JP7618963B2 - Ultrasonic inspection method for round bar material - Google Patents

Description

The present invention relates to an ultrasonic inspection method for round bar material, and in particular to an ultrasonic inspection method that can easily distinguish and detect surface defects and surface layer defects that occur inside the round bar material just below the surface.

When detecting flaws near the surface of a round bar material, the transverse waves of the ultrasonic beam Ub for flaw detection are used with the refraction angle (sector scan angle) set to about 45 degrees, as shown in Figure 7. However, with this method, the flaw echo signals (Figures 8 (1) and 8 (2)) from the surface flaw M1 (Figure 7 (1)) opening on the surface of the round bar material M and the surface flaw M2 (Figure 7 (2)) occurring just below the surface of the round bar material M can appear at the same time with almost the same magnitude, making it often difficult to distinguish between the two.

For example, Patent Document 1 discloses a flaw detection method in which different sector scan angles are set and when the magnitude of the flaw echo signal obtained at each sector scan angle exceeds a predetermined threshold, it is determined that there is a surface flaw or a subsurface flaw.

Patent Publication 2012-225887

However, the conventional method described above requires changing the sector scan angle and repeating the same inspection, which takes time, and there is a problem in that it is not possible to quickly inspect round bar material flowing through the line.

Therefore, the present invention aims to solve these problems by providing an ultrasonic flaw detection method for round bar material that can quickly and reliably distinguish between surface flaws and subsurface flaws.

In order to achieve the above object, in the first invention, an ultrasonic wave for angle beam inspection (Ub) is transmitted toward a round bar material (M) while scanning the round bar material (M) in a direction along the peripheral surface thereof, and reflected ultrasonic waves reflected and returned by surface defects (M1) of the round bar material (M) or surface defects (M2) occurring inside the round bar material (M) just below the surface are received as defect echo signals, and defect echo images are created from each scanning position of the ultrasonic wave for angle beam inspection (Ub) and the time change in the signal intensity of the defect echo signal at the scanning position, and a required number of the defect echo images are convoluted as learning data images to perform neural network analysis. In this ultrasonic inspection method for round bar material, the flaw echo image is fed to a convolutional neural network (CNN) (3) for learning, and a new flaw echo image is fed to the trained convolutional neural network (3) to determine whether the flaw corresponding to the flaw echo image is the surface flaw (M1) or the surface flaw (M2). In the flaw echo image, predetermined areas (X) containing flaw echo signals are cut out for the surface flaw (M1) and the surface flaw (M2), respectively, and images extracted from the predetermined area (X) at scanning positions fewer than all scanning positions in the predetermined area (X) are used as the learning data image.

In the first invention, a flaw echo image is created, and a new flaw echo image is given to a neural network that has been trained in advance using the flaw echo image to determine whether the flaw corresponding to the flaw image is a surface flaw or a subsurface flaw, so that surface flaws and subsurface flaws can be quickly and reliably determined. Also, by appropriately changing the number of extracted flaw echo signals, the accuracy of determination and the speed of determination, which are in a trade-off relationship with this, can be appropriately selected according to the object to be detected.

In the second invention, a phased array probe (1) is used to transmit and receive the ultrasonic waves for angle beam inspection.

According to the second invention, the phased array probe can transmit and scan ultrasonic waves for angle beam inspection, and receive reflected ultrasonic waves in a simple and compact manner.

In the third invention , in the flaw echo image, a portion where the signal intensity of the flaw echo signal is high is displayed with high brightness, and a portion where the signal intensity of the flaw echo signal is low is displayed with low brightness.

In the fourth invention , in the flaw echo image, a portion where the flaw echo signal has a high signal intensity is displayed in low brightness, and a portion where the flaw echo signal has a low signal intensity is displayed in high brightness.

In the fifth invention , the round bar material (M) is a black surface material.

The symbols in parentheses above are for reference purposes only and indicate the corresponding relationship with the specific means described in the embodiments below.

As described above, the ultrasonic inspection method for round bar material of the present invention can quickly and reliably distinguish between surface defects and subsurface defects. More specifically, the method can be used for round bar material that has been through a rolling process, a heat treatment process, and a bending straightening process, and whose surface is covered with black rust, which is a metal oxide. Since black skin material has black rust on the outermost layer, it is particularly difficult to distinguish between surface defects and subsurface defects, but by applying the method of the present invention, it is possible to quickly and reliably distinguish between surface defects and subsurface defects.

FIG. 1 is a diagram showing the configuration of an apparatus for carrying out the method of the present invention. FIG. 13 is a diagram showing an example of a flaw echo image. FIG. 2 is a diagram showing an image obtained by cutting out a predetermined region including a flaw echo signal from a flaw echo image. FIG. 1 is a diagram showing a schematic configuration of a convolutional neural network. FIG. 13 is a diagram showing another example of a flaw echo image. FIG. 13 is a diagram showing yet another example of a flaw echo image. FIG. 1 is a schematic cross-sectional view showing a conventional flaw detection method. FIG. 1 is a diagram showing a flaw echo signal in a conventional flaw detection method.

The embodiment described below is merely an example, and various design improvements made by those skilled in the art that do not deviate from the gist of the present invention are also included in the scope of the present invention.

Figure 1 shows the configuration of an apparatus for carrying out the ultrasonic flaw detection method of the present invention. In Figure 1, a phased array probe 1 is provided facing the outer periphery of a metal round bar M. In the phased array probe 1, a number of ultrasonic transducers (not shown) are arranged adjacent to each other in a known structure so as to form a transmission/reception surface 1a that curves to follow the outer periphery of the round bar M. In this embodiment, a shear wave ultrasonic beam Ub, which is an ultrasonic beam for angle flaw detection that has a sector scan angle of approximately 45 degrees and converges on the surface of the round bar and the surface layer immediately below it, is generated by vibrating a predetermined number of adjacent ultrasonic transducers with an excitation signal having a predetermined time difference output from a control device 2 having a built-in computer. Then, by sequentially moving the range of the predetermined number of ultrasonic transducers to be vibrated, the shear wave ultrasonic beam Ub is scanned in an angle range D of approximately 90 degrees, and the reflected ultrasonic waves from the surface flaw M1 and surface flaw M2 in this range are received again by the phased array probe 1 and output to the control device 2 as a flaw echo signal.

When detecting surface defects M1 and surface defects M2 around the entire circumference of the round bar material M, multiple (four) phased array probes 1 of similar configuration are provided around the entire circumference of the round bar material M, or the round bar material M is rotated or the phased array probe 1 is rotated around the round bar material M.

Figure 2 shows a flaw echo image (B-scope) drawn in the control device 2 by scanning with an ultrasonic beam Ub, with the horizontal axis representing the scanning position and the vertical axis representing the time change in signal strength of the flaw echo signal. Here, the white line portion in the area surrounded by a dashed line in Figure 2 corresponds to the flaw echo signal. Note that the flaw echo image in this case was obtained at a scanning position interval of 97 lines (scanning positions)/45 degrees.

In the flaw echo signal shown in Figures 8(1) and (2) above, the reflection intensity waveform is converted into an absolute value, so it does not represent both positive and negative waveforms, and the resolution is low, which affects the accuracy of flaw determination. In contrast, the flaw echo image (B-scope) shown in Figure 2 is not converted into an absolute value, and the signal intensity reflects both positive and negative flaw echo signals, so it has twice the resolution compared to the flaw echo signals in Figures 8(1) and (2), clearly demonstrating high resolution.

Furthermore, the defect echo image has high resolution because the parts of the defect echo signal with high signal strength are displayed at high brightness and the parts of the defect echo signal with low signal strength are displayed at low brightness, and thus can be suitably used as a learning data image, as described below. Note that the defect echo image can also have high resolution even if the parts of the defect echo signal with high signal strength are displayed at low brightness and the parts of the defect echo signal with low signal strength are displayed at high brightness, and thus can be suitably used as a learning data image, as described below.

A specific region X (Figure 2) containing the defect echo signal from the specified number of obtained defect echo images is cut out for the surface defect M1 and the surface defect M2 as shown in Figures 3 (1) and (2), respectively, and an appropriate number of learning data images and test data images are allocated for each. As an example, 309 learning data images and 1236 test data images were obtained for the surface defect M1, and 260 learning data images and 1040 test data images were obtained for the surface defect M2.

Then, a convolutional neural network (CNN) of a known configuration having an appropriate number of convolutional layers 31, pooling layers 32, and a fully connected layer 33 as shown in FIG. 4, which was constructed in the control device 2, was given training data images (309 images + 260 images) to learn from, and then given test data images (1236 images + 1040 images) to obtain a test accuracy rate. This is shown in Table 1 (1).

As shown in Table 1 (1), the test accuracy rate for both surface defects M1 and surface defects M2 was 100%, which is an extremely good result. However, on the other hand, the data transfer speed, which is problematic when considering use on a production line, is relatively slow, and the processing time for generating the learning model is also relatively long. Note that the data transfer speed ratios and processing time ratios for generating the learning model in Table 1 are ratios when those in Table 1 (3) described below are taken as 1.0.

Therefore, an image (FIG. 5) was obtained by extracting the above-mentioned predetermined region X (FIG. 2) containing the flaw echo signal only at the scanning position of three lines, and from this, 372 learning data images of the surface flaw M1, 1125 test data images, 299 learning data images of the surface flaw M2, and 900 test data images were obtained. Next, the above-mentioned learning data images (372 images + 299 images) were given to the CNN for learning, and then the test data images (1125 images + 900 images) were given to obtain the test accuracy rate. This is shown in Table 1 (2). According to this, the test accuracy rate of the surface flaw M1 was 100%, and the test accuracy rate of the surface flaw M2 was 99.4%. Both the data transfer speed and the processing time for generating the learning model have been greatly improved.

Furthermore, an image (FIG. 6) was obtained by extracting the above-mentioned predetermined region X (FIG. 2) containing the flaw echo signal only at the scanning position of one line, and 298 learning data images of the surface flaw M1 and 1203 test data images were obtained, and 246 learning data images and 992 test data images were obtained for the surface flaw M2. Next, the above-mentioned learning data images (298 images + 246 images) were given to the CNN for learning, and then the test data images (1203 images + 992 images) were given to obtain the test accuracy rate. This is shown in Table 1 (3). According to this, the test accuracy rate for the surface flaw M1 was 98.3%, and the test accuracy rate for the surface flaw M2 was 98.0%. Both the data transfer speed and the processing time for generating the learning model have been further improved.

In this way, if more defect echo signals are extracted to obtain the defect echo image to be provided to the CNN (if the resolution of the defect echo image is increased), the CNN test accuracy rate (judgment result) will improve, but on the other hand, the data transfer speed will be slow and the processing time for generating the learning model will be long, so it is best to optimally select the number of defect echo signals extracted to obtain the defect echo image to be provided to the CCN depending on the application.

1...phased array probe, 2...control device, 3...convolutional neural network, M...round bar material, M1...surface defect, M2...surface defect, Ub...transverse wave ultrasonic beam (ultrasonic for angle beam inspection).

Claims (5)

While transmitting ultrasonic waves for angle beam inspection toward a round bar material, the ultrasonic waves are scanned in a direction along the circumferential surface of the round bar material, and reflected ultrasonic waves reflected and returned from surface defects of the round bar material or surface defects occurring inside the round bar material just below the surface are received as defect echo signals. A defect echo image is created from each scanning position of the ultrasonic waves for angle beam inspection and the time change in signal intensity of the defect echo signal at the scanning position, and a required number of the defect echo images are given to a convolutional neural network (CNN) as learning data images to learn the defect echo image. a new flaw echo image is provided to the trained convolutional neural network to determine whether the flaw corresponding to the flaw echo image is the surface flaw or the subsurface flaw, wherein a predetermined area including a flaw echo signal is cut out from the flaw echo image for each of the surface flaw and the subsurface flaw, and an image extracted from the predetermined area at a scanning position that is fewer than all scanning positions in the predetermined area is used as the training data image. The ultrasonic inspection method for round bar material described in claim 1, in which a phased array probe is used for transmitting and receiving the ultrasonic beam inspection. 3. The ultrasonic inspection method for a round bar material according to claim 1 or 2, wherein the defect echo image is displayed with high brightness in parts of the defect echo signal where the signal strength is high, and with low brightness in parts of the defect echo signal where the signal strength is low. 3. The ultrasonic inspection method for a round bar material according to claim 1 or 2, wherein the defect echo image is displayed with low brightness in parts of the defect echo signal where the signal strength is high, and with high brightness in parts of the defect echo signal where the signal strength is low. 5. The ultrasonic flaw detection method for a round bar material according to claim 1, wherein the round bar material is a black-skinned material.