Method and device for intelligently identifying near-surface defects of large ring piece based on ultrasonic phased array

Abstract

The invention provides a method and a device for intelligently identifying near-surface defects of a large ring piece based on an ultrasonic phased array, and relates to the field of ultrasonic nondestructive testing. The method comprises the following steps: scanning along the surface of the large ring piece through an ultrasonic phased array probe to obtain a ring piece material and surface structure characteristics corresponding to the large ring piece; the method comprises the following steps: constructing ultrasonic phased array detection parameters and an ultrasonic phased array full-focusing imaging algorithm of a surface area of a large ring part according to a ring part material and surface structure characteristics; according to the ultrasonic phased array detection parameters, an ultrasonic phased array probe is controlled to collect ultrasonic detection signals in real time and upload the ultrasonic detection signals to an upper computer; an ultrasonic detection signal is processed in real time through an upper computer, and the ultrasonic detection signal is converted into an ultrasonic full-focus image based on an ultrasonic phased array full-focus imaging algorithm; generating an ultrasonic phased array full-focus detection video according to the ultrasonic full-focus image; and inputting the ultrasonic phased array full-focus detection video into the moving target detection neural network model, and outputting the near surface defect of the large ring based on the spatio-temporal characteristics of the video sequence frame through the moving target detection neural network model. According to the invention, the problem that the existing ultrasonic detection technology and the defect intelligent identification method are poor in near-surface defect detection precision is solved.

Description

Technical Field

[0001] This application relates to the field of ultrasonic non-destructive testing, and particularly to an intelligent recognition method and device for near-surface defects of large ring parts based on ultrasonic phased array. Background Art

[0002] Large ring parts are important supporting and transmission components of high-end equipment, and are widely used in the fields of aerospace, wind energy, and large-scale construction machinery, such as the ring of an aero-engine housing, the ring of a large wind power bearing, the rim of a mining vehicle, and the ring of a large mechanical slewing bearing. Such ring parts have a complex cross-sectional shape and a large wall thickness dimension, and usually have structures such as raceways, flanges, and fillets. During the rolling and forming manufacturing process of large ring parts, non-metallic inclusions on the surface of the ring parts are rolled into the ring parts, forming near-surface defects.

[0003] Currently, the pulse echo ultrasonic testing technology is usually used to identify near-surface defects of large ring parts. However, due to the certain width of the reflected echoes of ultrasonic waves at the workpiece interface and the bottom, the defect signals near the interface and the bottom are blocked, and the defect signals and image features cannot be identified, resulting in a detection blind area near the workpiece interface and the bottom. For an ultrasonic probe with a center frequency of 2.5 - 15 MHz, the near-surface blind area range is usually 2 - 3 mm. At this time, the existing ultrasonic testing technology and defect intelligent recognition method have poor detection accuracy for near-surface defects. Especially for defects within 2 mm near the workpiece surface, it is difficult for the existing ultrasonic intelligent detection method to identify.

[0004] Therefore, there is an urgent need for an intelligent recognition method and device for near-surface defects of large ring parts based on ultrasonic phased array. Summary of the Invention

[0005] This application provides an intelligent recognition method and device for near-surface defects of large ring parts based on ultrasonic phased array, which solves the problem of poor detection accuracy of the existing ultrasonic testing technology and defect intelligent recognition method for near-surface defects.

[0006] In the first aspect of the present application, an intelligent identification method for near-surface defects of large ring parts based on ultrasonic phased array is provided. The method includes: scanning the surface of the large ring part with an ultrasonic phased array probe to obtain the ring part material and surface structure characteristics corresponding to the large ring part; constructing ultrasonic phased array detection parameters and an ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring part according to the ring part material and surface structure characteristics; controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and uploading them to the upper computer; processing the ultrasonic detection signals in real time by the upper computer, and converting the ultrasonic detection signals into ultrasonic full-focus images based on the ultrasonic phased array full-focus imaging algorithm; generating an ultrasonic phased array full-focus detection video according to the ultrasonic full-focus images; inputting the ultrasonic phased array full-focus detection video into a moving target detection neural network model, and outputting the near-surface defects of the large ring part based on the spatio-temporal characteristics of the video sequence frames through the moving target detection neural network model.

[0007] Optionally, controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and uploading them to the upper computer specifically includes: constructing ultrasonic phased array detection parameters for the surface area of the large ring part according to the ring part material and surface structure characteristics, where the ultrasonic phased array detection parameters include the upper surface phased array probe frequency parameter, the lower surface phased array probe frequency parameter, the probe element number parameter, the element spacing parameter, and the element width parameter; controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and uploading them to the upper computer.

[0008] Optionally, constructing an ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring part according to the ring part material and surface structure characteristics, where the ultrasonic phased array full-focus imaging algorithm is used to calculate the full-focus imaging amplitude, specifically includes: constructing an ultrasonic phased array full-focus imaging area according to the probe element number parameter, the element spacing parameter, and the element width parameter; dividing the ultrasonic phased array full-focus imaging area into multiple grid points in a preset manner; obtaining the target coordinate point among the multiple grid points, where the target coordinate is any coordinate point among the multiple grid points; transmitting a preset number of ultrasonic signals by each element of the ultrasonic phased array probe, and receiving the ultrasonic signals simultaneously by each element, where the preset number is determined by the probe element number parameter; calculating the full-focus imaging amplitude corresponding to the target coordinate point according to the multiple ultrasonic signals by the amplitude superposition method.

[0009] Optionally, before inputting the ultrasonic phased array full-focus detection video into the moving target detection neural network model, it is necessary to construct a moving target detection neural network model, which specifically includes: constructing a reference block according to the ring material and surface structure characteristics. The reference block is a large ring with different materials and structures, and the reference block includes reference near-surface defects with different sizes and depths; scanning along the reference block with an ultrasonic phased array probe to collect in real time a reference ultrasonic phased array full-focus detection video containing reference near-surface defects; annotating the reference near-surface defects in the reference ultrasonic phased array full-focus detection video to establish a near-surface defect sample video dataset, and the near-surface defect sample video dataset is expanded through a method combining VideoMix video data augmentation and image data augmentation; training the model based on the near-surface defect sample video dataset to construct a moving target detection neural network model.

[0010] Optionally, the moving target detection neural network model includes a background modeling module, a current frame feature extraction module, and a dynamic feature fusion module. Through the moving target detection neural network model, based on the spatio-temporal features of the video sequence frames, the near-surface defects of the large ring are output, which specifically includes: through the background modeling module, based on the adaptive bilateral filtering algorithm for image noise elimination and the background modeling algorithm, obtaining the background features of each frame image in the ultrasonic phased array full-focus detection video; through the current frame feature extraction module, based on the sequential frame attention mechanism and the ShuffleNetV2 lightweight backbone network, obtaining the current frame features of each frame image in the ultrasonic phased array full-focus detection video; based on the background features and the current frame features, obtaining the spatio-temporal features of the video sequence frames through the dynamic feature fusion module, and outputting the near-surface defects of the large ring according to the spatio-temporal features of the video sequence frames.

[0011] Optionally, through the moving target detection neural network model, based on the spatio-temporal features of the video sequence frames, the near-surface defects of the large ring are output, which specifically includes: through the moving target detection neural network model based on the spatio-temporal features of the video sequence frames, recording the coordinates of the near-surface defects identified in each frame image of the ultrasonic phased array full-focus detection video; performing real-time trajectory tracking on the near-surface defects through the near-surface defect coordinates, and outputting the near-surface defects.

[0012] Optionally, after inputting the ultrasonic phased array full-focus detection video into the moving target detection neural network model and outputting the near-surface defects of the large ring through the moving target detection neural network model based on the spatio-temporal features of the video sequence frames, the method further includes: obtaining the near-surface defect features of the sequence frames in the ultrasonic phased array full-focus detection video according to the moving target detection neural network model; calculating the correlation features of adjacent frames according to the near-surface defect features of the sequence frames; predicting the positions of the near-surface defects in the missing frames according to the correlation features.

[0013] In a second aspect of the present application, an intelligent identification device for near-surface defects of large ring parts based on ultrasonic phased array is provided. The device includes an acquisition module, a full-focus detection video generation module, and a processing module. Among them,

[0014] The acquisition module is used to scan along the surface of the large ring part through an ultrasonic phased array probe to obtain the ring part material and surface structure characteristics corresponding to the large ring part.

[0015] The full-focus detection video generation module is used to construct ultrasonic phased array detection parameters and an ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring part according to the ring part material and surface structure characteristics; control the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and upload them to the upper computer; process the ultrasonic detection signals in real time through the upper computer, and convert the ultrasonic detection signals into ultrasonic full-focus images based on the ultrasonic phased array full-focus imaging algorithm; generate an ultrasonic phased array full-focus detection video according to the ultrasonic full-focus images.

[0016] The processing module is used to input the ultrasonic phased array full-focus detection video into a moving target detection neural network model, and output the near-surface defects of the large ring part based on the spatio-temporal characteristics of the video sequence frames through the moving target detection neural network model.

[0017] In a third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method as described in any one of the above.

[0018] In a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to perform the method as described in any one of the above.

[0019] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0020] 1. The ultrasonic phased array probe is used to scan along the surface of the large ring to obtain the ring material and surface structure characteristics corresponding to the large ring. According to the ring material and surface structure characteristics, the ultrasonic phased array detection parameters and the ultrasonic phased array full focusing imaging algorithm for the surface area of the large ring are constructed. According to the ultrasonic phased array detection parameters, the ultrasonic phased array probe is controlled to collect ultrasonic detection signals in real time and upload them to the upper computer. The upper computer processes the ultrasonic detection signals in real time and, based on the ultrasonic phased array full focusing imaging algorithm, converts the ultrasonic detection signals into ultrasonic full focusing images. An ultrasonic phased array full focusing detection video is generated according to the ultrasonic full focusing images. The ultrasonic phased array full focusing detection video is input into the moving target detection neural network model, and the near-surface defects of the large ring are output based on the spatio-temporal characteristics of the video sequence frames through the moving target detection neural network model. Thus, the motion characteristics of the near-surface defects are extracted using the historical frame information of the ultrasonic detection video, the occluded near-surface defects are effectively identified, and the detection accuracy of the near-surface defects of the large ring is greatly improved.

[0021] 2. According to the ring material and surface structure characteristics, a comparison test block is constructed. The comparison test block is a large ring with different materials and structures, and the comparison test block includes comparison near-surface defects with different sizes and different depths. The ultrasonic phased array probe is used to scan along the comparison test block to collect in real time a comparison ultrasonic phased array full focusing detection video containing the comparison near-surface defects. The comparison near-surface defects in the comparison ultrasonic phased array full focusing detection video are labeled to establish a near-surface defect sample video data set. The near-surface defect sample video data set is expanded through a method combining VideoMix video data enhancement and image data enhancement. Model training is carried out according to the near-surface defect sample video data set to construct a moving target detection neural network model. Thus, by combining VideoMix and image enhancement, the images in the video are data-enhanced and then recombined into a video, and the video is reorganized in terms of time and space, effectively expanding the number of near-surface defect detection video samples. Furthermore, the robustness and generalization ability of the trained moving target detection neural network model for near-surface defects under multi-scale, complex structures, and different material backgrounds are improved, and its recognition stability in high-noise interference, weak defect response, and occlusion environments is enhanced.

[0022] 3. According to the moving target detection neural network model, the sequence frame near-surface defect characteristics in the ultrasonic phased array full focusing detection video are obtained. According to the sequence frame near-surface defect characteristics, the correlation characteristics of adjacent frames are calculated. According to the correlation characteristics, the position of the near-surface defect in the missing frame is predicted. Thus, using the position of the near-surface defect in the known video sequence frames, the motion trajectory of the near-surface defect is predicted to predict the position of the defect in the video missing frame, realizing real-time detection and position tracking of the near-surface defects of the large ring. Description of the Drawings

[0023] Figure 1 It is a schematic flowchart of an intelligent recognition method for near-surface defects of large ring parts based on ultrasonic phased array provided by an embodiment of the present application;

[0024] Figure 2 It is a schematic diagram of the near-surface blind area in ultrasonic phased array detection of large ring parts provided by an embodiment of the present application;

[0025] Figure 3 It is a schematic diagram of an immersion ultrasonic phased array detection system for large ring parts provided by an embodiment of the present application;

[0026] Figure 4 It is a schematic diagram of a contrast test block for near-surface defects of large ring parts provided by an embodiment of the present application;

[0027] Figure 5 It is a schematic diagram of a frame of an ultrasonic phased array detection video sequence for near-surface defects of large ring parts provided by an embodiment of the present application;

[0028] Figure 6 It is a schematic diagram of ultrasonic detection video data enhancement for near-surface defects provided by an embodiment of the present application;

[0029] Figure 7 It is a schematic diagram of the structure of a neural network model for moving target detection provided by an embodiment of the present application;

[0030] Figure 8 It is a schematic diagram of the structure of a background modeling module provided by an embodiment of the present application;

[0031] Figure 9 It is a schematic diagram of the SFAM-ShuffleNetV2 network structure provided by an embodiment of the present application;

[0032] Figure 10 It is a schematic diagram of the structure of a current frame feature extraction module provided by an embodiment of the present application;

[0033] Figure 11 It is a schematic diagram of the structure of a dynamic feature fusion module provided by an embodiment of the present application;

[0034] Figure 12 It is a schematic diagram of the intelligent recognition result of near-surface defects of large ring parts by ultrasonic images provided by an embodiment of the present application;

[0035] Figure 13 It is a schematic diagram of the defect tracking result of near-surface ultrasonic detection video of large ring parts provided by an embodiment of the present application;

[0036] Figure 14 It is a schematic diagram of the modules of an intelligent recognition device for near-surface defects of large ring parts based on ultrasonic phased array provided by an embodiment of the present application;

[0037] Figure 15 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0038] Explanation of reference numerals: 141, acquisition module; 142, full-focus detection video generation module; 143, processing module; 1501, processor; 1502, communication bus; 1503, user interface; 1504, network interface; 1505, memory. Specific embodiments

[0039] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments.

[0040] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. As used in the specification of the present application, the singular forms "a", "an", "the", "above", "the foregoing", "this" are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to and includes any or all possible combinations of one or more of the listed items.

[0041] Hereinafter, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0042] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0043] Please refer to Figure 1 , which shows a schematic flow chart of a method for intelligent identification of near-surface defects of large ring parts based on ultrasonic phased array provided by an embodiment of the present application. The flow chart mainly includes the following steps: S101 to S106.

[0044] In step S101, an ultrasonic phased array probe is used to scan along the surface of the large ring part to obtain the ring part material and surface structure characteristics corresponding to the large ring part.

[0045] Specifically, for the current pulse-echo ultrasonic testing technology, since the reflected echoes of ultrasonic waves at the workpiece interface and bottom have a certain width, the defect signals near the interface and bottom are blocked, making it impossible to identify the defect signals and image features, forming a detection blind area near the workpiece interface and bottom. For ultrasonic probes with a center frequency of 2.5 - 15 MHz, the near-surface blind area range is usually 2 - 3 mm. Please refer to Figure 2 , which shows a schematic diagram of the near-surface blind area in the ultrasonic phased array detection of a large ring in the embodiment of the present application. In addition, due to the severe scattering of ultrasonic waves on the curved surface of the workpiece, the noise interference in the near-surface area of the workpiece is strong, making it more difficult to identify near-surface defects. Therefore, the existing ultrasonic testing technology and defect intelligent recognition methods have poor detection accuracy for near-surface defects. Especially for defects within 2 mm from the workpiece surface, the existing ultrasonic intelligent detection methods are difficult to identify. The detection of near-surface defects still relies on manual experience to analyze data, which is both time-consuming and laborious.

[0046] The technical solution in this application is to solve the above defects. First, scan the surface of the large ring: Place the large ring in a container filled with a coupling agent, and adopt a fully immersed ultrasonic phased array detection method to make the ultrasonic phased array probe scan uniformly along the entire circumferential surface of the large ring in a coupled state. During the scanning process, receive the echo signals through the probe and combine with the preset excitation parameters to obtain the acoustic response data of the surface area of the large ring. Based on the acoustic response data, parse out the ring material and surface structure characteristics corresponding to the large ring through the recognition model integrated in the upper computer. The ring materials include alloy steel, stainless steel, titanium alloy, superalloy, aluminum alloy, and other high-strength and tough metal materials, and the surface structure characteristics include structural composition characteristics such as curvature, flange, fillet, and raceway. Further, according to the changes in the propagation characteristics of ultrasonic waves in different materials and different geometric contour regions, combined with the acoustic impedance, reflection coefficient, and echo amplitude distribution characteristics, extract the basic physical property parameters for subsequent imaging modeling, as the basic input for constructing the ultrasonic phased array detection parameters and the full-focus imaging algorithm. This step provides structural adaptability support for the subsequent detection area parameter configuration and imaging accuracy control, ensuring accurate detection and image restoration under complex topography and material composition conditions. Please refer to Figure 3 , which shows a schematic diagram of a large-ring water-immersion ultrasonic phased array detection system provided in the embodiment of the present application.

[0047] Step S102, construct the ultrasonic phased array detection parameters and the ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring according to the ring material and surface structure characteristics.

[0048] Specifically, for different types of ring material and complex and variable surface structure features, an ultrasonic phased array probe with a suitable frequency range is selected. For the upper surface area, a high-frequency probe above 10 MHz is preferably used, and for the lower surface area, a low-frequency probe below 5 MHz is selected. Combining with the geometric profile of the ring, parameters such as the number of probe elements, element spacing, and element width are determined to form a set of detection parameters with structural matching. On this basis, the imaging area range is set and grid division is performed, and an ultrasonic phased array full-focus imaging algorithm model based on full matrix acquisition and amplitude superposition is established to improve the imaging resolution and defect identifiability in the near-surface area.

[0049] Step S103: According to the ultrasonic phased array detection parameters, control the ultrasonic phased array probe to collect ultrasonic detection signals in real time and upload them to the host computer.

[0050] Specifically, based on the detection parameters such as the probe frequency, number of elements, element spacing, and element width that have been constructed, drive the ultrasonic phased array probe to perform element excitation and reception operations in sequence during the scanning process to complete the real-time acquisition of full matrix ultrasonic signals. The collected ultrasonic detection signals are processed by the probe and the control module and then uploaded to the host computer in real time through the data transmission interface, providing continuous raw echo data input for subsequent image reconstruction and defect identification.

[0051] In a possible implementation manner, step S103 further includes: constructing ultrasonic phased array detection parameters for the surface area of the large ring according to the ring material and surface structure features. The ultrasonic phased array detection parameters include the upper surface phased array probe frequency parameter, the lower surface phased array probe frequency parameter, the probe element number parameter, the element spacing parameter, and the element width parameter; according to the ultrasonic phased array detection parameters, control the ultrasonic phased array probe to collect ultrasonic detection signals in real time and upload them to the host computer.

[0052] Specifically, for the detection of the upper surface area of the large ring, since the near-surface defects are small in size and shallow in distribution, in order to obtain a high-resolution image, a high-frequency phased array probe with a center frequency above 10 MHz should be selected, and according to the surface structure and topography features of the ring, set the number of probe elements to N, the element spacing to p, and the element width to e to form a high-density sound field coverage of the near-field area of the upper surface; for the lower surface area at the far end of the large ring, since the signal needs to penetrate a thicker material layer, in order to enhance the penetration depth and suppress the material attenuation effect, a low-frequency phased array probe with a center frequency below 5 MHz should be selected, and configure the corresponding probe structure parameters with reference to the upper surface setting method; the above detection parameters are input as control instructions to the phased array instrument, driving the ultrasonic phased array probe to execute the signal acquisition process according to the preset element excitation sequence and time step, and at the same time uploading the collected full matrix ultrasonic detection signals to the host computer in real time through the data interface to ensure the continuity and integrity of the acoustic data required for image reconstruction and defect identification.

[0053] Step S104: Process the ultrasonic detection signal in real time through the host computer, and based on the ultrasonic phased array full focus imaging algorithm, convert the ultrasonic detection signal into an ultrasonic full focus image.

[0054] Specifically, after the host computer receives the full matrix detection signal from the phased array probe, it calls the preset ultrasonic phased array full focus imaging algorithm to perform delay compensation and energy superposition processing on the signal, so as to achieve the focused expression of the echo energy in the detection area, thereby constructing an ultrasonic image with higher spatial resolution and stronger contrast.

[0055] In a possible implementation manner, step S104 further includes: constructing an ultrasonic phased array full focus imaging area according to the probe element number parameter, element spacing parameter, and element width parameter; dividing the ultrasonic phased array full focus imaging area into multiple grid points in a preset manner; obtaining the target coordinate points among the multiple grid points, where the target coordinate is any coordinate point among the multiple grid points; transmitting a preset number of ultrasonic signals through each element of the ultrasonic phased array probe, and receiving the ultrasonic signals simultaneously by each element, and the preset number is determined by the probe element number parameter; calculating the full focus imaging amplitude corresponding to the target coordinate point through the amplitude superposition method according to the multiple ultrasonic signals.

[0056] Specifically, in order to obtain a higher-resolution ultrasonic image in the near-surface area, the ultrasonic full focus imaging algorithm is selected to reconstruct the image in the near-surface area; determine the ultrasonic phased array full focus imaging area, where the imaging area width is D = p(N - 1) + e, and the imaging area depth is the near-surface L n . Divide the imaging area into a×b grid points, where the coordinate of any point is F(x, y); use each element of the ultrasonic phased array probe to sequentially transmit ultrasonic signals, and all elements receive the signals simultaneously, and a total of N×N groups of ultrasonic signals are collected; use the full matrix signal to perform amplitude superposition at each grid point F(x, y) in turn, and the full focus imaging amplitude of this coordinate point is obtained as:

[0057]

[0058]

[0059] Among them, I F (x, y) represents the full focus imaging amplitude of the grid point F(x, y), S ij represents the signal transmitted by the i-th element and received by the j-th element, t ij (x, y) is the total time for the ultrasonic wave to propagate from the transmitting element to the receiving element, L 1i and L 2i are the distances that the ultrasonic waves emitted by the elements propagate in water and steel respectively, L 2j and L 1jare the distances traveled by the ultrasonic wave after reflection in steel and water respectively, and c1 and c2 are the sound velocities in water and steel respectively.

[0060] Step S105: Generate an ultrasonic phased array full focus detection video based on the ultrasonic full focus image.

[0061] Specifically, as the ultrasonic phased array probe continuously scans along the surface of the large ring, the ultrasonic full focus image sequences obtained by the upper computer in real-time processing are arranged in chronological order, and this image sequence is constructed into a detection video with continuity between frames, thereby forming an ultrasonic phased array full focus detection video that reflects the dynamic evolution of the surface structure and near-surface defects of the large ring, providing a spatio-temporal feature input basis for subsequent defect recognition based on the frames of the video sequence.

[0062] In a possible implementation manner, step S105 further includes: constructing a comparison test block according to the material and surface structure characteristics of the ring. The comparison test block is a large ring with different materials and structures, and the comparison test block includes comparison near-surface defects with different sizes and different depths; scanning along the comparison test block through the ultrasonic phased array probe to collect in real-time a comparison ultrasonic phased array full focus detection video containing the comparison near-surface defects; annotating the comparison near-surface defects in the comparison ultrasonic phased array full focus detection video to establish a near-surface defect sample video data set, and the near-surface defect sample video data set is expanded by combining VideoMix video data enhancement and image data enhancement; training a model according to the near-surface defect sample video data set to construct a moving target detection neural network model. Please refer to Figure 4 , which shows a schematic diagram of a comparison test block for near-surface defects of a large ring provided by an embodiment of the present application.

[0063] Specifically, the process of constructing a moving target detection neural network model is as follows: Simulate the material and structural characteristics of large ring parts, design contrast test blocks of large ring parts with different materials and structures, and process near-surface defects with different sizes and depths on the contrast test blocks; Use an ultrasonic phased array instrument and a probe to detect all the contrast test blocks. The ultrasonic phased array probe scans along the surface of the contrast test blocks. During the movement of the probe, a near-surface defect ultrasonic phased array detection video is collected in real time; Label the near-surface defects in the ultrasonic detection video, generate an XML tag file, and establish a near-surface defect sample video data set; Construct a moving target detection algorithm model and train it through the ultrasonic phased array detection video data set to obtain a trained moving target detection model. Moreover, when establishing the near-surface defect ultrasonic phased array detection sample video data set, the method of combining VideoMix video data augmentation and image data augmentation is used to expand the video data set; Randomly select three videos from the original video data, convert the videos into sequence frame images, and randomly process the images by adding Gaussian noise, dropout, and changing the contrast, etc., to enhance the richness of the sequence frame images in the videos; Then randomly select regions to reconstruct the videos, including the center point coordinates, length, and width of the videos, and merge the three videos into time, space, and spacetime respectively to obtain new videos, increasing the richness of the large ring part near-surface defect ultrasonic phased array detection video data set. Please refer to Figure 5 , which shows a schematic diagram of the sequence frames of the ultrasonic phased array detection video of the near-surface defects of a large ring part provided by an embodiment of the present application.

[0064] Step S106, input the ultrasonic phased array full-focus detection video into the moving target detection neural network model, and output the near-surface defects of the large ring part based on the spatio-temporal characteristics of the video sequence frames through the moving target detection neural network model.

[0065] Specifically, the ultrasonic phased array full-focus detection video is image-enhanced through the moving target detection neural network model, so as to obtain enhanced spatio-temporal characteristics, and the near-surface defects of the large ring part are output based on the spatio-temporal characteristics of the video sequence frames. Please refer to Figure 6 , which shows a schematic diagram of the enhancement of the ultrasonic detection video data of the near-surface defects provided by an embodiment of the present application.

[0066] In a possible implementation, step S106 further includes: through the background modeling module, performing image noise elimination and background modeling algorithm based on the adaptive bilateral filtering algorithm to obtain the background features of each frame of the ultrasonic phased array full-focus detection video; through the current frame feature extraction module, based on the sequential frame attention mechanism and the ShuffleNetV2 lightweight backbone network, obtaining the current frame features of each frame of the ultrasonic phased array full-focus detection video; based on the background features and the current frame features, obtaining the spatio-temporal features of the video sequence frames through the dynamic feature fusion module, and outputting the near-surface defects of the large ring parts according to the spatio-temporal features of the video sequence frames. Please refer to Figure 7 , which shows a schematic structural diagram of a moving target detection neural network model provided by an embodiment of the present application.

[0067] Specifically, the moving target detection neural network algorithm model proposed in the present application is divided into three parts: the background modeling module (ABF-STBM), the current frame feature extraction module (SFAM-ShuffeNetV2), and the dynamic feature fusion module (DFFM). Among them, the background modeling module is composed of the adaptive bilateral filtering algorithm (ABF) and the background modeling algorithm (STBM); the current frame feature extraction module is composed of the sequential frame attention mechanism (SFEM) and the ShuffleNetV2 network; the dynamic feature fusion module (DFFM) is to fuse the background features and the current frame features to enhance the accurate recognition ability of the near-surface defects occluded under the high-noise background.

[0068] For the background modeling module, the adaptive bilateral filtering (ABF) is used to eliminate the noise in the ultrasonic image. Please refer to Figure 8 , which shows a schematic structural diagram of a background modeling module provided by an embodiment of the present application. ABF is a non-linear filtering method that uses spatial proximity and pixel value similarity to process images. Pixel values at different distances are given Gaussian weights by the Gaussian variance, and the image edge details are retained, which can suppress the noise points in the ultrasonic image and obtain an ultrasonic image with a higher signal-to-noise ratio. Assume that A: I → R is the original image and B: I → R is the output result. The ABF algorithm is defined as:

[0069]

[0070] Among them, B(i) represents the output image gray value at the pixel position i after filtering, and k i : R → R is the kernel of the local Gaussian range, and Ω is a window centered at the origin. W (i) is a normalization parameter. k i(t) is the intensity similarity kernel, which is used to measure the difference in gray values between the current neighborhood pixels and the central pixel. θ(i) is the central information. σ(i) is the spatial variation function of the kernel g: g: Ω → R. g(j) represents the weight in the spatial domain, and ρ represents the standard deviation in g(j). A(i - j) represents the gray value of the original image corresponding to the pixel offset by j in the window centered at i. The spatio-temporal background modeling (STBM) method used in the background modeling module is a network model that stacks multiple depthwise separable convolutions (DSConv). DSConv can effectively reduce the computational complexity of the model. By stacking multiple DSConv models, the background information of historical frames is fully extracted, and 3×3Conv is used for the adaptive selection of multi-scale features. The depth of the input image is calculated channel by channel to obtain the output features of the same channel, and then 1×1×n convolution is used for pointwise calculation to weight and combine the features of each channel dimension to obtain the output features. The calculation of the output features of each depthwise separable convolution is as follows:

[0071]

[0072] Among them, F is the input feature, Γ τ is the output feature, τ is the serial number of DSConv, τ ∈ [1, 3]. φ represents the BN layer and the activation function. D n,2τ-1,2τ-1 is DSConv, is the convolution operation. The input feature is convolved using DSConv of different sizes. Channel concatenation and 1×1 convolution are used for feature fusion, and the output background feature is obtained by 3×3 convolution. Feature fusion is defined as:

[0073]

[0074]

[0075] Figure 9 This application also proposes a lightweight network architecture ShuffeNetV2, which consists of two networks with strides of 1 and 2 respectively. The main branches of both structures consist of three convolutional layers: 1×1 convolution, 3×3 depthwise convolution (DWConv), and 1×1Conv. Please refer to Figure 9, which shows a schematic diagram of the SFAM-ShuffleNetV2 network structure provided by this application. In the backbone network, the input features are realized through channel splitting, and the side branches perform channel shuffling. In the downsampling module, the features are directly processed by two branches, and the side branch has 3×3 DWConv and 1×1Conv. Finally, channel shuffling is used to achieve feature fusion of the two branches. By adding the current frame feature extraction module to the last block of each stage in ShuffleNetV2, the input size of the feature map remains unchanged. Please refer to Figure 10 , which shows a schematic diagram of the structure of a current frame feature extraction module provided by an embodiment of this application. In the figure, SFAM can simultaneously focus on the features at different frame timings, extract the motion features of near-surface defects, and enhance the anti-interference ability of ShuffleNetV2. The defect features of the sequence frames are obtained by the ShuffleNetV2 backbone network, and each frame feature can be recalibrated by adding weights. Assume that the input sequential frame features are {Z m}, m = 1, 2…M, where m is the number of different frames. Set the weight calibration formula for different frames as follows:

[0076]

[0077] where, is the convolution operation, * is the multiplication operation, and Z′ m is the calibrated feature map. The calibration weight consists of the initial weight V a and the adaptive weight δ m for each frame, and δ m is obtained from the continuous frame time context information. By performing global average pooling on each frame feature map in the spatial dimension, the video frame adaptive weight is obtained as:

[0078] δ m = GAP f (Z m )

[0079] where, δ m is the video frame adaptive weight, and GAP f (.) represents performing global average pooling processing on the input sequential frame features in the spatial dimension. The video frame adaptive weight not only considers the information of the current frame but also integrates the temporal attention information between related frame sequences. Then, through global average pooling in the temporal dimension, the global description factor representing each frame is obtained:

[0080] G0 = GAP s (δ m )

[0081] where, G0 is the global description factor for each frame, and GAP s(.) represents global average pooling processing of the frequency-frame adaptive weight in the time dimension. By performing two convolutions and normalization operations on the feature map, additional time-adaptive calibration weights are obtained. Therefore, the whole process of the adaptive convolution module is as follows:

[0082] F A (Z m ) = Conv(θ(Conv(γ m +FC(G0))))

[0083] Among them, F A is the mapping function of Z m and the frequency-frame adaptive weight δ m . Conv is a 1×1 convolution, γ m is the frame description factor, θ represents the ReLU and normalization operations, and FC represents global pooling. Through adaptive convolution, the current frame features of each frame image in the ultrasonic phased array full-focus detection video are finally obtained.

[0084] For the dynamic feature fusion module, it is used to finally integrate the background feature (BF) and the current frame feature (CFF). Please refer to Figure 11 , which shows a schematic diagram of the structure of a dynamic feature fusion module provided by an embodiment of the present application. In the figure, it consists of two branches. One branch is used to obtain the global channel information W ch from BF and CFF. W ch is used to calibrate CFF, and the calibrated feature is obtained by 1×1 convolution. In addition, the global spatial information W sp is extracted from BF and CFF, and W spThe features are recalibrated to generate fused features. Finally, the fused features are input into the subsequent recognition layer of the neural network. By combining the spatial structure changes and temporal motion trajectories between frames, high-dimensional spatio-temporal features in the video sequence frames are extracted. Then, based on the spatio-temporal features of the video sequence frames through the moving object detection neural network model, the coordinates of near-surface defects recognized in each frame image of the ultrasonic phased array full-focus detection video are recorded; real-time trajectory tracking of the near-surface defects is performed based on the near-surface defect coordinates, and the near-surface defects are output: Through the inference and recognition of each frame image in the detection video by the moving object detection neural network model, the spatial position coordinates of all near-surface defect targets in each frame are extracted and recorded, and a coordinate time series sequence is established with the frame number as the time identifier to form a preliminary motion trajectory of the defect object in the video. Further, by calculating the feature similarity and spatial position continuity between the same defect candidate regions in adjacent frames, a frame-to-frame correlation matrix is constructed using the Gaussian kernel function to evaluate the temporal consistency of defect targets between different frames. If the recognition of a certain defect in the current frame is interrupted in subsequent frames, based on the average speed and motion trend of the defect object in the previous several frames, the trajectory prediction function is used to complete the defect coordinates in the missing frames, so as to realize the complete trajectory restoration of the near-surface defects. The finally output video-level recognition results include: the position, existence duration, and motion trajectory of each defect object in each frame, ensuring stable cross-frame tracking and real-time dynamic positioning of shallow defects close to the workpiece surface, and greatly improving the recognition accuracy and defect detection integrity. Please refer to Figure 12 , which shows a schematic diagram of the intelligent recognition result of near-surface defects in ultrasonic images of large ring parts provided by its own embodiment. Please refer to Table 1, which shows the test results of the moving object detection model for near-surface defects in large ring parts.

[0085] [Table 1]

[0086]

[0087] In a possible implementation manner, step S106 further includes: obtaining the near-surface defect features of the sequence frames in the ultrasonic phased array full-focus detection video according to the moving object detection neural network model; calculating the correlation features of adjacent frames according to the near-surface defect features of the sequence frames; predicting the positions of the near-surface defects in the missing frames according to the correlation features.

[0088] Specifically, the proposed moving object detection neural network model is used to recognize the near-surface defects in the ultrasonic full-focus detection video sequence frame images to obtain the near-surface defect features of the sequence frames; by calculating the correlation of adjacent frames in the video, the features of the next frame can be predicted to realize near-surface defect tracking and prediction, where the calculation formula of the correlation function of adjacent frames is:

[0089]

[0090] Among them, G mov (X k-1 , X k ) represents the correlation function value between the (k - 1)-th frame and the k-th frame, X k-1 represents the feature of the previous frame, X ′ represents the feature of the adjacent frame, and v is the average speed of the target in k consecutive frames. The Gaussian kernel function is used to calculate the correlation similarity matrix of adjacent frames:

[0091]

[0092] Among them, G(X k-1 , X k ) represents the correlation similarity matrix between the (k - 1)-th frame and the k-th frame, K represents the Gaussian kernel function, and ξ is the weight coefficient. Through the recognition of near-surface defects and the tracking of adjacent frames in the video, the positions of near-surface defects F0(x, y), F1(x, y), F2(x, y)... in each frame image of the ultrasonic full-focus detection video are obtained, and the tracking of the movement trajectory of near-surface defects is realized. Please refer to Figure 13 , which shows a schematic diagram of the defect tracking result of the near-surface ultrasonic detection video of a large ring provided by the embodiment itself. Please refer to Table 2, which shows the intelligent recognition results of the near-surface defects of a certain large ring by ultrasonic phased array.

[0093] [Table 2]

[0094]

[0095] Please refer to Figure 14 , which shows a schematic diagram of the modules of an intelligent near-surface defect recognition device for large ring parts based on ultrasonic phased array provided by the embodiment of the present application. The device includes an acquisition module 141, a full-focus detection video generation module 142, and a processing module 143. Among them,

[0096] The acquisition module 141 is used to scan along the surface of the large ring part through an ultrasonic phased array probe to obtain the ring material and surface structure characteristics corresponding to the large ring part.

[0097] The full-focus detection video generation module 142 is used to construct the ultrasonic phased array detection parameters and the ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring part according to the ring material and surface structure characteristics; control the ultrasonic phased array probe to collect ultrasonic detection signals in real time and upload them to the host computer according to the ultrasonic phased array detection parameters; process the ultrasonic detection signals in real time through the host computer, and convert the ultrasonic detection signals into ultrasonic full-focus images based on the ultrasonic phased array full-focus imaging algorithm; generate an ultrasonic phased array full-focus detection video according to the ultrasonic full-focus images.

[0098] The processing module 143 is configured to input the ultrasonic phased array full-focus detection video into the moving target detection neural network model, and output the near-surface defects of the large ring part based on the spatio-temporal features of the video sequence frames through the moving target detection neural network model.

[0099] In a possible implementation manner, the full-focus detection video generation module 142 is configured to control the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and upload them to the host computer, specifically including: constructing the ultrasonic phased array detection parameters of the surface area of the large ring part according to the ring part material and surface structure features, where the ultrasonic phased array detection parameters include the upper surface phased array probe frequency parameter, the lower surface phased array probe frequency parameter, the probe element number parameter, the element spacing parameter, and the element width parameter; controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and upload them to the host computer.

[0100] In a possible implementation manner, the full-focus detection video generation module 142 is configured to construct the ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring part according to the ring part material and surface structure features, and the ultrasonic phased array full-focus imaging algorithm is used to calculate the full-focus imaging amplitude, specifically including: constructing the ultrasonic phased array full-focus imaging area according to the probe element number parameter, the element spacing parameter, and the element width parameter; dividing the ultrasonic phased array full-focus imaging area into multiple grid points in a preset manner; obtaining the target coordinate point among the multiple grid points, where the target coordinate is any coordinate point among the multiple grid points; transmitting a preset number of ultrasonic signals through each element of the ultrasonic phased array probe, and receiving the ultrasonic signals by each element simultaneously, where the preset number is determined by the probe element number parameter; calculating the full-focus imaging amplitude corresponding to the target coordinate point according to the multiple ultrasonic signals by the amplitude superposition method.

[0101] In a possible implementation manner, before the full-focus detection video generation module 142 inputs the ultrasonic phased array full-focus detection video into the moving target detection neural network model, it is necessary to construct the moving target detection neural network model, specifically including: constructing a comparison test block according to the ring part material and surface structure features, where the comparison test block is a large ring part with different materials and structures, and the comparison test block includes comparison near-surface defects with different sizes and different depths; scanning the comparison test block along the ultrasonic phased array probe to collect the comparison ultrasonic phased array full-focus detection video containing the comparison near-surface defects in real time; annotating the comparison near-surface defects in the comparison ultrasonic phased array full-focus detection video to establish a near-surface defect sample video data set, and the near-surface defect sample video data set is expanded by the method of combining VideoMix video data enhancement and image data enhancement; performing model training according to the near-surface defect sample video data set to construct the moving target detection neural network model.

[0102] In a possible implementation manner, the moving target detection neural network model includes a background modeling module, a current frame feature extraction module, and a dynamic feature fusion module. The processing module 143 is configured to output near-surface defects of the large ring part based on the spatio-temporal features of the video sequence frames through the moving target detection neural network model, specifically including: through the background modeling module, performing image noise elimination and background modeling algorithm based on the adaptive bilateral filtering algorithm to obtain the background features of each frame image in the ultrasonic phased array full-focus detection video; through the current frame feature extraction module, obtaining the current frame features of each frame image in the ultrasonic phased array full-focus detection video based on the sequential frame attention mechanism and the ShuffleNetV2 lightweight backbone network; based on the background features and the current frame features, obtaining the spatio-temporal features of the video sequence frames through the dynamic feature fusion module, and outputting the near-surface defects of the large ring part according to the spatio-temporal features of the video sequence frames.

[0103] In a possible implementation manner, the processing module 143 is configured to output near-surface defects of the large ring part based on the spatio-temporal features of the video sequence frames through the moving target detection neural network model, specifically including: based on the spatio-temporal features of the video sequence frames through the moving target detection neural network model, recording the coordinates of the near-surface defects recognized in each frame image of the ultrasonic phased array full-focus detection video; performing real-time trajectory tracking on the near-surface defects through the near-surface defect coordinates, and outputting the near-surface defects.

[0104] In a possible implementation manner, after the processing module 143 inputs the ultrasonic phased array full-focus detection video into the moving target detection neural network model and outputs the near-surface defects of the large ring part based on the spatio-temporal features of the video sequence frames through the moving target detection neural network model, the method further includes: obtaining the near-surface defect features of the sequence frames in the ultrasonic phased array full-focus detection video according to the moving target detection neural network model; calculating the correlation features of adjacent frames according to the near-surface defect features of the sequence frames; predicting the positions of the near-surface defects in the missing frames according to the correlation features.

[0105] It should be noted that: when the device provided in the above embodiment realizes its functions, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0106] This application also provides an electronic device. Refer to Figure 15 , Figure 15It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. The electronic device may include: at least one processor 1501, at least one communication bus 1502, a user interface 1503, at least one network interface 1504, and a memory 1505.

[0107] Among them, the communication bus 1502 is used to realize the connection and communication between these components.

[0108] Among them, the user interface 1503 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 1503 may further include a standard wired interface and a wireless interface.

[0109] Among them, the network interface 1504 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0110] Among them, the processor 1501 may include one or more processing cores. The processor 1501 connects various parts within the entire server through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 1505, and by calling data stored in the memory 1505, it executes various functions of the server and processes data. Optionally, the processor 1501 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 1501 may integrate a combination of one or several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, and application programs, etc.; the GPU is responsible for the rendering and drawing of the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above modem may not be integrated into the processor 1501 and may be implemented separately by a single chip.

[0111] Among them, the memory 1505 may include a Random Access Memory (RAM), or may also include a Read-Only Memory. Optionally, the memory 1505 includes a non-transitory computer-readable storage medium. The memory 1505 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 1505 may include a program storage area and a data storage area. Among them, the program storage area can store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area can store the data involved in the above-mentioned various method embodiments. The memory 1505 may optionally also be at least one storage device located far from the aforementioned processor 1501. Refer to Figure 15 , in the memory 1505 as a computer storage medium, it may include an operating system, a network communication module, a user interface module, and an intelligent recognition application program for near-surface defects of large ring parts based on ultrasonic phased arrays.

[0112] In Figure 15 In the electronic device shown, the user interface 1503 is mainly used to provide an input interface for the user to obtain the data input by the user; and the processor 1501 can be used to call the intelligent recognition application program for near-surface defects of large ring parts based on ultrasonic phased arrays stored in the memory 1505. When executed by one or more processors 1501, the electronic device is caused to execute one or more of the methods as described in the above embodiments. It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0113] The present application also provides a computer-readable storage medium, and the computer-readable storage medium stores instructions. When executed by one or more processors, the electronic device is caused to execute one or more of the methods as described in the above embodiments.

[0114] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0115] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling or communication connection between each other can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in electrical or other forms.

[0116] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0117] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0118] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present application. The aforementioned memory includes: various media such as USB flash drives, mobile hard disks, magnetic disks or optical discs that can store program codes.

[0119] The above are only exemplary embodiments disclosed in the present application and cannot be used to limit the scope of the disclosure of the present application. That is, all equivalent changes and modifications made in accordance with the teachings disclosed in the present application still fall within the scope covered by the disclosure of the present application.

[0120] The present application aims to cover any variations, uses or adaptive changes of the disclosure of the present application. These variations, uses or adaptive changes follow the general principles of the disclosure of the present application and include common general knowledge or conventional technical means in the technical field not recorded in the disclosure of the present application.

Claims

1. An intelligent recognition method for near-surface defects of large ring parts based on ultrasonic phased array, characterized in that, The method includes: Scanning along the surface of the large ring piece by an ultrasonic phased array probe to obtain the ring piece material and surface structure characteristics corresponding to the large ring piece; Constructing ultrasonic phased array detection parameters and an ultrasonic phased array full focus imaging algorithm for the surface area of the large ring piece according to the ring piece material and the surface structure characteristics; Controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and uploading them to the host computer; Processing the ultrasonic detection signals in real time by the host computer and converting the ultrasonic detection signals into ultrasonic full focus images based on the ultrasonic phased array full focus imaging algorithm; Generating an ultrasonic phased array full focus detection video according to the ultrasonic full focus images; Inputting the ultrasonic phased array full focus detection video into a moving target detection neural network model, and outputting the near-surface defects of the large ring piece based on the spatio-temporal characteristics of the video sequence frames through the moving target detection neural network model.

2. The method according to claim 1, characterized in that, The step of controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and uploading them to the host computer specifically includes: Constructing the ultrasonic phased array detection parameters for the surface area of the large ring piece according to the ring piece material and the surface structure characteristics, where the ultrasonic phased array detection parameters include the upper surface phased array probe frequency parameter, the lower surface phased array probe frequency parameter, the probe element number parameter, the element spacing parameter, and the element width parameter; Controlling the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and uploading them to the host computer.

3. The method according to claim 2, characterized in that, Constructing an ultrasonic phased array full focus imaging algorithm for the surface area of the large ring piece according to the ring piece material and the surface structure characteristics, where the ultrasonic phased array full focus imaging algorithm is used to calculate the full focus imaging amplitude, and specifically includes: Constructing an ultrasonic phased array full focus imaging area according to the probe element number parameter, the element spacing parameter, and the element width parameter; Dividing the ultrasonic phased array full focus imaging area into multiple grid points in a preset manner; Obtaining target coordinate points among the multiple grid points, where the target coordinate is any coordinate point among the multiple grid points; Transmitting a preset number of ultrasonic signals through each element of the ultrasonic phased array probe, and simultaneously receiving the ultrasonic signals by each element, where the preset number is determined by the probe element number parameter; Calculating the full focus imaging amplitude corresponding to the target coordinate point through the amplitude superposition method according to the multiple ultrasonic signals.

4. The method according to claim 1, wherein Before inputting the ultrasonic phased array full focus detection video into the moving target detection neural network model, it is necessary to construct the moving target detection neural network model, which specifically includes: Constructing a comparison test block according to the ring piece material and the surface structure characteristics, where the comparison test block is a large ring piece with different materials and structures, and the comparison test block includes comparison near-surface defects with different sizes and depths; Scanning along the comparison test block by the ultrasonic phased array probe to collect a comparison ultrasonic phased array full focus detection video containing the comparison near-surface defects in real time; Annotate the near-surface defects in the contrast ultrasonic phased array full-focus detection video to establish a near-surface defect sample video dataset, and expand the dataset by combining VideoMix video data augmentation and image data augmentation methods for the near-surface defect sample video dataset; Perform model training based on the near-surface defect sample video dataset to construct the moving target detection neural network model.

5. The method according to claim 1, wherein The moving target detection neural network model includes a background modeling module, a current frame feature extraction module, and a dynamic feature fusion module. Through the moving target detection neural network model, the near-surface defects of the large ring parts are output based on the spatio-temporal features of the video sequence frames, specifically including: Through the background modeling module, based on the adaptive bilateral filtering algorithm for image noise elimination and background modeling algorithm, obtain the background features of each frame image in the ultrasonic phased array full-focus detection video; Through the current frame feature extraction module, based on the sequential frame attention mechanism and the ShuffleNetV2 lightweight backbone network, obtain the current frame features of each frame image in the ultrasonic phased array full-focus detection video; Based on the background features and the current frame features, obtain the spatio-temporal features of the video sequence frames through the dynamic feature fusion module, and output the near-surface defects of the large ring parts according to the spatio-temporal features of the video sequence frames.

6. The method according to claim 1, characterized in that, The output of the near-surface defects of the large ring parts through the moving target detection neural network model based on the spatio-temporal features of the video sequence frames specifically includes: Through the moving target detection neural network model based on the spatio-temporal features of the video sequence frames, record the coordinates of the near-surface defects identified in each frame image of the ultrasonic phased array full-focus detection video; Perform real-time trajectory tracking on the near-surface defects through the near-surface defect coordinates and output the near-surface defects.

7. The method according to claim 1, characterized in that, After inputting the ultrasonic phased array full-focus detection video into the moving target detection neural network model and outputting the near-surface defects of the large ring parts through the moving target detection neural network model based on the spatio-temporal features of the video sequence frames, the method further includes: Obtain the near-surface defect features of the sequence frames in the ultrasonic phased array full-focus detection video according to the moving target detection neural network model; Calculate the correlation features of adjacent frames according to the near-surface defect features of the sequence frames; Predict the positions of the near-surface defects in the missing frames according to the correlation features.

8. An intelligent recognition device for near-surface defects of large ring parts based on ultrasonic phased array, characterized in that, The device includes an acquisition module, a full-focus detection video generation module, and a processing module, where The acquisition module is used to scan along the surface of the large ring part through an ultrasonic phased array probe to obtain the ring part material and surface structure features corresponding to the large ring part; The full-focus detection video generation module is configured to construct ultrasonic phased array detection parameters and an ultrasonic phased array full-focus imaging algorithm for the surface area of the large ring part according to the ring part material and the surface structure features; control the ultrasonic phased array probe to collect ultrasonic detection signals in real time according to the ultrasonic phased array detection parameters and upload them to the host computer; process the ultrasonic detection signals in real time through the host computer, and convert the ultrasonic detection signals into ultrasonic full-focus images based on the ultrasonic phased array full-focus imaging algorithm; generate an ultrasonic phased array full-focus detection video according to the ultrasonic full-focus images. The processing module is configured to input the ultrasonic phased array full-focus detection video into a moving target detection neural network model, and output the near-surface defects of the large ring part based on the spatio-temporal features of the video sequence frames through the moving target detection neural network model.

9. An electronic device, characterized in that, It includes a processor, a communication bus, a user interface, a network interface, and a memory. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions which, when executed, execute the method according to any one of claims 1 to 7.

Images