DEFECT ANALYSIS APPARATUS, DEFECT ANALYSIS METHOD, PROGRAM, AND LEARNING APPARATUS

Abstract

To provide a defect analyzer, a defect analysis method, a program, and a training device with which it is possible to effectively analyze defects.SOLUTION: The defect analyzer comprises: a model storage unit that stores a trained model in which teaching information that represents the positions, shapes, and sizes of a plurality of defects planted in a standard test piece and detection information regarding the plurality of defects detected on the basis of a plurality of standard ultrasonic images acquired from the standard test piece are associated; an object image acquisition unit that acquires a plurality of object ultrasonic images from an inspection object; and an enhanced image creation unit that creates a plurality of enhanced images in which the plurality of defects are enhanced, on the basis of the trained model and the plurality of object ultrasonic images. The trained model has been trained using standard ultrasonic images as to which of the plurality of defects should be detected in which order of priority.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a defect analysis device, a defect analysis method, a program, and a learning device. [Background technology]

[0002] Patent Document 1 describes the following defect detection system. That is, the defect detection system described in Patent Document 1 uses, for example, an ultrasonic flaw detector to emit ultrasonic waves to an object while changing the emission position, and measures the reflected waves at each emission position. Then, an original image is created in which the luminance is changed for each intensity distribution of the reflected waves and the reflected waves are plotted in a two-dimensional matrix for each emission position. Furthermore, the system judges whether or not there is a defect based on the feature amount of a defect candidate region based on the luminance of the image calculated from the original image and the range of the feature amount calculated from a learning original image in which the location of the defect has been detected in advance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6786035 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the defect detection system described in Patent Document 1, when preparing multiple original images for learning, the original images would have to be obtained by performing multiple measurements, for example, which could take a lot of time and effort in some cases.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a defect analysis device, a defect analysis method, a program, and a learning device that are capable of efficiently analyzing defects. [Means for solving the problem]

[0006] In order to solve the above problems, the defect analysis device disclosed herein includes a model storage unit that stores a trained model that associates teacher information representing the positions, shapes, and sizes of multiple defects incorporated in a standard test piece with detection information of the multiple defects detected based on multiple standard ultrasound images acquired from the standard test piece, a target image acquisition unit that acquires multiple target ultrasound images from an inspection object, and an enhanced image creation unit that creates multiple enhanced images in which the multiple defects are enhanced based on the trained model and the multiple target ultrasound images, and the trained model learns to prioritize defects to be detected among the multiple defects in each standard ultrasound image.

[0007] The defect analysis method according to the present disclosure includes the steps of acquiring a plurality of target ultrasound images from an inspection object, and creating a plurality of enhanced images in which the plurality of defects are enhanced based on a trained model in which teacher information representing the positions, shapes, and sizes of a plurality of defects incorporated in a standard test piece and detection information of the plurality of defects detected based on the plurality of standard ultrasound images acquired from the standard test piece is associated with each other, and the plurality of target ultrasound images, wherein the trained model learns to prioritize defects to be detected among the plurality of defects in each standard ultrasound image.

[0008] The program of the present disclosure is a program that causes a computer to execute the steps of acquiring a plurality of target ultrasound images from an inspection target, and creating a plurality of enhanced images that enhance the plurality of defects based on a trained model that associates teacher information representing the position, shape, and size of a plurality of defects incorporated in a standard test piece and detection information of the plurality of defects detected based on the plurality of standard ultrasound images acquired from the standard test piece, and the plurality of target ultrasound images, wherein the trained model learns to prioritize defects to be detected among the plurality of defects in each standard ultrasound image.

[0009] The learning device of the present disclosure includes a machine learning unit that generates a trained model by learning based on teacher information representing the positions, shapes, and sizes of multiple defects incorporated in a standard test piece and detection information of the multiple defects detected based on multiple standard ultrasound images acquired from the standard test piece, and the trained model learns to prioritize defects to be detected among the multiple defects in each standard ultrasound image. Effect of the Invention

[0010] According to the defect analysis device, defect analysis method, program, and learning device disclosed herein, defects can be analyzed efficiently. [Brief description of the drawings]

[0011] [Figure 1] 1 is a block diagram showing an example configuration of a defect analysis device according to an embodiment of the present disclosure. [Diagram 2] FIG. 1 is a diagram showing an example of an ultrasonic image obtained by measuring a standard test piece according to an embodiment of the present disclosure. [Diagram 3] FIG. 13 is a diagram illustrating an example of a measurement result of a reflected wave according to an embodiment of the present disclosure. [Figure 4] FIG. 13 is a diagram illustrating an example of a measurement result of a reflected wave according to an embodiment of the present disclosure. [Diagram 5] FIG. 13 is a diagram illustrating an example of a measurement result of a reflected wave according to an embodiment of the present disclosure. [Figure 6] FIG. 1 illustrates a number of example standard ultrasound images according to an embodiment of the present disclosure. [Figure 7] FIG. 13 illustrates an example of labeling defects in a plurality of standard ultrasound images according to an embodiment of the present disclosure. [Figure 8] FIG. 2 is a diagram showing an example of an ultrasound image obtained by measuring an object according to an embodiment of the present disclosure. [Figure 9] FIG. 13 is a diagram illustrating an example of an emphasis image according to an embodiment of the present disclosure. [Figure 10] FIG. 2 is a diagram illustrating an example of a plurality of binarized images according to an embodiment of the present disclosure. [Figure 11]FIG. 2 is a diagram illustrating an example of a plurality of binarized integrated images according to an embodiment of the present disclosure. [Figure 12] FIG. 13 is a diagram illustrating an example of multiple overlaid images according to an embodiment of the present disclosure. [Figure 13] 10 is a flowchart showing an example of an operation of a learning process in the defect analysis device according to the embodiment of the present disclosure. [Figure 14] 10 is a flowchart showing an example of an operation of a determination process in a defect analysis device according to an embodiment of the present disclosure. [Figure 15] FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, a defect analysis device, a defect analysis method, a program, and a learning device according to an embodiment of the present disclosure will be described with reference to FIGS. 1 to 15. FIG. 1 is a block diagram showing a configuration example of a defect analysis device according to an embodiment of the present disclosure. FIG. 2 is a diagram showing an example of an ultrasonic image obtained by measuring a standard test piece according to an embodiment of the present disclosure. FIGS. 3 to 5 are diagrams showing an example of a measurement result of a reflected wave according to an embodiment of the present disclosure. FIG. 6 is a diagram showing an example of a plurality of standard ultrasonic images according to an embodiment of the present disclosure. FIG. 7 is a diagram showing an example of labeling defects in a plurality of standard ultrasonic images according to an embodiment of the present disclosure. FIG. 8 is a diagram showing an example of an ultrasonic image obtained by measuring an object according to an embodiment of the present disclosure. FIG. 9 is a diagram showing an example of an emphasized image according to an embodiment of the present disclosure. FIG. 10 is a diagram showing an example of a plurality of binarized images according to an embodiment of the present disclosure. FIG. 11 is a diagram showing an example of a plurality of binarized integrated images according to an embodiment of the present disclosure. FIG. 12 is a diagram showing an example of a plurality of superimposed images according to an embodiment of the present disclosure. FIG. 13 is a flowchart showing an operation example of a learning process in a defect analysis device according to an embodiment of the present disclosure. FIG. 14 is a flowchart showing an operation example of a determination process in a defect analysis device according to an embodiment of the present disclosure. 15 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. Note that in each drawing, the same or corresponding components are denoted by the same reference numerals and descriptions thereof will be omitted as appropriate.

[0013] (Configuration of defect analysis equipment) The defect analysis device 1 shown in FIG. 1 can be configured using a computer such as a server, a personal computer, or a tablet terminal, and includes the following units as a functional configuration including a combination of hardware such as the computer and peripheral devices, and software such as a program executed by the computer. That is, the defect analysis device 1 includes a standard image acquisition unit 11, a teacher information input unit 12, a machine learning unit 13, a model storage unit 14, a target image acquisition unit 15, an enhanced image creation unit 16, and an output unit 17. The model storage unit 14 stores a trained model 141. Note that a part or all of the functional configuration in the defect analysis device 1 may be configured on a cloud. Note that the defect analysis device 1 is also an example of a configuration of a learning device according to the present disclosure.

[0014] The defect analysis device 1 uses a trained model 141 that associates multiple standard ultrasonic images acquired by the ultrasonic waveform acquisition device 2 from the standard test piece 3 with teacher information representing the position, shape, and size of multiple defects embedded in the standard test piece 3, to create multiple highlighted images that highlight the defects contained in the inspection object 4 based on multiple target ultrasonic images acquired by the ultrasonic waveform acquisition device 2 from the inspection object 4.

[0015] The ultrasonic waveform acquisition device 2 is a device for performing ultrasonic flaw inspection, for example, called an ultrasonic flaw detector. In this case, the ultrasonic waveform acquisition device 2 acquires reflected ultrasonic waves in two dimensions using a probe that is kept at a certain distance from the standard test piece 3 or the inspection object 4 via water and can scan in a plane. An image in which the position of the probe is the two-dimensional pixel position and the height (intensity, amplitude) of the reflected wave is the value of each pixel is called a C-Scan image. A C-Scan image is a flaw inspection result display image mapped two-dimensionally. In this embodiment, a C-Scan image is also called an ultrasonic image.

[0016] The ultrasonic waveform acquisition device 2 measures an echo waveform, for example, as shown in FIG. 3, at each position of the probe, and outputs the measurement results for all positions to the defect analysis device 1. The measurement results include time series data of the measurement values ​​of the height of the echo waveform at each position. The measurement results also include data in which the height of the waveform other than the three peaks shown in FIG. 3 is approximately zero. The measurement results of the ultrasonic waveform acquisition device 2 are also called waveform data. FIG. 3 shows the change in the echo wave with the horizontal axis representing time (or distance) and the vertical axis representing echo intensity (for example, amplitude). The peak at the left end is the surface echo SE, the central peak is the intermediate echo IE, and the peak at the right end is the bottom echo BE. The surface echo SE is a wave reflected from the surface of the standard test piece 3 or the inspection object 4. The intermediate echo IE is a wave reflected from an intermediate structure of the standard test piece 3 or the inspection object 4. In this case, the intermediate echo IE is not a wave reflected from a defect. The bottom echo BE is a wave reflected from the bottom of the standard test piece 3 or the inspection object 4. FIG. 4 shows an example in which an intermediate echo IE does not occur, and FIG. 5 shows an example in which an echo due to a defect (defect echo DE) is measured.

[0017] FIG. 2 shows an example of a C-Scan image of a standard test piece 3. The standard test piece 3 is formed by stacking multiple layers of sheet-like members in which multiple pseudo defects (artificial defects) with different positions, shapes, sizes (depth, length, width, etc.), types, and overlapping patterns are arranged. However, the standard test piece 3 is not limited to those manufactured as standard test pieces, and may be, for example, a part after a natural defect simulation test or the like has been performed, or a part used in an actual machine. If only small pseudo defects are inserted into the standard test piece, overlearning of small defects may occur, and large defects may not be detected by the defect analysis device. Therefore, it is preferable to insert pseudo defects with a relatively large size. The C-Scan image of the standard test piece 3 shown in FIG. 2 is an image in which the maximum echo intensity measured within a gate having a predetermined time width in FIG. 3 is used as the brightness value at each position. The lighter (whitish) color indicates a higher echo intensity. The inspection target 4 is, for example, a composite material part for an aircraft.

[0018] The standard image acquisition unit 11 acquires a plurality of standard ultrasonic images acquired from the standard test piece 3. The target image acquisition unit 15 acquires a plurality of target ultrasonic images from the test object 4. The standard image acquisition unit 11 and the target image acquisition unit 15 acquire ultrasonic images (waveform data) from the ultrasonic waveform acquisition device 2, for example, via a communication line or by using a predetermined storage medium. In this embodiment, the standard image acquisition unit 11 and the target image acquisition unit 15 set a plurality of gates F11 to F15 or gates F11 to F13 as shown in Figs. 3 to 5, and acquire ultrasonic images for each set gate. The standard ultrasonic image is an ultrasonic image of the standard test piece 3. The target ultrasonic image is an ultrasonic image of the test object 4. The gates F11 to F15 are defined in terms of width (referred to as a gate range) and height (referred to as a gate level). For example, for the gate F11, only echoes that are located within the time width (or distance) indicated by each end (circle mark) and have a level higher than the line connecting each end are measured. For example, in the example shown in Fig. 5, the defect echo DE can be detected in the ultrasonic image of the gate F12, the ultrasonic image of the gate F13, and the ultrasonic image of the gate F15. On the other hand, the defect echo DE cannot be detected in the ultrasonic image of the gate F11 and the ultrasonic image of the gate F14. In this case, the gate ranges and gate levels of the multiple gates F11 to F15 are different from each other. The number of gates, the gate ranges, and the gate levels are specified, for example, for each component.

[0019] Fig. 6 shows an example of standard ultrasonic images of the gates F11 to F13 shown in Fig. 4. The standard image acquisition unit 11 acquires three standard ultrasonic images shown in Fig. 6 from a single measurement result of the standard test piece 3.

[0020] As shown in FIG. 7, the teacher information input unit 12 labels (annotates) all defects that can be detected from the standard ultrasound images of gates F11 to F13 by, for example, a nondestructive testing (NDT) engineer to provide teacher information indicating the positions, shapes, and sizes of multiple defects. Furthermore, each defect is classified into defects that should be detected in the ultrasound image and defects that should be detected in another ultrasound image, and a priority order is set. In the example shown in FIG. 7, among the multiple defects that have been set, defects with a high priority order are represented by a shaded area D11, and defects with a low priority order are represented by an unshaded area D12. For example, the trained model can be trained appropriately by increasing the priority order of the ultrasound image of a gate in which the feature amount of the defect is well displayed. Note that, although each defect is displayed in a simplified area in FIG. 7, data indicating the shape of the defect can be included in the teacher information using, for example, multiple position information (coordinate information) or vector information. Priority information can also be included in the teacher information in association with position information, etc.

[0021] The machine learning unit 13 uses the standard ultrasound images of the gates F11 to F13 labeled by the teacher information input unit 12 as teacher data, inputs each ultrasound image, and performs machine learning (or re-machine learning) of an unlearned machine learning model (a model before machine learning of the trained model 141) or the trained model 141, outputting an image (called an enhanced image) that indicates the possibility of a defect by pixel brightness values ​​(the higher the brightness value, the more likely the defect is). The machine learning unit 13 stores the trained trained model 141 in the model storage unit 14.

[0022] The emphasized image creating unit 16 creates a plurality of emphasized images in which a plurality of defects are emphasized, based on the learned model 141 and a plurality of target ultrasound images acquired by the target image acquiring unit 15 as described above. FIG. 8 shows an example of a target ultrasound image. FIG. 9 shows an example of an emphasized image in which a plurality of emphasized images are superimposed. Note that the ultrasound image before binarization shown in FIG. 8, which will be described later, is also referred to as an original image.

[0023] Further, the enhanced image creating unit 16 obtains a plurality of binary images obtained by binarizing a plurality of enhanced images, and integrates the plurality of binary images to create a binary integrated image representing a plurality of defects. FIG. 10 shows an example of a plurality of binary images corresponding to the gates F11 to F13. FIG. 11 shows an example of a binary integrated image representing a plurality of defects created by integrating the plurality of binary images. The enhanced image creating unit 16, for example, sets a different threshold value for each of the gates F11 to F13, and binarizes each enhanced image. In this case, the enhanced image creating unit 16 may set the threshold value or adjust it from a predetermined set value so that the area that becomes black or white falls within a predetermined range. This makes it possible to appropriately suppress, for example, the possibility of erroneously determining that there is a defect.

[0024] Further, the enhanced image creation unit 16 creates a superimposed image by superimposing the original image of the multiple target ultrasound images and the binarized integrated image. Fig. 12 shows an example of the superimposed image. Note that Fig. 12 shows the superimposed image in grayscale, but for example, the binarized integrated image may be an image with a color other than black instead of a black and white image, and then superimposed on the original image.

[0025] The output unit 17 outputs the superimposed image from a predetermined output device such as a display.

[0026] (Example of the learning process) Next, an example of the operation of the defect analysis device 1 in the process of creating the trained model 141 will be described with reference to FIG. 13. In the process shown in FIG. 13, first, the standard image acquisition unit 11 acquires the measurement results of the standard test piece 3 from the ultrasonic waveform acquisition device 2 (step S11). Next, the standard image acquisition unit 11 generates (acquires) multiple standard ultrasonic images based on multiple gates (step S12). Next, the teacher information input unit 12 labels defect information for each standard ultrasonic image (step S13). Next, the machine learning unit 13 creates the trained model 141 by supervised machine learning and stores it in the model storage unit 14 (step S14).

[0027] (Example of the judgment process) Next, an example of the operation of the defect analysis device 1 in the defect determination process will be described with reference to FIG. 14. In the process shown in FIG. 14, first, the target image acquisition unit 15 acquires the measurement result of the inspection target 4 from the ultrasonic waveform acquisition unit 2 (step S21). Next, the enhanced image creation unit 16 creates a plurality of target ultrasonic images corresponding to each gate, and creates a plurality of enhanced images using the trained model 141 (step S22). Next, the enhanced image creation unit 16 binarizes the plurality of enhanced images to create a plurality of binary images (step S23). Next, the enhanced image creation unit 16 acquires a plurality of binary images obtained by binarizing the plurality of enhanced images, integrates the plurality of binary images, and creates a binary integrated image representing a plurality of defects (step S24). Next, the enhanced image creation unit 16 creates a superimposed image by superimposing the original images of the plurality of target ultrasonic images and the binary integrated image, and the output unit 17 outputs the superimposed image (step S25).

[0028] (Action and effect) The defect analysis device, defect analysis method, and program configured as described above create multiple emphasized images in which multiple defects are emphasized, based on a trained model based on multiple standard ultrasonic images acquired from a standard test piece and multiple target ultrasonic images acquired from an inspection target. Also, the trained model is trained by prioritizing defects to be detected among multiple defects in each standard ultrasonic image. Therefore, according to the defect analysis device, defect analysis method, and program of the embodiment, defect analysis using an emphasized image in which defects to be detected are prioritized and emphasized is possible. Therefore, defects can be analyzed efficiently.

[0029] In addition, in the above embodiment, multiple target ultrasound images are created using multiple gates with different gate ranges and gate levels, so the probability of detecting defect echoes with an appropriate S / N ratio (signal-to-noise ratio) can be increased. In addition, in the above embodiment, waveform data using multiple gates with different gate ranges and gate levels is aggregated into two-dimensional data, so the processing time can be reduced compared to, for example, the case where all the waveform data is used as input information for a machine learning model.

[0030] In addition, in the above embodiment, a binarized integrated image can be created by binarizing each emphasized image, which can reduce erroneous determination of defects.

[0031] Furthermore, in the above embodiment, the original images of the multiple target ultrasound images and the binarized integrated image are superimposed, so that the original images of the target ultrasound images and the binarized integrated image can be easily compared.

[0032] Furthermore, in the above embodiment, since the inspection object 4 is a composite material, defects in the composite material are easily detected.

[0033] Furthermore, when generating the trained model 141 based on multiple standard ultrasonic images acquired from a standard test piece, the learning device 1 configured as described above trains the defects to be detected among multiple defects in each standard ultrasonic image by prioritizing them. Therefore, the learning device 1 enables defect analysis using a trained model in which defects to be detected are trained with priority. This allows defects to be analyzed efficiently.

[0034] (Other embodiments) Although the embodiments of the present disclosure have been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and design changes and the like that do not depart from the gist of the present disclosure are also included.

[0035] Computer Configuration FIG. 15 illustrates a computer configuration according to at least one embodiment. The computer 90 comprises a processor 91, a main memory 92, a storage 93, and an interface 94. The above-mentioned defect analysis device 1 is implemented in a computer 90. The operations of each of the above-mentioned processing units are stored in the form of a program in a storage 93. A processor 91 reads the program from the storage 93, loads it in a main memory 92, and executes the above-mentioned processing in accordance with the program. The processor 91 also secures storage areas in the main memory 92 corresponding to each of the above-mentioned storage units in accordance with the program.

[0036] The program may be for realizing a part of the functions to be performed by the computer 90. For example, the program may be for realizing the functions by combining with other programs already stored in the storage or by combining with other programs implemented in other devices. In another embodiment, the computer may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or instead of the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions to be realized by the processor may be realized by the integrated circuit.

[0037] Examples of the storage 93 include a hard disk drive (HDD), a solid state drive (SSD), a magnetic disk, a magneto-optical disk, a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), and a semiconductor memory. The storage 93 may be an internal medium directly connected to the bus of the computer 90, or an external medium connected to the computer 90 via an interface 94 or a communication line. In addition, when the program is distributed to the computer 90 via a communication line, the computer 90 that receives the program may load the program into the main memory 92 and execute the above-mentioned process. In at least one embodiment, the storage 93 is a non-transitory tangible storage medium.

[0038] <Additional Notes> The defect analysis device 1 described in the above embodiment can be understood, for example, as follows.

[0039] (1) The defect analysis device 1 according to the first embodiment includes a model storage unit 14 storing a trained model 141 in which teacher information representing the positions, shapes, and sizes of a plurality of defects built into a standard test piece 3 is associated with detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece 3, a target image acquisition unit 15 acquiring a plurality of target ultrasonic images from an inspection target 4, and an enhanced image creation unit 16 creating a plurality of enhanced images in which the plurality of defects are enhanced based on the trained model 141 and the plurality of target ultrasonic images, and the trained model learns by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. According to this embodiment, defect analysis using an enhanced image in which defects to be detected are preferentially emphasized is possible. Therefore, defects can be analyzed efficiently.

[0040] (2) A defect analysis apparatus 1 according to a second aspect is the defect analysis apparatus 1 of (1), in which the plurality of standard ultrasonic images are a plurality of C-Scan images.

[0041] (3) A defect analysis device 1 according to a third aspect is the defect analysis device 1 according to (1) or (2), in which the gate ranges and gate levels of the multiple C-Scan images are different from each other. According to this aspect, multiple target ultrasonic images are created using multiple gates with different gate ranges and gate levels, so that the probability of detecting defect echoes with an appropriate S / N ratio (signal-to-noise ratio) can be increased. Also, according to this aspect, waveform data using multiple gates with different gate ranges and gate levels is aggregated into two-dimensional data, so that processing time can be shortened.

[0042] (4) A defect analysis device 1 according to a fourth aspect is any one of the defect analysis devices 1 according to (1) to (3), in which the target ultrasonic images are C-Scan images having different gate ranges and gate levels, and the emphasized image creation unit 16 acquires a plurality of binary images obtained by binarizing the plurality of emphasized images, integrates the plurality of binary images, and creates a binary integrated image representing the plurality of defects. According to this aspect, a binary integrated image can be created from a plurality of binary images binarized according to each emphasized image. This makes it possible to suppress erroneous determination of defects.

[0043] (5) A defect analysis device 1 according to a fifth aspect is any one of the defect analysis devices 1 according to (1) to (4), and the enhanced image creation unit 16 creates a superimposed image by superimposing the original images of the plurality of target ultrasound images and the binary integrated image. According to this aspect, it is easy to compare the original images of the target ultrasound images with the binary integrated image.

[0044] (6) A defect analysis device 1 according to a sixth aspect is the defect analysis device 1 according to any one of (1) to (5), in which the inspection object 4 is a composite material. According to this aspect, defects in the composite material can be easily detected.

[0045] (7) A defect analysis method according to a seventh aspect includes a step of acquiring a plurality of target ultrasonic images from an inspection target 4, and a step of creating a plurality of enhanced images in which the plurality of defects are enhanced based on a trained model 141 in which teacher information representing the positions, shapes, and sizes of a plurality of defects embedded in a standard test piece 3 and detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece 3 are associated with each other, and the plurality of target ultrasonic images, and the trained model learns by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. According to this aspect, defect analysis using an enhanced image in which defects to be detected are preferentially emphasized becomes possible. Therefore, defects can be analyzed efficiently.

[0046] (8) A program according to an eighth aspect causes a computer to execute the steps of acquiring a plurality of target ultrasonic images from an inspection target 4, and creating a plurality of enhanced images in which the plurality of defects are enhanced based on a trained model 141 in which teacher information representing the positions, shapes, and sizes of a plurality of defects embedded in a standard test piece 3 and detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece 3 are associated with each other, and the plurality of target ultrasonic images, and the trained model learns by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. According to this aspect, defect analysis using an enhanced image in which defects to be detected are preferentially emphasized is possible. Therefore, defects can be analyzed efficiently.

[0047] (9) A learning device 1 according to a ninth aspect includes a machine learning unit 13 that generates a trained model 141 by learning based on teacher information representing the positions, shapes, and sizes of a plurality of defects embedded in a standard test piece 3 and detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece 3, and the trained model 141 trains by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. According to this aspect, defect analysis can be performed using a trained model in which defects to be detected are trained with priority. This allows defects to be analyzed efficiently. [Explanation of symbols]

[0048] 1...Defect analysis device (learning device) 2...Ultrasonic waveform acquisition device 3…Standard test piece 4. Subject of inspection 11…Standard image acquisition section 12…Teacher information input section 13…Machine Learning Department 14…Model storage area 15…Target image acquisition section 16…Enhanced image creation section 141…Trained model

Claims

1. a model storage unit that stores a trained model in which teacher information representing the positions, shapes, and sizes of a plurality of defects embedded in a standard test piece and detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece are associated with each other; a target image acquisition unit for acquiring a plurality of target ultrasound images from an examination target; An enhanced image creation unit that creates a plurality of enhanced images in which the plurality of defects are enhanced based on the trained model and the plurality of target ultrasound images, The trained model trains by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. Defect analysis equipment.

2. The plurality of standard ultrasound images are a plurality of C-Scan images. The defect analysis device according to claim 1 .

3. The gate ranges and gate levels of the plurality of C-Scan images are different from each other. The defect analysis device according to claim 2 .

4. The plurality of target ultrasound images are a plurality of C-Scan images having different gate ranges and gate levels, The enhanced image creation unit obtains a plurality of binary images by binarizing the plurality of enhanced images, integrates the plurality of binary images, and creates a binary integrated image representing the plurality of defects. The defect analysis device according to claim 3 .

5. The emphasized image creating unit creates a superimposed image by superimposing the original images of the plurality of target ultrasound images and the binarized integrated image. The defect analysis device according to claim 4.

6. The test object is a composite material. The defect analysis device according to claim 5 .

7. acquiring a plurality of target ultrasound images from a subject; and creating a plurality of enhanced images in which the plurality of defects are enhanced based on the plurality of target ultrasonic images and a trained model in which teacher information representing the positions, shapes, and sizes of a plurality of defects embedded in a standard test piece and detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece are associated with each other, The trained model trains by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. Defect analysis methods.

8. acquiring a plurality of target ultrasound images from a subject; A step of creating a plurality of enhanced images in which the plurality of defects are enhanced based on a trained model in which teacher information representing the positions, shapes, and sizes of a plurality of defects embedded in a standard test piece and detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece are associated with each other, and the plurality of target ultrasonic images; Run the following on your computer: The trained model trains by prioritizing defects to be detected among the plurality of defects in each standard ultrasonic image. program.

9. a machine learning unit that generates a trained model by learning based on teacher information that indicates positions, shapes, and sizes of a plurality of defects embedded in a standard test piece and detection information of the plurality of defects that are detected based on a plurality of standard ultrasonic images acquired from the standard test piece; The trained model trains by prioritizing defects that are highlighted based on brightness values ​​among the plurality of defects in each standard ultrasound image as defects to be detected. Learning device.

Images