Scanning path planning method for full coverage of large-scale structural member

Abstract

The invention belongs to the technical field of three-dimensional scanning path planning and industrial automatic detection, and particularly relates to a full-coverage scanning path planning method for a large complex structural part. The revolutionary optimization of the three-dimensional scanning path of the large structural member is realized by constructing a comprehensive technical system of curvature self-adaptive sampling, pose collaborative optimization, a three-stage Weibull motion strategy, real-time collision detection and whole-process efficiency improvement. According to the scheme, geometric feature perception, motion continuity guarantee, algorithm adaptability enhancement and operation safety control are creatively integrated, a technical breakthrough is formed in three dimensions of non-blind area coverage, high motion efficiency and zero collision risk, and meanwhile, through parameterization design and a cross-platform compatible framework, the operation safety is improved. Deployment flexibility and system robustness in an industrial scene are remarkably improved, and a complete solution with high precision, high efficiency and high reliability is provided for automatic detection of a complex curved surface.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional scanning path planning and industrial automation inspection technology, specifically involving a full-coverage scanning path planning method for large and complex structural components (including ship sections, spacecraft components, and automotive body panels). Background Technology

[0002] Large structural components (such as automobile hoods, ship curved sections, and aerospace fairings) present the following common industry challenges in automated 3D scanning due to their large size, complex surfaces, and heterogeneous features:

[0003] Insufficient Coverage: Traditional uniform sampling methods cannot adapt to the geometric features of high-curvature regions (such as edges and ridges), resulting in frequent scanning blind spots. Existing technologies ignore dynamic changes in curvature, leading to insufficient sampling density in areas of abrupt surface changes, with measured coverage rates far below industrial inspection standards. Low Motion Efficiency: Mainstream path planning methods decouple equipment position and posture for optimization, resulting in discontinuous robotic arm trajectories. Frequent posture adjustments during scanning significantly increase idle travel time, resulting in poor path smoothness and overall scanning efficiency failing to meet production line cycle time requirements. Significant Safety Risks: Multi-site scanning of large components relies on manual experience for planning, leading to a high risk of equipment collisions. Traditional collision detection algorithms have high computational complexity and cannot meet real-time path correction requirements, especially in confined spaces. Algorithm Adaptability Defects: Existing swarm intelligence optimization algorithms (such as genetic algorithms) are prone to getting trapped in local optima at a scale of hundreds of viewpoints, and their convergence speed cannot meet engineering timeliness requirements. More importantly, the optimization of position distance and posture continuity is fragmented, resulting in large fluctuations in coverage rates for complex curved surfaces.

[0004] Limitations of existing scanning planning techniques: In the viewpoint generation stage: uniform sampling ignores surface feature differences, resulting in insufficient sampling in high-curvature regions; the scanner's field of view (FOV) and working distance (DOF) are not strictly constrained, leading to invalid viewpoints; viewpoint pose is not optimized synchronously with position, affecting scanning effectiveness. In the path optimization stage: position distance and pose continuity are optimized independently, resulting in poor path smoothness; traditional algorithms have weak exploration capabilities and suffer from severe premature convergence; collision detection and pose continuity constraints are not integrated into the core optimization framework, resulting in poor real-time performance.

[0005] Therefore, this invention overcomes the above-mentioned challenges by using curvature dynamic segmentation, visual cone pose modeling, a three-stage Weibull motion strategy, and pose-aware local optimization, providing a full-coverage, high-efficiency, and robust scanning solution for large structural components. Summary of the Invention

[0006] In view of this, this invention proposes a full-coverage scanning path planning method for large structural components such as ship sections, spacecraft parts, and automotive body panels. This method solves the problems of insufficient coverage, discontinuous motion, and equipment collisions in scanning complex curved surfaces through two core technologies: adaptive viewpoint generation based on curvature features and pose-coordinated path optimization. The technical solution includes the following steps: First, an adaptive viewpoint set is generated based on the curvature features of the point cloud to ensure blind-spot-free surface coverage; second, the device position trajectory and scanning attitude are simultaneously optimized through an improved marine predator algorithm; finally, a safe path is output by integrating real-time collision detection and pose continuity constraint mechanisms. This solution overcomes the limitations of traditional methods that optimize position and attitude separately, achieving a synergistic improvement in scanning efficiency and safety, and providing a full-coverage, high-efficiency, and robust scanning solution for large structural components.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A scanning path planning method for full coverage of large structural components includes the following steps:

[0009] Digital Discretization of Freeform Surfaces

[0010] Firstly, the curvature-based sampling discretization method analyzes the curvature distribution of free-form surfaces in the CAD model, dividing high-curvature regions (such as sharp edges and complex surfaces) into feature-sensitive areas and setting high sampling density, while low-curvature flat areas are sampled sparsely. After generating an initial point set using Poisson disk sampling or parametric mesh, redundant points are removed by voxel filtering and the normal vector direction is corrected. Finally, a point cloud model that balances feature accuracy and data efficiency is output, thus completing the generation of the corresponding point cloud model for the CAD model of the free-form surface of a large structural component.

[0011] Dynamic curvature feature sampling

[0012] The point cloud model generated in the previous step is initially uniformly sampled, and principal component analysis (PCA) is used to accelerate neighborhood search by KD-Tree to calculate local curvature feature values. The calculation results are then normalized to the [0,1] interval. Based on the characteristics of large structural components, a dynamic curvature threshold is determined, and the curvature gradient of each point is calculated. Points with gradient values ​​greater than twice the threshold are marked as gradient abrupt change regions, which determines the chord height threshold. Finally, all high curvature points are retained, and flat points are sampled a second time to retain some points. Midpoints (new sampling points) are inserted between adjacent gradient abrupt change points to enhance feature continuity.

[0013] Scan pose physical constraint modeling

[0014] Based on the scanner parameters and constraints, including field of view constraints, working distance constraints, maximum tilt angle constraints, etc., a visible cone boundary is constructed. Then, the sampling point is offset along its normal vector direction, and the offset length is the optimal working distance (the sum of the minimum working distance and half the depth of field) to obtain the position of the viewpoint. Finally, the direction vector pointing from the viewpoint to the corresponding sampling point is defined as the orientation of the scanner at this viewpoint position, and it is used as the Z-axis reference direction of the scanner coordinate system to ensure that the scanner always points accurately to the target surface.

[0015] Spatial viewpoint simplification and optimization

[0016] To address the issue of redundant viewpoints, a K-means clustering algorithm is employed to compress the initial viewpoint set. This algorithm dynamically divides cluster centers based on the spatial distribution of viewpoints, merges neighboring viewpoints, and simultaneously evaluates the contribution of each viewpoint using view frustum coverage analysis. For viewpoints whose overlapping coverage exceeds a threshold, those with minimal impact on global coverage are preferentially deleted; for areas with weak coverage, new viewpoints are added through interpolation. Finally, while ensuring full model coverage, the number of viewpoints is optimized to generate the final optimized viewpoint pose information, with the viewpoint set being V = {v1, v2, ... v...}. n}

[0017] Modeling of pose collaboration costs

[0018] Using the viewpoint set obtained in the previous step, calculate the viewpoint position distance matrix and attitude difference matrix; divide each element in the position distance matrix by the global maximum distance to map it to the [0,1] interval. Then, fuse the attitude differences by performing a linear transformation on the attitude matrix based on the minimum / maximum angle in the attitude difference matrix, normalizing the attitude differences as follows:

[0019]

[0020] Where: θ min θ max θ is the minimum and maximum angle between the two sides in the attitude difference matrix. ij Let d be the angle between the direction vectors of viewpoints i and j. o It is the normalized result corresponding to the angle between the direction vectors of the two viewpoints i and j.

[0021] A weighted linear combination method is used to fuse the normalized position distance and attitude difference into a comprehensive distance matrix, as shown in the formula:

[0022]

[0023] Where: d max It is the maximum distance in the position distance matrix, d ij Let be the Euclidean distance between viewpoints i and j, r be the position weight coefficient, and d be the distance between viewpoints i and j. cIt is the combined distance between viewpoints i and j.

[0024] Improved marine predator algorithm

[0025] A hybrid initialization strategy, combining the advantages of greedy algorithms and random generation, provides high-quality and diverse initial solutions for the ocean predator algorithm.

[0026] The Weibull distribution is used to replace Brownian motion to simulate the random movement of prey in the early stage of iteration. By adjusting the shape parameter k and the scale parameter λ, a variety of movement patterns are generated to enhance the algorithm's exploration ability and flexibly adapt to complex search spaces. In order to improve computational efficiency while ensuring the randomness of prey movement, the following comprehensive Weibull distribution is adopted.

[0027] R W =αf s +βf m +λf l (3)

[0028] Where α, β, and λ are the scaling factors for short-range, medium-range, and long-range searches, respectively, f s f m and f l The Weibull distributions for short-range, medium-range, and long-range searches are set for different scale parameters and shape parameters.

[0029] The adaptive step size factor follows the idea of ​​inertial weights and introduces a nonlinear adjustment method on this basis. By introducing a sine function and a control factor, the flexibility and adaptability of the ocean predator algorithm are increased.

[0030]

[0031] Where, ω max and ω min These remain the maximum and minimum values ​​of the inertia weight, where t is the current iteration number, and Max... iter The maximum number of iterations is denoted by α, which is a control factor. The value of α controls the shape of the sine function, thereby affecting the rate and pattern of change of the adaptive step size factor.

[0032] An improvement on the traditional 2-OPT algorithm is made by introducing attitude continuity constraints to optimize the path. When the attitude continuity value falls below a set threshold, a penalty factor is applied to the gain (i.e., cost reduction), thereby reducing the priority of the swap operation. When optimizing the path length, the algorithm tends to select swap operations that will not cause abrupt attitude changes (i.e., excessively large angles between direction vectors), thus shortening the path while maintaining the smoothness of attitude changes.

[0033] Finally, a two-level collision detection strategy is used, which includes spatial accelerated structure detection based on the separating axis theorem and spatial relationship detection and determination based on adaptive sampling and nearest neighbor query. The safety distance should be set to be greater than the scanner radius (or half width) to ensure that the scanner does not collide with the workpiece during movement.

[0034] Using the improved marine predator algorithm obtained from the above optimization, the optimal scanning path is finally obtained by taking into account the cost, position, attitude and collision penalty.

[0035] In the dynamic curvature feature sampling stage, the criterion for determining curvature gradient abrupt change zones is: the curvature gradient value is greater than twice the dynamic curvature threshold, and a midpoint viewpoint is inserted between adjacent gradient abrupt change points to enhance feature continuity.

[0036] In the physical constraint modeling stage of scanning pose, the construction of the visible cone boundary must strictly meet the constraints of the scanner's field of view and working distance, and the consistency of the scanning direction must be ensured through normal vector unification technology.

[0037] In the two-level collision detection algorithm, the first level uses spatial acceleration structure detection based on the separating axis theorem to quickly eliminate non-collision line segments, and the second level performs accurate collision detection on the remaining line segments through adaptive sampling and nearest neighbor query.

[0038] This invention relates to the field of 3D scanning path planning technology for large structural components, proposing a full-coverage scanning method based on curvature adaptive sampling and pose collaborative optimization. This method is applicable to automated inspection scenarios of complex curved surface structures such as ship sections, spacecraft components, and automotive body panels. The method achieves efficient, blind-spot-free scanning path planning through two core technologies: curvature feature-driven viewpoint generation and pose collaborative path optimization. Viewpoint generation stage: Principal component analysis is used to calculate the curvature features of the point cloud. Dynamic curvature and chord high-threshold segmentation technology is used to automatically identify strong feature regions. The sampling point density is increased in high curvature regions, and midpoint viewpoints are inserted at the points of abrupt curvature gradient change. Key points are sparsely retained in flat regions. Based on the "visible cone" theory, the scanner's field of view and working distance constraints are strictly satisfied. Normal vector unification technology is used to ensure the consistency of the scanning direction. The optimal pose of the scanner is calculated for each viewpoint—the viewpoint position is set at the optimal working distance offset from the sampling point along the normal vector direction, and the scanning posture is determined as the direction vector from the viewpoint to the corresponding sampling point. Finally, K-means clustering algorithm is used to compress redundant viewpoints, and the viewpoint layout is dynamically adjusted in combination with viewpoint cone coverage analysis to minimize the number of viewpoints while ensuring full coverage. Path optimization stage: A comprehensive distance matrix integrating position Euclidean distance and attitude angle is constructed, and motion efficiency and attitude continuity are balanced through weighting coefficients. An improved marine predator algorithm is used for global optimization: a hybrid initialization strategy (greedy + random) is employed to enhance population diversity; a three-stage Weibull motion model is designed (global exploration uses Weibull distribution perturbation, mid-term balancing Levy flight and elite guidance, and local development implements three elite vector weighting); a sinusoidal modulation adaptive step size control mechanism is introduced to dynamically balance exploration and development capabilities; attitude continuity constraints are embedded in pose-aware 2-opt local optimization to reduce the frequency of abrupt changes in robot arm attitude; an efficient collision detection algorithm is developed to monitor path feasibility in real time, applying collision penalties to areas with distances less than a safety threshold, and adding motion stability penalties to paths with non-same-side viewpoints. This method solves core problems such as insufficient coverage of curved surface blind spots, discontinuous robot arm motion, and equipment collisions in the scanning of large structural components, enabling fully automated high-precision path planning and significantly improving the efficiency and safety of industrial quality inspection.

[0039] The beneficial effects of this invention are that it achieves revolutionary optimization of the 3D scanning path for large structural components by constructing a comprehensive technical system that integrates curvature adaptive sampling, pose collaborative optimization, a three-stage Weibull motion strategy, real-time collision detection, and overall process efficiency improvement. This solution innovatively integrates geometric feature perception, motion continuity assurance, algorithm adaptability enhancement, and operational safety control, achieving technological breakthroughs in three dimensions: blind-spot-free coverage, high motion efficiency, and zero collision risk. Furthermore, through parametric design and a cross-platform compatible architecture, it significantly improves deployment flexibility and system robustness in industrial scenarios, providing a complete solution for the automated inspection of complex curved surfaces with high precision, high efficiency, and high reliability. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the implementable scanning path planning method for full coverage of large structural components according to the present invention.

[0041] Figure 2 This is a CAD model drawing of a car's hood (model format is STL).

[0042] Figure 3 An optimized viewpoint diagram generated for a car hood model;

[0043] Figure 4 This is a schematic diagram of the scan path generated for a certain car hood model. Detailed Implementation

[0044] The technical solutions in the embodiments of the present invention will be further described in detail below.

[0045] Figure 1 This is a flowchart illustrating the feasible scanning path planning method for full coverage of large structural components according to the present invention. The specific process is as follows:

[0046] Feature-driven initial viewpoint pose generation

[0047] First, the CAD model (e.g., STL format) is converted into point cloud data with normal vectors and curvature information, completing the mapping from the geometric model to its digital representation. Then, for high-curvature regions (e.g., edges, grooves), gradient abrupt interpolation is used to densify the sampling points, while for low-curvature regions (e.g., planes), random downsampling is used to reduce redundancy and retain key feature information. Finally, the optimal working distance is offset along the normal vector direction of the sampling points to generate the viewpoint position, ensuring that the scanner's optical axis points to the sampling point (attitude constraint), and invalid viewpoints outside the view frustum range (e.g., exceeding the maximum tilt angle) are removed. If no sampling is performed and the feature is not satisfied, the interpolation parameters are adjusted, iteratively optimizing until the feature coverage requirement is met. This results in an initial set of viewpoint poses (including position and attitude information) that meets the feature coverage requirements.

[0048] Viewpoint optimization and pose coordination cost modeling

[0049] First, adjacent redundant viewpoints are merged using clustering algorithms (such as K-means), retaining the cluster centers as core viewpoints, significantly reducing the number of viewpoints. Then, based on view frustum coverage analysis, insufficiently covered areas (such as areas with too small an intersection area) are detected, and new viewpoints are added through linear interpolation, further refining the data until no weak areas exist. Finally, a comprehensive cost matrix is ​​constructed by fusing positional distance and pose differences, quantifying the cost weights of viewpoint switching. This results in an optimized set of viewpoint poses (without redundancy or weak coverage areas) and a pose-co-cost matrix (providing optimization basis for path planning).

[0050] Pose Coordination Path Optimization and Output

[0051] First, multiple initial paths are generated by combining a greedy strategy (selecting the neighboring viewpoint with the minimum cost from the starting point) and a random strategy to enhance solution diversity and avoid local optima. Short, medium, and long-distance search modes (with Weibull distribution parameters dynamically adjusted) are used to optimize the local, mid-level, and global viewpoint order, respectively, improving the algorithm's exploration capability. Then, the step size is dynamically adjusted based on the attitude angle (e.g., increasing the step size in high-curvature regions to ensure stability, and decreasing the step size in low-curvature regions to improve efficiency) to balance path continuity and optimization speed. Next, an improved 2-OPT algorithm is used to adjust the path order and limit the attitude angle between adjacent viewpoints to ensure the continuity of the robotic arm's movement; two-level collision detection (fast geometric detection + precise spatial query) ensures path safety. Finally, position, attitude, and collision cost are fused to output the optimal path that satisfies coverage and zero-collision constraints. The optimal scanning path (including viewpoint access order, path length, and comprehensive cost) obtained by the improved Ocean Predator algorithm is thus derived.

[0052] The following describes the implementation process in detail, following the flowchart logic of the scanning planning method (feature-driven viewpoint generation → viewpoint optimization and cost modeling → pose-coordinated path optimization), combined with the characteristics of a car hood.

[0053] Figure 2 This is a CAD model drawing of a car engine hood (model format is STL). This structural component has large dimensions: 1660mm×1070mm×170mm, and complex curvature: including edges (curvature abrupt change areas), arc surfaces (high curvature areas), and planes (low curvature areas).

[0054] Feature-driven viewpoint generation

[0055] (1) Discretization of freeform surfaces

[0056] Point cloud generation and normal vector calculation: The CAD model of the car hood (such as STL format) is converted into point cloud data with normal vector and curvature information. Redundant points are removed by voxel filtering and the direction of the normal vector is corrected to ensure that the point cloud model accurately reflects the geometric features of the hood (such as surface curvature and edge position).

[0057] Curvature feature analysis: PCA is used to calculate the local curvature of the point cloud, dividing it into high curvature regions (such as edges and logo recesses) and low curvature regions (such as the central plane). The chamfered edges, reinforcing ribs, and logo recesses of the hood need to be marked as high curvature feature sensitive areas, while the central plane is a low curvature flat area.

[0058] (2) Dynamic curvature feature sampling

[0059] Initial uniform sampling: Perform initial uniform sampling on the point cloud to generate an initial sampling point set.

[0060] Dynamic curvature threshold segmentation: Calculate the curvature gradient of the sampling points and mark points with gradient values ​​greater than twice the dynamic curvature threshold as curvature abrupt change regions (such as the intersection of a stiffener and a plane). Increase the sampling point density in high curvature regions (e.g., retain all high curvature points) and sparsely sample in flat regions (e.g., retain 30% of flat points).

[0061] Midpoint viewpoint insertion: Inserting a midpoint viewpoint between adjacent gradient abrupt change points enhances feature continuity (such as the transition area of ​​the LOGO groove).

[0062] Scan pose physical modeling

[0063] (1) Viewpoint position generation

[0064] Offset the sampling points generated in the previous step by 500mm along the normal vector direction to the optimal working distance to generate the viewpoint position. Due to the large tilt angle of the hood's curved surface, it is necessary to ensure that all viewpoints are located on the scanning side. If the nearest distance from the generated viewpoint to the point cloud is less than the safety threshold, the viewpoint position at that point needs to be adjusted. This safety threshold considers the scanner's physical dimensions and safety margins.

[0065] (2) Scanning attitude constraints

[0066] The scanner's attitude at each viewpoint is defined as follows: The direction vector pointing from the viewpoint to the corresponding sampling point is used as a reference, and this direction vector must be opposite to the direction of the normal vector at the sampling point. Based on this, this direction vector is determined as the attitude representation direction of the scanner at that viewpoint position, and it is used as the Z-axis reference direction of the scanner coordinate system. This setting ensures that the scanner always accurately points to the target surface during the scanning process, thus meeting the precise attitude control requirements of the scanning operation.

[0067] Initial viewpoint verification

[0068] (1) Visual cone coverage inspection:

[0069] Based on the scanner parameters and constraint information, invalid viewpoints that exceed the scanner's field of view (e.g., 51°), working distance (300-700mm), or maximum tilt angle constraint of 60° are eliminated.

[0070] (2) Coverage assessment:

[0071] Calculate the coverage of the curved surface region by the initial viewpoint. If the coverage is less than 100%, adjust the sampling parameters (such as increasing the number of interpolations in the high curvature region) and return to dynamic curvature feature sampling for re-optimization.

[0072] Output an initial viewpoint pose set (including position and attitude information) that meets the requirements for hood feature coverage.

[0073] Viewpoint optimization and cost modeling

[0074] (1) Simplification and optimization of spatial viewpoint

[0075] K-means clustering: The K-means algorithm is used to merge adjacent redundant viewpoints, retaining the cluster centers as core viewpoints. Hood feature adaptation: The hood surface has good continuity, but sufficient viewpoints need to be retained at corners to avoid coverage blind spots. During clustering, a balance needs to be struck between global coverage and local density.

[0076] Coverage gaps: By analyzing the cone coverage, areas that are not fully covered (such as the corner where the hood meets the body) are detected, and new viewpoints are added using linear interpolation.

[0077] Figure 3 This is an optimized viewpoint diagram generated by the present invention for a certain car hood model; the number of viewpoints is 49, including information on spatial position and attitude.

[0078] (2) Modeling of pose coordination costs

[0079] Position distance and attitude difference matrix: Calculate the Euclidean distance matrix between viewpoints and the angle matrix between direction vectors, and normalize to [0,1]. Comprehensive cost matrix: Use a weighted linear combination to fuse the normalized position distance and attitude difference into a comprehensive distance matrix, with the formula:

[0080]

[0081] Where: d max It is the maximum distance in the position distance matrix, d ij Let be the Euclidean distance between viewpoints i and j, r be the position weight coefficient with a value of 0.5, and d be the distance between viewpoints i and j. c It is the combined distance between viewpoints i and j.

[0082] Pose Co-location Path Optimization and Output

[0083] (1) Improved MPA algorithm initialization

[0084] Hybrid initialization strategy: Combine greedy algorithm (10% of the population) and random initialization (90% of the population) to generate multiple initial paths.

[0085] (2) Weibull's Three-Stage Movement

[0086] By adjusting the shape parameter k and the scale parameter λ, diverse motion patterns are generated, enhancing the algorithm's exploration capabilities and allowing it to flexibly adapt to complex search spaces. To improve computational efficiency while ensuring the randomness of prey movement, the following comprehensive Weibull distribution is adopted.

[0087] R W =αf s +βf m +λf l (2)

[0088] Wherein, α, β, and λ are the scaling factors for short-range, medium-range, and long-range searches, respectively. In this study, their values ​​are 0.6, 0.3, and 0.1, respectively. The scaling factor settings take into account that excessive exploration with large step sizes in the early stages of iteration may introduce more uncertainty. Therefore, the proportion of long-range searches is relatively small, while the proportion of short-range searches is the largest. s f m and f l Weibull distributions for short-range, medium-range, and long-range searches were set for different scale and shape parameters. By integrating the Weibull distributions, the search diversity and computational efficiency of the marine predator algorithm in the early stages of iteration were enhanced, thus enabling the marine predator algorithm to adapt to more flexible and varied search environments.

[0089] (3) Adaptive step size adjustment

[0090] Step size factor calculation: A sine function and a control factor are introduced to increase the flexibility and adaptability of the marine predator algorithm.

[0091]

[0092] Where, ω max and ω min The inertia weights remain the maximum and minimum values, with the weight value being ω. max =1 and ω min =0.001, the algorithm maintains good performance, ensuring the step size curve conforms to previous step size increases, improving global search capability, and gradually decreasing the step size interval as the number of iterations increases, achieving fast convergence. t is the current iteration round, Max... iter The maximum number of iterations is denoted by α, which is a control factor. The control factor α is set to 0.5, which makes the change in the adaptive step size factor more gradual, facilitating more detailed local searches in the later stages of the algorithm and thus achieving better solution accuracy.

[0093] (4) Collision detection and path smoothing

[0094] Fast geometric detection: Based on the Separating Axis Theorem (SAT), non-collision line segments are quickly eliminated. Precise spatial query: Collision risk of remaining line segments is detected through adaptive sampling and nearest neighbor query. Path smoothing: The improved 2-OPT algorithm is used to optimize the path order, prioritizing the retention of exchange operations with high pose continuity.

[0095] (5) Comprehensive cost calculation and output

[0096] By calculating the comprehensive path cost incorporating distance, attitude, and collision costs, the optimal scanning path output by the improved marine predator algorithm is obtained. The optimal scanning path (including viewpoint access order, path length, and comprehensive cost) enables efficient, blind-spot-free, and safe scanning of car hoods. Figure 4 This is a schematic diagram of the scan path generated for a certain car hood model.

[0097] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. 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.

Claims

1. A scanning path planning method for full coverage of large structural components, characterized in that... Includes the following steps: The first step is to generate a point cloud model based on the CAD model of the freeform surface of the large structural component; The second step is to collect the dynamic curvature features of large structural components in the point cloud model generated in the first step. The third step is to establish a physical constraint model for the scanning pose based on the dynamic curvature features collected in the second step. The fourth step is to optimize the initial viewpoint in the scanning pose physical constraint model established in the third step. Fifth, establish a pose cooperative cost model based on the viewpoint set optimized in the fourth step; The sixth step is to plan the scanning path based on the pose-cooperative cost model established in the fifth step.

2. The scanning path planning method for full coverage of large structural components according to claim 1, characterized in that: The specific method for generating a point cloud model from the CAD model of the free-form surface of the large structural component in the first step is as follows: Firstly, the curvature-based sampling discretization method analyzes the curvature distribution of free-form surfaces in the CAD model, dividing high-curvature regions (such as sharp edges and complex surfaces) into feature-sensitive areas and setting high sampling density, while low-curvature flat areas are sampled sparsely. After generating an initial point set using Poisson disk sampling or parametric mesh, redundant points are removed by voxel filtering and the normal vector direction is corrected. Finally, a point cloud model that balances feature accuracy and data efficiency is output, thus completing the generation of the corresponding point cloud model for the CAD model of the free-form surface of a large structural component.

3. The scanning path planning method for full coverage of large structural components according to claim 1, characterized in that: In the second step, the method for collecting dynamic curvature features is as follows: The generated point cloud model is initially uniformly sampled, and principal component analysis (PCA) is used. KD-Tree is used to accelerate neighborhood search to calculate local curvature feature values. The calculated local curvature feature values ​​are normalized to the [0,1] interval. Then, based on the large structural components, the dynamic curvature threshold is determined, the curvature gradient of each point is calculated, and points with gradient values ​​greater than twice the threshold are marked as high curvature points, and points with gradient values ​​not greater than twice the threshold are marked as flat points. All high curvature points are retained, and some points are retained by secondary sampling of the flat points. Midpoints are inserted between adjacent high curvature points and flat points.

4. The scanning path planning method for full coverage of large structural components according to claim 1, characterized in that: In the third step, the scanning method for establishing the physical constraint model of the scanning pose is as follows: The visible cone boundary is constructed based on the scanner parameters and constraints. Then, the sampling points are offset along the normal vector direction of the sampling points to obtain the position of the viewpoint. The direction vector from the viewpoint to the corresponding sampling point is defined as the attitude of the scanner at this viewpoint position, and the attitude is used as the Z-axis reference direction of the scanner coordinate system.

5. The scanning path planning method for full coverage of large structural components according to claim 4, characterized in that: The constraints include field of view constraints, working distance constraints, and maximum tilt angle constraints.

6. The scanning path planning method for full coverage of large structural components according to claim 4, characterized in that: The length of the offset is the optimal working distance, which is the sum of the minimum working distance and half the depth of field.

7. The scanning path planning method for full coverage of large structural components according to claim 1, characterized in that: In the fourth step, the method for optimizing the initial viewpoint in the scanning pose physical constraint model is as follows: To address the issue of redundant viewpoints, a K-means clustering algorithm is used to compress the initial viewpoint set. Cluster centers are dynamically assigned based on the spatial distribution of viewpoints, and neighboring viewpoints are merged. Simultaneously, viewpoint contribution is evaluated using view frustum coverage analysis. For viewpoints whose overlapping coverage exceeds a threshold, those with minimal impact on global coverage are preferentially deleted. For areas with weak coverage, new viewpoints are added through interpolation. Finally, while ensuring full model coverage, the number of viewpoints is optimized to generate the final optimized viewpoint pose information. The viewpoint set is V = {v1, v2, ... v...} n }, where n is the number of optimized viewpoints, i = 1, 2, 3, ..., n, and vi is the pose of the i-th viewpoint.

8. The scanning path planning method for full coverage of large structural components according to claim 7, characterized in that: In the fifth step, the method for establishing the pose cooperative cost model based on the optimized viewpoint set is as follows: Using the optimized viewpoint set, calculate the viewpoint position distance matrix and attitude difference matrix; Divide each element in the location distance matrix by the global maximum distance and map it to the [0,1] interval; The attitude differences are normalized by incorporating a linear transformation of the attitude matrix based on the minimum and maximum angles between the two sides, as shown in the formula: Where: θ min θ max θ is the minimum and maximum angle between the two sides in the attitude difference matrix. ij Let d be the angle between the direction vectors of viewpoints i and j. o It is the normalized result corresponding to the angle between the direction vectors of the two viewpoints i and j; A weighted linear combination method is used to fuse the normalized position distance and attitude difference into a comprehensive distance matrix, as shown in the formula: Where: d max It is the maximum distance in the position distance matrix, d ij Let be the Euclidean distance between viewpoints i and j, r be the position weight coefficient, and d be the distance between viewpoints i and j. c It is the combined distance between viewpoints i and j.

9. The scanning path planning method for full coverage of large structural components according to claim 7, characterized in that: In the sixth step, the scanning path planning based on the pose cooperative cost model using the improved marine predator algorithm includes: Hybrid initialization strategy; The random movement of prey in the early stage of the iteration is simulated using the Weibull distribution, and diverse movement patterns are generated by adjusting the shape parameter k and the scale parameter λ. The adaptive step size CF is: Where, ω max and ω min The maximum and minimum values ​​of the inertia weights are given by t, where t is the current iteration number. iter The maximum number of iterations is α, which is a control factor. The value of α controls the shape of the sine function. An attitude continuity constraint is introduced to optimize the path. When the attitude continuity value is lower than a set threshold, a penalty factor is applied to the gain, thereby reducing the priority of the swap operation. When optimizing the path length, swap operations that do not cause abrupt attitude changes, i.e., the angle between the direction vectors is too large, will be selected, thus maintaining the smoothness of attitude changes while shortening the path. Use a two-level collision detection strategy; The safety distance should be set greater than the scanner radius or half the width to ensure that the scanner does not collide with the workpiece during movement.

10. The scanning path planning method for full coverage of large structural components according to claim 9, characterized in that: The Weibull distribution is as follows: R W =αf s +βf m +λf l (4) Where α, β, and λ are the scaling factors for short-range, medium-range, and long-range searches, respectively, f s f m and f l Set the Weibull distributions for short-range, medium-range, and long-range searches for different scale and shape parameters respectively; The two-level collision detection strategy consists of spatial accelerated structure detection based on the separation axis theorem and spatial relationship detection and determination based on adaptive sampling and nearest neighbor query.

Images