Ultrasonic flaw detection device and ultrasonic flaw detection method

Abstract

To provide an ultrasonic flaw detection device or the like that can improve defect detection accuracy and detect defect positions with high accuracy in ultrasonic flaw detection using a pulse compression method.SOLUTION: An ultrasonic flaw detection device 100 comprises: an ultrasonic probe 1; a prediction model obtaining unit 4 which obtains a prediction model; a correlation waveform calculation unit 31 which calculates a correlation waveform between a reference wave and a reception wave based on a transmission wave at the ultrasonic probe 1; a correction correlation waveform calculation unit 32 which calculates a correction correlation waveform represented by the function f(t) shown in the expression (1) when a time position at which the correlation waveform shows a peak is set as t0 by inputting the correlation waveform into the prediction model; and a defect detection unit 33 which detects defects based on the correction correlation waveform. In the expression (1), t indicates the time position. a, b are predetermined values that satisfy the following expression (2). |a| / 3≥|b|...(2)SELECTED DRAWING: Figure 2

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 improve the defect detection accuracy and enable the defect position to be detected with high accuracy in ultrasonic flaw detection using a pulse compression method. [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 testing method uses an ultrasonic probe to transmit ultrasonic waves as transmission waves to the material being tested, and receives the ultrasonic waves (echoes) reflected inside the material as reception waves to detect defects inside the material being tested.

[0003] A more specific ultrasonic flaw detection method is the pulse-echo method, which transmits ultrasonic waves with a short time duration as a transmission wave and detects the presence and location of defects that are reflection sources from the change in the amplitude of the received wave over time. For example, when the material to be detected is a metal material such as steel and defects such as cracks, cavities, and inclusions present inside the material are to be detected, the ultrasonic waves propagating inside the material are reflected at the interface between the defects. Echoes from defects (defect echoes) generated by this reflection appear as changes in amplitude in the time waveform of the received wave, so the presence and location of defects present inside the material can be detected from the presence or absence of changes in this amplitude and the time position at which the changes appear.

[0004] The pulse-echo method uses the time waveform of the received wave to detect defects. Therefore, when noise with a relatively large amplitude is superimposed on the received wave compared to the defect echo, the ratio of the defect echo to noise (S / N) drops significantly, and there is a risk that the defect cannot be detected accurately.

[0005] In order to address the above-mentioned problems with the pulse echo method, a technique called pulse compression has recently been used (see, for example, Patent Document 1). Figure 1 is a diagram illustrating a pulse compression method. In the most basic pulse compression method, a waveform modulated by amplitude modulation, frequency modulation (or a combination of these), or the like, shown by the dashed lines in Figures 1(a) and 1(b), is used as the transmitted wave. Then, the time position of the transmitted wave is shifted relative to the received wave, shown by the solid lines in Figures 1(a) and 1(b), and a correlation waveform, such as that shown in Figure 1(c), is calculated, which indicates how closely the received wave matches the transmitted wave at each shifted time position. In the correlation waveform, the horizontal axis represents the time position (the time position of the transmitted wave) and the vertical axis represents the degree of match (the degree of match between the transmitted waveform and the received waveform). The pulse compression method assumes that echoes reflected inside the test material will appear in the received wave with a waveform similar to that of the transmitted wave. Therefore, at the time position where the received wave and the time-shifted transmitted wave coincide (the time position shown in Figure 1(b)), the degree of match of the correlation waveform shown in Figure 1(c) is high, while at the shifted time position (the time position shown in Figure 1(a)), the degree of match of the correlation waveform shown in Figure 1(c) is low. Furthermore, random noise with a waveform different from the transmitted wave will not match highly regardless of the time position. Therefore, even if random noise other than the transmitted and received waves is present in the test material, the pulse compression method focuses only on signals with a high degree of match, allowing it to extract only the received wave generated in response to the transmitted wave, enabling high-precision measurements without being affected by noise. Furthermore, if the received wave for which the correlation waveform shown in Figure 1(c) is calculated contains a defect echo resulting from the transmitted wave being reflected by a defect inside the material being inspected, the time position (b) in Figure 1(c) will correspond to the time position corresponding to the round-trip distance to the defect. Therefore, if the sound speed in the material being inspected is determined in advance, the defect position can be detected based on the time position of the point with the highest correlation match. Furthermore, since the correlation waveform match value has a positive correlation with the magnitude of the defect echo, it is also possible to estimate the magnitude of the defect echo, and therefore the size of the defect, from the magnitude of this match.

[0006] The advantage of using pulse compression is that by devising the waveform of the transmitted wave, it is possible to narrow the range of time positions where the degree of coincidence of the correlation waveform is large (thus improving the accuracy of detecting the defect position), and to increase the S / N ratio of the correlation waveform (thus improving the accuracy of detecting the defect).This gives the appearance of obtaining a signal waveform compressed in the time axis direction, hence the name pulse compression method.

[0007] However, the pulse compression method has the problem that if the waveform of the received wave changes from that of the transmitted wave, the correlation waveform deteriorates, making it impossible to determine the time position where the degree of match is high, and therefore making it impossible to measure accurately. Specifically, the range of time positions where the degree of match of the correlation waveform is high becomes wide, resulting in a correlation waveform with a poor S / N ratio. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Patent No. 7186816 Summary of the Invention [Problem to be solved by the invention]

[0009] 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 improve the accuracy of defect detection and can detect the defect position with high accuracy in ultrasonic flaw detection using the pulse compression method. [Means for solving the problem]

[0010] In order to solve the above problems, the present inventors conducted extensive research and came up with the idea of ​​correcting a degraded correlation waveform in the pulse compression method to an appropriate correlation waveform (a correlation waveform with a high S / N ratio and a narrow time position range where the degree of coincidence of the correlation waveforms is high) by using a prediction model generated by machine learning. They then came up with the idea that if defects present inside the material to be detected are detected based on this corrected correlation waveform (corrected correlation waveform), the accuracy of defect detection will be improved and the defect position can be detected with high accuracy. The present invention has been completed based on the above findings of the present inventors.

[0011] That is, in order to solve the above-mentioned problems, the present invention provides an ultrasonic flaw detection device that detects defects present inside a material to be detected, comprising: an ultrasonic probe that transmits ultrasonic waves to the material to be detected as a transmission wave and receives echoes reflected inside the material to be detected as a reception wave; a prediction model acquisition unit that acquires a prediction model; a correlation waveform calculation unit that calculates a correlation waveform between a reference wave based on the transmission wave and the reception wave; a corrected correlation waveform calculation unit that inputs the correlation waveform into the prediction model and calculates a corrected correlation waveform expressed by the function f(t) shown in the following equation (1) when the time position at which the correlation waveform shows a peak is t0; and a defect detection unit that detects defects present inside the material to be detected based on the corrected correlation waveform.

number

[0012] In the present invention, a "reference wave" is a waveform having the same amplitude and frequency components as a transmitted wave. In equation (1) of the present invention, "t=t0" is not limited to a single time position. If the received wave used to generate the correlation waveform contains defect echoes reflected from multiple defects, multiple time positions t=t0 will exist accordingly. Furthermore, in equation (1), "t≠t0" is not necessarily limited to representing all time positions excluding t=t0, but also includes a range of time positions excluding t=t0 and the vicinity of t=t0 (i.e., t0-α≦t≦t0+α, where α is an infinitesimal positive real number). In this case, within the range of time positions t of t0-α≦t≦t0+α, f(t) is an appropriate function with a width, such as a normal distribution, where f(t)=a when t=t0 and f(t)=b when t=t0-α and t=t0+α.

[0013] According to the present invention, the correlation waveform calculation unit calculates a correlation waveform between a reference wave based on a transmission wave transmitted from an ultrasonic probe and a reception wave received by the ultrasonic probe. In other words, a correlation used in the pulse compression method is obtained. Then, the corrected correlation waveform calculation unit calculates a corrected correlation waveform using the prediction model acquired by the prediction model acquisition unit. Specifically, when a correlation waveform (correlation waveform before correction) is input to this prediction model, a corrected correlation waveform expressed by the function shown in equation (1) is output. In other words, even if the correlation waveform is degraded, it is corrected to an appropriate correlation waveform (as shown in equation (1), a correlation waveform in which the degree of coincidence of the correlation waveform is high (the range of time position t (t=t0) where f(t)=a is narrow and the S / N is high (S / N=|a| / |b|≧3))). Therefore, if the defect detection unit detects defects present inside the material being inspected based on this appropriately corrected correlation waveform, the defect detection accuracy is improved compared to conventional pulse compression methods (pulse compression methods that do not correct the correlation waveform), and the defect position can be detected with high accuracy. In addition, the prediction model used in the corrected correlation waveform calculation unit of the ultrasonic flaw detection device of the present invention may be a prediction model generated by another ultrasonic flaw detection device and used in the corrected correlation waveform calculation unit, or the ultrasonic flaw detection device of the present invention itself may generate the prediction model.

[0014] Preferably, the prediction model acquisition unit calculates a learning correlation waveform, which is a correlation waveform between a reference wave based on an arbitrary transmitted wave and a received wave corresponding to the arbitrary transmitted wave, and generates a learning corrected correlation waveform based on the learning correlation waveform, which is expressed by the function f(t) shown in equation (1) and satisfies equation (2) when the time position at which the learning correlation waveform shows a peak is t0, and acquires the prediction model, which is generated by training a machine learning model, either internally or by obtaining it from an external source, so that the learning corrected correlation waveform is output when the learning correlation waveform is input.

[0015] According to the above-described preferred configuration, when a learning correlation waveform (correlation waveform before correction) is input, the prediction model acquisition unit acquires a prediction model generated by training a machine learning model so that the prediction model is expressed by the function shown in equation (1) and outputs a learning corrected correlation waveform that satisfies equation (2), either by generating it internally (i.e., by the prediction model acquisition unit itself) or by obtaining it externally (i.e., by obtaining a prediction model generated by another device). The prediction model acquired by the prediction model acquisition unit is generated by training a machine learning model as described above, and therefore, by performing sufficient machine learning, it is possible to relatively easily obtain a prediction model that can calculate an appropriate corrected correlation waveform that satisfies equations (1) and (2).

[0016] In the present invention, in order to improve the prediction accuracy of the prediction model (correction accuracy of the corrected correlation waveform calculation section), it is preferable to use different prediction models depending on the type of defect. That is, preferably, the prediction model acquisition unit acquires a plurality of different prediction models according to the types of the defects.

[0017] According to the above preferred configuration, each prediction model corresponding to a defect type is generated for each defect type by, for example, machine learning using a learning correlation waveform and a corrected learning correlation waveform corresponding to the defect type, and is used depending on the type of defect to be detected. For example, if the type of defect to be detected is a cavity, a prediction model generated by machine learning using a learning correlation waveform and a corrected learning correlation waveform calculated for the cavity is used, and if the type of defect to be detected is an inclusion, a prediction model generated by machine learning using a learning correlation waveform and a corrected learning correlation waveform calculated for the inclusion is used. This is expected to improve the prediction accuracy of the prediction model (the correction accuracy of the corrected correlation waveform calculation unit), further improve the defect detection accuracy, and enable the defect position to be detected with even greater accuracy.

[0018] In the present invention, the learning correlation waveform may be generated using a test material instead of the material to be inspected, or may be generated using ultrasonic simulation. In the former case, the learning correlation waveform is preferably calculated as a correlation waveform between a reference wave based on the transmitted wave and the received wave, by the ultrasonic probe transmitting ultrasonic waves as a transmitted wave to a test material having an internal defect instead of the material to be inspected, and receiving an echo reflected inside the test material as a received wave. In the latter case, the learning correlation waveform is preferably calculated by calculating, by ultrasonic simulation, the received wave, which is an echo reflected inside the test material when an arbitrary ultrasonic wave is transmitted as a transmission wave to the test material, and calculating the correlation waveform between the reference wave based on the transmitted wave and the received wave. It is not limited to either the former or the latter, but it is also possible to generate a part of the learning correlation waveform using a test material instead of the material to be inspected, and calculate the remainder of the learning correlation waveform by ultrasonic simulation.

[0019] In addition, in order to solve the above-mentioned problems, the present invention is also provided as an ultrasonic flaw detection method for detecting defects present inside a material to be detected, comprising: an ultrasonic transmission / reception step of using an ultrasonic probe to transmit ultrasonic waves to the material to be detected as a transmission wave and receiving echoes reflected inside the material to be detected as a reception wave; a prediction model acquisition step of using a prediction model acquisition unit to acquire a prediction model; a correlation waveform calculation step of using a correlation waveform calculation unit to calculate a correlation waveform between a reference wave based on the transmission wave and the reception wave; a corrected correlation waveform calculation step of using a corrected correlation waveform calculation unit to input the correlation waveform into the prediction model, thereby calculating a corrected correlation waveform represented by the function f(t) shown in the following equation (1), when the time position at which the correlation waveform shows a peak is t0; and a defect detection step of using a defect detection unit to detect defects present inside the material to be detected based on the corrected correlation waveform.

number

[0020] According to the present invention, in ultrasonic flaw detection using the pulse compression method, the accuracy of defect detection is improved and the defect position can be detected with high accuracy. [Brief explanation of the drawings]

[0021] [Figure 1] FIG. 1 is a diagram for schematically explaining a pulse compression method. [Figure 2] 1 is a diagram illustrating a schematic configuration of an ultrasonic flaw detection device according to an embodiment of the present invention. [Figure 3] 2 is a diagram illustrating how the correlation waveform calculated by the correlation waveform calculation unit 31 shown in FIG. 1 changes. FIG. [Figure 4]3 is a diagram illustrating the state of a correlation waveform calculated by the correlation waveform calculation unit 31 shown in FIG. 2 for a material S to be detected that has a defect. [Figure 5] 3 is a diagram for explaining the operation of the teacher data generating unit 41 shown in FIG. 2. FIG. DETAILED DESCRIPTION OF THE INVENTION

[0022] An ultrasonic flaw detector according to one embodiment of the present invention will be described below with reference to the accompanying drawings as needed. Fig. 2 is a diagram showing a schematic configuration of an ultrasonic flaw detection device according to this embodiment. As shown in Fig. 2, the ultrasonic flaw detection device 100 according to this embodiment is a device that detects defects present inside a material S to be detected, and includes an ultrasonic probe 1, a flaw detector 2, a calculation processing unit 3, and a prediction model acquisition unit 4. The calculation processing unit 3 includes a correlation waveform calculation unit 31, a corrected correlation waveform calculation unit 32, and a defect detection unit 33, and the prediction model acquisition unit 4 includes a teacher data generation unit 41 and a prediction model generation unit 42. Each of the components included in the ultrasonic flaw detection device 100 will be described below.

[0023] <Ultrasonic probe 1> The ultrasonic probe 1 is placed opposite the material to be detected S, transmits ultrasonic waves U to the material to be detected S as a transmission wave, and receives as a received wave an echo reflected inside the material to be detected S. If a defect F exists inside the material to be detected S and is located on the propagation path of the ultrasonic waves U, the received wave will contain a defect echo E reflected from the defect, which appears as a change in amplitude in the time waveform of the received wave. 2 has a single transducer, and performs two-dimensional scanning relative to the material to be detected S as needed. However, the ultrasonic probe 1 is not limited to this, and it is also possible to use a linear array type ultrasonic probe having a plurality of transducers arranged in a line, or a matrix array type ultrasonic probe having a plurality of transducers arranged in a matrix.

[0024] <Flaw detector 2> The flaw detector 2 is electrically connected to the 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 U (transmitted waves) from the transducer possessed by the ultrasonic probe 1, a receiver for receiving echoes (received waves) at the transducer possessed by the ultrasonic probe 1, an amplifier for amplifying an analog flaw detection signal corresponding to the received wave output from the receiver, and an A / D converter for converting the analog signal output from the amplifier into a digital signal.

[0025] <Calculation processing unit 3> The arithmetic processing unit 3 is configured, for example, by a general-purpose computer electrically connected to the flaw detector 2, and stores programs for causing the arithmetic processing unit 3 to function as the correlation waveform calculation unit 31, the corrected correlation waveform calculation unit 32, and the defect detection unit 33. The arithmetic processing unit 3 can then function as the correlation waveform calculation unit 31, the corrected correlation waveform calculation unit 32, and the defect detection unit 33 by executing these programs.

[0026] [Correlation waveform calculation unit 31] The correlation waveform calculation unit 31 calculates a correlation waveform between a reference wave based on a transmission wave transmitted from the ultrasound probe 1 and a received wave. Specifically, a digital signal identical to the signal output from the flaw detector 2 (a pulser provided in the flaw detector 2) to the ultrasonic probe 1 (this signal is for causing the ultrasonic probe 1 to transmit a transmission wave, and has the same amplitude and frequency components as the transmission wave) is branched off from the flaw detector 2 and input to the correlation waveform calculation unit 31 as a reference wave. Also input is a flaw detection signal obtained from the reception wave received by the ultrasonic probe 1, i.e., a digital flaw detection signal output from the flaw detector 2 (an A / D converter provided in the flaw detector 2) (hereinafter, for convenience, this flaw detection signal input to the correlation waveform calculation unit 31 will also be referred to as a reception wave). The correlation waveform calculation unit 31 calculates a correlation waveform between the reference wave and reception wave input as described above.

[0027] Fig. 3 is a diagram for explaining how the correlation waveform calculated by the correlation waveform calculation unit 31 changes. Specifically, Fig. 3 shows the results of calculating the correlation waveform between the transmitted wave and the received wave by calculating, using a known ultrasonic simulation, what the received wave will be for each model when the bottom echo generated by the reflection of ultrasonic waves on the bottom surface (the surface opposite to the surface where the ultrasonic waves are incident) of the material S to be inspected, in two models that are defect-free and have different average grain sizes of the structure that constitute the material S to be inspected, and Assuming the transmitted wave shown in Figure 3(a), the received wave calculated for the model with a large average grain size of tissue shown in Figure 3(b) and the received wave calculated for the model with a small average grain size of tissue shown in Figure 3(c) will both be distorted waveforms that differ from the transmitted wave waveform. This is because the attenuation rate of ultrasound differs for each frequency component of the transmitted wave, so the amplitude component of the transmitted wave does not change uniformly. Furthermore, as can be seen by comparing the waveform of the received wave shown in Figure 3(b) with the waveform of the received wave shown in Figure 3(c), the way in which the waveform distorts will differ depending on the average grain size of the tissue.

[0028] Figure 3(d) shows the correlation waveform between the transmitted wave and the received wave when it is assumed that the received wave is identical to the transmitted wave shown in Figure 3(a). Figure 3(e) shows the correlation waveform between the transmitted wave shown in Figure 3(a) and the received wave shown in Figure 3(c). As can be seen by comparing the correlation waveforms shown in Figure 3(d) and 3(e), when the waveform of the received wave changes from the waveform of the transmitted wave (Figure 3(e)), the range of time positions where the degree of match of the correlation waveforms is large becomes wider than when there is no change (Figure 3(d)), resulting in a correlation waveform with a poor S / N ratio.

[0029] The example shown in Figure 3 is for a test material S that does not contain defects, but even for a test material S that does contain defects, the waveform of the defect echo will change from the waveform of the transmitted wave depending on the shape and type of the defect, resulting in a deterioration of the correlation waveform. Fig. 4 is a diagram for explaining the state of the correlation waveform calculated by the correlation waveform calculation unit 31 for a material S to be inspected that has a defect. Specifically, Fig. 4 shows the results of calculating the correlation between the transmitted wave and the received wave by using a known ultrasonic simulation to calculate what the received wave will be like, including a defect echo generated by the reflection of ultrasonic waves at the defect, for a model of the material S to be inspected that has a defect. As shown in Fig. 4(a), when the received wave containing a defect echo changes from the waveform of the transmitted wave (not shown) and becomes a distorted waveform, the correlation waveform shown in Fig. 4(b) also deteriorates. As a result, the range of time positions where the correlation waveform matches well becomes wider, and the S / N ratio in defect detection deteriorates, making it difficult to determine the time position of a peak that should occur according to the defect position.

[0030] [Corrected correlation waveform calculation unit 32] The corrected correlation waveform calculation unit 32 inputs the correlation waveform into the prediction model, and calculates a corrected correlation waveform expressed by a function f(t) shown in the following equation (1), assuming that the time position at which the correlation waveform shows a peak is t0.

number

[0031] [Defect detection section 33] The defect detection unit 33 detects defects present inside the material to be inspected S by calculating the distance to the defect from the sound speed in the material to be inspected S and the time position at which the corrected correlation waveform shows a peak based on the corrected correlation waveform calculated by the corrected correlation waveform calculation unit 32. Specifically, for example, if the degree of coincidence of the corrected correlation waveform is greater than or equal to a predetermined threshold value, it is possible to determine that a defect exists inside the material S to be inspected, and to detect the defect position based on the time position at which the corrected correlation waveform shows a peak.

[0032] <Prediction model acquisition part 4> The prediction model acquisition unit 4 acquires a prediction model. The prediction model acquisition unit 4 is configured, for example, by a general-purpose computer, and stores programs for functioning as the teacher data generation unit 41 and the prediction model generation unit 42. The prediction model acquisition unit 4 executes these programs and causes them to function as the teacher data generation unit 41 and the prediction model generation unit 42, thereby generating and acquiring a prediction model. Note that the prediction model acquisition unit 4 does not generate a prediction model itself using the teacher data generation unit 41 and the prediction model generation unit 42, but may instead acquire a prediction model similar to the one generated using the teacher data generation unit 41 and the prediction model generation unit 42 by obtaining it from an external device via a communication line or the like (not shown). When the calculation processing unit 3 and the prediction model acquisition unit 4 are configured from general-purpose computers, they may be configured from separate computers or from the same computer.

[0033] [Teacher data generation unit 41] FIG. 5 is a diagram for explaining the operation of the teacher data generating unit 41. As shown in FIG. The training data generating unit 41 generates training data that is a combination of the learning correlation waveform and the learning corrected correlation waveform. Specifically, as shown in FIG. 5( a), the teacher data generating unit 41 generates a learning correlation waveform by calculating the correlation between a reference wave based on an arbitrary transmission wave and a received wave corresponding to the arbitrary transmission wave. For example, the learning correlation waveform is generated by the ultrasonic probe 1 transmitting ultrasonic waves as transmission waves to a test material having an internal defect instead of the material to be detected S, receiving echoes reflected inside the test material as received waves, and the teacher data generating unit 41 calculating the correlation between the reference wave based on the transmission wave and the received wave. Alternatively, the learning correlation waveform is generated by the teacher data generating unit 41 transmitting arbitrary ultrasonic waves as transmission waves to the material to be detected S, calculating received waves that are echoes reflected inside the material to be detected S by ultrasonic simulation, and calculating the correlation between the reference wave based on the transmission wave and the received wave. It is also possible to employ an embodiment in which a portion of the learning correlation waveform is generated using a test material instead of the material to be detected S, and the remaining portion of the learning correlation waveform is calculated by ultrasonic simulation.

[0034] Next, as shown in FIG. 5(b), when the time position at which the learning correlation waveform (FIG. 5(a)) shows a peak is set to t0, the teacher data generation unit 41 generates a learning corrected correlation waveform that is expressed by the function f(t) shown in the above-mentioned equation (1) and satisfies the above-mentioned equation (2). In addition, when generating a learning corrected correlation waveform, b in equation (2) is set to a value corresponding to a degree of match of 0 or substantially 0 (i.e., even if the degree of match value contains a noise component, a small amount is acceptable), and a is preferably set to a value corresponding to a degree of match that is larger than the value in the case of b and can be regarded as a peak (a degree of match that is substantially different from 0 and can be regarded as having a peak in the pulse compression method).

[0035] In the example shown in Fig. 5(b), in Equation (1), "t≠t0" represents all time positions except t=t0. However, the present invention is not limited to this, and as shown in Fig. 5(c), "t≠t0" of the function f(t) representing the learning corrected correlation waveform also represents a range of time positions excluding t=t0 and the vicinity of t=t0 (i.e., t0-α≦t≦t0+α, where α is an infinitesimal positive real number). In this case, within the range of time position t of t0-α≦t≦t0+α, f(t) employs an appropriate function with a width, such as a normal distribution, where f(t)=a when t=t0 and f(t)=b when t=t0-α and t=t0+α. In both cases of FIG. 5(b) and FIG. 5(c), it is preferable to set b=0.

[0036] The learning corrected correlation waveform can also be calculated by applying a wavelet transform or a short-time frequency Fourier transform to the learning correlation waveform. Furthermore, when the prediction model generation unit 42 described below generates multiple different prediction models according to the type of defect (thus, when the corrected correlation waveform calculation unit 32 uses multiple different prediction models according to the type of defect), it is preferable to generate multiple sets of training data according to the type of defect. Furthermore, when the prediction model generation unit 42 described below generates multiple different prediction models according to the distance from the ultrasonic probe 1 (i.e., according to the deterioration of the waveform based on the propagation distance of the ultrasonic waves) (therefore, when the corrected correlation waveform calculation unit 32 uses multiple different prediction models according to the distance from the ultrasonic probe 1), it is preferable to generate multiple sets of training data according to the distance from the ultrasonic probe 1. Furthermore, when different ultrasonic probes 1 are used depending on the material S to be inspected, and the prediction model generation unit 42 described below generates multiple different prediction models depending on the ultrasonic probe 1 to be used (thus, when multiple different prediction models depending on the ultrasonic probe 1 to be used in the corrected correlation waveform calculation unit 32 are used), it is preferable to generate multiple sets of training data depending on the ultrasonic probe 1. Furthermore, when generating training data to be used for machine learning of the same prediction model, it is preferable to use the same transmission wave (reference wave). The teacher data generation unit 41 outputs the generated teacher data to the prediction model generation unit 42.

[0037] [Prediction model generation unit 42] The prediction model generation unit 42 generates a prediction model by training a machine learning model using the training data generated by the training data generation unit 41 so that when a training correlation waveform is input, the machine learning model outputs a training corrected correlation waveform. As the machine learning model, a neural network related to deep learning can be used, but is not limited to this, and a regression model, for example, can be used. Furthermore, a deep learning model can also be used by regarding the training correlation waveform as an image. When the corrected correlation waveform calculation unit 32 uses a plurality of different prediction models according to the type of defect, etc., the prediction model generation unit 42 generates a plurality of different prediction models. The prediction model generating unit 42 outputs the generated prediction model to the corrected correlation waveform calculating unit 32, and the corrected correlation waveform calculating unit 32 stores the input prediction model. Therefore, when the correlation waveform calculated by the correlation waveform acquisition unit 31 is input to the prediction model in the corrected correlation waveform calculation unit 32, if the time position at which the correlation waveform shows a peak is set to t0, it becomes possible to calculate a corrected correlation waveform expressed by the function f(t) shown in the above-mentioned equation (1).

[0038] As described above, according to the ultrasonic flaw detection device 100 of this embodiment, the correlation waveform calculation unit 31 calculates a correlation waveform between a reference wave based on a transmission wave transmitted from the ultrasonic probe 1 and a reception wave received by the ultrasonic probe 1. In other words, a correlation used in the pulse compression method is acquired. Then, the corrected correlation waveform calculation unit 32 calculates a corrected correlation waveform expressed by the function shown in Equation (1) using the prediction model acquired by the prediction model acquisition unit 4. This prediction model is generated by training a machine learning model so that, when a learning correlation waveform (correlation waveform before correction) is input, the machine learning model outputs a corrected learning correlation waveform expressed by the function shown in Equation (1). In other words, even if the learning correlation waveform is degraded, the machine learning model is trained to correct it to an appropriate correlation waveform (as shown in Equation (1)), a correlation waveform in which the degree of correlation waveform agreement is high (the range of time position t (t = t0) where f(t) = a is narrow and the S / N ratio is high (S / N = |a| / |b| ≧ 3)). Therefore, by using this prediction model, the corrected correlation waveform calculation unit 32 can calculate a corrected correlation waveform, which is an appropriate correlation waveform, even if the correlation waveform calculated by the correlation waveform calculation unit 31 is degraded. Therefore, if the defect detection unit 33 detects defects present inside the material S to be inspected based on this appropriately corrected correlation waveform, the defect detection accuracy is improved compared to conventional pulse compression methods (pulse compression methods that do not correct the correlation waveform), and the defect position can be detected with high accuracy. [Explanation of symbols]

[0039] 1...Ultrasonic probe 2...Flaw detector 3. Processing unit 4. Prediction model acquisition section 31 Correlation waveform calculation section 32 Corrected correlation waveform calculation unit 33 Defect detection section 41. Training data generation unit 42 Prediction model generation unit 100...Ultrasonic flaw detection equipment S...Material to be tested

Claims

1. An ultrasonic flaw detection device for detecting defects present inside a material to be inspected, an ultrasonic probe that transmits ultrasonic waves to the test object as a transmission wave and receives echoes reflected inside the test object as a reception wave; a prediction model acquisition unit that acquires a prediction model; a correlation waveform calculation unit that calculates a correlation waveform between a reference wave based on the transmission wave and the reception wave; a corrected correlation waveform calculation unit that calculates a corrected correlation waveform expressed by a function f(t) shown in the following equation (1) when the time position at which the correlation waveform shows a peak is set to t0 by inputting the correlation waveform to the prediction model; a defect detection unit that detects defects present inside the material to be inspected based on the corrected correlation waveform; An ultrasonic flaw detection device comprising: [Equation 4] In the above formula (1), t represents the time position, and a and b are predetermined values ​​that satisfy the following formula (2). |a| / 3≧|b| ...(2)

2. The prediction model acquisition unit calculating a learning correlation waveform that is a correlation waveform between a reference wave based on an arbitrary transmission wave and a reception wave corresponding to the arbitrary transmission wave; generating a corrected learning correlation waveform based on the learning correlation waveform, the corrected learning correlation waveform being expressed by a function f(t) shown in equation (1) and satisfying equation (2), where t0 is the time position at which the learning correlation waveform shows a peak; The prediction model is generated by training a machine learning model so that the corrected correlation waveform for learning is output when the correlation waveform for learning is input, The ultrasonic flaw detection device according to claim 1 , wherein the ultrasonic flaw detection signal is obtained by generating the signal internally or obtaining the signal externally.

3. The ultrasonic flaw detection device according to claim 1 , wherein the prediction model acquisition unit acquires a plurality of different prediction models according to the types of the defects.

4. The learning correlation waveform is 3. The ultrasonic flaw detection device of claim 2, wherein the ultrasonic probe transmits ultrasonic waves as a transmission wave to a test material having an internal defect instead of the material to be detected, receives an echo reflected inside the test material as a received wave, and calculates a correlation waveform between a reference wave based on the transmitted wave and the received wave.

5. The learning correlation waveform is 3. The ultrasonic flaw detection device of claim 2, wherein when an arbitrary ultrasonic wave is transmitted as a transmission wave to the material to be detected, a received wave, which is an echo reflected inside the material to be detected, is calculated by ultrasonic simulation, and is calculated as a correlation waveform between a reference wave based on the transmitted wave and the received wave.

6. An ultrasonic flaw detection method for detecting defects present inside a material to be inspected, comprising: an ultrasonic wave transmitting / receiving step of transmitting ultrasonic waves as a transmission wave to the test object using an ultrasonic probe and receiving an echo reflected inside the test object as a reception wave; a prediction model acquisition step of acquiring a prediction model using a prediction model acquisition unit; a correlation waveform calculation step of calculating a correlation waveform between a reference wave based on the transmission wave and the reception wave using a correlation waveform calculation unit; a corrected correlation waveform calculation step of calculating a corrected correlation waveform expressed by a function f(t) shown in the following formula (1), when the time position at which the correlation waveform shows a peak is set to t0, by inputting the correlation waveform to the prediction model using a corrected correlation waveform calculation unit; a defect detection step of detecting defects present inside the material to be inspected based on the corrected correlation waveform using a defect detection unit; The ultrasonic flaw detection method has the following features. [Equation 5] In the above formula (1), t represents the time position, and a and b are predetermined values ​​that satisfy the following formula (2). |a| / 3≧|b| ...(2)

Images