Space constraint-based carriage side plate cutting path dynamic planning method and system

Abstract

The application provides a space constraint-based carriage side plate cutting path dynamic programming method and system, comprising the following steps: performing laser scanning on a carriage side plate to obtain carriage side plate point cloud data, and constructing a three-dimensional model of the carriage side plate based on the carriage side plate point cloud data; presetting a cutting head envelope space by using the three-dimensional model of the carriage side plate, and establishing a cutting head occupied space constraint; constructing a cutting path programming model based on the three-dimensional model of the carriage side plate and the cutting head occupied space constraint; and cutting the carriage side plate based on the cutting path programming model. The technical scheme of the application can significantly improve cutting precision, cutting efficiency and cutting quality, enhance the safety of the cutting process, and adapt to various complex carriage side plate structures.

Description

Technical Field

[0001] This invention belongs to the field of path planning technology, specifically relating to a dynamic planning method and system for cutting paths of carriage side panels based on spatial constraints. Background Technology

[0002] Traditional automated cutting path planning methods are often based on simplified two-dimensional models or fixed templates, which cannot accurately reflect the actual three-dimensional shape and structure of the carriage side panels. This may lead to collisions between the cutting head and the carriage side panels, or the inability to accurately cut the required shape, thus affecting cutting quality and production efficiency. For complex carriage side panel structures, such as side panels with reinforcing ribs, connecting components, and other areas that cannot be cut, traditional cutting path planning methods struggle to effectively identify and avoid these areas, easily causing cutting errors or damage to the carriage side panels.

[0003] During the cutting process, the movement range and posture changes of the cutting head need to be constrained to avoid collisions with the side panels of the vehicle body or other components. Pre-setting the cutting head envelope space and establishing space constraints for the cutting head can effectively limit its movement range, ensuring the safety and reliability of the cutting process. Traditional cutting head envelope spaces are usually set based on experience or simplified models, which cannot accurately reflect the dynamic changes of the cutting head during actual movement. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method and system for dynamic planning of cutting paths for carriage side panels based on spatial constraints, which avoids collisions between the cutting head and the carriage side panels and further improves cutting accuracy.

[0005] A dynamic programming method for cutting paths of carriage side panels based on spatial constraints includes: Laser scanning is performed on the side panels of the carriage to obtain point cloud data of the side panels, and a three-dimensional model of the side panels is constructed based on the point cloud data of the side panels. Using the 3D model of the carriage side panel, the envelope space of the cutting head is preset, and the space constraint of the cutting head is established. Based on the three-dimensional model of the carriage side panel and the space constraints of the cutting head, a cutting path planning model is constructed. Based on the cutting path planning model, the cutting of the carriage side panel is completed.

[0006] Preferably, the method for constructing the three-dimensional model of the carriage side panel includes: Dynamic laser scanning is performed on the side panels of the carriage, and multi-view registration is performed on the point cloud data of the carriage side panels acquired from different angles to obtain registered point cloud data. The registered point cloud data is subjected to adaptive filtering and denoising to obtain denoised point cloud data; The denoised point cloud data is segmented into multi-scale regions by growing and marking indivisible regions and key inflection points to establish an obstacle model for the side panel of the carriage. Based on the obstacle model of the carriage side panel, hierarchical voxel modeling is performed on the carriage side panel to obtain the three-dimensional model of the carriage side panel.

[0007] Preferred methods for establishing obstacle models of the carriage side panels include: Calculate the Gaussian curvature and local point density of each point in the denoised point cloud data; Using preset curvature thresholds and preset density thresholds, the denoised point cloud data is divided into flat region seed points and concave-convex feature seed points. The seed points of the flat region are merged and the seed points of the concave and convex features are segmented using a multi-scale region growing and segmentation strategy to obtain point cloud data after multi-scale region growing and segmentation. Based on the point cloud data after multi-scale region growing and segmentation, geometric feature matching of the carriage side panel is performed to identify the inscissible regions, and a topological map of the carriage side panel components is established, with semantic marking of the inscissible regions. Based on the topology map of the carriage side panel components, the component intersection points are identified, and key inflection point detection is performed by combining the point cloud data after multi-scale region growing and segmentation. An obstacle model for the side panel of the carriage is established based on the indivisible area and key inflection points.

[0008] Preferably, the method for establishing space constraints for the cutting head includes: A local coordinate system for the cutting head is established based on the length, maximum sway angle, and radial safety distance of the cutting head. Based on the local coordinate system of the cutting head, the sweep body cross section of the cutting head in each posture is calculated; Based on the swept body section, the minimum cut head envelope space in the attitude space is generated using the convex hull algorithm. Store the envelope space of all the minimum cutting heads and construct a dynamic constraint database; Based on the obstacle model and the dynamic constraint database, a dynamic interference detection model is established; Based on the dynamic interference detection model, a space constraint for the cutting head is established.

[0009] Preferably, the method for calculating the swept body cross-section of the cutting head in each posture includes: Calculate the cross-sectional circle of the cutting head using the cutting head radius and radial safety distance, and obtain the radius of the basic cross-section. Obtain the current pitch angle, yaw angle, and velocity of the cutting head, and calculate the major axis of the rotated ellipse based on the radius of the base section. The extension length of the cutting head along the direction of motion is obtained by using convolution kernel dilation; The sweep body section is calculated by taking the radius of the basic section as the minor axis and the major axis of the ellipse as the major axis, and outputting the extension length.

[0010] Preferred methods for establishing dynamic interferometric detection models include: Complexity analysis is performed on the obstacle model, and the local complexity is evaluated by calculating the curvature change of the obstacle surface; A depth layer is projected along the main axis of the cutting head, and the depth layer is dynamically updated based on the local complexity assessment results and the movement trajectory of the cutting head. The dynamic margin is calculated using the cutting head's movement speed and acceleration; wherein, the dynamic margin represents the maximum positional deviation of the cutting head during its movement. By introducing a probabilistic interference judgment method, and combining the dynamic margin, sensor accuracy, and updated depth layer, the probability of interference occurring at each sampling point is calculated, thus completing the construction of the dynamic interference detection model.

[0011] Preferably, the methods for constructing a cutting path planning model include: Based on the obstacle model, an obstacle repulsion potential field and a key inflection point guiding potential field are generated to construct a composite potential field. Gradient sampling is performed on the composite potential field to generate sparse nodes, and adaptive path search based on the composite potential field is performed to obtain the initial cutting path. Based on the preset optimization objectives, an evolutionary optimization model for Pareto frontier perception is established; The initial path is optimized based on the Pareto front-aware evolutionary optimization model to obtain the final cutting path, thus completing the construction of the cutting path planning model.

[0012] Preferably, the preset optimization targets include cutting path length, safety margin, and inflection point deviation.

[0013] This invention also provides a dynamic planning system for cutting paths of carriage side panels based on spatial constraints, used to implement the method, including: The 3D scanning module is used to perform laser scanning on the side panel of the carriage to obtain point cloud data of the side panel, and to construct a 3D model of the side panel based on the point cloud data of the side panel. The spatial constraint module is used to establish space constraints for the cutting head by using the 3D model of the side panel of the carriage and presetting the envelope space of the cutting head. The path planning module is used to construct a cutting path planning model based on the three-dimensional model of the carriage side panel and the space constraints occupied by the cutting head; The side panel cutting module is used to cut the side panels of the carriage based on the cutting path planning model.

[0014] Preferably, the three-dimensional scanning module includes: The point cloud registration unit is used to perform dynamic laser scanning on the side panel of the carriage and to perform multi-view registration on the point cloud data of the side panel of the carriage from different angles to obtain registered point cloud data. A point cloud denoising unit is used to perform adaptive filtering and denoising on the registered point cloud data to obtain denoised point cloud data. The obstacle recognition unit is used to perform multi-scale region growth segmentation on the denoised point cloud data, mark the indivisible region and key inflection point, and establish an obstacle model of the carriage side panel. The voxel modeling unit is used to perform hierarchical voxel modeling on the side panel of the carriage based on the obstacle model of the side panel, so as to obtain the three-dimensional model of the side panel of the carriage.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: By acquiring point cloud data of the carriage side panel through laser scanning and constructing a high-precision 3D model based on this data, the cutting path planning can more accurately reflect the actual shape and size of the carriage side panel. This allows the cutting head to cut along a precise path, thereby improving cutting accuracy and reducing cutting errors. Pre-setting the cutting head envelope space using the 3D model of the carriage side panel can more precisely limit the movement range and attitude changes of the cutting head, avoiding collisions between the cutting head and the carriage side panel, further improving cutting accuracy. The cutting path planning model constructed based on the 3D model of the carriage side panel and the space constraints occupied by the cutting head can automatically generate the optimal cutting path. Compared with traditional manual planning methods, this method can significantly reduce planning time and adjustment time during the cutting process, thereby improving cutting efficiency. Through precise cutting path planning, the cutting head can complete the cutting task more efficiently, reducing unnecessary movement and pauses, further improving cutting efficiency. Pre-setting the cutting head envelope space and establishing space constraints for the cutting head can effectively limit the movement range of the cutting head, avoiding collisions between the cutting head and the carriage side panel or other components, thereby enhancing the safety of the cutting process. By precisely planning the cutting path, the cutting head can move within a safe range, reducing dangerous factors during the cutting process and further improving the safety of the cutting process. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a dynamic planning method for cutting paths of carriage side panels based on spatial constraints, according to an embodiment of the present invention. Detailed Implementation

[0018] 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.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Example 1

[0020] like Figure 1 As shown, the dynamic programming method for cutting the side panel of the carriage based on spatial constraints includes: S1: Perform laser scanning on the side panel of the carriage to obtain point cloud data of the side panel, and construct a three-dimensional model of the side panel based on the point cloud data of the side panel.

[0021] A further implementation method involves constructing a three-dimensional model of the carriage side panel, including: S11: Perform dynamic laser scanning on the side panels of the carriage and register the point cloud data from different angles to obtain registered point cloud data. Specifically, low-resolution scanning (5mm point pitch) is used for flat areas to improve efficiency; high-resolution scanning (1mm point pitch) is used for complex areas (reinforcing ribs, weld protrusions) to ensure feature integrity. The scan data from different angles are fused using the ICP (Iterative Closest Point) algorithm to eliminate blind spots.

[0022] S12: Adaptive filtering and denoising are performed on the registered point cloud data to obtain denoised point cloud data. In this embodiment, statistical outlier removal (SOR) is used to filter out floating noise points. For reflective metal areas, a curvature-based smoothing algorithm is used to avoid erroneously removing real features. Normal vector clustering is used to identify planes and cylindrical surfaces (such as rivet heads). Edge enhancement algorithms are used for welds and uneven structures to improve the accuracy of subsequent segmentation.

[0023] S13: Perform multi-scale region growing and segmentation on the denoised point cloud data, mark indivisible regions and key inflection points, and establish a vehicle side panel obstacle model. A further implementation method for establishing the vehicle side panel obstacle model includes: S131: Calculate the Gaussian curvature and local point density of each point in the denoised point cloud data. Specifically, for each point in the point cloud, its Gaussian curvature reflects the degree of local curvature of the surface at that point. A k-nearest neighbor is constructed for the point cloud to obtain a neighborhood point set; the value of k is increased in flat regions and decreased in high-curvature regions. Then, PCA is used to fit a local quadratic surface to the neighborhood point set, and the first and second fundamental form functions of the surface are calculated. The principal curvature is obtained through the eigenvalues ​​of the Venn-Katen matrix. The local density of each point in the point cloud reflects the density of points in its neighborhood. First, the point cloud kernel density is estimated, an octree is constructed to accelerate neighborhood queries, the number of points in the spherical neighborhood of each point is calculated, and the point cloud kernel density is normalized to obtain the local point density.

[0024] S132: Using preset curvature and density thresholds, the denoised point cloud data is divided into flat region seed points and concave / convex feature seed points. In this embodiment, a curvature-density dual-threshold seed classification method is proposed. By calculating the normalized curvature entropy value of each point, a dynamic classification threshold between curvature and local point density is calculated, achieving intelligent selection of seed points. Specifically, points with low curvature and high density are selected as flat region seed points, and points with high curvature and low density are selected as concave / convex feature seed points. Non-maximum suppression is used to avoid excessive seed point density.

[0025] S133: A multi-scale region growing segmentation strategy is used to merge seed points in flat regions and segment seed points for concave and convex features to obtain point cloud data after multi-scale region growing segmentation. In this embodiment, a specific multi-scale region growing segmentation strategy is provided: Large-scale growth (10mm neighborhood): Used for quickly merging flat areas. The growth conditions are: normal vector angle <15° and point spacing <8mm.

[0026] Small-scale growth (3mm neighborhood): used for fine segmentation of concave and convex features. The growth conditions are: the angle between the normal vectors is <5° and the distance between points is <3mm.

[0027] Specifically, this embodiment also includes dynamic adjustment of the growth threshold. A region complexity evaluation function is constructed using a curvature weighting factor, the normalized average curvature of a preset region, a density weighting factor, and the local point density variation coefficient of the preset region. A comprehensive complexity score for the preset region is calculated, and this score is used to divide complex and simple regions, allowing for the reduction or expansion of neighborhoods. Hole repair based on Delaunay triangulation is performed on the segmented regions; Laplacian sharpening is applied to the edge points of concave and convex features to enhance feature continuity.

[0028] S134: Based on the point cloud data after multi-scale region growing and segmentation, geometric feature matching of the carriage side panel is performed to identify indivisible regions and establish a topological map of the carriage side panel components, semantically labeling the indivisible regions. In this embodiment, the Laplacian shrinkage algorithm is used to extract the topological skeleton for each segmented region and calculate the persistent coherence features of the skeleton. Using the region principal direction vector and the density distribution of concentric rings, an asymmetric descriptor is constructed at the geometric feature points. Using the persistent coherence features of the skeleton and the asymmetric descriptor, macroscopic skeleton extraction and microscopic texture analysis of the carriage side panel are realized respectively, simulating the "whole-part" cognitive mechanism of human vision. The obtained microscopic and macroscopic features are positively matched using a standard template library of carriage side panel components, and physical infeasibility matching is eliminated using kinematic constraints to complete bidirectional matching verification, realizing the biomimetic feature matching of the carriage side panel. In this embodiment, a graph neural network with embedded physical rules is also introduced. The carriage side panel components identified after feature matching are used as initial nodes, and hidden components are supplemented through functional necessity analysis to construct derived nodes. Based on contact analysis, physical connection edges between nodes are established, and functional association edges between nodes are established using knowledge graph-driven logical reasoning to complete the construction of the topology diagram of the carriage side panel component.

[0029] Using the biomimetic feature matching results of the carriage side panels, i.e., the geometric feature matching results, and the topological context information obtained from the topological diagram of the carriage side panel components, semantic labeling of indivisible regions is performed: For areas that cannot be cut: Match marks (such as reinforcing ribs, weld points, connecting parts, etc.) using predefined templates; For areas that can be cut but need to be avoided: assign them a higher weight and prioritize detours during path planning (such as around drainage ditches).

[0030] S135: Based on the topology map of the carriage side panel components, identify the intersection points of the components, and combine the point cloud data after multi-scale regional growth and segmentation to detect key inflection points; specifically, during path planning, set a protection radius for key inflection points (such as drainage channels and structural connection points) to ensure that the cutting path does not damage their structural integrity; construct a spherical protection zone (radius 30-50mm) centered on the inflection point; prohibit path crossing within the protection zone, and use Bézier curves to detour instead.

[0031] S136: Based on the indivisible region and key inflection points, establish an obstacle model for the side panel of the carriage.

[0032] S14: Based on the obstacle model of the carriage side panel, hierarchical voxel modeling is performed on the carriage side panel to obtain a 3D model of the carriage side panel. Specifically, coarse-grained voxels are used for large flat areas to reduce computational load, storing only surface normal vectors; fine-grained voxels are used for densely obstacle-prone areas to accurately represent the concave and convex structures, storing SDF (symbolic distance field) values. The truncated symbolic distance field TSDF is constructed using the distance from the voxel to the nearest obstacle and the radius of the cutting head. , In the formula, This represents the distance of voxel v to the nearest obstacle. Set it to twice the radius of the cutting head.

[0033] S2: Using the 3D model of the carriage side panel, a pre-defined envelope space for the cutting head is established, and space constraints for the cutting head are created. A further implementation method for establishing space constraints for the cutting head includes: S21: Establish a local coordinate system for the cutting head based on its length, maximum sway angle, and radial safety distance; S22: Based on the local coordinate system of the cutting head, calculate the sweep body section of the cutting head in each posture; where the sweep body section refers to the two-dimensional projection profile of the space swept by the cutting head along the direction of motion in a specific posture on a certain plane. It reflects the instantaneous workspace occupancy of the cutting head in that posture, and its included elements are geometry, safety margin and dynamic changes. The geometry is determined by the shape of the cutting head (such as cylindrical or conical) and its orientation (yaw angle, pitch angle).

[0034] The safety margin includes a radial safety distance (a buffer zone to prevent collisions).

[0035] Dynamic changes refer to real-time updates that occur as the cutting head moves.

[0036] Specifically, for each possible attitude, the pitch and yaw angles of the cutting head are calculated using the cutting head coordinate system to obtain the swept body cross-section. In this embodiment, for each point to be cut, its maximum intrusion depth along the cutting head's movement direction is calculated (the maximum deviation depth of the cutting head along the movement direction that may exceed the expected path, determined by factors such as motion control error and mechanical vibration). A safety exclusion zone is generated by convolution kernel expansion, and a cutting head swept body layer is superimposed on the TSDF. Incremental TSDF updates are used, and when a new obstacle is detected, only the affected voxels are locally updated.

[0037] This embodiment provides a specific method for calculating the cross-section of the swept body: Calculate the cross-sectional circle of the cutting head using the cutting head radius and safety distance to obtain the basic cross-sectional radius; Obtain the current pitch angle, yaw angle, and velocity of the cutting head. Combined with the radius of the base section, calculate the major axis of the ellipse after rotation. Use convolution kernel expansion to obtain the extension length of the cutting head along the direction of motion. Finally, use the radius of the base section as the minor axis of the swept body section and the major axis of the ellipse as the major axis of the swept body section, and output the extension length.

[0038] S23: Based on the swept body cross-section, a minimum cutting head envelope space within the attitude space is generated using the convex hull algorithm. Specifically, the extracted cross-section is shape optimized to more accurately reflect the actual working range of the cutting head. An energy-minimization-based shape adjustment algorithm is used. Specifically, an energy function is defined, which includes geometric constraints on the cross-section (keeping the cross-section area constant) and smoothness constraints on adjacent cross-sections (shape changes between adjacent cross-sections should be smooth). By minimizing this energy function, the shape of the cross-section is adjusted. This avoids inaccurate calculation of the cutting head envelope space due to cross-section extraction errors. For example, if noise causes sharp local changes in the cross-section shape during extraction, energy minimization adjustment can make the cross-section shape smoother and more consistent with the actual cutting head shape.

[0039] This paper combines traditional 2D and 3D convex hull algorithms. First, the convex hull of each section is calculated using a 2D convex hull algorithm on the 2D plane containing the section. Then, these 2D convex hulls are combined in 3D space, and a 3D convex hull algorithm is used to generate the minimum cutter head envelope space within the entire pose space. During the 3D convex hull generation process, the spatial relationships and pose changes between sections are considered. When the convex hulls of two adjacent sections intersect in 3D space, the shape of the 3D convex hull is adjusted according to their pose and positional relationships to accurately contain the convex hulls of all sections. A time-dimensional convex hull algorithm is introduced. For the pose changes of the cutter head at different time points, time is treated as an additional dimension to construct a 4D convex hull. This allows for a more comprehensive consideration of the dynamic behavior of the cutter head.

[0040] Machine learning algorithms are used to optimize the convex hull algorithm. A large amount of cutting head pose data and corresponding cross-sectional data are collected as training samples. A neural network model is trained, which learns the complex mapping relationship between cross-sectional data and the shape of the convex hull. In practical applications, new cross-sectional data is input into the model, and the model can quickly predict the shape of the convex hull. This method can improve the efficiency and accuracy of the convex hull algorithm.

[0041] S24: Store the envelope space of all minimum cutting heads and build a dynamic constraint database; specifically, pre-calculate the envelope space under typical poses and store it as an octree lookup table, and quickly obtain the current envelope size through pose interpolation during real-time planning.

[0042] S25: Establish a dynamic interference detection model based on obstacle models and dynamic constraint databases; A further implementation method involves establishing a dynamic interferometric detection model, including: Complexity analysis is performed on the obstacle model, and the local complexity is evaluated by calculating the curvature change of the obstacle surface; A depth layer is projected along the main axis of the cutting head, and the depth layer is dynamically updated based on the local complexity assessment results and the movement trajectory of the cutting head. The dynamic margin is calculated using the cutting head's movement speed and acceleration; wherein, the dynamic margin represents the maximum positional deviation of the cutting head during its movement. By introducing a probabilistic interference judgment method, and combining the dynamic margin, sensor accuracy, and updated depth layer, the probability of interference occurring at each sampling point is calculated, thus completing the construction of the dynamic interference detection model.

[0043] Specifically, a depth layer is projected along the main axis of the cutting head, with each layer recording the maximum intrusion depth and surface normal vector. The interlayer spacing is adaptively adjusted according to the cutting head radius. First, a local complexity analysis is performed on the obstacle model. Local complexity is assessed by calculating the curvature changes of the obstacle surface. In areas with large curvature changes (such as obstacle edges and grooves), the interlayer spacing is reduced to more accurately record the maximum intrusion depth and surface normal vector; in areas with small curvature changes (such as flat parts of the obstacle), the interlayer spacing is appropriately increased to reduce computational load. The depth layer is dynamically updated as the cutting head moves. Utilizing dynamic changes in the obstacle model (such as obstacle movement and deformation), the recorded content of the depth layer is adjusted in real time.

[0044] By utilizing the dynamic margin of the cutting head's motion velocity, real-time interference judgment is performed on each sampling point of the cutting head's swept body. First, the dynamic margin is calculated based on the cutting head's motion velocity and acceleration. The dynamic margin represents the maximum possible positional deviation of the cutting head during its motion. This dynamic margin is taken into account when performing interference judgment, expanding the detection range of the cutting head's swept body. A probabilistic interference judgment method is introduced. Considering the uncertainty of the cutting head's motion and sensor measurement errors, a probability value is assigned to the interference judgment result. By establishing a probabilistic model and combining the cutting head's motion parameters, sensor accuracy, and the uncertainty of the depth layer, the probability of interference occurring at each sampling point is calculated.

[0045] S26: Based on the dynamic interference detection model, establish the space constraint of the cutting head.

[0046] S3: Based on the 3D model of the carriage side panel and the space constraints of the cutting head, a cutting path planning model is constructed. A further implementation method for constructing the cutting path planning model includes: S31: Based on the obstacle model, generate an obstacle repulsion potential field and a key inflection point guiding potential field to construct a composite potential field. The repulsion potential field is generated using the SDF value of the obstacle. A guiding potential energy function is defined, which is related to the shape and location of the potential energy trap. The guiding potential field is established using the potential energy traps set within the key inflection point protection zone. Using preset weighting coefficients for each potential field, the repulsion and guiding potential fields are weighted and summed to construct the composite potential field. The determination of the weighting coefficients considers the effect of the obstacle repulsion potential field at a large scale to avoid collisions between the cutting head and the obstacle; at a small scale, the effect of the key inflection point guiding potential field is mainly considered to accurately guide the cutting head to avoid the key inflection point. Multi-scale fusion improves the adaptability and robustness of the composite potential field.

[0047] Specifically, the formula for the repulsive potential field of the obstacle is as follows: , In the formula, It is dynamically adjusted according to the cutting head speed v. This represents the distance from point q to the nearest obstacle. Represents the dynamic influence radius. Represents the radius of the cutting head. represents the repulsive potential field strength coefficient, and k represents the velocity influence coefficient.

[0048] The formula for calculating the potential field at the key inflection point is as follows: , In the formula, a softening factor is introduced. Avoiding singularities Represents the coordinates of the i-th critical inflection point. This represents the inflection point attraction weight.

[0049] S32: Gradient sampling is performed on the composite potential field to generate sparse nodes. Adaptive path search based on the composite potential field is then performed to obtain an initial cutting path. Specifically, candidate nodes are generated along the gradient direction, and the sampling step size is adaptively adjusted according to the rate of change of the potential field. In this embodiment, when node expansion gets trapped in a local minimum, a tangential perturbation potential field is temporarily added to achieve dynamic redirection. Specifically, the gradient calculation formula is as follows: , In the formula, h represents the difference step size, and (x,y) represents the planar coordinate components of point q.

[0050] Design heuristic functions: , In the formula, Represents the coordinates of the current node. Represents the coordinates of the target point. Represents the potential field weight.

[0051] A bidirectional search strategy is implemented using heuristic functions. Starting from the origin, low-potential nodes are expanded first, and backtracking is performed from the destination to check path safety. When the potential difference between the bidirectional search paths meets a preset threshold, the paths merge. When a change in the potential field is detected (such as the addition of a new obstacle), the potential field model is locally updated, and a cone-shaped region re-search is initiated for the affected path segments.

[0052] S33: Based on the preset optimization objectives, establish an evolutionary optimization model for Pareto front perception; a further implementation method is that the preset optimization objectives include cutting path length (minimize the total cutting distance), safety margin (maximize the minimum distance between the path and obstacles), and inflection point deviation (minimize the intrusion depth of the path into the protection zone of key inflection points).

[0053] In this embodiment, the specific method for establishing a Pareto front-aware evolutionary optimization model includes: partitioning multiple optimization objectives, generating an initial population, optimizing evolutionary operations, and extracting and refining the Pareto front. Specifically, the optimization objectives are divided into core objectives and auxiliary objectives, a dynamic weight vector is constructed, and the objective space is hierarchically layered. The grid resolution is automatically adjusted according to the distribution density of the front solutions, and non-dominated solutions plus the most diverse optimal solutions are retained within each grid. A crowding entropy index is introduced to retain the final elite solutions. The Adaptive Gridded Objective Space (AGOS) method is used to solve the front degradation problem in high-dimensional objective optimization.

[0054] This paper proposes a hybrid evolutionary strategy that integrates Genetic Expression Programming (GEP) and Differential Evolution (DE), overcoming the limitations of traditional algorithms through a multi-level collaborative optimization mechanism. Specifically, in the initial stage of the population, a heterogeneous individual generation strategy is adopted. 30% of the initial individuals are directly encoded from safe path segments generated from sparse nodes, ensuring the preservation of the geometric features of the original point cloud; the remaining 70% of individuals construct symbolic expression trees through GEP, which encode engineering rules into evolvable logic (e.g., "IF (safety margin > 3mm) THEN (prioritize shortening the path)"), achieving knowledge-driven initial population construction. Compared with traditional random initialization, this strategy improves the quality of the initial population by more than 60%.

[0055] In terms of mutation operations, a geometric awareness mechanism is innovatively introduced. Directional mutation guided by potential field gradients is applied to path nodes. The learning rate in the mutation formula adopts an adaptive adjustment strategy: a smaller value is used in regions with drastic changes in the potential field (such as near obstacles) to ensure stability, while a larger value is used in flat regions to accelerate exploration. Simultaneously, Gaussian noise conforming to the N(0,1) distribution is superimposed to maintain the directionality of gradient guidance while avoiding getting trapped in local optima.

[0056] Cross-operation optimization proposes a topology-preserving constraint. Individual similarity is measured by calculating path edit distance, allowing only parent individuals with similar connection structures to exchange segments. In the specific implementation, the path is first encoded with topological features (such as inflection point sequences and curvature distribution histograms), and then the dynamic time warping (DTW) algorithm is used to calculate the edit distance. When the similarity exceeds a threshold (usually 0.7), sub-path segments are exchanged while maintaining topological consistency.

[0057] After each generation of evolution, a refined local search is initiated on the Pareto front solution. First, a constraint graph of the path nodes is established to identify key locations such as curvature abrupt changes and safety thresholds. Then, the projected gradient method is used for constraint optimization to minimize the path length while ensuring a safe distance, thus completing the construction of an evolutionary optimization model based on Pareto front awareness.

[0058] S34: Optimize the initial path using an evolutionary optimization model based on Pareto front awareness to obtain the final cutting path, completing the construction of the cutting path planning model. In this embodiment, it also includes constructing a path knowledge graph containing historical cases to perform multi-dimensional evaluation of the newly generated Pareto solution set: predicting cutting quality scores using convolutional neural networks, estimating thermal deformation based on finite element analysis, and recommending the optimal feed rate using a process database. Finally, 3-5 optimal solutions are output, with quantitative indicators for each solution in terms of safety, efficiency, and cost.

[0059] S4: Based on the cutting path planning model, complete the cutting of the carriage side panel. Example 2

[0060] This invention also provides a dynamic planning system for cutting paths of carriage side panels based on spatial constraints, and a method for implementing it, including: The 3D scanning module is used to perform laser scanning on the side panels of the carriage to obtain point cloud data of the side panels, and to construct a 3D model of the side panels based on the point cloud data of the side panels. The spatial constraint module is used to establish space constraints for the cutting head by using the 3D model of the side panel of the carriage and presetting the envelope space of the cutting head. The path planning module is used to construct a cutting path planning model based on the 3D model of the carriage side panel and the space constraints occupied by the cutting head; The side panel cutting module is used to cut the side panels of the carriage based on the cutting path planning model.

[0061] A further embodiment of the invention includes a 3D scanning module comprising: The point cloud registration unit is used to perform dynamic laser scanning on the side panel of the carriage and to perform multi-view registration on the point cloud data of the side panel of the carriage from different angles to obtain registered point cloud data. The point cloud denoising unit is used to perform adaptive filtering and denoising on the registered point cloud data to obtain denoised point cloud data. The obstacle recognition unit is used to perform multi-scale region growth and segmentation on the denoised point cloud data, mark the indivisible areas and key inflection points, and establish an obstacle model for the side panel of the carriage. The voxel modeling unit is used to perform hierarchical voxel modeling on the side panel of the carriage based on the obstacle model of the side panel, so as to obtain the three-dimensional model of the side panel of the carriage.

[0062] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A dynamic programming method for cutting paths of carriage side panels based on spatial constraints, characterized in that, include: Laser scanning is performed on the side panels of the carriage to obtain point cloud data of the side panels, and a three-dimensional model of the side panels is constructed based on the point cloud data of the side panels. Using the 3D model of the carriage side panel, the envelope space of the cutting head is preset, and the space constraint of the cutting head is established. Based on the three-dimensional model of the carriage side panel and the space constraints of the cutting head, a cutting path planning model is constructed. Based on the cutting path planning model, the cutting of the carriage side panel is completed; The method for constructing the three-dimensional model of the carriage side panel includes: Dynamic laser scanning is performed on the side panels of the carriage, and multi-view registration is performed on the point cloud data of the carriage side panels acquired from different angles to obtain registered point cloud data. The registered point cloud data is subjected to adaptive filtering and denoising to obtain denoised point cloud data; The denoised point cloud data is segmented into multi-scale regions by growing and marking indivisible regions and key inflection points to establish an obstacle model for the side panel of the carriage. Based on the obstacle model of the carriage side panel, hierarchical voxel modeling is performed on the carriage side panel to obtain the three-dimensional model of the carriage side panel. Methods for establishing obstacle models for the side panels of the carriage include: Calculate the Gaussian curvature and local point density of each point in the denoised point cloud data; Using preset curvature thresholds and preset density thresholds, the denoised point cloud data is divided into flat region seed points and concave-convex feature seed points. The seed points of the flat region are merged and the seed points of the concave and convex features are segmented using a multi-scale region growing and segmentation strategy to obtain point cloud data after multi-scale region growing and segmentation. Based on the point cloud data after multi-scale region growing and segmentation, geometric feature matching of the carriage side panel is performed to identify the inscissible regions, and a topological map of the carriage side panel components is established, with semantic marking of the inscissible regions. Based on the topology map of the carriage side panel components, the component intersection points are identified, and key inflection point detection is performed by combining the point cloud data after multi-scale region growing and segmentation. Based on the indivisible areas and key inflection points, an obstacle model for the side panel of the carriage is established. Methods for establishing dynamic interferometric detection models include: Complexity analysis is performed on the obstacle model, and the local complexity is evaluated by calculating the curvature change of the obstacle surface; A depth layer is projected along the main axis of the cutting head, and the depth layer is dynamically updated based on the local complexity assessment results and the movement trajectory of the cutting head. The dynamic margin is calculated using the cutting head's movement speed and acceleration; wherein, the dynamic margin represents the maximum positional deviation of the cutting head during its movement. By introducing a probabilistic interference judgment method, and combining the dynamic margin, sensor accuracy, and updated depth layer, the probability of interference occurring at each sampling point is calculated, thus completing the construction of the dynamic interference detection model.

2. The dynamic programming method for cutting paths of carriage side panels based on spatial constraints according to claim 1, characterized in that, Methods for establishing space constraints for the cutting head include: A local coordinate system for the cutting head is established based on the length, maximum sway angle, and radial safety distance of the cutting head. Based on the local coordinate system of the cutting head, the sweep body cross section of the cutting head in each posture is calculated; Based on the swept body section, the minimum cut head envelope space in the attitude space is generated using the convex hull algorithm. Store the envelope space of all the minimum cutting heads and construct a dynamic constraint database; Based on the obstacle model and the dynamic constraint database, a dynamic interference detection model is established; Based on the dynamic interference detection model, a space constraint for the cutting head is established.

3. The dynamic programming method for cutting paths of carriage side panels based on spatial constraints according to claim 2, characterized in that, Methods for calculating the swept body cross-section of the cutting head at each orientation include: Calculate the cross-sectional circle of the cutting head using the cutting head radius and radial safety distance, and obtain the radius of the basic cross-section. Obtain the current pitch angle, yaw angle, and velocity of the cutting head, and calculate the major axis of the rotated ellipse based on the radius of the base section. The extension length of the cutting head along the direction of motion is obtained by using convolution kernel dilation; The sweep body section is calculated by taking the radius of the basic section as the minor axis and the major axis of the ellipse as the major axis, and outputting the extension length.

4. The dynamic programming method for cutting paths of carriage side panels based on spatial constraints according to claim 1, characterized in that, Methods for constructing a cutting path planning model include: Based on the obstacle model, an obstacle repulsion potential field and a key inflection point guiding potential field are generated to construct a composite potential field. Gradient sampling is performed on the composite potential field to generate sparse nodes, and adaptive path search based on the composite potential field is performed to obtain the initial cutting path. Based on the preset optimization objectives, an evolutionary optimization model for Pareto frontier perception is established; The initial cutting path is optimized based on the Pareto front-aware evolutionary optimization model to obtain the final cutting path, thus completing the construction of the cutting path planning model.

5. The dynamic programming method for cutting paths of carriage side panels based on spatial constraints according to claim 4, characterized in that, The preset optimization targets include cutting path length, safety margin, and inflection point deviation.

6. A dynamic planning system for cutting paths of carriage side panels based on spatial constraints, used to implement the method described in any one of claims 1-5, characterized in that, include: The 3D scanning module is used to perform laser scanning on the side panel of the carriage to obtain point cloud data of the side panel, and to construct a 3D model of the side panel based on the point cloud data of the side panel. The spatial constraint module is used to establish space constraints for the cutting head by using the 3D model of the side panel of the carriage and presetting the envelope space of the cutting head. The path planning module is used to construct a cutting path planning model based on the three-dimensional model of the carriage side panel and the space constraints occupied by the cutting head; The side panel cutting module is used to cut the side panels of the carriage based on the cutting path planning model.

7. The dynamic planning system for cutting paths of carriage side panels based on spatial constraints according to claim 6, characterized in that, The three-dimensional scanning module includes: The point cloud registration unit is used to perform dynamic laser scanning on the side panel of the carriage and to perform multi-view registration on the point cloud data of the side panel of the carriage from different angles to obtain registered point cloud data. A point cloud denoising unit is used to perform adaptive filtering and denoising on the registered point cloud data to obtain denoised point cloud data. The obstacle recognition unit is used to perform multi-scale region growth segmentation on the denoised point cloud data, mark the indivisible region and key inflection point, and establish an obstacle model of the carriage side panel. The voxel modeling unit is used to perform hierarchical voxel modeling on the side panel of the carriage based on the obstacle model of the side panel, so as to obtain the three-dimensional model of the side panel of the carriage.

Images