Automatic light beam centering method and system based on industrial CT

Abstract

The invention relates to the technical field of industrial CT, discloses an automatic light beam centering method and system based on industrial CT, and aims at solving the technical defects that in the welding process, due to thermal deformation, the centering precision of a light beam and a welding seam is low, and the reliability is poor. The method comprises the steps that before welding, three-dimensional CT volume data of a workpiece to be welded are obtained to generate a nominal welding seam trajectory; establishing a thermal-mechanical coupling finite element analysis model, and simulating a welding process to generate a predicted thermal teratogenic field map; superimposing the predicted thermal teratogenesis to a nominal trajectory to generate a dynamic welding trajectory; in the welding process, a laser contour measurement sensor is driven to collect three-dimensional contour data of a welding seam area in real time and extract the actual welding seam center position; and establishing a Kalman filtering state observer. By the adoption of the technical scheme, continuous and high-precision tracking of the center position of the welding seam can be achieved, the influence of thermal teratogenesis is overcome, and the welding quality and the yield of complex structural parts are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-destructive testing, and in particular relates to an automatic light beam centering method and system based on industrial CT. Background Art

[0002] The field of non-destructive testing technology includes multiple branches such as industrial computed tomography, ultrasonic testing, X-ray testing, and magnetic particle testing. The core content of this technology is to use physical or chemical methods to inspect and test the internal and surface structures of materials, parts, and equipment without damaging or affecting the performance of the object being tested. Industrial computed tomography technology refers to the use of X-rays to penetrate objects and be received by detectors, and then reconstruct images through computers to obtain the core means of obtaining the three-dimensional structural information of the object's internal structure. With the improvement of precision manufacturing and quality control requirements, industrial CT technology has been widely used in aerospace, automotive industry, materials science and other fields. The development of this technology has not only promoted the accuracy of product quality inspection, but also brought profound changes in aspects such as reverse engineering and failure analysis.

[0003] Among them, the automatic beam centering method refers to the use of automated algorithms and control systems to achieve precise alignment between the X-ray source, the rotation center of the workpiece being measured, and the detector during the industrial CT scanning process. The theme proposes an automated solution to the time-consuming and unstable alignment process in pre-scan preparation. This method calculates the offset between the workpiece rotation center and the ideal scanning center by analyzing the characteristics of the initial projection data or preview image, and drives the multi-axis motion platform for compensation correction. In this method, the system collects projection data from a small number of angles and uses image processing algorithms to identify the workpiece contour or specific marks to calculate the precise center position. This method aims to replace traditional manual visual alignment or calibration processes that rely on standard samples to improve detection efficiency and image reconstruction quality.

[0004] Existing beam alignment technologies often rely on manual operator adjustments or semi-automatic calibration based on simple geometric models. These methods lack the ability to adapt to complex workpiece shapes and cannot guarantee high-precision alignment consistency. For example, when dealing with asymmetric workpieces or those with complex internal structures, edge detection-based algorithms often suffer from centering errors due to feature extraction errors. The alignment process is often a static, single-shot calibration, unable to monitor and compensate for centering deviations caused by thermal drift or mechanical vibration during the scanning process, resulting in motion artifacts and blur in the reconstructed image. Alignment algorithms inefficiently utilize projection data, typically analyzing only a single or a few frames, failing to fully integrate all projection information to achieve a global optimal solution, resulting in insufficient alignment accuracy. Existing methods lack a direct feedback mechanism between alignment accuracy and image quality, making it impossible to quantify the impact of alignment errors on the final inspection results. These issues are particularly prominent in high-resolution inspection scenarios such as precision welding and additive manufacturing, directly impacting defect detection rates and dimensional measurement accuracy. Summary of the Invention

[0005] The purpose of the present invention is to overcome the technical defects in the prior art, such as low precision and poor reliability of beam-weld alignment caused by thermal deformation during the welding process, thereby providing an automatic beam alignment method and system based on industrial CT.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for automatic beam centering based on industrial CT, comprising the following steps:

[0007] S1: Acquire three-dimensional industrial computed tomography (CT) volume data of the workpiece to be welded before welding, perform image segmentation processing on the three-dimensional industrial CT volume data to extract a three-dimensional point cloud model of the weld area, and calculate and generate a nominal weld trajectory line representing the weld geometry based on the three-dimensional point cloud model;

[0008] S2: establishing a thermomechanical coupling finite element analysis model that matches the material properties of the workpiece to be welded and preset welding process parameters, inputting the nominal weld trajectory line as the heat source movement path into the thermomechanical coupling finite element analysis model, performing a numerical simulation of the welding process, thereby calculating the four-dimensional time-varying displacement field of each spatial node on the workpiece caused by the heat input during the welding process, and generating a predicted thermally induced distortion field map;

[0009] S3: superimposing the displacement vectors corresponding to the points on the nominal weld trajectory line in the predicted thermal distortion field map in the time series to the spatial coordinates of the corresponding points on the nominal weld trajectory line to generate a dynamic welding trajectory that has been compensated for thermal distortion;

[0010] S4: During the welding process, a laser profile measurement sensor having a fixed spatial position relationship with the welding beam is driven to continuously collect real-time three-dimensional profile data of the weld area in front of the welding beam action point at a preset sampling frequency, and the current actual weld center position is extracted from the real-time three-dimensional profile data;

[0011] S5: Establish a Kalman filter state observer, use the weld center position indicated by the dynamic welding trajectory at the current moment as the predicted value of the system state, use the current actual weld center position extracted by the laser profile measurement sensor as the observed value of the system state, perform optimal state estimation through the Kalman filter state observer, calculate the deviation vector between the predicted position and the actual position, and generate a real-time position correction control signal for the welding beam actuator based on the deviation vector.

[0012] As a further solution of the present invention, generating the nominal weld trajectory line in step S1 specifically includes:

[0013] S111: Processing the three-dimensional industrial CT volume data using a threshold-based region growing algorithm, setting a lower limit threshold of the voxel grayscale value to the grayscale value corresponding to the workpiece material density and an upper limit threshold to the grayscale value corresponding to the air density, identifying and segmenting the voxels in the weld gap and groove area to form independent weld geometry data;

[0014] S112: Applying a three-dimensional surface extraction algorithm, such as a Marching Cubes algorithm, to the weld geometry data to generate a triangular mesh model representing the inner and outer surfaces of the weld, namely, the three-dimensional point cloud model;

[0015] S113: Slicing the three-dimensional point cloud model along a predetermined welding direction to obtain a series of two-dimensional weld cross-sectional profiles, applying a geometric center calculation or a bottom V-shaped cusp recognition algorithm to each two-dimensional weld cross-sectional profile to determine the coordinates of the center point of the cross section;

[0016] S114: Connecting the center point coordinates of the series of two-dimensional weld cross-sectional profiles in three-dimensional space, and fitting them using a B-spline curve or a cubic spline interpolation algorithm to generate a smooth, continuous three-dimensional space curve, which is the nominal weld trajectory line.

[0017] As a further solution of the present invention, generating the predicted thermal distortion field map in step S2 specifically includes:

[0018] S211: Constructing a geometric model of the thermal-mechanical coupling finite element analysis model, wherein the geometric model accurately reproduces the geometrical shape of the workpiece to be welded and the weld; setting material property parameters of the model according to the material specification of the workpiece, wherein the material property parameters include thermal conductivity, specific heat capacity, density, elastic modulus, Poisson's ratio, and thermal expansion coefficient that vary with temperature;

[0019] S212: The heat source effect of the welding beam is equivalent to a three-dimensional Gaussian distribution or double ellipsoid volume heat source model, and its mathematical expression is:

[0020]

[0021] Where P is the beam power, η is the thermal efficiency, a, b, c are the geometric size parameters of the heat source model, is the energy distribution coefficient of the front and rear ellipsoids, and the parameters are set according to the preset welding process;

[0022] S213: discretizing the nominal weld trajectory into a series of path points. In the simulation, driving the volume heat source model to move along the path points at a preset welding speed; solving the unsteady-state heat conduction equation in each time step;

[0023]

[0024] Calculate the three-dimensional temperature field distribution of the entire workpiece model;

[0025] S214: The three-dimensional temperature field calculated at each time step is used as a load and input into the structural mechanics analysis solver to solve the elastic-plastic constitutive relationship equation based on the thermal strain increment. ,in is the stress increment, is the elastic-plastic matrix, is the total strain increment, is the thermal strain increment, thereby calculating the displacement vectors of all nodes on the workpiece caused by the temperature field, and collecting the node displacement vectors of all time steps to form the predicted thermal distortion field map.

[0026] As a further solution of the present invention, generating a dynamic welding trajectory in step S3 specifically includes:

[0027] S311: The nominal weld trajectory line Expressed as an arc length is a function of its parameters, ;

[0028] S312: From the predicted thermal distortion field map, obtain the corresponding The predicted 3D displacement vector of the position , , where t(l) is the time when the welding beam reaches the arc length l;

[0029] S313: performing vector addition on the displacement vector and the coordinates of the corresponding point on the nominal weld trajectory to obtain the dynamic welding trajectory , and its calculation formula is ;

[0030] S314: Discretize the dynamic welding trajectory into a series of six-degree-of-freedom path points containing position and posture information, whose timestamps correspond to the welding speed, to form an instruction sequence that can be directly executed by the welding robot controller.

[0031] As a further solution of the present invention, extracting the current actual weld center position in step S4 specifically includes:

[0032] S411: The laser profile measurement sensor consists of a line laser emitter and a high-speed industrial camera. The line laser emitter emits a fan-shaped laser beam with a central wavelength of 658 nanometers and a power of 50 milliwatts, which is irradiated on the weld area to form a laser line reflecting the cross-sectional profile of the weld. The high-speed industrial camera is equipped with a narrow-band filter with a central wavelength of 658 nanometers and a bandwidth of 10 nanometers, and captures images of the laser line at a frame rate of 200 Hz.

[0033] S412: Preprocess each frame of the collected image, including image denoising and binarization, and then use the centroid method or Steger algorithm to extract the center pixel coordinates of the laser line to generate a two-dimensional point set ,in and is the pixel coordinate;

[0034] S413: According to the pre-calibrated camera intrinsic parameter matrix and laser plane equation, the two-dimensional point set is converted from the pixel coordinate system to the three-dimensional space coordinate system under the welding robot base coordinate system to obtain a real-time three-dimensional point cloud of the weld cross section ;

[0035] S414: Perform feature recognition on the real-time 3D point cloud, and determine the actual weld center position at the current moment by searching for the lowest point in the point cloud or calculating the intersection of the two slope lines of the V-groove. .

[0036] As a further solution of the present invention, the calculation of the deviation vector by the Kalman filter state observer in step S5 specifically includes:

[0037] S511: Define the state vector of the system ,in and are the position deviations of the welding beam center in the horizontal and vertical directions, and is the rate of change of the deviation;

[0038] S512: Establish a prediction model for the system state, and its state transition equation is ,in is the prior state estimate at time k, is the posterior state estimate at time k-1, F is the state transfer matrix, which is in the form of [[1,0,Δt,0],[0,1,0,Δt],[0,0,1,0],[0,0,0,1]], Δt is the sampling time interval, is the process noise, and its covariance matrix Q represents the uncertainty of the thermal distortion model prediction;

[0039] S513: Establish the observation model of the system, and its observation equation is ,in is the observation value at time k, that is, the deviation between the actual weld center position measured by the laser profile measurement sensor and the weld center position indicated by the dynamic welding trajectory at that time; H is the observation matrix, which is in the form of [[1,0,0,0],[0,1,0,0]], To observe the noise, its covariance matrix R is calibrated according to the measurement accuracy of the laser profilometry sensor;

[0040] S514: Execute the Kalman filter update step, first calculate the Kalman gain ,in is the prior error covariance matrix; then use the observed values Update the state estimate to get the posterior state estimate ;Finally update the error covariance matrix ;

[0041] S515: From the optimal estimated posterior state Extract the deviation component As the deviation vector, the deviation vector is transmitted to the underlying controller of the welding beam actuator to generate a position correction signal superimposed on the main motion instruction.

[0042] An automatic beam centering system based on industrial CT, used to perform any of the aforementioned methods, comprising:

[0043] a CT data processing module configured to acquire three-dimensional industrial CT volume data of the workpiece to be welded, generate a three-dimensional point cloud model of the weld area through image segmentation and three-dimensional reconstruction, and calculate the nominal weld trajectory line based on the point cloud model;

[0044] The welding process simulation module is configured to construct a thermomechanical coupling finite element analysis model that matches the workpiece, simulate and calculate the four-dimensional time-varying displacement field of the workpiece caused by the heat input based on preset welding process parameters and using the nominal weld trajectory as the heat source path, and generate a predicted thermally induced distortion field map;

[0045] a distortion compensation trajectory generation module configured to spatially and temporally superimpose the displacement vector corresponding to the predicted thermal distortion field map with the nominal weld trajectory to generate a dynamic welding trajectory that has been compensated for thermal distortion;

[0046] A real-time geometric sensing module, comprising a laser profile measurement sensor fixedly connected to the welding beam, configured to collect real-time weld cross-sectional profile data in front of the beam while welding is in progress and extract the current actual weld center position from it;

[0047] The state estimation and control module integrates a Kalman filter state observer and is configured to receive the predicted position indicated by the dynamic welding trajectory and the actual position measured by the real-time geometric sensing module, fuse the information of the two through state estimation, accurately calculate the deviation vector between the two, and output the deviation vector as a real-time position correction control signal for the welding beam actuator.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are:

[0049] This invention uses industrial CT data to establish a high-precision weld seam geometry model before welding. Combined with thermomechanical coupled finite element analysis, it pre-calculates the complex, nonlinear workpiece distortion caused by heat input during welding, thereby generating a dynamic welding trajectory with forward-looking distortion compensation. This transforms the control problem from a purely lagging feedback correction to a predictive control based on a physical model. During the welding execution phase, the invention does not rely on traditional visual sensors, which are susceptible to interference from welding arc light and smoke, but instead employs interference-resistant laser profiling technology to obtain real-time, accurate weld seam geometry information. Core to this, the invention establishes a Kalman filter state observer that optimally fuses predictions based on the physical model with observations based on real-time sensing. This not only fully utilizes the global, forward-looking information provided by offline simulation, but also provides real-time correction for model uncertainty and random perturbations through online measurement. This closed-loop control framework, combining prediction and correction, enables the beam centering system to overcome the significant effects of thermally induced distortion during welding and achieve continuous, high-precision tracking of the weld seam center position, significantly improving the quality and yield of welding complex structural components. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1A schematic diagram of the overall process of a method for automatic centering of a light beam provided by an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of the functional modules of an automatic light beam centering system provided by an embodiment of the present invention;

[0052] Figure 3 Schematic diagram of data flow in the process of generating dynamic welding trajectory in an embodiment of the present invention;

[0053] Figure 4 Schematic diagram of a real-time closed-loop control process based on Kalman filtering in an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of an application scenario of a method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the specific embodiments described herein are only used to explain the present invention, rather than to limit the scope of protection of the present invention. In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0056] In a specific embodiment, referring to Figure 1 The present invention provides an automatic beam centering method based on industrial CT, comprising the following steps:

[0057] Step S1: Acquire three-dimensional industrial computed tomography (ICT) volume data of the workpiece to be welded before welding, perform image segmentation processing on the three-dimensional industrial CT volume data to extract a three-dimensional point cloud model of the weld area, and calculate and generate a nominal weld trajectory line representing the weld geometry based on the three-dimensional point cloud model.

[0058] Step S2: Establish a thermomechanical coupling finite element analysis model that matches the material properties of the workpiece to be welded and the preset welding process parameters, input the nominal weld trajectory line as the heat source movement path into the thermomechanical coupling finite element analysis model, and perform a numerical simulation of the welding process, thereby calculating the four-dimensional time-varying displacement field of each spatial node on the workpiece caused by the heat input during the welding process, and generating a predicted thermally induced distortion field map.

[0059] Step S3, superimpose the displacement vectors corresponding to the points on the nominal weld trajectory in the predicted thermal distortion field map in the time series to the spatial coordinates of the corresponding points on the nominal weld trajectory to generate a dynamic welding trajectory that has been compensated for thermal distortion. The data flow of this process is shown in the attached figure. Figure 3 shown.

[0060] Step S4: During the welding process, a laser profile measurement sensor having a fixed spatial position relationship with the welding beam is driven to continuously collect real-time three-dimensional profile data of the weld area in front of the welding beam action point at a preset sampling frequency, and the current actual weld center position is extracted from the real-time three-dimensional profile data.

[0061] Step S5: Establish a Kalman filter state observer, use the weld center position indicated by the dynamic welding trajectory at the current moment as the predicted value of the system state, use the current actual weld center position extracted by the laser profile measurement sensor as the observed value of the system state, perform optimal state estimation through the Kalman filter state observer, calculate the deviation vector between the predicted position and the actual position, and generate a real-time position correction control signal for the welding beam actuator based on the deviation vector. The real-time closed-loop control process is shown in the attached figure. Figure 4 shown.

[0062] Furthermore, each of the above steps is elaborated in detail.

[0063] The process of acquiring the three-dimensional industrial CT volume data of the workpiece to be welded and generating the nominal weld trajectory line in step S1 can be specifically decomposed into the following sub-steps.

[0064] Step S111 processes the three-dimensional industrial CT volume data using a threshold-based region growing algorithm. In a specific application scenario, the workpiece to be welded is a V-groove joint formed by butting two 12 mm thick GH4169 high-temperature alloy plates. The industrial CT scanning system uses a 450 kV microfocus X-ray source and a flat-panel detector. During scanning, the tube voltage is set to 420 kV and the tube current is 1.2 mA. 2,400 projection views are acquired through 360-degree rotation. The reconstructed three-dimensional volume data has a voxel size of 0.08 mm x 0.08 mm x 0.08 mm. The volume data is represented as a three-dimensional array, with each element representing a 16-bit grayscale value. The lower and upper voxel grayscale thresholds are determined based on the significant difference in X-ray attenuation coefficient between the workpiece base material and the weld gap (typically filled with air or shielding gas), which is reflected as a difference in grayscale value. Specifically, the upper threshold is determined by selecting multiple pure air regions in the volume data, statistically analyzing the distribution of their voxel grayscale values, and taking the mean plus three times the standard deviation as the result. For example, the calculated grayscale value is 850. The lower threshold is determined by selecting multiple sample regions within the dense matrix material area away from the weld, statistically analyzing the distribution of their voxel grayscale values, and taking the mean minus three times the standard deviation as the result. For example, the calculated grayscale value is 51200. Therefore, voxels with grayscale values ​​in the range of [850, 51200] are identified as belonging to the weld gap and groove area. The region growing algorithm is executed as follows: First, a seed point is manually or automatically selected through image moment analysis on any 2D slice image containing the weld. This seed point must be located within the weld gap. The algorithm then uses this seed point as a starting point and checks its 26 adjacent voxels in three-dimensional space. If the grayscale value of an adjacent voxel falls within the preset range of [850, 51200], it is added to the target region and used as a new seed point. This process is iterated until no new voxels meet the criteria and can be added. Ultimately, the collection of all voxels included in the target region constitutes the independent weld geometry data.

[0065] As an alternative, image segmentation can also utilize a deep learning-based semantic segmentation network, such as a 3D U-Net. This approach requires pre-constructing a labeled dataset containing CT volume data of various typical workpieces and weld regions accurately segmented manually by professionals as labels. A 3D U-Net model is trained on this dataset to learn the mapping from raw CT voxel grayscale to "weld" or "non-weld" characteristics. In practice, the CT volume data of the workpiece to be welded is input into the trained 3D U-Net model, which directly outputs a probability map of the same size as the raw volume data. The value of each voxel in the map represents the probability of belonging to the weld region. By setting a probability threshold (for example, 0.95), voxels with a probability above the threshold are identified as weld geometry data. This approach is more robust when dealing with complex weld morphologies or in the presence of artifacts.

[0066] Step S112, applying a three-dimensional surface extraction algorithm to the weld geometry data to generate a three-dimensional point cloud model. Specifically, a marching cubes algorithm is used. The algorithm processes cubic units consisting of 8 adjacent voxels one by one. For each cubic unit, 2^8=256 combinations can be formed depending on whether its 8 vertices are located inside the segmented weld geometry data. Each combination corresponds to a preset triangular facet topology structure. By querying a pre-calculated lookup table, it is determined to generate one or more triangular facets inside the cubic unit to approximate the isosurface. After traversing all cubic units, the vertex set of all generated triangular facets constitutes a three-dimensional point cloud model representing the inner and outer surfaces of the weld. The point cloud density of the model is determined by the original CT voxel resolution. In one embodiment, the generated triangular mesh model contains approximately 2.5 million vertices and 5 million triangular facets.

[0067] Step S113, slicing the three-dimensional point cloud model along a predetermined welding direction. The predetermined welding direction is usually defined as the main direction of the weld length extension, which can be specified by the operator or automatically determined by performing principal component analysis (PCA) on the point cloud model, where the first principal component direction is the welding direction. Set a slice spacing, for example 0.5 mm, and generate a series of parallel slice planes along this direction. Each slice plane intersects with the three-dimensional point cloud model to form a point set of a two-dimensional weld cross-sectional profile. For each two-dimensional cross-sectional profile, a geometric center calculation algorithm is applied to determine the coordinates of its center point. Specifically, for a set of N points The closed contour of the composition, its geometric center The calculation formula is:

[0068]

[0069] As an alternative, particularly for V- or U-shaped grooves, a bottom V-shaped cusp identification algorithm can be used. This algorithm first performs piecewise linear fitting on the 2D cross-sectional contour point set to identify the two main straight line segments that form the groove sidewall. The intersection of these two straight line segments is then calculated and used as the center point of the cross section. This method is more suitable for scenarios where welding energy must be precisely applied to the weld root because it directly locates the physical root position. During the processing, contour irregularities may be encountered due to CT noise or inaccurate segmentation. In this case, the RANSAC (Random Sample Consensus) algorithm can be applied to the 2D contour point set to perform straight line or curve fitting to eliminate outliers and enhance the stability of feature point positioning.

[0070] Step S114: Connect and fit the center point coordinates of the series of two-dimensional weld cross-sectional profiles in three-dimensional space. Through step S113, an ordered three-dimensional point sequence is obtained;

[0071] , where each point is the center point of the corresponding section. In order to generate a smooth, continuous and differentiable trajectory line, the cubic spline interpolation algorithm is used to fit the point sequence. This algorithm ensures that the generated curve not only passes through all data points, but is also continuous in the first and second order derivatives, which is crucial to ensure the smoothness of the welding robot movement. The final three-dimensional space curve is the nominal weld trajectory line, which is expressed in the parameterized form , where l is the arc length along the curve.

[0072] The process of establishing the thermomechanical coupling finite element model and generating the predicted thermal distortion field map in step S2 is shown in the attached figure. Figure 2 As shown in the figure, the functions of the welding process simulation module can be broken down into the following sub-steps.

[0073] In step S211, a geometric model of the thermal-mechanical coupling finite element analysis model is constructed and material properties are set. The geometric model directly adopts the three-dimensional model of the workpiece base portion segmented in step S1. This model has accurately reproduced the macro-geometric morphology of the workpiece to be welded and the weld groove. The geometric model is meshed using unstructured tetrahedral elements. In order to balance calculation accuracy and efficiency, a non-uniform mesh strategy is adopted. In the estimated welding heat-affected zone (HAZ) and fusion zone, that is, the cylindrical area with a radius of 5 mm along the nominal weld trajectory line, the unit feature size is set to 0.3 mm; in the base portion away from the welding area, the unit size gradually transitions to 3 mm. The final generated finite element mesh contains approximately 2.1x10^6 tetrahedral elements and 4.5x10^5 nodes.

[0074] The temperature-dependent material property parameters in the model were set based on the material specification of the workpiece material GH4169 and a publicly available academic literature database. The accuracy of these parameters directly affects the reliability of the simulation results. In a specific embodiment, the temperature-dependent thermophysical and mechanical property parameter data used are shown in the following table:

[0075] Temperature (℃) Thermal conductivity (W / (m·K)) Specific heat capacity (J / (kg·K)) Density (kg / m³) Elastic modulus (GPa) Poisson's ratio <![CDATA[Coefficient of thermal expansion (10⁻ 6 / K)]]> Yield strength (MPa) 20 11.4 435 8240 205 0.294 12.8 1100 200 13.2 485 8190 198 0.301 13.5 1050 400 15.1 540 8130 189 0.309 14.2 1020 600 17.3 590 8060 175 0.318 15.1 980 800 20.5 650 7980 140 0.325 16.0 450 1000 24.1 710 7900 80 0.330 16.8 200 1200 28.0 760 7820 30 0.335 17.5 50 1400 32.5 810 7750 5 0.340 18.2 10

[0076] In the finite element solver, linear interpolation is used between these data points to calculate the material property values ​​at any temperature.

[0077] Step S212: Mathematically model the heat source effect of the welding beam. The Goldak double ellipsoid volume heat source model is used because it can asymmetrically describe the temperature gradient before and after the welding pool, which is more consistent with the physical reality of high-speed welding. Its mathematical expression is:

[0078] For the front half of the ellipsoid (x ≥ 0):

[0079]

[0080] For the posterior semi-ellipsoid (x<0):

[0081]

[0082] Where P is the beam power, η is the thermal efficiency, , , b, c are the geometric size parameters of the heat source model, and f_r are the energy distribution coefficients of the front and rear ellipsoids and + =2. These parameters are set according to the preset welding process. For example, for a laser wire welding process, the parameters are set as follows:

[0083] Beam power P = 4000 watts.

[0084] Welding speed v = 15 mm / s.

[0085] Heat input efficiency η = 0.70. This value is calibrated through calorimetric experiments or selected based on empirical data from similar processes.

[0086] Length of the front half ellipsoid =2.5 mm.

[0087] Length of the rear half ellipsoid =5.0 mm.

[0088] The semi-axis width of the weld is b = 2.0 mm.

[0089] The semi-axis of penetration depth c = 3.0 mm.

[0090] Energy distribution coefficient =0.6, =1.4.

[0091] These geometric parameters are calibrated by conducting test welding on a test plate of the same material and thickness as the actual workpiece, then making a metallographic cross-section of the weld, measuring the actual size of the molten pool, and then adjusting the model parameters through reverse optimization so that the error between the simulated molten pool morphology and the actual metallographic morphology (such as the error in melt width and melt depth) is less than 5%.

[0092] Step S213, perform transient thermal analysis. First, the nominal weld trajectory generated in step S114 is Discretize into a series of path points. The discrete step length Δl is determined by the welding speed v and the solution time step Δt, that is, Δl = v * Δt. To ensure numerical convergence and accuracy, the time step Δt is set to 0.02 seconds, and the path point spacing is 0.3 mm. During the simulation, the double ellipsoid volume heat source model is driven to move along the discrete path points at a welding speed of 15 mm / s. Within each time step, the unsteady heat conduction equation is solved;

[0093]

[0094] in, , , are the density, specific heat capacity and thermal conductivity that vary with temperature, is a double ellipsoid heat source term. The boundary conditions are set to allow convection and radiation between all exterior surfaces of the workpiece and the environment. The ambient temperature is set to 25°C, the convection heat transfer coefficient h is set to 20 W / (m²·K), and the surface emissivity ε is set to 0.8. By solving the entire welding process and the subsequent cooling process (for example, cooling to 500 seconds after the start of welding), the three-dimensional temperature field distribution T(x, y, z, t) is calculated for all nodes in the workpiece model at each time.

[0095] Step S214, perform sequential coupled structural mechanics analysis. The transient temperature field T(x, y, z, t) calculated in step S213 is used as a thermal load and applied to the structural model at each time step. Before the mechanical analysis, the mechanical boundary conditions need to be set. According to the actual clamping situation, three fixed constraint points are set at the bottom of the workpiece to limit its rigid body displacement and rotation. In each incremental step, the elastic-plastic constitutive relationship equation based on the thermal strain increment is solved. .in is the stress increment, is the elastic-plastic stiffness matrix considering the temperature effect, is the total strain increment. Thermal strain increment according to Calculate, where is the temperature-dependent thermal expansion coefficient, is the temperature increment of the current step. The constitutive model uses a combination of the vonMises yield criterion and the kinematic hardening model to more accurately describe the mechanical behavior of the material during cyclic heating and cooling. By solving the entire time series, the three-dimensional displacement vectors of all finite element nodes on the workpiece at each moment caused by the temperature field change are finally calculated. The set of displacement vectors of all nodes at all time steps constitutes the predicted thermal distortion field map. This map is a high-dimensional dataset whose data structure is an associative array or hash table, with the key being the node ID and timestamp (node_id, t) and the value being the corresponding three-dimensional displacement vector [Ux, Uy, Uz].

[0096] As a more computationally efficient alternative, the inherent strain method can be used. This method first performs a detailed thermomechanical coupling analysis on a representative two-dimensional weld cross section to calculate the residual plastic strain distribution within the cross section, that is, the inherent strain. This two-dimensional inherent strain distribution is then mapped along the nominal weld trajectory line of the entire three-dimensional workpiece to form an initial strain field. Finally, only one linear elastic static structural analysis is performed on the entire workpiece model, and the initial strain field is applied as a load to quickly calculate the final welding deformation. This method greatly reduces the calculation time, but sacrifices the accurate description of the dynamic deformation process during welding. It is suitable for scenarios where the prediction of the final deformation state is high but the dynamic accuracy of the process is not high.

[0097] The data flow of the process of generating the dynamic welding trajectory after thermal distortion compensation in step S3 is as shown in the attached figure. Figure 3 As shown, it can be broken down into the following sub-steps.

[0098] Step S311: the nominal weld trajectory The smooth curve generated in step S114 is stored as a set of densely sampled three-dimensional coordinate points, with each point P_i being accompanied by a cumulative arc length value from the starting point. For example, if a weld with a length of 300 mm is discretized with an accuracy of 0.1 mm, a lookup table containing 3001 (x, y, z, l) quadruplets will be generated. The coordinates corresponding to the arc length l at any intermediate position are , which can be obtained by performing linear interpolation or higher-order interpolation between two adjacent points in the lookup table, thereby realizing the function =[x(l),y(l),z(l)].

[0099] Step S312: from the predicted thermal distortion field map, obtain the value corresponding to the arc length l when welding is performed. The predicted three-dimensional displacement vector of the position. This query process includes matching and interpolation in both time and space dimensions. First, time matching is performed. The time t(l) corresponding to the welding beam traveling to the arc length l is given by;

[0100]

[0101] Calculation shows that v is the preset constant welding speed. For example, when the welding speed v is 15 mm / s, the time t(90) corresponding to the arc length l=90 mm is 6.0 seconds. Secondly, spatial interpolation is performed. Since the point The coordinates of usually do not coincide with a node of the finite element model, and the displacement of the point needs to be interpolated based on the displacement values ​​of the surrounding nodes. Specifically, first locate the point in the finite element mesh. The positioning process can be accelerated by constructing a spatial index structure such as an octree. After finding the tetrahedral unit, the displacement vectors U1, U2, U3, U4 of its four vertices (set as N1, N2, N3, N4) at time t(l) can be directly obtained from the predicted thermal distortion field map. The displacement vector It can be calculated by barycentric coordinate interpolation. The coordinates of the center of gravity are (w1,w2,w3,w4), then its displacement vector is As an alternative spatial interpolation scheme, the inverse distance weighted method (IDW) can be used. All k nodes within the radius R, according to the nodes to The inverse of the distance is used as the weight to perform weighted average of the displacement vector. This process is performed on each discrete point on the nominal weld trajectory. Repeat the process to obtain a displacement vector sequence corresponding to each .

[0102] Step S313, the displacement vector is added to the coordinates of the corresponding point on the nominal weld trajectory to obtain the dynamic welding trajectory. This calculation is performed independently for each discrete point on the trajectory. The calculation formula is: Taking a specific point as an example, assuming that the arc length l = 90 mm, the nominal trajectory point coordinates (90) is [150.2, 55.8, 12.0] (unit: mm). Through the query and interpolation in step S312, the predicted thermal displacement vector of the point at t=6.0 seconds is obtained. (90,6.0) is [0.0,-0.45,0.25]. Then the coordinates of the corresponding points on the dynamic welding trajectory after compensation are (90) is [150.2, 55.8, 12.0] + [0.0, -0.45, 0.25] = [150.2, 55.35, 12.25]. This calculation reveals that at this moment, the actual position of the weld is offset by -0.45 mm in the y direction (lateral direction) and +0.25 mm in the z direction (height) compared to its initial position. Performing this vector superposition operation on all points on the nominal trajectory generates a complete set of dynamic welding trajectory points.

[0103] Step S314: discretize the dynamic welding trajectory into a six-degree-of-freedom path point sequence containing position and posture information. This sequence is an instruction set that can be directly executed by the welding robot controller. For each dynamic trajectory point , its position information [x, y, z] has been determined. Its posture information, that is, the direction of the welding beam axis, needs additional calculation. In a typical embodiment, the beam axis should always be perpendicular to the workpiece surface and point to the center of the weld. Therefore, the Z-axis direction of the tool coordinate system (usually the beam direction) can be determined by the tangent direction of the dynamic trajectory at that point and the normal direction of the weld cross section. Specifically, the dynamic trajectory at point The tangent vector of It can be calculated by the central difference method:

[0104]

[0105] This tangent vector defines one degree of freedom of the robot tool posture (e.g., rotation around the Y axis). Another posture vector, the X axis direction (lateral direction) of the tool coordinate system, can be defined as perpendicular to and lies within the local horizontal plane. By orthogonalizing these vectors, a complete six-degree-of-freedom pose (usually expressed as Euler angles or quaternions) can be uniquely determined. The resulting instruction sequence is a list, each element of which contains: the path point number, the three-dimensional coordinates [x, y, z], the three-dimensional pose [Rx, Ry, Rz], the linear velocity to reach that point, and the corresponding timestamp. This instruction sequence file is formatted in a text format compatible with the specific robot controller for easy loading and execution.

[0106] The process of collecting three-dimensional contour data in real time and extracting the actual weld center position during the welding process in step S4 can be specifically decomposed into the following sub-steps.

[0107] Step S411: Configure and use a laser profilometry sensor. The sensor is physically attached to the welding head via a precision-machined rigid bracket, ensuring a constant relative position and orientation relative to the welding beam focal point. In one specific embodiment, the sensor lens optical axis is parallel to the welding beam axis, and the measurement point is located 20 mm in front of the welding beam's impact point. This "lead distance" provides the necessary system response time. The laser emitter emits a fan-shaped laser beam with a central wavelength of 658 nanometers and a power of 50 milliwatts. The 658 nanometer wavelength was chosen because it falls within the visible red light range, facilitating debugging and observation. It also differs significantly from the emission spectrum of common welding plasma or reflected lasers, facilitating signal filtering. The high-speed industrial camera is equipped with a narrowband interference filter with a central wavelength precisely matching the laser wavelength (658 nanometers) and a bandwidth (FWHM) of 10 nanometers. This extremely narrow bandwidth effectively suppresses the broad spectrum of background noise generated during welding, such as intense arc light, spatter, and thermal radiation, allowing only the reflected laser light to pass through, significantly improving the image's signal-to-noise ratio. The camera continuously captures laser line images at a frame rate of 200 Hz. This sampling frequency is sufficient to capture the rapid dynamic changes in weld position caused by thermal deformation, clamping stress release, etc., and meet the bandwidth requirements of real-time closed-loop control.

[0108] Step S412, pre-process each frame of the collected image and extract the center of the laser line. Each frame of the original image first goes through the pre-processing process. In the image denoising stage, a 5x5 Gaussian filter is used to perform a convolution operation on the image to smooth the random noise caused by the dark current or readout noise of the sensor, while retaining the edge features of the laser line. Subsequently, binarization is performed, but it is not a simple global threshold binarization. Instead, an adaptive threshold method is used to address the problem of uneven brightness of the laser line under different substrate surface reflectivities. More preferably, binarization is not performed, and the center is directly extracted on the grayscale image. Specifically, the Steger algorithm is used to extract the sub-pixel center coordinates of the laser line. The algorithm first calculates the Hessian matrix of each pixel in the image and solves its eigenvalues ​​and eigenvectors. The center point of the laser line is defined as the point with the maximum second-order directional derivative in the direction perpendicular to the line. By solving a quadratic equation, the precise sub-pixel positions of these points can be analytically calculated. Compared to the centroid method which can only provide pixel-level accuracy, the Steger algorithm can achieve an accuracy of 0.1 pixel or even higher, which is crucial for the accuracy of subsequent 3D reconstruction. After this step, each frame of the image is converted into a 2D sub-pixel coordinate point set. , where i ranges from 1 to N, and N is usually hundreds of points, which together constitute the projection of the laser line on the target surface of the image sensor.

[0109] Step S413, convert the two-dimensional pixel coordinate point set into three-dimensional space coordinates. This conversion relies on a pre-completed sensor calibration process. The calibration process includes intrinsic parameter calibration and extrinsic parameter (hand-eye) calibration. The intrinsic parameter calibration uses a plane calibration plate such as a chessboard, and calculates the camera's intrinsic parameter matrix K (including focal length) through the Zhang Zhengyou calibration method. , and main point , ) and distortion coefficients. External parameter calibration determines the equation of the laser plane in the camera coordinate system For each extracted two-dimensional sub-pixel point ( , ), first use the inverse of the intrinsic parameter matrix K and the distortion model to reversely project it into a three-dimensional ray in the camera coordinate system. The direction vector of this ray is;

[0110]

[0111] Then, calculate the intersection of the ray and the calibrated laser plane, and the three-dimensional coordinates of the intersection This is the actual position of the point in the camera coordinate system. Finally, the pre-calibrated homogeneous transformation matrix from the camera coordinate system to the welding robot base coordinate system is used. , all three-dimensional points in the camera coordinate system Transform to the robot base coordinate system to obtain the final real-time 3D point cloud of the weld cross section The entire conversion process is completed within each sampling period (5 milliseconds).

[0112] Step S414: perform feature recognition on the real-time 3D point cloud to determine the actual weld center position. For a typical V-groove, a robust straight line fitting algorithm based on RANSAC is used. First, the real-time 3D point cloud is acquired. The point is projected onto a two-dimensional plane perpendicular to the welding direction. Two RANSAC linear fits are then performed on this two-dimensional point set. The first fit identifies the point set that forms the inclined wall on one side of the groove and obtains its line equation, L_1. After removing the inliers belonging to L_1 from the point set, a second RANSAC linear fit is performed on the remaining point set to obtain the line equation, L_2, for the inclined wall on the other side. The intersection of these two lines, L_1 and L_2, is precisely determined as the current weld center (root) position, Z_k. The advantage of this method is that the RANSAC algorithm is effectively immune to interference from local contour outliers caused by surface oxide scale, temporary tack welds, or small amounts of spatter, providing a very stable and reliable weld center location result. For U-shaped or rectangular grooves, the feature recognition algorithm is adjusted accordingly to fit a curve or search for a central axis of symmetry. The final output, Z_k, is a three-dimensional coordinate vector in the robot's base coordinate system.

[0113] The process of calculating the deviation vector by the Kalman filter state observer in step S5 is as shown in the attached figure. Figure 4 As shown, it can be broken down into the following sub-steps.

[0114] Step S511, define the state vector of the system. is defined as .in, and These represent the positional deviation of the welding beam center perpendicular to the weld seam (transverse direction) and along the beam axis (height or gap). These two directions constitute the most critical planes for centering control. and Including the velocity component in the state vector allows the filter to not only estimate the current position deviation but also predict the deviation trend at the next moment, thus achieving smoother and more forward-looking control.

[0115] Step S512: Establish a prediction model for the system state. This model describes how the system state evolves naturally over time. Its state transition equation is: . is the prior state estimate at time k, that is, the optimal estimate based on the previous moment Predict the current state. The state transition matrix F is constructed based on Newton's laws of motion, assuming that the rate of change of the deviation is approximately constant within a sampling period as short as Δt (for example, 5 milliseconds). Therefore, the specific form of the F matrix is ​​[[1,0,Δt,0],[0,1,0,Δt],[0,0,1,0],[0,0,0,1]]. This matrix indicates that the current position deviation is equal to the position deviation at the previous moment plus the product of velocity and time, and the current velocity is the same as the previous velocity. is the process noise term, which represents the uncertainty of the state model itself, mainly coming from the thermal distortion finite element model ( ) and unmodeled dynamic disturbances (such as small changes in the workpiece clamping force). The process noise is assumed to be zero-mean Gaussian white noise, and its covariance matrix Q is a 4x4 matrix that requires careful tuning. The size of the diagonal elements of Q reflects the degree of confidence in the model prediction deviation and its rate of change. For example, if the simulation model's prediction accuracy in the lateral direction is lower than that in the height direction, the Q matrix will be and The corresponding diagonal element value should be set larger, indicating that the model has greater uncertainty in that direction.

[0116] Step S513: Establish an observation model for the system. This model associates the system states that cannot be directly observed with the observable quantities that can be directly measured. Its observation equation is: The observation value z_k is the actual weld center position measured by the laser profile measurement sensor at time k. (from step S414) and the weld center position indicated by the dynamic welding trajectory at that moment The difference vector between .because and are all three-dimensional coordinates, is also a three-dimensional vector, but we only care about its components in the horizontal and height directions, so Is a two-dimensional vector [Δx, Δy]^T. The role of the observation matrix H is to transform the four-dimensional state vector Extract the component that directly corresponds to the observed quantity. Under this definition, the form of H is [[1,0,0,0],[0,1,0,0]], which directly converts the position deviation component in the state vector and Extract it. is the observation noise term, which represents the uncertainty in the measurement process. It mainly comes from the inherent measurement error of the laser profilometry sensor, the accuracy limitation of the image processing algorithm, and the calibration error of the coordinate transformation. The observation noise is also assumed to be zero-mean Gaussian white noise, and its covariance matrix R is a 2x2 diagonal matrix. The diagonal elements of R are and The R matrix is ​​a quantitative measure of the sensor's measurement variance in the lateral and height directions. These values ​​can be determined offline by aligning the sensor with a static V-groove standard block of known geometry and performing repeated measurements over a long period of time (e.g., thousands of times). The variance of the measurement results in the lateral and height directions is then statistically analyzed to calibrate the R matrix.

[0117] Step S514: perform a recursive update step of the Kalman filter. This step is the core of information fusion and is executed cyclically in each sampling period.

[0118] First, predict the error covariance of the next state: This formula converts the posterior error covariance of the previous moment into By forward propagating the state transfer model and adding the influence of process noise Q, the prior error covariance at the current moment is obtained .

[0119] Next, calculate the Kalman gain: Kalman gain is a 4x2 matrix, which is essentially a dynamically calculated weight. It weighs the uncertainty of the prior state estimate (given by and the uncertainty of the new observation (represented by R). If the sensor is very accurate (R is small), will become larger, making the filter more trustworthy. On the contrary, if the physical model is very reliable ( Very small), will become smaller, making the filter more confident in its predictions.

[0120] Then, the state estimate is updated using the observations: This is the filter calibration step. It is called "innovation" or "measurement residual", which represents the difference between the actual measurement value and the model prediction value. This difference is multiplied by the Kalman gain After that, it is used to correct the prior estimate , thus obtaining the optimal posterior state estimate at the current moment .

[0121] Finally, update the error covariance matrix This step calculates the uncertainty of the system state estimate after integrating the new observation information. The updated covariance P_k is usually smaller than the prior covariance , which quantitatively reflects the accuracy improvement brought by information fusion. The next iteration will be .

[0122] Step S515, extract the deviation vector from the optimal estimated a posteriori state and generate a control signal. After that, directly extract its first two components, namely , as the most reliable weld deviation vector. This deviation vector is transmitted to the underlying motion controller of the welding robot. An independent, high-speed closed-loop position control algorithm, such as a proportional-integral-derivative (PID) controller, is run inside the controller. The PID controller uses the deviation vector As input, the goal is to drive this deviation to zero. The controller outputs a correction velocity or displacement signal, which is superimposed in real time with the main motion command from P_dynamic being executed by the robot's main controller. For example, if the deviation vector is [-0.1, 0.05] (mm), the PID controller generates a small corrective motion in the direction of [+0.1, -0.05], driving the welding beam precisely back to the actual weld center. This predictive-corrective control framework based on the Kalman filter ensures smooth, fast, and precise response of the entire system.

[0123] Reference Figure 2 and Figure 5 The present invention also provides an industrial CT-based automatic beam alignment system for performing the aforementioned method. This system is integrated into an automated welding workstation, which includes a six-axis industrial robot, a laser welding head, a wire feeder, and corresponding workpiece fixtures. The system specifically includes the following functional modules:

[0124] A CT data processing module, typically deployed on a high-performance workstation, receives raw 3D volume data files (e.g., in DICOM format) from industrial CT equipment and performs all operations described in steps S111 to S114. Specifically, it integrates threshold-based region growing or 3D U-Net segmentation algorithms, marching cubes algorithms, multi-section analysis, and B-spline interpolation algorithms. The module's final output is a nominal weld trajectory file representing the original weld geometry.

[0125] A welding process simulation module is also deployed on a high-performance computing server or cluster. The core of this module is a commercial or open-source finite element analysis software solver. This module is configured to receive the workpiece geometry model and nominal weld trajectory generated by the CT data processing module and, based on the user-entered material property library and welding process parameters (described in steps S211 and S212), perform the coupled thermal-mechanical simulation in steps S213 and S214. The module's output is a large data file containing the displacement vectors of all nodes at all time steps, i.e., the predicted thermally induced distortion field map.

[0126] A distortion-compensated trajectory generation module can be deployed on the same computer as the previous two modules. This module is configured to receive the nominal weld trajectory and the predicted thermally induced distortion field map as inputs and perform the calculations in steps S311 to S314. Through spatiotemporal interpolation, vector superposition, and robot pose calculation, it ultimately generates a distortion-compensated dynamic welding trajectory program that can be directly downloaded to the robot controller.

[0127] A real-time geometric sensing module is physically mounted on the robot's end-effector. The core of this module is a laser profiling sensor (consisting of a line laser emitter and a high-speed camera) and its accompanying image acquisition card. During the welding process, the module is configured to trigger and capture contour images of the weld front at a frequency of 200 Hz based on synchronization signals from the robot controller. It then executes the image processing and 3D reconstruction algorithms in steps S411 to S414. The output is the current actual weld center position Z_k at each sampling time k, expressed in the robot's base coordinate system.

[0128] A state estimation and control module, which usually runs on the robot controller or an embedded industrial PC that communicates with it at high speed. The core of this module is the Kalman filter state observer described in step S5. It is configured to receive the predicted position from the distortion compensation trajectory generation module at each sampling period. and the measured position Z_k from the real-time geometric sensing module. By executing the filtering algorithm from steps S511 to S515, the module calculates the optimal deviation vector and outputs it as a real-time correction signal to the robot's underlying servo controller, thus closing the entire high-precision alignment control loop.

[0129] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A beam automatic centering method based on industrial CT, characterized in that: The following steps are involved: S1. Acquire 3D industrial CT volume data of a workpiece to be welded, perform image segmentation and 3D reconstruction on the 3D industrial CT volume data, extract a 3D point cloud model of the weld area, and generate a nominal weld trajectory representing the weld geometry based on the 3D point cloud model; S2. Establishing a thermomechanical coupling finite element analysis model that matches the material properties and welding process parameters of the workpiece to be welded, performing a numerical simulation of the welding process using the nominal weld trajectory as the heat source movement path, and generating a predicted thermally induced distortion field map; S3, superimposing the displacement vectors corresponding to the nominal weld trajectory line in the time series in the predicted thermal distortion field map onto the nominal weld trajectory line to generate a dynamic welding trajectory that has been compensated for thermal distortion; S4. During the welding process, a laser profile measurement sensor fixedly connected to the welding beam is used to collect real-time three-dimensional profile data of the weld area in front of the welding beam action point, and the current actual weld center position is extracted from the real-time three-dimensional profile data.

2. The automatic beam centering method based on industrial CT according to claim 1 is characterized in that: The three-dimensional point cloud model includes a vertex coordinate set and a triangular mesh topology structure; the predicted thermal distortion field map includes a finite element model node identifier, a time step, and a corresponding three-dimensional displacement vector; the dynamic welding trajectory includes a six-degree-of-freedom path point sequence sorted by timestamp and containing position and posture information; the real-time three-dimensional contour data includes a set of three-dimensional space point coordinates representing the weld cross-section.

3. The automatic beam centering method based on industrial CT according to claim 1, characterized in that: The steps for generating the nominal weld trajectory are specifically as follows: S111. Processing the three-dimensional industrial CT volume data using a threshold-based region growing algorithm or a deep learning semantic segmentation network to identify and segment the weld gap and groove area to form independent weld geometry data; S112, applying a three-dimensional surface extraction algorithm to the weld geometry data to generate the three-dimensional point cloud model representing the inner and outer surfaces of the weld; S113, slicing the three-dimensional point cloud model along a predetermined welding direction to obtain a series of two-dimensional weld cross-sectional profiles, and determining the center point coordinates of each cross section using a geometric center calculation or a bottom V-shaped cusp recognition algorithm; S114 , connecting a series of center point coordinates in a three-dimensional space, and fitting them using a B-spline curve or a cubic spline interpolation algorithm to generate the nominal weld trajectory line.

4. The automatic beam centering method based on industrial CT according to claim 3 is characterized in that: The steps for generating the predicted thermal distortion field map are specifically as follows: S211, constructing a finite element geometric model that reproduces the geometric shape of the workpiece to be welded, and setting temperature-dependent material property parameters such as thermal conductivity, specific heat capacity, density, elastic modulus, Poisson's ratio, and thermal expansion coefficient; S212, equating the heat source effect of the welding beam to a three-dimensional Gaussian distribution or a double ellipsoid volume heat source model, and setting its power, thermal efficiency, and geometric size parameters according to a preset welding process; S213, discretizing the nominal weld trajectory, driving the volume heat source model to move along the nominal weld trajectory, solving the unsteady heat conduction equation in each time step, and calculating the three-dimensional temperature field distribution of the entire workpiece model; S214. Using the three-dimensional temperature field distribution as a thermal load, solving the elastic-plastic constitutive relationship equation based on the thermal strain increment, calculating the displacement vectors of all nodes on the workpiece at each time step, and forming the predicted thermally induced distortion field map.

5. The method for automatic beam centering based on industrial CT according to claim 4, characterized in that: The steps for generating the dynamic welding trajectory are specifically as follows: S311, expressing the nominal weld trajectory as a function with arc length as a parameter; S312. Querying and obtaining, from the predicted thermal distortion field map, a predicted three-dimensional displacement vector corresponding to a nominal weld trajectory position when welding reaches a specific arc length; S313, performing vector addition on the predicted three-dimensional displacement vector and the coordinates of the corresponding point on the nominal weld trajectory to obtain the dynamic welding trajectory; S314. Discretize the dynamic welding trajectory into a series of six-degree-of-freedom path points containing position and posture information to form an instruction sequence for execution by the welding robot controller.

6. The method for automatic beam centering based on industrial CT according to claim 5, characterized in that: The steps for extracting the current actual weld center position are specifically as follows: S411, collecting a laser line image reflecting the cross-sectional profile of the weld by using a line laser transmitter and a high-speed industrial camera; S412, performing denoising and binarization processing on the collected image, and extracting the center pixel coordinates of the laser line using the centroid method or Steger algorithm to generate a two-dimensional point set; S413, converting the two-dimensional point set from a pixel coordinate system to a three-dimensional space coordinate system according to a pre-calibrated camera intrinsic parameter matrix and a laser plane equation to obtain the real-time three-dimensional contour data; S414: Search the real-time three-dimensional contour data for the intersection of the V-groove slope lines or the lowest point of the contour to determine the current actual weld center position.

7. The method for automatic beam centering based on industrial CT according to claim 1, characterized in that: The method further comprises step S5: S5. Establish a Kalman filter state observer, use the weld center position indicated by the dynamic welding trajectory as a system state prediction value, use the actual weld center position as a system state observation value, perform optimal state estimation, calculate the deviation vector between the predicted position and the actual position, and generate a real-time position correction control signal for the welding beam actuator; The deviation vector includes a lateral position deviation and a longitudinal position deviation.

8. The method for automatic beam centering based on industrial CT according to claim 7, characterized in that: The steps for generating the real-time position correction control signal are specifically as follows: S511. Define a system state vector including position deviation and deviation change rate; S512, establishing a system state prediction model based on the state transfer matrix to perform prior state estimation; S513, establishing a system observation model based on an observation matrix, and taking the difference between the dynamic welding trajectory indication position and the actual weld center position as an observation value; S514, executing a Kalman filter update step, calculating a Kalman gain, and updating a state estimate in combination with the observation value to obtain a posterior state estimate; S515 . Extract the position deviation component from the posterior state estimation as the deviation vector, and generate the real-time position correction control signal.

9. An automatic beam centering system based on industrial CT, characterized in that: The system is used to implement the automatic beam centering method based on industrial CT according to any one of claims 1 to 8, and the system includes: The nominal trajectory generation module obtains the 3D industrial CT volume data of the workpiece to be welded, generates a 3D point cloud model through image segmentation and 3D reconstruction, and performs slicing and center point fitting along the weld direction based on the model to calculate and generate the nominal weld trajectory line; A thermal distortion simulation module constructs a thermal-mechanical coupling finite element model based on the geometry and material properties of the workpiece, simulates the welding process using the nominal weld trajectory as the heat source path, solves the unsteady-state heat conduction and elastic-plastic mechanics equations, and generates a predicted thermal distortion field map; A dynamic trajectory compensation module, based on the predicted thermal distortion field map, extracts the displacement vectors corresponding to the time and space of each point on the nominal weld trajectory line, performs distortion compensation through vector superposition operation, and generates a dynamic welding trajectory; The real-time contour sensing module drives the laser contour measurement sensor during the welding process to collect real-time 3D contour data in front of the weld and extracts the current actual weld center position through image processing and coordinate transformation; The state estimation and correction module establishes a Kalman filter state observer, fuses the predicted position given by the dynamic welding trajectory with the actual position measured by the real-time contour sensing module, performs optimal state estimation, calculates the deviation vector and generates a real-time position correction control signal for the welding beam actuator.

Images