using multiple different learning data. On the other hand, if the result of the determination in step S160 shows that the learning completion condition is satisfied, the learning unit 100 completes learning of the defect detection model in step S170.

[0048] Next, a method for deriving a digitized radiation image and performing a radiographic test (RT) according to an embodiment of the present invention will be described. Fig. 4 is a flowchart illustrating a method for deriving a digitized radiation image and performing a radiographic test (RT) according to an embodiment of the present invention.

[0049] 4, in step S210, the radiographic inspection device RTA performs radiography on a welded object with an Image Quality Indicator (IQI) attached, and extracts a radiographic inspection film. Here, the object may be a pipe, a tube, or the like.

[0050] Then, in step S220, the scanner SC scans the radiographic inspection film to generate a digitized radiographic image.

[0051] After the radiation image is generated, the inspection device 10 receives the digitized radiation image in step S230 and analyzes the input radiation image to inspect the welded portion. At this time, the digitized radiation image may be input directly from the scanner SC or input via other media. Also, information about the object of the radiation image may be input by the user.

[0052] Next, a method for inspecting defects in a weld based on radiographic testing (RT) according to an embodiment of the present invention will be described. FIG. 5 is a flowchart for explaining a method for inspecting defects in a weld based on radiographic testing according to an embodiment of the present invention. Specifically, FIG. 5 is a detailed description of step S230 in FIG. 4. FIGS. 6 to 11 are example screens for explaining a method for inspecting defects in a weld based on radiographic testing according to an embodiment of the present invention.

[0053] 5, in step S310, a radiation image may be input to the inspection apparatus 10. The radiation image is obtained by attaching an Image Quality Indicator (IQI) to the welding target object and then converting the radiation film obtained by radiography into a digital image, as described in FIG.

[0054] When the radiation image is input in this manner, the reading region processing unit 200 performs image processing in step S320 to highlight the features of the reading region (ROI) including the welding portion of the object, i.e., the weld bead portion, in the radiation image. This image processing is image processing that highlights the portion of interest, i.e., the edge of the weld bead.

[0055] Next, in step S330, the reading area processing unit 200 detects a reading area (ROI) that indicates the area occupied by the weld bead in the radiation image.

[0056] According to one embodiment of step S330, the reading area processing unit 200 extracts a signal indicating the pixel level of the radiation image through a polar search-based detection model, detects edges indicating the weld bead from the extracted signal from the maximum and minimum points of the extracted signal, and can detect an area including the edges indicating the weld bead in the radiation image using a bounding box BB1 as the reading area (ROI), as shown in FIG.

[0057] According to another embodiment of step S330, the reading area processing unit 200 can use a reading area detection model, which is a learning model, to detect the reading area (ROI), which is an area that includes all edges that indicate the weld bead in the radiation image, using a bounding box BB2, as shown in Figure 6.

[0058] According to another embodiment of step S330, the region of interest (ROI) can be detected using both the pole search-based detection model and the read region detection model. Specifically, the read region processing unit 200 first extracts signals indicating pixel levels of the radiation image through the pole search-based detection model, detects edges indicating weld beads from the signals extracted from the signal maxima and minima, and detects an area in the radiation image that includes the edges indicating the weld beads using a first bounding box BB1 as the region of interest (ROI), as shown in FIG.

[0059] Then, using a reading region detection model (RDM), the reading region processing unit 200 can detect, as the reading region (ROI), an area that includes the edge indicating the weld bead using the second bounding box BB2, as shown in Fig. 6. Next, the reading region processing unit 200 can detect the intersection area of ​​the first bounding box and the second bounding box as the final reading region (ROI).

[0060] Next, the image quality processor 300 performs image processing in step S340 to make the features of the penetrator, especially the wire features of the penetrator, more prominent in the radiation image. This image processing can be summarized as follows.

[0061] The image quality processor 300 can remove noise from the radiation image and increase brightness and contrast by performing contrast limited adaptive histogram equalization (CLAHE) processing on the wire region image W03 to increase edge definition in the radiation image.

[0062] The image quality processor 300 can perform frequency domain filtering. The image quality processor 300 converts a time-domain radiation image into a frequency domain image to generate a frequency domain image, performs frequency domain filtering on the frequency domain image using a composite filter, and then inversely converts the frequency domain image from the frequency domain to the time domain. The composite filter includes a low-pass filter that passes only low-band components of the frequency domain image, a vertical weighting filter that multiplies vertical components in the frequency domain image by a predetermined weight, an hourglass-shaped filter that eliminates upper and lower regions of a region divided by two different line segments that pass through a reference point (e.g., the midpoint of the frequency domain image) in a vertical symmetrical manner in the frequency domain image, and a Butterworth filter that minimizes phase distortion in boundary regions of frequencies filtered in the frequency domain image. In particular, the Butterworth filter is expressed by Equation 1:

number

[0063] where w is the frequency and n is the filtering order.

[0064] The image quality processor 300 performs filtering using a Sobel filter in both directions based on a line segment perpendicular to the longitudinal direction of the edge in the radiation image so that the boundary of the perpendicular component to the longitudinal direction of the edge present in the radiation image can be detected.

[0065] The image quality processor 300 binarizes the radiation image to derive a binarized radiation image. At this time, the image quality processor 300 may binarize the radiation image using a threshold value that is a predetermined percentage of the most significant pixel values ​​in the pixel value distribution of the radiation image. That is, the image quality processor 300 binarizes the radiation image by dividing pixels having pixel values ​​equal to or greater than the threshold value and the remaining pixels.

[0066] The image quality processor 300 performs a morphology operation on the radiation image. At this time, the image quality processor 300 may perform at least one of an erosion operation to remove noise from the radiation image and a dilation operation to dilate edges of the radiation image, or may perform the erosion operation and the dilation operation sequentially.

[0067] Such image processing allows the image quality processor 300 to detect as many wires as an inspector would visually assess.

[0068] Next, in step S350, the image quality processor 300 counts the number of penetrometer wires in the radiation image. Referring to Figure 7, the penetrometer includes a text area TA containing text indicating the penetrometer specifications and a wire area WA in which multiple wires of different thicknesses are arranged so that the quality can be determined. The penetrometer area IA is an area including the text area IA and the wire area WA.

[0069] According to one embodiment of step S350, the image quality processor 300 detects text from the radiation image and identifies a text area TA, which is an area occupied by the text of the transmissometer. Optical Character Recognition (OCR) may be used for this purpose. The image quality processor 300 then identifies a wire area WA, which is an area occupied by multiple wires of the transmissometer, based on the text area TA. Since the wire area WA is located in a predetermined area below the text area based on the text direction, the image quality processor 300 can identify the wire area WA based on the text direction of the identified text area. Next, the image quality processor 300 counts the number of wires of the transmissometer in the wire area WA. To this end, the image quality processor 300 performs a Hough transform on the wire area WA to detect multiple lines from the wire area WA. To this end, as shown in FIG. 8, the image quality processor 300 converts the image W01 of the wire area WA in a two-dimensional Cartesian coordinate system into a polar coordinate system to generate a polar coordinate system image W02 having multiple curves. The image quality processor 300 then creates an accumulation array of points where the multiple curves in the polar coordinate system image W02 intersect. Next, the image quality processor 300 maps the multiple points whose accumulation array values ​​are equal to or greater than a threshold onto the image of the wire area WA in the Cartesian coordinate system to create an image W03 of the wire area WA having multiple lines. When multiple lines are detected in this manner, the image quality processor 300 clusters the multiple lines in the image W03 of the wire area WA having multiple lines based on their density, as shown in FIG. 9, to derive one or more line clusters LC. Next, the image quality processor 300 counts the number of wires of the penetrometer according to the number of line crowds LC derived from the image W04 of the wire area WA having the derived line crowds LC.

[0070] According to one embodiment of step S350, the image quality processor 300 uses the transmittance detector detection model to detect wire areas WA, which are areas occupied by multiple transmittance detector wires, in the radiation image using bounding boxes. Then, as shown in Fig. 10, the image quality processor 300 uses the count model to detect count wire areas, which are areas occupied by each of the multiple transmittance detector wires, in the wire area WA using bounding boxes BB1 to BB6, and counts the number of transmittance detector wires according to the number of detected count wire areas.

[0071] Next, in step S360, the image quality processor 300 evaluates whether the quality of the radiation image satisfies predefined quality conditions based on the number of counted wires. In step S360, the image quality processor 300 determines that the quality of the radiation image is suitable for defect inspection if the number of counted wires in step S350 is equal to or greater than the reference value of the predefined quality conditions. Conversely, the image quality processor 300 may determine that the quality of the radiation image is unsuitable for defect inspection if the number of counted wires in step S350 is less than the reference value of the predefined quality conditions. That is, the better the quality of the radiation image, the more counted wires there are, and the worse the quality of the radiation image, the fewer counted wires there are. Thus, the quality of the radiation image can be determined by comparing the number of counted wires with predefined quality conditions and determining whether the number of counted wires is equal to or greater than the reference value of the predefined quality conditions. If the image quality processor 300 determines in step S360 that the quality of the radiation image is suitable for defect inspection, the image quality processor 300 analyzes the radiation image and proceeds to step S380 to start defect inspection. On the other hand, if the image quality processor 300 determines that the quality of the radiation image is not suitable for defect inspection, the image quality processor 300 outputs a re-imaging command to re-imaging the target object in step S370.

[0072] If the quality of the radiation image satisfies a preset reference value, i.e., if it is determined that the quality of the radiation image is suitable for defect inspection based on the number of counted wires, the defect processing unit 400 performs image processing to highlight the defect features of the welded portion in the reading region (ROI) in step S380.

[0073] Thereafter, the defect processing unit 400 detects defects in the welded portion in the region of interest (ROI) in step S390. At this time, the defect processing unit 400 performs a weighting operation in which the learned weights are applied to the region of interest (ROI) of the radiographic image to derive a defect vector. The defect vector includes a defect bounding box BB and a defect class. The defect bounding box indicates the area occupied by the defect in the region of interest (ROI) of the radiographic image. Such a defect bounding box BX may be expressed by the center coordinates, width, and height of the bounding box. The defect class indicates the type of the detected defect. FIG. 11 shows an example in which the defect class is a crack class (CRACK), which indicates that a crack has occurred in the weld bead.

[0074] If a defect is detected, the output unit 500 generates and outputs a defect report in step S400, which indicates the defect location and defect area using a defect bounding box and the defect type using a defect class. Such a defect report may be provided to an inspector for further evaluation. Alternatively, the defect location, defect area, and defect class in the defect report may be compared with the defect location, defect area, and defect class created by the inspector, and further evaluation may be performed on matching defects.

[0075] 12 is a diagram illustrating a computing device according to an embodiment of the present invention. The computing device TN100 of FIG. 12 may be a device described herein (e.g., inspection device 10, etc.).

[0076] 12, computing device TN100 may include at least one processor TN110, a transceiver TN120, and a memory TN130. Computing device TN100 may further include a storage device TN140, an input interface device TN150, an output interface device TN160, etc. The components included in computing device TN100 may be connected by a bus TN170 to communicate with each other.

[0077] Processor TN110 can execute program commands stored in at least one of memory TN130 and storage device TN140. Processor TN110 can refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. Processor TN110 can be configured to implement procedures, functions, methods, etc. described in connection with embodiments of the present invention. Processor TN110 can control each component of computing device TN100.

[0078] Memory TN130 and storage device TN140 can each store various information related to the operation of processor TN110. Memory TN130 and storage device TN140 can each be configured with at least one of a volatile storage medium and a non-volatile storage medium. For example, memory TN130 may be configured with at least one of a read-only memory (ROM) and a random access memory (RAM).

[0079] The transceiver TN120 can transmit or receive wired or wireless signals and can communicate with a network.

[0080] In particular, the above-mentioned learning unit 100, reading area processing unit 200, image quality processing unit 300, defect processing unit 400, and output unit 500 according to an embodiment of the present invention may be realized in the form of a program readable via a computer means, stored in memory TN130, and executed by processor TN110, or may become a lower module of processor TN110.

[0081] Meanwhile, the various methods according to the above-described embodiments of the present invention can be realized in the form of a program readable by various computer means and recorded on a computer-readable recording medium. Here, the recording medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the recording medium may be specially designed and constructed for the present invention, or may be known and available to those skilled in the art of computer software. For example, recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include not only machine language, such as that produced by a compiler, but also high-level language codes executable by a computer using an interpreter, etc. Such hardware devices can be configured to operate as one or more software modules to perform the operations of the present invention, or vice versa.

[0082] Although one embodiment of the present invention has been described above, a person having ordinary skill in the art may modify and change the present invention in various ways by adding, changing, deleting or adding components within the scope of the concept of the present invention as set forth in the claims, and this also falls within the scope of the present invention.

Claims

1. a reading area processing unit performing image processing to make the features of the reading area including the welded portion of the object stand out in the radiation image; a step in which the reading area processing unit detects the reading area in the radiation image; an image quality processing unit performing image processing to make the transmissometer feature prominent in the radiation image; the image quality processor counting the number of wires of the penetrameter in the radiation image; the image quality processor evaluating the quality of the radiation image according to the number of the penetrameter wires counted; If the quality of the radiation image satisfies a predetermined standard value, a defect processing unit performs image processing to highlight the defect characteristics of the welded portion in the reading area; the defect processing unit detecting defects in the welded portion in the reading area; an output unit outputting a defect report including defect positions, defect areas, and defect types in accordance with the detected defects; Including, A method for inspecting weld defects.

2. The step of detecting the reading area includes: The reading area processing unit extracts a signal indicating a pixel level of the radiation image through a pole search based detection model; Detecting edges representing weld beads from the extracted signal from the maximum and minimum points of the extracted signal; detecting an area occupied by an edge indicating the weld bead in the radiation image as the reading area using a bounding box; 10. The method for inspecting weld defects of claim 1.

3. The step of detecting the reading area includes: the reading area processing unit detects an area occupied by the reading area in the radiation image as a bounding box via a reading area detection model, which is a learning model; 10. The method for inspecting weld defects of claim 1.

4. The step of detecting the reading area includes: the reading area processing unit extracts a signal indicating a pixel level of the radiation image through a pole search based detection model, and detects an edge indicating a weld bead from the signal extracted from the maximum and minimum points of the signal; detecting an area occupied by an edge indicating a weld bead in the radiation image as the reading area using a first bounding box; the reading area processing unit detects an area occupied by the reading area using a second bounding box by using a reading area detection model; the reading area processing unit detecting an intersection area of ​​the first bounding box and the second bounding box as a final reading area; Including, 10. The method for inspecting weld defects of claim 1.

5. The step of detecting the number of wires of the penetrometer includes: the image quality processor identifying a text area in the radiation image, the area being an area occupied by the transmissometer text; the image quality processor identifying a wire region based on the text region, the wire region being an area occupied by a penetrometer wire; the image quality processor counting the number of penetrameter wires in the wire region; Including, 10. The method for inspecting weld defects of claim 1.

6. Counting the number of penetrometer wires in the wire region includes: the image quality processor performs a Hough transform on the wire region to construct an image of the wire region having a plurality of lines; the image quality processor clustering the plurality of lines based on a density of the plurality of lines to derive one or more line clusters; the image quality processor counting the number of wires of a penetrameter according to the number of the derived line clusters; Including, 6. The method for inspecting weld defects of claim 5.

7. The step of detecting the number of wires of the penetrometer includes: the image quality processor detects a wire region in the radiation image by a bounding box using a transmissometer detection model, the wire region being an area occupied by a plurality of wires of the transmissometer; the image quality processing unit detects count wire regions, which are regions occupied by each of the plurality of wires of the penetrometer, in the wire region by a bounding box using a count model, and counts the number of wires of the penetrometer according to the number of detected count wire regions; Including, 10. The method for inspecting weld defects of claim 1.

8. The step of detecting defects in the welded portion includes: the defect processing unit performs a weight calculation in which the learned weights are applied to the read area of ​​the radiation image to derive a defect vector; The defect vector includes a defect bounding box indicating an area occupied by the defect in a read area of ​​the radiation image, and a defect class indicating a type of the detected defect.

10. The method for inspecting weld defects of claim 1.

9. before the step of processing the image so that the features of the reading area stand out, A learning unit loads learning data including learning radiation images and target vectors; a learning unit inputting the learning radiation image into a defect detection model having weights for which learning has not been completed; performing a weighting operation in which the defect detection model applies weights that have not yet been learned to the training radiographic image, and deriving a defect vector including a defect bounding box that detects an area occupied by a defect in the training radiographic image and a defect class that indicates a type of the detected defect; the learning unit calculating a loss indicating a difference between the target vector and the defect vector using a loss function; a step in which the learning unit performs optimization to correct weights of the defect detection model so as to minimize the calculated loss; further comprising: A method for inspecting weld defects according to any one of claims 1 to 8.

10. The learning radiation image includes a radiation image of an object having a defect in a weld portion, which is generated by combining an image of the defect with a radiation image of an object having no defect in a weld portion, and the target vector includes a target bounding box indicating an area occupied by a defect in the training radiological image, and a target class indicating a class of a defect present in the training radiological image; 10. The method for inspecting weld defects of claim 9.

11. Image processing is performed to highlight the features of the reading area, including the welded part of the object, in the radiographic image. a reading area processing unit that detects the reading area in the radiation image; image processing is performed to make the characteristics of the transmissometer stand out in the radiation image; Counting the number of wires of the penetrameter in the radiographic image; an image quality processing unit that evaluates the quality of the radiation image according to the number of wires counted by the transmittance meter; If the quality of the radiation image satisfies a preset standard value, image processing is performed to highlight the characteristics of defects in the welded portion in the reading area; a defect processing unit for detecting defects in the welded portion in the reading area; an output unit that outputs a defect report including defect positions, defect areas, and defect types in accordance with the detected defects; Including, Equipment for inspecting weld defects.

12. The reading area processing unit extracting a signal indicating a pixel level of the radiation image through a pole search-based detection model; Detecting edges representing weld beads from the extracted signal from the maximum and minimum points of the extracted signal; detecting an area occupied by an edge indicating the weld bead in the radiation image as the reading area using a bounding box; 12. The apparatus for inspecting weld defects of claim 11.

13. The reading area processing unit Detecting an area occupied by the reading area in the radiation image as a bounding box through a reading area detection model, which is a learning model; 12. The apparatus for inspecting weld defects of claim 11.

14. The reading area processing unit extracting a signal indicating a pixel level of the radiation image through a pole search-based detection model, and detecting an edge indicating a weld bead from the signal extracted from the maximum and minimum points of the signal; Detecting an area occupied by an edge indicating a weld bead in the radiation image as the reading area using a first bounding box; Using a reading area detection model, an area occupied by the reading area is detected by a second bounding box; Detecting an intersection area of ​​the first bounding box and the second bounding box as a final reading area; 12. The apparatus for inspecting weld defects of claim 11.

15. The image quality processing unit identifying a text area in the radiographic image, the area occupied by the transmissometer text; Identifying a wire region based on the text region, the region being an area occupied by a wire of a penetrometer; Counting the number of penetrometer wires in the wire region; 12. The apparatus for inspecting weld defects of claim 11.

16. The image quality processing unit performing a Hough transform on the wire region to construct an image of the wire region having a plurality of lines; clustering the plurality of lines based on a density of the plurality of lines to derive one or more line clusters; Counting the number of wires of the penetrometer according to the number of line clusters derived; 16. An apparatus for inspecting weld defects according to claim 15.

17. The image quality processing unit Using a transmissometer detection model, a wire region is detected in the radiation image by a bounding box, the wire region being an area occupied by a plurality of wires of the transmissometer; Using a count model, a count wire region, which is a region occupied by each of the plurality of wires of the penetrometer, is detected in the wire region by a bounding box, and the number of wires of the penetrometer is counted according to the number of the detected count wire regions.

12. The apparatus for inspecting weld defects of claim 11.

18. The defect processing section includes: A weight calculation is performed to apply the learned weights to the read area of ​​the radiation image to derive a defect vector; The defect vector includes a defect bounding box indicating an area occupied by the defect in a read area of ​​the radiation image, and a defect class indicating a type of the detected defect.

12. The apparatus for inspecting weld defects of claim 11.

19. Load training data including training radiology images and target vectors; inputting the training radiological image into a defect detection model having weights that have not yet been trained; The defect detection model performs a weighting operation to apply weights that have not yet been learned to the learning radiographic image, and derives a defect vector including a defect bounding box that detects an area occupied by a defect in the learning radiographic image and a defect class that indicates the type of the detected defect. Calculating a loss indicating a difference between the target vector and the defect vector using a loss function; a learning unit that performs optimization by correcting the weights of the defect detection model so as to minimize the calculated loss; further comprising:

19. An apparatus for inspecting defects in a weld according to any one of claims 11 to 18.

20. The learning radiation image includes a radiation image of an object having a defect in a weld portion, which is generated by combining an image of the defect with a radiation image of an object having no defect in a weld portion, and the target vector includes a target bounding box indicating an area occupied by a defect in the training radiological image, and a target class indicating a class of a defect present in the training radiological image; 20. The apparatus for inspecting weld defects of claim 19.

Images