A robot phased array nondestructive testing defect positioning method based on human-computer cooperation

Abstract

The application discloses a kind of based on man-machine cooperation's robot phased array nondestructive testing defect positioning method, comprising: obtaining the profile image of workpiece to be scanned;It is converted to profile point cloud data, the splicing and filtering of point cloud data are completed;Pick up multiple points in different directions and complete point cloud cutting;Curve surface is obtained by fitting, and robot scanning track is generated;Extract normal information at the feature point of profile on track, determine the robot end posture, generate nondestructive testing scanning path;Robot scans according to scanning path, and phased array scanning probe carries out damage echo data analysis;Scanning data is fitted, and damage cloud chart is formed, the pose information of robot corresponding to defect region is extracted, whether secondary scanning is needed is judged, and the defect damage type scanned is distinguished.If needed, secondary scanning is carried out, the detection process of the application is simple and convenient, manual coordinate system calibration is not needed, secondary scanning of defect position is facilitated, and can be suitable for multi-material, multi-shape workpiece detection operation.

Description

Technical Field

[0001] This invention belongs to the field of nondestructive testing technology for materials, specifically relating to a method for defect localization in robot phased array nondestructive testing based on human-machine collaboration. Background Technology

[0002] In the manufacturing process of advanced composite materials, the difficulty in precisely controlling various process parameters leads to unstable quality and a degree of randomness in the composite structure. Static loads, mechanical damage, fatigue, creep, and overheating during use can also cause damage in composite materials. The generation, propagation, and accumulation of damage exacerbate environmental and stress corrosion, accelerate material aging, severely degrade thermal properties, and drastically reduce strength and stiffness, significantly shortening the structure's service life. Therefore, non-destructive testing of composite structures before and during use is extremely important. Furthermore, the high costs of spacecraft launch and operation necessitate minimizing the weight of spacecraft structures. To reduce the weight and manufacturing costs of aerospace composite structures, damage detection technology must be employed to accurately detect and identify various defects and damage within the material.

[0003] Phased array nondestructive testing (PAD) technology, as an effective means of material defect detection, can detect microscopic defects compared to other testing technologies. It boasts advantages such as high sensitivity, strong penetration capability, precise positioning, harmlessness to materials, and environmental friendliness. Both domestically and internationally, ultrasonic testing is widely used as an important means of controlling and assessing product quality. However, in many industries, PAD still relies on workers manually scanning parts with probes, consuming significant manpower and exhibiting low efficiency. Furthermore, manual scanning greatly affects the reliability of measurement results, easily leading to missed or false defects, causing losses in subsequent production. Additionally, for some large-scale, multi-curvature products, digital models are unavailable, or the materials and components have poor rigidity and large deformation. Traditional methods of generating PAD paths using theoretical digital models are complex and cannot address the discrepancies between the theoretical model and the actual component caused by accumulated errors during actual component manufacturing and assembly. Moreover, complex calibration work is required to ensure the accuracy of NC machining. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for defect localization in robot phased array nondestructive testing based on human-machine collaboration, which addresses the shortcomings of the prior art.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0006] A human-machine collaborative robot phased array nondestructive testing defect localization method includes the following steps:

[0007] S1. Take multiple photos using a binocular camera fixed to the robot's end effector to obtain the outline image of the workpiece to be scanned.

[0008] S2. Convert the contour image into contour point cloud data in the robot base coordinate system, complete the stitching of the point cloud data, and perform filtering processing on the point cloud data.

[0009] S3. On the point cloud image, multiple points in different directions are picked up by human-computer interaction, and the interactive points are connected in sequence according to the picking order to automatically complete the point cloud cropping and obtain the cropped point cloud outline set.

[0010] S4. Perform NURBS surface fitting on the clipped point cloud contour set to obtain the curve surface S. Set the parameters of the scanning trajectory on the host computer and automatically generate the robot scanning trajectory C on the curve surface S according to the set process parameters.

[0011] S5. Extract contour feature points P on the robot's scanning trajectory C. i The normal information at the location is used as the processing path points with the contour feature points, and the posture of the robot end is determined according to the normal information to generate the robot non-destructive testing scanning path;

[0012] S6. The host computer directly sends the non-destructive testing scanning path to the robot. The robot executes the motion trajectory. After the robot moves to the area to be scanned, the host computer sends a command to open the phased array scanning probe and prepare for scanning.

[0013] S7. Data recording is performed using a synchronous task periodic triggering method. At the same time, the task synchronization and data collection threads are started. The robot moves according to the scanning path, and the phased array performs damage scanning. The host computer synchronously collects the phased array scanning echo data and robot motion data at a set frequency and records the timestamp.

[0014] S8. After the scan is completed, the host computer will align the returned damage echo data with the robot motion data according to the timestamp, and construct a visual damage cloud map to analyze the distribution and severity of the damage and determine the defect areas that need to be scanned a second time.

[0015] S9. Using the damage cloud map data obtained from each scan as a sample set, train and establish a defect classification model, and then automatically identify and display the category of the scanned defects.

[0016] S10. For defective areas that require secondary scanning, the host computer automatically extracts the robot's pose information corresponding to the defective area and sends the extracted pose information back to the robot for secondary scanning.

[0017] To optimize the above technical solution, the specific measures also include:

[0018] In step S1 above, the overall detection range of the binocular camera is a square area with a pixel count of 2.3MP, and the camera performs multiple acquisitions of adjacent areas.

[0019] In step S2 above, Gaussian filtering is used to filter the point cloud data, and voxel grid method is used to simplify the point cloud data.

[0020] Step S3 above involves interactively picking multiple points from different orientations on the point cloud image and connecting the interactive points sequentially according to the picking order to automatically complete the point cloud cropping and obtain the cropped point cloud contour set. The specific process includes:

[0021] S31. Pick three or more points in different directions according to processing requirements;

[0022] S32. The three-dimensional position information of the picked point is combined with the normal information to form the slicing plane P. l Slicing plane P l The point cloud data band K, which is divided and enclosed by the point cloud surface E, is the clipped point cloud contour set.

[0023] In step S4 above, the scanning trajectory parameters include scanning trajectory type, scanning point interval, robot moving speed, and turning area radius.

[0024] The scanning preparation in step S6 above refers to the robot moving to the area to be scanned, continuing to execute motion commands to compact the area to be scanned by the non-destructive testing end effector, turning on the water inlet pump on the end effector, and turning on the water outlet pump installed on the end effector after the end effector cavity is filled to achieve dynamic balance between water inlet and outlet.

[0025] In step S7 above, a synchronous task periodic triggering method is adopted, simultaneously starting the task synchronization and data collection threads. A custom synchronous task triggering mechanism is used to trigger data acquisition signals at fixed time intervals. Based on TTL pulse synchronization, phased array scan echo data and robot motion data are acquired, and timestamps are recorded. The acquired phased array scan corresponding damage echo signal data set is P = {q1, q2, ..., q...} n The robot's current set of poses is P. F ={p1, p2, ..., p n This enables the matching of robot pose with damage echo signals during the scanning process.

[0026] Step S8 above specifically includes the following steps:

[0027] S81. The host computer will align the robot pose data and damage echo signal data transmitted back during the scanning process with timestamps, map the probe amplitude information arranged in sequence in the damage echo signal data to the depth of color, and construct the damage cloud map of the workpiece through color coding.

[0028] S82. When the cloud map shows that there is a damage defect in a certain area of ​​the workpiece, the damage cloud map is searched, and the searched defect areas are numbered sequentially. Each time a defect area is added, the count is incremented by 1 until the final search is completed, thus realizing defect counting and defect numbering.

[0029] S83. The area that needs to be scanned again can be manually selected or the damage cloud map can be automatically analyzed by the system according to the preset threshold. When the feature value of the damaged area exceeds the preset threshold, the area will be automatically marked as a defect area that needs to be scanned again, and the robot pose information of the current area will be extracted.

[0030] Step S9 above specifically includes the following steps:

[0031] S91. Classify the damage based on the damage cloud map;

[0032] S92. Classify defects into porosity defects, crack defects, inclusion defects, pit defects, and scratch defects, and label the defect images to form a damage image dataset. Randomly divide the dataset into training set, validation set, and test set.

[0033] S93. Introduce the GAM attention mechanism. GAM includes two attention modules: channel attention and spatial attention. It uses a three-dimensional arrangement to maintain the integrity of information.

[0034] S94. Normalized Wasserstein distance is introduced as the optimization loss function. Each pixel in the anchor box is weighted and its anchor box is modeled as a two-dimensional Gaussian distribution.

[0035] S95. A defect classification model is constructed using the GAM attention mechanism and normalized Wasserstein distance to output the defect classification for a specific cloud map defect.

[0036] The above step S10 specifically includes the following steps:

[0037] S101. Extract the phased array damage echo signal data set corresponding to the point cloud that needs to be scanned a second time. If the damage echo signal data to be extracted is Q = {q1, q2, ..., q...} m If}, then the set of robot poses corresponding to the defect area during scanning is Q. F ={p1, p2, ..., p m}, and send the corresponding robot pose information to the robot;

[0038] S102. Use the EtherCAT industrial real-time Ethernet protocol to transmit signals to the robot. Read the robot's motion pose data through the TCP / IP protocol and compile it into a custom XML format robot pose data. The robot reads the XML pose data and moves to the assigned position, or the operator manually drags the robot's end effector to the location of the workpiece defect to perform a secondary scan.

[0039] The present invention has the following beneficial effects:

[0040] This invention uses a robot as a carrier for non-destructive testing scanning, eliminating the need to obtain the theoretical digital model of the traditional product. This avoids the model registration difficulties caused by production errors between the actual product and the theoretical digital model, effectively eliminating the difficulty in obtaining digital models of some large-configuration, multi-curvature products, as well as the model registration difficulties caused by production errors between the actual product and the theoretical digital model. It also solves the problem that the scanning path of robot NC machining based on the theoretical digital model of the traditional product cannot meet the requirements of non-destructive testing.

[0041] This invention visualizes point cloud data. Operators can interactively select non-destructive testing (NDT) scanning areas in the visualization window according to their own process operation habits. The point cloud data is converted into the motion path of the robot's NDT scanning. The path planning of the NDT scanning is completed, which effectively detects damage and locates defects, making it convenient for the robot to perform secondary scanning. The detection process is simple and convenient to operate, and the workpiece to be scanned can be placed freely without the need for manual coordinate system calibration.

[0042] This invention constructs a workpiece defect cloud map based on the scanning information to display the internal scanning situation of the workpiece. It converts the scanning situation into a three-dimensional cloud map for display, which can realize the quantitative extraction of defects and the robot pose positioning and extraction of defect locations, facilitating secondary scanning of defect locations.

[0043] This invention effectively avoids the problems of low efficiency and high labor intensity of traditional manual inspection, realizes quantitative analysis of defects, and locates and extracts the robot pose information corresponding to the defects, driving the robot to perform a second inspection and scan, improving human-machine interaction and scanning efficiency. Moreover, it does not require product digital models or multiple product coordinate system calibrations, and can automatically convert 3D point clouds and process requirements into robot motion instructions. It is simple and convenient to operate and can be applied to flaw detection operations of workpieces with multiple materials and shapes. Attached Figure Description

[0044] Figure 1 A system flowchart provided for embodiments of the present invention;

[0045] Figure 2 Hardware configuration diagram provided for embodiments of the present invention;

[0046] Figure 3 A flowchart for point cloud trajectory generation provided in an embodiment of the present invention;

[0047] Figure 4 A schematic diagram of a point cloud contour set interactively picked up by a detection system block diagram provided in an embodiment of the present invention;

[0048] Figure 5 This is a block diagram of a detection system provided in an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0050] Although the steps in this invention are arranged by reference numerals, this is not intended to limit the order of the steps. Unless the order of the steps is explicitly stated or the execution of a step requires other steps as a basis, the relative order of the steps can be adjusted. It is understood that the term "and / or" as used herein refers to and covers any and all possible combinations of one or more of the associated listed items.

[0051] Example 1

[0052] refer to Figure 1-5 This invention discloses a human-machine collaborative robot phased array nondestructive testing defect localization method, implemented using a six-degree-of-freedom robot, a nondestructive testing end effector, a phased array probe, and a host computer. The specific implementation includes the following steps:

[0053] S1. Take multiple photos using a binocular camera fixed to the robot's end effector to obtain the outline image of the workpiece to be scanned.

[0054] S2. Convert the contour image into contour point cloud data in the robot base coordinate system, complete the stitching of the point cloud data, and perform filtering processing on the point cloud data.

[0055] S3. Following the non-destructive testing process, the operator interactively picks up multiple points from different directions on the point cloud image and connects the interactive points in the order of picking to automatically complete the point cloud cropping and obtain the cropped point cloud contour set.

[0056] S4. Perform NURBS surface fitting on the clipped point cloud contour set to obtain the curve surface S. The operator sets the parameters of the scanning trajectory on the host computer and automatically generates the robot scanning trajectory C on the curve surface S according to the set process parameters.

[0057] S5. Extract contour feature points P on the robot's scanning trajectory C. i The normal information at the location is used as the processing path points with the contour feature points, and the posture of the robot end is determined according to the normal information to generate the robot non-destructive testing scanning path;

[0058] S6. The host computer directly sends the non-destructive testing scanning path to the robot. The robot executes the motion trajectory. After the robot moves to the area to be scanned, the host computer sends a command to open the phased array scanning probe and prepare for scanning.

[0059] S7. Data recording is performed using a synchronous task periodic triggering method. At the same time, the task synchronization and data collection threads are started. The robot moves according to the scanning path, and the phased array performs damage scanning. The host computer collects phased array scanning echo data and robot motion data at a set frequency and records the timestamp.

[0060] S8. After the scan is completed, the host computer will align the returned damage echo data with the robot motion data according to the timestamp, and construct a visual damage cloud map to analyze the distribution and severity of the damage and identify the defect areas that need to be scanned a second time. The damage cloud map uses color coding to represent the severity of the damage, which can be used by operators to judge the defect areas that need to be scanned a second time based on the damage cloud map, or the system can automatically identify the defect areas that need to be scanned a second time based on the preset threshold.

[0061] S9. Using the damage cloud map data obtained from each scan as a sample set, train and establish a defect classification model, and then automatically identify and display the category of the scanned defects.

[0062] S10. For defective areas that require secondary scanning, the host computer automatically extracts the robot's pose information corresponding to the defective area and sends the extracted robot pose at the defective location back to the robot, driving the robot to move to the defective location of the workpiece according to the pose command for secondary non-destructive scanning, or the operator manually drags the robot's end effector to the defective location of the workpiece for secondary scanning.

[0063] In this embodiment, in step S1, the part to be scanned is placed on the fixture and fixed. After the camera calibration is completed, the 3D camera fixed to the end effector of the robot non-destructive testing takes pictures of the workpiece to be scanned (the area to be processed). The robot end effector is moved so that the binocular camera takes multiple pictures of the area. The detection range of the binocular camera is: the overall detection range of the binocular camera is a square area with a pixel count of 2.3MP. The camera takes multiple pictures of adjacent areas.

[0064] In step S2, Gaussian filtering is used to filter the point cloud data, and voxel grid method is used to simplify the point cloud data.

[0065] In step S2, the information acquired by the camera is converted into a 3D point cloud model. Multiple point cloud images acquired by the camera are stitched together, and the point cloud is transformed into the robot's base coordinate system (that is, multiple sets of point cloud data are rotated into the robot's base coordinate system to form the scanning point cloud data). This stitching of the point cloud further expands the non-destructive testing area, and the obtained 3D point cloud data is filtered. The specific process includes the following steps:

[0066] S21. Convert the images of adjacent areas captured by the camera multiple times into contour point cloud data in the robot's base coordinate system, and complete the stitching of the point cloud data.

[0067] S22. Define a smooth window of size Z×Z, and pre-define a convolution kernel for this window. Let a size x be the Euclidean distance from a certain position within the convolution kernel to the center pixel, f(x) be the value at that position, and e be the standard deviation of the Gaussian function;

[0068] S23. Use a convolution kernel to traverse each pixel in the image to complete the smoothing process of Gaussian filtering;

[0069] S24. Use the voxel grid method to simplify the point cloud data.

[0070] In S24, the specific process includes the following steps:

[0071] S241. Establish a three-dimensional voxel in the point cloud, and use the centroid of the voxel to approximate the data points in the three-dimensional voxel.

[0072] S242, Setting n grid (i) represents the number of point clouds in a voxel, and the number of 3D points P in the voxel. ij (x ij ,y ij ,z ij The center of gravity P grid (i) The calculation method is as follows:

[0073]

[0074] In step S3, based on the stitched point cloud, the stitched point cloud is cropped in the human-computer interaction interface. The operator picks the corner points of the area to be non-destructive tested according to the detection / actual scanning requirements, and crops the stitched point cloud with the picked points as the boundary. The specific process includes the following steps:

[0075] S31. The operator selects three or more interaction points according to the processing requirements.

[0076] S32. The three-dimensional position information of the picked surface feature interaction points is combined with the normal information to form the slicing plane P. l Slicing plane P l The point cloud data band K, which is segmented and enclosed by the point cloud surface E, is the point cloud contour set for interactive picking.

[0077] In step S4, the point cloud data is fitted into a NURBS surface based on the cropped point cloud data, and the scanning trajectory parameters are set according to the scanning requirements to automatically generate the running trajectory / scanning process that the robot can directly execute.

[0078] Manually setting scanning trajectory parameters includes, but is not limited to, setting the scanning trajectory type (Z-shaped, square-shaped, etc.), scanning point interval, robot moving speed, turning area radius, etc.

[0079] In step S5, the specific process includes the following steps:

[0080] S51, with tangent vector n y Normal vector n z sum vector n x These represent the Y-axis, Z-axis, and X-axis directions of the robot's end-effector posture coordinate system, respectively.

[0081] S52. Let the rotation vectors of the X, Y, and Z axes of the robot's end effector coordinate system be normalized to obtain the direction vectors as follows:

[0082] S53, based on the direction vector Calculate the Euler angles of rotation, which represent the pose of the robot's end effector. The expression for the Euler angles of rotation is:

[0083]

[0084] In the above formula, These are the direction vectors obtained after normalizing the rotation vectors of the X, Y, and Z axes of the robot's end effector posture coordinate system.

[0085] The non-destructive testing end effector is also equipped with an inlet pump and an outlet pump. In step S6, the host computer directly sends the non-destructive testing scanning path to the robot. The robot executes the motion trajectory. After the robot moves to the area to be scanned, the host computer sends a command to turn on the phased array scanning probe and prepare for scanning. The scanning preparation refers to turning on the inlet pump on the end effector after the robot moves to the non-destructive testing area and the non-destructive testing end effector compacts the area to be scanned. After the inner cavity of the non-destructive testing end effector is filled, the outlet pump installed on the end effector is turned on to achieve a dynamic balance between water inlet and outlet.

[0086] In step S7, the host computer software initiates the task synchronization and data collection thread. First, it simultaneously sends start signals to both the robot and the phased array controller to ensure data acquisition synchronization. Then, using a custom synchronization task triggering mechanism, it triggers data acquisition signals at fixed time intervals, achieving microsecond-level synchronous acquisition of phased array scan echo data and robot motion data via TTL pulses, and recording timestamps. To ensure low latency and stability in communication with the equipment, the EtherCAT industrial real-time Ethernet protocol is used for signal transmission with the robot, while the robot's motion pose data is read via TCP / IP protocol. A custom XML format for robot pose data is also used to reduce data size and latency. Similarly, in communication with the phased array controller, USB 3.0 is used for high-speed data transmission. Based on the controller's existing UDP protocol, a custom data transmission format is used to reduce the amount of interactive data, ensuring the real-time performance and integrity of the scan echo signal.

[0087] After the scan is completed, the host computer obtains the damage echo signal data set P = {q1, q2, ..., q...} n} and the robot's motion pose set P F ={p1, p2, ..., p n The damage echo signal is the amplitude information of the B-scan echo from the phased array probe. Aligning the robot pose data with the damage echo signal data by timestamp allows for precise matching of pose and signal during the scanning process.

[0088] In step S8, robot pose information and phased array defect information are read periodically. After registering the robot pose information and scanning data, a workpiece defect cloud map is constructed based on the scanning information. Defects are counted, and robot pose localization and extraction of defect locations are achieved. The corresponding robot pose located based on the damage information displayed on the damage cloud map can be easily driven to perform secondary damage detection. Specifically, the steps include:

[0089] S81. The host computer will align the robot pose data and damage echo signal data (including the intensity and position information of the reflected signal during the scan) returned during the scanning process with timestamps, and map the probe amplitude information arranged in sequence in the damage echo signal data to the shades of colors. Red indicates a higher signal amplitude, and blue or light colors indicate a lower signal amplitude. Through color coding, a damage cloud map of the workpiece will be constructed and displayed on the host computer interface.

[0090] S82. When the cloud map shows that there is a damage defect in a certain area of ​​the workpiece, the scan cloud map is searched, and the searched defect areas are numbered sequentially. Each time a defect area is added, the count is incremented by 1 until the final search is completed, thus realizing defect counting and defect numbering.

[0091] S83. The system automatically analyzes the damage cloud map based on the preset threshold and manually selects the area that needs to be scanned a second time. The threshold is set based on the severity of the damage (such as the intensity of the reflected signal, the area of ​​damage, etc.). The threshold can be manually adjusted. When the feature value of the damaged area exceeds the preset threshold, the area will be automatically marked as a defect area that needs to be scanned a second time, and the robot pose information of the current area will be extracted.

[0092] Step S9 specifically includes the following steps:

[0093] S91. Data Acquisition: Operators classify damage based on the damage cloud map;

[0094] S92. Data annotation: Defects are classified into spherical, crack, inclusion, ptis, and scratch defects, etc., and the defect images are annotated using the image annotation tool Labelimg to form a damage image dataset. The dataset is randomly divided into training set, validation set and test set in a 6:2:2 ratio.

[0095] S93. Introduce the GAM attention mechanism. GAM contains two key attention modules: channel attention and spatial attention. It uses three-dimensional arrangement to maintain the integrity of information.

[0096] S94. Normalized Wasserstein Distance (NWD) is introduced as the optimization loss function. Each pixel in the anchor box is weighted and its anchor box is modeled as a two-dimensional Gaussian distribution.

[0097] S95. A defect classification model is constructed using the GAM attention mechanism and normalized Wasserstein distance. After the initial sample accumulation is completed, the defect classification of the region can be output for specific cloud map defects without manual classification.

[0098] In step S10, the robot is driven to perform a secondary scan of the defect location, which specifically includes the following steps:

[0099] S101. Extract the phased array damage echo signal data set corresponding to the point cloud that needs to be scanned a second time. If the damage echo signal data to be extracted is Q = {q1, q2, ..., q...} m If}, then the set of robot poses corresponding to the defect area during scanning is Q. F ={p1, p2, ..., p m}, and send the corresponding robot pose information to the robot;

[0100] S102. Use the EtherCAT industrial real-time Ethernet protocol to transmit signals to the robot. Read the robot's motion pose data through the TCP / IP protocol and compile it into a custom XML format robot pose data. The robot reads the XML pose data and moves to the assigned position, or the operator manually drags the robot's end effector to the location of the workpiece defect to perform a secondary scan.

[0101] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0102] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for defect localization in robot phased array nondestructive testing based on human-machine collaboration, characterized in that, Includes the following steps: S1. Take multiple photos using a binocular camera fixed to the robot's end effector to obtain the outline image of the workpiece to be scanned. S2. Convert the contour image into contour point cloud data in the robot base coordinate system, complete the stitching of the point cloud data, and perform filtering processing on the point cloud data. S3. On the point cloud image, multiple points in different directions are picked up by human-computer interaction, and the interactive points are connected in sequence according to the picking order to automatically complete the point cloud cropping and obtain the cropped point cloud outline set. S4. Perform NURBS surface fitting on the clipped point cloud contour set to obtain the curve surface S. Set the parameters of the scanning trajectory on the host computer and automatically generate the robot scanning trajectory C on the curve surface S according to the set process parameters. S5. Extract contour feature points P on the robot's scanning trajectory C. i The normal information at the location is used as the processing path points with the contour feature points, and the posture of the robot end is determined according to the normal information to generate the robot non-destructive testing scanning path; S6. The host computer directly sends the non-destructive testing scanning path to the robot. The robot executes the motion trajectory. After the robot moves to the area to be scanned, the host computer sends a command to open the phased array scanning probe and prepare for scanning. S7. Data recording is performed using a synchronous task periodic triggering method. At the same time, the task synchronization and data collection threads are started. The robot moves according to the scanning path, and the phased array performs damage scanning. The host computer synchronously collects the phased array scanning echo data and robot motion data at a set frequency and records the timestamp. S8. After the scan is completed, the host computer will align the returned damage echo data with the robot motion data according to the timestamp and build a visual damage cloud map to analyze the distribution and severity of the damage and identify the defect areas that need to be scanned a second time. S9. Using the damage cloud map data obtained from each scan as a sample set, train and establish a defect classification model, and then automatically identify and display the category of the scanned defects. S10. For defective areas that require secondary scanning, the host computer automatically extracts the robot's pose information corresponding to the defective area and sends the extracted pose information back to the robot for secondary scanning.

2. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, In step S1, the overall detection range of the binocular camera is a square area with a pixel count of 2.3MP, and the camera performs multiple acquisitions of adjacent areas.

3. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, In step S2, Gaussian filtering is used to filter the point cloud data, and voxel grid method is used to simplify the point cloud data.

4. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration as described in claim 1, characterized in that, Step S3 involves interactively picking multiple points from different orientations on the point cloud image and connecting these points sequentially according to the picking order. This automatically completes the point cloud cropping, resulting in a cropped point cloud contour set. The specific process includes: S31. Pick three or more points in different directions according to processing requirements; S32. The three-dimensional position information of the picked point is combined with the normal information to form the slicing plane P. l Slicing plane P l The point cloud data band K, which is divided and enclosed by the point cloud surface E, is the clipped point cloud contour set.

5. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, In step S4, the scanning trajectory parameters include scanning trajectory type, scanning point interval, robot moving speed, and turning area radius.

6. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, The scanning preparation in step S6 refers to the robot moving to the area to be scanned, continuing to execute motion commands to compact the area to be scanned by the non-destructive testing end effector, turning on the water inlet pump on the end effector, and turning on the water outlet pump installed on the end effector after the end effector cavity is filled to achieve dynamic balance between water inlet and outlet.

7. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, In step S7, a synchronous task periodic triggering method is adopted, simultaneously starting the task synchronization and data collection threads. A custom synchronous task triggering mechanism is used to trigger data acquisition signals at fixed time intervals. Phased array scan echo data and robot motion data are synchronously acquired based on TTL pulses, and timestamps are recorded. The acquired phased array scan corresponding damage echo signal data set is P = {q1, q2, ..., q...} n The robot's current set of poses is P. F ={p1, p2, ..., p n This enables the matching of robot pose with damage echo signals during the scanning process.

8. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, Step S8 specifically includes the following steps: S81. The host computer will align the robot pose data and damage echo signal data transmitted back during the scanning process with the timestamp, map the probe amplitude information arranged in sequence in the damage echo signal data into the color depth, and construct the damage cloud map of the workpiece through color coding. S82. When the cloud map shows that there is a damage defect in a certain area of ​​the workpiece, the damage cloud map is searched, and the searched defect areas are numbered sequentially. Each time a defect area is added, the count is incremented by 1 until the final search is completed, thus realizing defect counting and defect numbering. S83. The area that needs to be scanned again can be manually selected or the damage cloud map can be automatically analyzed by the system according to the preset threshold. When the feature value of the damaged area exceeds the preset threshold, the area will be automatically marked as a defect area that needs to be scanned again, and the robot pose information of the current area will be extracted.

9. The method for defect localization in robot phased array nondestructive testing based on human-machine collaboration according to claim 1, characterized in that, Step S9 specifically includes the following steps: S91. Classify the damage based on the damage cloud map; S92. Classify defects into porosity defects, crack defects, inclusion defects, pit defects, and scratch defects, and label the defect images to form a damage image dataset. Randomly divide the dataset into training set, validation set, and test set. S93. Introduce the GAM attention mechanism. GAM includes two attention modules: channel attention and spatial attention. It uses a three-dimensional arrangement to maintain the integrity of information. S94. Normalized Wasserstein distance is introduced as the optimization loss function. Each pixel in the anchor box is weighted and its anchor box is modeled as a two-dimensional Gaussian distribution. S95. A defect classification model based on YOLOv7 is constructed using the GAM attention mechanism and normalized Wasserstein distance to output the defect classification of the region for specific cloud map defects.

10. A method for defect localization in robot phased array nondestructive testing based on human-machine collaboration as described in claim 1, characterized in that, Step S10 specifically includes the following steps: S101. Extract the phased array damage echo signal data set corresponding to the point cloud that needs to be scanned a second time. If the damage echo signal data to be extracted is Q = {q1, q2, ..., q...} m If}, then the set of robot poses corresponding to the defect area during scanning is Q. F ={p1, p2, ..., p m }, and send the corresponding robot pose information to the robot; S102. Use the EtherCAT industrial real-time Ethernet protocol to transmit signals to the robot. Read the robot's motion pose data through the TCP / IP protocol and compile it into a custom XML format robot pose data. The robot reads the XML pose data and moves to the assigned position, or the operator manually drags the robot's end effector to the location of the workpiece defect to perform a secondary scan.

Images