Ultrasonic wave flaw detection device and ultrasonic wave flaw detection method

Abstract

To improve the resolution of an ultrasonic image.SOLUTION: An ultrasonic wave flaw detection device detects defects inside an object using ultrasonic waves. The ultrasonic wave flaw detection device includes: a plurality of ultrasonic transmission elements that transmit ultrasonic waves toward the object; a plurality of ultrasonic reception elements that receive intensities of the ultrasonic waves reflected by the object as signals; and a data interpolation section that estimates a signal that is not actually received on the basis of the signals received by the ultrasonic reception elements. The data interpolation section acquires a neural network, which uses coordinates as input to output a corresponding signal, as a trained neural network, by training using coordinates of the plurality of ultrasonic reception elements and the signals received by the plurality of ultrasonic reception elements, and estimates a signal that would be received if the ultrasonic reception elements are present at specific positions by inputting coordinates of the specific positions into the trained neural network for specific positions in an area where the plurality of ultrasonic reception elements are not present.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an ultrasonic flaw detection device and an ultrasonic flaw detection method. [Background technology]

[0002] In many industrial fields, there is a growing trend to use classifiers trained by machine learning methods to extract experience and knowledge from huge amounts of data and use this knowledge to automate processes. In particular, in ultrasonic flaw detection, which is used as a form of non-destructive testing in the manufacturing industry, many machine learning models based on neural networks, including deep learning, have been introduced, and efforts are underway to improve the efficiency of inspections that previously relied on human labor.

[0003] In addition to efforts to detect anomalies by applying deep learning to images of signals obtained by ultrasonic flaw detection, in recent years deep learning has also begun to be introduced to the signals themselves before they are imaged. For example, Patent Document 1 proposes a method for identifying detection targets such as defects using a machine learning model trained on 3D voxel data from a non-destructive testing device.

[0004] Furthermore, Patent Document 2 proposes a deep learning model in which an ultrasonic signal is used as input data and an appropriate corresponding filter control signal is used as output data.

[0005] In parallel with the focus on machine learning, ultrasonic flaw detection systems are using an increasing number of elements in their transducers, as more elements improve spatial resolution. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2022-66637 [Patent Document 2] Japanese Patent Application Publication No. 2023-106098 Summary of the Invention [Problem to be solved by the invention]

[0007] However, as the number of elements increases, manufacturing costs rise sharply, and there are also problems with susceptibility to element failure. This problem is particularly pronounced in linear arrays, where multiple elements are arranged in one dimension, and matrix arrays, where elements are arranged in a grid pattern in two dimensions. In particular, in the manufacturing industry, where ultrasonic waves are used for flaw detection, improving resolution by increasing the number of elements is desirable, but physically increasing the number of elements directly leads to increased costs and increased maintenance load due to failures, making it unacceptable.

[0008] In the application of a deep learning model to 3D voxel data proposed in Patent Document 1, the deep learning model is used with the aim of revealing abnormalities from signals obtained from a predetermined number of elements, and the improvement in resolution of ultrasound images that would be achieved by increasing the number of elements is not taken into consideration at all in the machine learning process.

[0009] Furthermore, the deep learning model proposed in Patent Document 2 also assumes that the number of elements is constant, and does not take into account the improvement in resolution of ultrasound images that would be achieved by increasing the number of elements.

[0010] The present disclosure has been made in consideration of the above circumstances, and provides an ultrasonic flaw detection device and an ultrasonic flaw detection method that can improve the resolution of ultrasonic images. [Means for solving the problem]

[0011] In order to solve the above problem, according to one aspect of the present disclosure, there is provided an ultrasonic flaw detection device that uses ultrasonic waves to detect internal defects of an object, the ultrasonic flaw detection device comprising: a plurality of ultrasonic transmitting elements that transmit ultrasonic waves toward the object; a plurality of ultrasonic receiving elements that receive signals representing the intensity of the ultrasonic waves reflected by the object; and a data interpolation unit that estimates signals that are not actually received based on the signals received by the ultrasonic receiving elements, wherein the data interpolation unit acquires a trained neural network that takes coordinates as input and outputs corresponding signals by learning using the coordinates of the plurality of ultrasonic receiving elements and the signals received by the plurality of ultrasonic receiving elements, and provides an ultrasonic flaw detection device that estimates a signal that would be received if the ultrasonic receiving elements were present at a specific position by inputting the coordinates of the specific position into the trained neural network for a specific position in an area where the plurality of ultrasonic receiving elements are not present.

[0012] In order to solve the above problem, according to one aspect of the present disclosure, there is provided an ultrasonic flaw detection method for detecting internal defects of an object using ultrasonic waves, the ultrasonic flaw detection method including: an ultrasonic flaw detection device having a plurality of ultrasonic transmitting elements that transmit ultrasonic waves toward the object, a plurality of ultrasonic receiving elements that receive signals representing the intensity of ultrasonic waves reflected by the object as signals, and a data interpolation unit that estimates signals that are not actually received based on the signals received by the ultrasonic receiving elements; an acquisition step using the data interpolation unit to learn a neural network that inputs coordinates and outputs corresponding signals using the coordinates of the plurality of ultrasonic receiving elements and the signals received by the plurality of ultrasonic receiving elements, thereby acquiring the trained neural network; and an estimation step for inputting the coordinates of a specific position in an area where the plurality of ultrasonic receiving elements are not present into the trained neural network, thereby estimating a signal that would be received if the ultrasonic receiving element were present at the specific position. [Effects of the Invention]

[0013] According to the present disclosure, the resolution of ultrasound images can be improved. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 2 is a block diagram showing the hardware configuration of the ultrasonic flaw detection device. [Figure 2] FIG. 2 is a plan view of an ultrasonic transmitting element array. [Figure 3] FIG. 2 is a plan view of an ultrasonic receiving element array. [Figure 4] FIG. 2 is a functional block diagram of the ultrasonic flaw detector. [Figure 5] FIG. 10 is a plan view showing an example of a specific position. [Figure 6] 10 is a flowchart of an ultrasonic flaw detection process. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0016] 1 shows the hardware configuration of an ultrasonic flaw detector 10 according to the present disclosure. The ultrasonic flaw detector 10 is a device that detects defects inside an object using ultrasonic waves.

[0017] The target is an object to be inspected when the ultrasonic flaw detector 10 inspects the target for defects, and may be, for example, a strip-shaped steel plate such as a rolled steel plate or a slab in a steel mill, etc., but is not limited to this. The target may be made of other materials than iron, such as aluminum or titanium, and may have various shapes other than a flat strip-like shape, such as a columnar billet, a steel bar, a wire rod, a tubular steel pipe, or an H-shaped cross-section steel.

[0018] Furthermore, defects include, but are not limited to, scratches, inclusions, bubbles, and the like inside the object.

[0019] 1, the ultrasonic flaw detector 10 includes a controller 12. The controller 12 is configured as a device including a general computer.

[0020] The controller 12 includes a central processing unit (CPU) 12A, a read-only memory (ROM) 12B, a random access memory (RAM) 12C, and an input / output interface (I / O) 12D. The CPU 12A, ROM 12B, RAM 12C, and I / O 12D are connected to each other via a bus 12E. The bus 12E includes a control bus, an address bus, and a data bus.

[0021] Furthermore, the ultrasonic transmitting element array 14, the ultrasonic receiving element array 16, the operation unit 18, the display unit 20, the communication unit 22, and the storage unit 24 are connected to the I / O 12D.

[0022] FIG. 2 is a plan view of the ultrasonic transmitter array 14. As shown in FIG. 2, the ultrasonic transmitter array 14 has a configuration in which a plurality of ultrasonic transmitter elements 30 are arranged in M ​​rows and N columns. Note that M≧1 and N≧2. In the example of FIG. 2, M=5, N=7, and the ultrasonic receiver elements 32 are arranged two-dimensionally, but it is also possible to have a configuration in which M=1 and the ultrasonic receiver elements 32 are arranged one-dimensionally. It is also possible to have a configuration in which a single ultrasonic receiver element 32 scans one-dimensionally or two-dimensionally.

[0023] Fig. 3 is a plan view of the ultrasonic receiver array 16. As shown in Fig. 3, the ultrasonic receiver array 16 has a configuration in which a plurality of ultrasonic receiver elements 32 are arranged in m rows x n. Note that m ≥ 1 and n ≥ 2. In the example of Fig. 3, M = 5 and N = 7, and the ultrasonic receiver elements 32 are arranged two-dimensionally, but it is also possible to have a configuration in which m = 1 and the ultrasonic receiver elements 32 are arranged one-dimensionally. It is also possible to have a configuration in which a single ultrasonic receiver element 32 scans one-dimensionally or two-dimensionally.

[0024] 1, each ultrasonic transmitting element 30 of the ultrasonic transmitting element array 14 transmits ultrasonic waves US toward an object X. Each ultrasonic receiving element 32 of the ultrasonic receiving element array 16 receives the intensity of the ultrasonic waves reflected by the object X as a signal RV.

[0025] The operation unit 18 includes, for example, a mouse and a keyboard, and is a mechanism for operating the ultrasonic flaw detector 10 and performing input according to the operator's intentions.

[0026] The display unit 20 is configured by, for example, a liquid crystal display or the like, and is a mechanism for conveying the progress and final results of the processing performed by the ultrasonic flaw detector 10 to the operator by displaying them or the like.

[0027] The communication unit 22 is an interface for performing data communication with an external device such as an external server.

[0028] The storage unit 24 is configured as a non-volatile external storage device such as a hard disk, etc. As shown in Fig. 1, the storage unit 24 stores an ultrasonic flaw detection program 26 and a trained neural network 28.

[0029] 4 is a block diagram showing the functional configuration of the CPU 12A of the ultrasonic flaw detector 10. As shown in FIG.

[0030] The CPU 12A reads and executes the ultrasonic flaw detection program 26 stored in the storage unit 24, thereby functioning as each of the functional units of a data interpolation unit 34 and a defect detection unit 36.

[0031] The data interpolation unit 34 is a functional unit that estimates signals that are not actually received based on signals received by the ultrasonic receiving elements 32 .

[0032] The data interpolation unit 34 acquires a trained neural network 28 by training a neural network that takes coordinates as input and outputs corresponding signals using the coordinates of multiple ultrasonic receiving elements 32 and the signals received by the multiple ultrasonic receiving elements 32, and by inputting the coordinates of an arbitrary position (for example, a specific position in an area where no ultrasonic receiving elements 32 are present) into the trained neural network 28, it estimates the signal that would be received if an ultrasonic receiving element 32 were present at the specific position.

[0033] Note that the trained neural network 28 may be acquired by training a neural network that inputs time in addition to coordinates and outputs a corresponding signal using the coordinates and time of the multiple ultrasonic receiving elements 32 and the signals received by the multiple ultrasonic receiving elements 32. In this case, by inputting the coordinates and time of an arbitrary position (for example, a specific position in an area where no ultrasonic receiving elements 32 are present) into the trained neural network 28, it is possible to estimate the signal that would be received if the ultrasonic receiving element 32 were present at that arbitrary position at that time.

[0034] The neural network used here is called a neural field (NF). A neural field is a neural network that takes spatial coordinates and time as input and outputs a desired signal. The desired signal can be a scalar or a vector. Furthermore, of the spatial coordinates and time input to the neural field, the time can be omitted.

[0035] Neural networks are generally used for image classification and object detection, while neural fields are used for applications such as estimating signals and states that arise from any physical phenomenon, or generating signals that complement missing data signals when generating an image.

[0036] When training a neural network, it is desirable to use as training data, for example, the coordinates of ultrasonic receiving elements 32 located in spatially dispersed positions and the signals received by the ultrasonic receiving elements 32. Spatially dispersed ultrasonic receiving elements 32 are, in other words, ultrasonic receiving elements 32 having various spatial coordinates, and by training a neural field using the coordinates of such ultrasonic receiving elements 32 and the signals received by the ultrasonic receiving elements 32, it is possible to make the neural network learn the behavior of global physical phenomena compared to when training using only the coordinates of ultrasonic receiving elements 32 located in local positions and the signals received by the ultrasonic receiving elements 32 as training data.

[0037] The specific position is preferably set in an area where there are no ultrasonic receiving elements 32. For example, the specific position is set in a position around the ultrasonic receiving elements 32, such as an area between adjacent ultrasonic receiving elements 32, that is different from the position where the existing ultrasonic receiving elements 32 are present. By doing so, it is possible to estimate a signal that would be obtained if an element were hypothetically placed in a position where an element is not physically placed or where an element cannot be placed due to device restrictions or the like.

[0038] By inputting the coordinates of a specific position set in an area where no ultrasonic receiving elements 32 exist into the trained neural network 28 in this way, it is possible to estimate a signal that would be received if an ultrasonic receiving element 32 were present at the specific position. This not only makes it possible to obtain ultrasonic signals from existing ultrasonic receiving elements 32, but also makes it possible to obtain estimated ultrasonic signals even at positions where no ultrasonic receiving elements 32 exist and no received signals would be obtained. Therefore, the same effect as essentially increasing the number (or density) of measurement points by the ultrasonic receiving elements 32 is obtained, and the amount of information used to generate an ultrasonic image can be increased, thereby improving measurement resolution. Therefore, it is possible to generate high-resolution ultrasonic images and detect defects with high accuracy.

[0039] For example, as shown in Fig. 5, by inputting the coordinates of a specific position 38 set in an area between diagonally adjacent ultrasonic receiving elements 32 into the trained neural network 28, it is possible to acquire a signal that would be received if an ultrasonic receiving element 32 were present at the specific position 38. In the example of Fig. 5, the same number of specific positions 38 as the ultrasonic receiving elements 32 are set, so the resolution of the ultrasound image can be improved by two times.

[0040] The defect detection unit 36 ​​is a functional unit that detects defects inside the object X. Specifically, the defect detection unit 36 ​​detects defects inside the object X based on the signals received by the ultrasonic receiving elements 32, or the signals obtained by the ultrasonic receiving elements 32 and the signals estimated by the data interpolation unit 34.

[0041] For example, when generating an ultrasound image by the so-called aperture synthesis method, an ultrasound wave US is transmitted from one ultrasound transmitting element 30 to the object X, and the intensity of the ultrasound reflected by the object X is received as a signal RV by all ultrasound receiving elements 32. This process is performed the same number of times as the number of ultrasound transmitting elements 30. As a result, signals RV are obtained in the number equal to the number of ultrasound transmitting elements 30 x the number of ultrasound receiving elements 32. Then, the defect detection unit 36 ​​generates a three-dimensional ultrasound image of the object X based on the obtained signals RV in the number equal to the number of ultrasound transmitting elements 30 x the number of ultrasound receiving elements 32.

[0042] Furthermore, the defect detection unit 36 ​​may generate an ultrasonic image of the object X based on signals equal to the number of ultrasonic transmitting elements 30 multiplied by the number of ultrasonic receiving elements 32, and the signals estimated by the data interpolation unit 34, i.e., signals that would be received if an ultrasonic receiving element 32 were present at the specific position 38, obtained by inputting the coordinates of the specific position 38 into the trained neural network 28. This makes it possible to improve the resolution of the ultrasonic image compared to when an ultrasonic image is generated without using signals that would be received if an ultrasonic receiving element 32 were present at the specific position 38.

[0043] The defect detection unit 36 ​​may perform known image processing such as binarization and noise removal on the generated ultrasonic image of the object X to detect the position, shape, etc. of the defect. Alternatively, a deep learning model different from the neural field used to estimate the ultrasonic signal may be separately prepared to determine the position and type of the defect from the ultrasonic image. Alternatively, the defect detection unit 36 ​​may output the generated ultrasonic image of the object X to the display unit 20 for display, allowing an operator to visually detect the defect.

[0044] Next, the processing of the ultrasonic flaw detection method executed by the CPU 12A of the ultrasonic flaw detection device 10 will be described. Fig. 6 shows a flowchart of the defect detection method according to the present disclosure. The processing of the ultrasonic flaw detection method shown in Fig. 6 is executed by the CPU 12 reading the ultrasonic flaw detection program stored in the storage unit 24 into the RAM 12C and executing it. Below, a case where an ultrasonic image is generated by the aperture synthesis method will be described.

[0045] In step S100, the CPU 12A causes one of the ultrasonic transmitting elements 30 to transmit an ultrasonic wave US.

[0046] In step S101, the CPU 12A acquires the signals RV received by all the ultrasonic receiving elements 32.

[0047] In step S102, the CPU 12A determines whether or not ultrasonic waves US have been transmitted from all of the ultrasonic transmitting elements 30. If ultrasonic waves US have been transmitted from all of the ultrasonic transmitting elements 30, the process returns to step S100, and ultrasonic waves US are transmitted from the ultrasonic transmitting elements 30 that have not yet transmitted ultrasonic waves US. On the other hand, if ultrasonic waves US have been transmitted from all of the ultrasonic transmitting elements 30, the process proceeds to step S103. In this way, the ultrasonic transmitting elements 30 are sequentially switched to transmit ultrasonic waves US, and the process of receiving signals RV from all of the ultrasonic receiving elements 32 is executed for each of the ultrasonic transmitting elements 30.

[0048] In step S103, the CPU 12A acquires the trained neural network 28 stored in the storage unit 24 by reading it into the RAM 12C.

[0049] In step S104, the CPU 12A inputs the coordinates of the specific position 38 into the trained neural network 28 and executes a process for acquiring the signal output from the trained neural network 28 that would be received if an ultrasonic receiving element 32 were present at the specific position 38, the process being repeated for each specific position 38.

[0050] In step S105, the CPU 12A generates an ultrasound image of the object X based on signals RV obtained by executing the processing of step S101 for the number of ultrasound transmitting elements 30 multiplied by the number of ultrasound receiving elements 32, and on signals obtained in step S104 that would be received if an ultrasound receiving element 32 were present at each of the specific positions 38.

[0051] In step S106, the CPU 12A performs known image processing, including binarization processing, on the ultrasonic image generated in step S105 to detect the position, shape, etc. of the defect.

[0052] In step S107, the CPU 12A outputs the detection results such as the position and shape of the defect detected in step S106 to the display unit 16 for display.

[0053] As described above, in this embodiment, for a specific position 38 in an area where no ultrasonic receiving elements 32 are present, the coordinates of the specific position 38 are input to the trained neural network 28, and an ultrasonic image is generated by estimating the signal that would be received if an ultrasonic receiving element were present at the specific position 38. This improves the resolution of the ultrasonic image compared to when an ultrasonic image is generated using only the signals received by the ultrasonic receiving elements 32. Furthermore, detecting defects in the object X based on such an ultrasonic image enables highly accurate defect detection. Furthermore, since there is no need to add ultrasonic receiving elements 32, the cost of the ultrasonic flaw detection device 10 can be reduced. Furthermore, the specific position 38 can be set anywhere where a signal is needed without any limitations on the number or location. Therefore, the specific position 38 can be set appropriately where a signal is needed, enabling flexible signal acquisition depending on the defect.

[0054] Although the present embodiment has been described above, the present disclosure is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the gist of the present disclosure.

[0055] Furthermore, the configuration of the ultrasonic flaw detection device 10 described in the above embodiment (see Figure 1) is one example, and it goes without saying that unnecessary parts may be deleted or new parts may be added within the scope of the present disclosure.

[0056] Furthermore, in the present embodiment, a case has been described in which the technology of the present disclosure is applied to an ultrasonic flaw detector, but the technology of the present disclosure can also be applied to other non-destructive testing devices. [Explanation of symbols]

[0057] 10 Ultrasonic flaw detection equipment 12 Controller 14 Ultrasonic transmitting element array 16 ultrasonic receiving element array 16 Display 18 Control section 20 Display section 22 Communications Department 24 Memory section 26 Ultrasonic Flaw Inspection Program 28 Trained Neural Networks 30 ultrasonic transmitting element 32 ultrasonic receiving element 34 Data Interpolation Unit 36 Defect detection section 38 Specific location 40 Areas of Interest RV signal US ultrasound X Object

Claims

1. An ultrasonic flaw detector that detects internal defects of an object using ultrasonic waves, a plurality of ultrasonic transmitting elements that transmit ultrasonic waves toward the target; a plurality of ultrasonic receiving elements that receive signals representing the intensities of ultrasonic waves reflected by the object; a data interpolation unit that estimates signals that are not actually received based on signals received by the ultrasonic receiving elements; and The data interpolation unit A neural network that receives coordinates as input and outputs corresponding signals is learned using the coordinates of the plurality of ultrasonic receiving elements and the signals received by the plurality of ultrasonic receiving elements, thereby obtaining a learned neural network; An ultrasonic flaw detection device that estimates a signal that would be received if the ultrasonic receiving element were present at a specific position in an area where the multiple ultrasonic receiving elements are not present by inputting the coordinates of the specific position into the trained neural network.

2. a defect detection unit that detects defects inside the object, The defect detection unit 2. The ultrasonic flaw detection device according to claim 1, wherein defects inside the object are detected based on signals received by the ultrasonic receiving elements, or based on signals obtained by the ultrasonic receiving elements and signals estimated by the data interpolation unit.

3. An ultrasonic flaw detection method for detecting internal defects of an object using ultrasonic waves, comprising: a plurality of ultrasonic transmitting elements that transmit ultrasonic waves toward the target; a plurality of ultrasonic receiving elements that receive signals representing the intensities of ultrasonic waves reflected by the object; a data interpolation unit that estimates signals that are not actually received based on signals received by the ultrasonic receiving elements; Using an ultrasonic flaw detector having Using the data interpolation unit, An acquisition step of acquiring a trained neural network by learning a neural network that receives coordinates as input and outputs corresponding signals using the coordinates of the plurality of ultrasonic receiving elements and the signals received by the plurality of ultrasonic receiving elements; an estimation step of estimating a signal that would be received if the ultrasonic receiving element were present at a specific position in an area where the plurality of ultrasonic receiving elements are not present by inputting coordinates of the specific position into the trained neural network; The ultrasonic flaw detection method has the following features.

4. a defect detection step of detecting a defect inside the object using a defect detection unit, The defect detection step includes:

4. The ultrasonic flaw detection method according to claim 3, further comprising detecting an internal defect of the object based on a signal received by the ultrasonic receiving element, or a signal obtained by the ultrasonic receiving element and a signal estimated by the data interpolation unit.

Images