Industrial mechanical arm path optimization system based on electromagnetic scanning

Abstract

The invention discloses an industrial mechanical arm path optimization system based on electromagnetic scanning, and the system comprises an image recognition module which obtains target area information, and a mechanical arm executes electromagnetic scanning according to a positioning result, and collects electromagnetic data in a space. The system constructs a space environment model fusing images and electromagnetic information, extracts obstacle boundaries and physical constraints, and forms a dynamic convex constraint set. In the path planning process, a differential evolution algorithm is adopted to generate a variation path, and the path is mapped to a constraint range through a POCS algorithm after each round of evolution, so that iterative convergence of the path under environmental constraint is realized. And finally, an optimal path is selected according to the adaptability index of the path and used for controlling the mechanical arm to execute actions. According to the invention, joint modeling and dynamic optimization control of path planning and environment data are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent control technology, specifically to an industrial robotic arm path optimization system based on electromagnetic scanning. Background Technology

[0002] With the accelerated development of intelligent manufacturing and industrial automation, industrial robotic arms play a core role in various scenarios such as welding, handling, and spraying. Currently, intelligent planning of robotic arm motion paths in complex environments has become one of the key technologies to ensure operational efficiency and safety. Common path planning methods generally rely on fixed geometric models or simple obstacle detection mechanisms, employing vision-based static environment modeling methods, and combining traditional evolutionary algorithms or heuristic search algorithms for path optimization.

[0003] Existing path planning techniques still have significant limitations in handling dynamic, unstructured electromagnetic environments. On the one hand, existing systems often only construct static environment models based on image or LiDAR data, making it difficult to perceive physical factors affecting the stable operation of robotic arms, such as electromagnetic interference, invisible obstacles, or abnormal field strength. This makes path planning susceptible to interference in real-world environments, resulting in poor robustness and incomplete identification of feasible regions. On the other hand, traditional evolutionary algorithms, such as differential evolution algorithms, lack effective integration of real-time environmental constraints during path optimization. This leads to issues such as path point drift, failure to meet obstacle avoidance or posture constraints, especially in multi-dimensional path evolution where constraint failure is particularly prominent.

[0004] Therefore, how to provide an industrial robotic arm path optimization system based on electromagnetic scanning has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide an industrial robotic arm path optimization system based on electromagnetic scanning, including an image recognition module, a scanning control module, an environment modeling module, a constraint generation module, a path constraint module, and a path output module;

[0006] The image recognition module acquires images of the target area through an image acquisition device installed on an industrial robotic arm, and extracts image boundary information and spatial positioning data of the target area based on color features;

[0007] The scanning control module controls the industrial robotic arm to move to the target area based on the spatial positioning data, and performs near-field electromagnetic sampling through the electromagnetic scanning device at the end of the industrial robotic arm to obtain the raw electromagnetic data;

[0008] The environment modeling module jointly models the image boundary information with the original electromagnetic data to construct a spatial environment model for path planning.

[0009] The constraint generation module extracts obstacle information and physical boundary conditions related to path planning based on the spatial environment model, and forms a dynamically updated set of convex constraints.

[0010] The path evolution module generates multiple initial path solutions that satisfy convex constraints in the path search variable space, and performs differential evolution operations on these initial path solutions to obtain multiple mutated path candidate solutions.

[0011] The path constraint module calls the POCS algorithm to map the candidate solutions of the mutated paths to the set of convex constraints, thereby obtaining a set of paths that satisfy the environmental constraints.

[0012] The path output module selects the optimal path based on the adaptability index of each path in the set of paths that meet environmental constraints, and converts the optimal path into control commands to execute the path movements of the robotic arm.

[0013] Furthermore, the image recognition module includes an image acquisition unit, an image conversion unit, a parameter setting unit, a mask generation unit, a boundary extraction unit, and a coordinate analysis unit;

[0014] The image acquisition unit acquires color image frame data of the target area through a camera device installed at the end of an industrial robotic arm;

[0015] The image conversion unit converts color image frame data into an HSV color space image;

[0016] The parameter setting unit stores a set of parameters for color segmentation;

[0017] The mask generation unit extracts pixel regions that conform to the set color characteristics in the HSV color space image according to the parameter set, and generates a mask image containing the target region.

[0018] When extracting pixel regions that meet the set color characteristics, perform prior region verification, neighborhood feature verification and / or temporal consistency detection on the pixel regions, and delete misjudged pixel regions;

[0019] Prior area verification refers to pre-setting a fixed occurrence range for the target and marking pixel areas that meet the set color characteristics but exceed the fixed occurrence range of the target as misjudged pixel areas;

[0020] Neighborhood feature verification refers to calculating the texture / gradient features of neighboring pixels in the mask region. If there are no pre-stored typical texture features of the target in the surrounding area, the pixel region is marked as a misjudged pixel region.

[0021] Temporal consistency detection refers to comparing the mask positions of adjacent frames. If the position change of a pixel region does not conform to a preset motion pattern, then the pixel region is marked as a misjudged pixel region.

[0022] The boundary extraction unit analyzes the continuous pixel structure in the mask image and extracts the image boundary information of the target region in the image.

[0023] The coordinate analysis unit analyzes the spatial positioning data of the target area in the image space based on the image boundary information and the imaging parameters of the image acquisition device.

[0024] Furthermore, the set of parameters used for color segmentation includes upper and lower thresholds for hue, saturation, and brightness to identify target regions. These upper and lower thresholds are determined using sample images.

[0025] Furthermore, the scanning control module includes a target input unit, a motion planning unit, an attitude control unit, a position confirmation unit, a scan triggering unit, and a data binding unit.

[0026] The target input unit receives spatial positioning data output by the image recognition module and converts the spatial positioning data into target pose data in the three-dimensional workspace;

[0027] The motion planning unit generates a sequence of motion path instructions for the industrial robotic arm based on the target pose data.

[0028] The attitude control unit receives a sequence of motion path instructions, controls the coordinated movement of each joint of the robotic arm, and adjusts the end of the electromagnetic scanning device to a vertical orientation toward the target area.

[0029] The position confirmation unit collects the real-time difference between the current position and the target pose during the movement of the robotic arm, and updates the movement path according to the real-time difference so that the robotic arm reaches the target pose.

[0030] When the attitude control unit completes the attitude adjustment and the position confirmation unit determines that the target pose has been reached, the scanning trigger unit controls the electromagnetic scanning device to perform near-field sampling of electromagnetic data at a preset scanning height h, and records the spatial pose corresponding to the sampling time.

[0031] The data binding unit binds the raw electromagnetic data acquired by the sampling triggering unit with the three-dimensional pose information provided by the target receiving unit.

[0032] Furthermore, the environment modeling module includes an image data receiving unit, an electromagnetic data receiving unit, a coordinate registration unit, an attribute fusion unit, a mesh construction unit, and a model generation unit;

[0033] The image data receiving unit receives the image boundary information output by the image recognition module;

[0034] The electromagnetic data receiving unit receives the raw electromagnetic data with three-dimensional pose information bound to it, output by the scanning control module.

[0035] The coordinate registration unit performs coordinate mapping between the pixel coordinates contained in the image boundary information and the three-dimensional pose information to generate the correspondence between the image space and the actual working space.

[0036] The attribute fusion unit matches each image boundary region with the electromagnetic signal strength and electromagnetic gradient value of the corresponding three-dimensional pose information according to the spatial correspondence, and forms a multi-dimensional joint attribute set including coordinates, electromagnetic amplitude, change gradient and image boundary.

[0037] The mesh construction unit divides the workspace into three-dimensional grid cells and injects physical information from the multi-dimensional joint attribute set into the corresponding three-dimensional grid cells based on the multi-dimensional joint attribute set; the physical information includes coordinates, electromagnetic amplitude, gradient change, and image features;

[0038] The model generation unit uses the three-dimensional grid generated by the mesh construction unit as nodes to establish spatial connection relationships between each three-dimensional grid unit and constructs a spatial environment model that includes node coordinates, electromagnetic properties and image identifiers.

[0039] Furthermore, the constraint generation module includes a model receiving unit, an obstacle identification unit, an obstacle boundary construction unit, a gradient analysis unit, an attitude constraint generation unit, and a constraint construction unit;

[0040] The model receiving unit receives a spatial environment model; the spatial environment model includes the image boundary labels, electromagnetic signal amplitude, electromagnetic gradient value and corresponding spatial coordinates of each three-dimensional grid unit;

[0041] The obstacle recognition unit filters three-dimensional grid units with image boundary labels and three-dimensional grid units with electromagnetic signal amplitude exceeding a set threshold in the spatial environment model, and defines the set of center coordinates of the filtered three-dimensional grid units as the obstacle point set.

[0042] The obstacle boundary construction unit performs a three-dimensional geometric envelope algorithm on the set of obstacle points to generate a minimum convex boundary volume containing all obstacle points;

[0043] The gradient analysis unit calculates the rate of change of electromagnetic signal gradient between any adjacent grid cells in the spatial environment model, marks the corresponding grid area where the rate of change of electromagnetic signal gradient exceeds a preset threshold as an electromagnetic interference area, and records the spatial coordinate range of the interference area in three-dimensional space.

[0044] The attitude restriction generation unit combines the spatial location of the electromagnetic interference zone with the coordinate axis direction marked in the spatial environment model to calculate the allowable attitude orientation range of the end effector in the corresponding spatial region, and defines the spatial attitude that does not meet the corresponding attitude orientation as an infeasible region.

[0045] The constraint construction unit maps the minimum convex boundary volume, the spatial coordinate range of the electromagnetic interference zone, and the direction of the infeasible region into convex constraint forms in the path search variable space, and constructs a dynamic convex constraint set that includes geometric obstacle constraints, electromagnetic intensity constraints, and attitude restriction constraints.

[0046] Among them, the geometric obstacle constraint means that path points must not fall inside the polyhedron defined by the convex hull;

[0047] Electromagnetic strength constraint means that path points are not allowed to appear in regions where the gradient strength exceeds a threshold.

[0048] Attitude constraints refer to the requirement that the attitude vector of a path point within a specific region must belong to the set of allowed directions.

[0049] Furthermore, the path evolution module includes an initial path construction unit, a differential perturbation generation unit, a constraint feedback adjustment unit, a cross-combination unit, and an effectiveness screening unit;

[0050] The initial path construction unit generates multiple initial path solutions that satisfy convex constraints in the path search variable space.

[0051] For each initial path solution, the differential perturbation generation unit constructs a path difference vector based on multiple other path solutions, and synthesizes the difference vector with the base path to generate a first mutated path candidate solution;

[0052] After the first variant path candidate solution is generated, the constraint feedback adjustment unit receives the convex constraint set update information, judges the trend of the path point deviating from the feasible region based on the update information, and dynamically corrects the differential perturbation direction to obtain the second variant path candidate solution.

[0053] The steps to correct the direction of differential perturbation are as follows: select two or more different path individuals from the current population, calculate the vector difference between the corresponding path points of these path individuals, and add the weighted difference vector as a perturbation term to another path individual to generate a new second mutation path candidate solution;

[0054] The cross-combination unit randomly combines the current path initial solution and the second variant path candidate solution at the path point level to generate a cross-path solution that integrates the original structure and the perturbation structure.

[0055] The validity screening unit performs convex constraint checks on each path point in the cross-path solution, identifies the set of path points that meet the constraints, eliminates infeasible path points, and constructs a continuous structure to obtain the final variant path candidate solution.

[0056] Furthermore, the candidate solutions for the mutated path include multiple consecutive path points, each of which has coordinate information in the path search space;

[0057] Furthermore, the path constraint module includes a path candidate receiving unit, a constraint set loading unit, a path point parsing unit, a path projection calculation unit, and a path reconstruction unit;

[0058] The path candidate receiving unit receives the variant path candidate solution;

[0059] The constraint set loading unit receives the dynamic convex constraint set of the current optimization round; the convex constraint set includes multiple types of environmental constraints formed by obstacle boundaries, electromagnetic interference zones, and attitude direction restrictions;

[0060] The path point parsing unit parses each path point in the candidate solution of the variant path one by one, maps each path point to the constraint space defined by the convex constraint set, and establishes the constraint correspondence between the path point and the constraint set.

[0061] The path projection calculation unit performs a stepwise constraint projection operation on each path point according to the constraint correspondence, and sequentially maps the path point to all convex constraint sets until the spatial distance between two consecutive updates of the path point is lower than a preset threshold.

[0062] The path reconstruction unit summarizes and sorts the converged path point results, reconnects them according to the spatial sequence of the original path, and generates a complete path structure within the range of the convex constraint set.

[0063] Furthermore, the path output module includes an adaptive evaluation unit and a path optimization unit;

[0064] The adaptive evaluation unit evaluates each path in the path set output by the path constraint module. The evaluation includes path length index, electromagnetic interference accumulation index, attitude deviation index, smoothness index, and constraint proximity index.

[0065] Among them, the path length index is the total movement distance of the path in three-dimensional space;

[0066] The cumulative electromagnetic interference index is the degree of exposure of a path within an electromagnetic interference area;

[0067] The attitude deviation index is the average angle of deviation between the end attitude and the target attitude during the execution of the path;

[0068] The smoothness index is the rate of change of the angle between adjacent segments of the path;

[0069] The constraint proximity index is the distribution of the minimum distance from each path point to the constraint boundary in the path;

[0070] The path selection unit calculates the total adaptability score of each path using a weighted comprehensive scoring method based on the adaptability index, and determines the path with the highest score as the optimal path, which serves as the target path for the industrial robotic arm control execution.

[0071] The technical effects of this invention are undeniable, and its beneficial effects are:

[0072] (1) By constructing a system architecture consisting of an image recognition module, an electromagnetic scanning module, an environmental modeling module and a path optimization module, this invention integrates image boundaries, spatial positioning and electromagnetic properties, and realizes accurate identification and spatial modeling of key obstacles and interference areas in complex industrial scenarios, effectively improving the environmental perception capability and modeling accuracy of path planning.

[0073] (2) This invention introduces an improved differential evolution algorithm, combines a path adaptive feedback mechanism to dynamically adjust the direction of disturbance, and introduces convex constraints for screening, thereby improving the adaptability of the path evolution process to complex environmental constraints, enhancing the convergence stability of the path search and the global optimal solution acquisition rate.

[0074] (3) This invention proposes a path projection constraint mechanism based on POCS, which gradually projects path points to multiple types of convex constraint spaces and performs convergence judgment to achieve full constraint convergence of the path under obstacle boundaries, electromagnetic interference zones and attitude constraints, thus ensuring the environmental feasibility and attitude compliance of the final path.

[0075] (4) This invention comprehensively evaluates path traversability, electromagnetic safety and motion smoothness through path adaptability index, supports intelligent selection of the optimal path in multi-path solutions, and automatically generates adaptive control commands to ensure stable, continuous and efficient execution of industrial robotic arms in complex interference environments, effectively improving the intelligence and robustness of industrial path planning systems.

[0076] In summary, this invention comprehensively applies electromagnetic sensing, image recognition, and intelligent optimization algorithms. It utilizes an electromagnetic scanning device to acquire real-time environmental and obstacle information, and combines image recognition to extract spatial boundary features, constructing a dynamic environment model. Through a fusion mechanism of differential evolution and projection to convex set algorithms, it achieves synchronous iteration of global search and constraint correction during path evolution, and selects the optimal motion trajectory through path adaptability evaluation. This invention realizes high-precision path planning and safe control of industrial robotic arms in complex electromagnetic environments, possessing advantages such as high intelligence, high path optimization efficiency, and good environmental adaptability. Attached Figure Description

[0077] Figure 1 This is a flowchart of the electromagnetic scanning-based industrial robotic arm path optimization system proposed in this invention;

[0078] Figure 2 A flowchart illustrating the path evolution of the differential evolution algorithm. Detailed Implementation

[0079] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0080] Example 1:

[0081] See Figures 1 to 2 An industrial robotic arm path optimization system based on electromagnetic scanning includes an image recognition module, a scanning control module, an environment modeling module, a constraint generation module, a path constraint module, and a path output module.

[0082] The image recognition module acquires images of the target area through an image acquisition device installed on an industrial robotic arm, and extracts image boundary information and spatial positioning data of the target area based on color features;

[0083] The scanning control module controls the industrial robotic arm to move to the target area based on the spatial positioning data, and performs near-field electromagnetic sampling through the electromagnetic scanning device at the end of the industrial robotic arm to obtain the raw electromagnetic data;

[0084] The environment modeling module jointly models the image boundary information with the original electromagnetic data to construct a spatial environment model for path planning.

[0085] The constraint generation module extracts obstacle information and physical boundary conditions related to path planning based on the spatial environment model, and forms a dynamically updated set of convex constraints.

[0086] The path evolution module generates multiple initial path solutions that satisfy convex constraints in the path search variable space, and performs differential evolution operations on these initial path solutions to obtain multiple mutated path candidate solutions.

[0087] The path constraint module calls the POCS algorithm to map the candidate solutions of the mutated paths to the set of convex constraints, thereby obtaining a set of paths that satisfy the environmental constraints.

[0088] The path output module selects the optimal path based on the adaptability index of each path in the set of paths that meet environmental constraints, and converts the optimal path into control commands to execute the path movements of the robotic arm.

[0089] Example 2:

[0090] The path optimization system for an industrial robotic arm based on electromagnetic scanning is the same as in Embodiment 1. Further, the image recognition module includes an image acquisition unit, an image conversion unit, a parameter setting unit, a mask generation unit, a boundary extraction unit, and a coordinate analysis unit.

[0091] The image acquisition unit acquires color image frame data of the target area through a camera device installed at the end of an industrial robotic arm;

[0092] The image conversion unit converts color image frame data into an HSV color space image;

[0093] The parameter setting unit stores a set of parameters for color segmentation;

[0094] The mask generation unit extracts pixel regions that conform to the set color characteristics in the HSV color space image according to the parameter set, and generates a mask image containing the target region.

[0095] When extracting pixel regions that meet the set color characteristics, perform prior region verification, neighborhood feature verification and / or temporal consistency detection on the pixel regions, and delete misjudged pixel regions;

[0096] Prior area verification refers to pre-setting a fixed occurrence range for the target and marking pixel areas that meet the set color characteristics but exceed the fixed occurrence range of the target as misjudged pixel areas;

[0097] Neighborhood feature verification refers to calculating the texture / gradient features of neighboring pixels in the mask region. If there are no pre-stored typical texture features of the target in the surrounding area, the pixel region is marked as a misjudged pixel region.

[0098] Temporal consistency detection refers to comparing the mask positions of adjacent frames. If the position change of a pixel region does not conform to a preset motion pattern, then the pixel region is marked as a misjudged pixel region.

[0099] Specifically, the spatial distribution patterns of the target (such as the target typically appearing in specific areas of the image and its relative position to other objects) are considered to determine whether a misjudgment has occurred.

[0100] Prior region constraint: If the target has a fixed range of occurrence (such as the license plate being in the lower 1 / 3 of the image), then pixels of the same color in the mask that are outside this region are directly marked as misjudged;

[0101] Neighborhood feature verification: Calculate the texture / gradient features of pixels surrounding the mask region. If there is no typical texture of the target in the surrounding area, it will be a misjudgment.

[0102] Temporal consistency detection (video scene): Because it is a video stream, the mask positions of adjacent frames can be compared. The position changes of the real target conform to the motion law (such as uniform speed, continuous), while the misjudged area is usually unrelated between frames → marked as misjudged.

[0103] The boundary extraction unit analyzes the continuous pixel structure in the mask image and extracts the image boundary information of the target region in the image.

[0104] The coordinate analysis unit analyzes the spatial positioning data of the target area in the image space based on the image boundary information and the imaging parameters of the image acquisition device.

[0105] Example 3:

[0106] The industrial robotic arm path optimization system based on electromagnetic scanning has the same technical content as any one of Embodiments 1-2. Further, the parameter set used for color segmentation includes upper and lower thresholds for hue, saturation, and brightness of the target region. These upper and lower thresholds are determined through sample images. Specifically, statistical intervals are calculated as initial thresholds by acquiring HSV data of the target samples.

[0107] Sample collection: Select 10-20 sample images of typical industrial scenarios, covering different working conditions, including target areas with different lighting (strong light / shadow), different angles (different scanning directions of electromagnetic scanning), and different surface conditions (clean / slight oil stains).

[0108] Manually select the pure target area (no background, no reflective points) in the sample, and use the image processing tool OpenCV to extract the H, S, and V values ​​of all pixels in these areas.

[0109] Statistical calculation threshold:

[0110] Calculate the statistical feature values ​​for the collected H, S, and V pixel values: mean μ and standard deviation σ.

[0111] Example 4:

[0112] The path optimization system for an industrial robotic arm based on electromagnetic scanning has the same technical content as any one of embodiments 1-3. Further, the scanning control module includes a target input unit, a motion planning unit, a posture control unit, a position confirmation unit, a scanning trigger unit, and a data binding unit.

[0113] The target input unit receives spatial positioning data output by the image recognition module and converts the spatial positioning data into target pose data in the three-dimensional workspace;

[0114] Specifically, the robotic arm aligns the center of the object with the center of the mask image, and in order to reduce scanning errors, it converts distance errors (pixels) into angle errors, that is, it converts the image coordinates into actual spatial coordinates.

[0115] The error calibration method is as follows:

[0116] error_x = obj_x - center_x # Horizontal error

[0117] error_y = obj_y - center_y # Vertical error

[0118] The motion planning unit generates a sequence of motion path instructions for the industrial robotic arm based on the target pose data.

[0119] The attitude control unit receives a sequence of motion path instructions, controls the coordinated movement of each joint of the robotic arm, and adjusts the end of the electromagnetic scanning device to a vertical orientation toward the target area.

[0120] The position confirmation unit collects the real-time difference between the current position and the target pose during the movement of the robotic arm, and updates the movement path according to the real-time difference so that the robotic arm reaches the target pose.

[0121] When the attitude control unit completes the attitude adjustment and the position confirmation unit determines that the target pose has been reached, the scanning trigger unit controls the electromagnetic scanning device to perform near-field sampling of electromagnetic data at a preset scanning height h, and records the spatial pose corresponding to the sampling time.

[0122] The data binding unit binds the raw electromagnetic data acquired by the sampling triggering unit with the three-dimensional pose information provided by the target receiving unit.

[0123] Example 5:

[0124] The path optimization system for an industrial robotic arm based on electromagnetic scanning has the same technical content as any one of embodiments 1-4. Furthermore, the environment modeling module includes an image data receiving unit, an electromagnetic data receiving unit, a coordinate registration unit, an attribute fusion unit, a mesh construction unit, and a model generation unit.

[0125] The image data receiving unit receives the image boundary information output by the image recognition module;

[0126] The electromagnetic data receiving unit receives the raw electromagnetic data with three-dimensional pose information bound to it, output by the scanning control module.

[0127] The coordinate registration unit maps the pixel coordinates contained in the image boundary information to the three-dimensional pose information to generate a correspondence between the image space and the actual workspace. The coordinate mapping method is: pan_angle = pan_angle + errorPan / 75 # horizontal direction (75 is an empirical value obtained after debugging), tilt_angle = tilt_angle - errorTilt / 75 # vertical direction.

[0128] The attribute fusion unit matches each image boundary region with the electromagnetic signal strength and electromagnetic gradient value of the corresponding three-dimensional pose information according to the spatial correspondence, and forms a multi-dimensional joint attribute set including coordinates, electromagnetic amplitude, change gradient and image boundary.

[0129] The mesh construction unit divides the workspace into three-dimensional grid cells and injects physical information from the multi-dimensional joint attribute set into the corresponding three-dimensional grid cells based on the multi-dimensional joint attribute set; the physical information includes coordinates, electromagnetic amplitude, gradient change, and image features;

[0130] The model generation unit uses the three-dimensional grid generated by the mesh construction unit as nodes to establish spatial connection relationships between each three-dimensional grid unit and constructs a spatial environment model that includes node coordinates, electromagnetic properties and image identifiers.

[0131] Example 6:

[0132] The path optimization system for an industrial robotic arm based on electromagnetic scanning has the same technical content as any one of embodiments 1-5. Further, the constraint generation module includes a model receiving unit, an obstacle identification unit, an obstacle boundary construction unit, a gradient analysis unit, an attitude constraint generation unit, and a constraint construction unit.

[0133] The model receiving unit receives a spatial environment model; the spatial environment model includes the image boundary labels, electromagnetic signal amplitude, electromagnetic gradient value and corresponding spatial coordinates of each three-dimensional grid unit;

[0134] The obstacle recognition unit filters three-dimensional grid units with image boundary labels and three-dimensional grid units with electromagnetic signal amplitude exceeding a set threshold in the spatial environment model, and defines the set of center coordinates of the filtered three-dimensional grid units (three-dimensional grid units with electromagnetic signal amplitude exceeding the set threshold and three-dimensional grid units with image boundary labels) as the obstacle point set.

[0135] The obstacle boundary construction unit performs a three-dimensional geometric envelope algorithm on the set of obstacle points to generate a minimum convex boundary volume containing all obstacle points;

[0136] The gradient analysis unit calculates the rate of change of electromagnetic signal gradient between any adjacent grid cells in the spatial environment model, marks the corresponding grid area where the rate of change of electromagnetic signal gradient exceeds a preset threshold as an electromagnetic interference area, and records the spatial coordinate range of the interference area in three-dimensional space.

[0137] The attitude restriction generation unit combines the spatial location of the electromagnetic interference zone with the coordinate axis direction marked in the spatial environment model to calculate the allowable attitude orientation range of the end effector in the corresponding spatial region, and defines the spatial attitude that does not meet the corresponding attitude orientation as an infeasible region.

[0138] The constraint construction unit maps the minimum convex boundary volume, the spatial coordinate range of the electromagnetic interference zone, and the direction of the infeasible region into convex constraint forms in the path search variable space, and constructs a dynamic convex constraint set that includes geometric obstacle constraints, electromagnetic intensity constraints, and attitude restriction constraints.

[0139] Among them, the geometric obstacle constraint means that path points must not fall inside the polyhedron defined by the convex hull;

[0140] Electromagnetic strength constraint means that path points are not allowed to appear in regions where the gradient strength exceeds a threshold.

[0141] Attitude constraints refer to the requirement that the attitude vector of a path point within a specific region must belong to the set of allowed directions.

[0142] Example 7:

[0143] The path optimization system for industrial robotic arms based on electromagnetic scanning has the same technical content as any one of embodiments 1-6. Further, the path evolution module includes an initial path construction unit, a differential disturbance generation unit, a constraint feedback adjustment unit, a cross combination unit, and an effectiveness screening unit.

[0144] The initial path construction unit generates multiple initial path solutions that satisfy convex constraints in the path search variable space.

[0145] For each initial path solution, the differential perturbation generation unit constructs a path difference vector based on multiple other path solutions, and synthesizes the difference vector with the base path to generate a first mutated path candidate solution;

[0146] After the first variant path candidate solution is generated, the constraint feedback adjustment unit receives the convex constraint set update information, judges the trend of the path point deviating from the feasible region based on the update information, and dynamically corrects the differential perturbation direction to obtain the second variant path candidate solution.

[0147] The steps to correct the direction of differential perturbation are as follows: select two or more different path individuals from the current population, calculate the vector difference between the corresponding path points of these path individuals, and add the weighted difference vector as a perturbation term to another path individual to generate a new second mutation path candidate solution;

[0148] The cross-combination unit randomly combines the current path initial solution and the second variant path candidate solution at the path point level to generate a cross-path solution that integrates the original structure and the perturbation structure.

[0149] The validity screening unit performs convex constraint checks on each path point in the cross-path solution, identifies the set of path points that meet the constraints, eliminates infeasible path points, and constructs a continuous structure to obtain the final variant path candidate solution.

[0150] Example 8:

[0151] The industrial robotic arm path optimization system based on electromagnetic scanning has the same technical content as any one of embodiments 1-7. Further, the variant path candidate solution includes multiple continuous path points, and each path point has coordinate information in the path search space.

[0152] Example 9:

[0153] The path optimization system for an industrial robotic arm based on electromagnetic scanning has the same technical content as any one of embodiments 1-8. Further, the path constraint module includes a path candidate receiving unit, a constraint set loading unit, a path point parsing unit, a path projection calculation unit, and a path reconstruction unit.

[0154] The path candidate receiving unit receives the variant path candidate solution;

[0155] The constraint set loading unit receives the dynamic convex constraint set of the current optimization round; the convex constraint set includes multiple types of environmental constraints formed by obstacle boundaries, electromagnetic interference zones, and attitude direction restrictions;

[0156] The path point parsing unit parses each path point in the candidate solution of the variant path one by one, maps each path point to the constraint space defined by the convex constraint set, and establishes the constraint correspondence between the path point and the constraint set.

[0157] The path projection calculation unit performs a stepwise constraint projection operation on each path point according to the constraint correspondence, and sequentially maps the path point to all convex constraint sets until the spatial distance between two consecutive updates of the path point is lower than a preset threshold.

[0158] The path reconstruction unit summarizes and sorts the converged path point results, reconnects them according to the spatial sequence of the original path, and generates a complete path structure within the range of the convex constraint set.

[0159] Example 10:

[0160] The path optimization system for industrial robotic arms based on electromagnetic scanning has the same technical content as any one of embodiments 1-9. Furthermore, the path output module includes an adaptive evaluation unit and a path optimization unit.

[0161] The adaptive evaluation unit evaluates each path in the path set output by the path constraint module. The evaluation includes path length index, electromagnetic interference accumulation index, attitude deviation index, smoothness index, and constraint proximity index.

[0162] Among them, the path length index is the total movement distance of the path in three-dimensional space;

[0163] The cumulative electromagnetic interference index is the degree of exposure of a path within an electromagnetic interference area;

[0164] The attitude deviation index is the average angle of deviation between the end attitude and the target attitude during the execution of the path;

[0165] The smoothness index is the rate of change of the angle between adjacent segments of the path;

[0166] The constraint proximity index is the distribution of the minimum distance from each path point to the constraint boundary in the path;

[0167] The path selection unit calculates the total adaptability score of each path using a weighted comprehensive scoring method based on the adaptability index, and determines the path with the highest score as the optimal path, which serves as the target path for the industrial robotic arm control execution.

[0168] Example 11:

[0169] An industrial robotic arm path optimization system based on electromagnetic scanning includes:

[0170] The image recognition module is used to acquire images of the target area through an image acquisition device installed on an industrial robotic arm, and to extract the image boundary information and spatial positioning data of the target area based on color features;

[0171] The scanning control module is used to control the industrial robotic arm to move to the target area based on spatial positioning data, and to perform near-field electromagnetic sampling through the electromagnetic scanning device at its end to obtain raw electromagnetic data;

[0172] The environment modeling module is used to jointly model image boundary information with raw electromagnetic data to construct a spatial environment model for path planning.

[0173] The constraint generation module is used to extract obstacle information and physical boundary conditions related to path planning based on the spatial environment model, and form a dynamically updatable set of convex constraints.

[0174] The path evolution module is used to perform differential evolution operations on multiple initial solutions of paths based on a set of convex constraints to obtain candidate solutions of mutated paths;

[0175] The path constraint module is used to call the POCS algorithm after each differential mutation to map the candidate solutions of the mutated path to the convex constraint set, thereby obtaining a set of paths that satisfy the environmental constraints.

[0176] The path output module is used to select the optimal path based on the adaptability index of each path in the path set, and convert the optimal path into control commands to execute the path movements of the robotic arm.

[0177] In this embodiment, the image recognition module includes:

[0178] The image acquisition unit is used to acquire color image frame data of the target area through a camera device installed at the end of an industrial robotic arm;

[0179] The image conversion unit is used to convert color image frames into HSV color space images for color feature extraction processing.

[0180] The parameter setting unit is used to set the upper and lower limit thresholds of hue, saturation and brightness for identifying target areas in the HSV color space, and generate a set of parameters for color segmentation.

[0181] The mask generation unit is used to extract pixel regions that conform to the set color characteristics in the HSV color space image according to the parameter set, and generate a mask image containing the target region.

[0182] The boundary extraction unit is used to analyze the continuous pixel structure in the mask image and extract the image boundary information of the target region in the image;

[0183] The coordinate analysis unit is used to analyze the spatial positioning data of the target area in the image space based on the image boundary information and the imaging parameters of the image acquisition device, and output the spatial positioning data to the scanning control module for the movement positioning of the robotic arm.

[0184] Setting upper and lower thresholds for the hue, saturation, and brightness of the target area refers to setting the minimum and maximum ranges for the hue (H), saturation (S), and brightness (V) values ​​of the measured area in the HSV color space, used for pixel filtering in the image. The set threshold range covers the main color distribution of the target area while excluding color components from non-target areas, thus allowing the extraction of the target area's outline using a binary mask. This threshold setting is based on the identifiability of the target color in the image and is adjusted in conjunction with actual ambient lighting conditions.

[0185] Defining pixel regions based on color characteristics involves comparing all pixels in an image within the HSV color space with preset hue, saturation, and brightness threshold ranges, selecting the set of pixels that meet the corresponding color conditions, and marking these pixel regions as target regions. This process uses mask image generation to highlight the distribution of the target regions within the entire image.

[0186] Analyzing the structure of continuous pixels involves determining the connectivity of all pixels in a masked image that are in an active state (e.g., white or "1"), identifying adjacent or connected pixel groups, and dividing them into one or more region contours based on their topological relationships within the image. This operation is often based on connected component analysis in image processing algorithms to extract the closed boundaries of target regions in the image for subsequent localization and contour tracking.

[0187] Imaging parameters of an image acquisition device refer to a set of fixed or adjustable values ​​used to describe the imaging characteristics of a camera. These mainly include focal length, principal point position, pixel size, image resolution, field of view, and distortion coefficient. These parameters are usually obtained through a calibration process to map pixels in the image coordinate system to the actual spatial coordinate system, thus supporting the conversion between image recognition results and the robotic arm's motion space.

[0188] In this embodiment, the scanning control module includes:

[0189] The target input unit is used to receive the positioning data of the target region in the image space output by the image recognition module and convert it into target pose data in the three-dimensional workspace.

[0190] The motion planning unit is used to generate a sequence of motion path instructions for the industrial robotic arm based on the target pose data and output it to the attitude control unit.

[0191] The attitude control unit is used to receive motion path command sequences, control the coordinated movement of each joint of the robotic arm, and adjust the end of the electromagnetic scanning device to a vertical orientation toward the target area.

[0192] The position confirmation unit is used to collect the real-time difference between the current position and the target pose during the movement of the robotic arm, and use the real-time difference to update the motion path to achieve fine-tuning of the position;

[0193] The scanning trigger unit is used to control the electromagnetic scanning device to perform near-field sampling of electromagnetic data at a specified scanning height and record the spatial pose corresponding to the sampling time when the attitude control unit completes attitude adjustment and the position confirmation unit determines that the target pose has been reached.

[0194] The data binding unit is used to bind the raw electromagnetic data acquired by the sampling trigger unit with the three-dimensional pose information provided by the target receiving unit, and output it to the environment modeling module as modeling input data.

[0195] Near-field electromagnetic data sampling refers to placing an electromagnetic probe at a predetermined distance above the surface of a target area, typically on the order of millimeters. A vector network analyzer is used to excite and collect the reflected or transmitted signals received by the probe, obtaining the near-field electromagnetic parameters at the corresponding point, such as amplitude, phase, or S-parameters. During sampling, a robotic arm maintains a constant scanning height and moves the probe point-by-point or continuously to cover the target area, achieving spatial distribution measurement of the electromagnetic signal.

[0196] In this embodiment, the environment modeling module includes:

[0197] The image data receiving unit is used to receive image boundary information output by the image recognition module;

[0198] The electromagnetic data receiving unit is used to receive the raw electromagnetic data with three-dimensional pose information bound to it, output by the scanning control module.

[0199] The coordinate registration unit is used to map the pixel coordinates contained in the image boundary information to the three-dimensional pose information, thereby generating the correspondence between the image space and the actual working space.

[0200] The attribute fusion unit is used to match the electromagnetic signal intensity and electromagnetic gradient value of each image boundary region with the corresponding three-dimensional pose information according to the spatial correspondence, and form a multi-dimensional joint attribute set containing coordinates, electromagnetic amplitude, change gradient and image boundary.

[0201] Mesh building units are used to divide the entire workspace into three-dimensional grid units and inject physical information from the multi-dimensional joint attribute set into the corresponding three-dimensional grid units based on the multi-dimensional joint attribute set.

[0202] The model generation unit is used to establish spatial connection relationships between each 3D raster unit with the 3D raster generated by the mesh building unit as nodes, and to construct a spatial environment model containing node coordinates, electromagnetic properties and image labels.

[0203] The model output unit is used to transmit the spatial environment model to the constraint generation module for obstacle detection, electromagnetic constraint extraction and attitude limitation judgment in subsequent path planning;

[0204] Coordinate mapping refers to converting the spatial pixel coordinates of the image acquired by the image recognition module into actual three-dimensional coordinates in the robotic arm's workspace. This process is based on the camera's imaging parameters, including focal length, principal point position, image resolution, and installation posture, and maps two-dimensional image coordinates to spatial points in the robotic arm's coordinate system. During the mapping process, spatial calibration information from image acquisition is incorporated to compensate for geometric distortions in the field of view. Furthermore, based on the positions of boundary points extracted from the image in the pixel coordinate system, their position vectors in three-dimensional space are inferred, generating a target point set with spatial pose information. This set serves as the data foundation for robotic arm localization and subsequent modeling.

[0205] Establishing spatial connections between each 3D grid cell refers to constructing a topological structure between nodes based on the physical adjacency of the grid cells after dividing the workspace into equally spaced 3D grids. Each grid cell acts as a spatial node, establishing first-order connections with its adjacent grid cells in the six directions (X, Y, Z) according to the hexahedral adjacency principle. For each pair of adjacent grid cells, the index offset or distance vector between them is recorded and represented as directed or undirected edges in the modeling structure. This connection relationship is used to calculate the connectivity and cost evaluation between candidate paths during path search, and also provides a spatial topological basis for constraint generation and path optimization. The connection information can be stored as an adjacency matrix or a graph structure.

[0206] The spatial environment model is a three-dimensional structured representation built upon image recognition boundary information and electromagnetic scanning data, using the industrial robotic arm's workspace as its scope. The model consists of regularly divided three-dimensional grid cells. Each grid cell contains its center coordinates within the workspace and a set of physical attributes corresponding to that location. These attributes include electromagnetic signal amplitude, electromagnetic gradient values, and image boundary labels. By establishing spatial connections between grid cells, the model forms a graph representation with an adjacency topology, used to describe obstacle distribution, electromagnetic interference zones, and safe passage areas. The spatial environment model can serve as input to the path planning and constraint generation module, supporting the avoidance and dynamic updating of potential risk areas in the robotic arm's path, thus coupling environmental perception and motion control.

[0207] In this embodiment, the constraint generation module includes:

[0208] The model receiving unit is used to receive a spatial environment model, which includes the image boundary labels, electromagnetic signal amplitude, electromagnetic gradient value and corresponding spatial coordinates of each three-dimensional grid unit.

[0209] The obstacle recognition unit is used to filter three-dimensional grid cells with image boundary labels and grid cells with electromagnetic signal amplitude exceeding a set threshold in the received spatial environment model, and define the set of their center coordinates as the obstacle point set.

[0210] An obstacle boundary construction unit is used to perform a three-dimensional geometric envelope algorithm on the set of obstacle points to generate a minimum convex boundary volume containing all obstacle points. The minimum convex boundary volume serves as the geometric obstacle boundary of the impassable region in the path search space.

[0211] The gradient analysis unit is used to calculate the rate of change of electromagnetic signal gradient between any adjacent grid cells in the spatial environment model. If the rate of change exceeds a preset threshold, the corresponding grid area is marked as an electromagnetic interference area, and the spatial coordinate range of the interference area in three-dimensional space is recorded.

[0212] The attitude constraint generation unit is used to combine the spatial location of the electromagnetic interference zone with the coordinate axis direction marked in the spatial environment model to calculate the allowable attitude orientation range of the end effector in the corresponding spatial region, and define the spatial attitude that does not meet the corresponding attitude orientation as an infeasible region.

[0213] The constraint construction unit is used to map the spatial coordinate range of the minimum convex boundary body, the electromagnetic interference zone, and the direction of the infeasible region into convex constraint forms in the path search variable space, and to construct a dynamic convex constraint set that includes geometric obstacle constraints, electromagnetic strength constraints, and attitude restriction constraints.

[0214] The constraint output unit is used to output the dynamic convex constraint set generated by the constraint construction unit to the path evolution module, and is used to perform constraint verification and correction operations on the candidate path solutions during differential evolution.

[0215] A threshold-defined grid cell refers to a 3D grid cell within a spatial environment model where the physical attribute values ​​(such as electromagnetic signal amplitude, electromagnetic gradient, etc.) are compared to a preset range. When the attribute value within a grid cell exceeds the corresponding threshold, the grid is marked as an "abnormal" or "high-risk" area and used for subsequent obstacle or interference zone identification. For example, if the electromagnetic signal amplitude exceeds a set maximum allowable intensity threshold, or the gradient change rate exceeds a set abrupt change threshold, the corresponding grid cell can be marked as a cell requiring exclusion or path restriction. This threshold can be set according to equipment safety specifications, material shielding capabilities, or mission requirements, and supports dynamic adjustment.

[0216] The 3D geometric envelope algorithm refers to constructing a minimum convex hull to enclose all identified obstacle points in a spatial environment model, using their coordinates in 3D space. Specifically, each point in the obstacle set is treated as a node in 3D space, and algorithms such as Quickhull or Incremental are used to calculate the convex polyhedron enclosing these points, generating a 3D geometric structure containing a vertex set, edge set, and face set. This structure represents the boundary of impassable obstacle regions and serves as input to the constraint construction unit as the region to be avoided in path planning. The generated geometric boundary is closure and convex, facilitating intersection judgment and constraint projection processing with path points or solution vector spaces.

[0217] The convex constraint form in the path search variable space refers to converting path planning constraints such as obstacle boundaries, electromagnetic interference regions, and attitude restrictions into a mathematical expression structure that can be used for optimization calculations. Specifically, the path is represented as a set of continuous discrete points or control variables, each defined in a three-dimensional coordinate space, and constraints are applied to these variables in the form of convex sets. For example, the constraint for an obstacle region is "path points must not fall inside the polyhedron defined by the convex hull," the constraint for an electromagnetic interference region is "path points are not allowed to appear in regions where the gradient intensity exceeds a threshold," and the attitude restriction corresponds to "the attitude vector of a path point within a specific region must belong to the set of allowed directions." These constraints are uniformly modeled as the intersection of multiple convex sets, and projection operators such as POCS are used to keep the path variables within the feasible region during optimization.

[0218] In this embodiment, the path evolution module includes:

[0219] An initial path construction unit is used to generate multiple initial path solutions that satisfy convex constraints in the path search variable space. The initial path solutions consist of a series of spatial path points.

[0220] The differential perturbation generation unit is used to construct a path difference vector based on multiple other path solutions for each initial path solution, and synthesize the difference vector with the base path to generate the first mutated path candidate solution;

[0221] The constraint feedback adjustment unit is used to receive the convex constraint set update information after the first variant path candidate solution is generated, determine the trend of the path point deviating from the feasible region based on the update information, and dynamically correct the differential perturbation direction to obtain the second variant path candidate solution.

[0222] The cross-combination unit is used to randomly combine the current path initial solution and the second variant path candidate solution at the path point level to generate a cross-path solution that integrates the original structure and the perturbation structure.

[0223] The validity screening unit is used to perform convex constraint checks on each path point in the cross-path solution, identify the set of path points that meet the constraints, eliminate infeasible path points, and construct a continuous structure to obtain the final variant path candidate solution.

[0224] Constructing path difference vectors refers to selecting two or more distinct path individuals from the current population during differential evolution, calculating the vector difference between their corresponding path points, and then weighting this difference vector and adding it as a perturbation term to another path individual to generate a new mutated path. Specifically, in the path point dimension, two sets of path individuals are selected. and The coordinate difference of its corresponding position Perform linear scaling and connect with the third path individual The coordinates are added point by point to generate a new sequence of variable path point coordinates. By modeling the differences in each path point dimension, effective perturbation of the path structure can be achieved.

[0225] Synthesizing the difference vector with the base path involves superimposing the path difference vector as a perturbation term onto a base path to construct a new candidate path. Specifically, a base path is selected, consisting of multiple consecutive path points. For each path point, its corresponding coordinate value is extracted, and the difference vector component of the same dimension is added to this coordinate value according to a set scaling factor, thus obtaining the new path point position. This process is performed sequentially across all path points, ultimately forming a new path sequence where the overall structure remains continuous, but the path shape undergoes perturbation and change.

[0226] The tendency of a path point to deviate from the feasible region is determined by comparing the positional relationship between the mutated path point and the current constraint boundary. Specifically, for each path point, the shortest distance between it and the nearest convex constraint boundary surface (such as an obstacle boundary, electromagnetic interference zone boundary, or attitude restriction zone boundary) is calculated. If this distance is less than a set threshold, or its direction points towards the interior of the infeasible region, it is considered to have a deviation tendency. Furthermore, the angle between the changing gradient direction of the path point and the normal direction of the constraint boundary can be introduced. If the angle is less than a set angle threshold, the mutated direction of the path point is considered to be trending towards the infeasible region. If either of these two conditions is met, the path point is determined to have a tendency to deviate from the feasible region.

[0227] Convex constraint checking refers to the process of evaluating each path point during path optimization to confirm whether it lies within a predefined convex feasible region. This process involves comparing the path point's 3D coordinates with multiple sets of convex constraints, comprised of geometric obstacles, electromagnetic interference zones, and attitude restrictions. If a path point satisfies the inclusion condition of all constraint sets—that is, its coordinates fall within or on the boundary of each convex set—then the path point is considered to satisfy the convex constraint conditions; otherwise, it is considered an invalid path point.

[0228] In this embodiment, the path constraint module includes:

[0229] A path candidate receiving unit is used to receive variant path candidate solutions, wherein the variant path candidate solutions are composed of multiple consecutive path points, and each path point has coordinate information in the path search space.

[0230] The constraint set loading unit is used for the dynamic convex constraint set of the current optimization round. The convex constraint set includes multiple types of environmental constraints formed by obstacle boundaries, electromagnetic interference zones and attitude direction restrictions, and the set is passed to the subsequent path point constraint calculation unit.

[0231] The path point parsing unit is used to parse each path point in the candidate solution of the variant path, map each path point to the constraint space defined by the convex constraint set, and establish the constraint correspondence between the path point and the constraint set.

[0232] The path projection operation unit is used to perform a stepwise constraint projection operation on each path point according to the constraint correspondence, sequentially mapping the path point to all convex constraint sets. After each projection is completed, the coordinates of the path point are updated, and the updated path point is used as the input point of the next constraint. After multiple rounds of cyclic projection, if the spatial distance between two consecutive updates of the path point is lower than a preset threshold, it is determined that the corresponding path point has reached a convergence state in all convex constraint sets, and the convergence result is recorded.

[0233] The path reconstruction unit is used to summarize and sort the converged path point results, reconnect them according to the spatial sequence of the original path, and generate a complete path structure within the scope of the convex constraint set.

[0234] The stepwise constraint projection operation refers to the process of sequentially mapping a path point to multiple convex constraint sets. The specific steps are as follows: Select a path point to be processed as the initial input; then, according to a preset constraint order, project the path point onto each individual convex constraint set. Each projection uses the current path point as input, calculates its nearest point within the set based on the geometric or physical definition of the current constraint, and updates the path point's position using this nearest point; this process is iterated multiple times until all constraint sets have been processed or the path point's position change is below a set threshold, finally outputting the path point position that satisfies all constraints.

[0235] A preset threshold is a numerical standard used to determine whether a path point has reached convergence during the progressive constraint projection process. Specifically, this threshold measures the change in spatial position of a path point after two consecutive projection operations. When the Euclidean distance between the coordinates of a path point after the k-th and (k+1)-th projections is lower than this threshold, the path point is considered to have reached a stable state under all convex constraint sets, meaning it no longer changes significantly, thus satisfying the convergence condition. This threshold is typically a positive number smaller than the path step size or spatial resolution, such as 0.01 meters or less, set according to the path accuracy requirements in the specific application scenario, and consistently applied to all path points.

[0236] In this embodiment, the path output module includes:

[0237] An adaptive evaluation unit is used to evaluate each path in the path set output by the path constraint module. The evaluation includes the following adaptive metrics:

[0238] The path length metric is used to calculate the total distance traveled by a path in three-dimensional space.

[0239] The electromagnetic interference cumulative index is used to calculate the exposure degree of a path in an electromagnetic interference area based on the cumulative value of electromagnetic signal strength at each path point.

[0240] The attitude deviation index is used to calculate the average angle of deviation between the end attitude and the target attitude during the execution of the path.

[0241] Smoothness index is used to calculate the rate of change of angle between adjacent segments of a path based on the continuity of a path point sequence.

[0242] The constraint proximity index is used to evaluate the distribution of the minimum distances from each path point to the constraint boundary in the path, and is used to reflect the path safety margin.

[0243] The path selection unit is used to calculate the total adaptability score of each path based on the adaptability index using a weighted comprehensive scoring method, and to determine the path with the highest score as the optimal path, which is used as the target path for the industrial robotic arm control execution.

[0244] Example 12:

[0245] The verification of an industrial robotic arm path optimization system based on electromagnetic scanning includes:

[0246] To verify the feasibility of this invention in practice, it was applied in a high-end manufacturing workshop. Addressing the need for high-precision surface micro-defect detection of components, the company deployed an intelligent inspection platform integrating the "Electromagnetic Scanning-Based Industrial Robotic Arm Path Optimization System" of this invention. This system primarily serves the non-destructive testing of customized aerospace structural components after welding. These workpieces have complex shapes and strong material heterogeneity, and traditional fixed-path inspection methods suffer from low efficiency and significant blind spots.

[0247] In this embodiment, the industrial robotic arm is mounted on a three-dimensional movable platform, with a high-frequency near-field electromagnetic scanning device and a high-definition image camera at its end effector. Upon initial operation, the image acquisition unit activates its imaging function, capturing color image frames containing images of the workpiece surface and sending them to the image recognition module in real time. The image conversion unit within the image recognition module converts the acquired image into HSV color space format. The parameter setting unit sets upper and lower thresholds for hue, saturation, and brightness based on preset color characteristics of paint, metal oxidation, or cleaning residue on the workpiece surface. The mask generation unit generates the corresponding mask image.

[0248] The boundary extraction unit extracts the boundary contour of the target area, while the coordinate analysis unit calculates the target area's positioning information in three-dimensional space based on camera imaging parameters. After receiving this spatial pose data, the scanning control module calls the motion planning module to generate the robotic arm's motion trajectory. The attitude control unit precisely aligns the robotic arm with the detection area, and finally performs a high-density near-field electromagnetic scan on the target surface, binding the three-dimensional pose during the scan.

[0249] The raw electromagnetic data and image boundary information are input into the environment modeling module. The coordinate registration unit completes the accurate mapping from image space to real space. Subsequently, the attribute fusion unit fuses the boundary image, pose, and electromagnetic signal intensity to generate a multi-dimensional attribute set. The mesh construction module divides the entire workpiece inspection surface into regular three-dimensional grid units and injects physical attributes, ultimately establishing a spatial environment model corresponding to the image boundary, electromagnetic response, electromagnetic gradient, and coordinates.

[0250] The constraint generation module extracts obstacle grids from the spatial model, such as structural edge regions indicated by color labels, areas with large electromagnetic interference fluctuations, and attitude restriction regions unsuitable for perpendicular incidence detection. These obstacle data are mapped into dynamic convex constraints in the path search space, establishing a complete set of convex constraints.

[0251] The path evolution module then executes an improved differential evolution algorithm. By differentially combining multiple initial path solutions and introducing a constraint feedback mechanism to adjust the perturbation direction, the search is guided to converge rapidly towards the feasible region, generating multiple candidate solutions for mutated paths. The path constraint module uses the POCS algorithm to progressively project path points from the mutated paths into the constraint set, and reconstructs the complete path set after reaching a convergent and stable state through threshold judgment.

[0252] The path output module evaluates the paths based on their adaptability metrics from the path set. These metrics include the total path length, the proportion of the path traversed by electromagnetic interference (EMI) regions, the average electromagnetic gradient change rate of the path segments, and the rotation amplitude of the robotic arm. The system uses a weighted scoring model, outputting the path with the highest score as the optimal path. This path is then converted into joint motion control commands and sent to the industrial robotic arm controller to execute the path-following task. Actual deployment data is as follows:

[0253] Table 1. Experimental data comparing path planning efficiency

[0254] Table 2 Comparison of Detection Accuracy and Execution Efficiency

[0255]

[0256] Based on the experimental data in Tables 1 and 2, the electromagnetic scanning-based industrial robotic arm path optimization system proposed in this invention demonstrates significant advantages in both path planning efficiency and detection performance. In path planning, the average path planning time is reduced from 9.4 seconds using traditional methods to 3.1 seconds, an improvement of 67%, while the path length is also reduced by 21.4%, effectively reducing robotic arm motion redundancy. Regarding electromagnetic interference crossing, this system reduces the proportion of paths crossing interference zones from 12.5% ​​to 3.2%, and the detection blind zone area is reduced by nearly 71%, indicating that its path generation is more intelligent and accurate. Further, Table 2 shows that in identifying micro-defects, the system achieves a recognition rate of 94.2%, significantly higher than the 81.5% of traditional methods; the detection cycle is only 6.5 minutes, more than half the time of the traditional 13.4 minutes, and the frequency of robotic arm posture adjustments is also reduced by 64%. These data comprehensively verify that this invention has significant overall performance improvements in terms of improving detection accuracy, reducing path redundancy, increasing execution efficiency, and adapting to complex interference environments.

[0257] In this embodiment of the invention, by deeply integrating image recognition, electromagnetic sensing, and constraint-guided intelligent path evolution algorithms, a highly adaptive electromagnetic detection path planning system is constructed, which significantly improves the automation level and fine control capability of robotic arm detection operations, and solves problems such as invalid and redundant paths, poor environmental adaptability, and blind spot missed detection in existing path planning methods.

Claims

1. An industrial robotic arm path optimization system based on electromagnetic scanning, characterized in that: It includes an image recognition module, a scanning control module, an environment modeling module, a constraint generation module, a path constraint module, and a path output module; The image recognition module acquires images of the target area through an image acquisition device installed on an industrial robotic arm, and extracts image boundary information and spatial positioning data of the target area based on color features; The scanning control module controls the industrial robotic arm to move to the target area based on the spatial positioning data, and performs near-field electromagnetic sampling through the electromagnetic scanning device at the end of the industrial robotic arm to obtain the raw electromagnetic data; The environment modeling module jointly models the image boundary information with the original electromagnetic data to construct a spatial environment model for path planning. The constraint generation module extracts obstacle information and physical boundary conditions related to path planning based on the spatial environment model, and forms a dynamically updated set of convex constraints. The path evolution module generates multiple initial path solutions that satisfy convex constraints in the path search variable space, and performs differential evolution operations on these initial path solutions to obtain multiple mutated path candidate solutions. The path constraint module calls the POCS algorithm to map the candidate solutions of the mutated paths to the set of convex constraints, thereby obtaining a set of paths that satisfy the environmental constraints. The path output module selects the optimal path based on the adaptability index of each path in the set of paths that meet environmental constraints, and converts the optimal path into control commands to execute the path movements of the robotic arm.

2. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The image recognition module includes an image acquisition unit, an image conversion unit, a parameter setting unit, a mask generation unit, a boundary extraction unit, and a coordinate analysis unit; The image acquisition unit acquires color image frame data of the target area through a camera device installed at the end of an industrial robotic arm; The image conversion unit converts color image frame data into an HSV color space image; The parameter setting unit stores a set of parameters for color segmentation; The mask generation unit extracts pixel regions that conform to the set color characteristics in the HSV color space image according to the parameter set, and generates a mask image containing the target region. When extracting pixel regions that meet the set color characteristics, perform prior region verification, neighborhood feature verification and / or temporal consistency detection on the pixel regions, and delete misjudged pixel regions; Prior area verification refers to pre-setting a fixed occurrence range for the target and marking pixel areas that meet the set color characteristics but exceed the fixed occurrence range of the target as misjudged pixel areas; Neighborhood feature verification refers to calculating the texture / gradient features of neighboring pixels in the mask region. If there are no pre-stored typical texture features of the target in the surrounding area, the pixel region is marked as a misjudged pixel region. Temporal consistency detection refers to comparing the mask positions of adjacent frames. If the position change of a pixel region does not conform to a preset motion pattern, then the pixel region is marked as a misjudged pixel region. The boundary extraction unit analyzes the continuous pixel structure in the mask image and extracts the image boundary information of the target region in the image. The coordinate analysis unit analyzes the spatial positioning data of the target area in the image space based on the image boundary information and the imaging parameters of the image acquisition device.

3. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 2, characterized in that, The set of parameters used for color segmentation includes upper and lower thresholds for hue, saturation, and brightness to identify target regions; these thresholds are determined from sample images.

4. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The scan control module includes a target input unit, a motion planning unit, an attitude control unit, a position confirmation unit, a scan triggering unit, and a data binding unit. The target input unit receives spatial positioning data output by the image recognition module and converts the spatial positioning data into target pose data in the three-dimensional workspace; The motion planning unit generates a sequence of motion path instructions for the industrial robotic arm based on the target pose data. The attitude control unit receives a sequence of motion path instructions, controls the coordinated movement of each joint of the robotic arm, and adjusts the end of the electromagnetic scanning device to a vertical orientation toward the target area. The position confirmation unit collects the real-time difference between the current position and the target pose during the movement of the robotic arm, and updates the movement path according to the real-time difference so that the robotic arm reaches the target pose. When the attitude control unit completes the attitude adjustment and the position confirmation unit determines that the target pose has been reached, the scanning trigger unit controls the electromagnetic scanning device to perform near-field sampling of electromagnetic data at a preset scanning height h, and records the spatial pose corresponding to the sampling time. The data binding unit binds the raw electromagnetic data acquired by the sampling triggering unit with the three-dimensional pose information provided by the target receiving unit.

5. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The environment modeling module includes an image data receiving unit, an electromagnetic data receiving unit, a coordinate registration unit, an attribute fusion unit, a mesh construction unit, and a model generation unit. The image data receiving unit receives the image boundary information output by the image recognition module; The electromagnetic data receiving unit receives the raw electromagnetic data with three-dimensional pose information bound to it, output by the scanning control module. The coordinate registration unit performs coordinate mapping between the pixel coordinates contained in the image boundary information and the three-dimensional pose information to generate the correspondence between the image space and the actual working space. The attribute fusion unit matches each image boundary region with the electromagnetic signal strength and electromagnetic gradient value of the corresponding three-dimensional pose information according to the spatial correspondence, and forms a multi-dimensional joint attribute set including coordinates, electromagnetic amplitude, change gradient and image boundary. The mesh construction unit divides the workspace into three-dimensional grid cells and injects physical information from the multi-dimensional joint attribute set into the corresponding three-dimensional grid cells based on the multi-dimensional joint attribute set; the physical information includes coordinates, electromagnetic amplitude, gradient change, and image features; The model generation unit uses the three-dimensional grid generated by the mesh construction unit as nodes to establish spatial connection relationships between each three-dimensional grid unit and constructs a spatial environment model that includes node coordinates, electromagnetic properties and image identifiers.

6. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The constraint generation module includes a model receiving unit, an obstacle identification unit, an obstacle boundary construction unit, a gradient analysis unit, an attitude constraint generation unit, and a constraint construction unit. The model receiving unit receives a spatial environment model; the spatial environment model includes the image boundary labels, electromagnetic signal amplitude, electromagnetic gradient value and corresponding spatial coordinates of each three-dimensional grid unit; The obstacle recognition unit filters three-dimensional grid units with image boundary labels and three-dimensional grid units with electromagnetic signal amplitude exceeding a set threshold in the spatial environment model, and defines the set of center coordinates of the filtered three-dimensional grid units as the obstacle point set. The obstacle boundary construction unit performs a three-dimensional geometric envelope algorithm on the set of obstacle points to generate a minimum convex boundary volume containing all obstacle points; The gradient analysis unit calculates the rate of change of electromagnetic signal gradient between any adjacent grid cells in the spatial environment model, marks the corresponding grid area where the rate of change of electromagnetic signal gradient exceeds a preset threshold as an electromagnetic interference area, and records the spatial coordinate range of the interference area in three-dimensional space. The attitude restriction generation unit combines the spatial location of the electromagnetic interference zone with the coordinate axis direction marked in the spatial environment model to calculate the allowable attitude orientation range of the end effector in the corresponding spatial region, and defines the spatial attitude that does not meet the corresponding attitude orientation as an infeasible region. The constraint construction unit maps the minimum convex boundary volume, the spatial coordinate range of the electromagnetic interference zone, and the direction of the infeasible region into convex constraint forms in the path search variable space, and constructs a dynamic convex constraint set that includes geometric obstacle constraints, electromagnetic strength constraints, and attitude restriction constraints. Among them, the geometric obstacle constraint means that path points must not fall inside the polyhedron defined by the convex hull; Electromagnetic strength constraint means that path points are not allowed to appear in regions where the gradient strength exceeds a threshold. Attitude constraints refer to the requirement that the attitude vector of a path point within a specific region must belong to the set of allowed directions.

7. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The path evolution module includes an initial path construction unit, a differential perturbation generation unit, a constraint feedback adjustment unit, a cross-combination unit, and an effectiveness screening unit. The initial path construction unit generates multiple initial path solutions that satisfy convex constraints in the path search variable space. For each initial path solution, the differential perturbation generation unit constructs a path difference vector based on multiple other path solutions, and synthesizes the difference vector with the base path to generate a first mutated path candidate solution; After the first variant path candidate solution is generated, the constraint feedback adjustment unit receives the convex constraint set update information, judges the trend of the path point deviating from the feasible region based on the update information, and dynamically corrects the differential perturbation direction to obtain the second variant path candidate solution. The steps to correct the direction of differential perturbation are as follows: select two or more different path individuals from the current population, calculate the vector difference between the corresponding path points of these path individuals, and add the weighted difference vector as a perturbation term to another path individual to generate a new second mutation path candidate solution; The cross-combination unit randomly combines the current path initial solution and the second variant path candidate solution at the path point level to generate a cross-path solution that integrates the original structure and the perturbation structure. The validity screening unit performs convex constraint checks on each path point in the cross-path solution, identifies the set of path points that meet the constraints, eliminates infeasible path points, and constructs a continuous structure to obtain the final variant path candidate solution.

8. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 7, characterized in that, The candidate solutions for the variant paths include multiple consecutive path points, each of which has coordinate information in the path search space.

9. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The path constraint module includes a path candidate receiving unit, a constraint set loading unit, a path point parsing unit, a path projection calculation unit, and a path reconstruction unit; The path candidate receiving unit receives the variant path candidate solution; The constraint set loading unit receives the dynamic convex constraint set of the current optimization round; the convex constraint set includes multiple types of environmental constraints formed by obstacle boundaries, electromagnetic interference zones, and attitude direction restrictions; The path point parsing unit parses each path point in the candidate solution of the variant path one by one, maps each path point to the constraint space defined by the convex constraint set, and establishes the constraint correspondence between the path point and the constraint set. The path projection calculation unit performs a stepwise constraint projection operation on each path point according to the constraint correspondence, and sequentially maps the path point to all convex constraint sets until the spatial distance between two consecutive updates of the path point is lower than a preset threshold. The path reconstruction unit summarizes and sorts the converged path point results, reconnects them according to the spatial sequence of the original path, and generates a complete path structure within the range of the convex constraint set.

10. The industrial robotic arm path optimization system based on electromagnetic scanning according to claim 1, characterized in that, The path output module includes an adaptive evaluation unit and a path optimization unit; The adaptive evaluation unit evaluates each path in the path set output by the path constraint module. The evaluation includes path length index, electromagnetic interference accumulation index, attitude deviation index, smoothness index, and constraint proximity index. Among them, the path length index is the total movement distance of the path in three-dimensional space; The cumulative electromagnetic interference index is the degree of exposure of a path within an electromagnetic interference area; The attitude deviation index is the average angle of deviation between the end attitude and the target attitude during the execution of the path; The smoothness index is the rate of change of the angle between adjacent segments of the path; The constraint proximity index is the distribution of the minimum distance from each path point to the constraint boundary in the path; The path selection unit calculates the total adaptability score of each path using a weighted comprehensive scoring method based on the adaptability index, and determines the path with the highest score as the optimal path, which serves as the target path for the industrial robotic arm control execution.

Images