Ultrasonic flaw detection device and ultrasonic flaw detection method

Abstract

To provide an ultrasonic flaw detection device, etc., capable of shortening flaw detection time without compromising detection performance.SOLUTION: An ultrasonic flaw detection device 100 comprises: a first ultrasonic probe 1; a first image generation unit 41 that generates a first image of the material S to be inspected based on an echo received by the first ultrasonic probe; an inspection condition estimation unit 42 that has a prediction model generated by performing machine learning in a teacher data generation unit 45 using the teacher data generated in the teacher data generation unit 45, and estimates optimal inspection conditions by inputting the first image into the prediction model; a second ultrasonic probe 2 that uses only transmitting and receiving transducers determined by the optimal inspection conditions, as the number of combinations of transmitting and receiving transducers is greater than the number of combinations in the first ultrasonic probe; a second image generation unit 43 that generates a first aperture synthesis image of the material to be inspected based on the echoes received by the second ultrasonic probe; and a defect detection unit 44 that detects defects present inside the material to be inspected based on the first aperture synthesis image.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an ultrasonic flaw detection device and an ultrasonic flaw detection method for detecting flaws in a test material such as a steel material using ultrasonic waves, and in particular to an ultrasonic flaw detection device and an ultrasonic flaw detection method that can reduce the flaw detection time without impairing the flaw detection performance. [Background technology]

[0002] 2. Description of the Related Art Conventionally, ultrasonic flaw detection methods have been used to detect defects present inside a material to be flaw-detected, such as a steel material. The ultrasonic flaw detection method is a method of detecting defects inside a material to be inspected by transmitting ultrasonic waves to the material using a vibrator in the ultrasonic probe and receiving the ultrasonic waves (echoes) reflected inside the material using a vibrator in the ultrasonic probe. The same vibrator as that used to transmit the ultrasonic waves may be used to receive the echoes, or a vibrator different from that used to transmit may be used.

[0003] Among the ultrasonic flaw detection methods described above, the aperture synthesis method has attracted attention. This method uses an ultrasonic probe with a large number of transmitting and receiving transducers, and performs aperture synthesis processing to add up flaw detection signal waveforms obtained from echoes received by all combinations of transmitting and receiving transducers, thereby generating an aperture synthesis image that visualizes the internal state of the material being detected (see, for example, Patent Document 1). The advantages of aperture synthesis include improved detection ability (high S / N) by averaging out noise by adding up many flaw detection signal waveforms, improved spatial resolution through post-processing, and the ability to obtain information on the distribution of defects through imaging.

[0004] However, the aperture synthesis method has the problem that it processes many flaw detection signal waveforms, which results in a large calculation load and a long flaw detection time (= processing time to acquire flaw detection signal waveforms + imaging time). [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-203037 Summary of the Invention [Problem to be solved by the invention]

[0006] The present invention has been made to solve the problems of the above-mentioned conventional technology, and has an object to provide an ultrasonic flaw detection device and an ultrasonic flaw detection method that can reduce the calculation load and shorten the flaw detection time without impairing the flaw detection performance. [Means for solving the problem]

[0007] In order to solve the above problems, the inventors have conducted extensive research and have come up with the idea of ​​generating an aperture synthetic image without using any of the transducers (thinning out any of the transducers) among the multiple transmitting transducers and multiple receiving transducers of an ultrasonic probe in order to reduce the calculation load and shorten the inspection time. In this case, the more transducers are thinned out (the number of unused transducers), the more the inspection time can be shortened, but the more the number of transducers is thinned out, the more the inspection performance (for example, the ability to distinguish two adjacent defects into two) decreases, so it is necessary to increase the number of thinning outs within a limit that does not impair the inspection performance. According to the findings of the present inventors, an appropriate method for thinning out the transducers (a method for thinning out that does not impair the flaw detection performance) is affected by the defect position. Therefore, the present inventors have come up with the idea of ​​providing an ultrasonic probe (referred to as a "first ultrasonic probe" in this specification) for grasping the rough defect position in advance, separate from an ultrasonic probe (referred to as a "second ultrasonic probe" in this specification) that performs ultrasonic testing to detect defects present inside the test material, generating an image of the inside of the test material using the first ultrasonic probe (if a defect exists in the test material, this image will reflect the defect position), and constructing a prediction model based on this image to estimate an appropriate method for thinning out the transducers of the second ultrasonic probe. Specifically, we used machine learning to generate a predictive model that takes the image generated by the first ultrasonic probe as input and outputs the optimal flaw detection conditions (the number and position of transducers after thinning, with a large thinning number within the limit that does not impair flaw detection performance).We then found that by inputting the image generated by the first ultrasonic probe into this predictive model, it is possible to estimate an appropriate thinning method for the second ultrasonic probe, with a large thinning number within the limit that does not impair flaw detection performance. The present invention has been completed based on the above findings of the present inventors.

[0008] That is, in order to solve the above-mentioned problems, the present invention provides an ultrasonic flaw detection device for detecting defects present inside a material to be detected, the ultrasonic flaw detection device comprising: a first ultrasonic probe having a plurality of transmitting vibrators and a plurality of receiving vibrators, which transmits ultrasonic waves to the material to be detected using the transmitting vibrators and receives echoes reflected inside the material to be detected using the receiving vibrators; a first image generation unit which generates a first image that visualizes the state inside the material to be detected based on the echoes received by the first ultrasonic probe; and a flaw detection condition estimation unit which has a prediction model generated using machine learning and which, by inputting the first image into the prediction model, estimates suitable flaw detection conditions that do not impair the flaw detection performance of the material to be detected, and which are composed of the number of transmitting vibrators, the positions of the transmitting vibrators, the number of receiving vibrators, and the positions of the receiving vibrators. and a second ultrasonic probe that transmits ultrasonic waves to the material to be detected using only the transmitting oscillators determined by the preferred flaw detection conditions among the plurality of transmitting oscillators and the receiving oscillators that are determined by the preferred flaw detection conditions, and receives echoes reflected inside the material to be detected using only the receiving oscillators determined by the preferred flaw detection conditions among the plurality of receiving oscillators; a second image generation unit that generates a first aperture synthetic image that visualizes the state inside the material to be detected based on the echoes received by the second ultrasonic probe; and a defect detection unit that detects defects present inside the material to be detected based on the first aperture synthetic image.

[0009] In the present invention, the "transmitting vibrator" and the "receiving vibrator" may be separate vibrators, or may be the same vibrator (i.e., one vibrator is used to transmit ultrasound and also to receive echoes). In the present invention, the "first ultrasonic probe" is not necessarily limited to an ultrasonic probe that is physically separate from the second ultrasonic probe, but also includes the concept of using some of the multiple transmitting transducers and multiple receiving transducers of the second ultrasonic probe as the multiple transmitting transducers and multiple receiving transducers of the first ultrasonic probe. In the present invention, the "first image" may be a two-dimensional image or a three-dimensional image. Furthermore, the "first image" may be an aperture synthetic image or an image expressed by normal luminance values ​​without aperture synthetic processing. In the present invention, "suitable flaw detection conditions" refer to conditions consisting of the number of transmitting and receiving transducers and the positions of the transmitting and receiving transducers among the multiple transmitting and receiving transducers possessed by the second ultrasonic probe, which do not impair the flaw detection performance of the material to be detected; in other words, conditions consisting of the number of transmitting and receiving transducers after thinning out, which is large enough to allow for a large number of thinning outs within the limit that does not impair the flaw detection performance, and the positions of the transmitting and receiving transducers after thinning out. In the present invention, examples of "flaw detection performance" include the ability to detect the position of a defect, the ability to detect the size of a defect, the ability to distinguish between two adjacent defects, etc. "Flaw detection performance is not impaired" means that, when comparing a case where a defect is detected based on a synthetic aperture image obtained using all of the transmitting transducers and all of the receiving transducers of the second ultrasonic probe with a case where a defect is detected based on a synthetic aperture image (first synthetic aperture image) obtained using only the transmitting transducers and receiving transducers determined by the preferred flaw detection conditions out of all of the transmitting transducers and all of the receiving transducers of the second ultrasonic probe, the difference in flaw detection performance between the two is within an allowable range. In the present invention, the "first aperture synthetic image" may be a two-dimensional image or a three-dimensional image. However, when the "first image" is a two-dimensional image, the "first aperture synthetic image" is also a two-dimensional image, and when the "first image" is a three-dimensional image, the "first aperture synthetic image" is also a three-dimensional image.

[0010] According to the present invention, the first image generation unit generates a first image that visualizes the state of the inside of the material to be inspected based on the echo received by the first ultrasonic probe. Then, the inspection condition estimation unit estimates suitable inspection conditions. Specifically, the first image is input to a prediction model generated by machine learning in the inspection condition estimation unit, and suitable inspection conditions are output from the prediction model. Next, according to the present invention, ultrasonic waves are transmitted to the material to be detected using only the transmitting transducers (i.e., only the transmitting transducers after thinning out) determined by the suitable flaw detection conditions out of the multiple transmitting transducers and multiple receiving transducers possessed by the second ultrasonic probe, and echoes reflected inside the material to be detected are received using only the receiving transducers (i.e., only the receiving transducers after thinning out) determined by the suitable flaw detection conditions, and a first aperture synthetic image that visualizes the state inside the material to be detected is generated by the second image generation unit based on the echoes received by the second ultrasonic probe. Finally, according to the present invention, the defect detection section detects defects present inside the material to be inspected based on the first aperture synthetic image. As described above, according to the present invention, the first aperture synthetic image, which is the target for detecting defects present inside the material to be inspected, is generated based on echoes obtained using only the transmitting transducers and receiving transducers determined by the preferred inspection conditions out of the multiple transmitting transducers and multiple receiving transducers possessed by the second ultrasonic probe. Therefore, it is possible to reduce the calculation load and shorten the inspection time without impairing the inspection performance.

[0011] When the performance of flaw detection is considered to be the performance of detecting a flaw position (hereinafter, this will be referred to as the "first aspect" where appropriate), the ultrasonic flaw detection device according to the present invention preferably comprises a training data generation unit that generates training data that is a combination of a training image and a training-use suitable flaw detection condition, and a prediction model generation unit that generates the prediction model by machine learning using the training data, wherein the first ultrasonic probe transmits ultrasonic waves to a test material having a defect therein instead of the material to be flaw-detected, receives echoes reflected inside the test material, and the first image generation unit generates the echoes received by the first ultrasonic probe. a second image that visualizes the state inside the test material based on the echoes received by all the receiving transducers of the second ultrasonic probe; the second ultrasonic probe transmits ultrasonic waves to the test material using all the transmitting transducers of the second ultrasonic probe and receives echoes reflected inside the test material using all the receiving transducers of the second ultrasonic probe; the second image generation unit generates a second aperture synthetic image that visualizes the state inside the test material based on the echoes received by all the receiving transducers of the second ultrasonic probe; Ultrasonic waves are transmitted to the test material using m (m is an integer greater than 1 and equal to or less than the number of all transmitting transducers) transmitting transducers, and echoes reflected inside the test material are received using n (n is an integer greater than 1 and equal to or less than the number of all receiving transducers) receiving transducers among all receiving transducers possessed by the second ultrasonic probe, and the second image generating unit generates a third aperture synthetic image that visualizes the state inside the test material based on the echoes received by the n receiving transducers possessed by the second ultrasonic probe, and estimates, based on the second aperture synthetic image, a condition related to the flaw detection performance. the estimated value of the defect position in the test material estimated based on the third aperture synthetic image is compared with the estimated value of the defect position in the test material estimated based on the third aperture synthetic image, and m and n that minimize the product m×n are determined under the condition that the absolute value of the difference between the estimated values ​​of the defect position is equal to or less than a predetermined value, and when the positions of m transmitting transducers and n receiving transducers are determined in the second ultrasonic probe, the sum of the processing time required to transmit ultrasonic waves from the determined m transmitting transducers and receive echoes from the determined n receiving transducers and the time required to image based on the received echoes is minimized,The teacher data generation unit sets the obtained m as the number of transmitting transducers, the obtained m positions of the transducers as the positions of the transmitting transducers, the obtained n as the number of receiving transducers, and the obtained n positions of the transducers as the positions of the receiving transducers as the suitable flaw detection conditions for learning, and generates the teacher data using the second image as the learning image, and the prediction model generation unit generates the prediction model by machine learning a machine learning model using the teacher data.

[0012] In the ultrasonic flaw detection device according to the first aspect, the "second image" may be a two-dimensional image or a three-dimensional image. However, when the "first image" is a two-dimensional image, the "second image" is also a two-dimensional image, and when the "first image" is a three-dimensional image, the "second image" is also a three-dimensional image. The "second image" may be an aperture synthetic image, or an image expressed with normal brightness values ​​without aperture synthetic processing. However, when the "first image" is an aperture synthetic image, the "second image" is also an aperture synthetic image, and when the "first image" is an image expressed with normal brightness values, the "second image" is also an image expressed with normal brightness values. The same applies to the ultrasonic flaw detection devices according to the second and third aspects described below. In the ultrasonic flaw detection device according to the first aspect, the "second aperture synthetic image" and the "third aperture synthetic image" may be two-dimensional images or three-dimensional images. However, when the "second image" is a two-dimensional image, the "second aperture synthetic image" and the "third aperture synthetic image" are also two-dimensional images, and when the "second image" is a three-dimensional image, the "second aperture synthetic image" and the "third aperture synthetic image" are also three-dimensional images. The same applies to the ultrasonic flaw detection devices according to the second and third aspects described below.

[0013] In the ultrasonic flaw detection device according to the first aspect, an estimate of the defect location in the test material estimated based on the second synthetic aperture image (an aperture synthetic image of the interior of the test material obtained using all transmitting transducers and all receiving transducers of the second ultrasonic probe) is compared with an estimate of the defect location in the test material estimated based on the third synthetic aperture image (an aperture synthetic image of the interior of the test material obtained using m transmitting transducers and n receiving transducers among all transmitting transducers and all receiving transducers of the second ultrasonic probe). Under the condition that the absolute value of the difference between the estimated defect locations is equal to or less than a predetermined value, m and n that minimize the product m×n are determined. Furthermore, the positions of the m transmitting transducers and the n receiving transducers are determined so that the sum of the processing time required to transmit ultrasonic waves from the determined m transmitting transducers and receive echoes with the determined n receiving transducers, plus the imaging time required to image the received echoes, is minimized. The processing time and imaging time can be calculated based on the specifications of the elements constituting the ultrasonic flaw detection device. The task of finding m and n that minimize the above product m×n, as well as the task of finding the positions of m transmitting transducers and n receiving transducers, may be performed by a human using the second aperture synthetic image and a plurality of third aperture synthetic images obtained by changing the values ​​of m and n, and input to the teacher data generation unit, or may be performed automatically by the teacher data generation unit itself. According to the ultrasonic flaw detection device of the first aspect, the teacher data generation unit sets the number of transmitting transducers m determined as above, the positions of the m transducers determined as above as the positions of the transmitting transducers, the number of receiving transducers n determined as above, and the positions of the n transducers determined as above as the positions of the receiving transducers to set the suitable flaw detection conditions as the learning suitable flaw detection conditions. Furthermore, the teacher data generation unit sets the second image (an image of the inside of the test material obtained using the first ultrasonic probe) as the learning image. In this way, the teacher data generation unit generates teacher data, which is a combination of the learning image and the learning suitable flaw detection conditions. According to the ultrasonic flaw detection device of the first aspect, the prediction model is generated by the prediction model generation unit training the machine learning model using the training data. Specifically, the prediction model is generated by training the machine learning model so that when training images constituting the training data are input to the machine learning model, training suitable flaw detection conditions constituting the training data are output. According to the ultrasonic flaw detection device of the first aspect, a predictive model is generated by machine learning using training data that is the preferred flaw detection conditions for learning corresponding to m and n obtained under conditions where the absolute value of the difference between the estimated value of the defect location in the test material estimated based on the second aperture synthetic image and the estimated value of the defect location in the test material estimated based on the third aperture synthetic image is less than a predetermined value, in other words, conditions where the flaw detection performance is not impaired.Therefore, it is possible to shorten the flaw detection time without impairing the defect location detection performance.

[0014] When the flaw detection performance is considered to be the performance for detecting the size of a defect (hereinafter, this will be referred to as the "second embodiment" where appropriate), the ultrasonic flaw detection device according to the present invention preferably comprises a training data generation unit that generates training data which is a combination of a training image and a training suitable flaw detection condition, and a prediction model generation unit that generates the prediction model by machine learning using the training data, wherein the first ultrasonic probe transmits ultrasonic waves to a test material having a defect therein instead of the material to be detected, receives echoes reflected inside the test material, and the first image generation unit generates the first ultrasonic probe. a second image that visualizes the state inside the test material based on echoes received by the second ultrasonic probe, the second ultrasonic probe transmits ultrasonic waves to the test material using all of the transmitting transducers of the second ultrasonic probe and receives echoes reflected inside the test material using all of the receiving transducers of the second ultrasonic probe, the second image generation unit generates a second aperture synthetic image that visualizes the state inside the test material based on the echoes received by all of the receiving transducers of the second ultrasonic probe, and transmitting ultrasonic waves to the test material using m (m is an integer greater than 1 and not greater than the number of all transmitting transducers) transmitting transducers among all transmitting transducers of the second ultrasonic probe, and receiving echoes reflected inside the test material using n (n is an integer greater than 1 and not greater than the number of all receiving transducers) receiving transducers among all receiving transducers of the second ultrasonic probe, and the second image generating unit generates a third aperture synthetic image that visualizes the state inside the test material based on the echoes received by the n receiving transducers of the second ultrasonic probe, and As a condition, an estimated value of the size of the defect in the test material estimated based on the second aperture synthetic image is compared with an estimated value of the size of the defect in the test material estimated based on the third aperture synthetic image, and m and n that minimize the product m×n are determined under the condition that the absolute value of the difference between the estimated values ​​of the defect size is equal to or less than a predetermined value, and the sum of the processing time required to transmit ultrasonic waves from the determined m transmitting transducers and receive echoes from the determined n receiving transducers in the second ultrasonic probe and the time required to create an image based on the received echoes is minimized.When the positions of m transmitting transducers and n receiving transducers are determined, the teacher data generation unit sets the determined m to the number of transmitting transducers, the determined positions of the m transducers to the positions of the transmitting transducers, the determined n to the number of receiving transducers, and the determined positions of the n transducers to the positions of the receiving transducers as the suitable flaw detection conditions for learning, and generates the teacher data using the second image as the learning image, and the prediction model generation unit generates the prediction model by machine learning a machine learning model using the teacher data.

[0015] The ultrasonic flaw detection device according to the second embodiment differs from the ultrasonic flaw detection device according to the first embodiment described above, which considers the defect position detection performance, only in that the defect size detection performance is considered as the flaw detection performance. Specifically, the ultrasonic flaw detection device of the second embodiment generates a predictive model by machine learning using training data that is the preferred flaw detection conditions for learning corresponding to m and n obtained under conditions where the absolute value of the difference between the estimated value of the flaw size in the test material estimated based on the second aperture synthetic image and the estimated value of the flaw size in the test material estimated based on the third aperture synthetic image is less than a predetermined value, in other words, conditions where the flaw detection performance is not impaired.Therefore, it is possible to shorten the flaw detection time without impairing the flaw detection performance.

[0016] When considering the ability to distinguish two adjacent defects into two as the flaw detection performance (hereinafter, this will be referred to as the "third aspect" where appropriate), the ultrasonic flaw detection device according to the present invention preferably comprises a training data generation unit that generates training data that is a combination of a training image and a training-use suitable flaw detection condition, and a prediction model generation unit that generates the prediction model by machine learning using the training data, and the first ultrasonic probe transmits ultrasonic waves to a test material having two adjacent defects inside it instead of the material to be detected, receives echoes reflected inside the test material, and generates the first image. a generating unit that generates a second image that visualizes the state of the interior of the test material based on the echoes received by the first ultrasonic probe, the second ultrasonic probe that transmits ultrasonic waves to the test material using all of the transmitting transducers of the second ultrasonic probe and receives echoes reflected inside the test material using all of the receiving transducers of the second ultrasonic probe, the second image generating unit that generates a second aperture synthetic image that visualizes the state of the interior of the test material based on the echoes received by all of the receiving transducers of the second ultrasonic probe, ultrasonic waves are transmitted to the test material using m (m is an integer greater than 1 and not greater than the number of all transmitting transducers) transmitting transducers among all transmitting transducers possessed by the second ultrasonic probe, and echoes reflected inside the test material are received using n (n is an integer greater than 1 and not greater than the number of all receiving transducers) receiving transducers among all receiving transducers possessed by the second ultrasonic probe, and the second image generating unit generates a third aperture synthetic image that images the state inside the test material based on the echoes received by the n receiving transducers possessed by the second ultrasonic probe, As a condition related to the function, under the condition that in the second aperture synthetic image, pixel regions corresponding to two defects in the test material are separated into two at a predetermined brightness value or more, and in the third aperture synthetic image, pixel regions corresponding to two defects in the test material are separated into two at a predetermined brightness value or more, m and n that minimize the product m×n are obtained, and in the second ultrasonic probe, the sum of the processing time required to transmit ultrasonic waves from the obtained m transmitting transducers and receive echoes from the obtained n receiving transducers and the time required to create an image based on the received echoes is minimized.When the positions of m transmitting transducers and n receiving transducers are determined, the teacher data generation unit sets the determined m to the number of transmitting transducers, the determined positions of the m transducers to the positions of the transmitting transducers, the determined n to the number of receiving transducers, and the determined positions of the n transducers to the positions of the receiving transducers as the suitable flaw detection conditions for learning, and generates the teacher data using the second image as the learning image, and the prediction model generation unit generates the prediction model by machine learning a machine learning model using the teacher data.

[0017] The ultrasonic flaw detection device of the third aspect differs from the ultrasonic flaw detection device of the aforementioned first aspect, which considers defect position detection performance, only in that the third aspect considers identification performance, which is the ability to distinguish two adjacent defects into two, as its flaw detection performance. Specifically, the ultrasonic flaw detection device of the third aspect generates a predictive model by machine learning using as training data the preferred flaw detection conditions for learning corresponding to m and n obtained under conditions where the pixel areas corresponding to two defects in the test material in the second aperture synthetic image are separated into two at a predetermined brightness value or above, and the pixel areas corresponding to two defects in the test material in the third aperture synthetic image are separated into two at a predetermined brightness value or above, in other words, conditions where the discrimination performance of being able to distinguish two adjacent defects into two is not impaired.Therefore, it is possible to shorten the inspection time without impairing the discrimination performance of the two defects.

[0018] The ultrasonic flaw detection devices according to the first to third aspects described above may be configured individually, or may be configured as an ultrasonic flaw detection device capable of simultaneously performing multiple aspects of the first to third aspects, such as an ultrasonic flaw detection device capable of simultaneously performing the first and second aspects (i.e., an ultrasonic flaw detection device that does not impair both the performance of detecting the defect location and the performance of detecting the defect size).

[0019] Further, in order to solve the above-mentioned problems, the present invention provides an ultrasonic testing method for detecting defects present inside a material to be tested, the method comprising: a first ultrasonic transmission / reception step of using a first ultrasonic probe having a plurality of transmitting vibrators and a plurality of receiving vibrators, transmitting ultrasonic waves to the material to be tested using the transmitting vibrators, and receiving echoes reflected inside the material to be tested using the receiving vibrators; a first image generation step of using a first image generation unit to generate a first image that visualizes the state inside the material to be tested based on the echoes received by the first ultrasonic probe; and a flaw detection condition estimation step of using a flaw detection condition estimation unit having a prediction model generated using machine learning to input the first image into the prediction model, thereby estimating suitable flaw detection conditions that do not impair the flaw detection performance of the material to be tested, the suitable flaw detection conditions being composed of the number of transmitting vibrators, the positions of the transmitting vibrators, the number of receiving vibrators, and the positions of the receiving vibrators. It is also provided as an ultrasonic flaw detection method, comprising: a second ultrasonic transmission / reception step of using a second ultrasonic probe having multiple transmitting vibrators and multiple receiving vibrators, the number of combinations of transmitting vibrators and receiving vibrators being greater than the number of combinations of transmitting vibrators and receiving vibrators of the first ultrasonic probe, transmitting ultrasonic waves to the material to be detected using only the transmitting vibrators of the multiple transmitting vibrators that are determined by the preferred flaw detection conditions, and receiving echoes reflected inside the material to be detected using only the receiving vibrators of the multiple receiving vibrators that are determined by the preferred flaw detection conditions; a second image generation step of using a second image generation unit to generate a first aperture synthetic image that images the interior of the material to be detected based on the echoes received by the second ultrasonic probe; and a defect detection step of using a defect detection unit to detect defects present inside the material to be detected based on the first aperture synthetic image. [Effects of the Invention]

[0020] According to the present invention, it is possible to reduce the calculation load and shorten the inspection time without impairing the inspection performance. [Brief explanation of the drawings]

[0021] [Figure 1]1 is a diagram illustrating a schematic configuration of an ultrasonic flaw detection device according to an embodiment of the present invention. [Figure 2] 10A and 10B are diagrams for explaining the influence of thinning out transducers; [Figure 3] 10A and 10B are diagrams showing the vicinity of a pixel region corresponding to a defect in the second aperture synthetic image and the third aperture synthetic image. [Figure 4] 10A and 10B are diagrams for explaining the relationship between the size of a defect and the maximum luminance value of a pixel region corresponding to the defect in the second aperture synthetic image and the third aperture synthetic image. [Figure 5] 10A and 10B are diagrams for schematically explaining an example of a criterion for determining whether pixel regions corresponding to two defects are separated into two. [Figure 6] 2 is a diagram for schematically explaining the execution contents of a prediction model generating unit 46 and a flaw detection condition estimating unit 42 shown in FIG. 1. FIG. [Figure 7] FIG. 10 is a diagram for schematically explaining an example of a method for estimating whether or not a certain flaw detection condition satisfies a condition related to flaw detection performance. DETAILED DESCRIPTION OF THE INVENTION

[0022] Hereinafter, with reference to the accompanying drawings as appropriate, an ultrasonic flaw detection device according to one embodiment of the present invention will be described, taking as an example a case where a material to be detected is transported in a horizontal direction. Fig. 1 is a diagram showing a schematic configuration of an ultrasonic flaw detection device according to this embodiment. In Fig. 1, the X direction indicates the conveyance direction of the material S to be detected, the Y direction indicates the horizontal direction perpendicular to the conveyance direction of the material S to be detected, and the Z direction indicates the thickness direction of the material S to be detected (the direction of transmitting and receiving ultrasonic waves). The same applies to the other figures described below. 1, the ultrasonic flaw detection device 100 according to this embodiment is a device for detecting defects present inside a material to be detected S, and includes a first ultrasonic probe 1, a second ultrasonic probe 2, flaw detectors 3a and 3b, and a calculation processing unit 4. The calculation processing unit 4 has a first image generation unit 41, a flaw detection condition estimation unit 42, a second image generation unit 43, a defect detection unit 44, a training data generation unit 45, and a prediction model generation unit 46. Each of the components included in the ultrasonic flaw detection device 100 will be described below.

[0023] <1st ultrasound probe 1> The first ultrasonic probe 1 has a plurality of transmitting vibrators and a plurality of receiving vibrators (not shown) arranged opposite each other in the thickness direction (Z direction) of the material to be inspected S. As the first ultrasonic probe 1, for example, a linear array type ultrasonic probe having a plurality of vibrators (each vibrator functions as both a transmitting vibrator and a receiving vibrator) arranged in a line along the Y direction can be used. The first ultrasonic probe 1 transmits ultrasonic waves to the material S to be detected using a transmitting oscillator, and receives echoes reflected inside the material S to be detected using a receiving oscillator. In this embodiment, an ultrasonic probe that is physically separate from the second ultrasonic probe 2 described below is used as the first ultrasonic probe 1, but the present invention is not limited to this. For example, when the material to be detected S is stationary, it is also possible to use some of the multiple transmitting vibrators and multiple receiving vibrators of the second ultrasonic probe 2 as the multiple transmitting vibrators and multiple receiving vibrators of the first ultrasonic probe 1. In this case, it is also possible to adopt a configuration that includes a single common flaw detector electrically connected to the first ultrasonic probe 1 and the second ultrasonic probe 2, instead of the flaw detectors 3a and 3b described below.

[0024] <Second ultrasound probe 2> The second ultrasonic probe 2 is arranged downstream of the position of the first ultrasonic probe 1 in the transport direction of the material S to be detected, facing the thickness direction (Z direction) of the material S to be detected, and has multiple transmitting vibrators and multiple receiving vibrators (not shown). The number of combinations of transmitting vibrators and receiving vibrators that the second ultrasonic probe 2 has is greater than the number of combinations of transmitting vibrators and receiving vibrators that the first ultrasonic probe 1 has. In other words, the first ultrasonic probe 1 only needs to be able to detect echoes that allow the rough location of defects present inside the material to be detected S, whereas the second ultrasonic probe 2 needs to be able to detect echoes with a higher density in order to detect defects with high accuracy. For this reason, the first ultrasonic probe 1 and the second ultrasonic probe 2 are configured so that the number of combinations of transmitting vibrators and receiving vibrators of the second ultrasonic probe 2 is greater than that of the first ultrasonic probe 1. The second ultrasonic probe 2 may be, for example, a matrix array ultrasonic probe having a plurality of transducers (each of which functions as both a transmitting transducer and a receiving transducer) arranged in a matrix along the X and Y directions. The second ultrasonic probe 2 transmits ultrasonic waves to the material S to be detected using a transmitting vibrator, and receives echoes reflected inside the material S using a receiving vibrator. At this time, of the multiple transmitting vibrators, only the transmitting vibrators determined by the preferred flaw detection conditions described below are used to transmit ultrasonic waves to the material S to be detected, and of the multiple receiving vibrators, only the receiving vibrators determined by the preferred flaw detection conditions are used to receive echoes reflected inside the material S to be detected.

[0025] <Flaw detector 3a, 3b> The flaw detector 3a is electrically connected to the first ultrasonic probe 1 and is equipped with well-known components similar to those of flaw detectors equipped in general ultrasonic flaw detection devices, such as a pulser for transmitting ultrasonic waves from the transmitting transducer of the first ultrasonic probe 1, a receiver for receiving echoes from the receiving transducer of the first ultrasonic probe 1, an amplifier for amplifying the analog flaw detection signal output from the receiver, and an A / D converter for converting the analog signal output from the amplifier into a digital signal. The flaw detector 3b is electrically connected to the second ultrasonic probe 2 and is equipped with well-known components similar to those of flaw detectors equipped in general ultrasonic flaw detection devices, such as a pulser for transmitting ultrasonic waves from the transmitting transducer of the second ultrasonic probe 2, a receiver for receiving echoes in the receiving transducer of the second ultrasonic probe 2, an amplifier for amplifying the analog flaw detection signal output from the receiver, and an A / D converter for converting the analog signal output from the amplifier into a digital signal.

[0026] <Calculation processing unit 4> The arithmetic processing unit 4 is composed of, for example, a general-purpose computer electrically connected to the flaw detectors 3a and 3b, and stores programs for causing the arithmetic processing unit 4 to function as a first image generating unit 41, a flaw detection condition estimating unit 42, a second image generating unit 43, a defect detecting unit 44, a teacher data generating unit 45, and a prediction model generating unit 46. The arithmetic processing unit 4 can then function as the first image generating unit 41, the flaw detection condition estimating unit 42, the second image generating unit 43, the defect detecting unit 44, the teacher data generating unit 45, and the prediction model generating unit 46 by executing these programs.

[0027] <First image generation unit 41> The first image generating unit 41 generates a first image that visualizes the state of the inside of the material S to be flaw-detected, based on the echoes received by the first ultrasonic probe 1. Specifically, the first image generating unit 41 receives a flaw detection signal obtained from an echo received by the first ultrasonic probe 1, which is composed of a relatively small number of transducers. That is, the digital flaw detection signal output from the flaw detector 3a (the A / D converter provided in the flaw detector 3a). Based on the input flaw detection signal, the first image generating unit 41 generates two-dimensional brightness value data (data in which the value of each pixel in the YZ cross section is represented by a brightness value corresponding to the intensity of the flaw detection signal) composed of brightness values ​​corresponding to the intensity of the flaw detection signal. The first image generating unit 41 of this embodiment then generates a first image, which is a three-dimensional image, by synthesizing the two-dimensional brightness value data for multiple YZ cross sections along the X direction that are sequentially generated as the flaw detection material S is transported. That is, the first image of this embodiment is a three-dimensional image represented by normal brightness values ​​without aperture synthesis processing. However, if the imaging time does not pose a problem in the processing of the ultrasonic flaw detection device 100, it may be an image that has undergone aperture synthesis processing. If a defect exists inside the material to be inspected S, the first image will be an image that reflects the approximate position of the defect. Furthermore, since the first image is generated using data from a relatively smaller number of transducers than the number of transducers in the second ultrasonic probe 2, the first image can be generated in a relatively short time.

[0028] <Flaw detection condition estimation section 42> The inspection condition estimation unit 42 has a prediction model generated by machine learning. In this embodiment, as will be described later, the prediction model generation unit 46 generates a prediction model by machine learning using training data generated by the training data generation unit 45, and the inspection condition estimation unit 42 acquires the prediction model generated by the prediction model generation unit 46 (the trained prediction model) in advance prior to defect detection. The flaw detection condition estimation unit 42 inputs the first image generated by the first image generation unit 41 into a prediction model to estimate suitable flaw detection conditions that do not impair the flaw detection performance of the material S to be detected and are composed of the number of transmitting transducers, the positions of the transmitting transducers, the number of receiving transducers, and the positions of the receiving transducers. That is, by inputting the first image into the prediction model, the suitable flaw detection conditions are output from the prediction model. The suitable flaw detection conditions are conditions composed of the number of transmitting transducers and receiving transducers and the positions of the transmitting transducers and receiving transducers among the multiple transmitting transducers and multiple receiving transducers possessed by the second ultrasonic probe 2 that do not impair the flaw detection performance of the material S to be detected. In other words, the suitable flaw detection conditions are conditions composed of the number of transmitting transducers and receiving transducers after thinning out, and the positions of the transmitting transducers and receiving transducers after thinning out, with the thinning number (the number of unused transducers) maximized within the limit that does not impair the flaw detection performance.

[0029] Figure 2 is a diagram for explaining the effect of thinning out the transducers. As shown in Figure 2(a), consider the case where two defects (cavities) F1 and F2 present inside the test material S are detected using a second ultrasonic probe 2. 2(b), the second ultrasonic probe 2 is a matrix-array ultrasonic probe having nine transducers 21 (each transducer 21 functions both as a transmitting transducer and a receiving transducer) arranged in a matrix of three transducers each along the X and Y directions, and all of these transducers 2 are used to transmit ultrasonic waves to the material S to be detected and all of the transducers 2 are used to receive echoes reflected inside the material S to be detected (in FIG. 2, the transducers 21 used for transmitting and receiving ultrasonic waves are filled in black. The same applies to FIGS. 6 and 7 described below). In this case, as shown in the lower diagram of FIG. 2(b), the pixel areas F1' and F2' corresponding to the two defects F1 and F2, respectively, are separated into two in an aperture synthetic image (three-dimensional image) that visualizes the interior of the material S to be detected based on the received echoes. In practice, the second ultrasonic probe 2 generally has more transducers 21, but for convenience, FIG. 2 illustrates the probe having nine transducers 21.

[0030] Here, as shown in the upper diagram of FIG. 2(c), even when one transducer 21 is thinned out (in FIG. 2, the thinned transducer 21 is indicated by a dashed line; the same applies to FIGS. 6 and 7 described below), pixel regions F1' and F2' corresponding to the two defects F1 and F2, respectively, are separated into two in the aperture synthesis image obtained based on the received echoes, as shown in the lower diagram of FIG. 2(c). However, when two transducers 21 are thinned out, as shown in the upper diagram of FIG. 2(d), pixel regions corresponding to the two defects F1 and F2 become a single unseparated pixel region F3' in the aperture synthesis image obtained based on the received echoes, as shown in the lower diagram of FIG. 2(d). In such a case, although a thinning number of 2 as shown in FIG. 2(d) can shorten the flaw detection time, it is inadequate in terms of flaw detection performance (the ability to distinguish two adjacent defects into two). For this reason, it is necessary to use a prediction model to estimate the optimum flaw detection conditions corresponding to the thinning number = 1 as shown in Fig. 2(c), which can shorten the flaw detection time without impairing the flaw detection performance. The specific content of the prediction model will be described later.

[0031] The flaw detection condition estimation unit 42 outputs the preferable flaw detection conditions estimated using the prediction model to the flaw detector 3b. The flaw detector 3b drives the pulser and receiver based on the preferable flaw detection conditions input from the flaw detection condition estimation unit 42, so that, as described above, the second ultrasonic probe 2 transmits ultrasonic waves to the material S to be detected using only the transmitting transducers determined by the preferable flaw detection conditions out of the multiple transmitting transducers, and receives echoes reflected inside the material S to be detected using only the receiving transducers determined by the preferable flaw detection conditions out of the multiple receiving transducers. Note that, because the input to the prediction model for estimating the optimum flaw detection conditions is the first image generated based on the echo received by the first ultrasonic probe 1, the setting of the optimum flaw detection conditions by the flaw detector 3b needs to be performed at the timing when the defect detected by the first ultrasonic probe 1 reaches the second ultrasonic probe 2. As shown in FIG. 1, if the separation distance between the first ultrasonic probe 1 and the second ultrasonic probe 2 in the conveying direction (X direction) of the flaw detection material S is L and the conveying speed of the flaw detection material S is V, the timing when the defect detected by the first ultrasonic probe 1 reaches the second ultrasonic probe 2 is L / V after the defect is located directly below the first ultrasonic probe 1. For example, by inputting this separation distance L and conveying speed V into the flaw detector 3b, the flaw detector 3b can set the optimum flaw detection conditions at the appropriate timing.

[0032] <Second image generation unit 43> The second image generation unit 43 generates a first aperture synthetic image that visualizes the internal state of the material S to be inspected based on the echoes received by the second ultrasonic probe 2 (echoes obtained using only the transmitting transducer and the receiving transducer determined by the suitable inspection conditions). Specifically, the second image generation unit 43 receives a flaw detection signal obtained from the echo received by the second ultrasonic probe 2, i.e., a digital flaw detection signal output from the flaw detector 3b (the A / D converter provided in the flaw detector 3b). The second image generation unit 43 generates a first aperture synthetic image by performing aperture synthesis processing on the input flaw detection signal. The first aperture synthetic image in this embodiment is a three-dimensional image, but may also be a two-dimensional image formed of, for example, a YZ cross section. Since known aperture synthesis processing can be applied as is, a description of the specific contents of the aperture synthesis processing will be omitted here.

[0033] <Defect detection unit 44> The defect detection unit 44 detects defects present inside the material S to be inspected based on the first aperture synthetic image generated by the second image generation unit 43. Specifically, pixel regions with a predetermined luminance value or more are detected as defects in the first aperture synthetic image.

[0034] <Teacher data generation unit 45> The teacher data generating unit 45 generates teacher data that is a combination of learning images and learning suitable flaw detection conditions. The teacher data generated by the teacher data generation unit 45 can vary depending on whether the inspection performance is considered as defect position detection performance (first mode), defect size detection performance (second mode), or identification performance capable of distinguishing two adjacent defects into two (third mode). The contents of the teacher data generated by the teacher data generating unit 45 will be described below for each of the first to third modes.

[0035] [First aspect] In a first aspect in which the defect location detection performance is considered as the flaw detection performance, a test material S' having an internal defect is prepared instead of the test material S. Then, with the test material S' placed opposite the first ultrasonic probe 1, ultrasonic waves are transmitted to the test material S' using the first ultrasonic probe 1, and echoes reflected inside the test material S' are received. Then, using the first image generation unit 41, a second image that visualizes the state inside the test material S' is generated based on the echoes received by the first ultrasonic probe 1. Like the first image of the test material S, this second image is a three-dimensional image represented by normal brightness values ​​without aperture synthesis processing in this embodiment, but it may also be a three-dimensional image with aperture synthesis processing. Next, with the test material S' placed opposite the second ultrasonic probe 2, the second ultrasonic probe 2 is used to transmit ultrasonic waves to the test material S' using all of the transmitting transducers of the second ultrasonic probe 2, and echoes reflected inside the test material S' are received using all of the receiving transducers of the second ultrasonic probe 2. Then, using the second image generation unit 43, a second aperture synthetic image is generated that visualizes the state inside the test material S' based on the echoes received by all of the receiving transducers of the second ultrasonic probe 2. Like the first aperture synthetic image of the test material S, this second aperture synthetic image is a three-dimensional image in this embodiment, but it may also be a two-dimensional image.

[0036] Next, with the test material S' also placed opposite the second ultrasonic probe 2, the second ultrasonic probe 2 is used to transmit ultrasonic waves to the test material S' using m (m is an integer greater than 1 and not greater than the number of all transmitting transducers) transmitting transducers among all transmitting transducers possessed by the second ultrasonic probe 2, and echoes reflected inside the test material S' are received using n (n is an integer greater than 1 and not greater than the number of all receiving transducers) receiving transducers among all receiving transducers possessed by the second ultrasonic probe 2. Then, using the second image generator 43, a third aperture synthetic image is generated that visualizes the state inside the test material S' based on the echoes received by the n receiving transducers possessed by the second ultrasonic probe 2. This third aperture synthetic image is generated by the same imaging method as the second aperture synthetic image. Then, while gradually decreasing the values ​​of m and n (and while changing the positions of the m transmitting transducers and the n receiving transducers), a third aperture synthetic image is generated for each value of m and n (and for each position of the m transmitting transducers and the n receiving transducers), and as a condition related to flaw detection performance, the estimated value of the defect position in the test material S' estimated based on the second aperture synthetic image is compared with the estimated value of the defect position in the test material S' estimated based on the third aperture synthetic image, and m and n that minimize the product m × n are found under the condition that the absolute value of the difference between the estimated values ​​of the defect position is less than a predetermined value.

[0037] 3 is a diagram showing the vicinity of a pixel region corresponding to a defect in a predetermined YZ cross section of the second aperture synthetic image and the third aperture synthetic image. In the second aperture synthetic image and the third aperture synthetic image, the defect position (position of defect F) is estimated, for example, by the coordinates (X0, Y0, Z0) of the pixel having the maximum brightness value in the pixel region corresponding to defect F. Then, as a condition for the defect position to be equal to or less than a predetermined value, for example, it can be considered to set the condition that the absolute value of the difference between the estimated values ​​of the defect position is equal to or less than one wavelength of the ultrasonic wave transmitted from the transmitting transducer.

[0038] After determining m and n that minimize the product m×n as described above, the positions of the m transmitting transducers and the n receiving transducers are determined in the second ultrasonic probe 2 so that the sum of the processing time required to transmit ultrasound waves from the determined m transmitting transducers and receive echoes from the determined n receiving transducers and the time required to create an image based on the received echoes is minimized. In this embodiment, the work of finding m and n that minimize the above product m×n, as well as the work of finding the positions of m transmitting transducers and n receiving transducers, is performed by a human using the second aperture synthetic image and a plurality of third aperture synthetic images obtained by changing the values ​​of m and n, and is input to the teacher data generation unit 45. The teacher data generation unit 45 sets the obtained m (input m) as the number of transmitting transducers, the obtained positions of the m transducers (the positions of the input m transducers) as the positions of the transmitting transducers, the obtained n (input n) as the number of receiving transducers, and the obtained positions of the n transducers (the positions of the input n transducers) as the positions of the receiving transducers, as the suitable flaw detection conditions for learning. Also, the second image is set as the learning image. Then, teacher data is generated which is a combination of the learning image and the suitable flaw detection conditions for learning. The teacher data generation unit 45 outputs the generated teacher data to the prediction model generation unit 46.

[0039] [Second mode] In the second aspect, training data is generated taking into consideration the performance of detecting the size of a defect as the flaw detection performance. Below, differences from the first aspect will be described, and descriptions of the same content as the first aspect will be omitted. In the second aspect, while gradually decreasing the values ​​of m and n (and while changing the positions of m transmitting transducers and n receiving transducers), a third aperture synthetic image is generated for each value of m and n (and for each position of m transmitting transducers and n receiving transducers), and as a condition related to flaw detection performance, the estimated value of the size of the defect in the test material S' estimated based on the second aperture synthetic image is compared with the estimated value of the size of the defect in the test material S' estimated based on the third aperture synthetic image, and m and n that minimize the product m × n are found under the condition that the absolute value of the difference between the estimated values ​​of the defect size is less than a predetermined value.

[0040] FIG. 4 is a diagram for explaining the relationship between the size of a defect and the maximum luminance value of a pixel region corresponding to the defect in the second aperture synthetic image and the third aperture synthetic image. Generally, the larger the defect, the greater the echo intensity, so the size of the defect is estimated from the maximum brightness value of the pixel area corresponding to the defect. In actual ultrasonic flaw detection, the size of the defect to be detected is specifically defined; for example, defects of 3 mm or larger are considered harmful and are subject to detection. For this reason, the test material S' is, for example, one that has a defect of about 3 mm in size inside, which is the lower limit of the size of the defect to be detected. As shown in FIG. 4, when there is no transducer thinning (thinning number = 0, i.e., in the case of the second aperture synthetic image), the change in the maximum brightness value relative to the change in defect size is large around a defect size of 3 mm (i.e., the gradient of the maximum brightness value / defect size is large). Therefore, the size of defects around 3 mm can be accurately estimated using the maximum brightness value (harmful and harmless defects can be accurately distinguished). However, as shown in FIG. 4, when the thinning number is large, for example, when the thinning number = a, the change in the maximum brightness value relative to the change in defect size around 3 mm becomes small. Therefore, even if the defect size is actually 3 mm, which is harmful, if it is estimated using the maximum brightness value, it may be erroneously estimated as a defect less than 3 mm, which is considered harmless. In other words, the estimated size of the defect in the test material S' estimated based on the third aperture synthetic image may differ significantly from the estimated size of the defect in the test material S' estimated based on the second aperture synthetic image. Therefore, as described above, by finding m and n that minimize the product m×n under the condition that the absolute value of the difference between the estimated values ​​of the defect size is equal to or less than a predetermined value, training data can be generated that does not impair the detection performance of the defect size.

[0041] [Third aspect] In the third embodiment, training data is generated with consideration given to the flaw detection performance of being able to distinguish two adjacent defects into two. Below, differences from the first embodiment will be explained, and explanations of the same content as the first embodiment will be omitted. In the third embodiment, a test material S' having two defects located close to each other inside is used instead of the test material S. As the two defects in the test material S', it is preferable to use the two defects that are closest to each other and that are intended to be determined as two defects by the ultrasonic flaw detection device 100. Then, in the third aspect, while gradually decreasing the values ​​of m and n (and while changing the positions of the m transmitting transducers and the n receiving transducers), a third aperture synthetic image is generated for each value of m and n (and for each position of the m transmitting transducers and the n receiving transducers), and m and n that minimize the product m×n are found under the condition that, as a condition related to flaw detection performance, in the second aperture synthetic image, the pixel areas corresponding to the two defects in the test material S' are separated into two at a predetermined brightness value or above, and in the third aperture synthetic image, the pixel areas corresponding to the two defects in the test material S' are separated into two at a predetermined brightness value or above.

[0042] FIG. 5 is a diagram for explaining an example of a criterion for determining whether pixel regions corresponding to two defects are separated into two. As the predetermined brightness value, for example, a brightness value that is −6 dB with respect to the maximum brightness value of the pixel area corresponding to the defect is used. In the second aperture synthetic image and the third aperture synthetic image, when the vicinity of the pixel areas corresponding to the two defects is in a state as shown in the left diagram of Figure 5(a), the brightness value profile along the line Z1 will be as shown in the right diagram of Figure 5(a), and the pixel areas will not be separated at brightness values ​​of -6 dB or more. In contrast, in the second aperture synthetic image and the third aperture synthetic image, when the vicinity of the pixel regions corresponding to the two defects is in a state as shown in the left diagram of Figure 5(b), the brightness value profile along the line Z1 will be as shown in the right diagram of Figure 5(b), and the pixel region will be separated into two at brightness values ​​of -6 dB or more. Therefore, by finding m and n that minimize the product m×n under the condition that both the second aperture synthetic image and the third aperture synthetic image are in the state shown in Figure 5(b), it is possible to generate training data that does not impair the ability to identify the two defects.

[0043] <Prediction model generation unit 46> The prediction model generation unit 46 generates a prediction model by performing machine learning on a machine learning model (i.e., a prediction model before learning) using the training data generated by the training data generation unit 45. The machine learning model is not limited to, but may be, for example, a neural network or a support vector machine.

[0044] 6A and 6B are diagrams illustrating the execution contents of the prediction model generation unit 46 and the flaw detection condition estimation unit 42. Fig. 6A shows the execution contents of the prediction model generation unit 46, and Fig. 6B shows the execution contents of the flaw detection condition estimation unit 42. 6(a), the prediction model generation unit 46 trains the machine learning model using training data (combinations of training images and suitable training inspection conditions) generated for multiple test materials S' that differ in the positions, sizes, numbers, etc. of defects F present inside. Specifically, the machine learning model is trained so that suitable training inspection conditions that constitute the training data are output when training images that constitute the training data are input to the machine learning model. By using the predictive model generated in this way, even if a defect F different from the training data is present inside the material to be inspected S, as shown in Figure 6(b), it is possible to estimate optimal inspection conditions by inputting the first image of the material to be inspected S into the predictive model.

[0045] As described above, according to the ultrasonic flaw detection device 100 of this embodiment, the first aperture synthetic image, which is the target for detecting defects present inside the material to be detected S, is generated based on echoes obtained using only the transmitting transducers and receiving transducers determined by the preferred flaw detection conditions out of the multiple transmitting transducers and multiple receiving transducers possessed by the second ultrasonic probe 2, so that the flaw detection time can be shortened without impairing the flaw detection performance.

[0046] In this embodiment, when determining suitable flaw detection conditions for learning, the values ​​of m and n are gradually decreased (and the positions of m transmitting transducers and n receiving transducers are changed) and a third aperture synthetic image is generated for each value of m and n (and for each position of m transmitting transducers and n receiving transducers). However, generating a large number of third aperture synthetic images is time-consuming. For this reason, for example, when the values ​​of m and n are the same but the positions of m transmitting transducers and n receiving transducers are different, it is also possible to adopt a mode in which some of them do not actually generate a third aperture synthetic image. Specifically, it is also possible to adopt a method of estimating, using another prediction model, whether or not the flaw detection conditions consisting of the number m of certain transmitting transducers, the positions of m transmitting transducers, the number n of receiving transducers, and the positions of n receiving transducers satisfy the conditions related to flaw detection performance (the conditions related to the first to third modes described above).

[0047] FIG. 7 is a diagram for schematically explaining an example of a method for estimating whether or not a certain flaw detection condition satisfies a condition related to flaw detection performance. As shown in FIG. 7(a), in this method, training data is generated which is a combination of the inspection conditions under which a third aperture synthetic image was actually generated (the inspection conditions are composed of the number of transmitting transducers, the positions of the transmitting transducers, the number of receiving transducers, and the positions of the receiving transducers, and are the "training inspection conditions" shown in FIG. 7(a)) and the results of an evaluation using the third aperture synthetic image to determine whether or not the inspection conditions satisfied conditions related to the inspection performance ("○" if satisfied, "×" if not satisfied). Then, this training data is used to train a second machine learning model. Specifically, the second machine learning model is trained so that when the training inspection conditions constituting the training data are input to the second machine learning model, the evaluation results constituting the training data are output. This generates a second prediction model. By using this second prediction model, as shown in Fig. 7(b), when a flaw detection condition that has not actually generated a third aperture synthetic image is input to the second prediction model, it is possible to estimate whether the flaw detection condition satisfies the conditions related to the flaw detection performance, thereby reducing the effort required to generate a third aperture synthetic image. If the flaw detection condition input to the second prediction model satisfies the conditions related to the flaw detection performance (for example, as shown in Fig. 7(b)), the flaw detection condition can be used as a candidate for the above-mentioned suitable flaw detection condition for learning.

[0048] Furthermore, in this embodiment, when determining the suitable learning inspection conditions, it cannot be denied that there is a possibility that the suitable learning inspection conditions will result in extreme inspection conditions consisting of only a transmitting transducer and a receiving transducer located directly above the defect. To avoid this, it is possible to evaluate whether or not a third aperture synthetic image generated under certain inspection conditions is a meaningful aperture synthetic image, and add this as a constraint when determining suitable inspection conditions for learning.As an evaluation index for whether or not a third aperture synthetic image is a meaningful aperture synthetic image, for example, it is possible to calculate the absolute value of the difference in brightness values ​​of adjacent pixels constituting the third aperture synthetic image for all pixels, and use as a constraint that the maximum value of the evaluation index is 2 dB or less.

[0049] In addition, in the present embodiment, a configuration has been described in which the arithmetic processing unit 4 of the same ultrasonic flaw detection device 100 includes all of the first image generation unit 41, the flaw detection condition estimation unit 42, the second image generation unit 43, the defect detection unit 44, the teacher data generation unit 45, and the prediction model generation unit 46, but the present invention is not limited to this. For example, when multiple similar ultrasonic flaw detection devices 100 are installed, each including the first ultrasonic probe 1, the second ultrasonic probe 2, the flaw detectors 3a and 3b, and the arithmetic processing unit 4, it is also possible to adopt a configuration in which only the arithmetic processing unit 4 of one ultrasonic flaw detection device 100 includes the teacher data generation unit 45 and the prediction model generation unit 46, and the arithmetic processing units 4 of the remaining ultrasonic flaw detection devices 100 do not include the teacher data generation unit 45 or the prediction model generation unit 46. In this case, for example, a predictive model can be generated using the teacher data generation unit 45 and the predictive model generation unit 46 possessed by the calculation processing unit 4 of one ultrasonic flaw detection device 100, and this predictive model can be stored in the flaw detection condition estimation unit 42 possessed by the calculation processing unit 4 of the remaining ultrasonic flaw detection device 100.

[0050] Furthermore, in common with the first, second, and third aspects of this embodiment, the region to be ultrasonically inspected in the material to be inspected S can be changed as desired depending on the need for inspection. For example, in a situation where only half of the region in the Z direction of the material to be inspected S needs to be inspected, the generation of a prediction model and the generation of the first aperture synthetic image, second aperture synthetic image, and third aperture synthetic image may be limited to half of the region in the Z direction. [Explanation of symbols]

[0051] 1...1st ultrasound probe 2...Second ultrasound probe 3a, 3b...Flaw detector 4. Processing unit 41 First image generating unit 42...Flaw detection condition estimation section 43 Second image generation unit 44 Defect detection section 45. Training data generation unit 46. ​​Prediction model generation unit 100...Ultrasonic flaw detection equipment S···The material being inspected

Claims

1. An ultrasonic flaw detection device for detecting defects present inside a material to be inspected, a first ultrasonic probe having a plurality of transmitting vibrators and a plurality of receiving vibrators, transmitting ultrasonic waves to the test object using the transmitting vibrators and receiving echoes reflected inside the test object using the receiving vibrators; a first image generating unit that generates a first image that visualizes the state of the interior of the test object based on the echo received by the first ultrasonic probe; a flaw detection condition estimation unit that has a prediction model generated using machine learning, and that estimates suitable flaw detection conditions that do not impair the flaw detection performance of the test material and are composed of the number of transmitting transducers, positions of the transmitting transducers, the number of receiving transducers, and positions of the receiving transducers by inputting the first image into the prediction model; a second ultrasonic probe having a plurality of transmitting transducers and a plurality of receiving transducers, the number of combinations of transmitting transducers and receiving transducers being greater than the number of combinations of transmitting transducers and receiving transducers of the first ultrasonic probe, and transmitting ultrasonic waves to the material to be detected using only the transmitting transducers determined by the preferred flaw detection conditions among the plurality of transmitting transducers, and receiving echoes reflected inside the material to be detected using only the receiving transducers determined by the preferred flaw detection conditions among the plurality of receiving transducers; a second image generating unit that generates a first aperture synthetic image that visualizes the state of the interior of the test object based on the echoes received by the second ultrasonic probe; a defect detection unit that detects defects present inside the material to be inspected based on the first aperture synthetic image; An ultrasonic flaw detection device comprising:

2. a training data generating unit that generates training data that is a combination of training images and training suitable inspection conditions; a prediction model generation unit that generates the prediction model by machine learning using the training data; Equipped with The first ultrasonic probe transmits ultrasonic waves to a test material having an internal defect instead of the material to be inspected, and receives echoes reflected inside the test material; The first image generating unit generates a second image that visualizes the internal state of the test material based on the echo received by the first ultrasonic probe, the second ultrasonic probe transmits ultrasonic waves to the test material using all of the transmitting transducers of the second ultrasonic probe, and receives echoes reflected inside the test material using all of the receiving transducers of the second ultrasonic probe; the second image generation unit generates a second aperture synthetic image that visualizes the state of the interior of the test material based on echoes received by all receiving transducers of the second ultrasonic probe; the second ultrasonic probe transmits ultrasonic waves to the test material using m transmitting transducers (m is an integer greater than 1 and not greater than the number of all transmitting transducers) among all transmitting transducers possessed by the second ultrasonic probe, and receives echoes reflected inside the test material using n receiving transducers (n is an integer greater than 1 and not greater than the number of all receiving transducers) among all receiving transducers possessed by the second ultrasonic probe; the second image generation unit generates a third aperture synthetic image that visualizes the state of the interior of the test material based on echoes received by the n receiving transducers of the second ultrasonic probe; As a condition related to the flaw detection performance, an estimated value of the defect position of the test material estimated based on the second aperture synthetic image is compared with an estimated value of the defect position of the test material estimated based on the third aperture synthetic image, and m and n that minimize the product m×n are found under the condition that the absolute value of the difference between the estimated values ​​of the defect position is equal to or less than a predetermined value; In the second ultrasonic probe, when the positions of the m transmitting transducers and the n receiving transducers are determined such that the sum of the processing time required to transmit ultrasound waves from the determined m transmitting transducers and receive echoes from the determined n receiving transducers and the time required to image based on the received echoes is minimized, the teacher data generation unit sets the obtained m as the number of transmitting transducers, the obtained m positions of the transducers as the positions of the transmitting transducers, the obtained n as the number of receiving transducers, and the obtained n positions of the transducers as the positions of the receiving transducers as the learning suitable flaw detection conditions, and generates the teacher data using the second image as the learning image; The ultrasonic flaw detection device according to claim 1 , wherein the prediction model generation unit generates the prediction model by performing machine learning on a machine learning model using the training data.

3. a training data generating unit that generates training data that is a combination of training images and training suitable inspection conditions; a prediction model generation unit that generates the prediction model by machine learning using the training data; Equipped with The first ultrasonic probe transmits ultrasonic waves to a test material having an internal defect instead of the material to be inspected, and receives echoes reflected inside the test material; The first image generating unit generates a second image that visualizes the internal state of the test material based on the echo received by the first ultrasonic probe, the second ultrasonic probe transmits ultrasonic waves to the test material using all of the transmitting transducers of the second ultrasonic probe, and receives echoes reflected inside the test material using all of the receiving transducers of the second ultrasonic probe; the second image generation unit generates a second aperture synthetic image that visualizes the state of the interior of the test material based on echoes received by all receiving transducers of the second ultrasonic probe; the second ultrasonic probe transmits ultrasonic waves to the test material using m transmitting transducers (m is an integer greater than 1 and not greater than the number of all transmitting transducers) among all transmitting transducers possessed by the second ultrasonic probe, and receives echoes reflected inside the test material using n receiving transducers (n is an integer greater than 1 and not greater than the number of all receiving transducers) among all receiving transducers possessed by the second ultrasonic probe; the second image generation unit generates a third aperture synthetic image that visualizes the state of the interior of the test material based on echoes received by the n receiving transducers of the second ultrasonic probe; As a condition related to the flaw detection performance, an estimated value of the size of the flaw in the test material estimated based on the second aperture synthetic image is compared with an estimated value of the size of the flaw in the test material estimated based on the third aperture synthetic image, and m and n that minimize the product m×n are found under the condition that the absolute value of the difference between the estimated values ​​of the flaw size is equal to or less than a predetermined value; In the second ultrasonic probe, when the positions of the m transmitting transducers and the n receiving transducers are determined such that the sum of the processing time required to transmit ultrasound waves from the determined m transmitting transducers and receive echoes from the determined n receiving transducers and the time required to image based on the received echoes is minimized, the teacher data generation unit sets the obtained m as the number of transmitting transducers, the obtained m positions of the transducers as the positions of the transmitting transducers, the obtained n as the number of receiving transducers, and the obtained n positions of the transducers as the positions of the receiving transducers as the learning suitable flaw detection conditions, and generates the teacher data using the second image as the learning image; The ultrasonic flaw detection device according to claim 1 , wherein the prediction model generation unit generates the prediction model by performing machine learning on a machine learning model using the training data.

4. a training data generating unit that generates training data that is a combination of training images and training suitable inspection conditions; a prediction model generation unit that generates the prediction model by machine learning using the training data; Equipped with The first ultrasonic probe transmits ultrasonic waves to a test material having two adjacent defects inside instead of the material to be inspected, and receives echoes reflected inside the test material; The first image generating unit generates a second image that visualizes the internal state of the test material based on the echo received by the first ultrasonic probe, the second ultrasonic probe transmits ultrasonic waves to the test material using all of the transmitting transducers of the second ultrasonic probe, and receives echoes reflected inside the test material using all of the receiving transducers of the second ultrasonic probe; the second image generation unit generates a second aperture synthetic image that visualizes the state of the interior of the test material based on echoes received by all receiving transducers of the second ultrasonic probe; the second ultrasonic probe transmits ultrasonic waves to the test material using m transmitting transducers (m is an integer greater than 1 and not greater than the number of all transmitting transducers) among all transmitting transducers possessed by the second ultrasonic probe, and receives echoes reflected inside the test material using n receiving transducers (n is an integer greater than 1 and not greater than the number of all receiving transducers) among all receiving transducers possessed by the second ultrasonic probe; the second image generation unit generates a third aperture synthetic image that visualizes the state of the interior of the test material based on echoes received by the n receiving transducers of the second ultrasonic probe; As a condition related to the flaw detection performance, m and n that minimize the product m×n are found under the condition that, in the second aperture synthetic image, pixel regions corresponding to the two defects in the test material are separated into two at a predetermined brightness value or more, and in the third aperture synthetic image, pixel regions corresponding to the two defects in the test material are separated into two at a predetermined brightness value or more; In the second ultrasonic probe, when the positions of the m transmitting transducers and the n receiving transducers are determined such that the sum of the processing time required to transmit ultrasound waves from the determined m transmitting transducers and receive echoes from the determined n receiving transducers and the time required to image based on the received echoes is minimized, the teacher data generation unit sets the obtained m as the number of transmitting transducers, the obtained m positions of the transducers as the positions of the transmitting transducers, the obtained n as the number of receiving transducers, and the obtained n positions of the transducers as the positions of the receiving transducers as the learning suitable flaw detection conditions, and generates the teacher data using the second image as the learning image; The ultrasonic flaw detection device according to claim 1 , wherein the prediction model generation unit generates the prediction model by performing machine learning on a machine learning model using the training data.

5. An ultrasonic flaw detection method for detecting defects present inside a material to be inspected, comprising: a first ultrasonic transmission / reception step of using a first ultrasonic probe having a plurality of transmitting vibrators and a plurality of receiving vibrators, transmitting ultrasonic waves to the material to be detected using the transmitting vibrators, and receiving echoes reflected inside the material to be detected using the receiving vibrators; a first image generating step of generating a first image, which is an image of the inside of the test material, using a first image generating unit based on the echo received by the first ultrasonic probe; a flaw detection condition estimation step of estimating suitable flaw detection conditions, which are configured by the number of transmitting transducers, the positions of the transmitting transducers, the number of receiving transducers, and the positions of the receiving transducers, and which do not impair the flaw detection performance of the material to be detected, by inputting the first image into a flaw detection condition estimation unit having a prediction model generated by machine learning; a second ultrasonic transmission / reception step of using a second ultrasonic probe having a plurality of transmitting vibrators and a plurality of receiving vibrators, the number of combinations of transmitting vibrators and receiving vibrators being greater than the number of combinations of transmitting vibrators and receiving vibrators of the first ultrasonic probe, transmitting ultrasonic waves to the material to be detected using only the transmitting vibrators determined by the preferred flaw detection conditions among the plurality of transmitting vibrators, and receiving echoes reflected inside the material to be detected using only the receiving vibrators determined by the preferred flaw detection conditions among the plurality of receiving vibrators; a second image generating step of generating a first aperture synthetic image, which is an image of the inside of the test material, based on the echoes received by the second ultrasonic probe, using a second image generating unit; a defect detection step of detecting defects present inside the material to be inspected based on the first aperture synthetic image using a defect detection unit; The ultrasonic flaw detection method has the following features.

Images