Information processing device, determination method, and computer-readable recording medium

Abstract

The determination results of each determination unit are considered according to the object data and a final determination result is derived. The information processing device (1) comprises: a reliability determination unit (103) that determines the reliability of each determination result of the determination units (102A to 102C) that determines a predetermined determination item based on the inspection image based on the inspection image; and a comprehensive determination unit (104) that determines the predetermined determination item using the respective determination results and the reliability.

Description

Technical Field

[0001] The present invention relates to an information processing device and the like that determines a predetermined determination item based on object data. Background Art

[0002] Patent Document 1 below discloses a technique for determining the presence and type of defects based on images of semiconductor substrates. Specifically, the technique in Patent Document 1 determines the final classification result based on the sum of the classification results of multiple classifiers multiplied by predetermined weights for each classifier. This improves classification accuracy compared to using only a single classifier.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Publication No. 2016-40650 Summary of the Invention

[0006] (1) Technical issues to be resolved

[0007] However, when detecting and classifying defects based on multiple images, the accuracy of each classifier can vary depending on the image due to various differences. Therefore, pre-determined weights are not always optimal. Furthermore, these non-optimal weights can affect the final accuracy of the determination.

[0008] For example, when using two classifiers, A and B, for one image, the classification of classifier A may be correct while the classification of classifier B may be incorrect, but the opposite may be true for another image. In this case, if the weight of classifier A is set to be larger than that of classifier B, the final classification result for one image may be correct while the final classification result for the other image may be incorrect.

[0009] This problem is not limited to classification using multiple classifiers, but is a common problem in situations where a final determination result is derived based on the results of determinations made by multiple determination units on predetermined determination items. Furthermore, this problem is not limited to determinations using images, but is a common problem in determinations based on any object data.

[0010] One aspect of the present invention has been made in view of the above-mentioned problem, and an object thereof is to realize an information processing device or the like that can appropriately consider the determination results of each determination unit according to target data and derive a final determination result.

[0011] (2) Technical solution

[0012] In order to solve the above-mentioned problem, an information processing device according to one embodiment of the present invention comprises: a reliability determination unit, which performs the following processing on a plurality of determination units respectively, namely: determining an indicator representing the possibility of a determination result of the determination unit, i.e., reliability, based on object data, and the determination unit determines a prescribed determination item based on one of the object data; and a comprehensive determination unit, which uses each of the determination results and the reliability determined by the reliability determination unit to determine the prescribed determination item.

[0013] In addition, in order to solve the above-mentioned problem, a judgment method of one embodiment of the present invention uses one or more information processing devices and includes: a reliability judgment step, performing the following processing on multiple judgment parts respectively, namely: judging the reliability, an indicator representing the possibility of the judgment result of the judgment part based on the object data, and the judgment part judges the prescribed judgment item based on one of the object data; and a comprehensive judgment step, using each of the judgment results and the reliability judged in the reliability judgment step to judge the prescribed judgment item.

[0014] (3) Beneficial effects

[0015] According to one aspect of the present invention, it is possible to derive a final determination result by appropriately considering the determination result of each determination unit according to the target data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a block diagram showing an example of the configuration of main parts of the information processing device according to the first embodiment of the present invention.

[0017] Figure 2 This is a diagram schematically showing an inspection system including the above-mentioned information processing device.

[0018] Figure 3 This is a diagram schematically showing an inspection performed using the above-mentioned information processing device.

[0019] Figure 4 This is a diagram showing a configuration example of a determination unit that performs determination using a generated model, and an example of a determination method for determining the presence or absence of a defect using the determination unit.

[0020] Figure 5 This figure shows an example of generating a heat map from an ultrasonic image and performing threshold processing on the generated heat map.

[0021] Figure 6 This diagram illustrates the relationship between the defect position, ultrasonic image, and thermal map.

[0022] Figure 7 A diagram illustrating a method for detecting a defective area.

[0023] Figure 8This is a diagram showing an example of areas set according to defect types.

[0024] Figure 9 This is a diagram explaining a method of detecting defects appearing in a plurality of ultrasonic images by integrating them into one defect.

[0025] Figure 10 This is a diagram explaining a method for calculating the thickness of a pipe end weld.

[0026] Figure 11 This is a diagram showing an output example of the inspection result.

[0027] Figure 12 This is a diagram showing an example of processing for constructing various models used in inspection and determining threshold values.

[0028] Figure 13 This is a diagram showing an example of an inspection method using the above-mentioned information processing device.

[0029] Figure 14 This is a flowchart showing an example of defect type determination processing for determining the defect type using a type determination model.

[0030] Figure 15 This is a flowchart showing an example of defect type determination processing for determining the defect type based on the position of the defect area. DETAILED DESCRIPTION

[0031] (System Overview)

[0032] based on Figure 2 An overview of an inspection system according to an embodiment of the present invention will be described. Figure 2 1 is a diagram schematically showing an inspection system 100. The inspection system 100 is a system for inspecting the presence or absence of defects in an inspection object based on an image of the inspection object, and includes an information processing device 1 and an ultrasonic flaw detection device 7.

[0033] In this embodiment, an example of using the inspection system 100 to inspect the tube end welds of a heat exchanger for defects is described. In addition, the tube end weld refers to a portion where a plurality of metal tubes constituting the heat exchanger and a metal tube sheet that bundles these tubes are welded. In addition, a defect in the tube end weld refers to a defect in which a void is generated inside the tube end weld. In addition, the tubes and tube sheets can be made of non-ferrous metals such as aluminum, or can be made of resin. In addition, the inspection system 100 can also be used to inspect the welds (root welds) between the tube seats and tubes of boiler equipment used in, for example, waste incineration facilities for defects. Of course, the inspection location is not limited to the weld, and the inspection object is not limited to the heat exchanger.

[0034] During the inspection, if Figure 2As shown, a probe coated with a contact medium is inserted from the pipe end. Ultrasonic waves are propagated through the probe from the inner wall of the pipe toward the pipe end weld, and the ultrasonic echoes are measured. If a defect, such as a void, occurs within the pipe end weld, the echo from this void can be measured, allowing the defect to be detected using this echo.

[0035] For example, in Figure 2 In the enlarged view of the probe periphery shown on the lower left, the ultrasonic wave, indicated by arrow L3, propagates toward a portion of the pipe end weld where there is no void. Therefore, the echo of the ultrasonic wave indicated by arrow L3 is not measured. On the other hand, the ultrasonic wave, indicated by arrow L2, propagates toward a portion of the pipe end weld where there is a void, so the echo of the ultrasonic wave reflected from this void is measured.

[0036] Furthermore, since ultrasonic waves are also reflected by the periphery of the pipe end weld, the echoes of the ultrasonic waves propagating toward the periphery are also measured. For example, the ultrasonic wave indicated by arrow L1 propagates toward the pipe end relative to the pipe end weld and, therefore, does not reach the pipe end weld but instead reflects off the pipe surface on the pipe end side of the pipe end weld. Therefore, the echo from the pipe surface indicated by arrow L1 is measured. Furthermore, the ultrasonic wave indicated by arrow L4 reflects off the pipe surface on the deeper side of the pipe end weld, and this echo is therefore measured.

[0037] Since the pipe end welds are present 360 degrees around the pipe, the probe is rotated in units of a specified angle (e.g., 1 degree) and repeated measurements are performed. Furthermore, data representing the probe measurement results is transmitted to the ultrasonic flaw detection device 7. For example, the probe may be an array probe composed of multiple array elements. If it is an array probe, it is possible to efficiently inspect pipe end welds having a width in the direction in which the pipe extends by configuring the array elements so that the arrangement direction is consistent with the extension direction of the pipe. Furthermore, the array probe may be a matrix array probe having multiple array elements arranged in the vertical and horizontal directions.

[0038] The ultrasonic flaw detection device 7 uses data representing the measurement results of the probe to generate an ultrasonic image by imaging the echoes of ultrasonic waves propagating toward the pipe and the pipe end weld. Figure 2 , which is an example of an ultrasonic image generated by the ultrasonic flaw detection device 7, is shown in FIG. Alternatively, the information processing device 1 may generate the ultrasonic image 111. In this case, the ultrasonic flaw detection device 7 transmits data indicating probe measurement results to the information processing device 1.

[0039] The measured echo intensity is represented as the pixel value of each pixel in the ultrasonic image 111. The image area of ​​the ultrasonic image 111 can be divided into a pipe area ar1 corresponding to the pipe, a weld area ar2 corresponding to the pipe end weld, and peripheral echo areas ar3 and ar4 where echoes from around the pipe end weld appear.

[0040] As described above, the ultrasonic wave propagating from the probe in the direction indicated by arrow L1 reflects off the pipe surface at the pipe end weld end. Furthermore, the ultrasonic wave reflects off the pipe's inner surface, with these reflections repeating. Consequently, repeated echoes a1 to a4 appear in the peripheral echo region ar3 along arrow L1 in ultrasonic image 111. Furthermore, the ultrasonic wave propagating from the probe in the direction indicated by arrow L4 also reflects repeatedly off both the pipe's outer and inner surfaces. Consequently, repeated echoes a6 to a9 appear in the peripheral echo region ar4 along arrow L4 in ultrasonic image 111. These echoes appearing in peripheral echo regions ar3 and ar4 are also referred to as bottom wall echoes.

[0041] Since the ultrasonic wave propagating from the probe in the direction indicated by arrow L3 is not reflected, no echo appears in the area along arrow L3 in ultrasonic image 111. On the other hand, since the ultrasonic wave propagating from the probe in the direction indicated by arrow L2 is reflected by the gap, i.e., the defect, within the weld of the pipe end, echo a5 appears in the area along arrow L2 in ultrasonic image 111.

[0042] The information processing device 1 analyzes the ultrasonic image 111 to check whether there are defects in the pipe end welds, as will be described in detail later. In addition, when the information processing device 1 determines that there are defects, it also automatically determines the type of the defect.

[0043] (Structure of Information Processing Device)

[0044] based on Figure 1 The configuration of the information processing device 1 will be described. Figure 1 1 is a block diagram showing an example of the structure of the main parts of the information processing device 1. Figure 1 As shown, the information processing device 1 includes a control unit 10 that collectively controls various units of the information processing device 1, and a storage unit 11 that stores various data used by the information processing device 1. Furthermore, the information processing device 1 includes an input unit 12 that receives input operations to the information processing device 1, and an output unit 13 for outputting data from the information processing device 1.

[0045] The control unit 10 includes an inspection image generating unit 101, a determination unit 102A, a determination unit 102B, a determination unit 102C, a reliability determination unit 103, a comprehensive determination unit 104, a heat map generating unit 105, a defect type determination unit 106, a thickness calculation unit 107, a comprehensive detection unit 108, and a defect length calculation unit 109. The storage unit 11 stores an ultrasonic image 111 and inspection result data 112. Hereinafter, when it is not necessary to distinguish between the determination unit 102A, the determination unit 102B, and the determination unit 102C, they are simply referred to as the determination unit 102.

[0046] The inspection image generating unit 101 cuts out an inspection target region from the ultrasonic image 111 and generates an inspection image for determining the presence or absence of defects in the inspection target object. The method of generating the inspection image will be described later.

[0047] The determination unit 102 determines a predetermined determination item based on the target data. In this embodiment, the following example is described: the inspection image generated by the inspection image generation unit 101 is the target data, and the presence or absence of welding defects in the welded portions of the heat exchanger tube ends shown in the inspection image is the predetermined determination item. Hereinafter, welding defects may be referred to simply as defects.

[0048] In addition, as for the definition of "defects" to be determined, it is sufficient to predetermine it according to the purpose of the inspection, etc. For example, if the quality inspection is for the pipe end weld of a manufactured heat exchanger, a "defect" can be defined as: the following echo is reflected in the inspection image, and the echo is caused by the internal gap of the pipe end weld or the unacceptable depression on the surface of the pipe end weld. Such depressions are caused by burn-through, for example. In other words, the presence or absence of defects means: whether there are parts that are different from normal products (abnormal parts). In addition, in the field of non-destructive testing, abnormal parts detected by ultrasonic waveforms or ultrasonic images are usually called "damage". This "damage" is also included in the category of the above-mentioned "defects". In addition, the above-mentioned "defects" also include defects or cracks.

[0049] The determination unit 102A, the determination unit 102B, and the determination unit 102C all determine the presence or absence of defects based on the inspection image generated by the inspection image generation unit 101 . However, as described below, their determination methods are different.

[0050] Determination unit 102A (generative model determination unit) uses a generated image generated by inputting the inspection image into the generative model to determine the presence of defects. Determination unit 102B (numerical analysis determination unit) analyzes the pixel values ​​of the inspection image to identify the inspection target area within the inspection image and determines the presence of defects based on the pixel values ​​of the identified inspection target area. Furthermore, determination unit 102C determines the presence of defects based on the output values ​​obtained by inputting the inspection image into the determination model. Details of the determinations performed by determination units 102A-102C and the various models used will be described later.

[0051] The reliability determination unit 103 determines the reliability, which is an index indicating the likelihood, of each determination result of the determination units 102A to 102C. The reliability determination is performed based on the inspection images used when the determination units 102A to 102C derive the determination results, as will be described in detail later.

[0052] Comprehensive determination unit 104 uses the determination results of determination units 102A-102C and the reliability determined by reliability determination unit 103 to determine the presence or absence of defects. This allows for a determination result that appropriately considers the determination results of determination units 102A-102C using the reliability corresponding to the inspection image. The determination method of comprehensive determination unit 104 will be described in detail later.

[0053] The heat map generating unit 105 generates a heat map using data obtained during the determination process of the determination unit 102A. The heat map is used for defect type determination by the defect type determination unit 106. The heat map will be described in detail later.

[0054] The defect type determination unit 106 determines the type of defect reflected in the inspection image determined as defective by the comprehensive determination unit 104. As described above, the heat map generated by the heat map generation unit 105 is used for type determination. The method for determining the defect type will be described later.

[0055] The thickness calculation unit 107 calculates the wall thickness of the pipe end weld. The wall thickness calculated by the thickness calculation unit 107 can be used as an indicator to determine whether the welding is performed properly. The wall thickness calculation method will be described later.

[0056] When the integrated determination unit 104 determines that a plurality of ultrasonic images 111, each corresponding to a portion of the inspection object, i.e., adjacent portions thereof, contain a defect, the integrated detection unit 108 detects the defects shown in the plurality of ultrasonic images 111 as a single defect. The integration of defects will be described in detail later.

[0057] The defect length calculation unit 109 calculates the length of the defect integrated by the integration detection unit 108. The method of calculating the defect length will be described later.

[0058] As described above, the ultrasonic image 111 is an image obtained by imaging the echoes of ultrasonic waves propagating toward the inspection object, and is generated by the ultrasonic flaw detection apparatus 7 .

[0059] Inspection result data 112 is data representing the defect inspection results of the information processing device 1. The inspection result data 112 records the defect presence / absence determination results of the integrated determination unit 104 on the ultrasonic image 111 stored in the storage unit 11. Furthermore, the inspection result data 112 records the defect type determination results of the defect type determination unit 106 on the ultrasonic image 111 determined to contain a defect. Furthermore, the inspection result data 112 records the defects integrated by the integrated detection unit 108, the integrated defect length calculated by the defect length calculation unit 109, and the wall thickness of the pipe end weld calculated by the thickness calculation unit 107.

[0060] (Overview of the inspection)

[0061] based on Figure 3 The inspection performed by the information processing device 1 will be briefly described. Figure 3 1 is a diagram schematically showing an inspection performed using the information processing device 1. Figure 3 2 shows a process after the ultrasonic image 111 generated by the ultrasonic flaw detection device 7 is stored in the storage unit 11 of the information processing device 1 .

[0062] First, the inspection image generation unit 101 extracts the inspection target region from the ultrasonic image 111 and generates the inspection image 111A. When extracting the inspection target region, an extraction model constructed by machine learning can be used. Figure 12 The extraction model will be described.

[0063] The inspection target area is an area sandwiched between two peripheral echo areas ar3 and ar4 where echoes from the peripheral portion of the inspection target portion of the inspection target object repeatedly appear. Figure 2 As shown, predetermined echoes (echoes a1 to a4 and a6 to a9) are repeatedly observed around the periphery of the inspection target site in ultrasonic image 111, caused by the shape of the periphery. Therefore, the region corresponding to the inspection target site in ultrasonic image 111 can be identified based on the positions of the peripheral echo regions ar3 and ar4, where these echoes repeatedly appear. Furthermore, the appearance of predetermined echoes around the periphery of the inspection target site is not limited to ultrasonic images 111 of pipe end welds. Therefore, the configuration of extracting the region surrounded by the peripheral echo region as the inspection target region can also be applied to inspections other than pipe end welds.

[0064] Next, the determination unit 102A, the determination unit 102B, and the determination unit 102C determine the presence or absence of a defect based on the inspection image 111A. The determination content will be described in detail later.

[0065] Furthermore, the reliability determination unit 103 is used to determine the reliability of each determination result of the determination unit 102A, the determination unit 102B, and the determination unit 102C. Specifically, the reliability of the determination result of the determination unit 102A is determined based on the following output value, which is the output value obtained by inputting the inspection image 111A into the reliability prediction model used by the determination unit 102A. Similarly, the reliability of the determination result of the determination unit 102B is determined based on the following output value, which is the output value obtained by inputting the inspection image 111A into the reliability prediction model used by the determination unit 102B. In addition, the reliability of the determination result of the determination unit 102C is determined based on the following output value, which is the output value obtained by inputting the inspection image 111A into the reliability prediction model used by the determination unit 102C.

[0066] Furthermore, comprehensive determination unit 104 uses the determination results of determination units 102A, 102B, and 102C, and the reliability of these determination results determined by reliability determination unit 103, to comprehensively determine the presence or absence of defects and outputs the result of this comprehensive determination. This result is added to inspection result data 112. Furthermore, comprehensive determination unit 104 may cause output unit 13 to output the result of the comprehensive determination.

[0067] In comprehensive determination, the determination results of determination unit 102 can be expressed numerically, and the reliability determined by reliability determination unit 103 can be used as a weight. For example, if determination units 102A, 102B, and 102C determine that a defect is present, they output a "1" as the determination result, and if they determine that there is no defect, they output a "-1" as the determination result. Furthermore, reliability determination unit 103 is configured to output a reliability value ranging from 0 to 1 (the closer to 1, the higher the reliability).

[0068] In this case, comprehensive determination unit 104 may calculate a total value obtained by adding the value obtained by multiplying the numerical values ​​of "1" or "-1" output by determination unit 102A, determination unit 102B, and determination unit 102C by the reliability output by reliability determination unit 103. Comprehensive determination unit 104 may then determine the presence or absence of a defect based on whether the calculated total value is greater than a predetermined threshold value.

[0069] For example, the threshold is set to "0," a value intermediate between "1" indicating a defect and "-1" indicating no defect. Furthermore, the output values ​​of determination units 102A, 102B, and 102C are set to "1," "-1," and "1," respectively, and their reliabilities are set to "0.87," "0.51," and "0.95," respectively.

[0070] In this case, the comprehensive determination unit 104 calculates 1×0.87+(−1)×0.51+1×0.95. The result of this calculation is 1.31, which is larger than the threshold value “0”, so the comprehensive determination result of the comprehensive determination unit 104 is that there is a defect.

[0071] (Reliability Correction)

[0072] Experience has shown that echoes caused by welding defects tend to appear upward from the center of the image area of ​​inspection image 111A. Therefore, when determination unit 102 determines that a defect is present, if the echo caused by the welding defect appears upward from the center of the image area of ​​inspection image 111A, the probability that the determination result is correct is high.

[0073] Therefore, within the image area of ​​inspection image 111A, a region where echoes due to welding defects appear frequently can be pre-defined. Furthermore, when determination unit 102 determines that a defect exists, reliability determination unit 103 can increase the reliability of the determination result by detecting echoes due to welding defects within this region. In this way, by correcting the reliability based on the tendency or characteristics of defects, it is possible to achieve a more appropriate reliability.

[0074] For example, the region above the center of the image region of inspection image 111A may be defined as the aforementioned region. Furthermore, when the reliability determination unit 103 detects an echo due to a welding defect within this region, it may add a predetermined constant to the reliability calculated using the reliability prediction model. Furthermore, when the position at which the echo due to a welding defect is detected is outside the aforementioned region, the reliability determination unit 103 may subtract the predetermined constant from the reliability calculated using the reliability prediction model.

[0075] However, when a constant is added to the reliability, it is preferable that the reliability after the addition does not exceed 1. Furthermore, when a constant is subtracted from the reliability, it is preferable that the reliability after the subtraction does not become less than 0.

[0076] Of course, the reliability correction method is not limited to the examples above. For example, the image area of ​​inspection image 111A can be further subdivided. If the location of the echo caused by the welding defect is within an area with a high frequency of welding defects, the value added to the reliability can be increased. Alternatively, for example, a value proportional to the distance from the detected echo location to the location with the highest frequency of welding defects can be added to the reliability, or a value inversely proportional to the distance can be subtracted from the reliability.

[0077] Furthermore, the reliability can be adjusted based on factors other than position. For example, even if a defect is determined and an echo caused by a welding defect is detected, if the pixel value of the echo is small, there is doubt that the determination result is incorrect. Therefore, the reliability can be adjusted to a smaller value as the pixel value of the echo caused by the welding defect is smaller, and the reliability can be adjusted to a larger value as the pixel value of the echo caused by the welding defect is larger.

[0078] The reliability correction described above is well-suited for the determination results obtained by determination units 102A and 102B. This is because the difference image calculated during determination unit 102A can be used to calculate the position or pixel value of an echo caused by a welding defect. Furthermore, during determination unit 102B, an echo caused by a welding defect is detected, and this detection result can be utilized.

[0079] (Determination by Determination Unit 102A)

[0080] As described above, determination unit 102A uses a generated image generated by inputting an inspection image into a generative model to determine the presence or absence of defects. This generative model is constructed to generate a new image with the same features as the input image through machine learning, using images of defect-free inspection objects as training data. The term "feature" refers to any information derived from an image, including, for example, the distribution and variance of pixel values ​​within the image.

[0081] The generative model is constructed through machine learning using images of defect-free inspection objects as training data. Therefore, when an image of a defect-free inspection object is input into the generative model as an inspection image, there is a high probability that a new image with the same characteristics as the inspection image will be output as the generated image.

[0082] On the other hand, when an image of a defective inspection object is input into the generation model as an inspection image, even if a defect of any shape and size is reflected at any position in the inspection image, the generated image is likely to have characteristics different from the inspection image.

[0083] Thus, a difference occurs in whether the target image input to the generation model is accurately restored between the generated image generated from the inspection image showing the defect and the generated image generated from the inspection image not showing the defect.

[0084] Therefore, the information processing device 1 can be used to determine with high precision whether there are defects with non-fixed position, size and shape. The information processing device 1 takes into account the judgment results of the judgment unit 102A to make a comprehensive judgment. The judgment unit 102A uses the generated image generated by the above-mentioned generation model to determine whether there are defects.

[0085] The following is based on Figure 4 The determination performed by the determination unit 102A will be described in detail. Figure 4 FIG. 1 shows an example of a configuration of the determination unit 102A and an example of a method for determining the presence or absence of a defect by the determination unit 102A. Figure 4 As shown, the determination unit 102A includes an inspection image acquisition unit 1021 , a restored image generation unit 1022 , and a defect presence determination unit 1023 .

[0086] The inspection image acquisition unit 1021 acquires an inspection image. As described above, the information processing device 1 includes the inspection image generation unit 101, so the inspection image acquisition unit 1021 acquires the inspection image generated by the inspection image generation unit 101. Alternatively, the inspection image may be generated by another device. In this case, the inspection image acquisition unit 1021 acquires the inspection image generated by the other device.

[0087] The restored image generator 1022 inputs the inspection image acquired by the inspection image acquisition unit 1021 into a generative model, thereby generating a new image with the same characteristics as the input inspection image. Hereinafter, the image generated by the restored image generator 1022 is referred to as the restored image. The generative model used to generate the restored image, also known as an autoencoder, is constructed through machine learning using images of defect-free inspection objects as training data, as described in detail below.

[0088] The defect determination unit 1023 determines whether the inspection object has defects using the restored image generated by the restored image generation unit 1022. Specifically, the defect determination unit 1023 determines that the inspection object has defects when the variance of the difference between each pixel in the inspection image and the restored image exceeds a predetermined threshold.

[0089] In the defect presence determination method using the determination unit 102A having the above configuration, the inspection image acquisition unit 1021 first acquires the inspection image 111A. The inspection image acquisition unit 1021 then transmits the acquired inspection image 111A to the restored image generation unit 1022. As described above, the inspection image 111A is generated by the inspection image generation unit 101 based on the ultrasonic image 111.

[0090] Next, the restored image generation unit 1022 inputs the inspection image 111A into the generation model and generates the restored image 111B based on the output value thereof.

[0091] The inspection image acquisition unit 1021 then removes the peripheral echo region from the inspection image 111A to generate a removed image 111C, and removes the peripheral echo region from the restored image 111B to generate a removed image (restored) 111D. Furthermore, if the inspection object is the same, the position and size of the peripheral echo region shown in the inspection image 111A are substantially constant. Therefore, the inspection image acquisition unit 1021 can remove a predetermined range in the inspection image 111A as the peripheral echo region. Furthermore, the inspection image acquisition unit 1021 can analyze the inspection image 111A to detect the peripheral echo region and remove the peripheral echo region based on the detection result.

[0092] By removing the peripheral echo region in the above manner, the defect determination unit 1023 determines the presence of defects in the remaining image region after removing the peripheral echo region from the restored image 111B. This prevents the influence of the peripheral echo region, allowing defect determination to be performed with improved accuracy.

[0093] Next, the defect determination unit 1023 determines the presence or absence of defects. Specifically, the defect determination unit 1023 first calculates the difference between the removed image 111C and the (restored) removed image 111D on a pixel-by-pixel basis. Next, the defect determination unit 1023 calculates the variance of the calculated difference. The defect determination unit 1023 then determines the presence or absence of defects based on whether the calculated variance exceeds a predetermined threshold.

[0094] Here, the difference value calculated for pixels reflecting the echo caused by the defect is larger than the difference values ​​calculated for other pixels. Therefore, the variance of the difference value calculated between the removed image 111C and the removed (restored) image 111D based on the inspection image 111A reflecting the echo caused by the defect increases.

[0095] On the other hand, the variance of the difference values ​​for subtracted image 111C and subtracted (restored) image 111D based on inspection image 111A, which does not reflect the echo due to the defect, is relatively small. This is because, when the echo due to the defect is not reflected, there may be areas where the pixel values ​​are relatively high due to the influence of noise, etc., but the probability of areas with extremely high pixel values ​​is low.

[0096] Thus, when the inspection object has a defect, the variance of the difference value increases, which is a characteristic phenomenon. Therefore, if the following structure is adopted, it is possible to appropriately determine whether there is a defect, that is, when the variance of the difference value exceeds a predetermined threshold, the defect determination unit 1023 determines that there is a defect.

[0097] Here, for inspection image 111A determined to have a defect, the defect type determination unit 106 determines the defect type based on the difference value of each pixel calculated by the defect presence determination unit 1023. Since the difference value of each pixel represents the difference between removed image 111C and removed image (restored) 111D, these difference values ​​are also referred to as a difference image.

[0098] The timing for removing the peripheral echo region is not limited to the above-mentioned example. For example, a difference image between the inspection image 111A and the restored image 111B may be generated, and the peripheral echo region may be removed from the difference image.

[0099] (Determination by Determination Unit 102B)

[0100] As described above, the determination unit 102B analyzes each pixel value of the inspection image, which is an image of the inspection object, to identify the inspection target portion in the inspection image, and determines the presence or absence of a defect based on the pixel value of the identified inspection target portion.

[0101] Conventional inspections using images involve inspectors visually identifying the inspection target area in the image and confirming whether the identified area contains defects such as damage or gaps that were not designed to be present. Automation is required for this visual inspection to reduce labor costs and stabilize accuracy.

[0102] Determination unit 102B analyzes the pixel values ​​of each image to identify the inspection target area and determines the presence of defects based on the pixel values ​​of the identified inspection target area. This allows for the automation of the aforementioned visual inspection. Furthermore, information processing device 1 makes its determination by comprehensively considering the determination results of determination unit 102B and those of other determination units 102, enabling highly accurate determination of the presence of defects.

[0103] The following describes in more detail the processing (numerical analysis) performed by the determination unit 102B. First, the determination unit 102B identifies two peripheral echo regions ( Figure 2 The region sandwiched by the peripheral echo regions ar3 and ar4 in the example is determined as the inspection target portion. The determination unit 102B then determines the presence or absence of a defect based on whether the determined inspection target portion includes a region (also referred to as a defect region) consisting of pixel values ​​exceeding a threshold value.

[0104] When detecting the peripheral echo region and the defect region, the determination unit 102B may first binarize the inspection image 111A using a predetermined threshold value to generate a binary image. Then, the determination unit 102B detects the peripheral echo region based on the binary image. For example, Figure 3 Echoes a1, a2, a6, and a7 are shown in the inspection image 111A shown. By binarizing the inspection image 111A using a threshold that distinguishes these echoes from noise components, the determination unit 102B can detect these echoes from the binarized image. Furthermore, the determination unit 102B can detect the ends of these detected echoes and identify the area surrounded by these ends as the inspection target site.

[0105] More specifically, the determination unit 102B determines the right end of echo a1 or a2 as the left end of the examination target site, and the left end of echo a6 or a7 as the right end of the examination target site. These ends constitute the boundary between the peripheral echo regions ar3 and ar4 and the examination target site. Similarly, the determination unit 102B determines the upper end of echo a1 or a6 as the upper end of the examination target site, and the lower end of echo a2 or a7 as the lower end of the examination target site.

[0106] In addition, if Figure 2 As shown in the ultrasonic image 111 , since the echo due to the defect appears above the echoes a1 and a6 , the determination unit 102B may set the upper end of the inspection target site above the position of the upper end of the echo a1 or a6 .

[0107] Furthermore, the determination unit 102B analyzes the inspection target portion identified in the binary image to determine whether an echo due to a defect is reflected. For example, if a continuous region consisting of a predetermined number of pixels or more exists in the inspection target portion, the determination unit 102B may determine that an echo due to a defect is reflected at the location of the continuous region.

[0108] The numerical analysis described above is merely an example, and the content of the numerical analysis is not limited to the examples described above. For example, if there is a substantial difference in the variance of the pixel values ​​of the inspection target area between the presence of a defect and the absence of a defect, the determination unit 102B may determine the presence of a defect based on the variance.

[0109] (Determination by Determination Unit 102C)

[0110] As described above, the determination unit 102C determines the presence or absence of defects based on the output value obtained by inputting the inspection image into the determination model. This determination model is constructed, for example, by performing machine learning using training data generated using ultrasonic images 111 of inspection objects with defects and training data generated using ultrasonic images 111 of inspection objects without defects.

[0111] The above-mentioned determination model can be constructed using any learning model suitable for image classification. For example, the determination model can be constructed using a convolutional neural network with excellent image classification accuracy.

[0112] (Heatmap and thresholding)

[0113] As mentioned above, heat maps are used to determine the defect type. Figure 5 The heat map generated by the heat map generating unit 105 and the threshold processing performed on the generated heat map will be described below. Figure 5 : is a diagram showing an example of generating a heat map from an ultrasonic image and performing threshold processing on the generated heat map. More specifically, Figure 5 The upper portion of FIG. 1 shows an example of an ultrasonic image 111 - a showing a defective portion in a pipe end weld. Figure 5 The lower portion of shows an example of an ultrasonic image 111 - b of a non-defective portion in the pipe end weld.

[0114] If based on Figure 4 As described, during the determination process by the determination unit 102A, the test image 111A and the restored image 111B are generated from the ultrasonic image 111. Furthermore, the subtracted image 111C is generated from the test image 111A, and the subtracted image (restored) 111D is generated from the restored image 111B.

[0115] exist Figure 5 In the example, a subtracted image 111C-a and a subtracted image (restored) 111D-a are generated from ultrasonic image 111-a. A difference image is generated from subtracted image 111C-a and subtracted image (restored) 111D-a. The heat map generation unit 105 generates a heat map that represents each pixel in the subtracted image using a color or shade corresponding to its pixel value.

[0116] Figure 5 Heat map 111E-a shows pixel values ​​from the lower limit to the upper limit using color gradations from black to white. As indicated by the hollow arrows in heat map 111E-a, the area corresponding to the defect (where pixels with large values ​​are concentrated) is where pixels close to white are concentrated. Therefore, the area corresponding to the defect is easy to visually identify in heat map 111E-a.

[0117] However, in heat map 111E-a, there are also areas where the pixel values ​​are increased due to noise, etc. Therefore, it is preferable that the heat map generation unit 105 performs threshold processing on the generated heat map to correct the pixel values ​​of the areas where the pixel values ​​are increased due to noise, etc. For example, the heat map generation unit 105 can set the pixel values ​​below the specified threshold in heat map 111E-a to zero (black). This generates heat map 111F-a with the noise component removed. Based on heat map 111F-a, the area corresponding to the defect can be more clearly identified.

[0118] Similarly, for ultrasonic image 111-b of a defect-free area, subtracted image 111C-b and subtracted image (restored) 111D-b are generated from ultrasonic image 111-b, and a difference image is generated from subtracted image 111C-b and subtracted image (restored) 111D-b. Furthermore, the heat map generation unit 105 generates a heat map 111E-b of this difference image and performs threshold processing on heat map 111E-b to generate heat map 111F-b. Comparing heat map 111F-a with heat map 111F-b shows that the presence or absence of a defect can be clearly determined. Furthermore, the location of the defect can be clearly determined in heat map 111F-a.

[0119] (Type of defect, ultrasonic image, thermal image)

[0120] Known defects in pipe end welds include, for example, poor initial penetration, poor fusion between weld beads, undercuts, and porosity. Poor initial penetration is a defect caused by a lack of fusion near the tube sheet, resulting in voids. Poor fusion between weld beads is a defect caused by a lack of fusion during multiple welds, resulting in voids. Undercuts are defects where the ends of weld beads appear concave, forming a groove. Porosity is a defect where spherical cavities form within the weld metal.

[0121] These defects occur at different locations. Therefore, the defect type can be determined based on the location of the defect-induced echo in ultrasonic image 111. Similarly, the defect type can be determined based on the location of the defect region in a heat map (preferably a thresholded heat map) generated based on ultrasonic image 111. Furthermore, as described above, a defect region is an area where the defect-induced echo appears and has larger pixel values ​​than other areas.

[0122] based on Figure 6 Determination of the defect type based on the position of the defect area will be described. Figure 6 This is a diagram that illustrates the relationship between the defect location, ultrasonic image, and thermal image. Figure 6 The left end of the first layer represents a cross-sectional surface of the pipe end weld where poor initial layer penetration occurs. Figure 6 The left side is the pipe end side, and the right side is the pipe depth side. Figure 6 The tube sheet is located below the outer surface of the tube. To determine the width of the welded portion of the tube end, place a ruler against the inner wall (inner surface) of the tube.

[0123] exist Figure 6 In the figure at the left end of the first layer, the dotted line indicates the tube sheet penetration zone produced during welding. The inverted triangle-shaped area to the left of the penetration zone is the weld metal, and the area combining these areas is the tube end weld. A void is present in the circled portion of this tube end weld. This void is located near the tube surface, close to the deep-side end of the tube end weld.

[0124] like Figure 6 As shown in the center of the first layer, an echo caused by the gap appears in the ultrasonic image 111-c at the location where the gap exists. Figure 6 As shown at the right end of the first layer of FIG, a region corresponding to the above-mentioned gap appears in the heat map 111F-c generated based on the ultrasonic image 111 - c as indicated by the hollow arrow in the figure.

[0125] exist Figure 6 The left end of the second layer shows a cross-section of the pipe end weld where poor fusion between weld beads occurs. A void has formed in the circled area. This void is located near the surface of the pipe, near the center of the pipe end weld in the thickness direction.

[0126] like Figure 6 As shown in the center of the second layer of FIG, an echo caused by the gap appears in the ultrasonic image 111-d at the location where the gap exists. Figure 6 As shown at the right end of the second layer, a region corresponding to the above-mentioned gap also appears in the heat map 111F-d generated based on the ultrasonic image 111-d as indicated by the hollow arrow in the figure. This region is located to the left of the first layer heat map 111F-c.

[0127] exist Figure 6 The left end of the third layer shows the pipe end weld where undercutting has occurred, as viewed from the pipe end. A void has formed in the circled area. This void is located near the pipe surface, at the end of the pipe end weld.

[0128] like Figure 6 As shown in the center of the third layer of FIG, an echo caused by the gap appears in the ultrasonic image 111-e of the part where the gap exists. Figure 6 As shown at the right end of the third layer, a region corresponding to the above-mentioned gap also appears in the heat map 111F-e generated based on the ultrasonic image 111-e as shown by the hollow arrow in the figure.

[0129] -d is located on the left in comparison.

[0130] exist Figure 6 The left end of the fourth layer shows a cross-section of the pipe end weld where a pore has formed. A void has formed in the circled area. This void is located closer to the inside of the pipe end weld than to the pipe surface, and its left-right position is near the center of the pipe end weld in the width direction.

[0131] like Figure 6 As shown in the center of the fourth layer of FIG, an echo caused by the gap appears in the ultrasonic image 111-f of the part where the gap exists. Figure 6 As shown at the right end of the fourth layer, a region corresponding to the above-mentioned gap also appears in the heat map 111F-f generated based on the ultrasonic image 111-f, as indicated by the hollow arrow in the figure. The horizontal position of this region is close to that of the second layer heat map 111F-d, and the vertical position is further downward.

[0132] As described above, there is a correlation between the defect type and the appearance of the heat map 111F. Therefore, based on this correlation, a type determination model can be constructed that determines the defect type based on the heat map 111F. Such a type determination model can be constructed through machine learning that uses a heat map of a difference image generated from an inspection image of an inspection object with a known defect type as training data. Furthermore, the defect type determination unit 106 can determine the defect type based on the output value obtained by inputting the heat map generated by the heat map generation unit 105 into such a determination model.

[0133] As described above, the heat map, which uses color or shading to represent the pixel values ​​of each pixel constituting the difference image, can reflect the differences in the types of defects reflected in the inspection image based on the difference image. Therefore, the above configuration can automatically determine the type of defect appropriately.

[0134] For example, a plurality of images generated based on the ultrasonic image 111 of the area where the initial layer poor penetration occurs may be prepared. Figure 6 Heat maps such as heat map 111F-c are used as training data. This allows the construction of a type determination model that outputs the probability that the defect type is poor initial layer penetration. Similarly, by using heat maps generated from ultrasonic images 111 of locations where other types of defects have occurred as training data for machine learning, a type determination model can be constructed that outputs the probabilities corresponding to various defects.

[0135] Therefore, the defect type determination unit 106 can determine the defect type based on the output value obtained by inputting the heat map into the type determination model. For example, the defect type determination unit 106 can determine that a defect of the type corresponding to the highest probability value among the various defect probability values ​​output from the type determination model has occurred.

[0136] (Another example of a method for determining defect types)

[0137] based on Figure 7 and Figure 8 Another example of a defect type determination method will be described. In the determination method described below, the defect type determination unit 106 detects a defect area from a difference image and determines the defect type of the defect area based on the position of the detected defect area in the image area of ​​the difference image.

[0138] based on Figure 7 A method for detecting defective areas will be described. Figure 7 is a diagram illustrating a method for detecting defective areas. Figure 7 , an example of detecting a defective area using a heat map is shown, but as described below, generating a heat map is not essential.

[0139] exist Figure 7 : a heat map 111E generated from an ultrasonic image 111 of a defective inspection object and a heat map 111F obtained by thresholding the heat map 111E. Figure 7 11F also shows an enlarged view of the upper left end portion of the heat map 111F, and in this enlarged view, the pixel value of each pixel of the heat map 111F is recorded.

[0140] In detecting a defective area, first, the defect type determination unit 106 detects the pixel with the largest pixel value in the heat map 111F. Figure 7 In this example, the maximum pixel value is 104, so this pixel is detected. Next, the defect type determination unit 106 detects pixels adjacent to the detected pixel whose pixel values ​​are equal to or greater than a predetermined threshold value.

[0141] The defect type determination unit 106 repeats this process until no adjacent pixels with pixel values ​​exceeding the threshold are detected. This allows the defect type determination unit 106 to detect a continuous region consisting of pixels with pixel values ​​exceeding the predetermined threshold as a defect region. Furthermore, the defect type determination unit 106 can detect the rectangular region ar5 containing the defect region detected in this manner as a defect region.

[0142] The above processing can be performed simply by having a difference image, data representing the difference between each pixel in inspection image 111A and restored image 111B. Specifically, the defect area can be detected by repeatedly detecting the pixel with the maximum pixel value in the difference image and then detecting adjacent pixels with pixel values ​​exceeding a predetermined threshold. Therefore, as described above, it is not necessary to generate heat maps 111E and 111F for defect area detection.

[0143] As described above, the defect type determination unit 106 detects an area in the difference image consisting of a plurality of pixels having pixel values ​​greater than a threshold as a defect area. In the difference image, the pixel values ​​of the pixels in the defect area are larger than the pixel values ​​of the pixels in the other areas. Therefore, this configuration enables automatic detection of appropriate defect areas.

[0144] If based on Figure 6 As described above, various types of defects are known for weld defects, including poor initial penetration and poor fusion between weld beads. These differences in defect type appear as positional differences in ultrasonic images. The defect type determination unit 106 determines the defect type of the detected defect region based on its position within the image area of ​​the difference image. This allows for automatic defect type determination.

[0145] For example, if regions corresponding to various types of defects are set in advance in the difference image, the defect type determination unit 106 can determine the defect type based on which region the defect region detected as described above is included.

[0146] Figure 8 is a diagram showing an example of areas set according to defect types. Figure 8 In the example of FIG11A , an area AR1 corresponding to the undercut is set in the upper left corner of the thermal map 111F, an area AR2 corresponding to poor fusion between weld beads is set in the upper center, and an area AR3 corresponding to poor initial layer penetration is set in the upper right corner. In addition, an area AR4 corresponding to pores is set slightly above the center. Such areas can be set in advance by analyzing differential images or thermal maps based on inspection images of various defective parts. Figure 8 In the example of FIG, since the defect region indicated by the hollow arrow is detected in the region AR3, the defect type determination unit 106 determines that the defect is a defect caused by poor primary layer penetration.

[0147] exist Figure 8 In the example shown in FIG, a portion of the region AR4 corresponding to the pore overlaps a portion of the regions AR1 to AR3. In this way, the region for determining the defect type may be set so as to partially overlap with other regions.

[0148] In this case, when a defective area is detected in an area where multiple areas overlap, the defect type determination unit 106 may output all types corresponding to these areas as the defect type determination result. For example, when a defective area is detected in the overlapping area of ​​areas AR1 and AR4, the defect type determination unit 106 may output both undercut and blowhole as the determination results.

[0149] Furthermore, the defect type determination unit 106 can limit the defect type determination results based on whether conditions unique to each defect type are met. For example, if the defect is characteristic in shape, a condition related to shape can be set, while if the defect is characteristic in size, a condition related to size can be set.

[0150] For example, a pore is a spherical cavity defect typically less than 2 mm in diameter. Therefore, if a single ultrasonic image 111 covers a width of approximately 1 mm across the inspection object, one pore is contained within approximately two to three ultrasonic images 111. Therefore, if a defect is detected consecutively in multiple ultrasonic images 111 corresponding to adjacent portions of the inspection object, and the number of ultrasonic images 111 is three or fewer, the defect is likely a pore. On the other hand, if the number of ultrasonic images 111 in which the defect is detected consecutively is four or more, the defect is more likely not a pore.

[0151] Therefore, when a defective area is detected in the overlapping area between area AR4 and other areas, the defect type determination unit 106 can determine the defect type as a pore under the condition that the number of ultrasonic images 111 in which defects are continuously detected is below a threshold value (for example, 3).

[0152] For example, in Figure 8 In this example, the defect area is detected in the overlapping area of ​​areas AR4 and AR2. In this case, the defect type determination unit 106 can determine the defect type as porosity when the number of ultrasonic images 111 in which defects are continuously detected is below a threshold value, and determine the defect type as poor fusion between weld beads when the number exceeds the threshold value.

[0153] Furthermore, as described above, since pores are spherical, when a single pore is detected across multiple ultrasonic images 111, the peak values ​​of the echoes caused by the pore often differ between the ultrasonic images 111. These differences in peak values ​​manifest as differences in pixel values ​​within the ultrasonic images 111. For example, suppose a single pore is detected across three ultrasonic images 111. In this case, if the peak value of the echo caused by the pore in the center ultrasonic image 111 of the three ultrasonic images 111 is set to 50%, the peak values ​​of the echoes caused by the pore in the ultrasonic images 111 before and after it will be 30% lower.

[0154] Therefore, when a defect area is detected in the overlapping area between area AR4 and another area, the defect type determination unit 106 can use the presence of a difference in the pixel values ​​of the defect area in each of the ultrasonic images 111 in which the defect is detected consecutively as a condition for determining the defect type to be a pore. For example, the defect type determination unit 106 can calculate the average value of the pixel values ​​of each pixel included in the defect area in each of the ultrasonic images 111 and determine that a difference exists when the difference between the average values ​​is greater than a threshold value.

[0155] The defect type determination unit 106 may perform both determination using the type determination model and determination based on the position of the defect region, or may perform only one of the two. Performing both determinations can improve the accuracy of the type determination result.

[0156] (Combination of defects)

[0157] The pipe end weld extends 360 degrees around the pipe. Therefore, as described above, the probe is rotated within the pipe by a predetermined angle to generate ultrasonic images 111 of each portion of the pipe end weld, and defects are detected based on each ultrasonic image 111. In this case, a single continuous defect may appear across multiple ultrasonic images, and while it is a single defect, it may be detected as multiple defects.

[0158] Therefore, the integrated detection unit 108 integrates the defects shown in the multiple ultrasonic images 111 and detects them as a single defect. More specifically, when the integrated determination unit 104 determines that multiple ultrasonic images 111 corresponding to portions of the pipe end weld, i.e., adjacent portions, are defects, the integrated detection unit 108 detects the defects shown in these multiple ultrasonic images 111 as a single defect. This enables appropriate detection based on the defect entity.

[0159] based on Figure 9 A comprehensive approach to defects is described. Figure 9 This figure explains a method of detecting defects that appear in a plurality of ultrasonic images 111 as a single defect. Figure 9 The upper left of the figure shows the cross section of the welded portion between the pipe and the pipe end. Figure 9 The lower left side of the figure shows the longitudinal section of the tube, the tube end weld, and the tube sheet.

[0160] exist Figure 9 In this example, a large weld defect occurs along the outer wall of a pipe. If the probe is rotated at a predetermined angle along the inner wall of the pipe and echoes are measured, the echo from the weld defect will be reflected in the measurement results of the area where the weld defect occurs. Figure 9As shown on the right side of , echoes due to the welding defect appear in ultrasonic images 111g to 111i generated based on the measurement results. Therefore, in the defect presence determination based on these ultrasonic images 111g to 111i, the comprehensive determination unit 104 determines that a defect exists.

[0161] Here, the ultrasonic images 111g to 111i correspond to adjacent portions of the pipe end welds, so the integrated detection unit 108 detects the defects shown in the ultrasonic images 111g to 111i determined as defects by the integrated determination unit 104 as one defect.

[0162] Furthermore, the integrated detection unit 108 can integrate these defects based on whether the defects detected from ultrasonic images 111g to 111i are located in the same or similar locations. Furthermore, as mentioned above, defects vary in location depending on their type. Therefore, the integrated detection unit 108 can integrate these defects based on the condition that the defects detected from ultrasonic images 111g to 111i are of the same type. This configuration improves the accuracy of defect integration.

[0163] The defect length calculator 109 calculates the length of the defects integrated by the above-described process. For example, the defect length calculator 109 may calculate the defect length by multiplying the average length of defects in one ultrasonic image 111 by the number of defects integrated by the integrated detector 108.

[0164] For example, for a pipe end weld formed 360 degrees around a pipe, the probe is moved along the inner wall of the pipe in 1-degree increments around the central axis of the pipe to perform 360 echo measurements. As a result, 360 ultrasonic images 111 are generated. In this case, the length of the defect shown in one ultrasonic image 111 is approximately (the outer diameter of the pipe) × π × 1 / 360. Therefore, when Figure 7 When the defects in the three ultrasonic images 111g to 111i are integrated in this manner, the defect length calculator 109 can calculate the length of the defect as (outer diameter of the tube)×π×3×1 / 360. Here, π is the ratio of pi.

[0165] (Calculation of the thickness of the pipe end weld)

[0166] based on Figure 10 The calculation method of the thickness (wall thickness) of the pipe end weld portion will be described. Figure 10 This is a diagram explaining the calculation method of the thickness of the pipe end weld. Figure 10 The lower side shows a longitudinal section of the pipe end weld portion, and the upper side shows an ultrasonic image 111 of the pipe end weld portion.

[0167] Figure 10 The thickness of the pipe end weld shown on the lower side including the penetration portion toward the tube sheet is X. Figure 2 As described above, the region of the pipe end weld shown in ultrasonic image 111 is sandwiched between two peripheral echo regions ar3 and ar4, where echoes from its peripheral edges repeatedly appear. Therefore, the thickness X of the pipe end weld can be calculated based on the distance Xi between the peripheral echo regions ar3 and ar4.

[0168] The distance Xi can be calculated by the thickness calculation unit 107 by analyzing the ultrasonic image 111 . However, since the determination unit 102B analyzes the ultrasonic image 111 , it is preferable to use the analysis result.

[0169] More specifically, based on Figure 3 As described above, the determination unit 102B detects peripheral echo regions ar3 and ar4 in the inspection image 111A generated from the ultrasonic image 111 and determines the presence or absence of defects in the area sandwiched between these regions. Therefore, based on the results of this detection by the determination unit 102B, the thickness calculation unit 107 can calculate the distance Xi from the right end of the peripheral echo region ar3 to the left end of the peripheral echo region ar4 detected by the determination unit 102B. Furthermore, if the scale of the inspection image 111A is determined in advance, the thickness calculation unit 107 can use this scale to calculate the thickness X of the pipe end weld.

[0170] As described above, the determination unit 102B detects the peripheral echo regions ar3 and ar4 in the process of determining the presence or absence of a defect. Therefore, the thickness calculation unit 107 can calculate the thickness of the inspection target portion using the detection results of the peripheral echo regions ar3 and ar4 by the determination unit 102B.

[0171] (Example of output of inspection results)

[0172] The result of the determination of whether the inspection object has defects by the information processing device 1 is outputted via the output unit 13. Figure 11 An example of inspection result output is described below. Figure 11 This is a diagram showing an output example of the inspection result.

[0173] exist Figure 11 The upper left of FIG shows a defect map 300. Defect map 300 shows a circled area 301 of a pipe end weld viewed from the pipe end side, with line segments 302 representing detected defects plotted. Defect map 300 makes it easy to identify the distribution of defects in the pipe end weld.

[0174] In addition, Figure 11 The tube sheet diagram 400 is shown in the upper right corner of FIG. The tube sheet diagram 400 schematically shows the Figure 2The heat exchanger having a plurality of tubes welded to a tube sheet is viewed from the tube end side. In the tube sheet diagram 400, a graph showing the defect inspection results of the tube end welds of each tube is drawn at the position of each tube, thereby showing the inspection results.

[0175] Specifically, the inspection results are plotted as white circles at the locations of tubes without detected defects, and as black circles at the locations of tubes with detected damage (defects). This makes it easy to identify the distribution of defective tube end welds. Furthermore, in tubesheet diagram 400, triangular marks are plotted at the locations of tubes not inspected, and square marks are plotted at the locations of tubes not subject to inspection. In this way, various information related to the inspection can also be included in tubesheet diagram 400.

[0176] In addition, Figure 11 The lower side shows an ultrasonic image group 500. The ultrasonic image group 500 includes three ultrasonic images (501 to 503). Ultrasonic image 501 is obtained by a sector scan on the tube end side, ultrasonic image 502 is obtained by a linear scan, and ultrasonic image 503 is obtained by a sector scan on the tube depth side.

[0177] Furthermore, a linear scan is a scan in a flaw detection direction perpendicular to the central axis of the pipe. The ultrasonic image 111 described above is also obtained through a linear scan. A pipe end side sector scan is a scan in which ultrasonic waves propagate in a flaw detection direction that is tilted from the direction perpendicular to the central axis of the pipe toward the pipe end. Furthermore, a pipe depth side sector scan is a scan in which ultrasonic waves propagate in a flaw detection direction that is tilted from the direction perpendicular to the central axis of the pipe toward the pipe end.

[0178] In these ultrasonic images, the reflected echoes corresponding to the detected defects are marked. By displaying the marked ultrasonic images as inspection results, the position of the defects can be easily identified.

[0179] Here, ultrasonic images 501-503 are all obtained by scanning the same location on the pipe end weld. However, due to the different detection directions, defects appear differently. Therefore, the inspection device 1 determines the presence of defects based on multiple ultrasonic images 111 with different detection directions. If a defect is determined in any one detection direction, the final determination result can be set as a defect, even if no defect is determined in other detection directions. This reduces the probability of missing defects. Alternatively, the inspection device 1 can use a composite image obtained by combining ultrasonic images obtained by linear scanning and ultrasonic images obtained by sector scanning as the basis for defect determination.

[0180] Furthermore, the information processing device 1 may output the entire defect map 300, tube sheet map 400, and ultrasonic image set 500 as inspection results, or may output only a portion thereof. Furthermore, the information processing device 1 may output information indicating the defect type determination result as an inspection result. Of course, these are merely examples, and the information processing device 1 may output the determination result in any format that allows a human to recognize the content.

[0181] (Processing flow before inspection)

[0182] Before using the information processing device 1 to perform defect inspection, it is necessary to build various models used in the inspection and determine the threshold value. Figure 12 The following describes the process of constructing various models used in inspection and determining threshold values. Figure 12 This is a diagram showing an example of processing for constructing various models used in an inspection and determining a threshold value. Note that these processes may be performed in the information processing device 1 or in another computer.

[0183] In S1, a smoothed ultrasonic image 111 is acquired. This ultrasonic image 111 includes images of an inspection object with defects and images of an inspection object without defects. Furthermore, the ultrasonic images 111 of the inspection object with defects are classified according to the type of defect.

[0184] Smoothing is a process that smoothes the changes in pixel values ​​between adjacent pixels. This smoothing process can be performed by the information processing device 1 or the ultrasonic flaw detection device 7. While smoothing is not essential, it is preferred because it facilitates the distinction between echoes caused by defects and noise components.

[0185] In S2, an extraction model is constructed. This extraction model is constructed through machine learning using training data created by associating extraction region information with the ultrasound image 111 as accurate data. The extraction region information indicates the region to be extracted from the ultrasound image 111, i.e., the examination target region. This extraction region information can be generated based on, for example, input from an operator inputting the region to be extracted when the ultrasound image 111 is displayed on a display device.

[0186] The extraction model can be constructed using any learning model suitable for extracting regions from an image. For example, the extraction model can be constructed using YOLO (You Only Look Once), which has excellent extraction accuracy and processing speed.

[0187] The area to be extracted only needs to include the pipe end weld area as the inspection target area. In addition, the area to be extracted preferably also includes at least a portion of the area where the echo from the peripheral portion is reflected. This is because: when there is no defect in the inspection target area, it is possible that no feature points that can be used for machine learning can be observed in the ultrasonic image 111. In this case, it is difficult to build an extraction model. For example, in Figure 2 In the ultrasonic image 111 shown, the region including a portion of echoes a1, a2, a6, and a7 can be selected as the region to be extracted. This allows the construction of an extraction model that extracts the region including the pipe end weld and the echoes from the peripheral portion.

[0188] In S3, a learning image is generated from the ultrasonic image 111 acquired in S1 using the extraction model constructed in S2. In S2, when an extraction model is constructed by machine learning using a region including an area where echoes from the peripheral portion are reflected as correct data, the inspection target region is extracted using this extraction model. Figure 2 As shown, the learning image has the characteristics that can be machine-learned, so it can automatically and accurately extract the inspection object part. In addition, since the learning image is generated using the same extraction model as the inspection image 111A, its appearance is the same as the inspection image 111A (the appearance of the inspection image 111A is referenced). Figure 3 ).

[0189] From S4 onward, using the learning images generated in S3, thresholds associated with each determination unit 102 are determined and a model is constructed. In S4, a generative model is constructed. This generative model is constructed using machine learning, using learning images generated from ultrasound images 111 of defect-free inspection objects as training data. As described above, the generative model can be an autoencoder. Alternatively, the generative model can be a model that is an improvement or modification of an autoencoder. For example, a variational autoencoder can be used as the generative model.

[0190] Furthermore, when an extraction model is constructed in S2 using the following machine learning, the training data used to construct the generative model also includes regions where echoes from the peripheral portion are reflected. The machine learning identifies regions where echoes from the peripheral portion are reflected as correct data. In an ultrasonic image 111 of a defect-free inspection object, the inspection target region contains no echoes, and although there are no feature points that require machine learning, a suitable generative model can be constructed using training data that includes regions where echoes from the peripheral portion are reflected.

[0191] In S5, the threshold for defect determination by the determination unit 102A is determined. Specifically, a test image is first input into the generative model constructed in S4 to generate a restored image. The test image is an image of the learning image generated in S3 that was not used to construct the generative model. The test images include images generated from ultrasonic images 111 of inspection objects without defects and images generated from ultrasonic images 111 of inspection objects with defects. Furthermore, the test images generated from ultrasonic images 111 of inspection objects with defects are classified according to the type of defect.

[0192] Next, the difference between the restored image generated as described above and the test image used as the basis for the restored image is calculated on a pixel-by-pixel basis, and the variance of the difference is calculated. Furthermore, a threshold value is determined so that the variance values ​​calculated for the multiple test images generated from ultrasonic images 111 of an inspection object without defects can be distinguished from the variance values ​​calculated for the multiple test images generated from ultrasonic images 111 of an inspection object with defects.

[0193] In S6, machine learning is used to construct a reliability prediction model for determination unit 102A. Determination unit 102A performs determinations using the generative model constructed in S4 and the threshold value determined in S5. This machine learning utilizes training data created by correlating the accuracy of the determination result of determination unit 102A based on the test image with the test image as correct data. The test image can be generated from an ultrasound image 111 whose presence or absence of defects is known.

[0194] When the reliability prediction model generated in this manner is input to the inspection image 111A, it outputs a value between 0 and 1 indicating the probability that the determination result of the determination unit 102A using the inspection image 111A is correct. Therefore, the reliability determination unit 103 can use the output value of the reliability prediction model as the reliability of the determination result of the determination unit 102A.

[0195] In S7, the heat maps generated from the test images of various defects are used as teacher data to build a category judgment model. Figure 6 As described above, features corresponding to the types of defects appear, so by using this heat map as teacher data for machine learning, a type determination model can be constructed.

[0196] The category determination model can be constructed using any learning model suitable for image classification. For example, a convolutional neural network with excellent image classification accuracy can be used to construct the category determination model.

[0197] In S8, the area for type determination is set. Specifically, first, from the heat map corresponding to each defect generated in S7, defect areas showing echoes caused by the defect are detected. Next, the area within the heat map image area where a defect of a certain type is detected is determined as the area where that defect type appears. This process is performed for each type to be determined.

[0198] Therefore, if Figure 8 As illustrated, the region where the defect occurs can be set according to the defect type.

[0199] Alternatively, either S7 or S8 may be omitted. If S7 is omitted, the defect type determination unit 106 determines the defect type based on the region set in S8. On the other hand, if S8 is omitted, the defect type determination unit 106 determines the defect type using the type determination model constructed in S7.

[0200] In S9, the learning image generated in S3 is used to determine the threshold value used for numerical analysis by the determination unit 102B. For example, when the determination unit 102B performs binarization processing, the threshold value used for binarization processing is determined.

[0201] In S10, machine learning is used to construct a reliability prediction model for determination unit 102B, which uses the threshold value determined in S9. This machine learning uses training data created by correlating the test image as correct data with the accuracy of the result determined by determination unit 102B based on the test image. The test image can be generated from an ultrasound image 111 whose presence or absence of defects is known.

[0202] In S11, a determination model is constructed by the determination unit 102C through machine learning to determine the presence or absence of defects. This machine learning uses the training data generated in S3, which is created by associating the presence or absence of defects with the training image as correct data. This allows the construction of a determination model that, when the inspection image 111A is input, outputs a value indicating the probability of a defect or the probability of no defect.

[0203] In S12, machine learning is used to construct a reliability prediction model for determination unit 102C, which uses the determination model constructed in S11. This machine learning uses training data, which correlates the accuracy of the determination results of determination unit 102C based on the test image with the test image as correct data. The test image can be generated from an ultrasound image 111 whose presence or absence of defects is known.

[0204] As described above, the reliability prediction model used by determination unit 102A can be constructed using machine learning, using training data in which information indicating whether the test image determined by determination unit 102A to have a defect is associated with the correct data. The same applies to the reliability prediction model used by determination unit 102B and the reliability prediction model used by determination unit 102C.

[0205] The reliability prediction model learns the correspondence between the test image determined by the determination unit 102 and whether the determination result is correct. Therefore, the output value obtained by inputting the inspection image 111A into the reliability prediction model represents the probability of the determination result when the determination unit 102 uses the inspection image 111A.

[0206] Therefore, the reliability determination unit 103 determines the reliability of the determination result of each determination unit 102 based on the output value obtained by inputting the inspection image 111A to the reliability prediction model for each determination unit 102, thereby being able to set an appropriate reliability based on past determination results.

[0207] (Process flow during inspection)

[0208] based on Figure 13 The flow of processing (determination method) during inspection will be described. Figure 13 This figure shows an example of an inspection method using the information processing device 1. Furthermore, the following describes that the storage unit 11 stores an ultrasonic image 111 obtained by imaging echoes from the pipe end weld and its peripheral edge, obtained by measuring while rotating the probe.

[0209] In S21 , the test image generator 101 generates the test image 111A. Specifically, the test image generator 102 obtains one ultrasonic image 111 stored in the storage 11 , inputs it into the extraction model, extracts the region representing the output value from the ultrasonic image 111 , and generates the test image 111A.

[0210] In S22 (determination step), each determination unit 102 uses the inspection image 111A generated by S21 to determine whether there is a defect. More specifically, in the determination unit 102A, the inspection image acquisition unit 1021 acquires the inspection image 111A generated by S21, and the restoration image generation unit 1022 uses the generation model constructed by S4 of S12 to generate a restoration image 111B from the inspection image 111A. In addition, the defect presence determination unit 1023 calculates the difference between each pixel of the inspection image 111A and the restoration image 111B, and then calculates the variance of the difference, and determines whether the value of the variance is greater than that of the restoration image 111B. Figure 12Furthermore, when generating the removed image 111C and the removed image (restored) 111D, the defect presence determination unit 1023 calculates the difference between these images.

[0211] In addition, the determination unit 102B uses Figure 12 The inspection image 111A generated in S21 is binarized using the threshold value determined in S9 to generate a binary image. The determination unit 102B then detects peripheral echo regions ar3 and ar4 from the generated binary image and determines the presence of a defect based on whether there is a defective region in the region sandwiched between these regions.

[0212] Then, the determination unit 102C inputs the inspection image 111A generated in S21 into the Figure 12 The determination model constructed by S11 is used to determine whether there is a defect based on its output value. For example, when a determination model that outputs a probability of defect is used, if the output value of the determination model exceeds a predetermined threshold, the determination unit 102C may determine that there is a defect. In addition, such a threshold is also Figure 12 Determined after the processing of S11.

[0213] In S23 (reliability determination step), the reliability determination unit 103 determines the reliability of the determination result of the determination unit 102 using the inspection image 111A generated in S21. Specifically, the reliability determination unit 103 determines the reliability of the determination result of the determination unit 102 based on the inspection image 111A generated in S21. Figure 12 The reliability prediction model constructed in S6 determines the reliability of the determination result of the determination unit 102A using the output value obtained by inputting the inspection image 111A.

[0214] For example, if the reliability prediction model outputs a value between 0 and 1 indicating the probability that the determination result of determination unit 102A is correct, reliability determination unit 103 can directly use this value as the reliability. Reliability determination unit 103 similarly determines the reliability of the determination results of determination units 102B and 102C. In this way, reliability determination unit 103 determines the reliability of the determination results of the presence or absence of defects for each of determination units 102A through 102C.

[0215] In S24 (comprehensive determination step), comprehensive determination unit 104 determines the presence or absence of a defect using the determination results of S22 and the reliability determined in S23. Specifically, comprehensive determination unit 104 determines the presence or absence of a defect using a numerical value obtained by adding together the numerical values ​​representing the determination results of determination units 102A to 102C, weighted according to their reliability.

[0216] For example, the determination results of the determination units 102A to 102C can be -1 (no defect) or 1 (no defect).

[0217] In this case, if the reliability is calculated using a numerical value between 0 and 1, the reliability value can be directly multiplied by the judgment result as a weight.

[0218] For example, the determination result of the determination unit 102A is that there is a defect, the determination result of the determination unit 102B is that there is no defect, and the determination result of the determination unit 102C is that there is a defect. In addition, the reliability of the determination results of the determination units 102A to 102C is 0.87, 0.51, and 0.95 respectively. In this case, the comprehensive determination unit 104 performs 1×0.87+(-1)×0.51.

[0219] The calculation of +1×0.95 yields a numerical value of 1.31.

[0220] Comprehensive determination unit 104 can then compare this value with a predetermined threshold value and determine a defect if the calculated value is greater than the threshold. If "-1" indicates no defect and "1" indicates a defect, the threshold value can be set to "0," the middle value between these values. In this case, since 1.31 > 0, comprehensive determination unit 104's final determination is a defect.

[0221] In S25, the comprehensive judgment unit 104 records the judgment result of S24 in the inspection result data 112. And, in S26, the defect type judgment process is performed. Figure 14 and Figure 15 The defect type determination process is described in detail.

[0222] In S27, it is determined whether the test image generator 101 has completed processing all of the ultrasound images 111 being tested. If it is determined that there are unprocessed ultrasound images 111 (No in S27), the process returns to S21, where the test image generator 101 reads the unprocessed ultrasound images 111 from the storage unit 11 and generates a test image 111A from the unprocessed ultrasound images 111. On the other hand, if it is determined that there are no unprocessed ultrasound images 111 (Yes in S27), the process proceeds to S28.

[0223] In S28, the integrated detection unit 108 integrates the defects detected by the integrated judgment unit 104. The integrated detection unit 108 records the integrated results in the inspection result data 112. Figure 9 If there is no defect that should be integrated, the processing of S28 and S29 is not performed, and the process proceeds to S30.

[0224] In S29, the defect length calculator 109 calculates the length of the defect integrated by the integrated detection unit 108. For example, the defect length calculator 109 may calculate the defect length by multiplying the average length of the defects in one ultrasonic image 111 by the number of defects integrated by the integrated detection unit 108. The defect length calculator 109 then records the calculation result in the inspection result data 112.

[0225] In S30, the thickness calculation unit 107 calculates the wall thickness of the pipe end weld and records the calculation result in the inspection result data 112. The wall thickness calculation method is based on Figure 10 After recording the wall thickness calculation results for all inspection images 111A, the process ends. Figure 13 processing.

[0226] (Defect type determination process: Using the type determination model)

[0227] based on Figure 14 Through Figure 13 The process of defect type determination processing performed in S26 is explained. Figure 14 This is a flowchart showing an example of defect type determination processing. In S41, the heat map generation unit 105 generates a heat map using the difference value (differential image) calculated when the determination unit 102A determines whether there is a defect. In S42, the heat map generation unit 105 performs threshold processing on the heat map generated by S41. Threshold processing is based on Figure 5 As explained above, the description will not be repeated here.

[0228] In S43, the defect type determination unit 106 uses the type determination model to determine the defect type. Specifically, the defect type determination unit 106 inputs the heat map, which has been thresholded in S42, into the type determination model and determines the defect type based on the output value. For example, if the type determination model outputs a numerical value indicating the probability of each pair of defect types belonging to that type, the defect type determination unit 106 may determine the type with the largest numerical value as the defect type.

[0229] In S44, the defect type determination unit 106 records the determination result of S43 in the inspection result data 112. This completes the defect type determination process.

[0230] (Defect type determination process: based on the location of the defect area)

[0231] The defect type determination unit 106 can replace Figure 14 Defect type determination and processing are carried out Figure 15 Determination and processing of defect types. Figure 15This is a flowchart showing an example of defect type determination processing for determining the defect type based on the position of the defect area.

[0232] In S51, the defect type determination unit 106 performs threshold processing on the difference value (difference image) calculated when the determination unit 102A determines whether or not a defect exists. The threshold processing in S51 is similar to Figure 14 The threshold processing of S42 is the same as that of S52. And, in S52, the defect type determination unit 106 detects the defect area based on the difference value after the threshold processing. The detection method of the defect area is as follows: Figure 7 As explained above, the description will not be repeated here.

[0233] In S53, the defect type determination unit 106 determines the defect type based on the position of the defect area determined in S52. For example, the defect type determination unit 106 may determine the defect type based on whether the defect area detected in S52 is included in the defect area. Figure 8 The defect type is determined by which of the areas AR1 to AR4 is used.

[0234] In S54, the defect type determination unit 106 records the determination result of S53 in the inspection result data 112. This completes the defect type determination process.

[0235] In addition, the defect type determination unit 106 can perform the following two processes, namely: Figure 14 Defect type determination and processing, and Figure 15 In this case, the defect type determination unit 106 only needs to record the determination results of both parties. In addition, the defect type determination unit 106 can combine the determination results of both parties to make a final determination of the defect type. In this case, the defect type determination unit 106 can calculate: Figure 14 The reliability of the defect type determination result, and Figure 15 The reliability of the defect type determination process is calculated, and the final defect type determination result is determined based on the calculated reliability. The reliability at this time can be calculated in the same way as the reliability of the determination result of the determination unit 102.

[0236] (Application Example)

[0237] In the above embodiment, an example of determining whether a pipe end weld has defects based on the ultrasonic image 111 is described. However, the setting of the determination items is arbitrary, as long as the object data used in the determination is set to any data corresponding to the determination items, and the present invention is not limited to the example in the above embodiment.

[0238] For example, the information processing device 1 can also be used to determine the presence or absence of defects (also referred to as abnormalities) in an inspection object during a radiographic test (RT). In this case, an image generated by the abnormality is detected based on image data obtained using an electronic device such as an imaging plate, rather than a radiographic radiograph.

[0239] In this case, the determination unit 102A can also determine the presence of an abnormal part using the generated model, and the determination unit 102C can determine the presence of an abnormal part using the determination model. Furthermore, the determination unit 102B can also determine the presence of an abnormal part by numerical analysis based on the pixel value or size of the image reflected in the image data.

[0240] Furthermore, in ultrasonic testing or RT, the presence of abnormalities can be determined using ultrasound echoes or radioactive signal waveform data, rather than image data. This allows the information processing device 1 to be applied to various nondestructive tests using a variety of data. Furthermore, in addition to nondestructive testing, the information processing device 1 can also be used for object detection based on still or moving images, as well as for classifying detected objects.

[0241] (Variation)

[0242] In the above embodiment, an example is described in which the output value obtained by inputting the inspection image into the reliability prediction model is used as the reliability. However, the reliability is not limited to this example as long as it is derived based on the data used by the determination unit 102 for determination.

[0243] For example, if determination unit 102B uses a binarized image obtained by binarizing an inspection image to determine the presence of defects, the reliability prediction model used by determination unit 102B can be a model that uses the binarized image as input data. On the other hand, if determination unit 102C uses the inspection image directly to determine the presence of defects, the reliability prediction model used by determination unit 102C can be a model that uses the inspection image as input data. In this way, the input data to the reliability prediction models used by each determination unit 102 does not need to be identical.

[0244] Furthermore, in the above embodiment, an example using three determination units 102 is described, but the number of determination units 102 may be two, or four or more. Furthermore, in the above embodiment, the determination methods of the three determination units 102 are different, but the determination methods of the determination units 102 may also be the same. For determination units 102 with the same determination method, it is sufficient to differ in the thresholds used in the determination or the teacher data used to construct the learned model used in the determination.

[0245] In addition, the execution subject of each process described in the above embodiments can be changed appropriately. For example, other information processing devices can also be used to execute: Figure 13 In the process, S21 (generation of inspection image), S23 (calculation using reliability determination model), S26 (defect type determination), S28 (defect integration), S29 (defect length calculation), and S30

[0246] (Calculation of wall thickness). Similarly, other information processing devices may be used to perform part or all of the processing performed by the determination units 102A to 102C. In addition, in these cases, the other information processing devices may be one or more. In this way, the functions of the information processing device 1 can be implemented through a variety of system structures. In addition, when building a system including multiple information processing devices, some of the information processing devices may be configured on the cloud. In other words, the functions of the information processing device 1 can also be implemented online using one or more information processing devices that perform information processing.

[0247] (Software-based implementation example)

[0248] The control module of the information processing device 1 (particularly, each unit included in the control unit 10 ) may be realized by a logic circuit (hardware) formed on an integrated circuit (IC chip) or the like, or may be realized by software.

[0249] In the latter case, the information processing device 1 includes a computer that executes commands of software that realizes various functions, namely, an information processing program. The computer, for example, includes one or more processors and a computer-readable recording medium that stores the information processing program. Furthermore, in the computer, the processor reads the information processing program from the recording medium and executes it, thereby achieving the purpose of the present invention. As the processor, for example, a CPU (Central Processing Unit) can be used. In addition, in addition to a processor such as a CPU, the information processing device 1 may also include a GPU.

[0250] (Graphics Processing Unit: Graphics Processing Unit). Using a GPU makes it possible to perform calculations using the above-mentioned various models at high speed. As the above-mentioned recording medium, a "non-temporary tangible medium" can be used. For example, in addition to ROM (Read Only Memory), a magnetic tape, an optical disk, a card, a semiconductor memory, a programmable logic circuit, etc. can also be used. In addition, a RAM (Random Access Memory) for expanding the above-mentioned program can also be provided. In addition, the above-mentioned program can also be provided to the above-mentioned computer via any transmission medium (communication network, radio waves, etc.) that can transmit the program. In addition, one embodiment of the present invention can also be implemented in the form of a data signal embedded in a carrier wave that is embodied by electronically transmitting the above-mentioned program.

[0251] The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims. Embodiments obtained by appropriately combining technical solutions disclosed in different embodiments are also included in the technical scope of the present invention.

[0252] Description of Reference Numerals

[0253] 1-Information processing device; 102A-Judgment unit (generation model judgment unit); 102B-Judgment unit (numerical analysis judgment unit); 102C-Judgment unit; 103-Reliability judgment unit; 104-Comprehensive judgment unit.

Claims

1. An information processing device comprising: The reliability determination unit performs the following processing on each of the plurality of determination units: using one object data, the reliability, which is an indicator indicating the likelihood of the determination result of the determination unit, is determined; The determination unit determines a predetermined determination item based on the object data; as well as The comprehensive determination unit determines the predetermined determination item using the determination results and the reliability determined by the reliability determination unit.

2. The information processing device according to claim 1, wherein The reliability determination unit determines the reliability of the determination result of each determination unit based on the output value obtained by inputting the object data into the reliability prediction model for each determination unit. The reliability prediction model is constructed by machine learning using teacher data obtained by associating information indicating whether the determination result is correct or not with respect to object data in which the determination unit has determined the determination matter as correct data.

3. The information processing device according to claim 1 or 2, characterized in that The object data is an image of an inspection object, The determination items are: whether there is any abnormal part in the inspection object, The plurality of determination units include a generation model determination unit that determines the presence or absence of an abnormal portion using a generation image generated by inputting the image into a generation model. The generation model is constructed so as to be able to generate a new image having the same features as an input image through machine learning using images of the inspection object without abnormal parts as training data.

4. The information processing device according to claim 3, wherein The plurality of determination units include a numerical analysis determination unit that determines an inspection target portion in the object data by analyzing each pixel value of the inspection target object, i.e., the object data, and determines the presence or absence of an abnormal portion based on the pixel value of the determined inspection target portion.

5. The information processing device according to claim 4, wherein The object data is an ultrasonic image obtained by imaging the echoes of ultrasonic waves propagating toward the inspection object. The numerical analysis and determination unit determines an area sandwiched between two peripheral echo areas where echoes from the peripheral portion of the inspection target part repeatedly appear in the ultrasonic image as the inspection target part, and determines the presence or absence of the abnormal part based on whether the determined inspection target part includes an area consisting of pixel values ​​greater than a threshold value. The information processing apparatus further includes a thickness calculator configured to calculate the thickness of the inspection target site based on a distance between the two peripheral echo regions.

6. A determination method, performed by one or more information processing devices, comprising: The reliability determination step performs the following processing on each of the plurality of determination units: using one object data, determining the reliability, which is an indicator indicating the possibility of the determination result of the determination unit, wherein The determination unit determines a predetermined determination item based on the object data; as well as The comprehensive determination step determines the predetermined determination item using the determination results and the reliability determined in the reliability determination step.

7. A computer-readable recording medium storing an information processing program for causing a computer to function as the information processing apparatus according to claim 1, that is, for causing the computer to function as the reliability determination unit and the comprehensive determination unit.

Images