Detection method for automatic scanning of fillet weld phased array of petrochemical device

Abstract

The invention discloses a detection method for automatic scanning of a fillet weld phased array of a petrochemical device. The method comprises the following steps: firstly, acquiring pipeline surface information in real time based on an environment sensing sensor, and synchronously generating a real-time three-dimensional environment model containing a fillet weld and a probe pose through an SLAM (Simultaneous Localization and Mapping) technology; extracting a characteristic track of the fillet weld as a reference scanning path; on the basis of the path and the real-time pose, the motion trail of the mechanical arm is dynamically planned, and phased array acoustic beam emission parameters are synchronously calculated and adjusted through the Snell law in a back-stepping mode according to the normal vector of the surface of the pipeline; the mechanical arm is controlled to operate along the planned track and execute ultrasonic detection; and finally, fusing the ultrasonic data with the real-time pose to generate a three-dimensional detection result corresponding to the three-dimensional environment model space. The full-process automation of fillet weld detection is realized, the problems of low efficiency, easy track deviation, inaccurate sound beam regulation and control, difficult data visualization and the like of a traditional method are solved, and the detection precision, efficiency and reliability are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of nondestructive testing technology and relates to a detection method for automatic phased array scanning of fillet welds in petrochemical plants. Background Technology

[0002] In petrochemical plants, pipe fillet welds, such as those at pipe seats, are critical to structural integrity, and their welding quality directly affects the safe operation of the plant. These welds often exhibit complex spatial geometries such as saddle shapes and are prone to internal defects such as cracks and lack of fusion under high temperature, high pressure, and corrosive media. Therefore, reliable non-destructive testing methods are required for their regular inspection.

[0003] Currently, the inspection of such fillet welds mainly relies on traditional manual ultrasonic scanning. This method has significant limitations: low inspection efficiency, results heavily influenced by operator experience; difficulty in maintaining stable probe coupling and precise path on complex curved surfaces, easily leading to missed detections; and the inability to dynamically adjust beam parameters based on probe posture in real time, affecting defect detection rate and positioning accuracy.

[0004] While some existing automated scanning devices can replace some manual operations, they still suffer from poor environmental adaptability. They mostly rely on preset paths or ideal CAD models and cannot adaptively adjust to the actual conditions of the pipe surface (such as rust and unevenness). Furthermore, they lack high-precision real-time positioning and environmental modeling capabilities, which leads to easy deviation of the scanning trajectory and inaccurate sound beam control.

[0005] Therefore, an automated inspection method is needed that integrates intelligent path planning, real-time acoustic beam control, and three-dimensional visualization to improve the accuracy, efficiency, and reliability of fillet weld inspection. Summary of the Invention

[0006] To address the problems existing in the background art, this invention proposes a detection method for automatic phased array scanning of fillet welds in petrochemical plants.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a detection method for automatic phased array scanning of fillet welds in petrochemical plants, comprising the following steps: Based on the real-time pipeline surface information acquired by environmental perception sensors, a real-time 3D environmental model including fillet welds and the real-time pose of the phased array probe are generated simultaneously through synchronous positioning and map building SLAM technology. Based on the real-time three-dimensional environment model, the feature trajectory of the fillet weld is extracted as the benchmark scanning path; Based on the baseline scanning path and the real-time pose, the motion trajectory of the robotic arm is dynamically planned, and the acoustic beam emission parameters of the phased array are adjusted synchronously. The robotic arm is controlled to move along the planned motion trajectory and to perform ultrasonic detection according to the adjusted sound beam emission parameters; The collected ultrasound data is fused with the corresponding real-time pose to generate a three-dimensional detection result corresponding to the real-time three-dimensional environment model space.

[0008] Specifically, the environmental perception sensor includes a 3D laser scanner and a vision sensor; The method of synchronously generating real-time 3D environment model and real-time pose through map construction SLAM technology specifically involves using a tightly coupled fusion algorithm to perform joint state estimation on laser point cloud data, visual image features, and inertial measurement unit data.

[0009] Specifically, the joint state estimation is achieved through factor graph optimization. The state variables of the optimized joint state estimation include the rotation, translation, velocity, gyroscope deviation, and accelerometer deviation of the phased array probe.

[0010] Specifically, the extraction of the feature trajectory of the fillet weld involves identifying and extracting the three-dimensional spatial ridge line of the fillet weld as the feature trajectory by calculating the principal curvature and orientation field of the vertices on the triangular mesh surface of the real-time three-dimensional environment model.

[0011] Specifically, the motion trajectory of the dynamic programming robotic arm is obtained using a time elastic band algorithm; The objective function of the time elastic band algorithm includes trajectory tracking deviation, distance to obstacles, motion dynamics smoothness, and phased array probe coupling state based on force sensor feedback.

[0012] Specifically, the synchronous adjustment of the phased array's acoustic beam emission parameters involves querying the local normal vector of the pipe surface from the real-time three-dimensional environment model based on the real-time pose. With the goal of focusing the sound beam at a predetermined focal point within the workpiece, the direction of the incident sound beam of each element of the phased array in the wedge is calculated by back-calculation based on Snell's law. The emission delay time of each array element is calculated based on the direction of the incident sound beam.

[0013] Specifically, the inverse calculation based on Snell's law satisfies the following vector relationship: ; in, For the first The unit vector of the incident sound beam in each array element within the wedge. The normal vector of the wedge's bottom surface. Let be the unit vector of the refracted sound beam in the workpiece pointing towards the focal point. Let be the local normal vector. It is the ratio of the sound velocity of the wedge to the sound velocity of the workpiece.

[0014] Specifically, the fusion of the acquired ultrasound data with the corresponding real-time pose involves using a precise time protocol to provide a synchronization timestamp for the ultrasound A-scan signal and the real-time pose data of the phased array probe. The ultrasound A-scan signal with consistent timestamps is correlated with the real-time pose data of the phased array probe to form a six-dimensional data point.

[0015] Specifically, generating the three-dimensional detection result involves filling all the six-dimensional data points into a three-dimensional volume data grid according to their real-time spatial pose. Using graphics processor-accelerated volume rendering technology, the three-dimensional volume data mesh is rendered to generate a three-dimensional visualization image that can be viewed and sectioned from any angle.

[0016] Specifically, while controlling the operation of the robotic arm, the motion trajectory is corrected in real time in a closed loop based on the deviation between the real-time pose and the reference scanning path, as well as the feedback from the force sensor.

[0017] Compared with existing technologies, this invention has the following advantages: By constructing a 3D environment model including the fillet weld in real time using SLAM technology and simultaneously locating the probe pose, the detection system achieves autonomous perception and precise positioning of complex pipeline environments. By extracting the 3D ridge line of the fillet weld as the reference path and using a time-elastic band algorithm to dynamically plan the robotic arm's trajectory, the probe can adapt to changes in pipeline curvature and stably track the weld. By combining Snell's law to calculate the incident direction of the sound beam in real time and adjusting the emission delay, the sound beam is guaranteed to always be precisely focused on the predetermined focal point within the workpiece. Finally, by synchronizing ultrasonic data and spatial pose through a precision time protocol and using GPU accelerators to generate 3D visualization results, the accurate positioning and intuitive presentation of the detection data and the spatial position of the weld are achieved. This method is fully automated, effectively solving the problems of low efficiency, easy trajectory deviation, inaccurate sound beam control, and difficulty in data visualization in traditional manual scanning, significantly improving the accuracy, efficiency, and reliability of fillet weld detection. Attached Figure Description

[0018] Figure 1 This is a flowchart of a detection method for automatic phased array scanning of fillet welds in petrochemical plants according to the present invention; Figure 2 This is a schematic diagram of the installation of the phased array scanning device for inspecting fillet welds of petrochemical plant pipe seats according to the present invention; Figure 3 This is a field implementation diagram of phased array testing of fillet welds on pipe seats in a petrochemical plant according to the present invention; Figure 4 This is a schematic diagram of the control unit of the phased array scanning device of the present invention; Figure 5This is a three-dimensional visualization result of the phased array corner weld inspection of the present invention; Figure 6 This is a schematic diagram of the circumferential detection position marking of the fillet weld of the pipe seat according to the present invention; Figure 7 This is a diagram showing the working status of the phased array scanning device for the fillet weld of the pipe seat according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figures 1-7 As shown, the technical solution adopted by the present invention is as follows: a detection method for automatic phased array scanning of fillet welds in petrochemical plants, comprising the following steps: Based on the real-time pipeline surface information acquired by environmental perception sensors, a real-time 3D environmental model including fillet welds and the real-time pose of the phased array probe are generated simultaneously through synchronous localization and map building SLAM technology.

[0021] Based on the real-time 3D environment model, the feature trajectory of the fillet weld is extracted as the reference scanning path.

[0022] Based on the baseline scanning path and the real-time pose, the motion trajectory of the robotic arm is dynamically planned, and the acoustic beam emission parameters of the phased array are adjusted synchronously.

[0023] The robotic arm is controlled to move along the planned motion trajectory and to perform ultrasonic detection based on the adjusted sound beam emission parameters.

[0024] The collected ultrasound data is fused with the corresponding real-time pose to generate a three-dimensional detection result corresponding to the real-time three-dimensional environment model space.

[0025] A detection method for automatic phased array scanning of fillet welds in petrochemical plants requires hardware implementation, including: moving and installation hardware, detection and sensing hardware, and control and processing hardware. The mobile and installation hardware are as follows: (1) Quick-release type chain: As the installation carrier of the walking device, the walking device can be quickly fixed on the straight pipe with the quick-release design, which is convenient for on-site installation and disassembly. It can also be adapted to the straight pipe diameter range of Φ100mm~Φ300mm, providing a stable installation foundation for subsequent components such as phased array probes and encoders, and ensuring the stability of the overall hardware relative position to the pipeline during testing.

[0026] (2) Small magnetic wheel automatic crawler: It can provide automatic movement power and can automatically scan the straight pipe axis. It adheres to the pipe surface with the help of magnetic adsorption force to prevent the crawler from shifting due to pipe tilting or vibration, ensuring the stability of the scanning trajectory. It can also be used with the handle button to control the start and stop of scanning and speed adjustment, reducing the intensity of manual intervention.

[0027] (3) Multi-degree-of-freedom robotic arm: used to carry end-effector detection tool heads (such as phased array probes, various sensors, etc.), and adjust the spatial attitude and position of the tool head through six degrees of freedom (translation along the X, Y, and Z axes and rotation around the three axes). It can adaptively adjust the angle of the phased array probe according to the real-time environment model (such as pipe curvature, weld position, etc.) to ensure that the phased array probe wedge is perpendicular to the pipe wall and well coupled, thus solving the problem of scanning trajectory deviation at complex structures such as pipe bends and tees.

[0028] The detection and sensing hardware are as follows: (1) Phased array probe and wedge: The phased array probe transmits and receives ultrasonic signals by controlling the array elements to realize ultrasonic detection of internal defects (such as cracks, pores, lack of fusion, etc.) in the fillet weld of the petrochemical unit. The wedge can adjust the incident angle of the ultrasonic beam so that the sound beam is accurately pointed to the weld area. At the same time, it fills the gap between the phased array probe and the pipe wall to ensure effective propagation of ultrasonic signals, improve detection sensitivity and defect detection rate, and can be adapted to pipes of different diameters. By adjusting the wedge parameters or the phased array probe posture, it can meet the full-area detection requirements of the saddle-shaped intersection fillet weld.

[0029] (2) Dual-axis encoder, including angle encoding and position encoding: can acquire position and angle data in real time. Position encoding is used to record the axial or circumferential movement distance of the phased array probe in the pipeline, avoiding defect coordinate offset caused by the accumulation of positioning errors of traditional encoders. Angle encoding is used to capture the relative angle change between the phased array probe and the pipeline surface, and combined with the pipe diameter data, realizes the real-time correspondence between the movement position of the phased array probe and the pipe seat contour line, providing accurate spatial coordinate reference for defect location, and ensuring that the position displayed in the detection map is consistent with the actual movement position of the phased array probe.

[0030] (3) Three-dimensional laser scanner: By emitting a laser beam and receiving the reflected signal, it acquires high-density three-dimensional point cloud data of the detection area (including the pipe surface, weld contour, surrounding obstacles, etc.). These data are used to construct a real-time three-dimensional environment model, accurately extract the three-dimensional ridge line of the fillet weld (as a global reference scanning path), and identify obstacles such as pipe surface corrosion and insulation layer support blocks, providing a geometric basis for adaptive path planning. Its positional accuracy reaches the millimeter level, which can help correct the flatness deviation of the phased array probe coupling surface and improve the detection reliability.

[0031] (4) Visual sensors are used to assist in environmental perception and feature recognition.

[0032] (5) Inertial Measurement Unit (IMU): It can provide high-frequency inertial data. Its built-in gyroscope measures the angular velocity (rotational motion) of the detection system (such as a robotic arm or phased array probe), and the accelerometer measures the linear acceleration (translational motion). The output data frequency reaches hundreds to thousands of hertz. In conjunction with laser point cloud and visual image features, it realizes real-time pose estimation of the system through a tightly coupled fusion algorithm, corrects the IMU zero drift error, avoids the positioning delay problem of relying solely on encoders or vision, and provides dynamic constraints for the motion control of multi-degree-of-freedom robotic arms, ensuring the stability of the phased array probe during movement and reducing the risk of missed detection.

[0033] (6) Force sensor: It can monitor the contact pressure between the phased array probe and the pipe wall in real time and feed the pressure data back to the adaptive clamp control system. When the pipe surface is uneven or the phased array probe moves to the curvature change area, the force sensor can capture the coupling pressure fluctuation, trigger the clamp to adjust the clamping force, ensure that the phased array probe is always well coupled with the pipe wall, and provide data support for the optimization of coupling quality items in path planning, and ensure the consistency of detection conditions during the scanning process.

[0034] The control and processing hardware includes: (1) an industrial-grade computer (including a high-performance GPU): the high-performance GPU is used to run the SLAM system (Simultaneous Localization and Mapping), adaptive path planning (TEB algorithm), dynamic beamforming and other core algorithms to realize the real-time fusion and computation of multi-sensor data (laser, vision, IMU, encoder, etc.). It stores the preset pipe CAD diameter data and detection parameters, and records the ultrasonic signals, phased array probe real-time pose, environmental model and other data during the detection process. It also provides a human-machine interface for operators, supports the generation and viewing of three-dimensional visualization detection reports, and realizes the traceability management of detection data.

[0035] (2) Phased array ultrasonic transducer: This is the ultrasonic signal processing and control component. It receives the ultrasonic echo signal from the phased array probe, and generates detection maps such as A-scan, B-scan, and C-scan through signal amplification, filtering, and digital processing. It can import CAD pipe diameter specification data and combine it with dual-axis encoder data to realize automatic calculation and correction of defect level and depth deviations. It communicates with an industrial-grade computer via high-speed Ethernet to synchronously receive control commands (such as beam focusing position and transmission delay) and feed back detection data to ensure real-time coordination between beam control and phased array probe movement, thereby improving defect positioning accuracy.

[0036] During the ultrasonic testing process, the basic signal acquired by the phased array ultrasonic instrument is the ultrasonic A-scan signal, which contains complete depth and amplitude information of a single detection point. By using a robotic arm to drive the probe along a predetermined path, the A-scan signals of multiple consecutive measurement points on a scan line are combined to generate a B-scan image that reflects the longitudinal cross-sectional shape of the weld. Furthermore, by using the probe to perform a two-dimensional gridded scan on the weld surface, the feature values ​​(such as maximum amplitude) of all A-scan signals within a specific depth range are extracted and mapped according to their spatial positions to synthesize a C-scan image that can intuitively display the planar distribution of defects.

[0037] (3) Synchronization trigger (including precision time protocol hardware): The precision time protocol (PTP) is used to provide microsecond-level timestamps for hardware such as phased array ultrasonic instruments, dual-axis encoders, three-dimensional laser scanners, vision sensors, and IMUs. This ensures that each ultrasonic A-scan signal, point cloud data, image frame, and inertial data is bound to a unique time identifier, avoiding real-time pose estimation deviations or defect coordinate misalignments caused by asynchronous data. This provides a foundation for data synchronization and three-dimensional visualization imaging, and enables precise binding of ultrasonic detection data with the spatial position of the phased array probe.

[0038] (4) Adaptive clamp (including force-controlled servo motor): Used to fix the phased array probe and adjust the coupling pressure. The force-controlled servo motor adjusts the clamping force of the clamp on the phased array probe in real time according to the feedback signal of the force sensor, adapting to the curvature changes of pipes with different diameters. The size of the clamping arm can be adjusted according to the pipe diameter specifications, solving the problem that the fixed-size magnetic wheels / tracks of existing scanning devices are only suitable for specific pipe diameters. In conjunction with the movement of the robotic arm, it ensures that the phased array probe maintains a stable posture during the scanning process, avoiding detection errors or missed detections caused by loose clamping.

[0039] Specifically, the environmental perception sensors include a 3D laser scanner and a vision sensor.

[0040] Furthermore, the 3D laser scanner, based on laser ranging technology, emits a high-frequency, highly stable laser beam at a preset scanning frequency and angle interval, continuously projecting laser signals into the detection area (including petrochemical pipes with fillet welds, surrounding structures, and potential obstacles). Upon contact with the target surface, the laser beam is reflected. The scanner's built-in high-precision photodetector receives the reflected signal. Combining the time difference (or phase difference) between laser emission and reception, and based on the principle of the constant speed of light (distance = speed of light × time difference / 2, the phase difference method calculates the optical path difference through phase change), the spatial distance between the scanner and each sampling point on the target surface is precisely calculated.

[0041] Meanwhile, the angle measurement module integrated into the scanner (such as a high-precision encoder) records the emission angle (horizontal and vertical angle) of each laser beam. Combined with the scanner's initial installation position parameters, the three-dimensional coordinates of each sampling point in the world coordinate system are determined through spatial geometric calculations (spherical coordinates to rectangular coordinates), ultimately forming the three-dimensional point cloud data of the detection area.

[0042] By acquiring high-density, high-precision 3D point clouds of complex structures such as pipe surfaces, fillet weld contours (e.g., saddle-shaped intersection lines), pipe bends, tees, and surrounding obstacles (e.g., insulation layer support blocks, field equipment), the point cloud density can reach hundreds of thousands to millions of points per square meter, and the positional accuracy can reach the millimeter level, providing fine-grained geometric data for subsequent weld model extraction and obstacle identification.

[0043] By capturing detailed features such as rust and uneven deformation on the pipe surface, this information can help determine the flatness of the phased array probe coupling surface, providing a reference for adjusting the clamping force of the adaptive clamp and avoiding phased array probe decoupling problems caused by uneven surfaces.

[0044] A single scanner can cover a range of several meters to tens of meters. Combined with the movement of a mobile platform, it can achieve complete point cloud acquisition of the inspection area of ​​corner welds in long-distance petrochemical pipelines, ensuring no blind spots.

[0045] The vision sensor employs a high-frame-rate, high-resolution industrial camera (typically a CCD or CMOS image sensor), paired with a customized optical lens (such as a fixed-focus lens to ensure image clarity and viewing angle stability) and a supplementary lighting unit (such as a high-brightness LED ring light source to provide uniform and stable illumination and eliminate ambient light interference). The camera continuously acquires two-dimensional images of the detection area at a preset frame rate (usually no less than 30 frames per second to ensure smooth capture of dynamic scenes). The image sensor converts the light signal into an electrical signal, and after analog-to-digital conversion and image preprocessing (such as noise reduction and white balance correction), it outputs a digital image containing information such as target contours, textures, and colors. By extracting visual information such as feature points (such as ORB feature points, which have rotation and scale invariance) and edges (such as the boundary line between pipes and welds, and the contour edges of obstacles) from the image, two-dimensional feature constraints are provided for subsequent fusion with laser point clouds and SLAM localization.

[0046] By comparing the positional changes of the same feature points in consecutive frames of images, the motion trajectory (translation and rotation) of the visual sensor (and the detection system equipped with it) can be initially estimated, providing basic data for the initial real-time pose estimation of SLAM. In particular, in areas where laser point cloud features are not obvious (such as smooth pipe surfaces), visual features can compensate for the deficiencies of laser perception.

[0047] By utilizing the color and grayscale differences in the image (such as the difference in grayscale between the weld area and the pipe base material), the approximate location and direction of the fillet weld can be quickly identified through image segmentation algorithms (such as threshold segmentation and edge detection). This provides initial area guidance for subsequent laser point cloud extraction of weld ridge lines, thereby improving weld extraction efficiency.

[0048] The supplementary lighting unit can automatically adjust its brightness according to the ambient light intensity to ensure clear images can be obtained in strong light, backlight or low light environments; the high frame rate feature can capture dynamic changes in the pipe surface and surrounding environment (such as sudden obstacles) during the movement of the detection system, providing timely visual feedback for real-time obstacle avoidance.

[0049] The method of synchronously generating real-time 3D environment model and real-time pose through map construction SLAM technology specifically involves using a tightly coupled fusion algorithm to perform joint state estimation on laser point cloud data, visual image features, and inertial measurement unit data.

[0050] A tightly coupled algorithm is used to perform joint state estimation on laser point cloud, visual image features, and IMU data. The steps are as follows: Laser point cloud processing: Point-to-surface matching is performed using the Iterative Closest Point (ICP) algorithm, with the objective function being: ; In the formula, For laser points, For CAD model points, for The normal vector is matched with a threshold of 1mm to correct positional deviations. This ensures that the laser point cloud closely matches the actual contour of the pipeline.

[0051] Visual feature extraction: Extract ORB feature points, 1000 per frame, pyramid level 8, FAST threshold 20, calculate reprojection error: ; In the formula For image feature points, For projection function, For camera internal parameters, The laser point cloud is a 3D point cloud. In textureless areas (such as smooth pipe surfaces), the matching accuracy decreases, leading to modeling errors. ORB feature points, however, are insensitive to illumination and can still be stably extracted on pipe surfaces; the reprojection error constrains the visual positioning accuracy, compensating for the shortcomings of laser-based point cloud analysis in textureless areas.

[0052] IMU pre-integration: Sampling rate 1000Hz, pre-integration error: ; In the formula For position increment, For angle increments, For acceleration, Angular velocity, , Zero bias. Laser and vision data acquisition have intermittent periods (e.g., laser 2000Hz, vision 100Hz), during which no positioning data is available, easily leading to real-time pose interruption. IMU pre-integration fills in these data intermittent periods, ensuring continuous real-time pose; multi-error fusion reduces the impact of single errors, achieving a positioning delay of <10ms, meeting real-time detection requirements.

[0053] The final joint state estimate is obtained through fusion. , , Achieving a positioning delay of <10ms meets the requirements of automated and high-precision detection.

[0054] Specifically, the joint state estimation is achieved through factor graph optimization. The state variables of the optimized joint state estimation include the rotation, translation, velocity, gyroscope deviation, and accelerometer deviation of the phased array probe.

[0055] Furthermore, the factor graph optimization framework is constructed using Google Ceres Solver, which includes three types of factors: Laser factor: based on ICP matching error Weight 3.

[0056] Visual Factor: Based on Reprojection Error Weight 3.

[0057] IMU factor: based on pre-integration error Weight 2.

[0058] Simultaneously, a pipe diameter curvature constraint factor is introduced: theoretical curvature is obtained by importing CAD pipe diameter data. Laser point cloud computing actual curvature Constraint error With a weight of 5, ensure that the model is compatible with pipe diameters from Φ100mm to Φ300mm.

[0059] In practical applications, the defect location needs to be determined based on the curvature of the specific detection location. Without pipe diameter curvature constraints, the model will not accurately match the actual pipe diameter. Therefore, the pipe diameter curvature constraint factor has the highest weight to ensure the model strictly adapts to pipe diameters from Φ100mm to Φ300mm (R=50~150mm). Multi-factor weighting is balanced to account for the accuracy characteristics of different sensors.

[0060] Definition of state variables for joint state estimation.

[0061] The state vector is the real-time pose sequence within the sliding window. The meanings and constraints of each variable are as follows: Rotation amount Rotation group parameterization (via quaternions) express, ,satisfy ), describes the phased array probe swing angle, ranging from ±15°, and adapts to the phased array probe clamping arm to adapt to the pipe curvature.

[0062] Translation amount Used to constrain three-dimensional coordinates (x, y, z) with an accuracy of ±0.1 mm, eliminating the accumulation of encoder positioning errors.

[0063] speed quantity Constrained axial movement speed, ranging from 5 to 10 mm / s, to meet the requirements. ( (To set the speed), ensure a smooth scan.

[0064] Gyroscope bias Range ±0.1 / h, accelerometer deviation Range ±0.01m / s 2 Sensor drift is avoided through IMU pre-integration correction.

[0065] The optimization solution adopts the Gauss-Newton method, with ≤10 iterations to ensure convergence of state estimation and meet the requirements of both real-time performance and accuracy.

[0066] In practical applications, manual scanning is difficult to guarantee stability, and unconstrained state variables can easily lead to motion deviations, affecting coupling and detection accuracy. Rotation constraints adapt to the phased array probe's swing angle requirements, translation constraints eliminate encoding errors, velocity constraints ensure smooth scanning, and zero-bias constraints reduce IMU drift, comprehensively ensuring detection stability.

[0067] Specifically, the extraction of the feature trajectory of the fillet weld involves identifying and extracting the three-dimensional spatial ridge line of the fillet weld as the feature trajectory by calculating the principal curvature and orientation field of the vertices on the triangular mesh surface of the real-time three-dimensional environment model.

[0068] Furthermore, the real-time 3D environment model is a triangular mesh, and the principal curvature calculation steps are as follows: Take any vertex in the grid Select its neighboring points (directly adjacent vertices). The number of neighboring points is adjusted according to four special points, namely: the two highest points and the two lowest points of the intersection line of the main fillet weld. These are not isolated points, but special points with distinct characteristics on the intersection line of the fillet weld (saddle shape). Each special point has a continuous feature area with the same curvature characteristics around it. Therefore, the number of neighboring points is adjusted according to four special points.

[0069] Two highest point regions: number of neighboring points This ensures the stability of planar features.

[0070] Regions with two lowest points: Number of neighboring points To capture arc-shaped features.

[0071] Transition region: Number of neighboring points To balance accuracy and efficiency.

[0072] This solution addresses the issue of varying curvature radii at different locations and significant differences in calculation accuracy between planar and arc-shaped regions due to the fixed number of neighboring points. In planar regions (highest points), a larger number of neighboring points ensures stable curvature calculations. In arc-shaped regions (lowest points), fewer neighboring points allow for precise capture of curvature changes. A balance between accuracy and efficiency is achieved in the transition region.

[0073] Construct the covariance matrix of the neighborhood points: ;in The three-dimensional coordinates of the neighboring points The mean of the neighborhood points; Covariance matrix, a 3×3 matrix, describes the spatial distribution characteristics of neighborhood points. : Number of neighboring points (adjusted by region: highest point) Lowest point transition zone ). : Matrix transpose.

[0074] For covariance matrix Perform eigenvalue decomposition to obtain eigenvalues. principal curvature , , corresponding feature vector .in, The main direction is the direction of the center line of the fillet weld. In the direction of maximum principal curvature, This is the direction of minimum principal curvature.

[0075] In differential geometry, the curvature of a surface at a point is typically described by two mutually perpendicular principal directions, which correspond to the maximum and minimum principal curvatures of the surface at that point. Taking a raised structure formed by a fillet weld as an example, at any point on its top, there are two naturally orthogonal directions.

[0076] The first principal direction (direction of maximum principal curvature): along the cross-sectional direction perpendicular to the centerline of the fillet weld. In this direction, the surface slopes sharply downwards from the highest point of the weld to the base metal surface, exhibiting the most dramatic change in curvature; therefore, the curvature in this direction is defined as the maximum principal curvature. .

[0077] The second principal direction (direction of minimum principal curvature): the tangent direction along the centerline of the fillet weld. In this direction, the surface changes relatively gently along the weld's extension path, and its curvature is defined as the minimum principal curvature. .

[0078] The direction of maximum principal curvature identified through eigenvector calculation is the aforementioned direction of maximum principal curvature. This direction is not directly the direction of the fillet weld centerline, but rather perpendicular to it. By first determining the direction of maximum principal curvature of all vertices, and then finding the continuity of their normal directions (i.e., the direction of minimum principal curvature), or by directly locating the curvature extrema, the three-dimensional spatial ridge of the fillet weld, i.e., the actual centerline scanning path, can be accurately traced and extracted.

[0079] The three-dimensional spatial ridge line extraction uses the Marching Lines algorithm to extract the fillet weld ridge lines. The steps are as follows: Along the direction of maximum principal curvature Search for extreme points: Perform linear interpolation on the grid edges, when the principal curvature... When the value changes from positive to negative, the interpolation point is the ridge point, satisfying the condition... ( (These are interpolation parameters), positioning accuracy ≤ 0.3mm.

[0080] The extreme point is the center line of the weld, which has high positioning accuracy and ensures that the trajectory completely coincides with the weld. Linear interpolation adapts to the discrete characteristics of the mesh, avoiding positioning deviations caused by mesh accuracy.

[0081] Connecting ridge points: The discrete ridge points are smoothed using cubic B-spline, and the spline equation is: In the formula: As control points, Use B-spline basis functions, with a control point spacing of 0.3 mm to ensure smooth ridge lines; Ridge alignment with encoder: Each 0.1mm ridge length corresponds to one position encoding pulse of the dual-axis encoder, ensuring that the trajectory is synchronized with the encoder data and avoiding deviation of the scanning trajectory.

[0082] B-spline smoothing eliminates ridge burrs, ensuring smooth robotic arm movement; alignment with the encoder enables bidirectional traceability of ridge position and encoded data, facilitating error verification and defect location.

[0083] Specifically, the motion trajectory of the dynamic programming robotic arm is obtained using a time elastic band algorithm; The objective function of the time elastic band algorithm includes trajectory tracking deviation, distance to obstacles, motion dynamics smoothness, and phased array probe coupling state based on force sensor feedback.

[0084] Furthermore, the extracted feature trajectory (reference scanning path) is discretized into N real-time pose points, where N is adjusted according to the pipe diameter: for Φ100mm, N=200 (1 point per 0.5mm); for Φ300mm, N=100 (1 point per 1mm). During the dynamic planning of the robotic arm's motion trajectory using the time-elastic band algorithm, the reference scanning path of the fillet weld is first discretized into a series of ordered real-time pose points, where the... A real-time pose point is represented by a symbol. Representation. Each real-time pose point. Both contain two sets of key information that together describe the spatial state of the phased array probe at the end of the robotic arm at that moment.

[0085] The first set is three-dimensional position information, consisting of three parameters, which represent the specific coordinates (usually in millimeters) of the phased array probe in the X-axis, Y-axis and Z-axis directions in the three-dimensional coordinate system, and are used to determine the accurate position of the phased array probe in space. The second set is three-dimensional attitude information, consisting of three parameters, representing the roll angle, pitch angle, and yaw angle (usually in degrees) of the phased array probe about its own X-axis. These parameters are used to determine the relative attitude of the phased array probe and the pipe surface, ensuring that the phased array probe wedge can be perpendicularly attached to the pipe surface.

[0086] Meanwhile, to ensure that the planned motion trajectory is synchronized with the actual control rhythm of the robotic arm, the period of the time elastic band is set to 3 milliseconds. This period is completely consistent with the control cycle of the robotic arm, meaning that the robotic arm will receive and execute a motion command corresponding to a real-time pose point every 3 milliseconds. This avoids trajectory deviations caused by a mismatch between the rhythm of trajectory command issuance and the rhythm of the robotic arm's response, ensuring that the phased array probe can move smoothly and accurately along the planned trajectory, meeting the real-time and stability requirements of automated scanning.

[0087] The objective function is optimized by constructing a weighted sum of four terms, as shown in the following formula: The meanings and parameters of each item are as follows: Total optimization error, the objective function value that integrates multiple constraints.

[0088] Tracking deviation term : In the formula, To plan real-time pose, (Based on real-time pose), weights =10, ensuring the trajectory fits the fillet weld.

[0089] Obstacle distance item : = In the formula, The weight represents the distance between the real-time pose point and the pipe protrusion. The protrusion of the tube seat affects the detection; setting... .

[0090] Motion smoothness item : In the formula, For the joint velocity of the robotic arm, weight Ensure that speed fluctuation is ≤5%.

[0091] Coupling state terms Based on a spring coupling structure, In the formula, The spring constant is... The spring compression is 20N, which is the optimal coupling pressure, and the weights are... ,when (F<15N) or When F>35N, Temporarily upgraded to 10.

[0092] Chain range constraint Based on quick-release chains, In the formula, To plan the axial length of the trajectory, (Effective walking length of the chain), weight This ensures that the trajectory does not exceed the range of the chain.

[0093] Multiple constraints cover trajectory accuracy, collision avoidance, smooth motion, stable coupling, and chain adaptation, ensuring comprehensive reliability of the scan.

[0094] The optimization problem is solved using the Levenberg-Marquardt algorithm, with ≤5 iterations and a solution time ≤2ms, meeting real-time requirements. The optimized trajectory is sent to the robotic arm driver via EtherCAT bus, ensuring a position tracking error of <0.05mm, adapting to scanning requirements for pipe diameters from Φ100mm to Φ300mm. The trajectory accurately follows the reference path; automated solving eliminates the need for manual intervention, improving inspection efficiency.

[0095] Specifically, the synchronous adjustment of the phased array's acoustic beam emission parameters involves querying the local normal vector of the pipe surface from the real-time three-dimensional environment model based on the real-time pose.

[0096] With the goal of focusing the sound beam at a predetermined focal point within the workpiece, the direction of the incident sound beam of each element of the phased array in the wedge is calculated by back-calculation based on Snell's law.

[0097] The emission delay time of each array element is calculated based on the direction of the incident sound beam.

[0098] Furthermore, the local normal vector query for the pipe surface is based on the imported main pipe diameter data from CAD. First, the theoretical local normal vector is calculated, and then corrected using laser point cloud. The steps are as follows: Theoretical normal vector calculation: Given pipe diameter D, center coordinates Coordinates of the detection point Then the theoretical normal vector is , =2 / D is the pipe radius.

[0099] Laser correction: Laser point cloud accuracy 0.05mm, calculate the normal vector of the laser point at the detection point. .

[0100] The data is obtained by high-density point cloud data acquired by a 3D laser scanner. Based on the spatial distribution characteristics of multiple laser sampling points in the neighborhood of the detection point, it reflects the actual micro-geometry of the pipe surface, including real surface features such as rust and unevenness.

[0101] The actual local normal vector is calculated by fusing the theoretical normal vector with the laser point normal vector. This process preserves the macroscopic geometric constraints of the pipe while incorporating actual surface features, ultimately serving as a key parameter for acoustic beam control. Specifically... , For correction amount, This ensures the normal vector accuracy is ≤0.5°. It also ensures the normal vector is perpendicular to the actual surface of the pipe, laying the foundation for precise sound beam control.

[0102] The calculation of the incident sound beam direction and emission delay time, and the sound beam control aimed at focusing the sound beam on the center of the fillet weld, are as follows: Focus position setting: fillet weld thickness =8~20mm (common specifications for petrochemical plants), focal depth Focal coordinates .

[0103] Calculation of refracted sound beam vector: The refracted sound beam points towards the focal point, and the unit vector is... In the formula, , , The coordinates are the center coordinates of the phased array probe element.

[0104] The focal point is located at the center of the weld, ensuring that the sound beam energy is concentrated in the defect-prone area; refraction unit vector Pointing to the focal point provides a target reference for adjusting the direction of the sound beam.

[0105] Inverse direction deduction of incident sound beam: Based on Snell's law, combined with the plexiglass wedge (speed of sound) ) and carbon steel workpiece (transverse wave velocity) ), incident sound beam unit vector satisfy .

[0106] Transmission delay calculation: Element spacing is 2mm (common specification for phased array probes), distance from element center to reference point. The delay time is This ensures that the acoustic beams of each array element are superimposed at the focal point, with a delay time accuracy of ≤0.01μs.

[0107] Snell's law ensures that the sound beam is accurately refracted within the workpiece (deviation ≤1mm), eliminating the need for post-processing correction; the delay time is highly accurate, and the sound beams of each array element are synchronously focused, improving the sensitivity of defect detection.

[0108] Specifically, the inverse calculation based on Snell's law satisfies the following vector relationship: ; in, : No. The unit vector of the incident sound beam of each phased array element in the wedge is oriented by the phased array probe tilt angle. When adapting to pipe curvature, and The included angle is ,satisfy .

[0109] : The normal vector of the wedge bottom surface and the local normal vector of the pipe surface Parallel, because the wedge is in contact with the pipe surface, ensuring that the wedge is perpendicular to the pipe wall, that is... .

[0110] The unit vector of the refracted sound beam in the workpiece pointing towards the focal point, which is the center of the fillet weld. The direction is along the weld thickness direction, satisfying... It is perpendicular to the surface of the pipe.

[0111] The local normal vector of the pipe surface is obtained by laser correction calculation, with high accuracy. The curvature is adjusted according to the pipe diameter from Φ100mm to Φ300mm. For Φ100mm... Change rate ≥ Φ300mm.

[0112] The ratio of the sound velocity of the wedge to the sound velocity of the workpiece.

[0113] pass It can accurately establish the correlation between the incident sound beam, the refracted sound beam, and the pipe curvature. Within the pipe diameter range of Φ100mm~Φ300mm, it ensures that the deviation between the refracted sound beam and the center of the fillet weld is ≤1mm, avoiding the need to calculate and correct the error of horizontal and depth deviation after marking, thus meeting the requirements of high-precision defect positioning.

[0114] Specifically, the fusion of the acquired ultrasound data with the corresponding real-time pose involves using a precise time protocol to provide a synchronization timestamp for the ultrasound A-scan signal and the real-time pose data of the phased array probe.

[0115] The ultrasound A-scan signal with consistent timestamps is correlated with the real-time pose data of the phased array probe to form a six-dimensional data point.

[0116] Furthermore, in the entire phased array automatic scanning and inspection process of fillet welds in petrochemical plants, the precise time protocol synchronization mechanism and the construction of six-dimensional data points are the core links for realizing the real-time correspondence between the phased array probe's moving position and the detection spectrum.

[0117] First, to ensure accurate matching between ultrasonic testing data and the spatial position data of the phased array probe, the IEEE 1588 precision time protocol is used as the synchronization standard, and a dual-axis encoder for positioning is used as the core of the synchronization trigger. The dual-axis encoder's position encoding is based on its own 0.1mm resolution. Whenever the phased array probe moves 0.1mm along the pipe axis (i.e., the position encoding value changes by 1 unit), a synchronization pulse signal is automatically generated. This pulse signal is rapidly transmitted via an FPGA (Field Programmable Gate Array) chip to the ultrasonic testing instrument and the phased array probe for real-time pose acquisition, effectively providing a unified data acquisition command signal to ensure that both start data recording at the same time.

[0118] Triggered by a synchronization pulse, the ultrasonic testing instrument and the phased array probe generate corresponding data timestamps in real-time pose acquisition. The ultrasonic instrument records the timestamp of the currently acquired ultrasonic A-scan signal, denoted as . The phased array probe's real-time pose acquisition module (combining SLAM system and dual-axis encoder data) records the timestamp of the current phased array probe's three-dimensional spatial real-time pose, denoted as . To avoid data misalignment due to time differences, the deviation between the two timestamps must be strictly controlled to ensure... This precision ensures a unique correspondence between the ultrasonic signal at a given moment and the spatial position of the phased array probe at that moment, resolving the disconnect between the image and location from a temporal perspective. Furthermore, to facilitate subsequent data processing categorized by inspection scenario (e.g., differentiating inspection results for different pipe diameters and weld locations), each timestamp is additionally marked with two types of key information: first, a pipe diameter identifier, specifying the diameter of the pipe being inspected, such as D=100mm, 200mm, or 300mm, suitable for pipe diameters ranging from Φ100mm to Φ300mm; second, an inspection location identifier, categorized according to the four special points of the fillet weld mentioned, indicating whether the current inspection is at the highest point, lowest point, or transition zone, providing a basis for subsequent location-based defect pattern analysis.

[0119] Based on the timestamp consistency condition established by the above synchronization mechanism, the construction of six-dimensional data points can be further completed. Ultrasonic A-scan signals with timestamp deviations ≤1μs are correlated one-to-one with the real-time pose data of the phased array probe to form structured six-dimensional data points, denoted as... The six core components of this data point each carry key information in different dimensions. The first dimension is the X-axis coordinate of the phased array probe, the second dimension is the Y-axis coordinate, and the third dimension is the Z-axis coordinate. These three dimensions together constitute the three-dimensional spatial position of the phased array probe, which is calculated and output by the SLAM system with an accuracy of ±0.1mm, ensuring that the specific coordinates of the phased array probe on the pipe surface can be accurately located.

[0120] The fourth dimension is the phased array probe's rotation angle around the X-axis, the fifth dimension is the phased array probe's rotation angle around the Y-axis, and the sixth dimension is the phased array probe's rotation angle around the Z-axis. These three dimensions together constitute the phased array probe's three-dimensional attitude. This attitude data is corrected by the angle encoding module of a dual-axis encoder, with the final accuracy controlled within ±0.5°, ensuring that the phased array probe's swing angle, adapted to the pipe curvature, can be accurately recorded. The ultrasonic A-scan signal acquired at the current moment is recorded at a sampling rate of 100MHz and a resolution of 12 bits. It contains key detection information such as the echo amplitude and defect depth of internal defects in the fillet weld, and is the core basis for determining the existence and severity of defects.

[0121] Furthermore, to meet subsequent data verification and traceability needs, such as investigating the source of positioning errors and verifying the reliability of test results, the six-dimensional data points also synchronously store the raw data from the dual-axis encoder. This includes the real-time swing angle of the phased array probe recorded by the angle encoding module, which serves as the original basis for attitude correction. The cumulative movement distance of the phased array probe recorded by the position encoding module serves as the original basis for triggering synchronization pulses. Through this complete link of time synchronization, data association, and raw record storage, it is ultimately possible to record a precise set of position, attitude, and ultrasonic data for every minute movement (0.1 mm) of the phased array probe.

[0122] Specifically, generating the three-dimensional detection result involves filling all the six-dimensional data points into a three-dimensional volume data grid according to their real-time spatial pose.

[0123] Using graphics processor-accelerated volume rendering technology, the three-dimensional volume data mesh is rendered to generate a three-dimensional visualization image that can be viewed and sectioned from any angle.

[0124] 3D volumetric data mesh construction. The scope of the 3D volumetric data mesh needs to be defined based on two core conditions: a pipe diameter of Φ100mm~Φ300mm and the fillet weld area. The specific scope division is as follows: 1. Radial range (X-axis): This range is defined by the X-coordinate of the pipe's center (denoted as...). Based on ), extend to the left to Subtract the pipe radius, then subtract 5mm. Extend to the right. Add the pipe radius R and then add 5mm. That is... The 5mm range is a redundancy range, designed to ensure that the mesh covers the pipe surface and surrounding microstructures, avoiding omission of fillet weld edge areas.

[0125] 2. Circumferential range (Y-axis): This range is defined by the Y-coordinate of the pipe's center (denoted as...). Based on ), extending downwards to Subtract the product of π and the pipe radius R, and extend upwards to... Adding the product of π and the pipe radius R, that is... This range covers exactly half of the pipe circumference and precisely corresponds to the inspection area on the straight pipe side, meeting the requirement of inspecting fillet welds only on the straight pipe side.

[0126] 3. Depth Range (Z-axis): This range is defined by the Z-coordinate of the pipe surface (denoted as...). Based on ), extending downwards to Subtract the fillet weld thickness (denoted as T), then subtract 5mm, extending upwards to... Add 5mm, that is The 5mm redundancy range can cover the heat-affected zone of the fillet weld, ensuring that no test data is missed.

[0127] The resolution of the 3D volumetric data mesh is set to 0.5mm. This resolution ensures the accuracy of defect detection (it can identify tiny defects ≥0.5mm) while avoiding the computational burden caused by an excessive number of meshes, thus achieving a balance between accuracy and computational efficiency.

[0128] The number of voxels is calculated as follows: The radial voxel count is obtained by dividing the total radial length (2R + 10mm) by the mesh resolution of 0.5mm; the circumferential voxel count is obtained by dividing the total circumferential length (2πR) by the mesh resolution of 0.5mm; and the depth voxel count is obtained by dividing the total depth length (fillet weld thickness + 10mm) by the mesh resolution of 0.5mm. Multiplying these three values ​​gives the total voxel count. In the data filling stage, the six-dimensional data points need to be filled into the corresponding voxels of the three-dimensional volumetric data grid according to their spatial location information. Each voxel needs to store two types of information simultaneously: first, the ultrasonic amplitude, which is the maximum value of the ultrasonic A-scan signal in the six-dimensional data point corresponding to that voxel, used to determine whether a defect exists. Second, the detection location, such as butt joints of flat plates, longitudinal welds, and transition zones. This type needs to be determined based on the four special points of the fillet weld (two highest points and two lowest points) and the division of the transition zone, providing a location scenario reference for subsequent defect analysis.

[0129] GPU-accelerated volume rendering and visualization utilizes GPU-accelerated raycasting volume rendering technology to visualize 3D detection results. The specific implementation steps are as follows: 1. Ray Generation: Starting from the origin position of the observation camera (denoted as o), a ray is emitted along the direction of each pixel of the camera. The mathematical equation of the ray is: ;in Let be the unit vector of the ray's direction. The step size parameter is used to traverse the voxels in the 3D volume data mesh and capture the detection information stored in the voxels.

[0130] 2. Voxel Sampling: When the ray passes through the voxels of the 3D volume data grid, the ultrasonic amplitude of each voxel is sampled. To ensure that the sampling accuracy is consistent with the grid resolution, the ray step size t is set to 0.5mm, that is, one voxel is sampled every 0.5mm, to avoid loss of defect information due to excessive step size or computational redundancy due to excessive step size.

[0131] 3. Color and Opacity Calculation: The color is determined based on the ratio of the ultrasonic amplitude to full scale of the voxel. If the ultrasonic amplitude is >80% of full scale, it is identified as a defect area and the color is set to red. If the ultrasonic amplitude is <20% of full scale, it is identified as a base material area and the color is set to blue. If the ultrasonic amplitude is in the transition region between 20% and 80% of full scale, a linear mapping is performed according to the ratio of ultrasonic amplitude to full scale to obtain a transition color, ensuring a clear distinction between the color of the defect and the base material.

[0132] Specifically In the formula, Final pixel color; : No. The color of each sampling point; : No. Opacity of the sampling points; N: Total number of sampling points along the ray; The cumulative product of light transmittance represents the distance the light reaches from the first ray. The proportion of samples that were not absorbed before the sampling point gradually decreases as the penetration depth increases.

[0133] The opacity of a voxel is determined based on the ratio of its stored ultrasonic amplitude to its full scale. For voxels identified as defect areas, their opacity is set to a higher level to ensure that the defects are clearly presented in the 3D visualization image and to avoid the defects being covered up due to insufficient opacity.

[0134] For voxels identified as part of the parent material, their opacity is set to a low level to prevent the image of the parent material from obscuring potential defects and to ensure that defects can be effectively identified. In the transition area between the defect area and the parent material area, the opacity of the voxels changes continuously and linearly with the corresponding ultrasonic amplitude, allowing the image display in the transition area to seamlessly connect the defect area and the parent material area, avoiding image breaks or abrupt visual effects.

[0135] Numerical integration, when using GPU-accelerated raycasting for volume rendering, requires color integration calculations along the direction of ray propagation for all voxels it passes through. First, a ray is emitted from the origin of the viewing camera, moving along the direction of each pixel. The ray then sequentially passes through the voxels in the 3D volumetric data mesh.

[0136] During integration, the color values ​​accumulated by the voxels that the current ray has passed through, as well as the opacity values ​​accumulated by these voxels, are used. The accumulated color values ​​are continuously updated by combining the color and opacity of the current voxel with the above accumulated values ​​according to a specific logic.

[0137] Once the ray has completed its traversal of all relevant voxels, the resulting cumulative color value is the color value of the corresponding pixel in the two-dimensional image. This process can transform the three-dimensional information of fillet weld detection stored in the three-dimensional volume data grid into directly observable two-dimensional image color information.

[0138] The final generated 3D visualization image possesses two core functions: observation from any angle and sectioning from any plane. When observing from any angle, the viewpoint can be adjusted to examine the overall 3D structure of the fillet weld from different directions and angles, as well as the distribution of internal defects, helping inspectors fully understand the shape and defect location of the fillet weld. When sectioning from any plane, the operation achieves high precision, obtaining a cross-sectional image of the fillet weld on a specific plane. This cross-sectional image clearly shows key parameters such as the depth and length of internal defects, enabling precise defect location and detailed analysis, fully meeting the need for intuitive defect localization during inspection.

[0139] Specifically, while controlling the operation of the robotic arm, the motion trajectory is corrected in real time in a closed loop based on the deviation between the real-time pose and the reference scanning path, as well as the feedback from the force sensor.

[0140] During the closed-loop real-time correction process of controlling the robotic arm's operation, real-time pose deviation monitoring requires real-time calculation of the deviation between the actual real-time pose of the phased array probe and the reference real-time pose corresponding to the reference scanning path. The set deviation trigger threshold is far lower than the defect detection accuracy requirements, enabling correction operations to be triggered before significant trajectory deviation occurs, thus preventing trajectory deviation from affecting the detection coverage. Simultaneously, real-time pose deviation monitoring is performed continuously without time delay, effectively preventing missed detections due to continuous deviation accumulation and ensuring the accuracy of the scanning trajectory.

[0141] During the closed-loop real-time correction process of controlling the robotic arm's operation, force feedback monitoring needs to collect in real time the deviation between the actual coupling pressure of the phased array probe and the target coupling pressure, as well as the actual adsorption force of the magnetic wheel automatic crawler. When the deviation between the actual coupling pressure and the target coupling pressure exceeds the set range, trajectory correction is triggered; when the actual adsorption force is lower than the set value, adsorption force compensation is triggered first. Pressure feedback enables real-time monitoring of coupling quality, ensuring that the actual coupling pressure of the phased array probe remains stable within the range required for detection, guaranteeing the effective propagation of ultrasonic signals. Simultaneously, by prioritizing adsorption force compensation, displacement or detachment of the magnetic wheel automatic crawler due to insufficient adsorption force can be avoided, effectively ensuring the safety of the on-site detection process.

[0142] In a specific embodiment, phased array automatic scanning inspection is performed on the fillet weld of a straight pipe seat with a diameter of Φ200mm in a petrochemical plant. The specific implementation process is as follows: 1. Synchronous generation of real-time 3D environment model and phased array probe real-time pose: Based on the environmental perception sensor composed of a 3D laser scanner and a vision sensor, the surface information of the Φ200mm straight pipe is acquired in real time; through synchronous localization and map building SLAM technology, a tightly coupled fusion algorithm is used to perform joint state estimation on laser point cloud data, visual image features and inertial measurement unit data, and synchronously generate a real-time 3D environment model including the fillet weld of the pipe seat, as well as the real-time pose of the phased array probe.

[0143] 2. Extracting the feature trajectory of fillet weld as the reference scanning path: On the surface of the triangular mesh of the generated real-time 3D environment model, the position of the fillet weld is identified by calculating the principal curvature and orientation field of the mesh vertices, and then the 3D spatial ridge line of the fillet weld is extracted and used as the reference scanning path.

[0144] 3. Dynamically plan the robotic arm's motion trajectory and adjust the acoustic beam emission parameters: Based on the extracted baseline scanning path and the real-time pose of the phased array probe, the time elastic band algorithm is used to dynamically plan the robotic arm's motion trajectory. The optimization objective function of this algorithm includes trajectory tracking deviation, distance to obstacles, motion dynamics smoothness, and the coupling state of the phased array probe based on force sensor feedback. Simultaneously, based on the real-time pose of the phased array probe, the local normal vector of the pipe surface is queried from the real-time 3D environment model. With the goal of focusing the acoustic beam on the center of the fillet weld within the pipe workpiece, the incident acoustic beam direction of each element of the phased array in the wedge is calculated based on Snell's law. Then, the emission delay time of each element is calculated based on the incident acoustic beam direction to complete the synchronous adjustment of the phased array acoustic beam emission parameters.

[0145] 4. Control the robotic arm to perform ultrasonic testing: Control the robotic arm to move along the planned motion trajectory. During the operation, according to the adjusted phased array acoustic beam emission parameters, control the phased array probe to perform ultrasonic testing and collect ultrasonic data of the fillet weld.

[0146] 5. Data Fusion to Generate 3D Detection Results: A precise time protocol is used to provide synchronization timestamps for the acquired ultrasonic A-scan signals and the real-time pose data of the phased array probe. The ultrasonic A-scan signals with consistent timestamps are associated with the real-time pose data of the phased array probe to form six-dimensional data points. All six-dimensional data points are filled into a three-dimensional volume data grid according to their real-time spatial pose. The volume rendering technology accelerated by the graphics processor is used to render the three-dimensional volume data grid to generate a three-dimensional visualization image that can be viewed and sectioned from any angle, i.e., the three-dimensional detection result corresponding to the real-time three-dimensional environment model space.

[0147] 6. Closed-loop real-time correction of motion trajectory: While controlling the operation of the robotic arm to perform ultrasonic testing, the motion trajectory of the robotic arm is corrected in a closed loop in real time based on the deviation between the real-time pose of the phased array probe and the reference scanning path, as well as the coupling pressure between the phased array probe and the pipe surface fed back by the force sensor. This ensures that the phased array probe always moves stably along the reference scanning path and maintains a good coupling state.

[0148] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A detection method for automatic phased array scanning of fillet welds in petrochemical plants, characterized in that, Includes the following steps: Based on the real-time pipeline surface information acquired by environmental perception sensors, a real-time 3D environmental model including fillet welds and the real-time pose of the phased array probe are generated simultaneously through synchronous positioning and map building SLAM technology. Based on the real-time three-dimensional environment model, the feature trajectory of the fillet weld is extracted as the benchmark scanning path; Based on the baseline scanning path and the real-time pose, the motion trajectory of the robotic arm is dynamically planned, and the acoustic beam emission parameters of the phased array are adjusted synchronously. The robotic arm is controlled to move along the planned motion trajectory and to perform ultrasonic detection according to the adjusted sound beam emission parameters; The collected ultrasound data is fused with the corresponding real-time pose to generate a three-dimensional detection result corresponding to the real-time three-dimensional environment model space.

2. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 1, characterized in that, The environmental perception sensor includes a 3D laser scanner and a vision sensor; The method of synchronously generating real-time 3D environment model and real-time pose through map construction SLAM technology specifically involves using a tightly coupled fusion algorithm to perform joint state estimation on laser point cloud data, visual image features, and inertial measurement unit data.

3. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 2, characterized in that, The joint state estimation is achieved through factor graph optimization. The state variables of the optimized joint state estimation include the rotation, translation, velocity, gyroscope deviation, and accelerometer deviation of the phased array probe.

4. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 1, characterized in that, The extraction of the feature trajectory of the fillet weld specifically involves identifying and extracting the three-dimensional spatial ridge line of the fillet weld as the feature trajectory by calculating the principal curvature and orientation field of the vertices on the triangular mesh surface of the real-time three-dimensional environment model.

5. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 1, characterized in that, The motion trajectory of the dynamic programming robotic arm is obtained using a time elastic band algorithm. The objective function of the time elastic band algorithm includes trajectory tracking deviation, distance to obstacles, motion dynamics smoothness, and phased array probe coupling state based on force sensor feedback.

6. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 1, characterized in that, The synchronous adjustment of the phased array's acoustic beam emission parameters specifically involves querying the local normal vector of the pipe surface from the real-time three-dimensional environment model based on the real-time pose. With the goal of focusing the sound beam at a predetermined focal point within the workpiece, the direction of the incident sound beam of each element of the phased array in the wedge is calculated by back-calculation based on Snell's law. The emission delay time of each array element is calculated based on the direction of the incident sound beam.

7. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 6, characterized in that, The inverse calculation based on Snell's law satisfies the following vector relationship: ； in, For the first The unit vector of the incident sound beam in each array element within the wedge. The normal vector of the wedge's bottom surface. Let be the unit vector of the refracted sound beam in the workpiece pointing towards the focal point. Let be the local normal vector. It is the ratio of the sound velocity of the wedge to the sound velocity of the workpiece.

8. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 1, characterized in that, The process of fusing the acquired ultrasound data with the corresponding real-time pose specifically involves using a precise time protocol to provide a synchronization timestamp for the ultrasound A-scan signal and the real-time pose data of the phased array probe. The ultrasound A-scan signal with consistent timestamps is correlated with the real-time pose data of the phased array probe to form a six-dimensional data point.

9. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 8, characterized in that, The generation of the three-dimensional detection result specifically involves filling all the six-dimensional data points into a three-dimensional volume data grid according to their real-time spatial pose. Using graphics processor-accelerated volume rendering technology, the three-dimensional volume data mesh is rendered to generate a three-dimensional visualization image that can be viewed and sectioned from any angle.

10. The detection method for automatic phased array scanning of fillet welds in petrochemical plants according to claim 1, characterized in that, While controlling the robotic arm to operate, the motion trajectory is corrected in real time in a closed loop based on the deviation between the real-time pose and the reference scanning path, as well as the feedback from the force sensor.

Images