Laser welding self-adaptive path planning method and system

Abstract

The invention relates to the technical field of intelligent welding systems, and discloses a laser welding self-adaptive path planning method and system.The method comprises the steps that weld joint three-dimensional point cloud data, a visual image and welding wire diameter data are obtained; identifying weld edge feature points after denoising, generating a gap width continuous change curve, and positioning abnormal feature points; if the process tolerance threshold value is exceeded, dividing a filling sub-region, and estimating a residual space to generate a current layer optimization path; simulating the shape of the remaining gap of the next layer, adjusting the width sequence to avoid a narrow area, and obtaining filling path parameters; and generating a self-adaptive welding track and calculating a friendliness score, and iteratively updating parameters until the parameters reach the standard to obtain a multi-layer path planning scheme. According to the method, self-adaptive multi-layer path planning under gap dynamic change can be achieved, and the strict requirements of laser welding for high-precision and high-consistency welding are met.

Description

Technical Field

[0001] This invention relates to the field of intelligent welding system technology, and in particular to a laser welding adaptive path planning method and system. Background Technology

[0002] Currently, in the field of intelligent welding system technology, with the continuous improvement of welding quality requirements in high-end manufacturing such as aerospace and energy equipment, and the increasing number of complex weld seam scenarios, laser welding path planning, as a core link to ensure weld seam forming accuracy and structural reliability, is directly related to the service life and operational safety of products.

[0003] Existing laser welding path planning methods in the industry mainly rely on fixed preset trajectories or offline programming of simple geometric features. For example, they use a uniform filling path to ignore irregular changes in the gap, or they plan based on theoretical models without considering actual clamping deviations, or single-layer planning does not take into account the filling needs of subsequent layers. However, this approach is clearly inadequate in complex operating environments. Because it cannot respond in real time to dynamic gap fluctuations caused by part deformation and clamping deviations, it easily leads to incomplete filling; the lack of coordinated multi-layer filling logic means that the current layer planning can easily cause narrow areas smaller than the welding wire diameter in subsequent layers, making them difficult to fill; and the lack of a dynamic adjustment mechanism, especially in scenarios with abrupt gap changes, easily leads to defects such as incomplete fusion and weld beads, affecting welding quality.

[0004] In summary, existing technologies are insufficient to achieve adaptive multi-layer path planning under dynamic gap changes, and cannot meet the stringent requirements of laser welding for high-precision and high-consistency welding. Summary of the Invention

[0005] This invention provides an adaptive path planning method and system for laser welding, which enables adaptive multi-layer path planning under dynamic gap changes, meeting the stringent requirements of laser welding for high-precision and high-consistency welding.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a laser welding adaptive path planning method, comprising: Acquire 3D point cloud data of the weld area, visual images of the weld, and welding wire diameter data; The three-dimensional point cloud data is denoised, and the edge feature points of the weld are identified. A continuous variation curve of the gap width is generated based on the spacing of the edge feature points. The abnormal points of the gap are identified from the continuously changing curve, and the correspondence between the abnormal points and the image pixel area is established by combining the visual image of the weld seam to determine the location of the abnormal feature points. If the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, then according to the three-dimensional geometric boundary of the ultra-wide region where the feature point location is located, a filling sub-region is divided, the remaining space value of the filling sub-region is estimated, and the optimized path of the current layer is generated based on the remaining space value. Based on the filling start point and width sequence of the optimized path, construct the solid contour model after filling the current layer, separate the gap areas not covered by the solid contour model, and extract the coordinate set of the closed boundary of the remaining gap in the next layer. Discretize the coordinate set along the direction perpendicular to the filling path, calculate the distance between boundary point pairs, generate the global gap width distribution, and simulate the shape of the remaining gap in the next layer. Traverse the width data of the remaining gap shape. If there is a narrow area smaller than the welding wire diameter data, adjust the width sequence to obtain the fill path parameters. Based on the filling path parameters, an adaptive welding trajectory for the current layer is generated, and a friendliness score is calculated. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme.

[0007] In a second aspect, the present invention provides a laser welding adaptive path planning system, comprising: The data acquisition module is used to acquire 3D point cloud data of the weld area, visual images of the weld, and welding wire diameter data; The curve generation module is used to denoise the three-dimensional point cloud data, identify the edge feature points of the weld, and generate a continuous curve of the gap width based on the spacing of the edge feature points. The feature localization module is used to identify abnormal points in the gap from the continuously changing curve, and to establish the correspondence between the abnormal points and the image pixel area by combining the weld visual image, thereby determining the location of the abnormal feature points. The path optimization module is used to divide the filling sub-region according to the three-dimensional geometric boundary of the ultra-wide region where the feature point is located if the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, estimate the remaining space value of the filling sub-region, and generate the optimized path of the current layer according to the remaining space value. The coordinate extraction module is used to construct the entity contour model after the current layer is filled based on the filling start point and width sequence of the optimized path, separate the gap area not covered by the entity contour model, and extract the coordinate set of the closed boundary of the remaining gap in the next layer. The shape simulation module is used to discretize the coordinate set along a direction perpendicular to the filling path, calculate the distance between boundary point pairs, generate the global gap width distribution, and simulate the shape of the remaining gap in the next layer. The parameter correction module is used to traverse the width data of the remaining gap shape. If there is a narrow area smaller than the welding wire diameter data, the width sequence is adjusted to obtain the filling path parameters. The update and optimization module is used to generate an adaptive welding trajectory for the current layer based on the filling path parameters and calculate a friendliness score. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains three-dimensional point cloud data, visual images and welding wire diameter data of the weld area, denoises the three-dimensional point cloud and identifies the edge features of the weld, generates a continuous curve of gap width change, and combines the image to locate abnormal feature points. It breaks through the limitation that the traditional fixed trajectory cannot respond to the dynamic fluctuation of the gap, explores the dynamic distribution characteristics of the weld gap in the whole domain, eliminates the interference of clamping deviation and part deformation, provides high-precision basic data support for path planning, effectively improves the accuracy of identifying abnormal points in the gap, and solves the problem of insufficient filling caused by fixed path.

[0009] (2) This invention divides the ultra-wide area into filling sub-regions, estimates the remaining space value to obtain optimized filling parameters, constructs a solid contour model to simulate the shape of the remaining gap in the next layer, traverses the width data to adjust the filling sequence, breaks through the limitation of traditional single-layer planning that does not take into account subsequent layers, accurately captures the spatial correlation characteristics of multi-layer filling, provides multi-dimensional basis for path correction, significantly improves the pertinence of path planning under complex gaps, and makes up for the defect of existing technology that small areas cannot be filled.

[0010] (3) The present invention generates an adaptive welding trajectory based on the filling path parameters, calculates the friendliness score to the subsequent layers, and iteratively updates the parameters until the standard is met. It solves the limitations of traditional methods that lack dynamic adjustment and multi-layer collaboration, provides accurate adaptive path basis for welding, solves defects such as non-fusion and weld beads, takes into account welding accuracy and process consistency, and meets the stringent requirements of high-end manufacturing for laser welding. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a laser welding adaptive path planning method provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a laser welding adaptive path planning system provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a laser welding adaptive path planning method, including the following steps: S101, acquire the 3D point cloud data of the weld area, the visual image of the weld, and the diameter data of the welding wire; S102, the three-dimensional point cloud data is denoised, and the edge feature points of the weld are identified. A continuous variation curve of the gap width is generated based on the spacing of the edge feature points. S103, Identify abnormal points in the gap from the continuously changing curve, establish the correspondence between the abnormal points and the image pixel area by combining the weld visual image, and determine the location of the abnormal feature points. S104, if the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, then according to the three-dimensional geometric boundary of the ultra-wide region where the feature point location is located, divide the filling sub-region, estimate the remaining space value of the filling sub-region, and generate the optimized path of the current layer according to the remaining space value. S105, Based on the filling start point and width sequence of the optimized path, construct the solid contour model after filling the current layer, separate the gap area not covered by the solid contour model, and extract the coordinate set of the closed boundary of the remaining gap in the next layer. S106, Discretize the coordinate set along the direction perpendicular to the filling path, calculate the distance between boundary point pairs, generate the gap width distribution of the entire domain, and simulate the shape of the remaining gap in the next layer. S107, Traverse the width data of the remaining gap shape. If there is a narrow area smaller than the welding wire diameter data, adjust the width sequence to obtain the filling path parameters. S108. Based on the filling path parameters, generate the adaptive welding trajectory of the current layer and calculate the friendliness score. If the friendliness score is lower than the preset friendliness threshold, update the filling path parameters and recalculate the friendliness score until the friendliness score meets the requirements, and obtain the final multi-layer path planning scheme.

[0014] In step S101, the three-dimensional point cloud data of the weld area, the visual image of the weld, and the diameter data of the welding wire are acquired, including: The weld area is scanned by a laser sensor, the laser echo signal is received and converted into three-dimensional point cloud data; Simultaneously acquire visual images of the weld area as weld visual images, and align the timestamps of the weld visual images with the timestamps of the three-dimensional point cloud data; Retrieve pre-stored welding wire diameter data, which is matched with welding process parameters.

[0015] It should be noted that, firstly, when scanning the weld area with a laser sensor, receiving the laser echo signal and converting it into three-dimensional point cloud data, a line structured light laser sensor is selected for example, with the scanning frequency set to 100Hz. The projected line laser covers the weld and a 10mm area on each side to ensure complete capture of the gap features. The echo signal received by the sensor is converted into three-dimensional spatial coordinates through triangulation to form initial point cloud data. After acquisition, the coordinates are first normalized, specifically using the minimum-maximum normalization method. The minimum value of each dimension is subtracted from the coordinate value of that dimension, and then divided by the range (the difference between the maximum and minimum values), thereby mapping the X, Y, and Z axis coordinates to the [0,1] interval.

[0016] Next, visual images of the weld area are acquired synchronously. When aligning the timestamps of the visual images with the 3D point cloud data, a high-definition industrial camera is used to acquire visual images synchronously. The camera frame rate is kept consistent with the laser sensor scanning frequency (100fps) to ensure that each frame of the point cloud has corresponding image data. Timestamp alignment is achieved through a hardware trigger signal, with the trigger error controlled within 1ms. After alignment, spatial consistency needs to be verified by calculating the mapping error between image pixels and point cloud coordinates. If it exceeds 0.1mm, images are re-acquired for calibration. For example, a prominent geometric feature point in the weld area is selected, and its 3D point cloud coordinates are projected onto the image plane using a calibration matrix to obtain theoretical pixel coordinates. The spatial distance deviation between these coordinates and the converted feature point pixel coordinates in the image is 0.05mm, which is less than the 0.1mm threshold and meets the accuracy requirements.

[0017] In this embodiment, when retrieving pre-stored welding wire diameter data, the welding wire diameter data is stored in the local database of the welding control system, categorized by welding material type (e.g., stainless steel, aluminum alloy) and welding method (e.g., consumable electrode, non-consumable electrode). The data is obtained through pre-measurement calibration, using calipers to measure three times and take the average value, with an accuracy controlled within ±0.01mm to ensure consistency with the actual welding wire used. During retrieval, the system automatically matches the corresponding welding wire diameter data according to the process parameters of the current welding task (e.g., material is stainless steel, welding method is consumable electrode). If the matching fails, manual confirmation is prompted. For example, if the current welding process is stainless steel consumable electrode laser welding, the system automatically retrieves welding wire data with a diameter of 1.2mm. This data is compatible with the welding process parameters such as welding current and wire feed speed, ensuring a good filling effect.

[0018] In step S102, the three-dimensional point cloud data is denoised, and edge feature points of the weld are identified. A continuous variation curve of the gap width is generated based on the spacing of the edge feature points, including: The 3D point cloud data is denoised to extract a set of effective depth information. Identify gradient abrupt change locations in the depth information set to determine the edge feature points of the weld; Calculate the Euclidean distance between the edge feature points to obtain discrete gap width values; The gap width value is fitted to generate a continuous variation curve of the gap width along the weld direction.

[0019] It should be noted that, firstly, when denoising the 3D point cloud data and extracting the effective depth information set, a combined denoising method of statistical filtering and radius filtering is used. Statistical filtering, based on the Gaussian assumption of the point cloud distribution, effectively removes random measurement noise (Gaussian noise) caused by sensor accuracy by calculating the average distance and standard deviation of the neighborhood. Radial filtering, on the other hand, accurately filters out isolated outliers (impulse noise) caused by welding spatter or airborne dust by determining the point cloud density in the local space. The two complement each other to preserve the geometric features of the weld edge to the greatest extent. The statistical filtering sets the number of neighborhood points to 50 to ensure that the sample size has sufficient statistical representativeness. The average distance between each point and its neighbors is calculated, and points exceeding the mean plus twice the standard deviation are removed. The radius filtering sets the search radius to 0.0025 (normalized value, corresponding to a physical scale of approximately 0.5 mm), and removes outliers with fewer than 10 points within the radius.

[0020] Subsequently, when identifying gradient abrupt change locations in the depth information set and determining the edge feature points of the weld, the gradient abrupt change locations are obtained by calculating the first-order difference of the depth information and comparing it with a preset gradient judgment threshold. The difference direction is along the laser scanning direction.

[0021] It is worth noting that the gradient judgment threshold is set based on the normalized depth variation statistics of welds of different materials over the past year. Based on historical data, 95% of the normal surface gradients of welds are below 0.6mm, therefore the basic threshold is set to 0.003 (corresponding to a physical scale of approximately 0.6mm). Stainless steel welds are harder and have higher surface reflectivity, resulting in more drastic noise fluctuations and depth variations during laser scanning. To prevent misjudgments, the judgment threshold needs to be increased, which can be adjusted to 0.004. Aluminum alloy welds are softer and have good diffuse reflection characteristics of the surface oxide layer, resulting in gentler depth variations. Therefore, the threshold needs to be lowered to capture subtle edges, which can be adjusted to 0.002. This threshold has been tested in multiple batches and can accurately distinguish weld edges from normal surfaces. When the absolute value of the difference exceeds the threshold for three consecutive points, it is determined to be a gradient abrupt change location. A set of abrupt change points is extracted from both the left and right sides, forming a pair of weld edge feature points.

[0022] For example, in the normalized depth information of the butt weld of stainless steel plate, the absolute difference values ​​of a certain continuous point are 0.0035, 0.0040, and 0.0045, which exceed the basic threshold of 0.003. It is determined to be the edge abrupt change position, and the coordinates of the left and right edge feature points are extracted as (x1, y1, z1) and (x2, y2, z2), respectively.

[0023] Next, the Euclidean distance between the feature points on the weld edge is calculated to obtain discrete gap width values. Then, the feature points on the left and right edges are paired sequentially along the weld length, and the three-dimensional Euclidean distance of each pair of feature points is calculated. This distance is the discrete gap width at the corresponding position. Before calculation, it is necessary to ensure that the feature points are paired accurately to avoid errors caused by cross-pairing. If the distance between a pair of feature points exceeds a reasonable range (0-5mm), it is marked as an invalid point and discarded during subsequent fitting.

[0024] For example, the normalized coordinates of a pair of edge feature points are (0.5010, 0.2515, 0.0505) and (0.5075, 0.2515, 0.0505), respectively. The calculated Euclidean distance is 0.0065. This value accurately represents the gap width on a dimensionless scale and serves as an effective discrete gap width value.

[0025] Finally, when fitting the discrete gap width values ​​to generate a continuous curve showing the gap width variation along the weld direction, a cubic spline curve fitting technique is used. This technique can smooth out discrete noise and accurately retain the true variation trend of the gap, avoiding distortion caused by overfitting. Before fitting, invalid discrete points are removed, and then the valid values ​​are arranged in order along the weld length direction before being input into the fitting function to generate a continuous curve.

[0026] In step S103, abnormal points of the gap are identified from the continuously changing curve, and the correspondence between the abnormal points and the image pixel regions is established by combining the weld visual image to determine the location of the abnormal feature points, including: Traverse the continuous change curve. If the rate of change of the gap width at a certain point exceeds the preset abrupt change judgment threshold, it is marked as an abnormal point of the gap. Establish a spatial index between the continuous change curve and the visual image of the weld, and clarify the image pixel range corresponding to the abnormal point; The image pixel range is subjected to feature response analysis. If the response intensity exceeds a preset feature determination threshold, it is determined to be an abnormal feature point location.

[0027] It should be noted that, firstly, by traversing the continuous variation curve, if the rate of change of the gap width at a certain point exceeds the preset abrupt change judgment threshold, it is marked as an abnormal point in the gap. The rate of change of the gap width is the difference in gap width between two adjacent points divided by the distance along the weld length, reflecting the degree of abrupt change in the gap width. The abrupt change judgment threshold is determined based on the statistical distribution range of the change rate of the qualified weld gap. Under normal assembly and welding heat deformation conditions, the fluctuation of the gap width mainly exhibits low-frequency gradual characteristics, with its normalized gradient value mainly concentrated in the range of 0 to 4.0. Therefore, the abrupt change judgment threshold is preset to 5.0 to filter out normal gradual fluctuations while accurately capturing high-frequency abrupt changes in the gap caused by positioning weld points, burrs, or foreign objects. Those skilled in the art will understand that this threshold can be fine-tuned according to the physical properties of the welding material: for stainless steel, which has high rigidity and severe stress release, it can be appropriately increased to 6.0; for aluminum alloy, which has good ductility and gentle deformation, it can be decreased to 4.0 to adapt to the deformation sensitivity of different materials. During the traversal, the rate of change is calculated point by point along the weld length. If three consecutive points exceed the threshold, they are marked as abnormal points of the gap.

[0028] For example, in a normalized continuous variation curve, the normalized gap width of a certain point increases abruptly from 0.150 to 0.164 within the range of normalized path increment of 0.002. The calculated gradient of this change is (0.164-0.150) / 0.002=7.0, which exceeds the threshold of 6.0 for stainless steel welds. This is determined to be a sudden change, and the segment is marked as an abnormal point of the gap.

[0029] Next, a spatial index is established between the continuously changing curve and the weld visual image. The spatial index is constructed through timestamp alignment and coordinate mapping. Each point on the continuously changing curve (normalized coordinates [0,1]) is associated with a precise sampling timestamp of a laser sensor. First, the weld visual image frame with the closest acquisition time is found using the timestamp index. Then, using a calibrated mapping matrix, the normalized one-dimensional coordinates P along the weld length direction are converted into two-dimensional pixel coordinates (u,v) in the image coordinate system, where u corresponds to the distribution of the weld along the image width and v corresponds to the image height.

[0030] After indexing, the corresponding image pixel range on the X-axis is determined based on the weld length range of the anomaly point. The Y-axis range covers the weld and 50 pixels on each side, ensuring complete coverage of the anomaly area. For example, if the weld length range corresponding to the anomaly point in the gap is 100-120mm, after coordinate mapping, the corresponding image pixel range on the X-axis is 370-450, and the pixel range on the Y-axis is 40-140, clearly defining the image pixel area corresponding to the anomaly point.

[0031] Finally, feature response analysis is performed on the image pixel range. If the response intensity exceeds the preset feature judgment threshold, the location is determined as an abnormal feature point. Here, feature response analysis is performed by using a convolutional neural network (CNN) to extract hidden high-dimensional visual features (such as edge breakage, texture abrupt change, etc.) in the image. The response intensity is represented by the confidence score output by the model. The closer the value is to 1, the higher the probability that there is a gap anomaly in the region.

[0032] It should be noted that the model architecture consists of 3 convolutional layers and 2 fully connected layers. The input is a grayscale image of the image pixel range (normalized to the [0,1] interval), and the output is the feature response intensity (0-1 interval).

[0033] It should be noted that this model is constructed using a supervised training mode, and the training set labels are obtained based on expert annotation and confirmation from historical quality inspections. Over 1000 sets of weld image samples were extracted from a historical welding database, and each sample was labeled by senior welding engineers based on post-weld X-ray inspection or metallographic examination records: areas with confirmed gap abrupt changes or forming defects were marked as 1 (positive samples), and continuous, smooth, normal weld areas were marked as 0 (negative samples). This constructed a dataset with precise labels. The dataset was divided into training and validation sets in an 8:2 ratio. The activation function was ReLU, the loss function was cross-entropy, and the initial learning rate was 0.005, decreasing by 0.1 every 30 epochs until it reached 0.0005. Training stopped when the validation set loss fluctuation was less than 0.001 for 15 consecutive epochs.

[0034] The feature determination threshold was set to 0.75. Validated on the training set, this threshold can balance false positives and false negatives. The response intensity in abnormal areas generally exceeds 0.75, while that in normal areas is below 0.5.

[0035] For example, when the image pixel range corresponding to a certain abnormal point is input into the CNN model, the output response intensity is 0.82, which exceeds the threshold of 0.75. The center pixel coordinates (410, 90) of this area are mapped back to the weld space coordinates and determined to be the abnormal feature point location.

[0036] In step S104, if the gap width corresponding to the feature point location exceeds a preset process tolerance threshold, then based on the three-dimensional geometric boundary of the ultra-wide region where the feature point location is located, a filling sub-region is divided, the remaining space value of the filling sub-region is estimated, and an optimized path for the current layer is generated based on the remaining space value, including: If the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, then extract the cross-sectional point cloud data of the weld seam in the ultra-wide area where the feature point location is located. The cross-sectional point cloud data is converted into a three-dimensional raster model, and virtual slices are made according to a preset layer thickness to divide the filled sub-regions. Identify the void cells within the filled sub-region, sum the volumes of the void cells, and obtain the remaining space value; Multiple initial filling paths are generated based on the remaining space values. The deviation of each path is iteratively optimized. If the deviation is lower than a preset convergence threshold, the optimal path is extracted as the optimized path for the current layer.

[0037] It should be noted that, firstly, if the gap width corresponding to the feature point exceeds the preset process tolerance threshold, when extracting the cross-sectional point cloud data of the weld in the ultra-wide region where the feature point is located, the preset process tolerance threshold is determined based on the maximum bridging span that the surface tension of the molten pool can support in laser filler wire welding. When the weld gap width exceeds a certain multiple (usually 1.0 to 1.2 times) of the filler wire diameter, the molten metal is very likely to overcome the surface tension under gravity and collapse or fail to weld. Therefore, taking the commonly used 1.2mm diameter welding wire in industry as an example, the basic threshold is preset to 1.2mm as the critical point to distinguish between conventional single-pass filler and ultra-wide gaps that require multi-pass planning. Those skilled in the art know that this threshold can be dynamically adjusted according to the melt rheological characteristics of the welding material: for stainless steel materials with high molten pool viscosity and high surface tension of liquid metal, it can be appropriately relaxed to 1.4mm; while for aluminum alloy materials with extremely high fluidity in the molten state and a high tendency to collapse defects, it needs to be strictly tightened to 1.0mm to ensure the feasibility of the process.

[0038] When extracting cross-sectional point cloud data, the weld area covering 15mm before and after the feature point is simultaneously acquired using a laser sensor. After acquisition, statistical filtering is performed based on the three-standard-deviation principle to remove noise, retaining valid cross-sectional data and ensuring a complete representation of the geometry of the ultra-wide region. For example, if the gap width corresponding to a feature point is 1.8mm, exceeding the basic threshold of 1.2mm, the weld cross-sectional point cloud data within 15mm before and after that point is extracted. After noise removal, the data clearly presents the cross-sectional outline of the ultra-wide region.

[0039] Next, the cross-sectional point cloud data is converted into a 3D raster model. Virtual slicing is performed according to a preset layer thickness. When dividing the sub-regions for filling, the raster resolution of the 3D raster model is set to 0.1mm to accurately reproduce the geometric details of the cross-sectional point cloud. The preset layer thickness is consistent with the single-pass deposition layer thickness of the welding process; the base layer thickness is set to 0.8mm, which can be adjusted to 1.0mm for thick plate welding and reduced to 0.6mm for thin plate welding, ensuring that the virtual slices match the actual welding layer thickness. When dividing the sub-regions for filling, each slice is divided into non-overlapping sub-regions according to the geometric boundaries of the virtual slices. The width of each sub-region does not exceed 3mm, facilitating precise control of the filling path.

[0040] For example, after converting the cross-sectional point cloud into a 3D raster model, virtual slicing is performed with a layer thickness of 0.8mm. The ultra-wide area of ​​a certain slice is divided into 3 filled sub-regions, each with a width of 2.5mm, 2.8mm, and 2.6mm, respectively, with clear boundaries and no overlap.

[0041] It is worth noting that when identifying void cells within the filled sub-region and accumulating their volume, the criterion for determining a void cell is that the distance from the center of the grid cell to the planned filling path is greater than half the single-pass deposition width plus a 0.1mm safety margin. The single-pass deposition width is determined based on the welding wire diameter and welding current, with a base width set at 2.4mm, which can be fine-tuned according to process parameters. The specific fine-tuning rules follow the principle of heat input energy balance: if the welding current increases by 10A from the base value, the molten pool spreads, and the preset value of the single-pass deposition width needs to be increased by approximately 0.1mm; conversely, if the welding speed increases by 10%, the amount of deposited metal decreases, and the preset width needs to be decreased by 0.1mm.

[0042] It is important to note that the quantitative mapping relationship between the aforementioned fine-tuning step size and process parameters is not solely based on theoretical derivation, but rather stems from statistical regression analysis of a large amount of historical process experimental data. Specifically, the system pre-constructed an orthogonal experimental sample library (sample size > 500 groups) containing different welding currents, scanning speeds, and single-pass deposition widths. A multiple linear regression algorithm was used to fit the experimental data, extracting the sensitivity coefficients of each process factor to the weld width. Statistical results show a significant positive correlation between welding current and deposition width (Pearson correlation coefficient > 0.92), and the regression slope verifies the statistical law that every 10A current increment induces approximately 0.1mm of width expansion. Similarly, the regression characteristics of the speed factor verify its inverse contraction effect on width. This fitted model was tested on an independent validation set, and the prediction error was controlled within ±5%, ensuring that the volume calculation model matches the actual physical deposition process. Each grid cell has a volume of 0.001 mm³ (0.1 mm × 0.1 mm × 0.1 mm). By traversing all grid cells within the filled sub-region, counting the number of empty cells and accumulating the volumes, the remaining space value is obtained.

[0043] For example, if 42 void cells are identified within a certain filling sub-region, the accumulated remaining space value is 0.042 mm³, accurately reflecting the volume of material to be filled in that region. It is worth noting that this remaining space value is the theoretical filling volume and does not account for material shrinkage during welding. In actual path planning, a 5% to 10% volume margin can be reserved.

[0044] Finally, multiple initial filling paths are generated based on the remaining space values. The deviation of each path group is iteratively optimized. If the deviation is lower than a preset convergence threshold, the optimal path is extracted. When generating the optimized path for the current layer based on the remaining space values, 100 initial filling paths are generated. Each path group contains different starting positions and width parameters, with the width parameter ranging from 2.0 to 3.0 mm. In this implementation, the iterative optimization uses a genetic algorithm. The deviation is the absolute difference between the cumulative filling volume of each path group and the remaining space value. The convergence threshold is set to 0.0005 mm³, which has been tested multiple times to ensure the filling accuracy of the optimal path.

[0045] It is worth noting that the specific algorithm implementation process is as follows: First, a real-number encoding method is used to map the starting coordinate offset and width sequence of the path to chromosome genes, initializing a population containing 100 individuals. Next, a fitness function is constructed as the reciprocal of the bias, such that individuals whose filling volume is closer to the remaining space have a higher probability of being selected. Then, a roulette wheel selection operator is executed to screen for high-quality parents. For the selected individual pairs, a single-point crossover operation is performed with a probability of 0.7 to generate new path combinations, and a non-uniform mutation operation is performed on gene loci with a probability of 0.1 to escape local optima. Finally, the above evaluation, selection, crossover, and mutation steps are repeated until the bias of the best individual in the population is lower than the convergence threshold or the upper limit of 50 generations is reached.

[0046] For example, 100 initial filling paths are generated. After 50 iterations, the deviation of a certain path is 0.0003 mm³, which is lower than the convergence threshold. The starting point of this path is extracted as 2 mm to the right of the weld center, and the width sequence is 2.1 mm, 2.3 mm, and 2.2 mm, which are used as the optimized filling parameters for the current layer.

[0047] In step S105, based on the filling start point and width sequence of the optimized path, a solid contour model after filling the current layer is constructed, the gap areas not covered by the solid contour model are separated, and the coordinate set of the closed boundaries of the remaining gaps in the next layer is extracted, including: Based on the preset single-pass deposition cross-section model, and according to the filling start point and width sequence of the optimized path, a solid contour model after the current layer is filled is constructed. The difference between the entity contour model and the preset theoretical slice boundary of the current layer is calculated to separate the uncovered gap area after the current layer is filled. Extract the boundary coordinates of the uncovered gap regions, arrange them in clockwise order, and form the coordinate set of the closed boundaries of the remaining gaps in the next layer.

[0048] It should be noted that, firstly, when constructing the solid contour model after filling the current layer based on the preset single-pass deposition cross-section model and the optimized filling start point and width sequence, the single-pass deposition cross-section model adopts a semi-elliptical model, with the major axis being the single-pass filling width (taken from the optimized width sequence), and the ratio of the minor axis to the major axis set to 0.3. It is calibrated based on 100+ sets of welding experimental data to ensure consistency with the actual deposition morphology.

[0049] It is worth noting that the model training set includes single-pass deposition cross-section data under different welding wire diameters and welding currents. Each parameter group has 50+ samples collected, and the semi-ellipse parameters are determined using least squares fitting. After training, the model's prediction error is controlled within 0.05mm. During construction, semi-elliptical cross-sections are added sequentially according to the optimized fill start point, employing a maximum value fusion principle—that is, taking the largest height value in the overlapping area rather than simply accumulating it—to simulate the leveling effect of liquid metal. The overlap rate between adjacent passes is set to 30% to avoid gaps or excessive overlap, ultimately forming a complete solid contour model. For example, with an optimized fill start point 2mm to the right of the weld center and width sequences of 2.1mm, 2.3mm, and 2.2mm, after sequentially adding semi-elliptical cross-sections, the solid contour model clearly presents the three-dimensional shape after the current layer is filled, without obvious breaks.

[0050] In this embodiment, when performing a difference calculation between the solid contour model and the preset theoretical slice boundary of the current layer to separate the gap area not covered after the current layer is filled, the theoretical slice boundary of the current layer is a two-dimensional contour corresponding to the preset weld layer thickness (the layer thickness is consistent with the virtual slice layer thickness in S104, with a base of 0.8mm), which is generated by CAD software based on the weld design model. The difference calculation uses the Boolean difference algorithm to calculate the geometric difference between the solid contour model and the theoretical slice boundary, retaining the area not covered by the solid contour, i.e., the unfilled gap area.

[0051] It is important to emphasize that the coordinate systems of both must be unified before the calculation to ensure the accuracy of the difference result. The specific implementation of coordinate unification adopts a homogeneous transformation method based on feature alignment. First, the path planning origin and the design origin of the theoretical slice boundary are extracted from the solid contour model. Then, a transformation matrix containing rotation and translation components is constructed to map the local coordinate system of the solid contour model to the global design coordinate system of the theoretical slice, eliminating the relative pose deviation caused by the path planning process. The difference calculation is specifically implemented using the Vatti polygon clipping algorithm based on computational geometry. This algorithm first discretizes the boundary of the solid contour model into closed polygons and defines the boundary of the theoretical slice as the target polygon. Then, a Boolean subtraction operation is performed to calculate the geometric difference between the two, that is, to geometrically retain the region that belongs to the target polygon but not to the solid contour polygon. After the calculation, if scattered small regions appear, area threshold filtering is required. This threshold is set to 0.1 square millimeters, based on the fact that regions smaller than this area are usually caused by discretization errors or numerical calculation accuracy limitations, and physically cannot accommodate the smallest droplets with diameters usually greater than 0.5 millimeters, therefore they are judged as computational noise and discarded.

[0052] For example, in a certain area of ​​the weld edge, the solid profile model does not completely cover the theoretical slice boundary. After the difference calculation, a narrow and elongated uncovered gap area is separated out, with a width of about 0.5 mm and a length of about 5 mm, and the boundary is clear.

[0053] Finally, the boundary coordinates of the uncovered gap area are extracted and arranged clockwise to form a closed boundary coordinate set. The boundary coordinate extraction uses an edge detection algorithm, traversing the edges of the gap area after the interpolation calculation, and collecting continuous boundary point coordinates. The coordinates are normalized to the [0,1] interval, based on the overall weld size mapping. During the arrangement, the leftmost point of the gap area is used as the starting point, and the coordinates are sorted clockwise to ensure continuity and a closed profile, avoiding intersections or breaks. After sorting, the closure needs to be verified by calculating the distance between the first and last points. If the distance exceeds 0.01mm, intermediate points are added to ensure the integrity of the profile.

[0054] For example, a total of 120 boundary coordinates of the uncovered gap area were extracted. After being sorted clockwise, the distance between the first and last points was 0.005mm, forming a closed polygon coordinate set that accurately reflects the gap boundary that needs to be filled in the next layer.

[0055] In step S106, the coordinate set is discretized and scanned along a direction perpendicular to the filling path, the distance between boundary point pairs is calculated, the global gap width distribution is generated, and the shape of the remaining gap in the next layer is simulated, including: Along the direction perpendicular to the filling path, the coordinate set is discretized and scanned at a preset step distance. The distance between the intersection points of each scan line and the coordinate set is calculated to obtain the gap width of each scan position. Summarize all the gap widths to generate a global gap width distribution, and simulate the remaining gap shape of the next layer based on the gap width distribution.

[0056] It should be noted that, firstly, when discretizing the coordinate set of the closed boundary along a direction perpendicular to the filling path and at a preset step distance, the preset step distance is set according to the accuracy requirements of the gap width measurement. The basic step distance is set to 0.08mm. For scenarios with high welding accuracy requirements, such as welding aerospace parts, it can be adjusted down to 0.05mm, and for mass production scenarios with high real-time requirements, it can be adjusted up to 0.1mm. This step distance has been verified through multiple sets of tests to achieve a balance between accuracy and efficiency. The scanning direction is strictly perpendicular to the filling path. When the filling path is along the weld length direction, the scanning direction is the weld width direction. Starting from one edge of the closed boundary, scanning line by line to the other edge ensures coverage of the entire gap area.

[0057] Next, the distance between the intersection points of each scan line and the closed boundary is calculated to obtain the gap width at each scan position. The intersection point pair identification adopts the ray method. Each scan line traverses the coordinates of the closed boundary from left to right and records the two intersection points with the boundary (the left intersection point and the right intersection point). If only one intersection point is identified or there is no intersection point, it is determined to be an invalid scan line and discarded. The distance is calculated using Euclidean distance, and the calculated Euclidean distance is the gap width.

[0058] Finally, all gap widths are aggregated to generate a global gap width distribution. When simulating the remaining gap shape in the next layer, all effective gap widths are aggregated in scan line order to form a global gap width dataset. This dataset includes scan position coordinates and corresponding gap widths. The simulation of the remaining gap shape uses cubic spline curve fitting technology, with the scan position as the abscissa and the gap width as the ordinate, to generate a continuous gap width variation curve. This curve is combined with the coordinates of the closed boundary to construct a three-dimensional model of the remaining gap shape.

[0059] For example, by summarizing the gap width of 1000 effective scan lines, the data shows a trend of gradually changing from 0.5mm to 1.2mm and then narrowing to 0.6mm. After three spline fittings, the remaining gap of the next layer is simulated to be a narrow and elongated shape that is wide in the middle and narrow at both ends, clearly reflecting the spatial shape that needs to be filled.

[0060] In step S107, the width data of the remaining gap shape is traversed. If there is a narrow region smaller than the welding wire diameter data, the width sequence is adjusted to obtain the fill path parameters, including: Traverse the gap width distribution; if there are positions with values ​​smaller than the welding wire diameter data, mark them as narrow regions and record the coordinates of the narrow regions. Based on the coordinates of the narrow area, backtrack to locate the corresponding filling path segment of the current layer; Calculate the difference between the width of the narrow region and the diameter of the welding wire, combine it with the spatial distribution range of the narrow region, generate a width correction increment, and superimpose it on the initial width of the filling path segment to obtain a corrected width sequence; The local entity outline is reconstructed based on the corrected width sequence. If the local entity outline shows no narrow area in the next layer, the corrected width sequence is used as the fill path parameter.

[0061] It should be noted that, firstly, when traversing the gap width distribution, if there are locations with values ​​smaller than the welding wire diameter, these are marked as narrow regions and their coordinates are recorded. The traversal is performed point-by-point in the order of discretized scanning, directly comparing the gap width at each scan location with the welding wire diameter data. The welding wire diameter data has already been retrieved from S101. During marking, not only are the three-dimensional coordinates of the location recorded (mapped to the global weld coordinate system), but the start and end points of the continuous narrow regions are also marked to form region range information. For example, if the welding wire diameter is 1.2mm, and during the traversal, a segment of five consecutive scan points has gap widths of 0.8mm, 0.9mm, 0.7mm, 1.0mm, and 0.9mm, all less than 1.2mm, the coordinate range of this region is marked as (x1, y1, z1) to (x5, y5, z5), thus identifying it as a narrow region.

[0062] Next, when tracing back to the filling path segment corresponding to the current layer based on the coordinates of the narrow area, the tracing relies on the spatial index relationship between the filling path and the weld coordinates constructed in S104. The index table records the coverage coordinate range of each filling path segment. By comparing the coordinates of the narrow area with the coverage range of the path segment, the specific path segment leading to the narrow area can be accurately located.

[0063] If multiple path segments have a combined impact, they are sorted by the percentage of the affected area, with the path segment having the highest percentage being adjusted first. For example, the coordinates of the aforementioned narrow area were backtracked and found to be mainly covered by the third filling path segment of the current layer. This path segment has an initial width of 2.0 mm and covers an area of ​​10-15 mm in the weld length direction, and is therefore determined to be a critical path segment that needs adjustment.

[0064] It should be noted that when generating the width correction increment based on the difference between the width of the narrow area and the welding wire diameter, combined with the spatial distribution range of the narrow area, a weighted summation method is used. The weight of the difference is set to 0.6, and the weight of the spatial range is set to 0.4. The difference is the difference between the welding wire diameter and the minimum width of the narrow area, and the spatial range is the length of the narrow area (along the weld direction). The larger the difference and the wider the range, the larger the correction increment. The upper limit of the increment is set to 0.5mm to avoid over-adjustment that could lead to fill overflow.

[0065] It should be further explained that the width difference is the core geometric constraint for determining whether the welding wire can physically enter the gap, and directly determines the physical feasibility of filling. Therefore, it is given a high weight as the dominant factor. The spatial range represents the continuous length and defect scale of the narrow area along the weld direction, reflecting the overall urgency of the correction requirement. It is given a second-highest weight as an auxiliary factor. The combination of the two can take into account both local geometric accessibility and overall process stability.

[0066] For example, suppose the welding wire diameter is 1.2 mm, the minimum width of the narrow area is 0.7 mm (difference of 0.5 mm), and the length is 5 mm. The correction increment is calculated as follows: 0.5 × 0.6 + 5 × 0.02 × 0.4 = 0.34 (mm). In this formula, the first term compensates for the width difference, and the second term compensates for the narrow space. The coefficient 0.02 originates from the fluid dynamics friction resistance principle, meaning that for every 1 mm increase in gap length, an additional 0.02 mm of width compensation is needed to overcome flow resistance. Finally, the 0.34 mm increment is added to the initial width of 2.0 mm, resulting in a corrected width of 2.34 mm.

[0067] Subsequently, the local entity outline is reconstructed and the remaining gap in the next layer is verified. If no narrow area is verified, the corrected local width sequence is locked. When updating the filling path parameters, the reconstructed local entity outline adopts the semi-elliptical single-pass deposition section model in S105, only replacing the corrected local width sequence, keeping the pass overlap rate unchanged at 30%.

[0068] It should be noted that during verification, the reconstructed contour is virtually scanned using the discretized scanning method of S106. If the gap width of all scan points is greater than the diameter of the welding wire, the verification is considered successful. If narrow areas still exist, the correction increment is repeatedly adjusted (increased or decreased by 0.05 mm each time) until the verification is successful.

[0069] For example, after reconstructing the local entity outline of the third path segment, the virtual scan found that the gap width of the original narrow area reached 1.3mm, which is greater than the welding wire diameter of 1.2mm. The verification was passed, and the corrected width of the path segment of 2.34mm was locked and updated to the global fill path parameters to obtain the final fill path parameters.

[0070] In step S108, an adaptive welding trajectory for the current layer is generated based on the filling path parameters, and a friendliness score is calculated. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme, including: Based on the filling path parameters and the welding torch posture constraints, an adaptive welding trajectory for the current layer is generated. The adaptive welding trajectory is mapped to a preset thermo-mechanical coupling model to predict the deposition surface morphology data; Extract the smoothness features of the deposition surface morphology data to identify regions of abrupt elevation changes and points of fusion risk; The friendliness score is calculated by weighting the distribution density of the elevation change region and the fusion risk point. If the friendliness score is lower than the preset friendliness threshold, the filling path parameters are adjusted and the score is recalculated until the score meets the requirements, thus obtaining the final multi-layer path planning scheme.

[0071] It should be noted that, firstly, when generating the adaptive welding trajectory of the current layer based on the fill path parameters in conjunction with the welding torch posture constraints, the welding torch posture constraints include the tilt angle range (±10°) and the rotation angle limit (±5°). These constraints are set based on the mechanical performance and process requirements of the welding equipment to avoid the posture exceeding the equipment's capabilities, which would lead to welding instability.

[0072] The fill path parameters include the starting coordinates, width sequence, and path direction. B-spline curve technology is used for trajectory generation to ensure a continuous and smooth trajectory, with the transition curvature between adjacent path segments not exceeding 0.5 rad / m to avoid abrupt changes in welding torch movement. During generation, the trajectory coverage is adjusted segment by segment according to the width sequence; the larger the width, the wider the trajectory coverage, and the overlap rate between adjacent trajectories is maintained at 30% to ensure dense filling.

[0073] For example, the starting point of the fill path parameters is 2mm to the right of the weld center, and the width sequence is 2.34mm, 2.2mm, and 2.1mm. Combined with the constraint of the welding torch tilt angle of 8°, the generated adaptive welding trajectory extends along the weld length direction, and the width of each segment matches the parameters, resulting in a smooth transition without any stuttering.

[0074] It should be noted that when mapping the adaptive welding trajectory to the preset thermo-mechanical coupling model to predict the deposition surface morphology data, the thermo-mechanical coupling model uses the Transient Thermal and StaticStructural modules of ANSYS Workbench for coupling. The material parameters in the model are set according to the actual welding material (such as the thermal conductivity of stainless steel is 16.3 W / (m·K) and the specific heat capacity is 502 J / (kg·K)).

[0075] In this embodiment, the model training set contains over 1000 sets of deposition surface data corresponding to different welding parameters (current, speed, path width). Each parameter set has 50+ samples. The model parameters are calibrated using least squares fitting, and the prediction error after training is controlled within 0.05mm. During mapping, the coordinates and width of the trajectory are input into the model to simulate heat conduction and material deposition during the welding process. The output is three-dimensional point cloud data of the deposited surface, normalized to the [0,1] interval, mapped based on the overall height range of the weld. For example, after inputting a certain adaptive welding trajectory into the model, the predicted deposition surface point cloud data shows that the height of the weld center region is approximately 0.65mm, and the edge region is approximately 0.5mm, exhibiting a gently transitioning morphology.

[0076] Subsequently, the smoothness characteristics of the depositional surface morphology data were extracted to identify regions of abrupt height changes and fusion risk points. The smoothness characteristics were obtained by calculating the height variance. The depositional surface was divided into 5mm × 5mm grids, and the standard deviation of the height values ​​within each grid was calculated. A larger standard deviation indicates poorer smoothness. The threshold for identifying regions of abrupt height changes was set at 0.18mm / mm; that is, a region was marked as an abrupt change if the height change rate between adjacent grids exceeded this value. Fusion risk points were grid cells with a variance greater than 0.10mm located in the path overlap area. These areas are prone to non-fusion due to concentrated heat or insufficient filling.

[0077] For example, the edge grid height variance of a certain deposition surface is 0.15 mm, and the height change rate of adjacent grids is 0.22 mm / mm, which is marked as a region of abrupt height difference; the variance of the three grids in the path overlap area is 0.12 mm, which is determined to be a fusion risk point. A total of 12 regions of abrupt height difference and 5 fusion risk points were identified.

[0078] Based on the distribution density of elevation change areas and fusion risk points, the friendliness score is calculated using a weighted average. The distribution density is the proportion of abnormal areas (change areas and risk points) to the total number of grid cells. The weight allocation is 0.6 for elevation change areas and 0.4 for fusion risk points. Elevation change areas directly disrupt the global smoothness of the deposition surface, severely hindering the stable deposition of subsequent layers and having a more significant impact on the forming quality; therefore, they are given a higher weight. Fusion risk points mainly induce local metallurgical defects, with a relatively limited impact range, and are therefore given a lower weight. The score ranges from 0 to 100, with lower density resulting in a higher score. The calculation first normalizes the distribution density to the [0,1] interval, then weights the distribution density to obtain the deduction ratio, multiplies it by 100 to obtain the deduction value, and subtracts the deduction value from 100 to obtain the friendliness score.

[0079] For example, with a total of 400 grid cells, 17 abnormal regions were identified, including 12 regions with abrupt elevation changes and 5 fusion risk points. First, the distribution density was calculated: the density of regions with abrupt elevation changes was 12 / 400 = 0.03, and the density of fusion risk points was 5 / 400 = 0.0125. Next, normalization was performed (based on precision welding process requirements, a density tolerance upper limit of 0.1 was set, i.e., the normalization value was 1 when the density reached 0.1): after normalization, the density of abrupt elevation changes was 0.03 / 0.1 = 0.3, and the density of fusion risk points was 0.0125 / 0.1 = 0.125. Finally, the weighted deduction value was calculated as (0.3 × 0.6 + 0.125 × 0.4) × 100 = (0.18 + 0.05) × 100 = 23 points. The friendliness score was 100 - 23 = 77 points. The score (77 points) is lower than the preset threshold (e.g., 80 points), indicating that although the absolute number of abnormal areas is not large, their distribution density has already posed a risk to the quality of multilayer deposition, and the system needs to automatically trigger the optimization and adjustment of the filling path parameters.

[0080] If the friendliness score is lower than the preset friendliness threshold, the filling path parameters are adjusted until the score meets the requirements, resulting in the final multi-layer path planning scheme. The friendliness threshold is determined based on the mapping relationship between interlayer surface morphology and multi-layer welding mechanical properties. Statistical studies show a significant negative correlation between the smoothness of the deposited surface (i.e., friendliness) and the incidence of interlayer non-fusion defects. Therefore, the basic threshold is set at 80 points, corresponding to an interlayer defect rate controlled within 0.05%. For high-precision scenarios such as aerospace and energy equipment subjected to high-cycle fatigue loads, to ensure fatigue life, the defect rate needs to be reduced to below 0.01%, hence the threshold is increased to 85 points. For mass-produced ordinary structural parts, only static strength requirements need to be met, and the defect rate tolerance is relatively high (e.g., 0.1%); to improve production cycle time, the threshold can be lowered to 75 points. This threshold system has been verified through multiple batches of welding metallographic experiments, accurately balancing welding quality reliability and production efficiency.

[0081] Finally, when adjusting the fill path parameters, for areas with abrupt changes in elevation (manifested as local bulges or depressions), the corresponding path width is adjusted first (increased or decreased by 0.05-0.1 mm). This utilizes the volume compensation principle to change the local metal deposition amount, achieving peak smoothing and valley filling. For fusion risk points (manifested as deep V-shaped grooves at the overlap), the starting position of the path is adjusted first, offsetting it by 0.1-0.2 mm along the weld direction. This utilizes the phase shift effect to change the overlap phase of adjacent weld layers, preventing weld joints from overlapping and accumulating in the vertical direction. The adjustment process is achieved through gradient descent. After adjusting the parameters, the trajectory is regenerated and the score is calculated. The maximum number of iterations is set to 5 to avoid infinite loops.

[0082] For example, if the initial friendliness score is 71 points, which is lower than the basic threshold of 80 points, and the main deduction item is identified as a sudden change in local elevation difference, the system will automatically adjust the width of the corresponding path segment from 2.34mm to 2.4mm to increase the filling amount. After recalculation, the score is 83 points, which meets the requirements. The parameter is locked, and the final multi-layer path planning scheme is obtained by combining the path planning of all layers.

[0083] In summary, this invention discloses an adaptive path planning method for laser welding, comprising: acquiring 3D point cloud of the weld, visual image, and welding wire diameter data; identifying weld edge features after denoising; generating a continuously changing curve of the gap width; locating abnormal gap feature points; dividing the gap into filling sub-regions if the gap exceeds the process tolerance; estimating the remaining space to obtain optimized filling parameters; constructing a solid contour model to simulate the shape of the remaining gap in the next layer; adjusting the filling sequence to avoid narrow areas; generating an adaptive welding trajectory; calculating the friendliness score of subsequent layers; iteratively updating parameters until the target is met, thus obtaining a multi-layer path planning scheme. This method achieves high-precision adaptive planning under dynamic gap changes, meeting the requirements for high-consistency welding.

[0084] Reference Figure 2 The second embodiment of the present invention provides a laser welding adaptive path planning system, comprising: The data acquisition module is used to acquire 3D point cloud data of the weld area, visual images of the weld, and welding wire diameter data; The curve generation module is used to denoise the three-dimensional point cloud data, identify the edge feature points of the weld, and generate a continuous curve of the gap width based on the spacing of the edge feature points. The feature localization module is used to identify abnormal points in the gap from the continuously changing curve, and to establish the correspondence between the abnormal points and the image pixel area by combining the weld visual image, thereby determining the location of the abnormal feature points. The path optimization module is used to divide the filling sub-region according to the three-dimensional geometric boundary of the ultra-wide region where the feature point is located if the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, estimate the remaining space value of the filling sub-region, and generate the optimized path of the current layer according to the remaining space value. The coordinate extraction module is used to construct the entity contour model after the current layer is filled based on the filling start point and width sequence of the optimized path, separate the gap area not covered by the entity contour model, and extract the coordinate set of the closed boundary of the remaining gap in the next layer. The shape simulation module is used to discretize the coordinate set along a direction perpendicular to the filling path, calculate the distance between boundary point pairs, generate the global gap width distribution, and simulate the shape of the remaining gap in the next layer. The parameter correction module is used to traverse the width data of the remaining gap shape. If there is a narrow area smaller than the welding wire diameter data, the width sequence is adjusted to obtain the filling path parameters. The update and optimization module is used to generate an adaptive welding trajectory for the current layer based on the filling path parameters and calculate a friendliness score. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme.

[0085] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0086] The specific embodiments described above further illustrate the purpose, technical solution, 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 laser welding adaptive path planning method, characterized in that, include: Acquire 3D point cloud data of the weld area, visual images of the weld, and welding wire diameter data; The three-dimensional point cloud data is denoised, and the edge feature points of the weld are identified. A continuous variation curve of the gap width is generated based on the spacing of the edge feature points. Identify abnormal points in the gap from the continuously changing curve, establish the correspondence between the abnormal points and the image pixel area by combining the weld visual image, and determine the location of the abnormal feature points. If the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, then according to the three-dimensional geometric boundary of the ultra-wide region where the feature point location is located, a filling sub-region is divided, the remaining space value of the filling sub-region is estimated, and the optimized path of the current layer is generated based on the remaining space value. Based on the filling start point and width sequence of the optimized path, construct the solid contour model after filling the current layer, separate the gap areas not covered by the solid contour model, and extract the coordinate set of the closed boundary of the remaining gap in the next layer. Discretize the coordinate set along the direction perpendicular to the filling path, calculate the distance between boundary point pairs, generate the global gap width distribution, and simulate the shape of the remaining gap in the next layer. Traverse the width data of the remaining gap shape. If there is a narrow area smaller than the welding wire diameter data, adjust the width sequence to obtain the fill path parameters. Based on the filling path parameters, an adaptive welding trajectory for the current layer is generated, and a friendliness score is calculated. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme.

2. The laser welding adaptive path planning method according to claim 1, characterized in that, The acquisition of the three-dimensional point cloud data of the weld area, the visual image of the weld, and the diameter data of the welding wire includes: The weld area is scanned by a laser sensor, the laser echo signal is received and converted into three-dimensional point cloud data; Simultaneously acquire visual images of the weld area as weld visual images, and align the timestamps of the weld visual images with the timestamps of the three-dimensional point cloud data; Retrieve pre-stored welding wire diameter data, which is matched with welding process parameters.

3. The laser welding adaptive path planning method according to claim 1, characterized in that, The step of denoising the three-dimensional point cloud data and identifying the edge feature points of the weld, and generating a continuous variation curve of the gap width based on the spacing of the edge feature points, includes: The 3D point cloud data is denoised to extract a set of effective depth information. Identify gradient abrupt change locations in the depth information set to determine the edge feature points of the weld; Calculate the Euclidean distance between the edge feature points to obtain discrete gap width values; The gap width value is fitted to generate a continuous variation curve of the gap width along the weld direction.

4. The laser welding adaptive path planning method according to claim 1, characterized in that, The process of identifying abnormal points in the gap from the continuously changing curve, establishing a correspondence between the abnormal points and image pixel regions based on the weld visual image, and determining the location of abnormal feature points includes: Traverse the continuous change curve. If the rate of change of the gap width at a certain point exceeds the preset abrupt change judgment threshold, it is marked as an abnormal point of the gap. Establish a spatial index between the continuous change curve and the visual image of the weld, and clarify the image pixel range corresponding to the abnormal point; The image pixel range is subjected to feature response analysis. If the response intensity exceeds a preset feature determination threshold, it is determined to be an abnormal feature point location.

5. The laser welding adaptive path planning method according to claim 1, characterized in that, If the gap width corresponding to the feature point location exceeds a preset process tolerance threshold, then based on the three-dimensional geometric boundary of the ultra-wide region where the feature point location is located, a filling sub-region is divided, the remaining space value of the filling sub-region is estimated, and an optimized path for the current layer is generated based on the remaining space value, including: If the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, then extract the cross-sectional point cloud data of the weld seam in the ultra-wide area where the feature point location is located. The cross-sectional point cloud data is converted into a three-dimensional raster model, and virtual slices are made according to a preset layer thickness to divide the filled sub-regions. Identify the void cells within the filled sub-region, sum the volumes of the void cells, and obtain the remaining space value; Multiple initial filling paths are generated based on the remaining space values. The deviation of each path is iteratively optimized. If the deviation is lower than a preset convergence threshold, the optimal path is extracted as the optimized path for the current layer.

6. The laser welding adaptive path planning method according to claim 1, characterized in that, Based on the fill start point and width sequence of the optimized path, a solid contour model after filling the current layer is constructed. The gap areas not covered by the solid contour model are separated, and the coordinate set of the closed boundaries of the remaining gaps in the next layer is extracted, including: Based on the preset single-pass deposition cross-section model, and according to the filling start point and width sequence of the optimized path, a solid contour model after the current layer is filled is constructed. The difference between the entity contour model and the preset theoretical slice boundary of the current layer is calculated to separate the uncovered gap area after the current layer is filled. Extract the boundary coordinates of the uncovered gap regions, arrange them in clockwise order, and form the coordinate set of the closed boundaries of the remaining gaps in the next layer.

7. The laser welding adaptive path planning method according to claim 1, characterized in that, The discretization scan of the coordinate set along a direction perpendicular to the filling path, calculation of the distance between boundary point pairs, generation of the global gap width distribution, and simulation of the remaining gap shape in the next layer include: Along the direction perpendicular to the filling path, the coordinate set is discretized and scanned at a preset step distance. The distance between the intersection points of each scan line and the coordinate set is calculated to obtain the gap width of each scan position. Summarize all the gap widths to generate a global gap width distribution, and simulate the remaining gap shape of the next layer based on the gap width distribution.

8. The laser welding adaptive path planning method according to claim 7, characterized in that, If, during the traversal of the width data of the remaining gap shape, there exists a narrow region smaller than the welding wire diameter data, the width sequence is adjusted to obtain the fill path parameters, including: Traverse the gap width distribution; if there are positions with values ​​smaller than the welding wire diameter data, mark them as narrow regions and record the coordinates of the narrow regions. Based on the coordinates of the narrow area, backtrack to locate the corresponding filling path segment of the current layer; Calculate the difference between the width of the narrow region and the diameter of the welding wire, combine it with the spatial distribution range of the narrow region, generate a width correction increment, and superimpose it on the initial width of the filling path segment to obtain a corrected width sequence; The local entity outline is reconstructed based on the corrected width sequence. If the local entity outline shows no narrow area in the next layer, the corrected width sequence is used as the fill path parameter.

9. The laser welding adaptive path planning method according to claim 1, characterized in that, The process involves generating an adaptive welding trajectory for the current layer based on the filling path parameters and calculating a friendliness score. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme. This includes: Based on the filling path parameters and the welding torch posture constraints, an adaptive welding trajectory for the current layer is generated. The adaptive welding trajectory is mapped to a preset thermo-mechanical coupling model to predict the deposition surface morphology data; Extract the smoothness features of the deposition surface morphology data to identify regions of abrupt elevation changes and points of fusion risk; The friendliness score is calculated by weighting the distribution density of the elevation change region and the fusion risk point. If the friendliness score is lower than the preset friendliness threshold, the filling path parameters are adjusted and the score is recalculated until the score meets the requirements, thus obtaining the final multi-layer path planning scheme.

10. A laser welding adaptive path planning system, characterized in that, include: The data acquisition module is used to acquire 3D point cloud data of the weld area, visual images of the weld, and welding wire diameter data; The curve generation module is used to denoise the three-dimensional point cloud data, identify the edge feature points of the weld, and generate a continuous curve of the gap width based on the spacing of the edge feature points. The feature localization module is used to identify abnormal points in the gap from the continuously changing curve, and to establish the correspondence between the abnormal points and the image pixel area by combining the weld visual image, thereby determining the location of the abnormal feature points. The path optimization module is used to divide the filling sub-region according to the three-dimensional geometric boundary of the ultra-wide region where the feature point is located if the gap width corresponding to the feature point location exceeds the preset process tolerance threshold, estimate the remaining space value of the filling sub-region, and generate the optimized path of the current layer according to the remaining space value. The coordinate extraction module is used to construct the solid contour model after the current layer is filled based on the filling start point and width sequence of the optimized path, separate the gap area not covered by the solid contour model, and extract the coordinate set of the closed boundary of the remaining gap in the next layer. The shape simulation module is used to discretize the coordinate set along a direction perpendicular to the filling path, calculate the distance between boundary point pairs, generate the global gap width distribution, and simulate the shape of the remaining gap in the next layer. The parameter correction module is used to traverse the width data of the remaining gap shape. If there is a narrow area smaller than the welding wire diameter data, the width sequence is adjusted to obtain the filling path parameters. The update and optimization module is used to generate an adaptive welding trajectory for the current layer based on the filling path parameters and calculate a friendliness score. If the friendliness score is lower than a preset friendliness threshold, the filling path parameters are updated and the friendliness score is recalculated until the friendliness score meets the requirements, thus obtaining the final multi-layer path planning scheme.

Images