Modal fusion ultrasonic defect signal generation method and system

Abstract

The invention provides a modal fusion ultrasonic defect signal generation method and system, and the method comprises the steps: S1, carrying out the scanning of a detected workpiece through ultrasonic detection equipment, collecting an original A signal, and generating a B scanning image and a C scanning image; s2, determining defect A signals, selecting a plurality of defect A signals from the determined defect A signals, and randomly selecting a plurality of non-defect A signals; s3, converting the defect A signal and the non-defect A signal into an n * n image; s4, inputting a defect A signal into a signal branch of a twinning generative adversarial network method framework, and inputting a generated image into an image branch of the twinning generative adversarial network method framework; s5, generating a first group of A signals through a signal branch; generating a second set of A signals via the image branch; and S6, denoising, filtering and combining the first group of A signals and the second group of A signals, inputting the signals into a discriminator for discrimination, and retaining defect A signals. According to the invention, the imbalance problem of ultrasonic data in training can be relieved.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a modal fusion method and system for generating ultrasonic defect signals. Background Technology

[0002] In modern industrial production, workpieces such as welded parts are prone to defects during production or long-term use, which can easily lead to accidents. Therefore, defect detection plays a vital role in ensuring safe production. In practice, non-destructive testing of workpieces is often required to ensure the usability of normal workpieces.

[0003] Ultrasonic testing is a crucial technique in non-destructive testing (NDT). It relies on an ultrasonic probe to collect ultrasonic signals, then generates images using imaging algorithms, and finally determines defects using technical means or manually. Artificial intelligence has made significant progress in various fields, such as facial recognition, ChatGPT, and image semantic segmentation. However, progress in defect identification has been slow. This is due to two main reasons: firstly, the material, defect type, ultrasonic probe frequency, and noise significantly affect imaging; secondly, defects are extremely rare within the overall workpiece, making defect identification a typical example of an imbalanced problem. Therefore, a large amount of defect data is needed, especially raw defect data, i.e., defect data at the ultrasonic signal (A-signal) level.

[0004] Currently, the generation of ultrasonic defect signals mainly relies on techniques such as inversion and translation, which alleviates the shortcomings of defect A-scan signals to a certain extent. However, this method can only be applied to some materials and some defect categories.

[0005] In view of this, the inventors of this application have designed a modal fusion method and system for generating ultrasonic defect signals in order to overcome the above-mentioned technical problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the lack of automatic identification means for ultrasonic defects in workpieces in the prior art, and to provide a modal fusion ultrasonic defect signal generation method and system.

[0007] The present invention solves the above-mentioned technical problems through the following technical solution:

[0008] This invention provides a modal fusion-based method for generating ultrasonic defect signals, characterized by the following steps: S1, using an ultrasonic testing device to scan the workpiece, collecting raw A signals, and generating B and C scan images; S2, combining the B and C scan images to determine defect A signals, selecting several defect A signals from the determined defect A signals, and randomly selecting several non-defect A signals; S3, converting the defect A signals and non-defect A signals into an n×n image; S4, inputting the defect A signals into the signal branch of a Siamese generative adversarial network (SGN) framework, and inputting the generated images into the image branch of the SGN framework; S5, generating a first group of A signals via the signal branch; generating a second group of A signals via the image branch; S6, after denoising, filtering, and merging the first and second groups of A signals, inputting them into a discriminator for discrimination, retaining the defect A signals.

[0009] According to one or more embodiments of the present invention, in step S2: a number of defect A signals with a signal-to-noise ratio greater than 6dB and having tip diffraction characteristics and a number of defect A signals with a signal-to-noise ratio less than 4dB are selected from the determined defect A signals, and a number of non-defect A signals are randomly selected.

[0010] According to one or more embodiments of the present invention, step S3 includes the following steps: S 31 Determine the value of n, calculate the maximum value of n N, and let the length of signal A be L. A ,but in, Indicates rounding down; S 32 Fill the n×n image with signal A.

[0011] According to one or more embodiments of the present invention, step S 32 Includes the following steps: S 321 Calculate the span of the sampling area: S 322 Double the span of the sampling area to M = 2*M;

[0012] S 323 Number of computational regions N A =L A / M;S 324 Extract the maximum and minimum amplitude values ​​from each region as fill data without losing the characteristics of the ultrasound A-signal; S 325 Perform image filling.

[0013] According to one or more embodiments of the present invention, in step S 325 In this process, when filling an image, the first and last positions are not filled, the last positions are filled, and the remaining positions are filled with the middle value.

[0014] According to one or more embodiments of the present invention, in step S5, noise is added to the signal branch, and a first group of A signals is generated using a sequence generation network; noise is added to the image branch, and a batch of images is generated through an image generation network; then the images are converted back into signals to generate a second group of A signals.

[0015] This invention also provides a modal fusion ultrasonic defect signal generation system, characterized in that the modal fusion ultrasonic defect signal generation system employs the modal fusion ultrasonic defect signal generation method described above. The ultrasonic defect signal generation system includes: a signal detection device for collecting raw A signals and generating B-scan and C-scan images; a signal filtering device for combining the B-scan and C-scan images to determine defective A signals, selecting several defective A signals from the determined defective A signals, and randomly selecting several non-defective A signals; a signal conversion device for converting defective A signals and non-defective A signals into an n×n image; a signal processing device for inputting defective A signals into the signal branch of a Siamese generative adversarial network (SGN) framework, inputting the generated images into the image branch of the SGN framework; generating a first group of A signals via the signal branch; generating a second group of A signals via the image branch; and a signal discrimination device for inputting the first and second groups of A signals, after denoising and filtering, into a discriminator for discrimination, retaining the defective A signals.

[0016] According to one or more embodiments of the present invention, the method for converting signals by the signal conversion device includes the steps of: determining the value of n, calculating the maximum value of n N, and assuming the length of signal A is L. A ,but in, This indicates rounding down; it fills the n×n image with signal A.

[0017] The present invention also provides an electronic device, characterized in that it includes a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions being executed by the processor to implement the modal fusion ultrasonic defect signal generation method as described above.

[0018] The present invention also provides a readable storage medium, characterized in that the readable storage medium stores a program or instructions that, when executed by a processor, implement the modal fusion ultrasonic defect signal generation method as described above.

[0019] The positive and progressive effects of this invention are as follows:

[0020] The modal fusion-based ultrasonic defect signal generation method of the present invention has at least the following advantages:

[0021] The present invention provides a modal fusion ultrasonic defect signal generation method and system that utilizes artificial intelligence to automatically discover internal defects and automatically generate defect A signals. This can supplement the defect data in the training of deep neural networks, alleviate the imbalance problem of ultrasonic data in training, i.e., the defect data is much less than the normal data, and achieve a more general and better generalization performance ultrasonic defect recognition model. Attached Figure Description

[0022] The above and other features, properties and advantages of the present invention will become more apparent from the following description taken in conjunction with the accompanying drawings and embodiments, in which the same reference numerals always denote the same features, wherein:

[0023] Figure 1 This is a schematic flowchart of the ultrasonic defect signal generation method based on modal fusion of the present invention.

[0024] Figure 2 This is a framework diagram of the Siamese Generative Adversarial Network (Siamese-GAN) method in the modal fusion ultrasonic defect signal generation method of this invention.

[0025] Figure 3 This is an example diagram illustrating the sampling problem in the modal fusion ultrasonic defect signal generation method of the present invention.

[0026] Figure 4a This is a schematic diagram of the original defect A signal.

[0027] Figure 4b This is a schematic diagram of the A signal generated by a standard GAN.

[0028] Figure 4c This is a schematic diagram of the A signal generated before denoising filtering.

[0029] Figure 4d This is a schematic diagram of the A-signal generated by the modal fusion ultrasonic defect signal generation method of the present invention. Detailed Implementation

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0031] Embodiments of the invention will now be described in detail with reference to the accompanying drawings. Preferred embodiments of the invention will now be described in detail, examples of which are illustrated in the drawings. Wherever possible, the same reference numerals will be used in all the drawings to denote the same or similar parts. Furthermore, although the terminology used herein is selected from commonly known and used terminology, some terms mentioned in this specification may have been chosen by the applicant at his or her discretion, and their detailed meanings are explained in the relevant sections of the description herein. Moreover, the invention should be understood not only by the actual terms used, but also by the meaning implied by each term.

[0032] like Figures 1-2 As shown, the present invention provides a method for generating ultrasonic defect signals by modal fusion, the method comprising the following steps:

[0033] Step S1: Use an ultrasonic testing device to scan the workpiece to be tested, collect the original A signal, and generate B and C scan images;

[0034] Step S2: Combine the B scan image and the C scan image to determine the defect A signal, select several defect A signals from the determined defect A signals, and randomly select several non-defect A signals.

[0035] Step S3: Convert the defect A signal and the non-defect A signal into an n×n image;

[0036] Step S4: Input the defect A signal into the signal branch of the Siamese-GAN method framework, and input the generated image into the image branch of the Siamese-GAN method framework.

[0037] Step S5: Generate the first group of A signals via signal branch; generate the second group of A signals via image branch;

[0038] Step S6: After the first group of A signals and the second group of A signals are denoised, filtered and merged, they are input into the discriminator for discrimination, and the defective A signals are retained.

[0039] Preferably, in step S1, ultrasonic testing equipment such as an ultrasonic microscope or a phased array system is used to scan the workpiece to be tested, collect the original A signal, and generate B and C scan images based on it.

[0040] Figure 1 This is a simplified flowchart of the implementation process of the modal fusion ultrasonic defect signal generation method of the present invention; Figure 2 This is a framework diagram of the Siamese-GAN method.

[0041] The modal fusion ultrasonic defect signal generation method of the present invention is a general ultrasonic defect A signal generation method. It can supplement the defect data in the training of deep neural networks, alleviate the imbalance problem, and achieve a more general and better generalization performance ultrasonic defect recognition model.

[0042] In a preferred embodiment of the ultrasonic defect signal generation method of modal fusion of the present invention, in step S2: a number of defect A signals with a signal-to-noise ratio greater than 6dB and tip diffraction characteristics and a number of defect A signals with a signal-to-noise ratio less than 4dB are selected from the determined defect A signals, and a number of non-defect A signals are randomly selected.

[0043] Several defect A signals with a signal-to-noise ratio greater than 6dB and tip diffraction characteristics are usually easily identifiable and obvious defect A signals. The ratio of such defect A signals to the adjacent noise signals is greater than 6dB, and they have tip diffraction characteristics or a certain length.

[0044] Several defect A signals with a signal-to-noise ratio of less than 4dB, preferably several defect A signals with a low signal-to-noise ratio of less than 4dB, which are usually less obvious defect signals.

[0045] In a preferred embodiment of the modal fusion ultrasonic defect signal generation method of the present invention, step S3 includes the following steps:

[0046] Step S 31 Determine the value of n, calculate the maximum value of n N, and let the length of signal A be L. A ,but in, This indicates rounding down; typically n is 32 or 64.

[0047] Step S 32 The A signal is applied to an n×n image.

[0048] As mentioned above, the length of signal A is L. A , The image size is n×n, where n≤N.

[0049] The following problems often arise in methods that fill images with A-signal:

[0050] The A signal collected after the ultrasound probe passes through the data acquisition card is itself a number from 0 to 255. Therefore, each data element in the A signal can correspond to a grayscale pixel in an image. However, typically n×n <L A For example, in the experiment of this invention, L A =3000, but the converted image only needs to be 32×32. Therefore, two problems need to be solved in the conversion process: (1) the sampling problem; (2) the tail problem.

[0051] As a preferred embodiment of the modal fusion ultrasonic defect signal generation method of the present invention, step S 32 Includes the following steps:

[0052] Step S 321 Calculate the span of the sampling area:

[0053] Step S 322 Double the span of the sampling area to M = 2*M;

[0054] Step S 323 Number of computational regions N A =L A / M;

[0055] Step S 324 Extract the maximum and minimum amplitude values ​​in each region as filler data without losing the characteristics of the ultrasound A-signal;

[0056] Step S 325 Perform image filling.

[0057] The above implementation provides a solution to the sampling problem.

[0058] Regarding the issue of sampling points: Ultrasound signals are very special signals, and important feature points should not be lost in the generated image. Feature points include the points with the maximum amplitude of the positive half-wave and the minimum amplitude of the negative half-wave in each sampling region. For example... Figure 3 As shown, point P1 represents the maximum amplitude of the positive half-wave, and point P2 represents the minimum amplitude of the negative half-wave. These two points should not be missed during the sampling process.

[0059] The solution to the sampling problem in the above embodiments is illustrated with examples based on experimental data from the method of the present invention:

[0060] Calculate the span of the sampling area: For example,

[0061] Double the span of the sampling area M = 2 * M, for example, M = 2 × 3 = 6.

[0062] Number of computational regions N A =L A / M, for example, N A =3000 / 6=500; then extract two values ​​from each region, namely the maximum amplitude and the minimum amplitude.

[0063] Extracting filling data N without losing ultrasound A-signal characteristics A *2, for example, 500 × 2 = 1000 significant numbers.

[0064] As a preferred embodiment of the ultrasonic defect signal generation method of modal fusion of the present invention, in step S 325 In this process, when filling an image, the first and last positions are not filled, the last positions are filled, and the remaining positions are filled with the middle value.

[0065] The above implementation provides a solution to the tail problem.

[0066] Regarding the tail problem: When sampling again, a tail problem will occur. The actual data needed may be larger than the number of valid samples. If zeros are filled in, it will change the characteristics of the A signal itself, that is, it will oscillate around the middle value.

[0067] Therefore, in this embodiment, the remaining positions are filled with the middle value, and the first and second positions are not filled, while the last position is filled, because signal A is a timing signal.

[0068] In a preferred embodiment of the ultrasonic defect signal generation method of modal fusion of the present invention, in step S5, noise is added to the signal branch, and a first group of A signals is generated using a sequence generation network; noise is added to the image branch, and a batch of images is generated through an image generation network; then the images are converted back into signals to generate a second group of A signals.

[0069] The ultrasonic defect signal generation method based on modal fusion of this invention can generate defect A signals in batches, providing data for automatic defect identification. Test results are as follows... Figures 4a to 4d As shown. This invention can automatically generate defect A signals, alleviating the imbalance problem during training and laying the foundation for further exploration of automatic identification models for ultrasonic defects.

[0070] The modal fusion-based ultrasonic defect signal generation method of this invention converts the defect A signal into a defect A signal image based on the characteristics of the defect A signal. Furthermore, during the defect A signal → defect A signal image conversion process, the point extraction and tail problems are solved through point extraction and filling methods, ensuring that no feature points of the A signal are lost.

[0071] The present invention provides a modal fusion-based ultrasonic defect signal generation method, which is a modal fusion (homogeneous heteromodal) adversarial generative ultrasonic defect A signal generation method: the original defect A signal is input into the first branch of the Siamese Generative Adversarial Network (Siamese-GAN), i.e., the signal branch, and the two-dimensional image generated by the defect A signal is input into the second branch of the Siamese Generative Adversarial Network (Siamese-GAN), i.e., the image branch. The two branches share weights for training, and then the signals generated by the two branches are merged and output, and finally input into a discriminator to determine whether it is a defect signal.

[0072] The modal fusion ultrasonic defect signal generation method of this invention mainly inputs the A signal and the image after A signal conversion into the modal fusion adversarial deep neural network model to generate defect A signal, which provides a data foundation for training the defect recognition network algorithm and solves the data imbalance problem of defect recognition.

[0073] This invention also provides a modal fusion ultrasonic defect signal generation system, wherein the modal fusion ultrasonic defect signal generation system employs the modal fusion ultrasonic defect signal generation method described above, and the ultrasonic defect signal generation system includes:

[0074] A signal detection device is used to collect the raw A signal and generate B and C scan images;

[0075] A signal filtering device is used to combine B-scan images and C-scan images to determine defect A signals, select several defect A signals from the determined defect A signals, and randomly select several non-defect A signals.

[0076] A signal conversion device is used to convert defect A signal and non-defect A signal into an n×n image;

[0077] A signal processing device is used to input defect A signals into the signal branch of the Siamese generative adversarial network (SGI) method framework, input the generated image into the image branch of the Siamese generative adversarial network (SGI) method framework, generate a first set of A signals via the signal branch, and generate a second set of A signals via the image branch.

[0078] The signal discrimination device is used to input the first group of A signals and the second group of A signals into the discriminator for discrimination after denoising, filtering and merging, and retain the defective A signals.

[0079] As a preferred embodiment of the ultrasonic defect signal generation system for modal fusion of the present invention, the method of signal conversion by the signal conversion device includes the steps of: determining the value of n, calculating the maximum value of n N, and assuming the length of signal A is L. A ,but in, This indicates rounding down; it fills the n×n image with signal A.

[0080] The present invention also provides an electronic device, including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions being executed by the processor to implement the modal fusion ultrasonic defect signal generation method as described above.

[0081] The present invention also provides a readable storage medium storing a program or instructions that, when executed by a processor, implement the modal fusion ultrasonic defect signal generation method described above.

[0082] The modal fusion ultrasonic defect signal generation method and system of this invention utilizes artificial intelligence to automatically discover internal defects and can automatically generate defect A signals, alleviating the imbalance problem of ultrasonic data during training, i.e., defect data is much less than normal data, laying the foundation for further exploration of automatic identification models for ultrasonic defects.

[0083] The present invention provides a modal fusion method and system for generating ultrasonic defect signals. It mainly inputs the A signal and the image after A signal conversion into a modal fusion adversarial deep neural network model to generate defect A signals, providing a data foundation for training the defect recognition network algorithm and solving the data imbalance problem in defect recognition.

[0084] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.

Claims

1. A method for generating ultrasonic defect signals through modal fusion, characterized in that, The ultrasonic defect signal generation method includes the following steps: S1. Use ultrasonic testing equipment to scan the workpiece to be tested, collect the original A signal, and generate B scan image and C scan image; S2. Combining the B scan image and the C scan image, determine the defect A signal, select several defect A signals from the determined defect A signals, and randomly select several non-defect A signals. S3. Convert the defect A signal and the non-defect A signal into an n×n image; S4. Input the defect A signal into the signal branch of the Siamese generative adversarial network method framework, and input the generated image into the image branch of the Siamese generative adversarial network method framework. S5. Generate the first group of A signals via the signal branch; generate the second group of A signals via the image branch; S6. After denoising, filtering and merging the first group A signal and the second group A signal, input them into the discriminator for discrimination and retain the defective A signal.

2. The method for generating ultrasonic defect signals by modal fusion as described in claim 1, characterized in that, In step S2: Select several defect A signals with a signal-to-noise ratio greater than 6dB and tip diffraction characteristics, and several defect A signals with a signal-to-noise ratio less than 4dB from the determined defect A signals, and randomly select several non-defect A signals.

3. The method for generating ultrasonic defect signals by modal fusion as described in claim 1, characterized in that, Step S3 includes the following steps: S 31 Determine the value of n, calculate the maximum value of n N, and let the length of signal A be L. A ,but in, Indicates rounding down; S 32 Fill the n×n image with signal A.

4. The method for generating ultrasonic defect signals by modal fusion as described in claim 3, characterized in that, The step S 32 Includes the following steps: S 321 Calculate the span of the sampling area: S 322 Double the span of the sampling area to M = 2*M; S 323 Number of computational regions N A =L A / M; S 324 Extract the maximum and minimum amplitude values ​​in each region as filler data without losing the characteristics of the ultrasound A-signal; S 325 Perform image filling.

5. The method for generating ultrasonic defect signals by modal fusion as described in claim 4, characterized in that, In step S 325 In this process, when filling an image, the first and last positions are not filled, the last positions are filled, and the remaining positions are filled with the middle value.

6. The method for generating ultrasonic defect signals by modal fusion as described in claim 1, characterized in that, In step S5, noise is added to the signal branch, and a sequence generation network is used to generate the first group of A signals; noise is added to the image branch, and a batch of images is generated through an image generation network. Then, the images are converted back into signals to generate the second group of A signals.

7. A modal fusion-based ultrasonic defect signal generation system, characterized in that, The modal fusion ultrasonic defect signal generation system employs the modal fusion ultrasonic defect signal generation method as described in any one of claims 1-6, wherein the ultrasonic defect signal generation system comprises: A signal detection device is used to collect the raw A signal and generate B and C scan images; A signal filtering device is used to combine B-scan images and C-scan images to determine defect A signals, select several defect A signals from the determined defect A signals, and randomly select several non-defect A signals. A signal conversion device is used to convert defect A signal and non-defect A signal into an n×n image; A signal processing device is used to input defect A signals into the signal branch of the Siamese generative adversarial network (SGI) method framework, input the generated image into the image branch of the Siamese generative adversarial network (SGI) method framework, generate a first set of A signals via the signal branch, and generate a second set of A signals via the image branch. The signal discrimination device is used to input the first group of A signals and the second group of A signals after noise reduction and filtering into the discriminator for discrimination, and retain the defective A signals.

8. The ultrasonic defect signal generation system based on modal fusion as described in claim 7, characterized in that, The signal conversion method of the signal conversion device includes the following steps: determining the value of n, calculating the maximum value of n N, and assuming the length of signal A is L. A ,but in, This indicates rounding down; it fills the n×n image with signal A.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to implement the modal fusion ultrasonic defect signal generation method as described in any one of claims 1-6.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the modal fusion ultrasonic defect signal generation method as described in any one of claims 1-6.

Images