Control method and system for automatically switching soldering iron of multi-size welding spots

Abstract

The invention relates to the technical field of welding, and discloses a self-adaptive welding control method and system for multi-size welding spots, and the method comprises the steps: collecting an original image set of a welding target through a preset angle and an illumination condition, and synchronously obtaining environment interference data; carrying out feature extraction by adopting a deep learning model after image preprocessing, and identifying a target size parameter; calculating target position data based on a three-dimensional space coordinate system, and generating an initial motion path in combination with a mechanical constraint condition; the path reliability is verified through virtual simulation, and a dynamic compensation algorithm is adopted to eliminate positioning deviation and optimize path parameters; fusing the environmental interference data to generate a final execution path; and the multi-specification soldering bits are driven to be intelligently switched to complete precise welding. According to the method, through multi-mode sensing and self-adaptive path planning, the welding positioning precision and the process adaptability under the complex working condition are improved, and the automatic welding efficiency and the quality stability are improved.

Description

Technical Field

[0001] The present invention relates to the field of welding technology, and in particular to a control method and system for automatically switching soldering irons for soldering points of multiple sizes. Background Art

[0002] In modern industrial control systems, welding tasks under complex working conditions place higher demands on system intelligence. As manufacturing evolves towards flexibility and customization, welding objects exhibit diverse geometric shapes and dynamically changing spatial distributions, posing significant challenges to the real-time perception and dynamic response capabilities of industrial control systems. Especially in scenarios involving multi-sized welds, the system must accurately identify the target contour, rapidly analyze dimensional parameters, and adjust the operation path online within milliseconds. This places multi-dimensional demands on the architecture of traditional industrial control systems. Because welding environments are often plagued by interference factors such as uneven lighting, mechanical vibration, and thermal deformation, industrial control systems must be able to integrate and process multi-source heterogeneous data to achieve robust target feature extraction. Furthermore, to meet the demands of high-precision welding processes, the system must maintain millimeter-level positioning accuracy under dynamic conditions, placing even higher standards on the deep integration of sensor collaborative control, real-time feedback mechanisms, and adaptive algorithms. Developing an intelligent industrial control system that can accommodate welds of varying sizes and adapt to environmental interference has become a key breakthrough in improving the efficiency of welding automation.

[0003] In one existing technique, welding multi-sized solder joints is typically accomplished using a fixed, programmed mechanical positioning method. The specific implementation process is as follows: First, the system uses a single vision sensor to capture two-dimensional image data of the workpiece from a fixed perspective, based on a preset solder joint size template. The sensor uses a basic grayscale threshold segmentation algorithm to extract the rough outline of the target area and map the solder joint position coordinates to a two-dimensional plane coordinate system. The system then performs a simple comparison of the coordinate data with a pre-stored library of standard solder joint sizes. If a match is successful, the corresponding fixed welding parameters (such as soldering iron temperature and dwell time) and a preset linear path trajectory are applied. The welding robot arm moves along a linear path according to a predetermined program, completing the solder joint operation at a constant speed. If the workpiece position or size deviates from the preset template, manual intervention is required to adjust the sensor position or recalibrate the coordinate system, and the path offset is corrected by manually entering parameters.

[0004] In existing technologies, grayscale segmentation is sensitive to lighting, leading to contour deviations; two-dimensional mapping cannot accurately reflect the dynamic three-dimensional workpiece position; fixed parameters are difficult to adapt to dimensional differences or surface features, requiring manual calibration; and open-loop control lacks real-time feedback and cannot correct for vibration or temperature disturbances. These issues prevent the system from achieving improved welding accuracy under complex working conditions. Summary of the Invention

[0005] The present invention provides a control method and system for automatically switching soldering irons for solder joints of multiple sizes, thereby improving soldering accuracy, process adaptability, and automation efficiency under complex working conditions.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a control method for automatically switching soldering irons with multiple-sized solder joints, comprising: Collecting a set of original images of the welding target and obtaining environmental interference data under a preset angle and preset lighting conditions; Preprocessing the original image set to obtain a standard image set; Based on the standard image set, a pre-trained deep learning model is used to extract features and determine target size parameters; Calculating the relative position of the target in space based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data; Determining a preliminary operation path based on the target location data and the acquired path constraints; Performing a simulation according to the preliminary operation path and obtaining a position offset; when the position offset exceeds a preset offset threshold, adjusting path parameters to determine an optimized operation path; Determining a final execution path based on the optimized operation path and in combination with the environmental interference data; According to the final execution path, the intelligent switching of the execution tool is driven to perform welding operations on the welding target.

[0007] In an optional implementation, preprocessing the original image set to obtain a standard image set includes: According to the original image set, dynamically denoising is performed by using a three-layer decomposition of the Symlets-8 wavelet basis and a Bayesian shrinkage threshold algorithm to output a denoised image set; According to the denoised image set, a limited contrast adaptive histogram equalization algorithm is used to locally enhance the contrast of the denoised image set, and a Reinhard color migration algorithm is used to unify the color gamut of the denoised image set under different lighting conditions to generate a standard image set.

[0008] In an optional embodiment, the method of extracting features and determining target size parameters using a pre-trained deep learning model based on the standard image set includes: According to the standard image set, a pre-trained deep learning model is used to extract contour features, output a mask map containing pixel-level contour probabilities, and obtain a contour boundary map after binarization; When there are broken contours in the contour boundary map, a morphological dilation operation is used to strengthen edge continuity, and then a contrast-limited adaptive histogram equalization algorithm is used to locally enhance the contrast to generate an optimized enhanced image set; Based on the enhanced image set, a histogram of oriented gradients algorithm is used to extract features, and the extracted features are input into a pre-trained support vector machine classifier to output a final image set containing geometric labels; According to the final image set, the Harris corner detection algorithm is used to extract the contour key points, and the cubic spline interpolation algorithm is used to perform smoothing to obtain a key point coordinate set; According to the set of key point coordinates, the least squares ellipse fitting algorithm is used to perform parameter estimation and the polygon convex hull algorithm is used to correct the parameter boundaries to determine the target size parameters.

[0009] In an optional embodiment, calculating the relative position of the target in space based on the target size parameter in combination with a preset three-dimensional space coordinate system to obtain target position data includes: According to the target size parameters, combined with a preset three-dimensional space coordinate system, a Perspective-n-Point algorithm is used to perform spatial mapping to obtain the initial position of the target in space; If the deviation between the initial pose and the actual space coordinates exceeds the preset error threshold, the Levenberg-Marquardt nonlinear optimization algorithm is called to output the optimized pose by minimizing the reprojection error of the 2D-3D point pair. According to the optimized posture, the posture is finely calibrated through the ICP point cloud registration algorithm, and the target position data is output.

[0010] In an optional implementation, determining a preliminary operation path based on the target location data and the acquired path constraint conditions includes: Obtaining the joint motion range constraints of the welding robot arm, combining the target position data, and generating an initial collision-free path through the RRT-Connect path planning algorithm; The initial collision-free path is smoothed using a cubic B-spline interpolation algorithm to generate an adjustment path that satisfies the kinematic constraints of the manipulator, thereby obtaining a preliminary operation path.

[0011] In an optional embodiment, performing a simulation according to the preliminary operation path and obtaining a position offset, and when the position offset exceeds a preset offset threshold, adjusting path parameters to determine an optimized operation path, includes: According to the preliminary operation path and the position offset, a model predictive control algorithm is used to dynamically adjust the path parameters to generate a preliminary corrected path; Perform simulation according to the preliminary corrected path and obtain the actual motion trajectory; Performing path matching between the preliminary corrected path and the actual motion trajectory using a dynamic time warping algorithm to obtain a path matching error; When the path matching error is higher than a preset matching error threshold, a nonlinear least squares optimization algorithm is called to perform frame-by-frame fine-tuning on the preliminary correction path to generate an optimized operation path.

[0012] In an optional implementation, determining a final execution path based on the optimized operation path and in combination with the environmental interference data includes: Inputting the optimized operation path and the environmental interference data into an adaptive model predictive control algorithm, dynamically adjusting the path parameters through rolling horizon optimization, and generating an interference-resistant execution path; When the environmental interference data exceeds a preset interference threshold, an online Gaussian process regression algorithm is called to predict the trajectory and generate an obstacle avoidance path offset; The anti-interference execution path and the obstacle avoidance path offset are superimposed and fused to generate a final execution path.

[0013] In a second aspect, the present invention provides a control system for automatically switching soldering irons for solder joints of multiple sizes, comprising: A data acquisition module is used to collect a set of original images of the welding target and obtain environmental interference data under a preset angle and preset lighting conditions; A data processing module, configured to preprocess the original image set to obtain a standard image set; An object size analysis module is used to extract features and determine object size parameters based on the standard image set using a pre-trained deep learning model; A target position determination module is used to calculate the relative position of the target in space based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data; A preliminary path determination module, configured to determine a preliminary operation path based on the target location data and the acquired path constraint conditions; an optimized path determination module, configured to perform a simulation operation according to the preliminary operation path and obtain a position offset; and when the position offset exceeds a preset offset threshold, adjust path parameters to determine an optimized operation path; A final path determination module, configured to determine a final execution path based on the optimized operation path and the environmental interference data; The execution module is used to drive the intelligent switching of the execution tool according to the final execution path to perform welding operations on the welding target.

[0014] In a third aspect, the present invention also provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the control method for automatically switching soldering irons for multi-sized solder joints as described in any one of the above is implemented.

[0015] In a fourth aspect, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned methods for automatically switching soldering irons for multi-sized solder joints.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) By combining the Symlets-8 wavelet basis and the Bayesian shrinkage threshold dynamic denoising algorithm with the Reinhard color migration technology, the problems of contour breakage and color gamut deviation caused by the sensitivity of traditional grayscale threshold segmentation to illumination are solved, and the consistency of image preprocessing under multiple illumination conditions is improved. (2) Based on Perspective-n-Point spatial mapping and Levenberg-Marquardt nonlinear optimization algorithm, integrated with ICP point cloud registration technology, it breaks through the limitation that two-dimensional coordinate mapping cannot reflect three-dimensional dynamic workpieces, achieves millimeter-level spatial positioning accuracy, and eliminates welding path deviation caused by workpiece posture offset. (3) The RRT-Connect path planning algorithm and dynamic time warping matching technology are used in combination with the model predictive control algorithm to correct the robot arm motion trajectory in real time, overcoming the collision risk and kinematic constraint conflict problems of the traditional fixed path under complex working conditions, and improving the adaptive adjustment efficiency of the welding path. (4) By integrating environmental interference data with the online Gaussian process regression algorithm, the positioning drift problem caused by thermal deformation and mechanical vibration in open-loop control is solved through the superposition optimization of the anti-interference execution path and the obstacle avoidance offset, and the welding stability under dynamic interference is achieved, ensuring the high-quality forming of multi-sized welds. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 1 is a flow chart of a control method for automatically switching soldering irons for solder joints of multiple sizes provided by the first embodiment of the present invention; Figure 2 1 is a schematic diagram of the control system structure of the automatic switching soldering iron for multi-size solder joints provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] Reference Figure 1 The first embodiment of the present invention provides a control method for automatically switching soldering irons with multiple-sized solder joints, comprising the following steps: S11, collecting an original image set of the welding target and obtaining environmental interference data at a preset angle and under preset lighting conditions; S12, preprocessing the original image set to obtain a standard image set; S13, extracting features using a pre-trained deep learning model based on the standard image set and determining target size parameters; S14, calculating the relative position of the target in space based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data; S15, determining a preliminary operation path based on the target location data and the acquired path constraint conditions; S16, performing a simulation operation according to the preliminary operation path and obtaining a position offset. When the position offset exceeds a preset offset threshold, adjusting the path parameters to determine an optimized operation path. S17, determining a final execution path based on the optimized operation path and the environmental interference data; S18, driving the intelligent switching of execution tools according to the final execution path, and performing welding operations on the welding target.

[0020] In step S11 , an original image set of the welding target is collected and environmental interference data is obtained under a preset angle and preset lighting conditions.

[0021] Specifically, the camera is mounted at the end of the robotic arm at a preset angle (e.g., 30-degree pitch and 45-degree horizontal yaw), ensuring a field of view covering the core area of ​​the solder joint. The light source utilizes a multi-band LED array, providing uniform illumination according to preset parameters (e.g., 800 lumens brightness, 5000K color temperature, and 60-degree angle of incidence) to avoid reflections or shadows. Image acquisition is triggered by the robotic arm's motion controller. When the robotic arm reaches the target coordinates, the camera continuously captures images at a rate of 30 frames per second, generating a raw image set consisting of multiple frames, each with a timestamp and robotic arm pose data. Environmental interference data is collected in real time using distributed sensors: triaxial accelerometers installed at the joints of the robotic arm record vibration spectrum data; infrared thermal imagers monitor temperature distribution in the solder joint area; and electromagnetic sensors capture electromagnetic interference intensity from surrounding equipment. All data is transmitted via a high-speed bus and strictly synchronized with the image acquisition system's timestamps to form a structured dataset.

[0022] The preset angles are determined based on the 3D point cloud reconstruction of the welds, selecting the optimal viewing angle combination that covers at least 90% of the feature points. Lighting parameters are determined through experimental optimization: light source parameters are adjusted for different materials, and image outline clarity is used as an evaluation metric. The parameter combination that maximizes the gradient amplitude is ultimately selected. Preset thresholds for vibration and electromagnetic interference are determined through offline data analysis. After collecting sample data from the welding process, statistical methods are used to eliminate outliers, and the mean value is used as the threshold reference. The output of this step includes a set of original images, synchronized interference data, and calibration parameters, providing highly consistent multimodal input for subsequent denoising, feature extraction, and path planning, ensuring the repeatability of the system under dynamic interference.

[0023] In step S12, the original image set is preprocessed to obtain a standard image set.

[0024] In a specific embodiment, preprocessing the original image set to obtain a standard image set includes: According to the original image set, dynamically denoising is performed by using a three-layer decomposition of the Symlets-8 wavelet basis and a Bayesian shrinkage threshold algorithm to output a denoised image set; According to the denoised image set, a limited contrast adaptive histogram equalization algorithm is used to locally enhance the contrast of the denoised image set, and a Reinhard color migration algorithm is used to unify the color gamut of the denoised image set under different lighting conditions to generate a standard image set.

[0025] Specifically, the original image set is used as input data and is first processed through a three-layer decomposition of the Symlets-8 wavelet basis. This decomposition process decomposes each image into low-frequency subbands and high-frequency subbands, where the high-frequency subbands contain image details and noise components. For the high-frequency subbands, a Bayesian shrinkage threshold algorithm is used for dynamic denoising: based on the statistical characteristics of the wavelet coefficients, the shrinkage threshold of each layer is automatically calculated, and the coefficients exceeding the shrinkage threshold are shrunk to retain the valid signal while suppressing the noise. The denoised subbands are recombined through the wavelet reconstruction algorithm to output a denoised image set. This step effectively eliminates the random noise caused by electromagnetic interference and mechanical vibration in the welding environment, while avoiding the edge blurring problem caused by traditional filtering methods.

[0026] The denoised image collection is further subjected to local contrast enhancement using a contrast-constrained adaptive histogram equalization algorithm. This algorithm divides each image into several local regions, calculates a histogram within each region, and performs equalization. The enhancement amplitude is controlled using a preset contrast limiting parameter to prevent local over-enhancement. This contrast limiting parameter is determined by analyzing the typical grayscale distribution range of the welding target, ensuring that the texture characteristics of welds of different materials are appropriately enhanced.

[0027] Finally, the image color gamut is unified using the Reinhard color migration algorithm. This algorithm uses a reference image captured under standard lighting conditions as the target color gamut. It calculates the statistical differences in color space between the denoised image and the reference image, including the mean and covariance matrix. The color distribution of the denoised image is then mapped to the color gamut of the reference image. The reference image is selected from typical solder joint images captured under standard lighting conditions during welding experiments, whose color gamut characteristics represent ideal conditions. This process eliminates color deviations caused by varying lighting conditions and ensures the stability of subsequent feature extraction.

[0028] The standard image set output from the preprocessing process features sufficient noise suppression, optimized local contrast, and a unified color gamut. This provides high-quality input data for subsequent deep learning models and improves the accuracy of feature extraction. Each algorithm parameter is optimized based on experimental data, eliminating the need for manual intervention and ensuring consistency and repeatability in the processing.

[0029] In step S13, based on the standard image set, a pre-trained deep learning model is used to perform feature extraction and determine target size parameters.

[0030] In a specific embodiment, the method of extracting features and determining target size parameters using a pre-trained deep learning model based on the standard image set includes: According to the standard image set, a pre-trained deep learning model is used to extract contour features, output a mask map containing pixel-level contour probabilities, and obtain a contour boundary map after binarization; When there are broken contours in the contour boundary map, a morphological dilation operation is used to strengthen edge continuity, and then a contrast-limited adaptive histogram equalization algorithm is used to locally enhance the contrast to generate an optimized enhanced image set; Based on the enhanced image set, a histogram of oriented gradients algorithm is used to extract features, and the extracted features are input into a pre-trained support vector machine classifier to output a final image set containing geometric labels; According to the final image set, the Harris corner detection algorithm is used to extract the contour key points, and the cubic spline interpolation algorithm is used to perform smoothing to obtain a key point coordinate set; According to the set of key point coordinates, the least squares ellipse fitting algorithm is used to perform parameter estimation and the polygon convex hull algorithm is used to correct the parameter boundaries to determine the target size parameters.

[0031] Specifically, a standard image set is used as input data and first fed into a pre-trained U-Net deep learning model. This model employs an encoder-decoder architecture. The encoder consists of five downsampling stages, each of which extracts features using two 3×3 convolutional layers and performs downsampling via 2×2 max pooling. The decoder then corresponds to five upsampling stages, each of which performs upsampling via 2×2 transposed convolutions and is concatenated with the feature map from the corresponding encoder layer. The final layer of the model uses a 1×1 convolution and a sigmoid activation function to output a mask of the same size as the input image, where each pixel value represents the probability that the location belongs to the outline of a solder joint.

[0032] The output probability map is binarized using a fixed processing threshold to generate a binary contour boundary map. This processing threshold is determined by analyzing the precision and recall curves corresponding to different processing thresholds on the validation set. The optimal processing threshold that maximizes the product of the two is selected. In the binarized image, a pixel value of 1 represents a contour point, and 0 represents background. When a contour break is detected, a morphological dilation operation is performed using a 3×3 square structuring element. The number of dilations is adaptively determined based on the degree of breakage. Specifically, the ratio of the length of the broken contour area to the perimeter of the intact contour is calculated and multiplied by the preset maximum number of dilations to obtain the actual number of dilations. After dilation, the image is then subjected to constrained contrast adaptive histogram equalization. The image is divided into 8×8 subregions, and histogram equalization is performed within each subregion. The contrast enhancement is limited to a preset range to generate a contrast-enhanced image.

[0033] When there is no break in the contour boundary map, the system directly skips the morphological dilation operation, retains the integrity of the original contour, and enters the contrast-limited adaptive histogram equalization processing flow.

[0034] Extract oriented gradient histogram features from the enhanced image. First, calculate the gradient magnitude and direction for each pixel. Divide the image into 16×16 cells. Compute the histogram of the gradient direction within each cell, dividing the direction range into nine intervals. Then, group adjacent 2×2 cells into blocks, and normalize the feature vectors within the block. Finally, concatenate the feature vectors of all blocks to form the final feature description vector.

[0035] The extracted feature vectors are fed into a pretrained support vector machine classifier. This classifier uses a radial basis function as the kernel, with kernel parameters determined through grid search and cross-validation. The classifier outputs a geometric shape label for each solder joint region, including basic types such as circle, ellipse, and rectangle.

[0036] Based on the classification results, the Harris corner detection algorithm is used to extract contour keypoints. First, the image gradients in the x and y directions are calculated. Then, the autocorrelation matrix of each pixel is calculated. The eigenvalues ​​of the matrix are used to determine the corner response function value. Points with response values ​​greater than a preset threshold are considered candidate corner points. This threshold is determined by analyzing the distribution of corner response values ​​of typical welds in the training set. Non-maximum suppression is applied to the detected corner points, and the local maximum points are retained as the final corner points.

[0037] Cubic spline interpolation is used to smooth the sequence of corner point coordinates. Using the discrete corner point coordinates as control points, a cubic spline curve is constructed, ensuring that the curve passes through all control points and has continuous first- and second-order derivatives. The interpolated curve point coordinates form a smooth contour description.

[0038] Finally, a least-squares ellipse fitting algorithm is used to calculate the weld spot size parameters. The smoothed contour point coordinates are used as input, and the optimal ellipse is fitted by minimizing the sum of the squares of the algebraic distances between the points and the ellipse. To eliminate the influence of outliers, a polygonal convex hull algorithm is first used to calculate the convex hull vertices of the contour points. Only these vertices are used for ellipse fitting. The fitting results include the ellipse center coordinates, major axis length, minor axis length, and rotation angle parameters, which are the final target size parameters.

[0039] This step enables the system to automatically match the optimal welding parameters (such as soldering iron temperature, contact pressure, etc.) through precise dimensional parameter measurement, solving the problem of over-soldering or cold soldering caused by inaccurate dimensional recognition of traditional welding equipment.

[0040] In step S14, the relative position of the target in space is calculated based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data.

[0041] In a specific embodiment, calculating the relative position of the target in space based on the target size parameter in combination with a preset three-dimensional space coordinate system to obtain target position data includes: According to the target size parameters, combined with a preset three-dimensional space coordinate system, a Perspective-n-Point algorithm is used to perform spatial mapping to obtain the initial position of the target in space; If the deviation between the initial pose and the actual space coordinates exceeds the preset error threshold, the Levenberg-Marquardt nonlinear optimization algorithm is called to output the optimized pose by minimizing the reprojection error of the 2D-3D point pair. According to the optimized posture, the posture is finely calibrated through the ICP point cloud registration algorithm, and the target position data is output.

[0042] Specifically, the input data for this step consists of two parts: the first is the target size parameters obtained in step S13, specifically the coordinates of the ellipse center in the image coordinate system, the lengths of the ellipse's major and minor axes, and the ellipse's rotation angle; the second is the preset three-dimensional space coordinate system parameters, specifically the camera's internal parameter matrix and lens distortion coefficients. Together, these input data form the basis for spatial positioning.

[0043] When using the Perspective-n-Point algorithm for initial pose estimation, a certain number of feature points must first be selected on the ellipse contour. This selection method involves sampling the ellipse contour at equal angles, typically selecting 8-12 evenly distributed feature points. The coordinates of each feature point in the image can be accurately calculated using the ellipse's parametric equations. Furthermore, the corresponding coordinates of these feature points in 3D space must be determined based on the 3D geometric model of the weld. By establishing the correspondence between these 2D image points and 3D space points, a set of camera projection equations can be constructed. To solve this set of equations, a direct linear transformation method is first used to obtain an initial solution. Then, through orthogonalization, a rotation matrix and translation vector that conform to rigid body transformation constraints are obtained. This process outputs an initial pose containing six degrees of freedom: three for rotation and three for translation.

[0044] After the initial pose estimation is complete, its accuracy needs to be evaluated. This is done by calculating the reprojection error: projecting the 3D model points onto the 2D image plane according to the current pose and computing the Euclidean distance between the projected points and the actual image feature points. The preset error threshold is derived through statistical analysis of a large amount of experimental data. This method involves collecting hundreds of sets of calibration data under different working conditions, calculating the reprojection error distribution for each set, and taking the 95th percentile as the error threshold. If the measured reprojection error exceeds this threshold, the initial pose is inaccurate and requires optimization.

[0045] When invoking the Levenberg-Marquardt nonlinear optimization algorithm, the pose parameters must be reparameterized into a 6-dimensional vector. During the optimization process, the reprojection error for the current pose is first calculated. Then, the partial derivatives of the error function with respect to each pose parameter are calculated using numerical differences to construct the Jacobian matrix. An incremental equation is constructed based on the Jacobian matrix and the error vector, and the pose parameter adjustments are calculated by solving this equation. During the iteration process, the damping coefficient is dynamically adjusted based on the error changes. When the error decreases, the damping coefficient is reduced to accelerate convergence, and when the error increases, the damping coefficient is increased to ensure stability. Iteration termination conditions include reaching the maximum number of iterations or the error change between two consecutive iterations being less than a set value.

[0046] After nonlinear optimization, a more refined pose calibration is required. This is achieved using an iterative closest point algorithm. First, the 3D model points are projected into the point cloud space based on the current pose. Then, for each model point, the nearest neighbor is found in the measured point cloud to establish a new point correspondence. By calculating the centroid and covariance matrix of the two point clouds, singular value decomposition is used to determine the optimal rigid body transformation. This process is repeated until the change in pose parameters falls below the set convergence threshold.

[0047] The final output target position data is represented as a 4×4 homogeneous transformation matrix, containing both position and posture information. This matrix fully describes the position and posture of the solder joint in the three-dimensional spatial coordinate system. The accuracy of the entire positioning process is quantitatively evaluated using the covariance matrix, which reflects the positioning uncertainty in each degree of freedom. This multi-stage spatial positioning method overcomes the limitations of a single algorithm through a stepwise refinement strategy, achieving submillimeter positioning accuracy while maintaining computational efficiency.

[0048] This step achieves a precise mapping from 2D image features to 3D spatial coordinates through multi-stage pose estimation and optimization. Compared to traditional monocular vision positioning methods, this solution improves positioning stability and anti-interference capabilities in complex working conditions by introducing a multi-level optimization strategy.

[0049] In step S15, a preliminary operation path is determined based on the target location data and the acquired path constraint conditions.

[0050] In a specific embodiment, determining the preliminary operation path based on the target location data and the acquired path constraint conditions includes: Obtaining the joint motion range constraints of the welding robot arm, combining the target position data, and generating an initial collision-free path through the RRT-Connect path planning algorithm; The initial collision-free path is smoothed using a cubic B-spline interpolation algorithm to generate an adjustment path that satisfies the kinematic constraints of the manipulator, thereby obtaining a preliminary operation path.

[0051] Specifically, the input data includes the target position data obtained in step S14 (including the 3D coordinates and posture angle of the weld center) and the motion constraint parameters of the welding robot (including the angle range, velocity limits, and acceleration limits of each joint). First, a 3D environmental model of the workspace must be constructed. This model includes the geometric information of the robot body, welding gun tool, workpiece, and surrounding obstacles. A hierarchical bounding box approach is used to represent the collision volume of each object. The robot links are represented by cylindrical bounding boxes, the welding gun by conical bounding boxes, and the workpiece and obstacles by rectangular bounding boxes.

[0052] When using the RRT-Connect algorithm for path planning, a target point is first randomly sampled in the configuration space of the robot arm. This target point must meet the posture requirements required for the welding gun end effector to reach the target position. The algorithm maintains two random trees, which grow from the initial configuration and the target configuration respectively. At each iteration, one tree randomly selects a growth direction, and the other tree attempts to connect directly. During the growth process, it is necessary to detect in real time whether the angles of each joint exceed the limit and the collision between the robot arm links and the environment. Collision detection is achieved by calculating the intersection test between the bounding boxes at each level. When any two bounding boxes intersect, it is determined to be a collision. The termination condition of the path search is that the two trees are successfully connected or the maximum number of iterations is reached. After successful connection, the path from the initial point to the target point consists of a series of intermediate configuration points, which record the angle values ​​of each joint of the robot arm.

[0053] Since the path generated by the RRT-Connect algorithm may have unnecessary jitter, it is necessary to use a cubic B-spline curve for smoothing. During the processing, the configuration points on the path are first used as control points to construct a spline curve with C2 continuity. When parameterizing the curve, the kinematic constraints of the robot arm need to be considered, and the node vectors are adjusted to ensure that the speed and acceleration of each joint do not exceed the limit value. The smoothed path needs to be re-collision verified to ensure that the collision-free characteristics are maintained while satisfying the kinematic constraints. The final preliminary operation path is represented by a set of densely sampled configuration point sequences, each of which contains a timestamp, joint angle, velocity and acceleration information.

[0054] The key parameters involved in this step include the maximum number of iterations and step size of the RRT-Connect algorithm, which are determined by analyzing the size and complexity of the robot's workspace. The smoothing strength parameter of the cubic B-spline curve is adjusted according to the dynamic performance of the robot to ensure a smooth and executable trajectory. The entire path planning process enables the rapid generation of a collision-free path that meets the robot's motion constraints in a complex environment, laying the foundation for subsequent path optimization and welding execution. Compared with traditional linear interpolation methods, this sampling-based path planning algorithm can better handle obstacle avoidance problems in complex environments while ensuring the smoothness and executable nature of the path.

[0055] In step S16, a simulation operation is performed according to the preliminary operation path, and a position offset is obtained. When the position offset exceeds a preset offset threshold, the path parameters are adjusted to determine an optimized operation path.

[0056] In a specific embodiment, performing a simulation according to the preliminary operation path and obtaining a position offset, and when the position offset exceeds a preset offset threshold, adjusting path parameters to determine an optimized operation path, includes: According to the preliminary operation path and the position offset, a model predictive control algorithm is used to dynamically adjust the path parameters to generate a preliminary corrected path; Perform simulation according to the preliminary corrected path and obtain the actual motion trajectory; Performing path matching between the preliminary corrected path and the actual motion trajectory using a dynamic time warping algorithm to obtain a path matching error; When the path matching error is higher than a preset matching error threshold, a nonlinear least squares optimization algorithm is called to perform frame-by-frame fine-tuning on the preliminary correction path to generate an optimized operation path.

[0057] Specifically, the input data includes the preliminary operation path obtained in step S15 (including the angle sequence, velocity curve, and acceleration curve of each joint of the manipulator) and the manipulator's dynamic parameters (including the mass of each link, the inertia matrix, and the joint friction coefficient). First, a dynamic model of the manipulator is established in a virtual simulation environment. This model accounts for nonlinear factors such as gravity compensation, Coriolis force, and centrifugal force. During the simulation run, the preliminary operation path is input into the controller, and the torque required for each joint is calculated by solving the inverse dynamics equation. The actual motion process is then simulated based on the dynamic characteristics of the manipulator.

[0058] During the simulation, virtual sensors collect data on the actual motion trajectory of the robot's end effector. This data records the end effector's position and posture at a fixed sampling frequency. The actual trajectory is compared with the expected path, and the position offset at each sampling moment is calculated. The offset consists of two components: translation error and rotation error. Preset offset thresholds are determined by analyzing the welding process requirements. The translation error threshold is set as a percentage of the weld diameter, and the rotation error threshold is determined based on the weld angle tolerance. When any error component exceeds the corresponding threshold, the path correction process is triggered.

[0059] When using a model predictive control algorithm for path correction, a rolling optimization problem is constructed, encompassing both the prediction and control domains. During each control cycle, the optimal control input is obtained by solving a constrained optimization problem based on the current robot state and the path reference values ​​for several future steps. The optimization objective function includes three terms: trajectory tracking error, rate of change of the controlled variable, and terminal state error. The weight coefficients for each term are adjusted based on the welding process requirements. The optimization process requires consideration of the robot's joint limits, speed constraints, and torque constraints to ensure that the generated corrected path is executable. The initial corrected path consists of a series of optimized joint angle sequences.

[0060] The preliminary corrected path is simulated again, and the new actual motion trajectory data is recorded. When using the dynamic time warping algorithm to evaluate the correction effect, the expected path and the actual trajectory are first time-aligned to eliminate the impact of execution timing differences. The Euclidean distance between the aligned trajectory points is then calculated, and the average distance across all points is taken as the path matching error. The matching error threshold is determined by statistically analyzing the results of multiple simulation experiments. When the error exceeds the matching error threshold, more refined path adjustments are required.

[0061] When using a nonlinear least-squares optimization algorithm for fine-tuning, the path parameters (joint angles, velocities, and accelerations) are used as optimization variables, constructing an optimization problem with the goal of minimizing path matching error. The trust region method is used to ensure convergence during the optimization process, and the finite difference method is used to calculate gradient information. After each iteration, resimulation is performed to verify the effectiveness of the adjustments until the matching error falls below the matching error threshold or the maximum number of iterations is reached. The resulting optimized operation path not only satisfies the kinematic constraints but also accurately tracks the desired trajectory, ensuring the accuracy and stability of the welding process.

[0062] This step uses a multi-level optimization strategy to gradually improve path tracking accuracy and address trajectory deviations caused by the robot's dynamic characteristics. Compared to open-loop control methods, this closed-loop feedback-based path optimization technique effectively compensates for model errors and external interference, improving welding quality and process reliability. The parameter settings for each optimization algorithm have been validated through extensive simulation experiments to ensure stable optimization results under different operating conditions.

[0063] In step S17, a final execution path is determined based on the optimized operation path and the environmental interference data.

[0064] In a specific embodiment, determining the final execution path based on the optimized operation path and in combination with the environmental interference data includes: Inputting the optimized operation path and the environmental interference data into an adaptive model predictive control algorithm, dynamically adjusting the path parameters through rolling horizon optimization, and generating an interference-resistant execution path; When the environmental interference data exceeds a preset interference threshold, an online Gaussian process regression algorithm is called to predict the trajectory and generate an obstacle avoidance path offset; The anti-interference execution path and the obstacle avoidance path offset are superimposed and fused to generate a final execution path.

[0065] Specifically, the input data includes the optimized operation path obtained in step S16 (including the angle, velocity, and acceleration sequence of each joint of the robot arm) and the environmental interference data continuously collected in step S11 (including vibration spectrum, temperature distribution, and electromagnetic field strength). First, a mapping relationship model between interference factors and robot motion errors is established. This model is obtained by analyzing historical data and records the typical deviation patterns of the robot arm end effector under different interference intensities.

[0066] When using an adaptive model predictive control algorithm to handle conventional disturbances, a closed-loop control framework is constructed, integrating the current state, future predictions, and feedback corrections. During each control cycle (e.g., 10 milliseconds), the algorithm performs the following operations: It reads the actual position feedback from the robot arm's joint encoders and compares it with the expected position of the optimized operation path to obtain the tracking error; it automatically adjusts the parameters of the prediction model based on real-time changes in environmental disturbance data; and it solves a constrained optimization problem within a rolling time window (e.g., one second into the future) to generate control instructions. The optimization objective function consists of three terms: first, minimizing the position error between the welding gun tip and the weld point; second, controlling the smoothness of the joint motion; and finally, optimizing energy consumption. Constraints include the robot arm's kinematic and dynamic limits. By adjusting the lengths of the prediction and control time windows in real time, the algorithm maintains good performance under varying disturbance intensities.

[0067] When environmental disturbance data (such as vibration amplitude or temperature gradient) exceeds a preset safety threshold, an online Gaussian process regression algorithm is activated for trajectory prediction. This safety threshold is determined through destructive experiments. Under laboratory conditions, the intensity of each disturbance factor is gradually increased, and the critical value that leads to unsatisfactory welding quality is recorded. A threshold of 80% is then used as the warning threshold. The Gaussian process regression algorithm establishes a probabilistic model based on historical disturbance data to predict the maximum deviation that the robot arm may experience over several future time steps. Based on the predicted results, the required obstacle avoidance path offset is calculated. This offset consists of two components: spatial position correction and attitude adjustment.

[0068] A weighted superposition approach is used to fuse the interference-avoidance execution path with the obstacle avoidance offset. For position correction, a weight coefficient is determined based on the interference intensity, with the obstacle avoidance offset weight being increased in cases of strong interference. For posture adjustment, priority is given to maintaining the perpendicularity of the welding gun to the workpiece surface. During the fusion process, the robot arm's self-collision and collisions with environmental obstacles are detected in real time, with secondary adjustments performed as necessary. The resulting execution path not only incorporates position and velocity information in the joint space but also incorporates the expected interference compensation at each point in time.

[0069] The innovation of this step is reflected in three aspects: first, adaptive model predictive control realizes the combination of interference feedforward compensation and feedback correction, which has stronger anti-interference ability than traditional PID control; second, Gaussian process regression is used for prediction, which can handle nonlinear and non-stationary interference changes; finally, the path fusion mechanism ensures safety under strong interference, while maintaining the optimality of the original path as much as possible, improving the stability of welding quality in complex environments.

[0070] In step S18, according to the final execution path, the intelligent switching of the execution tool is driven to perform the welding operation on the welding target.

[0071] Specifically, the input data is the final execution path generated in step S17. This path data includes the spatial position sequence of each joint of the robot arm, the motion velocity curve, the acceleration curve, and the spatial posture data of the welding gun end. Also input is a welding process parameter library, which stores process parameters such as welding current, voltage, wire feed speed, and shielding gas flow rate for different weld sizes.

[0072] The execution process begins with adaptive soldering tool switching. The system selects a matching soldering tip from the tool library based on the current solder joint's dimensional characteristics (the target dimensional parameters from step S13). The tool switching mechanism is driven by a servo motor, with closed-loop position control achieved through high-precision encoder feedback. During the switching process, the tool position signal is monitored in real time to ensure that the soldering tip is properly installed and making good contact. After the tool switch is complete, the system automatically loads the corresponding soldering process parameters and preheats the soldering tip to the set temperature.

[0073] During the welding path execution phase, the robot controller controls motion according to the spatiotemporal requirements of the final execution path. In Cartesian space, a position-velocity-acceleration third-order planning algorithm is used to generate a smooth terminal trajectory. In joint space, an inverse kinematics solution is used to convert the Cartesian trajectory into angle commands for each joint. During motion, a force sensor monitors the contact force between the welding gun and the workpiece in real time. If the contact force deviates from the set range, the robot's impedance control parameters are immediately adjusted to maintain a constant welding pressure.

[0074] During the welding process, the system simultaneously monitors the welding current waveform, weld pool morphology, and heat-affected zone temperature distribution. By comparing real-time welding parameters with a database of process specifications, it dynamically adjusts welding power output. When a risk of a weld defect (such as lack of penetration or over-burning) is detected, the process parameters for subsequent welds are automatically corrected. After each weld is completed, the vision system immediately inspects the weld quality, recording quality indicators such as weld formation and surface porosity.

[0075] First, adaptive welding of multi-sized solder joints is achieved through intelligent matching of welding tools and process parameters. Second, closed-loop control integrating force control and visual feedback ensures the stability of the welding process. Finally, a parameter self-optimization mechanism based on real-time quality detection improves the welding yield rate.

[0076] Reference Figure 2 The second embodiment of the present invention provides a control system for automatically switching soldering irons with multiple sizes of solder joints, comprising: A data acquisition module is used to collect a set of original images of the welding target and obtain environmental interference data under a preset angle and preset lighting conditions; A data processing module, configured to preprocess the original image set to obtain a standard image set; An object size analysis module is used to extract features and determine object size parameters based on the standard image set using a pre-trained deep learning model; A target position determination module is used to calculate the relative position of the target in space based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data; A preliminary path determination module, configured to determine a preliminary operation path based on the target location data and the acquired path constraint conditions; an optimized path determination module, configured to perform a simulation operation according to the preliminary operation path and obtain a position offset; and when the position offset exceeds a preset offset threshold, adjust path parameters to determine an optimized operation path; A final path determination module, configured to determine a final execution path based on the optimized operation path and the environmental interference data; The execution module is used to drive the intelligent switching of the execution tool according to the final execution path to perform welding operations on the welding target.

[0077] It should be noted that the control device for automatically switching soldering irons for multi-sized solder joints provided in an embodiment of the present invention is used to execute all the process steps of the control method for automatically switching soldering irons for multi-sized solder joints in the above embodiment. The working principles and beneficial effects of the two correspond one to one, and therefore will not be repeated here.

[0078] An embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a control program for automatically switching soldering irons for solder joints of multiple sizes. When the processor executes the computer program, the steps of the above-mentioned control method for automatically switching soldering irons for solder joints of multiple sizes are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are realized, such as a control module for automatically switching soldering irons with multiple-sized solder joints.

[0079] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0080] The electronic device may be a computing device such as a desktop computer, notebook, PDA, or smart tablet. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the aforementioned components are merely examples of electronic devices and do not constitute a limitation of the electronic device. The electronic device may include more or fewer components than those described above, or a combination of certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, and the like.

[0081] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the electronic device and connects various parts of the entire electronic device using various interfaces and lines.

[0082] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function); the data storage area may store data generated based on the use of the mobile phone (such as audio data, a phone book, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0083] If the module / unit integrated into the electronic device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content of the computer-readable medium can be appropriately increased or decreased based on the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, based on legislation and patent practice, computer-readable media does not include electric carrier signals and telecommunication signals.

[0084] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive effort.

[0085] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A control method for automatically switching soldering irons for solder joints of multiple sizes, characterized in that: include: Collecting a set of original images of the welding target and obtaining environmental interference data under a preset angle and preset lighting conditions; Preprocessing the original image set to obtain a standard image set; Based on the standard image set, a pre-trained deep learning model is used to extract features and determine target size parameters; Calculating the relative position of the target in space based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data; Determining a preliminary operation path based on the target location data and the acquired path constraints; Performing a simulation according to the preliminary operation path and obtaining a position offset; when the position offset exceeds a preset offset threshold, adjusting path parameters to determine an optimized operation path; Determining a final execution path based on the optimized operation path and in combination with the environmental interference data; According to the final execution path, the intelligent switching of the execution tool is driven to perform welding operations on the welding target.

2. The control method for automatically switching soldering irons with multiple-size solder joints according to claim 1, characterized in that: The preprocessing of the original image set to obtain a standard image set includes: According to the original image set, dynamically denoising is performed by using a three-layer decomposition of the Symlets-8 wavelet basis and a Bayesian shrinkage threshold algorithm to output a denoised image set; According to the denoised image set, a limited contrast adaptive histogram equalization algorithm is used to locally enhance the contrast of the denoised image set, and a Reinhard color migration algorithm is used to unify the color gamut of the denoised image set under different lighting conditions to generate a standard image set.

3. The control method for automatically switching soldering irons with multiple-sized solder joints according to claim 1, characterized in that: The method of extracting features using a pre-trained deep learning model based on the standard image set and determining target size parameters includes: According to the standard image set, a pre-trained deep learning model is used to extract contour features, output a mask map containing pixel-level contour probabilities, and obtain a contour boundary map after binarization; When there are broken contours in the contour boundary map, a morphological dilation operation is used to strengthen edge continuity, and then a contrast-limited adaptive histogram equalization algorithm is used to locally enhance the contrast to generate an optimized enhanced image set; Based on the enhanced image set, a histogram of oriented gradients algorithm is used to extract features, and the extracted features are input into a pre-trained support vector machine classifier to output a final image set containing geometric labels; According to the final image set, the Harris corner detection algorithm is used to extract the contour key points, and the cubic spline interpolation algorithm is used to perform smoothing to obtain a key point coordinate set; According to the set of key point coordinates, the least squares ellipse fitting algorithm is used to perform parameter estimation and the polygon convex hull algorithm is used to correct the parameter boundaries to determine the target size parameters.

4. The control method for automatically switching soldering irons with multiple-sized solder joints according to claim 1, characterized in that: The step of calculating the relative position of the target in space based on the target size parameter and in combination with a preset three-dimensional space coordinate system to obtain target position data includes: According to the target size parameters, combined with a preset three-dimensional space coordinate system, a Perspective-n-Point algorithm is used to perform spatial mapping to obtain the initial position of the target in space; If the deviation between the initial pose and the actual space coordinates exceeds the preset error threshold, the Levenberg-Marquardt nonlinear optimization algorithm is called to output the optimized pose by minimizing the reprojection error of the 2D-3D point pair. According to the optimized posture, the posture is finely calibrated through the ICP point cloud registration algorithm, and the target position data is output.

5. The control method for automatically switching soldering irons with multiple-sized solder joints according to claim 1, characterized in that: The step of determining a preliminary operation path based on the target location data and the acquired path constraint conditions includes: Obtaining the joint motion range constraints of the welding robot arm, combining the target position data, and generating an initial collision-free path through the RRT-Connect path planning algorithm; The initial collision-free path is smoothed using a cubic B-spline interpolation algorithm to generate an adjustment path that satisfies the kinematic constraints of the manipulator, thereby obtaining a preliminary operation path.

6. The control method for automatically switching soldering irons with multiple-sized solder joints according to claim 1, characterized in that: The simulation operation is performed according to the preliminary operation path, and a position offset is obtained. When the position offset exceeds a preset offset threshold, path parameters are adjusted to determine an optimized operation path, including: According to the preliminary operation path and the position offset, a model predictive control algorithm is used to dynamically adjust the path parameters to generate a preliminary corrected path; Perform simulation according to the preliminary corrected path and obtain the actual motion trajectory; Performing path matching between the preliminary corrected path and the actual motion trajectory using a dynamic time warping algorithm to obtain a path matching error; When the path matching error is higher than a preset matching error threshold, a nonlinear least squares optimization algorithm is called to perform frame-by-frame fine-tuning on the preliminary correction path to generate an optimized operation path.

7. The control method for automatically switching soldering irons with multiple-sized solder joints according to claim 1, characterized in that: Determining a final execution path based on the optimized operation path and in combination with the environmental interference data includes: Inputting the optimized operation path and the environmental interference data into an adaptive model predictive control algorithm, dynamically adjusting the path parameters through rolling horizon optimization, and generating an interference-resistant execution path; When the environmental interference data exceeds a preset interference threshold, an online Gaussian process regression algorithm is called to predict the trajectory and generate an obstacle avoidance path offset; The anti-interference execution path and the obstacle avoidance path offset are superimposed and fused to generate a final execution path.

8. A control system for automatically switching soldering irons for solder joints of multiple sizes, characterized in that: include: A data acquisition module is used to collect a set of original images of the welding target and obtain environmental interference data under a preset angle and preset lighting conditions; A data processing module, configured to preprocess the original image set to obtain a standard image set; An object size analysis module is used to extract features and determine object size parameters based on the standard image set using a pre-trained deep learning model; A target position determination module is used to calculate the relative position of the target in space based on the target size parameters and a preset three-dimensional space coordinate system to obtain target position data; A preliminary path determination module, configured to determine a preliminary operation path based on the target location data and the acquired path constraint conditions; an optimized path determination module, configured to perform a simulation operation according to the preliminary operation path and obtain a position offset; and when the position offset exceeds a preset offset threshold, adjust path parameters to determine an optimized operation path; A final path determination module, configured to determine a final execution path based on the optimized operation path and the environmental interference data; The execution module is used to drive the intelligent switching of the execution tool according to the final execution path to perform welding operations on the welding target.

9. An electronic device, characterized in that: The invention comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the control method for automatically switching soldering irons with multi-sized solder joints according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the control method for automatically switching soldering irons with multiple-sized solder joints according to any one of claims 1 to 7.

Images