Ultrasonic detection method and device for steel structure defect detection
The ultrasonic testing device, which combines a robotic arm and a host computer, enables efficient automatic scanning and accurate identification of defects in steel structures. This solves the problems of low efficiency and insufficient automation in existing technologies, improves detection accuracy and versatility, and reduces costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-03-27
AI Technical Summary
Existing ultrasonic testing methods are inefficient, lack automation, have poor versatility, and have limited accuracy in defect identification and quantification, failing to meet the needs of large-scale rapid testing.
An ultrasonic testing device combining a robotic arm and a host computer is used. It uses an ultrasonic probe on the gripper for adaptive bonding detection, combines the TFM algorithm and conditional diffusion model for image reconstruction and defect identification, and achieves automated quantitative positioning through the robotic arm and marking frame.
It improves detection efficiency and accuracy, reduces human intervention, enhances the versatility and automation of detection, can adapt to various steel structures, reduces detection costs, and provides reliable quantitative defect data support.
Smart Images
Figure CN121741010A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nondestructive testing technology, specifically to an ultrasonic testing method and apparatus for detecting defects in steel structures. Background Technology
[0002] Steel structures are widely used in various fields such as construction, bridges, machinery manufacturing, and aerospace due to their high strength, high stability, and good construction adaptability. Their internal quality is directly related to the safety and service life of products or projects. Therefore, accurate detection of defects such as cracks, inclusions, and porosity inside steel structures is of utmost importance.
[0003] Ultrasonic testing has become the mainstream method for detecting defects in steel structures due to its advantages such as large detection depth, high sensitivity, and non-destructive and environmentally friendly nature. However, existing methods have significant shortcomings: First, current testing methods rely on manual or semi-automated operations, requiring manual adjustment of parameters and movement of probes, which makes it difficult to meet the needs of large-scale rapid testing, resulting in low testing efficiency and insufficient automation. Second, they have poor versatility and high costs, requiring dedicated testing production lines for steel structures of different shapes and specifications, leading to high costs for equipment purchase, site layout, and switching and maintenance. Third, the accuracy of defect identification and quantification is limited, relying on simple signal threshold judgments, making it difficult to accurately distinguish defect types, and resulting in insufficient quantitative and location accuracy, failing to provide reliable data support.
[0004] Therefore, there is an urgent need for an ultrasonic testing method and device that can efficiently and automatically scan different steel structures, accurately identify defects, and complete quantitative positioning, so as to improve testing accuracy and efficiency. Summary of the Invention
[0005] To address the problems in the background art, this invention proposes an ultrasonic testing device for detecting defects in steel structures, comprising a robotic arm and a host computer. The robotic arm is slidably mounted on one side of a testing platform, on which the steel structure to be tested is placed. The load end of the robotic arm is equipped with a gripper. Multiple sliding rods are provided on the inner side of the gripper at the end away from the robotic arm. One end of each sliding rod is elastically and slidably connected to the gripper via a spring. A marking frame is slidably connected to the middle part of the gripper. The two ends of the marking frame extend to the outer and inner sides of the gripper, respectively. An ultrasonic probe and an imprint pad are respectively provided inside and outside the inner end of the marking frame located on the gripper. A motor for driving the sliding of the marking frame is fixed on the outer side of the gripper. The drive end of the motor is connected to the outer end of the marking frame located on the gripper. The host computer is electrically connected to both the motor and the ultrasonic probe. Preferably, the testing platform is provided with a slide rail on one side, and the end of the robotic arm away from its load is slidably connected to the slide rail.
[0006] Preferably, one end of the spring is connected to the gripper, and the other end of the spring is connected to the slide rod.
[0007] Preferably, the ultrasonic probe is connected to the host computer via an ultrasonic detection module.
[0008] An ultrasonic testing method for detecting defects in steel structures includes the following steps: S1. Place the steel structure to be inspected on the inspection table, move the robotic arm above the steel structure to be inspected, and use the grippers to hold the steel structure. S2. When the steel structure is clamped, the end of the slide rod away from the gripper contacts the surface of the steel structure. The slide rod is compressed, and the spring is in a compressed state, so as to achieve adaptive contact with the surface of the steel structure. S3. The ultrasonic probe corresponds to the detection point on the steel structure. The ultrasonic probe emits ultrasonic waves into the steel structure and receives the reflected ultrasonic signals. The ultrasonic signals are transmitted to the host computer, which executes the TFM algorithm to generate an ultrasonic image of the steel structure cross section. The ultrasonic image is used to detect whether there are defects. Simulation data is obtained using the ultrasonic image to train the conditional diffusion model. The trained model is used as the judgment standard for subsequent detection to continuously improve the detection accuracy and adaptability. When a defect is detected, the host computer controls the motor to start, and the motor drives the marking frame to move, bringing the printing pad close to the steel structure and marking the defect location on the steel structure surface. S4. The host computer has preset curve cluster data. By comparing the ultrasound image with the curve cluster data, the defect size is evaluated, and the defect is identified and quantified. S5. After the test is completed, the gripper is released and the steel structure is placed on the test table. The robotic arm rotates the gripper to change the gripping position of the steel structure so that the gripper grips the steel structure again and the ultrasonic probe corresponds to another test point of the steel structure. S6. Repeat steps S2-S5 to achieve a comprehensive ultrasonic scan of the steel structure and complete the defect detection of the steel structure.
[0009] Preferably, step S3 specifically comprises: S31. All signals of each transmitting and receiving crystal combination are detected and recorded by phased-array ultrasonic imaging to form a complete raw data matrix; S32. The original data matrix records the complete amplitude scan signals of all transmit-receive chip pairs, forming a three-dimensional data matrix. TFM is used to perform time-delay superposition processing on all amplitude scan signals in the original data matrix to reconstruct the amplitude of each pixel in the image. S321. First, define the imaging area by dividing a two-dimensional pixel grid inside the steel structure to be inspected. S322, Traverse each pixel in the imaging region; S323. For each transmit-receive chip pair in the original data matrix, calculate the theoretical propagation time required for the sound wave to travel from the transmitting chip to the pixel and then back from the pixel to the receiving chip. The formula for calculating the theoretical propagation time is:
[0010] In the formula, P represents a pixel, i represents the transmitting chip, j represents the receiving chip, and t represents the receiving chip. i,j (P) represents the theoretical propagation time, x represents the horizontal coordinate of the pixel, z represents the vertical coordinate of the pixel, and c represents the longitudinal wave velocity of the ultrasonic wave propagating in the material. S324. Extract the amplitude at the theoretical propagation time from the corresponding amplitude scan signal; S325. Coherently superimpose the amplitude values corresponding to all N×N amplitude scanning signals at the pixel point to obtain the reconstructed amplitude value of the pixel:
[0011] In the formula, I(P) is the reconstructed amplitude of the pixel, and A i,j (t i,j (P) represents the amplitude at the theoretical propagation time; S33. After traversing all pixels, I(P) is mapped to grayscale or color to form the final TFM image.
[0012] Preferably, the curve cluster data is a distance-amplitude-equivalent curve cluster, and the method for obtaining the curve cluster data is as follows: On a standard steel structure test block, the echo amplitude of artificial defects of different depths and known sizes was measured, and the distance-amplitude relationship curve was plotted, where the distance is the sound path from the defect to the ultrasonic probe. By using experiments or acoustic simulation software, the curvature of the distance-amplitude relationship curve is adjusted to obtain a distance-amplitude relationship that better reflects the actual situation.
[0013] Preferably, step S3 specifically comprises: A1. Generating synthetic ultrasound images with specific defects using a conditional diffusion model. A11. Diffusion process: During the training phase, the conditional diffusion model learns to gradually add noise to a real defect image until it becomes pure Gaussian noise, resulting in a gradually blurred image sequence. A12. Inverse diffusion process: The conditional diffusion model learns how to remove noise from the image and gradually reconstructs the original defective image from the blurred image. Defect condition control is added during the learning process. By injecting defect conditions as additional input information into the denoising process, the model is guided by defect conditions at each step of image generation. A13. Generate a synthetic ultrasound image that meets the specified conditions; A2. The FID, KID, and LPIPS metrics were used to evaluate the quality of ultrasound images, and a self-supervised auxiliary network was used to extract defect features from the ultrasound images. The extracted defect feature data was used for model training. A3. After performing ultrasound image quality assessment, if the ultrasound image quality is invalid, the ultrasound image is reprocessed through steps A1-A2; if the ultrasound image quality is valid, the valid ultrasound image is used for model training.
[0014] The beneficial effects of this invention are as follows: (1) The present invention uses a testing table to place the steel structure to be tested, uses a robotic arm to grab the steel structure, and uses an ultrasonic probe on the gripper of the robotic arm to test the steel structure, thereby reducing the degree of manual intervention and improving the testing efficiency and automation. (2) The gripper of the present invention is connected to multiple sliding rods by springs. When the gripper clamps the steel structure, the sliding rods first contact the surface of the steel structure. The steel structure pushes the sliding rods to compress the springs through the reaction force. The sliding rods realize adaptive clamping of steel structures with different surface shapes, avoiding detection omissions caused by uneven steel structure surfaces or positioning deviations of the robotic arm. At the same time, it improves the versatility of the detection, adapts to the detection of various steel structures, and eliminates the need to set up special detection devices for different steel structures, thus reducing detection costs. (3) This invention generates ultrasonic images of steel structure cross-sections by executing the TFM algorithm on a host computer. Defects are detected based on the ultrasonic images, improving detection efficiency and accuracy. Curve cluster data is used to quantitatively analyze the imaging results and obtain the size of defects, which is beneficial for defect identification. At the same time, this invention extracts data from the acquired ultrasonic images for training of the conditional diffusion model, continuously acquiring and learning various defect features to provide richer and more accurate judgment criteria for subsequent detection, and continuously improving the adaptability and accuracy of subsequent defect detection. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of the structure of an ultrasonic testing device for detecting defects in steel structures proposed in this invention. Figure 2 This is a schematic diagram of the robotic arm of an ultrasonic testing device for detecting defects in steel structures proposed in this invention. Figure 3 This is a schematic diagram showing the usage status of an ultrasonic testing device for detecting defects in steel structures proposed in this invention. Figure 4 This is a schematic diagram of the gripper structure proposed in this invention; Figure 5 This is a flowchart of an ultrasonic testing method for detecting defects in steel structures proposed in this invention.
[0016] The following are labeled in the diagram: 1. Robotic arm; 2. Gripper; 3. Slide bar; 4. Motor; 5. Marking frame; 6. Ultrasonic probe; 7. Imprint pad; 8. Slide rail; 9. Inspection table. Detailed Implementation
[0017] To make the present invention clearer and more understandable, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the given embodiments are only one of the implementation methods and do not represent all embodiments.
[0018] In this article, terms such as "inner," "outer," "upper," and "lower" are established based on the positional relationships shown in the attached drawings. Depending on the attached drawings, the corresponding positional relationships may also change. Therefore, they should not be interpreted as an absolute limitation on the scope of protection.
[0019] Combined with appendix Figure 1 - Appendix Figure 4 An ultrasonic testing device for detecting defects in steel structures includes a robotic arm 1 and a host computer. The robotic arm 1 is slidably mounted on one side of a testing platform 9, on which the steel structure to be tested is placed. The robotic arm 1 has a gripper 2 at its load end. Multiple sliding rods 3 are provided on the inner side of the gripper 2 away from the robotic arm 1. One end of each sliding rod 3 is elastically slidably connected to the gripper 2 via a spring. A marking frame 5 is slidably connected to the middle part of the gripper 2. The two ends of the marking frame 5 extend to the outer and inner sides of the gripper 2, respectively. An ultrasonic probe 6 and an imprint pad 7 are respectively provided inside and outside the inner end of the marking frame 5 located on the gripper 2. A motor 4 for driving the marking frame 5 to slide is fixed to the outer side of the gripper 2. The driving end of the motor 4 is connected to the outer end of the marking frame 5 located on the gripper 2. The host computer is electrically connected to the motor 4 and the ultrasonic probe 6.
[0020] Specifically, the motor 4 drives the marking frame 5 to move in the following manner: the marking frame 5 is slidably connected to the gripper 2, and the marking frame 5 and the gripper 2 do not rotate relative to each other (for example, the gripper 2 is provided with a groove along the sliding direction of the marking frame 5, and a slider is fixed on the marking frame 5, the slider being slidably disposed in the groove). The outer wall of the marking frame 5 is provided with an external thread, and a rotating ring is threadedly connected to the outer wall of the marking frame 5. The rotating ring is rotatably connected to the gripper 2. The driving end of the motor 4 is connected to the rotating ring through a gear or a synchronous belt.
[0021] The host computer is a monitoring host or a cloud processing platform.
[0022] Specifically, a slide rail 8 is provided on one side of the testing platform 9, and the end of the robotic arm 1 away from its load is slidably connected to the slide rail 8. The robotic arm 1 can move along the slide rail 8. More specifically, the testing platform 9 is a conveying platform, and the steel structure placed on the platform of the testing platform 9 can move from the beginning of the conveying process to the end of the conveying process.
[0023] Specifically, one end of the spring is connected to the gripper 2, and the other end of the spring is connected to the slide bar 3.
[0024] Specifically, the ultrasonic probe 6 is connected to the host computer via an ultrasonic testing module. More specifically, the ultrasonic testing module is a multi-channel ultrasonic testing module, which is an independent ultrasonic module. The host computer and the multi-channel ultrasonic testing module are designed separately to ensure better stability and higher performance. Furthermore, the host computer and the multi-channel ultrasonic testing module are connected only via a network cable, ensuring that the multi-channel ultrasonic testing module is not affected by electronic interference or software systems of the host computer.
[0025] Combined with appendix Figure 5 An ultrasonic testing method for detecting defects in steel structures includes the following steps: S1. Place the steel structure to be inspected on the inspection table 9, move the robotic arm 1 above the steel structure to be inspected, and use the gripper 2 to hold the steel structure. S2. When the steel structure is clamped, the end of the slide rod 3 away from the gripper 2 contacts the surface of the steel structure. The slide rod 3 is compressed, and the spring is in a compressed state, so as to achieve adaptive fitting with the surface of the steel structure. S3. The ultrasonic probe 6 corresponds to the detection point on the steel structure. The ultrasonic probe 6 emits ultrasonic waves into the steel structure and receives the reflected ultrasonic signals. The ultrasonic signals are transmitted to the host computer (the ultrasonic signals are processed by the ultrasonic detection module and converted into digital signals for transmission to the host computer). The host computer executes the TFM (Total Focusing Method) algorithm to generate ultrasonic images of the steel structure cross-section. It detects whether there are defects based on the ultrasonic images and uses the ultrasonic images to obtain simulation data for training the conditional diffusion model. The trained model is used as the judgment standard for subsequent detection to continuously improve the detection accuracy and adaptability. When a defect is detected, the host computer controls the motor 4 to start, and the motor 4 drives the marking frame 5 to move, so that the printing pad 7 is close to the steel structure, and marking is performed on the defect location on the steel structure surface to achieve defect location. Step S3 specifically involves: S31. All signals of each transmitting and receiving wafer combination are recorded by phased-array ultrasonic imaging (PAUT) to form a complete raw data matrix (FMC). S32. The original data matrix records the complete amplitude (A) scan signals of all transmit-receive chip pairs, forming a three-dimensional data matrix. The amplitude of each pixel in the image is reconstructed by performing time-delay superposition processing on all amplitude scan signals in the original data matrix using TFM. S321. First, define the imaging area by dividing a two-dimensional pixel grid inside the steel structure to be inspected. S322. Iterate through each pixel P(x, z) in the imaging region. S323. For each transmit-receive chip pair (i, j) in the original data matrix, calculate the theoretical propagation time required for the sound wave to travel from the transmitting chip i to the pixel and then reflect back from the pixel to the receiving chip j. The formula for calculating the theoretical propagation time is:
[0026] In the formula, P represents a pixel, i represents the transmitting chip, j represents the receiving chip, and t represents the receiving chip. i,j (P) represents the theoretical propagation time, x represents the horizontal coordinate of the pixel, z represents the vertical coordinate of the pixel, and c represents the longitudinal wave velocity of the ultrasonic wave propagating in the material. S324. Extract the amplitude at the theoretical propagation time from the corresponding amplitude scan signal; S325. Coherently superimpose the amplitude values corresponding to all N×N amplitude scanning signals at the pixel point to obtain the reconstructed amplitude value of the pixel:
[0027] In the formula, I(P) is the reconstructed amplitude of the pixel, and A i,j (t i,j (P) represents the amplitude at the theoretical propagation time; S33. After traversing all pixels, I(P) is mapped to grayscale or color to form the final TFM image.
[0028] The transmitting crystal and receiving crystal are components inside the ultrasonic probe 6, which is conventional technology and will not be described in detail here.
[0029] The specific steps of the model training process in step S3 are as follows: A1. Generating synthetic ultrasound images with specific defects using a conditional diffusion model. A11. Diffusion process: During the training phase, the conditional diffusion model learns to gradually add noise to a real defect image until it becomes pure Gaussian noise, resulting in a gradually blurred image sequence; the real defect image is the image actually obtained during the detection process that has not been processed by noise. A12. Inverse diffusion process: The conditional diffusion model learns how to remove noise from the image and gradually reconstructs the original defective image from the blurred image. Defect condition control is added during the learning process. By injecting defect conditions as additional input information into the denoising process, the model is guided by defect conditions at each step of image generation. A13. Generate a synthetic ultrasound image that meets the specified conditions; Synthetic ultrasonic images generated using a conditional diffusion model can help the device identify steel structure defects more quickly, resulting in more accurate results and enabling data feedback. It can also improve the detection rate: weak defect signals or rare defects may go undetected or lead to errors, but by pre-synthesizing defect images, the detection rate of such defects can be improved.
[0030] In addition, it allows for comparison of the quality of materials produced by different production line processes, thus enabling faster selection of the optimal process and reducing testing costs.
[0031] A2. The FID (Freche inception distance), KID (kernel inception distance), and LPIPS (learned perceptual patch similarity) metrics are used to evaluate the quality of ultrasound images, and a self-supervised auxiliary network is used to extract defect features from ultrasound images. The extracted defect feature data is used for model training. Among them, FID: calculates the distribution distance between the real image set and the generated image set in the feature space. The lower the value, the closer the distribution of the two is, and the better the overall realism and diversity of the generated images. KID: is less sensitive to the number of samples and has a more stable evaluation. LPIPS: uses a pre-trained neural network to extract features and calculates the perceptual distance between image pairs, which can better measure the similarity of images in visual perception and is consistent with human visual judgment.
[0032] Specifically, by using a self-supervised auxiliary network and an attention mechanism, defect condition information is forcibly incorporated into the image generation process.
[0033] Conditional diffusion model: used to generate synthetic TFM samples consistent with different defect locations and patterns. Proposed denoising network used to predict the noise removal requirements. θ (learning to predict noise) at each time step of the inverse process is primarily constructed by a deep convolutional neural network (CNN) with residual connections. The main function of the denoising network can be represented by a series of two-dimensional convolutions:
[0034] in It is the j-th feature map in the l-th hidden layer. It is the i-th feature map of the previous layer. as well as The function represents the weights and biases of the l-th hidden layer. This indicates a Rectifier Lug Unit (ReLU). These represent the regular convolution on the downsampling path and the transposed convolution on the upsampling path, respectively. In addition, each hidden layer is appended with a temporal embedding layer, enabling the denoising network to learn predictions at specific time steps.
[0035] Meanwhile, the introduced self-supervised auxiliary network can effectively guide defect information into the denoising process. The main motivation stems from the question of the extent to which the hidden layer conditioning features encode defect-related information to guide the TFM image generation process. To this end, a pre-trained fully dialogue neural network is proposed to reconstruct the defect-related information D itself, which includes the defect shape and coordinates, and these information are reproduced in both topological and spatial forms.
[0036] It is important to emphasize that the auxiliary network is able to extract meaningful hidden layer features without being limited to a finite number of TFM samples, because it can be self-supervised and trained on only hundreds of thousands of defect cases. After the pre-training process, the hierarchical feature maps retrieved from specific defect cases are conditionalized into a denoising network to guide the inverse process of the diffusion model.
[0037] The conditional strategy for feature maps retrieved from the auxiliary network is described as follows: Preferably, a hierarchical self-focus mechanism is adopted, using query... Notes , and value The reverse process adaptively reflects the extracted feature information:
[0038]
[0039]
[0040] in, , , Represents the learnable weight matrix , , In the first hidden layer; The feature map represents the time step t; This represents a feature map that integrates defect information D and time step t. (Attention-weighted feature map) Then, based on the scaling factor, the following calculation is performed: :
[0041] In this way, the denoising network is designed to achieve adaptive feature learning capability for each set of defect-related features during the reverse process.
[0042] By utilizing the conditional diffusion model, tens of thousands of synthetic TFM images (i.e., synthetic ultrasonic images) with precise pixel-level labels (such as specific weld types and common defects) can be generated according to actual inspection needs, making the inspection process and results faster and more accurate. Repeated inspection methods can make the generated model more and more realistic, the inspection system more and more intelligent, and the process design more and more optimized.
[0043] A3. After performing ultrasound image quality assessment, if the ultrasound image quality is invalid, the ultrasound image is reprocessed through steps A1-A2; if the ultrasound image quality is valid, the valid ultrasound image is used for model training.
[0044] S4. The host computer has preset curve cluster data. By comparing the ultrasonic image with the curve cluster data, the defect size can be evaluated, and the defect can be identified and quantified. Evaluating the defect size can ensure non-destructive testing of steel structure materials. The approximate defect size can be obtained through the curve cluster without cutting open the inside of the steel structure material. This can then be used to infer related problems on the steel structure material production line, such as excessively large welds, to help improve the production line.
[0045] S5. After the test is completed, the gripper 2 is released and the steel structure is placed on the test table 9. The robotic arm 1 rotates the gripper 2 to change the gripping position of the steel structure. After the gripper 2 grips the steel structure again, the ultrasonic probe 6 corresponds to another test point of the steel structure. S6. Repeat steps S2-S5 to achieve a comprehensive ultrasonic scan of the steel structure and complete the defect detection of the steel structure.
[0046] More specifically, the subsequent processing of defect-free workpieces: After the inspection is completed, the host computer sends a command to the robotic arm 1. The robotic arm 1 drives the gripper 2 to place the steel structure on the inspection table 9 and release the steel structure. At this time, the slide bar 3 is reset under the elastic force of the spring, and the marking frame 5 is reset under the action of the motor 4. The inspection table 9 continues to run, transporting the qualified steel structure to the next station, completing the inspection process of a defect-free workpiece. Detection and handling of defective workpieces: If the ultrasonic probe 6 detects a defect inside the steel structure during the detection process, the defect signal is processed by the ultrasonic detection module and transmitted to the host computer. The host computer calls the evaluation algorithm based on the distance-amplitude-equivalent curve cluster to perform quantitative analysis of the imaging results. At the same time, the host computer sends a control signal to the motor 4, which drives the marking frame 5 to approach the steel structure. The marking pad 7 at the top of the marking frame 5 contacts the surface of the defect area of the steel structure, thus completing the marking. After receiving the defect signal, the host computer immediately triggers the alarm system to alert the operator. The control system sends a command to the inspection station 9, which stops operating, the steel structure stops moving, and the inspection work of the inspection device is stopped. After receiving the alarm, the operator goes to the inspection station, handles the defective workpiece, and then starts the next normal inspection procedure.
[0047] The data from the experiment were used to train and optimize the conditional diffusion model, further improving the intelligence level of the entire detection device and making it more accurate in actual detection.
[0048] Specifically, the curve cluster data is a distance-amplitude-equivalent curve cluster, and the method for obtaining the curve cluster data is as follows: On a standard steel structure test block, the echo amplitude of artificial defects (such as flat-bottomed holes) of different depths and known sizes was measured, and the distance-amplitude relationship curve (DAC curve) was plotted, where the distance is the sound path from the defect to the ultrasonic probe 6; Through experiments (such as conducting testing experiments on specially designed standard curved steel structure test blocks and comparing the actual test data with the distance-amplitude relationship curve) or acoustic simulation software, the curvature of the distance-amplitude relationship curve is adjusted to obtain a distance-amplitude relationship that is more in line with the actual situation.
[0049] The distance-amplitude-equivalent curve model is used for amplitude correction and equivalent evaluation of defects at different depths (distances). For steel structural components with different curvatures and thicknesses, the parameters of the distance-amplitude-equivalent curve model are calibrated through experiments or simulations to establish a curve family. The curve family solves the problem that defects of the same size will display different amplitude heights on the screen due to different depths, achieving standardized quantification. Here, distance is the sound path from the defect to the ultrasonic probe 6, amplitude is the detected peak height of the defect echo (dB value), and equivalent is the reflection capability of the actual defect equivalently represented by the size of a regular reflector (such as a flat-bottomed hole, a horizontal hole, or a spherical hole).
[0050] It is important to understand that images are used to obtain the data information required for detection. For example, the initial ultrasound image contains the simulation data information required before training the model, while the synthetic ultrasound image contains the synthetic data used for model training.
[0051] Although embodiments of the invention have been shown and described, those skilled in the art will be able to make various changes, modifications, substitutions and alterations to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An ultrasonic testing device for detecting defects in steel structures, comprising a robotic arm (1) and a host computer, wherein the robotic arm (1) is slidably disposed on one side of a testing table (9), and the testing table (9) is used to place the steel structure to be tested, characterized in that: The robotic arm (1) has a gripper (2) at its load end. The gripper (2) has multiple sliding rods (3) on the inner side of the end away from the robotic arm (1). One end of the sliding rod (3) is elastically slidably connected to the gripper (2) by a spring. A marking frame (5) is slidably connected to the middle part of the gripper (2). The two ends of the marking frame (5) extend to the outer and inner sides of the gripper (2) respectively. An ultrasonic probe (6) and a printing pad (7) are respectively provided inside and outside the inner side of the marking frame (5) at the inner side of the gripper (2). A motor (4) for driving the marking frame (5) to slide is fixed on the outer side of the gripper (2). The driving end of the motor (4) is connected to the outer side of the marking frame (5) at the outer side of the gripper (2). The host computer is electrically connected to the motor (4) and the ultrasonic probe (6) respectively.
2. The ultrasonic testing device for detecting defects in steel structures according to claim 1, characterized in that: The testing platform (9) is provided with a slide rail (8) on one side, and the end of the robotic arm (1) away from its load is slidably connected to the slide rail (8).
3. The ultrasonic testing device for detecting defects in steel structures according to claim 1, characterized in that: One end of the spring is connected to the gripper (2), and the other end of the spring is connected to the slide bar (3).
4. The ultrasonic testing device for detecting defects in steel structures according to claim 1, characterized in that: The ultrasonic probe (6) is connected to the host computer through the ultrasonic detection module.
5. An ultrasonic testing method for detecting defects in steel structures, implemented based on the ultrasonic testing device for detecting defects in steel structures as described in any one of claims 1-4, characterized in that, Includes the following steps: S1. Place the steel structure to be tested on the testing table (9), move the robotic arm (1) above the steel structure to be tested, and use the gripper (2) to hold the steel structure. S2. When the steel structure is clamped, the end of the slide rod (3) away from the gripper (2) contacts the surface of the steel structure. The slide rod (3) is compressed and the spring is in a compressed state, so as to achieve adaptive fitting with the surface of the steel structure. S3. The ultrasonic probe (6) corresponds to the detection point on the steel structure. The ultrasonic probe (6) emits ultrasonic waves into the steel structure and receives the reflected ultrasonic signals. The ultrasonic signals are transmitted to the host computer. The host computer executes the TFM algorithm to generate an ultrasonic image of the steel structure cross section. It detects whether there are defects based on the ultrasonic image and uses the ultrasonic image to obtain simulation data for conditional diffusion model training. The trained model is used as the judgment standard for subsequent detection to continuously improve the detection accuracy and adaptability. When a defect is detected, the host computer controls the motor (4) to start, and the motor (4) drives the marking frame (5) to move, so that the printing pad (7) is close to the steel structure and marks the defect position on the steel structure surface; S4. The host computer has preset curve cluster data. By comparing the ultrasound image with the curve cluster data, the defect size is evaluated, and the defect is identified and quantified. S5. After the test is completed, the gripper (2) is released and the steel structure is placed on the test table (9). The robotic arm (1) rotates the gripper (2) to change the gripping position of the steel structure so that after the gripper (2) grips the steel structure again, the ultrasonic probe (6) corresponds to the position of another test point of the steel structure. S6. Repeat steps S2-S5 to achieve a comprehensive ultrasonic scan of the steel structure and complete the defect detection of the steel structure.
6. The ultrasonic testing method for detecting defects in steel structures according to claim 5, characterized in that, Step S3 specifically involves: S31. All signals of each transmitting and receiving crystal combination are detected and recorded by phased-array ultrasonic imaging to form a complete raw data matrix; S32. The original data matrix records the complete amplitude scan signals of all transmit-receive chip pairs, forming a three-dimensional data matrix. TFM is used to perform time-delay superposition processing on all amplitude scan signals in the original data matrix to reconstruct the amplitude of each pixel in the image. S321. First, define the imaging area by dividing a two-dimensional pixel grid inside the steel structure to be inspected. S322, Traverse each pixel in the imaging region; S323. For each transmit-receive chip pair in the original data matrix, calculate the theoretical propagation time required for the sound wave to travel from the transmitting chip to the pixel and then back from the pixel to the receiving chip. The formula for calculating the theoretical propagation time is: In the formula, P represents a pixel, i represents the transmitting chip, j represents the receiving chip, and t represents the receiving chip. i,j (P) represents the theoretical propagation time, x represents the horizontal coordinate of the pixel, z represents the vertical coordinate of the pixel, and c represents the longitudinal wave velocity of the ultrasonic wave propagating in the material. S324. Extract the amplitude at the theoretical propagation time from the corresponding amplitude scan signal; S325. Coherently superimpose the amplitude values corresponding to all N×N amplitude scanning signals at the pixel point to obtain the reconstructed amplitude value of the pixel: In the formula, I(P) is the reconstructed amplitude of the pixel, and A i,j (t i,j (P) represents the amplitude at the theoretical propagation time; S33. After traversing all pixels, I(P) is mapped to grayscale or color to form the final TFM image.
7. The ultrasonic testing method for detecting defects in steel structures according to claim 5, characterized in that: The curve cluster data is a distance-amplitude-equivalent curve cluster, and the method for obtaining the curve cluster data is as follows: On a standard steel structure test block, the echo amplitude of artificial defects of different depths and known sizes was measured, and the distance-amplitude relationship curve was plotted, where the distance is the sound path from the defect to the ultrasonic probe (6). By using experiments or acoustic simulation software, the curvature of the distance-amplitude relationship curve is adjusted to obtain a distance-amplitude relationship that better reflects the actual situation.
8. The ultrasonic testing method for detecting defects in steel structures according to claim 5, characterized in that, Step S3 specifically involves: A1. Generating synthetic ultrasound images with specific defects using a conditional diffusion model. A11. Diffusion process: During the training phase, the conditional diffusion model learns to gradually add noise to a real defect image until it becomes pure Gaussian noise, resulting in a gradually blurred image sequence. A12. Reverse diffusion process: The conditional diffusion model learns how to remove noise from the noise and gradually reconstructs the original defective image from the blurred image. Defect condition control is incorporated into the learning process. By injecting defect conditions as additional input information into the denoising process, the model is guided by defect conditions at each step of image generation. A13. Generate a synthetic ultrasound image that meets the specified conditions; A2. The FID, KID, and LPIPS metrics were used to evaluate the quality of ultrasound images, and a self-supervised auxiliary network was used to extract defect features from the ultrasound images. The extracted defect feature data was used for model training. A3. After performing ultrasound image quality assessment, if the ultrasound image quality is invalid, the ultrasound image is reprocessed through steps A1-A2; if the ultrasound image quality is valid, the valid ultrasound image is used for model training.