Wave-shaped guardrail spraying robot's spraying control method

Abstract

This invention discloses a spraying control method for a robotic arm used for spraying corrugated guardrails, relating to the fields of industrial automation and robot control technology. The method includes acquiring data on the relative positional changes between the robotic arm and the corrugated guardrail surface during rotation using a pre-established robotic arm rotation trajectory model, recording changes in spraying angle and dynamic distance adjustments at each time point, and establishing a dynamic spatial mapping relationship for the robotic arm's rotation state for spraying control. By triggering secondary respray control commands based on coating quality consistency feedback, the reliability and stability of the respray effect can be further guaranteed. This significantly improves the equipment utilization rate of the robotic arm and the automation and intelligence level of the spraying operation while enhancing coating uniformity and spraying quality consistency.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and robot control technology, specifically to a spraying control method for a wave-shaped guardrail spraying robot. Background Technology

[0002] Corrugated guardrails, as an important component of highway and urban road traffic safety facilities, primarily reduce the severity of traffic accidents by absorbing collision energy and guiding vehicle direction. Because corrugated guardrails are typically exposed to the elements for extended periods, they are susceptible to corrosion from rainwater, salt spray, ultraviolet radiation, and pollutants. Therefore, the quality of the anti-corrosion coating applied to their surface directly affects the guardrail's corrosion resistance, service life, and overall protective effect. Consequently, effectively controlling the coating process to achieve uniform, consistent, and reliable coating quality is a widely recognized technical challenge in the current guardrail manufacturing and maintenance field.

[0003] Current corrugated guardrail painting operations largely rely on painting robots to complete the coating process according to preset trajectories. The painting control methods are typically based on fixed posture parameters and static process settings, making it difficult to fully consider factors such as the guardrail surface structure characteristics, changes in the robot's posture during painting, and dynamic changes in the relative position between the spray gun and the guardrail surface. In actual production, due to factors such as equipment precision, environmental disturbances, and workpiece clamping errors, quality problems such as uneven coating thickness, localized missed areas, or runs are still prone to occur during the painting process. To ensure painting quality, existing technologies often involve manual visual inspection or independent testing equipment to check the guardrail surface after the painting process is completed, and then manual or semi-automatic repainting is performed after defects are found. This separation of painting, inspection, and repainting results in a lack of real-time feedback and dynamic control capabilities in the painting process, increasing the number of workpiece handling operations and production cycle time, and making it difficult to guarantee the accuracy of repainting locations and the consistency of painting quality. Furthermore, during the return journey of the spraying robot after completing the spraying of a single corrugated guardrail, the robot's posture and motion state continuously change, and the relative angle and distance between the spray gun and the guardrail surface are constantly adjusted. Existing spraying control methods typically fail to effectively utilize this dynamic process and lack the ability to perceive and judge coating quality in real time under changing robot motion conditions, and adjust spraying control parameters accordingly. This makes it difficult for the spraying robot to perform timely and accurate targeted respraying control when facing coating defects, and the equipment utilization efficiency and automation level need further improvement. Summary of the Invention

[0004] The purpose of this invention is to provide a spraying control method for a wave-shaped guardrail spraying robot, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a spraying control method for a corrugated guardrail spraying robot, comprising: acquiring relative position change data between the robot and the corrugated guardrail surface during rotation by using a pre-established robot rotation trajectory model; recording spraying angle changes and distance dynamic adjustment information at each time point to establish a dynamic spatial mapping relationship for the robot's rotation state for spraying control; based on the dynamic spatial mapping relationship, using a dynamic detection system to scan the corrugated guardrail surface in real time to acquire surface image data, and initially marking coating defect areas in the image to determine the initial position range for surface defect capture for spraying control; performing depth analysis on the initially marked defect areas, extracting specific feature information of the defects using real-time data processing technology, and combining the spatial position coordinate mapping data to determine the three-dimensional position of the defects on the corrugated guardrail surface to obtain accurate defect positioning results for spraying control; generating planning data for the reverse motion path for spraying control based on the accurate defect positioning results, and adjusting the path parameters according to the dynamic characteristics of the robot rotation process to match the path planning with the current operating state of the robot, and determining the execution order of the re-spraying task.

[0006] Preferably, the step of acquiring relative position change data between the robot and the corrugated guardrail surface during rotation by using a pre-established robot rotation trajectory model, and recording spray angle changes and distance dynamic adjustment information at each time point to establish a dynamic spatial mapping relationship for robot rotation state for spraying control includes: acquiring relative position change data between the robot and the corrugated guardrail surface using a pre-established rotation trajectory model; recording spray angle changes and distance dynamic adjustment information at each time point based on the relative position change data; establishing a dynamic spatial mapping relationship for robot rotation state based on the spray angle changes and distance dynamic adjustment information; and using the dynamic spatial mapping relationship to compensate for the curvature of the corrugated guardrail and determine an optimized spraying path scheme.

[0007] Preferably, the step of using a dynamic detection system to scan the surface of the wave-shaped guardrail in real time according to the dynamic spatial mapping relationship, acquiring surface image data, and initially marking the coating defect areas in the image to determine the initial position range for capturing surface defects for spraying control includes: driving the dynamic detection system to acquire real-time surface image data according to a pre-established dynamic spatial mapping relationship, the real-time surface image data containing the curvature features of the guardrail; identifying abnormal pixel sets in the real-time surface image data to generate a binary mask image of the coating defect area; performing registration calculations between the binary mask image of the coating defect area and the dynamic spatial mapping relationship, establishing a corresponding index table, and generating three-dimensional bounding box data according to the corresponding index table; and using the three-dimensional bounding box data to determine the initial position range for capturing surface defects for spraying control.

[0008] Preferably, the step of performing depth analysis on the initially marked defect area, extracting specific feature information of the defect using real-time data processing technology, and combining it with spatial coordinate mapping data to determine the three-dimensional position of the defect on the surface of the wave-shaped guardrail, and obtaining a precise defect location result for spraying control includes: acquiring a refined feature descriptor of the defect generated based on the initially marked area; determining a precise two-dimensional contour mask of the defect area based on the refined feature descriptor; mapping the precise two-dimensional contour mask onto a synchronously acquired depth data plane; extracting the depth value set within the corresponding range and calculating the normal vector distribution of the local surface; using a pre-calibrated spatial mapping matrix, converting the depth value set and normal vector distribution into a three-dimensional point cloud cluster in the coordinate system of the wave-shaped guardrail surface; and performing surface fitting calculation on the three-dimensional point cloud cluster to generate a precise defect location result containing three-dimensional position coordinates and attitude information.

[0009] Preferably, the step of generating planning data for reverse motion paths for spraying control based on the accurate defect location results, and adjusting path parameters according to the dynamic characteristics of the robot's rotation process to match the path planning with the robot's current operating state, and determining the execution order of the respraying task includes obtaining the accurate defect location results; calculating the reverse kinematics solution set based on the accurate defect location results; generating initial reverse motion path planning data based on the reverse kinematics solution set; extracting parameters from the initial reverse motion path planning data to calculate joint torque load; generating corrected path parameters if the joint torque load exceeds a threshold; and correcting the corrected path parameters with the robot's current operating state to determine the execution order of the respraying task that matches the robot's current operating state.

[0010] Preferably, the method further includes driving the robot arm to move through path planning data, generating control commands for adjusting the spraying posture for spraying control, adjusting the spray gun angle and the distance between the spray gun and the corrugated guardrail surface according to the defect location, so that the change in spraying angle and the dynamic adjustment of distance are consistent, and obtaining a spraying control execution scheme for the spraying action. Specifically, this includes obtaining a discrete trajectory point sequence in the path planning data, calculating the surface normal vector and tangential curvature feature data at the defect location based on the discrete trajectory point sequence; solving the six-degree-of-freedom target pose matrix and target spraying distance value based on the surface normal vector and tangential curvature feature data; calculating the attitude error quaternion between the actual end pose and the six-degree-of-freedom target pose matrix, and generating a spraying posture adjustment control command based on the attitude error quaternion; converting the spraying posture adjustment control command into target torque and speed control signals, and obtaining a spraying control execution scheme for the spraying action.

[0011] Preferably, the method further includes real-time monitoring of the robot's posture data during the re-spraying process according to the spraying control execution scheme, and combining the coating quality consistency information fed back by the dynamic detection system to make spraying control judgments. When the difference between the re-sprayed area and the surrounding coating exceeds a preset threshold, a secondary re-spraying control command is triggered to determine whether the re-spraying effect meets the standard. Specifically, this includes acquiring the spatial pose data of the end effector and the surface optical reflectivity distribution map during the robot's re-spraying, performing spatiotemporal registration of the surface optical reflectivity distribution map and the end effector spatial pose data to calculate the texture feature similarity coefficient; if the texture feature similarity coefficient is lower than the preset standard, determining the secondary re-spraying entry angle and spraying distance based on the coordinates of the local area with significant differences; generating a secondary re-spraying control command based on the secondary re-spraying entry angle and spraying distance, and driving the robot to perform the repair action according to the secondary re-spraying control command, acquiring images to calculate the feature fusion degree to obtain the verification status of whether the re-spraying effect meets the standard.

[0012] Preferably, the method further includes obtaining the final coating quality consistency data based on the execution result of the secondary spray control command, and recording the time consumption information of the integrated spray control process of detection and respray to improve equipment utilization efficiency, which serves as the basis for optimizing the automation level of the spray control method. Specifically, this includes obtaining the execution status feedback sequence and process timestamp after the execution of the secondary spray control command, generating coating quality consistency data based on the execution status feedback sequence, associating and mapping the coating quality consistency data with the process timestamp, and extracting the time consumption of each stage to calculate the equipment utilization efficiency index.

[0013] Preferably, the step of obtaining the final coating quality consistency data based on the execution result of the secondary spray control command, and recording the time consumption information of the integrated spray control process of detection and re-spraying to improve equipment utilization efficiency, as a basis for optimizing the automation level of the spray control method, also includes locating automation bottleneck nodes based on equipment utilization efficiency indicators and generating an optimization weight matrix.

[0014] Preferably, the step of obtaining the final coating quality consistency data based on the execution result of the secondary spraying control command, and recording the time consumption information of the integrated spraying control process of detection and respraying to improve equipment utilization efficiency, as a basis for optimizing the automation level of the spraying control method, also includes calculating the automation level score by combining the optimization weight matrix and the coating quality consistency data, as a basis for optimizing the automation level of the spraying control method.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0016] The spraying control method of this corrugated guardrail spraying robot introduces a dynamic spatial mapping relationship during the robot's rotation, achieving unified modeling and control of the robot's posture changes, spray gun angle, and spraying distance. This frees spraying control from being limited to fixed trajectories and static parameters, fundamentally improving the adaptability of the spraying process to complex working conditions. Simultaneously, a dynamic detection system performs real-time scanning and defect identification on the corrugated guardrail surface, directly completing defect location, path planning, and respray control during the spraying process. This creates a closed-loop control process for spraying, detection, and respray, effectively avoiding the problems of repetitive handling, low efficiency, and inconsistent quality caused by traditional separate detection and respray methods. Furthermore, by triggering secondary respray control commands based on coating quality consistency feedback, the reliability and stability of the respray effect can be further guaranteed. This significantly improves the robot's equipment utilization rate and the automation and intelligence level of the spraying operation while enhancing coating uniformity and spraying quality consistency. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the corrugated guardrail spraying production line of the present invention.

[0018] Figure 2 This is a flowchart of the spraying control method of the present invention.

[0019] Figure label:

[0020] 1. Corrugated guardrail; 2. Conveyor line; 3. Robotic arm. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, the specific application scenario of the technical solution provided by the present invention is as follows: during the process of spraying the anti-corrosion coating on the surface of the corrugated guardrail 1, multiple corrugated guardrails 1 are suspended sequentially on the conveyor line 2, and then pass through multiple robotic arms 3 that spray the surface anti-corrosion agent in sequence to perform the initial spraying and touch-up spraying of the surface anti-corrosion coating.

[0023] like Figure 2As shown, this invention provides a technical solution: a spraying control method for a corrugated guardrail spraying robot, comprising: acquiring relative position change data between the robot 3 and the surface of the corrugated guardrail 1 during rotation by using a pre-established rotation trajectory model of the robot 3; recording spraying angle changes and distance dynamic adjustment information at each time point to establish a dynamic spatial mapping relationship of the robot 3 in the rotation state for spraying control; based on the dynamic spatial mapping relationship, using a dynamic detection system to scan the surface of the corrugated guardrail 1 in real time to acquire surface image data, and initially marking the coating defect areas present in the images to determine the initial position range of surface defect capture for spraying control; performing depth analysis on the initially marked defect areas, extracting specific feature information of the defects using real-time data processing technology, and combining the spatial position coordinate mapping data to determine the three-dimensional position of the defects on the surface of the corrugated guardrail 1 to obtain accurate defect positioning results for spraying control; and generating planning data for the reverse motion path for spraying control based on the accurate defect positioning results. The path parameters are adjusted based on the dynamic characteristics of the robot arm 3 during its rotation to match the path planning with the current operating state of the robot arm 3, thus determining the execution sequence of the respraying task. The robot arm 3 is driven to move using the path planning data, generating control commands for adjusting the respraying posture. The spray gun angle and the distance between the spray gun and the surface of the corrugated guardrail 1 are adjusted for the defect location, ensuring that the changes in spray angle and the dynamic adjustment of distance are consistent, resulting in a spraying control execution scheme for the respraying action. Based on the spraying control execution scheme, the posture data of the robot arm 3 during the respraying process is monitored in real time, and spraying control judgment is made in conjunction with the coating quality consistency information fed back by the dynamic detection system. When the difference between the resprayed area and the surrounding coating exceeds a preset threshold, a secondary respraying control command is triggered to determine whether the respraying effect meets the standard. Based on the execution result of the secondary respraying control command, the final coating quality consistency data is obtained, and the time consumption information of the integrated detection and respraying spraying control process is recorded to improve equipment utilization efficiency, serving as a basis for optimizing the automation level of the spraying control method.

[0024] In the above implementation, by pre-establishing a rotation trajectory model of the robotic arm 3, the spatial relationship between the rotational motion of the robotic arm 3 and the wave-shaped guardrail 1 is parameterized, ensuring that the spray gun remains in a calculable and predictable spatial state during its rotation. The dynamic spatial mapping relationship is used to uniformly describe the changes in the spray gun position, spraying angle, and spraying distance over time, providing a spatial basis for subsequent defect identification and respray control.

[0025] The dynamic detection system continuously scans the surface of the corrugated guardrail 1 during the spraying or touch-up spraying process, acquiring the coating status in real time through image acquisition and processing technology, and initially marking any abnormal areas. Subsequently, real-time data processing technology is used to extract features from the defective areas, and combined with coordinate mapping in the dynamic spatial mapping relationship, the defect is accurately converted from a two-dimensional image to a three-dimensional spatial location.

[0026] Based on this, through reverse motion path planning, the spatial location of the defect is mapped back to the motion trajectory of robot arm 3. The path parameters are dynamically corrected according to the current rotation state of robot arm 3, ensuring that the robot arm 3 completes the repainting action without interrupting the overall painting rhythm. Finally, through closed-loop control of attitude monitoring and coating quality feedback, real-time evaluation and adjustment of the repainting effect are achieved.

[0027] By using a pre-established rotation trajectory model of the robotic arm 3, the relative position change data between the robotic arm 3 and the surface of the corrugated guardrail 1 during rotation is obtained. Spraying angle changes and distance dynamic adjustment information are recorded at each time point. A dynamic spatial mapping relationship for the robotic arm 3 in its rotation state is established for spraying control. This includes obtaining the relative position change data between the robotic arm 3 and the surface of the corrugated guardrail 1 using the pre-established rotation trajectory model; recording the spraying angle changes and distance dynamic adjustment information at each time point based on the relative position change data; establishing a dynamic spatial mapping relationship for the robotic arm 3 in its rotation state based on the spraying angle changes and distance dynamic adjustment information; and using the dynamic spatial mapping relationship to compensate for the curvature of the corrugated guardrail 1 and determine an optimized spraying path scheme.

[0028] In this embodiment, the establishment of the rotation trajectory model first revolves around "how the rotational motion of the robot arm 3 corresponds to the surface of the wave-shaped guardrail 1 in space." During implementation, a unified calibration is first completed between the coordinates of the robot arm 3 body, the spray gun installation coordinates, the guardrail fixture coordinates, and the dynamic detection system coordinates: At least three calibration points with stable geometric features are selected on the guardrail fixture. The end effector of the robot arm 3 is sequentially aligned with these calibration points, and the joint position data of the robot arm 3 at each alignment moment is read. Combined with the assembly dimension data of the spray gun installation position, the posture description of the spray gun's spray direction in the robot arm 3 base coordinates is obtained. The dynamic detection system simultaneously identifies the above calibration points and outputs their position descriptions in the detection coordinates. After aligning and solving the two sets of descriptions, a stable transformation relationship from "detection coordinates to robot arm 3 base coordinates" is formed. The number of calibration points is determined to be at least three (determined by: three points satisfying the minimum constraint of plane and pose alignment; if strong vibration or reflection on site causes recognition jitter, the number of calibration points is increased to five, and multi-point fitting is used to suppress errors). At this point, the rotary trajectory model has a unified coordinate system, and all subsequent spatial calculations are completed under the same coordinate system, avoiding deviations in spraying angle and distance calculations caused by coordinate drift.

[0029] After calibration, the "relative position change data between the robot arm 3 and the guardrail surface" is obtained through a pre-established rotation trajectory model. The process is as follows: During rotation, the robot arm 3 controller outputs joint position data at a fixed sampling period. The sampling period parameter is determined to be no greater than 0.01 seconds (determined by using the maximum angular velocity of the robot arm 3 rotation and the minimum allowable spatial resolution of the spraying path as constraints, ensuring that the projected displacement of the spray gun endpoint on the guardrail surface corresponding to two adjacent samples does not exceed the allowable resolution; the allowable resolution is jointly determined by the spray width and the overlap requirement of adjacent spray trajectories. The overlap requirement is determined according to the coating uniformity target, with an overlap coefficient of 0.3 to 0.6. The overlap coefficient is determined based on the target coating thickness fluctuation tolerance and the spray edge attenuation characteristics, and is fixed on-site after a one-time calibration of the spray edge attenuation curve). The controller converts the joint position data into a real-time position and attitude description of the spray gun end in the robot arm 3 base coordinates. The dynamic detection system simultaneously outputs a point cloud or depth map of the guardrail surface and reconstructs the guardrail surface in a unified coordinate system. Subsequently, a nearest neighbor search was performed between the spray gun tip and the guardrail surface: the surface point closest to the spray gun tip was selected from the guardrail surface point cloud as the "corresponding surface point," and the surface normal direction was obtained within the neighborhood of this surface point through local plane fitting; the spatial distance from the spray gun tip to the corresponding surface point was used as the spraying distance, and the angle between the spray gun spraying direction and the surface normal direction was used as the spraying angle. The spraying distance parameter was initially set to 0.2 meters to 0.35 meters (determined based on constraints of paint atomization particle size, spray width coverage, and rebound rate; too small a distance leads to increased risk of local accumulation and sagging, while too large a distance leads to drying of atomized particles and decreased adhesion rate; this distance range was calibrated once upon equipment delivery based on the atomization characteristic curve of the spray gun model and the coating adhesion rate target, and only fine-tuned during production when the paint viscosity or ambient temperature and humidity change). The spraying angle parameter is determined to be no greater than 30 degrees from the surface normal when it appears (determined by: the more the spraying direction deviates from the normal, the more obvious the masking effect of the coating at the junction of the crest and trough, resulting in thinner coating at the edges; the upper limit of the angle is determined by comprehensively considering the masking sensitivity of the steepest slope segment of the guardrail waveform, the atomization cone angle of the spray gun, and the tolerance of the target thickness uniformity, and is fixed on-site after measuring the coating thickness profile of the steepest slope segment). The above calculation is completed once at each time point, thus obtaining a continuous sequence of "relative position change data, spraying angle change, and distance dynamic adjustment information".

[0030] When recording the spraying angle changes and distance dynamic adjustment information at each time point for relative position change data, the implementation process is further refined into three consecutive actions: "recording, correction, and smoothing". First, the controller uses the spraying angle and spraying distance at each sampling moment as the original recorded values. Second, outlier removal is performed on the original recorded values. The outlier removal threshold is determined when "the angle jump of three adjacent sampling moments exceeds 5 degrees or the distance jump exceeds 0.02 meters" (threshold determination method: using the upper limit of the end-effector posture change of the robotic arm 3 under maximum acceleration as a physical constraint, and simultaneously using detection...). The system's depth noise upper limit serves as a measurement constraint, with the stricter one being chosen. This threshold was obtained during equipment acceptance testing through noise statistics from no-load rotation and static guardrail scanning and was solidified as a process parameter. Furthermore, the rejected sequence undergoes time-consistency smoothing to ensure the spray gun attitude adjustment aligns with the dynamic characteristics of the robot arm 3's rotation. The smoothing window length parameter is determined to be between 0.03 and 0.08 seconds when it appears (determined by covering at least three sampling points to reduce detection noise while not crossing the main dynamic response cycle of the robot arm 3's control loop to avoid introducing lag and insufficient attitude tracking). After the above processing, the spray angle change and distance dynamic adjustment information possess stability suitable for mapping and control.

[0031] When establishing the "dynamic spatial mapping relationship of the robot arm 3 in rotation state" based on the spraying angle change and distance dynamic adjustment information, the implementation process adopts the "indexing, association, and interpolation" method to form a real-time queryable mapping structure. The index uses a dual index of the rotation process time node and rotation angle, with the rotation angle calculated from the position data of the robot arm 3's rotation joint; the associated content includes the spray gun end position, spray gun spraying direction, corresponding surface point position, surface normal direction, spraying angle, spraying distance, and the current speed and acceleration status of the controller. To ensure that the mapping remains continuously usable when the rotation speed changes, an interpolation rule is established between adjacent index points: when the controller reading is between two recorded indices, the pose and distance change trends of adjacent indices are used for continuous transition, so that the spraying angle and spraying distance change continuously with time, avoiding abrupt changes that could lead to overspraying or missed spraying. The interpolation step size parameter is determined to be no greater than 0.005 seconds when it occurs (determined by ensuring that the projected displacement of the spray gun after interpolation is less than half of the aforementioned allowable resolution, so as to ensure that the path adjustment after curvature compensation will not produce jagged boundaries). The resulting dynamic spatial mapping relationship can map the "current state of robot arm 3" to the "spraying posture requirement of the spray gun relative to the guardrail surface" at any rotation moment.

[0032] After obtaining the dynamic spatial mapping relationship, the process of compensating for the curvature of the waveform guardrail 1 and determining the optimized spraying path is as follows. After the dynamic detection system outputs the point cloud of the guardrail surface, an ordered profile sequence is first generated along the length of the guardrail: taking the center line of the guardrail as the reference, a cross-sectional profile is cut at fixed intervals. The profile spacing parameter is determined to be 0.01 meters to 0.03 meters when it appears (determination method: the profile spacing is linked with the spray width and path spacing, and the profile spacing is less than the path spacing to ensure the capture and coverage of waveform changes; when the reflection of the guardrail surface causes the point cloud to be sparse, the profile spacing takes a larger value to improve the fitting stability). Within each profile, the local curvature of the waveform curve is estimated: several consecutive points in the profile are selected to form a local segment. The length parameter of the local segment is determined to be 0.2 to 0.4 times the period of the waveform when it appears (determination method: the length is too short and sensitive to noise, and the length is too long and will mix the peaks and troughs, resulting in curvature distortion; this range is determined based on the standard size of the guardrail waveform and the density of detection points, and is fixed during the first modeling). After curve fitting of a local segment, the curvature change trend of that segment is obtained, which is used to identify "areas with large curvature changes" and "areas with gentle curvature changes". The curvature change judgment threshold is determined when "the difference in curvature trends between two adjacent local segments causes the expected spraying distance deviation to exceed 0.01 meters" (threshold determination method: based on the sensitivity of spraying distance deviation to coating thickness uniformity, the thickness fluctuation corresponding to 0.01 meters can still be eliminated by attitude compensation within the target coating tolerance, exceeding this deviation will significantly increase the risk of thin coating at the peak edge; this threshold is obtained through thickness detection and distance disturbance test during the process finalization stage, but is managed as a fixed process parameter in the production document). When it is determined that a region with large curvature changes has been entered, the path optimization scheme performs two types of compensation actions on the spray gun attitude: the first is distance compensation, which moves the end of the spray gun slightly forward or backward along the surface normal direction, so that the actual spraying distance returns to the vicinity of the center of the aforementioned distance interval; the second is angle compensation, which makes the spraying direction closer to the surface normal direction, and prioritizes suppressing the shading effect at peaks, troughs and transition slope sections. The process for determining the compensation amount is as follows: the current position of the spray gun end and the corresponding surface normal are obtained from the dynamic space mapping relationship; the difference between the current spraying distance and the target spraying distance is calculated, and this difference is directly used as the displacement command along the normal; the difference between the current spraying angle and the upper limit of the target spraying angle is calculated, and this difference is converted into the rotation adjustment amount of the end attitude. The attitude adjustment amount is constrained by the upper limit of the speed of the 3 joints of the robot arm. When the upper limit of the speed parameter appears, it is taken as 0.8 times the factory limit of the controller (determination method: retain 0.2 times the dynamic margin to resist the load fluctuation and control delay during the rotation process, and avoid insufficient following leading to angle exceeding the limit).

[0033] The spray path optimization scheme further constrains the "coverage integrity and efficiency" of the path based on curvature compensation. During implementation, a nominal spray trajectory is first generated: parallel trajectories are arranged along the length of the guardrail, with the trajectory spacing parameter determined to be 0.4 to 0.7 times the spray width (determined by: the trajectory spacing being consistent with the aforementioned overlap coefficient; a larger overlap coefficient results in a smaller trajectory spacing, higher thickness uniformity, but decreased efficiency; this range is jointly determined by the target production capacity and thickness uniformity target, and fixed after production line cycle time calculation). Subsequently, curvature compensation correction is performed on each nominal trajectory: the trajectory points are projected onto the guardrail surface, and the spray angle and spray distance requirements for each trajectory point are calculated through dynamic spatial mapping, forming a compensation trajectory consistent with the rotation state. Finally, a continuity check is performed: if the attitude change between adjacent compensation trajectory points exceeds the aforementioned angle jump threshold or distance jump threshold, a transition point is inserted between the two points, and the attitude requirements are recalculated until the continuity requirements are met.

[0034] Based on the dynamic spatial mapping relationship, a dynamic detection system is used to scan the surface of the waveform guardrail 1 in real time to acquire surface image data. Preliminary marking of coating defect areas in the images is then performed to determine the initial location range for surface defect capture used in spraying control. This includes driving the dynamic detection system to acquire real-time surface image data based on a pre-established dynamic spatial mapping relationship. The real-time surface image data contains the guardrail curvature features. Abnormal pixel sets in the real-time surface image data are identified to generate a binary mask image of the coating defect area. The binary mask image of the coating defect area is registered with the dynamic spatial mapping relationship to establish a corresponding index table, and three-dimensional bounding box data is generated based on the corresponding index table. The three-dimensional bounding box data is then used to determine the initial location range for surface defect capture used in spraying control.

[0035] In this embodiment, the real-time scanning of the dynamic detection system does not operate at an independent rhythm, but is constrained by the dynamic spatial mapping relationship and synchronized with the rotation process of the robot arm 3, thereby ensuring that the "collection position, collection direction, and collection time node" are consistent with the spatial state of the spraying control. Specifically, at each sampling node of the rotation operation, the robot arm 3 controller outputs the current rotation state index and end pose state. After receiving the index, the dynamic detection system extracts the guardrail surface spatial window information corresponding to the same node from the dynamic spatial mapping relationship. This spatial window is jointly defined by the "relative position change data of the spray gun end to the guardrail surface, spraying angle change, and spraying distance dynamic adjustment information," and is used to constrain the detection field of view coverage, so that the detection field of view is aligned with the spraying-affected area. The collection rhythm parameter is set to no more than 0.02 seconds (determined by: based on the relative motion speed of the guardrail surface during rotation, combined with the minimum detectable defect size specified by the quality standard, ensuring that the projection displacement of two adjacent frames on the guardrail surface is less than 1 / 2 of the minimum detectable size; the minimum detectable size is determined by the acceptance requirements and the target for missing detection risk control, and is formed by sample defect verification before the production line is put into operation). To ensure that the real-time surface image data includes the curvature features of the guardrail, the dynamic detection system outputs surface structure information simultaneously with the image frame output. The structure information is expressed using a depth map or an equivalent surface point set. When the output is a surface point set, the point density is set to no less than 25 valid points per square centimeter (determined by: based on the minimum curvature change scale of the wave crests, troughs and transition slope areas of the waveform guardrail 1, ensuring that each curvature change segment has a sufficient number of sampling points during local surface reconstruction, avoiding jumps in curvature estimation due to sparse points; this point density is selected through continuous scanning of the point cloud integrity assessment after the on-site installation height and field of view are determined).

[0036] After acquiring real-time surface image data, initial defect labeling is completed through abnormal pixel set identification. The core objective is to distinguish between "non-defect texture fluctuations, reflective bright spots, and sensor noise" and "real coating defects," and to converge real defects into a binary mask that can be used for spatial registration. In specific implementation, illumination consistency correction is first performed: within the same frame, the brightness gradient is compensated based on the current surface normal change trend given by the dynamic spatial mapping relationship to avoid natural brightness and darkness changes caused by the curved surface of the guardrail being misjudged as defects; the correction reference area is taken from the normal coating area with continuous texture and stable reflection within the same frame. The normal coating area is jointly determined by "local texture continuity, small neighborhood brightness fluctuations, and historical consistency with the rotation state index." After correction, for each pixel, the comprehensive difference value between it and its neighboring pixels in terms of brightness difference, color difference, and texture difference is calculated. The difference value is calculated by "first calculating the local mean in a fixed neighborhood, then calculating the degree of deviation of the pixel from the local mean, and then superimposing the degree of texture direction change," ensuring that the source of difference simultaneously covers color difference and roughness change. The anomaly threshold is set to twice the upper limit of the normal area difference (threshold determination method: before production line startup, collect no less than 100 defect-free sample images, statistically analyze the maximum stable fluctuation range of the normal area difference value, and use this upper limit as the baseline for noise and reflection; set the threshold to twice this upper limit to suppress false alarms caused by reflection and random noise, while maintaining the difference response to thin coating, missing coating, blistering, and pinholes; the multiple is determined during the trial production stage after comprehensive evaluation of the false alarm rate and the false alarm rate). After pixels that meet the anomaly threshold are included in the anomaly pixel set, connectivity aggregation is performed to merge spatially adjacent anomaly pixels with consistent difference characteristics into connected components, and the size of the connected components is screened, with the size threshold set to no less than 9 pixels (threshold determination method: 9 pixels correspond to the minimum reliable imaging area of ​​the detection system under the current installation distance and resolution conditions; connected components smaller than this area mainly come from sensor noise spikes and small reflective points, and removing them can significantly reduce false detections). The selected connected components are subjected to hole filling and boundary smoothing. Hole filling aims to close the holes by targeting the "zero-value regions enclosed within the connected components." The boundary smoothing scale is set to no more than 2 pixels (determined by the following principle: too large a scale will cause the defect boundary to expand outward and engulf the normal coating area, while too small a scale cannot suppress the amplification effect of jagged boundaries on subsequent 3D bounding boxes; 2 pixels matches the magnitude of boundary jitter in the detection system, which can suppress jitter without significantly changing the true defect boundary). This forms a binary mask image of the coating defect area, with defect areas marked as 1 and non-defect areas marked as 0, completing the initial labeling of defects and providing input for registration calculations.

[0037] The registration calculation between the binarized mask image and the dynamic spatial mapping relationship is used to transform the "defect region in the image" into the "defect spatial range in the 3-coordinate system of the robot arm", thereby constructing the initial position range for defect capture for spraying control. In specific implementation, time consistency is first established: each frame image is bound to the rotation state index corresponding to its acquisition time, and the robot arm 3 end pose, the coordinate transformation relationship between the detection system and the robot arm 3, and the local geometric description of the guardrail surface are extracted from the dynamic spatial mapping relationship based on this index. Then, the pixel to 3D surface point is calculated for each defect pixel in the mask: first, the pixel position is converted into the detection line of sight based on the imaging calibration parameters of the detection system. The imaging calibration parameters are derived from the joint results of factory calibration and on-site calibration (determination method: factory calibration provides lens distortion and basic imaging model, on-site calibration is used to eliminate small offsets caused by installation angle deviation and temperature drift, and on-site calibration is performed by acquiring no less than 20 frames through calibration board and setting the value after the recognition error converges). After obtaining the line-of-sight direction, if the detection system synchronously outputs depth information, the pixel depth and line-of-sight direction are combined to obtain the corresponding 3D point. If the detection system outputs a set of surface points, the line-of-sight direction and the reconstructed guardrail surface are intersected to obtain the 3D point. The intersection process is completed in the order of "searching forward along the line-of-sight direction for the nearest intersection point with the guardrail surface and checking the consistency of the local surface normal". To avoid depth anomalies caused by reflection and occlusion, the depth back-calculated points and surface reconstructed points are checked for consistency. The check threshold is set to no more than 0.005 meters (the threshold is determined by the superposition of the upper limit of depth noise and the upper limit of surface reconstruction fitting error. If the threshold is exceeded, it indicates that the depth of the pixel or the intersection result is affected by reflection and occlusion. Continuing to use it for bounding box calculation will amplify the spatial range and introduce the risk of false spraying. Therefore, it is judged as an invalid point and discarded). For each valid defect pixel, an index relationship is formed, which includes the pixel position, 3D surface point position, acquisition time index, and rotation state index. This index relationship is included in the corresponding index table to trace any subsequent defect pixel back to a spatial point in the robot's 3-coordinate system. After completing the mapping of all pixels in the same connected region, all 3D points corresponding to the connected region are aggregated into a defect point set. The enclosing range of the defect point set is calculated within the robot's 3-coordinate system: first, the minimum and maximum coordinate values ​​of the defect point set in the three coordinate directions are counted respectively; then, the length, width, and height of the enclosing range are determined using these two sets of extreme values. At the same time, the main direction and normal fluctuation range of the surface normal within the enclosing range are counted to provide a basis for the initial estimation of the spray gun attitude. The margin of the encirclement range is set to 0.01 meters to 0.03 meters (determined by: the margin is used to cover the boundary error caused by the smoothness of the mask boundary, the boundary missing caused by the sparse point cloud, the tracking error of the robot arm 3 and the detection jitter; the lower limit of the margin needs to cover the boundary missing amount under the worst case, and the upper limit of the margin is constrained by the risk of accidental spraying of the adjacent normal coating. This range is determined by the statistical results of accidental spraying boundary and missed spraying boundary during the trial production stage).To avoid contamination of normal statistics by false points, anomaly removal is performed on the normal set. The removal threshold is set to deviate from the main normal by no more than 20 degrees (the threshold is determined by the physical upper limit of the normal variation of the waveform guardrail 1 within the bounding area scale; normals exceeding this deviation are mainly introduced by mismatched points and occlusion points. Retaining them will cause the initial estimation of the spray gun attitude to deviate, thereby expanding the capture range and reducing positioning efficiency. The 20-degree threshold is jointly determined by the upper limit of the normal variation of the maximum slope segment of the guardrail waveform at this scale and the detection noise). After completing the above calculations, 3D bounding box data is obtained. The 3D bounding box data represents the spatial envelope of the defect in the 3-coordinate system of the robot arm.

[0038] When using 3D bounding box data to determine the initial location range for surface defect capture, the core is to transform the bounding box into an "initial capture window" that can be directly invoked by the spraying control, and to provide a complete process of entering the window, encrypted scanning, and locking the defect range. Specifically, the center point of the bounding box is used as the initial pointing position for defect capture, the main normal of the bounding box is used as the initial reference direction for the detection field of view and the spray gun attitude, and the length, width, and height range of the bounding box is used as the initial capture boundary, forming the defect capture window. During the operation of the robotic arm 3, the controller calculates the distance between the spray gun end position and the center point of the bounding box in real time, and uses this distance to determine whether to enter the initial capture window. The entry distance threshold is set to 0.05 meters (threshold determination method: this distance needs to cover the end-positioning error and rotational dynamic error of the robotic arm 3, so that the robotic arm 3 maintains a smooth attitude when entering the window. At the same time, this distance matches the depth of field of the detection system, so that the clarity of the detected image after entering the window meets the boundary verification requirements; the threshold is determined jointly based on the maximum braking distance of the robotic arm 3, the control loop response, and the detection depth of field). Once the distance meets the entry conditions, the dynamic detection system performs encrypted acquisition of the bounding box range, with the acquisition density boost factor set to 2 times (determined by the real-time constraints of the processing link, ensuring that the total latency of image processing and index update does not exceed 0.05 seconds; a 2x boost can significantly improve defect boundary resolution and reduce the search space for subsequent precise positioning under sufficient bandwidth and computing power). After encrypted acquisition, the bounding box region is re-verified according to the aforementioned abnormal pixel identification and registration calculation process. If the change in the bounding box boundary of the verification result relative to the initial bounding box boundary is less than 0.01 meters, then the bounding box is used as a stable initial position range for the next stage of precise positioning; the 0.01-meter threshold is determined based on the lower limit of the expansion margin and the magnitude of boundary jitter. A change less than this threshold indicates that the defect boundary has converged and the error is under control; further expanding the range will increase the probability of false spraying without improving the capture reliability. Through the above linkage, identification, registration, bounding calculation, and window convergence, the dynamic detection system completes the initial marking of coating defects during the rotation process and outputs the initial position range for surface defect capture for spraying control.

[0039] By performing in-depth analysis on the initially marked defect areas and using real-time data processing technology to extract specific feature information of the defects, combined with the spatial coordinate mapping data, the three-dimensional position of the defects on the surface of the wave-shaped guardrail 1 is determined, and accurate defect localization results for spraying control are obtained. This includes acquiring a refined feature descriptor of the defects generated based on the initially marked areas, determining an accurate two-dimensional contour mask of the defect areas based on the refined feature descriptor, mapping the accurate two-dimensional contour mask onto the synchronously acquired depth data plane, extracting the depth value set within the corresponding range and calculating the normal vector distribution of the local surface, using a pre-calibrated spatial mapping matrix to convert the depth value set and normal vector distribution into a three-dimensional point cloud cluster in the coordinate system of the wave-shaped guardrail 1 surface, and performing surface fitting calculations on the three-dimensional point cloud cluster to generate accurate defect localization results containing three-dimensional position coordinates and attitude information.

[0040] In this embodiment, a refined feature descriptor for defects is first generated for the initially marked region, and a precise two-dimensional contour mask for the defect region is determined based on this descriptor. In implementation, the initially marked region is defined as the region of interest (ROI). Brightness consistency processing is first applied to the ROI to prevent natural brightness gradients caused by the curvature of the guardrail from participating in defect boundary determination. Brightness consistency processing establishes a reference based on the local brightness fluctuation range of the normal coating region within the same frame. The normal coating region is obtained by jointly filtering based on texture continuity and low fluctuation. Subsequently, multi-source boundary evidence calculation is performed on each pixel within the ROI: the brightness abrupt change intensity, color abrupt change intensity, and texture direction abrupt change intensity of the pixel and its neighboring pixels are calculated, and the three types of abrupt change intensities are normalized to form a comprehensive abrupt change score. The neighborhood scale is set to no more than 5 pixels (parameter determination method: the neighborhood scale covers the typical width of the defect boundary transition zone; when it exceeds 5 pixels, the gradually changing brightness caused by the curvature of the waveform guardrail 1 enters the neighborhood statistics, leading to boundary expansion; when it is less than 5 pixels, the proportion of sensor noise in the abrupt change score increases, leading to jagged boundaries). Pixels whose comprehensive mutation score exceeds the boundary candidate threshold are marked as boundary candidate points. The boundary candidate threshold is set to twice the upper limit of the comprehensive mutation score of the normal coating area (threshold determination method: before the production line is put into use, no less than 100 defect-free images are collected, and the maximum stable fluctuation value of the comprehensive mutation score of the normal area is used as the upper limit. The threshold is set to twice this value to suppress false boundaries triggered by reflection and random noise, while maintaining the mutation response caused by thin coating, missing coating, blistering, and pinholes; the multiple is determined based on the comprehensive results of false alarms and missed alarms during the trial production stage). After obtaining the boundary candidate points, boundary link construction is performed: using the boundary candidate points as nodes, they are connected according to the consistency of gradient direction and spatial continuity to form several boundary polylines; closure judgment is performed on the polylines. If the distance between the first and last ends of the polyline is less than 2 pixels, it is determined to be closed. The 2-pixel threshold is set (threshold determination method: this distance corresponds to the upper limit of endpoint jitter introduced by discrete sampling during the boundary extraction process. Exceeding this distance indicates that there is a break in the boundary, and it is necessary to return to supplement the boundary point connection). After closure, an initial fine contour is obtained, and then contour convergence calculation is performed: using the initial fine contour as the starting boundary, the difference in texture consistency between the inner and outer sides of each boundary point is compared, and the boundary point is moved along the direction of "more significant difference" until the maximum boundary movement in a single iteration is less than 1 pixel and this condition is met for 3 consecutive iterations; a 1-pixel threshold (threshold determination method: the spatial size corresponding to the pixel resolution is lower than the minimum effective improvement scale for defect boundary positioning by spray control, and continuing iteration introduces boundary jitter without improving the respray accuracy); a 3-round condition (parameter determination method: to offset the single-round misconvergence caused by occasional noise, and to meet the cycle time requirement under real-time processing delay constraints). This yields a precise 2D contour mask for the defect region.

[0041] Subsequently, the precise 2D contour mask is mapped onto the synchronously acquired depth data plane, the depth value set within the corresponding range is extracted, and the normal vector distribution of the local surface is calculated. In implementation, the dynamic detection system outputs image frames and depth frames at the same timestamp, so that the mask pixels and depth pixels have a one-to-one correspondence. If the system has multiple sensor channels, frame alignment is first performed based on a unified timestamp, and the frame alignment tolerance is set to no more than 0.005 seconds (parameter determination method: at the maximum relative speed of the rotation process, the surface displacement corresponding to 0.005 seconds is less than 1 / 4 of the minimum detectable size of the defect. Exceeding this tolerance will cause the mask projection and depth projection to deviate, thereby widening the 3D point cloud cluster). After alignment, the pixel positions marked as 1 in the mask are traversed, and the corresponding depth values ​​are extracted from the depth frame to form a depth value set. Depth anomaly removal is then performed: the deviation of each depth value from the median depth of its mask neighborhood is calculated, and deviations exceeding 0.005 meters are considered anomalies and removed. The 0.005-meter threshold is determined by combining the upper limit of depth sensing noise with the upper limit of the residual of guardrail surface reconstruction fitting. Depths exceeding this deviation are often related to reflections, occlusions, or ranging failures, and their participation in normal calculation will cause the pose output to deviate. A local neighborhood point set is constructed for the cleaned depth pixels, and the neighborhood radius used for normal calculation is set to no more than 3 pixels (the radius is determined by the following method: when the radius exceeds 3 pixels, the probability of the neighborhood crossing the slope change region of the peaks and troughs increases, causing the normal estimation to shift towards the global trend; when the radius is less than 3 pixels, the influence of depth noise on plane fitting increases, causing the normal distribution to diverge). For each neighborhood point set, local plane fitting is performed. The fitting steps are as follows: convert the depth pixels in the neighborhood into local 3D points, calculate the principal direction of the local 3D points, and take the direction perpendicular to the principal direction as the normal direction; repeat this process for all effective pixels in the mask to obtain the normal vector distribution. To suppress outlier normals, calculate the principal direction of the normal distribution and remove normals that deviate from the principal direction by more than 20 degrees; 20-degree threshold (threshold determination method: within the defect envelope scale, the normal change of the local surface of the waveform guardrail 1 has a physical upper limit. Deviations exceeding 20 degrees are usually caused by depth distortion and boundary aliasing points, and retaining them will pull the subsequent surface fitting attitude estimation).

[0042] After obtaining the depth value set and normal vector distribution, a coordinate transformation is completed using a pre-calibrated spatial mapping matrix to form a three-dimensional point cloud cluster in the surface coordinate system of the wave guardrail 1. In implementation, the spatial mapping matrix is ​​generated by the joint calibration between the detection coordinate system, the robot arm 3 base coordinate system, and the guardrail surface coordinate system. The joint calibration process uses the method of aligning the guardrail tooling reference point with the robot arm 3 end point to solve the coordinate transformation, so that the detection data and the spraying control coordinates are of the same origin. The conversion steps are as follows: For each effective depth pixel, the pixel position and depth value are first converted into a three-dimensional point in the detection coordinate system according to the imaging calibration parameters. The imaging calibration parameters are jointly determined by the factory calibration and the field calibration (the parameter determination method is as follows: the factory calibration provides the lens distortion and imaging model benchmark, the field calibration eliminates the small drift caused by the installation angle deviation and temperature drift, and the field calibration is determined by collecting no less than 20 frames through the calibration board and converging the reprojection error to a stable range). Then, the three-dimensional point in the detection coordinate system is transformed to the guardrail surface coordinate system through the spatial mapping matrix to obtain the three-dimensional point in the guardrail surface coordinate system. The normal direction corresponding to the pixel is subjected to a homologous transformation to obtain the normal direction in the guardrail surface coordinate system, ensuring that the point and the normal are in the same coordinate system. To ensure the long-term stability of the mapping relationship, the following criteria for updating the spatial mapping matrix are set: the benchmark point is periodically re-measured, and recalibration is triggered if the re-measurement error exceeds 0.003 meters; the 0.003-meter threshold is determined by the acceptable boundary obtained by superimposing the upper limit of the repeatability error of the robotic arm 3 with the upper limit of the residual error of the on-site calibration of the detection system. If the error exceeds this limit, it indicates that the tooling pose or sensor installation has drifted, and continuing to use the original mapping will directly transmit the error to the re-spraying path.

[0043] After forming a 3D point cloud cluster, surface fitting calculations are performed on the 3D point cloud cluster to output accurate defect location results containing 3D position coordinates and attitude information. The implementation first establishes point cloud cluster quality constraints: the point density of the point cloud cluster within the defect region's projection range is statistically analyzed. If the point density is less than 15 points per square centimeter, detection and densification are triggered, and the point cloud cluster is regenerated. A threshold of 15 points per square centimeter is used (the threshold is determined by the fact that surface fitting requires sufficient sampling points at both the boundary and interior; below this density, the fitting becomes sensitive to noise, and the position and attitude output jitter increases significantly, affecting the spray gun's attitude stability). After meeting the density requirements, surface fitting is performed: the principal direction of the normal vector distribution is used as the initial attitude constraint for fitting. A local surface within the point cloud cluster with the highest consistency with the principal direction is found as a reference surface, and the deviation of each point from the reference surface is calculated. The deviation is used to distinguish between the defect core area and the normal coating area; the set of points with significant deviation and spatial connectivity is determined as the defect core point set. To avoid outliers skewing the fit, an outlier removal threshold is set: points whose distance from the reference surface exceeds 0.004 meters are considered outliers and removed. The 0.004-meter threshold is determined by combining the depth noise residual, the local surface fitting residual, and the upper limit of the guardrail surface roughness; points exceeding this distance are often related to false reflections or boundary aliasing, and retaining them would cause deviations in attitude estimation. After obtaining a stable reference surface and a set of core defect points, the 3D position coordinates are calculated through the geometric center: the 3D coordinates of the core defect point set are balanced to obtain the center point as the 3D position coordinates of the defect; the attitude information is taken from the normal direction of the reference surface at the center point, serving as the local surface attitude reference for subsequent spray gun alignment and spraying angle control. To ensure consistency between the 2D contour and 3D positioning, a consistency check is performed: the 3D point cloud cluster is projected back onto the image plane and compared with the precise 2D contour mask. A boundary difference exceeding 0.01 meters is considered inconsistent and the process reverts to the depth frame alignment and depth anomaly removal stage for recalculation. A 0.01-meter threshold is set (determined by adding the boundary error introduced by 2D contour boundary smoothing to the pixel-to-3D conversion residual; exceeding this threshold indicates anomalies in frame alignment, depth quality, or spatial mapping, and continued output will lead to an expansion of the respraying range and increase the risk of mis-spraying). Through the above refined feature description, 2D contour convergence, depth and normal calculation, spatial mapping transformation, point cloud surface fitting, and consistency check, precise defect positioning results for spraying control are obtained. These results simultaneously include the 3D coordinates of the defect center, the local orientation of the defect, and the spatial scale of the defect.

[0044] Based on the accurate defect location results, planning data for the reverse motion path used for spraying control is generated. The path parameters are adjusted according to the dynamic characteristics of the robot arm 3 during its rotation process to match the path planning with the current operating state of the robot arm 3. The execution order of the re-spraying task is determined by: obtaining the accurate defect location results; calculating the inverse kinematics solution set based on the accurate defect location results; generating initial reverse motion path planning data based on the inverse kinematics solution set; extracting parameters from the initial reverse motion path planning data to calculate joint torque load; generating corrected path parameters if the joint torque load exceeds a threshold; and correcting the corrected path parameters with the current operating state of the robot arm 3 to determine the execution order of the re-spraying task that matches the current operating state of the robot arm 3.

[0045] First, accurate defect location results are obtained, and target pose constraints at the spray gun tip are constructed. The accurate defect location results include the three-dimensional coordinates of the defect center, the orientation of the defect's local surface, and the spatial scale of the defect. The control system uses the defect center as the end-point target position, the orientation of the defect's local surface as the alignment reference for the end-point spray direction, and converts the defect spatial scale into boundary constraints for the end-point coverage trajectory. The end-point target distance parameter is set to 0.2 meters to 0.35 meters (parameter determination method: matched with the spray gun atomization characteristics, adhesion rate target, and rebound control target; too small a distance increases the risk of local buildup, while too large a distance decreases the adhesion rate; this range is determined during equipment delivery and commissioning based on the spray gun model characteristic curve and coating thickness uniformity requirements). The upper limit of the angle between the end-point spray direction and the local surface orientation direction is set to 30 degrees (parameter determination method: increasing the angle will exacerbate the shading effect at the slope and edge of the wave guardrail 1 and expand the thin-coat area; 30 degrees is determined comprehensively based on the shading sensitivity of the guardrail waveform's maximum slope segment, the spray width edge attenuation characteristics, and the target thickness fluctuation tolerance). After the target pose constraint is generated, the system binds the constraint to the current rotation state index, so that the target pose corresponds to the rotation phase on the time axis. The rotation phase alignment tolerance is set to no more than 0.01 seconds (parameter determination method: the surface projection displacement under the maximum relative rotation speed is used as a constraint, so that the displacement is less than 1 / 2 of the minimum boundary size of the defect. Exceeding this tolerance will cause the re-spraying point to deviate from the defect boundary and increase the risk of mis-spraying).

[0046] After completing the target pose constraint, the inverse kinematics solution set is calculated based on the target pose constraint. During implementation, the control system reads the structural parameters of the robot arm 3, the joint limit range, the upper limit of joint velocity, and the upper limit of joint acceleration, and uses the current joint state as the initial value for the solution, and starts the inverse solution process to obtain multiple sets of joint solutions that satisfy the constraints. During the solution process, the accessibility is first determined: it is determined that the position of the end effector target is within the workspace of the robot arm 3, and the feasibility of the end effector target posture under the joint limit is checked. If the posture is not feasible, the end effector posture constraint is fine-tuned. The upper limit of the fine-tuning amplitude is set to 5 degrees (parameter determination method: 5 degrees corresponds to the posture adjustment margin under the upper limit of the spraying angle of 30 degrees, and this amplitude will not cause the spraying angle to exceed the limit or significantly reduce the consistency of the re-spraying quality). Multiple joint solutions were selected and sorted. The selection criteria combined "joint displacement cost, singularity risk cost, and consistency cost with the current rotation direction." Joint displacement cost was calculated by weighing the change in angle from the current angle to the target angle for each joint and summing the weights according to joint travel. The weights were set based on joint inertia and motor rated capacity, with higher weights for joints with larger inertia and load sensitivity (parameter determination method: weights were determined by combining factory dynamic calibration data and on-site load calibration data for the robot 3, aiming to minimize large-amplitude movements of high-inertia joints). Singularity risk cost was determined by checking the degree to which the target joint solution approached the singular attitude range. The proximity threshold was set to "joint angle margin with the singular range boundary not less than 3 degrees" (threshold determination method: 3 degrees is used to cover angle fluctuations caused by control errors and rotation disturbances, avoiding speed amplification and torque surges near singularities). Consistency cost was determined by comparing the relationship between the end-effector attitude change direction of the target solution and the current rotation direction, prioritizing solutions with coordinated attitude change and rotation directions to reduce relative motion impact during insertion segments. After the above screening, the inverse kinematics solution set is obtained, and the top 3 candidate solutions with the lowest cost are retained for path planning. The number of candidate solutions is set to 3 (parameter determination method: 3 candidate solutions can cover the feasibility differences of different joint configurations, while satisfying the constraints of real-time computing resources and scheduling window. Too many solutions will increase the path evaluation time and affect real-time performance).

[0047] After obtaining the inverse kinematics solution set, initial inverse motion path planning data is generated based on the solution set. In implementation, the current joint state of the robot arm 3 is used as the path starting point, and candidate joint solutions are used as the path ending point. First, a transition segment is generated in the joint space, and then a defect coverage segment is generated in the end-effector space. The two are then spliced ​​together according to temporal continuity to form a complete initial path. During transition segment generation, the system plans a time series of position, velocity, and acceleration for each joint. The sequence generation follows the constraint of "continuous velocity and continuous acceleration." Continuous velocity is used to avoid impacts, and continuous acceleration is used to avoid torque spikes. The minimum duration of the transition segment is set to be no less than 0.1 seconds (parameter determination method: matched with the controller interpolation cycle and servo response time; a duration that is too short will cause an increase in acceleration peaks and push up the joint torque load). During defect coverage segment generation, the defect space scale is converted into the boundary of the end-effector coverage trajectory, and the end-effector trajectory point sequence is generated according to the coverage order of "expanding from the center outwards." This ensures that the spray gun first covers the core defect area and then the boundary area, reducing the probability of edge overspray during the first re-spray. The spacing between coverage trajectory points is set to 0.01 meters to 0.03 meters (parameter determination method: the point spacing matches the spray width and trajectory overlap requirements; excessively large point spacing will result in missed coverage, while excessively small point spacing will increase repetitive spraying and cycle time burden; this range is determined based on the spray gun's spray width and the target overlap coefficient). End-point attitude and distance requirements are generated for each trajectory point. The attitude requirement is constrained by a spray angle limit of 30 degrees, and the distance requirement is constrained by 0.2 meters to 0.35 meters. The rotation phase is aligned using a dynamic spatial mapping relationship to ensure that the timestamp of each trajectory point matches the current rotation state index. The time resolution of the path planning data output is set to no more than 0.01 seconds (parameter determination method: ensuring that the projected displacement of adjacent path points on the guardrail surface is less than 1 / 2 of the defect boundary size, while also meeting the controller interpolation cycle requirements).

[0048] After the initial reverse path planning data is generated, path parameters are extracted to calculate joint torque loads, and corrected path parameters are generated based on thresholds. In implementation, the system extracts the target position, target velocity, and target acceleration of each joint from the path planning data node by node. It then combines this with the mass distribution of the manipulator's three-link linkage, joint inertia parameters, end-effector and pipeline load parameters, and the inertial coupling effect of the rotation process to calculate the driving load level of each joint at each node, forming a joint torque load sequence. The load calculation is completed in the following steps: first, the acceleration load component is obtained based on the target acceleration and joint inertia; then, the velocity-related load component is obtained based on the target velocity and changes in motion direction; next, the load transfer component is obtained based on the end-effector load and attitude changes; finally, the total load level is formed by combining the additional inertial component introduced during the rotation process. To ensure consistency between the calculation and the actual driving capability, the system calibrates the end-effector load parameters before equipment commissioning. The calibration method involves collecting servo current and position errors at different rotation speeds and back-calculating the equivalent load. The back-calculation results are used to correct the parameters of the load model, ensuring that the model's response is consistent with the driving current within the working range. The joint torque load threshold is set to 0.8 times the rated allowable torque of the joint (threshold determination method: 0.8 times is used to reserve dynamic margin to resist paint pipeline swaying, rotation speed fluctuations and friction changes, and to avoid overheating and protection triggering when approaching the rated torque; the coefficient is jointly determined by the allowable overload curve of the motor and reducer, the on-site temperature rise constraint and the continuous operation cycle). When any joint exceeds the threshold at any node, the system locates the path segment where the over-limit node is located and generates correction path parameters. The correction process is executed in order of priority: first, the peak joint acceleration of the segment is reduced, so that the acceleration load component decreases; the upper limit of the acceleration reduction is set to 30% (parameter determination method: 30% can significantly suppress the peak torque without significantly lengthening the cycle, and too large an amplitude will cause the re-spraying insertion time to be too long, affecting the main spraying cycle). If the acceleration still exceeds the limit after reduction, the peak velocity of that segment is reduced and the climb time is extended to decrease the velocity-related load components. The upper limit for velocity reduction is set to 20% (parameter determination method: velocity reduction has a more direct impact on the cycle time; 20% is used to balance cycle time and overload risk). If the limit is still exceeded, the candidate solutions in the inverse kinematics solution set are switched, and the solution with lower joint displacement cost and smaller high-inertia joint motion is selected to regenerate the transition segment. The switching trigger condition is "there are still nodes exceeding the limit after two consecutive rounds of correction" (trigger condition determination method: to avoid frequent switching due to insufficient correction at one time, and to avoid too many rounds of correction occupying the real-time calculation window; two rounds balance stability and efficiency in real-time scheduling). After each correction is completed, the system recalculates the joint torque load sequence until all nodes meet the threshold constraints or the solution set switching is completed and convergence is achieved.

[0049] After generating the corrected path parameters that satisfy the load constraints, the corrected path parameters are adjusted against the current operating state of the robot arm 3, and the execution order of the repainting tasks is determined. In implementation, the system obtains the current operating state of the robot arm 3, including the current rotation angle position, the current speed and acceleration of each joint, the current controller scheduling window length, the current painting task queue, and the latest update of the defect range by the dynamic detection system. First, time correction is performed: the start timestamp of the corrected path is aligned to the nearest scheduling window start point, ensuring that the inserted segment executes within the controller's acceptable time slot. The lower limit of the scheduling window length is set to 0.2 seconds (parameter determination method: it must cover the minimum duration of the transition segment (0.1 seconds) and the execution time of at least one trajectory segment of the covered segment, while reserving space for controller queue refresh). Then, spatial correction is performed: the difference between the current end pose and the starting pose of the correction path is calculated. If the difference causes the maximum velocity or maximum acceleration of the transition segment to exceed the upper limit, a short transition segment is inserted at the starting end to make the velocity and acceleration continuous at the connection point. The duration of the inserted segment is set to be no less than 0.05 seconds (parameter determination method: 0.05 seconds matches the servo response characteristics, enabling small-amplitude attitude transitions and suppressing shocks). After correction, the task sequence determination is performed: when multiple defect repainting tasks exist simultaneously, the system calculates a ranking index for each task and generates an execution order. The ranking index includes switching cost, overload risk cost, and defect priority. The switching cost is calculated by the total joint displacement, total attitude change, and insertion delay between adjacent tasks. The insertion delay is obtained by the extension of the main spraying cycle after the task is inserted. The upper limit of the insertion delay is set to no more than 0.3 seconds (parameter determination method: constrained by the maximum disturbance allowed by the production line cycle; exceeding this upper limit will significantly affect spraying continuity and production efficiency). The overload risk cost is determined by the degree to which the load of each joint in the task correction path approaches the threshold. The proximity judgment threshold is set to "0.9 times the load of the threshold" (threshold determination method: 0.9 times is used to identify segments approaching saturation in advance to avoid exceeding the threshold when on-site disturbances increase; this coefficient is used in conjunction with the 0.8 times rated threshold to form a safety buffer). Defect priority is determined based on the defect area and the quality sensitivity of the defect location. The defect area is calculated from the projected area of ​​the 3D point cloud cluster. The area calculation is completed by closing the boundary of the 3D boundary points corresponding to the 2D contour mask of the defect and counting the surface area covered by the closed area; the area priority threshold is set to not less than 0.0004 square meters (threshold determination method: corresponding to the minimum attention scale for visible defects in spraying acceptance; defects smaller than this area have less impact on the overall appearance and are given lower priority to ensure cycle time). When sorting, tasks with low switching costs and low overload risks are given priority, and the sorting weight of defect tasks with areas exceeding the threshold is increased, so that repainting first processes defects that have a more significant impact on appearance.The final output is the respraying task execution sequence that matches the current operating state of the robotic arm 3, and the final reverse motion path planning data that is consistent with this sequence is output to drive the subsequent respraying attitude adjustment and spraying control execution.

[0050] The robot arm 3 is driven to move by path planning data, generating control commands for adjusting the spraying posture for spraying control. The spray gun angle and the distance between the spray gun and the surface of the wave-shaped guardrail 1 are adjusted according to the defect location, ensuring that the changes in spraying angle and distance are dynamically adjusted in sync. The spraying control execution scheme for the supplementary spraying action includes: acquiring the discrete trajectory point sequence from the path planning data; calculating the surface normal vector and tangential curvature feature data at the defect location based on the discrete trajectory point sequence; solving the six-degree-of-freedom target pose matrix and target spraying distance value based on the surface normal vector and tangential curvature feature data; calculating the attitude error quaternion between the actual end-effector pose and the six-degree-of-freedom target pose matrix, and generating supplementary spraying posture adjustment control commands based on the attitude error quaternion; and converting the supplementary spraying posture adjustment control commands into target torque and speed control signals to obtain the spraying control execution scheme for the supplementary spraying action.

[0051] In this implementation, the discrete trajectory point sequence in the path planning data is first acquired, and the discrete trajectory point sequence is bound to time and rotation phase. In practice, each trajectory point in the path planning data has a corresponding timestamp and re-spraying task number. The controller reads the timestamp and aligns it with the current rotation state index of the robot arm 3, ensuring that the trajectory advancement and rotation phase are on the same time reference. The alignment tolerance is set to no more than 0.01 seconds (parameter determination method: using the surface projection displacement at the maximum relative rotation speed as a constraint, ensuring that this displacement is less than 1 / 2 of the minimum scale of the defect boundary; exceeding this tolerance will cause a visible offset of the spray gun's action point relative to the defect area). After alignment, the controller establishes a "local window" for each trajectory point. The local window consists of adjacent points before and after the trajectory point, and the number of window points is set to no less than 5 (parameter determination method: 5 points are used to cover local changes in the tangential direction of the trajectory and suppress single-point jitter; if the number of points is too small, the normal and curvature estimations are sensitive to noise; if the number of points is too large, the window will cross the transition section between peaks and troughs, causing the local geometry to be flattened). For each trajectory point within the local window, surface fitting is performed: the nearest surface point to the trajectory point is searched from the guardrail surface reconstruction data of the dynamic detection system, and the local plane consistency is checked in the neighborhood of the surface point. If the consistency is insufficient, the neighborhood is expanded forward and backward along the trajectory tangential direction and the local plane consistency is re-evaluated until the local plane consistency meets the requirements. The consistency criterion threshold is set to "the deviation of the neighborhood point from the local plane does not exceed 0.002 meters" (the threshold is determined as follows: 0.002 meters corresponds to the acceptable upper limit of surface reconstruction noise and point cloud discretization error. Exceeding this threshold indicates that the neighborhood contains reflective pseudo-points or crosses curvature abrupt change segments. Continuing to use it for normal calculation will lead to attitude jitter).

[0052] After completing the surface bonding of the local window, the surface normal vector and tangential curvature feature data at the defect location are calculated based on the discrete trajectory point sequence. In implementation, the controller performs local surface reconstruction on the set of surface points within the local window. The reconstruction process follows the order of "first calculating the principal direction of the local plane, then calculating the plane normal direction." The principal direction of the plane is statistically obtained from the main extension directions of the point set, and the normal direction is taken as the direction perpendicular to the plane containing the principal direction, and the direction is uniformly pointed outwards to align with the spray gun's aiming direction. To ensure normal stability, the normal is calculated for each surface point within the window, and the principal normal is statistically analyzed. Then, normals deviating from the principal normal by more than 20 degrees are removed. The 20-degree threshold is determined by the physical upper limit of the local normal variation of the waveform guardrail 1 within the window scale; deviations exceeding 20 degrees are mostly caused by mismatched point clouds, occlusion, or depth distortion. After removal, the principal normal better represents the true surface posture. The tangential direction is taken from the direction of the line connecting the preceding and following points of the trajectory point, and this direction is projected onto the tangential plane of the local surface as the tangential reference direction. Subsequently, the curvature change of the local surface is statistically analyzed along this tangential direction. The statistical process is based on the normal variation amplitude and surface height variation trend of adjacent projection points within the window to form tangential curvature feature data, which is used to reflect the steepness and direction of curvature change at that location. To avoid the curvature estimation being amplified by noise, the smoothing window length of the curvature feature is set to be no less than 3 adjacent projection points (parameter determination method: 3 points smoothly cover the smallest local curvature segment and suppress single-point noise, while not crossing significant segments of the waveform period to avoid mixing peaks and troughs).

[0053] After obtaining the surface normal vector and tangential curvature feature data, the six-degree-of-freedom target pose matrix and target spraying distance are calculated based on these data. In implementation, a target position is constructed for each trajectory point, and the target position is taken as the defect re-spraying point corresponding to that trajectory point. The target direction of the spray gun's spray axis is determined by the surface normal vector, and the target direction is opposite to the direction of the surface normal vector, ensuring the spray gun faces the guardrail surface. To limit the rotation angle of the spray gun around the spray axis, the tangential direction is used as an auxiliary alignment direction, aligning the long axis of the spray pattern with the tangential direction of the trajectory. This allows the re-spraying coverage to advance along the defect extension direction, reducing edge accumulation caused by lateral overlapping spraying. The target spraying distance is determined by superimposing the base distance and the curvature compensation distance: the base distance is the median value between 0.2 meters and 0.35 meters (parameter determination method: the median value corresponds to the stable zone of the spray gun atomization cone shape and the stable zone of the adhesion rate, resulting in more stable thickness consistency within the range of coating viscosity and environmental fluctuations); the curvature compensation distance is triggered by tangential curvature characteristic data. When the curvature change increases, the system limits the single-cycle change in distance adjustment to make the distance change smoother. The upper limit of the single-cycle change is set to 0.005 meters (parameter determination method: 0.005 meters corresponds to the upper limit of the followable displacement along the normal direction at the end within the control cycle. Exceeding this value will increase the joint acceleration and cause a torque peak. At the same time, this value is less than the sensitivity threshold of thickness uniformity to distance disturbance). Meanwhile, the spraying angle control constraint is maintained at an angle of no more than 30 degrees with the surface normal (parameter determination method: 30 degrees is used to suppress the sloping shading effect of the wave guardrail 1 and reduce the risk of thin coating). The six-degree-of-freedom target pose matrix is ​​determined by the target position and target attitude. The target spraying distance value serves as a normal distance constraint that is updated synchronously with the pose. It is updated in each control cycle as the trajectory point advances, thus forming a continuous sequence of target pose and target distance during the respraying process.

[0054] After generating the target pose and target distance sequence, the attitude error quaternion between the actual end effector pose and the target pose is calculated, and a supplementary spraying attitude adjustment control command is generated based on the attitude error. In implementation, the controller reads encoder data from each joint in real time and combines it with end effector installation calibration data to calculate the actual end effector pose. The actual end effector pose and the target pose are in the same coordinate system. The attitude error calculation is completed in the order of "first calculating relative rotation, then expressing it using quaternions": the rotation relationship between the target pose and the actual pose is used as the error rotation, and this error rotation is converted into an attitude error quaternion to obtain a unified expression of the rotation axis and rotation amplitude of the error. To suppress error abrupt changes caused by noise during the rotation process, the attitude error quaternion is subjected to amplitude limiting processing. The amplitude limiting is determined based on the error change between adjacent control cycles, with the upper limit of the change set to no more than 2 degrees (threshold determination method: 2 degrees corresponds to the angle that the servo system can stably follow in a single cycle at the current speed upper limit; exceeding this value will cause joint speed spikes and lead to the risk of instantaneous over-limit spraying angle). After limiting the spray width, a control command for adjusting the spraying attitude is generated. The control command includes the attitude adjustment direction, attitude adjustment range, and adjustment duration. Simultaneously, a distance adjustment command is generated. The distance adjustment command is obtained from the difference between the actual spraying distance and the target spraying distance, and is also constrained by the single-cycle change limit of 0.005 meters to ensure that the attitude adjustment and distance adjustment are carried out in tandem within the same control cycle, thereby maintaining the consistency between the spraying angle change and the dynamic adjustment of the distance.

[0055] After receiving the respraying attitude adjustment control command, the control command is converted into target torque and speed control signals, forming a spraying control execution scheme for the respraying action. In implementation, the controller first converts the end-effector attitude increment and normal distance increment into incremental motion requirements in the joint space. The conversion process is completed based on the current joint state and the kinematic relationship of the robot arm 3, ensuring that each joint receives corresponding angle increment, velocity increment, and acceleration increment. Subsequently, a joint target speed signal is generated. The target speed is obtained by superimposing the joint speed reference of the node in the path planning data with the velocity increment generated by the attitude and distance adjustment, and is constrained by the upper speed limit of each joint. The upper speed limit is taken as 0.8 times the controller's factory limit (parameter determination method: 0.8 times is used to reserve dynamic margin to resist rotational disturbances, pipeline sway, and friction changes, avoiding insufficient following when approaching the limit). The target torque signal for the joint is then generated. The target torque is determined by the joint acceleration increment and the equivalent load. The equivalent load is obtained by combining the link mass distribution, joint inertia, end-effector and pipeline load, and the inertial coupling state introduced by rotation. The load parameters are back-calibrated to a fixed value before the equipment is put into operation through servo current and position error, so that the target torque is consistent with the actual drive requirements. To avoid torque saturation, the upper limit of the target torque is set to 0.8 times the rated allowable torque of the joint (parameter determination method: consistent with the overload control threshold in the path planning stage; 0.8 times is used to maintain long-term stable operation and reserve thermal margin). When the target torque or target speed reaches the upper limit, the controller links the attitude limit threshold of 2 degrees and the single-cycle distance change limit of 0.005 meters, so that the end-effector adjustment action is distributed to be completed in more control cycles, thereby converging to the target pose and target distance without triggering overload.

[0056] During the respraying process, the controller continuously performs closed-loop verification and updates: each control cycle recalculates the actual end pose, actual spraying distance, and attitude error quaternion, and compares them with the target pose matrix and target spraying distance values. If the attitude error continuously exceeds 3 degrees or the distance error continuously exceeds 0.01 meters and does not decrease for three consecutive cycles, the trajectory point is stopped and a local recalculation is triggered. The 3-degree threshold (threshold determination method: 3 degrees is close to the acceptable error band under the upper limit of the spraying angle of 30 degrees. Continuous exceedance indicates insufficient following or abnormal local normal estimation, requiring a pause in advancement to avoid the formation of uneven stripes); the 0.01-meter threshold (threshold determination method: 0.01 meters corresponds to the sensitive boundary of distance disturbance on thickness uniformity. Continuous exceedance will result in visible thickness deviation in the resprayed area); the continuous 3-cycle condition (parameter determination method: single-cycle noise and transient disturbances are eliminated. 3 cycles correspond to 0.03 seconds of control cycle 0.01 seconds, which still meets the real-time requirements of respraying). When a local recalculation is triggered, the system reconstructs a local window centered on the current trajectory point, updates the surface normal and tangential curvature feature data, and regenerates the target pose and target distance sequence for that segment. This ensures that the supplementary spray control execution scheme remains stable, continuous, and responsive under conditions of rotational disturbances and curvature changes. This is achieved through the above-mentioned methods: obtaining the normal curvature from the discrete trajectory point, synchronously calculating the target pose and target distance, coordinated control driven by attitude error quaternions, and outputting joint-level torque and speed signals.

[0057] According to the spraying control execution plan, the posture data of the robotic arm 3 during the re-spraying process is monitored in real time, and the coating quality consistency information fed back by the dynamic detection system is used to judge the spraying control. When the difference between the re-sprayed area and the surrounding coating exceeds the preset threshold, a secondary re-spraying control command is triggered to determine whether the re-spraying effect meets the standard. Specifically, this includes acquiring the spatial pose data of the end effector and the surface optical reflectivity distribution map of the robotic arm 3 during re-spraying, and performing spatiotemporal registration of the surface optical reflectivity distribution map and the spatial pose data of the end effector to calculate the texture feature similarity coefficient. If the texture feature similarity coefficient is lower than the preset standard, the secondary re-spraying entry angle and spraying distance are determined based on the coordinates of the local area with significant difference features. A secondary re-spraying control command is generated based on the secondary re-spraying entry angle and spraying distance, and the robotic arm 3 is driven to perform the repair action according to the secondary re-spraying control command. After that, the image is collected and the feature fusion degree is calculated to obtain the verification status of whether the re-spraying effect meets the standard.

[0058] In this embodiment, the pose data acquisition for the respraying process is first completed by the robot arm 3 controller. The controller reads the spatial pose data of the end effector during each control cycle of the respraying execution. The pose data includes the end effector position, end effector orientation, and the actual distance from the end effector to the guardrail surface, along with a timestamp. The pose sampling period is set to no more than 0.01 seconds (parameter determination method: matching the motion control interpolation cycle, ensuring that the projected displacement of the end effector on the guardrail surface corresponding to two adjacent samples is less than 1 / 2 of the minimum scale of the respraying area boundary, thereby avoiding strip misalignment in the registration map). Simultaneously, the dynamic detection system acquires the surface optical reflectivity distribution map of the respraying area. The map frame rate is set to no less than 50 frames per second (parameter determination method: matching the temporal resolution of the 0.01-second pose sampling, ensuring a stable correspondence between the map frames and pose sampling on the time axis, and covering the relative surface motion caused by rotation, reducing errors introduced by time interpolation). Before entering the consistency calculation, the reflectivity distribution map undergoes illumination normalization: a normal coating reference band is selected from outside the repainted area as the normalization reference, with a width of 0.03 meters (parameter determination method: 0.03 meters covers the stable coating texture and avoids the transition zone of the repainted boundary; a width that is too small is significantly affected by noise disturbance, while a width that is too large introduces curvature and illumination gradient, reducing reference stability). The normalization process is as follows: the center level and fluctuation amplitude of reflectivity intensity are statistically analyzed within the reference band, the current frame map is corrected as a whole according to the center level, and the local contrast is compressed or stretched according to the fluctuation amplitude, so that the reflectivity distribution of the same material under different illumination angles is on a uniform scale.

[0059] After completing the acquisition of pose and atlas data, spatiotemporal registration is performed to establish the co-domain data required for calculating texture feature similarity coefficients. Temporal alignment is based on timestamps, pairing each frame's reflectance map with the most recent pose sample. The temporal alignment tolerance is set to no more than 0.005 seconds (threshold determination method: constrained by the projected displacement under the maximum relative velocity of the rotating surface, ensuring this displacement is less than half the scale of the texture feature window coverage; exceeding this tolerance will cause texture window placement to shift and significantly reduce similarity). When the time difference exceeds 0.005 seconds, temporal interpolation is performed using two adjacent points of the pose sample. The interpolation process limits the end-effector pose change to no more than 2 degrees (threshold determination method: 2 degrees corresponds to the stable following angle range of the servo system within a 0.01-second period; exceeding this will introduce unreliable interpolated poses and cause registration deviations). Spatial registration is completed based on the dynamic spatial mapping relationship and the calibration results from the detection system to the robot's 3D coordinates: First, the pixel positions in the reflectivity map are mapped to 3D surface points in the guardrail surface coordinate system. Then, a correspondence is established between these 3D surface points and the trajectory of the spraying action point corresponding to the end effector pose at the same moment, forming a spatial index of "map surface points to spraying trajectory points". When generating the index, nearest neighbor matching is used and a distance threshold is applied. The matching threshold is set to no more than 0.01 meters (the threshold is determined by covering the upper limit of the superposition of the detection point cloud discrete error, calibration residual, and pose calculation error within 0.01 meters. Exceeding this threshold indicates that the correspondence is unreliable, and continuing to participate in similarity calculation will result in misjudgment).

[0060] After registration, the texture feature similarity coefficient is calculated. Texture feature extraction is performed using a sliding window method: a window grid is established within the overspray area according to the registered surface coordinates, and the mean reflectance distribution, local contrast distribution, and texture direction consistency distribution are statistically analyzed within each window to form the texture feature vector for that window. The sliding window side length is set to be no less than 9 pixels (parameter determination method: 9 pixels correspond to the smallest stable texture unit scale at the detection resolution; smaller scales result in a higher noise ratio, while larger scales average out local differences, leading to weakened thin-coat and overspray boundaries). The surrounding reference area is selected as an annular band outside the overspray boundary, with a width of 0.03 meters (parameter determination method: consistent with the illumination normalization reference band, ensuring that the reference statistics and normalization benchmark are within the same stable bandwidth range, reducing the offset caused by differences in region selection). The similarity calculation process is as follows: The texture feature vector of each window in the repainting area is similar to the texture feature vector of the corresponding window in the reference area. The similarity measurement results for all windows are then weighted and summarized. The weight is determined by the distance from the window to the repainting boundary; the closer the distance to the boundary, the higher the weight (parameter determination method: uneven transitions are most likely to occur at the boundary; increasing the boundary weight helps capture visible differences and reduces missed detections of uniform central areas with abrupt edge changes). The summarized result serves as the texture feature similarity coefficient, used to characterize the consistency level between the repainting area and the surrounding coating.

[0061] When the texture feature similarity coefficient is lower than the preset standard, the secondary respray judgment and parameter calculation process begins. The preset standard is set to be no less than 0.85 (threshold determination method: during the trial production stage, reflectivity spectra of qualified coating samples are collected and the same registration and similarity calculation are performed. The stable lower bound of the similarity of qualified samples is statistically analyzed, and this lower bound is adjusted upward to 0.85, so that the decrease in similarity corresponding to the visible difference is covered, while suppressing false triggering caused by slight changes in illumination). When the similarity is lower than 0.85, the system performs difference localization within the respray area: for each sliding window, the difference degree between its texture feature vector and the reference area is calculated, and windows with a difference degree exceeding the difference threshold are marked as candidate difference blocks. The difference threshold is set to twice the upper bound of the reference area difference degree (threshold determination method: the upper bound of the reference area difference degree reflects the natural fluctuation level of the normal coating under changes in curvature and illumination. Twice of it is used to separate the natural fluctuation from the inconsistency of the actual respray, while reducing the false identification of reflective spots). Subsequently, connectivity aggregation is performed on the candidate difference blocks to form difference blocks. The difference blocks are then mapped to local region coordinates in the guardrail surface coordinate system using the aforementioned spatial index. The local region coordinates are represented by the geometric center point, boundary range, and local surface normal statistics of the difference blocks. Data that deviates from the main normal by more than 20 degrees is removed from the normal statistics (threshold determination method: within the difference block scale, there is a physical upper limit to the change of the guardrail local normal. Deviations exceeding 20 degrees are mostly introduced by point cloud mismatch or depth distortion caused by strong reflectivity. Retaining them will mislead the calculation of the cutting angle).

[0062] After obtaining the local coordinates of the difference block, the secondary spraying entry angle and spraying distance are calculated, with the calculation logic prioritizing the determination of the difference type of the difference block. The difference type determination is based on the combined offset of the average reflectance and local contrast of the difference block: when the average reflectance is low and the contrast is low, it is determined to be a thin coating or missed coating trend; when the average reflectance is high and the consistency of texture direction decreases, it is determined to be an over-spray or high roughness trend. Under the thin coating or missed coating trend, the secondary spraying entry angle is set to be no more than 15 degrees with the local surface normal (parameter determination method: thin coating repair needs to improve effective deposition and reduce shading; a smaller angle can enhance normal deposition efficiency; 15 degrees is used to control the risk of local accumulation while improving deposition efficiency and maintaining the continuity of spray coverage). Under the thin coating or missed coating trend, the spraying distance is set to 0.2 meters to 0.3 meters (parameter determination method: the distance is closer to the lower limit to improve deposition per unit area and shorten atomization diffusion; 0.2 meters is used as the lower limit to suppress the risk of sagging; 0.3 meters is used as the upper limit to avoid a decrease in deposition efficiency). When overspraying or high roughness is expected, the angle of entry for secondary spraying is set between 15 and 25 degrees with the local surface normal (parameter determination method: a slight deviation from the normal can flatten the deposition gradient and improve transition uniformity; 25 degrees is still within the controllable range of shading risk and avoids the formation of thin edge coatings). The spraying distance when overspraying or high roughness is expected is set between 0.28 meters and 0.35 meters (parameter determination method: increasing the distance reduces deposition per unit area and expands the transition range, which is beneficial for weakening local raised textures; 0.35 meters is used as the upper limit to avoid a significant decrease in adhesion rate). After the angle and distance are calculated, they are used together with the local area coordinates of the difference block to generate local path segment constraints for secondary spraying. The constraints include the entry posture, entry distance, coverage boundary, and coverage advancement direction. The advancement direction is taken along the long axis of the difference block (parameter determination method: advancing along the long axis of the difference block reduces lateral overspraying and reduces edge accumulation).

[0063] The generation and execution of secondary spray control commands are based on local path segment constraints as input. The control system converts the local coordinates of the difference block into an end-target pose sequence, and then converts the target pose sequence into a joint space control sequence. The control sequence includes attitude adjustment commands and distance adjustment commands, and is aligned with the scheduling window of the current rotation phase. The spacing between points on the secondary spray coverage trajectory is set to 0.01 meters to 0.02 meters (parameter determination method: secondary spray is for local difference repair, so a denser spacing is used to improve coverage accuracy and reduce the probability of missed repairs, while the spacing is not less than 0.01 meters to control cycle disturbances and avoid excessive overspraying). During the secondary spraying process, the end-effector attitude change limit is set to no more than 2 degrees per cycle (threshold determination method: matched with servo stability tracking capability to avoid sudden attitude changes causing instantaneous deviation of the spraying angle and forming uneven stripes), and the spraying distance change limit is set to no more than 0.005 meters per cycle (threshold determination method: 0.005 meters corresponds to the upper limit of the end-effector's followable displacement along the normal direction within the control cycle; exceeding this will increase joint acceleration and form a load peak, while this limit is less than the sensitive boundary of thickness uniformity to distance disturbance). After execution, verification and acquisition are performed. The dynamic detection system re-acquires reflectivity distribution maps of the repaired area and completes the same illumination normalization and spatiotemporal registration to ensure that the comparison before and after repair is on the same evaluation scale.

[0064] Whether the repainting effect meets the standard is verified by calculating the feature fusion degree. The feature fusion degree calculation consists of two parts: the first part is the similarity of texture features between the repainted area and the surrounding reference area after repair; the second part is the decrease in the difference between the difference of the difference blocks before repair and the difference after repair. The fusion degree calculation process is as follows: first, the similarity coefficient of texture features after repair is recalculated, and then the difference of the difference blocks before and after repair is compared. If the increase in similarity and the decrease in difference both meet the threshold, the repair is considered effective. The standard for compliance is set to a feature fusion degree of not less than 0.9 (the threshold is determined by the following method: after secondary repair, the visible difference between the repainted area and the surrounding coating should be close to disappearing. 0.9 is derived from the stable lower bound of similarity after repair of qualified samples in the trial production stage, and is adjusted upwards in combination with visual consistency requirements to form a quality threshold). To suppress transient lighting disturbances, a continuity condition is introduced for compliance determination: a compliance verification state is output only if the fusion degree of three consecutive frames is not less than 0.9 (parameter determination method: three frames correspond to 0.06 seconds at a frame rate of 50 frames per second, covering transient fluctuations caused by short-term reflections and vibrations, while not significantly lengthening the beat). If the fusion degree of three consecutive frames does not meet 0.9, a non-compliance verification state is output, and the coordinates of the local area of ​​the difference block and the corresponding entry angle and spraying distance are recorded as the basis for process optimization, for subsequent parameter revision of the spraying angle change and distance dynamic adjustment strategy. This process involves spatiotemporal registration of pose and reflectivity maps, similarity quantification determination, difference block localization, secondary entry angle and spraying distance calculation, secondary control command execution, and fusion degree verification.

[0065] Based on the execution results of the secondary respray control command, the final coating quality consistency data is obtained. The time consumption information of the integrated spraying control process for detection and respraying is recorded to improve equipment utilization efficiency. This serves as the basis for optimizing the automation level of the spraying control method. Specifically, this includes obtaining the execution status feedback sequence and process timestamp after the secondary respray control command is executed; generating coating quality consistency data based on the execution status feedback sequence; mapping the coating quality consistency data with the process timestamps to extract the time consumption of each stage to calculate the equipment utilization efficiency index; locating automation bottleneck nodes based on the equipment utilization efficiency index and generating an optimization weight matrix; and calculating the automation level score by combining the optimization weight matrix and the coating quality consistency data, which serves as the basis for optimizing the automation level of the spraying control method.

[0066] In this embodiment, the execution status feedback sequence after the secondary spraying control command is obtained by the control system sampling at fixed intervals. The sampling period is set to no more than 0.01 seconds (parameter determination method: consistent with the scheduling rhythm of the robot arm 3 motion control and spray gun opening and closing control, so that the status feedback covers the key nodes of each posture adjustment and distance adjustment, avoiding missed sampling that would cause a break in the quality sequence). The execution status feedback sequence includes at least the end effector pose following status, spraying opening and closing status, deviation status between the actual distance of the spray gun and the target distance, posture error status, texture feature similarity coefficient output by the dynamic detection system, feature fusion degree, residual difference degree of difference block, secondary spraying start flag, secondary spraying end flag, verification acquisition start flag, and verification acquisition end flag. The above flag-type statuses are set by the controller at the moment the corresponding action occurs and are maintained for several subsequent sampling periods to ensure that the event is reliably acquired; the number of maintenance periods is set to no less than 3 sampling periods (parameter determination method: 3 sampling periods correspond to 0.03 seconds, covering the risk of short-term missed sampling caused by communication jitter and sampling asynchrony, while not affecting the event sequence determination).

[0067] The process timestamps are generated by the controller to mark key events, with a timestamp resolution set to no less than 0.001 seconds (parameter determination method: stage time consumption statistics need to distinguish between adjacent events such as spray start, spray end, and verification acquisition; 0.001 seconds is below the minimum interval of these events, ensuring stable consistency between event sorting and stage segmentation). Key event timestamps include at least the time of issuance of the secondary spray control command, the time of secondary spray initiation, the time of start of secondary spray, the time of end of secondary spray, the time of exit of secondary spray, the time of start of verification acquisition, and the time of end of verification acquisition. If the system also includes secondary spray path recalculation or spray gun parameter switching, timestamps are also generated for the start and end of recalculation, and the start and end of parameter switching, so that the calculation and switching time consumption is counted separately to avoid confusion with motion time consumption. To avoid events being marked repeatedly, the controller sets a deduplication condition for the same event: the minimum interval between two adjacent markings of the same event is set to be no less than 0.02 seconds (the threshold is determined as follows: 0.02 seconds is higher than the sampling period of 0.01 seconds and covers the controller's event refresh period; if the interval is lower than this, repeated marking is mostly caused by signal jitter, and the stage boundary is more stable after deduplication).

[0068] When generating coating quality consistency data based on the execution status feedback sequence, the system first aligns the quality indicators output by the dynamic detection system according to timestamps to form a quality sequence, and then extracts the stable interval from the quality sequence as the final coating quality consistency data. The alignment process is completed according to the principle of "taking the nearest value within the same timestamp window": for each quality indicator, with its acquisition time as the center, the system finds the pose sampling point and spraying opening / closing state closest to that time, and binds the three to the same record; the binding tolerance is set to no more than 0.005 seconds (threshold determination method: under the maximum relative rotation speed, the surface projection displacement corresponding to 0.005 seconds is less than 1 / 2 of the texture window coverage scale. Exceeding this tolerance will cause the quality indicator to be misaligned with the actual re-spraying landing point, affecting the reliability of the consistency data). The length of the stable interval is set to no less than 0.06 seconds (parameter determination method: under the condition that the image acquisition frame rate is no less than 50 frames per second, 0.06 seconds covers 3 consecutive frames to suppress transient fluctuations caused by short-term reflection and vibration, so that the final data represents the stable coating state). The starting point of the stable interval is determined by delaying the end of the verification acquisition by 0.02 seconds (the parameter is determined by using 0.02 seconds to avoid potential exposure switching and motion stoppage disturbances at the end of the verification acquisition). Within the stable interval, the mean values ​​of texture feature similarity coefficient and feature fusion degree are calculated, and the maximum value of residual difference degree for difference blocks is calculated. The mean value is used to characterize the overall consistency level, and the maximum value is used to characterize the most unfavorable local difference. The reason for using the maximum value instead of the mean value for residual difference degree is that local differences are more likely to cause visible defects and trigger rework. The final coating quality consistency data is expressed as a combination of "mean similarity of stable interval, mean fusion degree of stable interval, and maximum residual difference degree of stable interval", and is associated with the corresponding task number for subsequent joint evaluation with time-consuming data.

[0069] When mapping coating quality consistency data to process timestamps, the system first segments the process into stages based on timestamps, then links the quality consistency data to the end of each stage to form a correspondence between "stage input and quality result." Stage segmentation is sequentially formed according to timestamp boundaries: from instruction issuance to entry is the preparation stage; from entry to spraying start is the attitude and distance positioning stage; from spraying start to spraying end is the secondary spraying stage; from spraying end to exit is the exit stage; and from verification data collection start to verification data collection end is the verification stage. If path recalculation or parameter switching occurs, recalculation and switching stages are inserted between the corresponding timestamps. The time consumed in each stage is obtained by subtracting the stage start timestamp from the stage end timestamp. To eliminate non-process wait times caused by external line stoppages or safety interlocks, the system identifies wait segments and removes them from the stage duration: the criteria for a wait segment are that the duration of a single segment exceeds 1.0 second and the spraying opening / closing state is closed (threshold determination method: under normal process cycle time, a single segment wait should not exceed 1.0 second; exceeding this value usually stems from external conveying, interlocks, or manual intervention; simultaneously, spraying closure indicates that the segment does not contribute to coating quality, and removing it makes the equipment utilization efficiency index closer to the process itself). The boundary of the wait segment is confirmed by the continuous occurrence of "movement speed close to 0 and spraying closed" in the execution status feedback sequence, with a continuous occurrence duration threshold of 0.2 seconds (threshold determination method: 0.2 seconds covers short pauses and interpolation transitions; exceeding this duration better matches the wait characteristics).

[0070] The equipment utilization efficiency index is calculated using two statistical methods: "time utilization" and "output utilization," forming a comprehensive index. The time utilization index is obtained by comparing effective operating time with total process time. Effective operating time is the cumulative time spent automatically executed by the system during the preparation, arrival, painting, departure, verification, and recalculation / switching phases. Total process time is the time span from the instruction issuance to the verification completion time. Comparing these two values ​​yields the equipment's time utilization level within the task cycle. The output utilization index is obtained through the number of qualified tasks per unit time. Tasks with a verified "meeting the standard" status are counted, and the total process time of the corresponding tasks is used as the denominator to obtain the ability to complete qualified repairs per unit time. To ensure comparability of indicators across different shifts and operating conditions, the system performs interval normalization on time utilization and output utilization separately. This normalization is achieved by using the upper and lower limits of historical stable operating intervals as reference boundaries. These reference boundaries are determined by statistical results from the trial and stable operation phases, with a statistical period of at least 100 tasks (parameter determination method: 100 tasks cover multiple defect types and multiple rotation phases, resulting in more stable statistical results). The comprehensive equipment utilization efficiency index is obtained by weighting time utilization and output utilization, with weights set at 0.6 and 0.4 respectively (parameter determination method: in the spray painting repair scenario, time utilization directly affects the production line cycle time and has a higher priority, hence the higher weight of time utilization; 0.6 and 0.4 emphasize cycle time constraints without sacrificing quality assessment).

[0071] When locating automation bottleneck nodes based on equipment utilization efficiency indicators, the system primarily filters bottleneck candidates by the percentage of time spent in each stage, and then combines this with quality benefits to determine whether there is inefficient investment. The percentage of time spent in a stage is obtained by dividing the time spent in a certain stage by the total process time. If the percentage exceeds 0.35, it is identified as a bottleneck candidate stage (threshold determination method: the stage distribution of the integrated process tends to be balanced, and a single stage percentage exceeding 0.35 means that the stage significantly slows down the cycle time and has stable identifiability). Quality benefits are further calculated for bottleneck candidate stages: quality benefits are the improvement in quality consistency data at the end of the stage compared to before the start of the stage. An improvement of less than 0.05 is considered inefficient investment (threshold determination method: 0.05 corresponds to the minimum visible improvement in similarity or fusion degree on a scale of 0 to 1; below this improvement means that the time spent has not been significantly converted into consistency benefits). After the bottleneck stage is determined, an optimization weight matrix is ​​generated. The optimization weight matrix is ​​arranged with stages as rows and optimization dimensions as columns. The optimization dimensions include at least the percentage of time spent, the degree of insufficient quality benefits, the number of secondary spray triggers, the number of recalculations, the number of posture positioning failures, and the number of distance positioning failures. The weights for each dimension are calculated in the following order: First, the time consumption percentage is normalized to a range of 0 to 1, with higher percentages having larger values. Next, the degree of insufficient quality gains is normalized to "target improvement minus actual improvement," with the target improvement set to 0.1 (parameter determination method: 0.1 corresponds to the expected visible improvement level in similarity or fusion after one or two re-sprays; a level below this indicates low efficiency of the repair strategy). Then, the number of events is normalized per unit task, with higher numbers having larger values. The final stage weights are obtained by summing the above normalized values ​​according to preset weights, with the time consumption percentage weight set at 0.5, the insufficient quality gains weight at 0.3, and the number of events weight at 0.2 (parameter determination method: bottleneck identification is centered on cycle time, with the time consumption percentage having the greatest impact; insufficient quality gains reflect an imbalance between input and output, and have a secondary weight; the number of events reflects stability issues and serves as an auxiliary constraint). After the weight matrix is ​​generated, the stage with the highest weight corresponds to the priority optimization node, used to guide subsequent resource allocation and parameter adjustment.

[0072] When calculating the automation level score by combining the optimized weight matrix (atrix) with coating quality consistency data, the system uniformly maps "quality compliance level, quality stability, time efficiency, secondary repair dependency, and bottleneck severity" into a score output. The score output range is set from 0 to 100 (parameter determination method: to facilitate integration with the equipment management system's scoring system and support hierarchical decision-making). The quality compliance level is determined by the average similarity and average fusion of the final coating quality consistency data; quality stability is determined by the maximum residual difference; time efficiency is determined by the comprehensive equipment utilization efficiency index; secondary repair dependency is determined by the proportion of secondary repainting triggers to the total number of tasks; and bottleneck severity is determined by the maximum weight value of the optimized weight matrix. Each indicator is first normalized separately, and then aggregated according to its weight to form a total score. The weights are set as follows: quality compliance 0.35, quality stability 0.15, time efficiency 0.30, secondary repair dependency 0.10, and bottleneck severity 0.10 (parameter determination method: quality compliance and time efficiency are the core objectives and have the highest weight; stability is used to control the most unfavorable local differences; secondary repair dependency and bottleneck severity are used to reflect automation maturity). The scoring thresholds are set at 80 and 60 (threshold determination method: 80 corresponds to a stable state where most tasks do not require secondary spraying and the stage time distribution is balanced; 60 corresponds to a state where secondary spraying is frequently triggered and the proportion of bottleneck stages exceeds the limit for a long time; the thresholds are determined by the scoring distribution boundary between stable production lines and problematic production lines during the trial operation phase). When the score is below 60, the system selects the stage with the highest weight as the primary optimization node based on the optimization weight matrix, and outputs the event frequency statistics, time consumption composition, and quality benefit deficiencies in this stage separately for optimization decision-making; when the score is between 60 and 80, the system prioritizes reducing the time consumption of inefficient investment stages and reducing the number of secondary spraying triggers; when the score is not lower than 80, the system records the task as a stable sample for subsequent parameter regression and comparative analysis. Through the unified management of the above execution status feedback and timestamps, the generation of final quality consistency data, the extraction of stage time consumption and the calculation of equipment utilization efficiency, the bottleneck location and the generation of optimization weight matrix, and the output of automation level scores.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of spray control for a wave barrier painting robot, characterized by, include: By using a pre-established robotic arm rotation trajectory model, the relative position change data between the robotic arm and the wave-shaped guardrail surface during the rotation process is obtained. The spraying angle change and distance dynamic adjustment information are recorded at each time point to establish a dynamic spatial mapping relationship of the robotic arm in the rotation state for spraying control. Based on the dynamic spatial mapping relationship, a dynamic detection system is used to scan the surface of the wave guardrail in real time to obtain surface image data, and the coating defect areas in the image are initially marked to determine the initial position range of surface defect capture for spraying control. By conducting in-depth analysis of the initially marked defect areas, using real-time data processing technology to extract specific feature information of the defects, and combining it with the mapping data of spatial coordinates, the three-dimensional position of the defects on the surface of the wave guardrail is determined, and accurate defect positioning results for spraying control are obtained. Based on the accurate defect location results, the planning data of the reverse motion path for spraying control is generated, and the path parameters are adjusted according to the dynamic characteristics of the robot arm rotation process to match the path planning with the current operating state of the robot arm and determine the execution order of the re-spraying task. The process involves acquiring data on the relative positional changes between the robot arm and the corrugated guardrail surface during rotation using a pre-established robot arm rotation trajectory model, and recording changes in the spraying angle and dynamic distance adjustments at each time point. This establishes a dynamic spatial mapping relationship for the robot arm's rotation state used for spraying control, including: By using a pre-established rotary trajectory model, data on the relative positional changes between the robotic arm and the corrugated guardrail surface are obtained. For relative position change data, record the changes in spraying angle and dynamic distance adjustment information at each time point; Based on the information on changes in spraying angle and dynamic adjustment of distance, a dynamic spatial mapping relationship is established in the rotation state of the robotic arm; By employing dynamic spatial mapping relationships to compensate for the curvature of the wave-shaped guardrail, an optimized spraying path scheme is determined. The process involves using a dynamic detection system to scan the surface of the wave-shaped guardrail in real time based on a dynamic spatial mapping relationship, acquiring surface image data, and initially marking areas with coating defects in the images to determine the initial location range for capturing surface defects for spraying control. The dynamic detection system is driven by a pre-established dynamic spatial mapping relationship to collect real-time surface image data, which includes the curvature features of the guardrail. Identify anomalous pixel sets in real-time surface image data to generate a binary mask image of coating defect regions; The binary mask image of the coating defect area is registered with the dynamic spatial mapping relationship, a corresponding index table is established, and three-dimensional bounding box data is generated according to the corresponding index table. The initial location range for surface defect capture used for spray control was determined using 3D bounding box data.

2. The method of claim 1, wherein: The process involves performing in-depth analysis on the initially marked defect areas, extracting specific feature information of the defects using real-time data processing technology, and combining this with spatial coordinate mapping data to determine the three-dimensional position of the defects on the surface of the corrugated guardrail, thereby obtaining precise defect location results for spraying control. Obtain the refined feature descriptor of the defect generated based on the initially marked region, and determine the accurate two-dimensional contour mask of the defect region based on the refined feature descriptor of the defect; A precise two-dimensional contour mask is mapped onto a plane of synchronously acquired depth data, the set of depth values ​​within the corresponding range is extracted, and the distribution of the normal vector of the local surface is calculated. Using a pre-calibrated spatial mapping matrix, the depth value set and normal vector distribution are converted into a three-dimensional point cloud cluster in the waveform guardrail surface coordinate system; Surface fitting calculations are performed on the 3D point cloud clusters to generate accurate defect location results containing 3D position coordinates and attitude information.

3. The method of claim 2, wherein: Based on the accurate defect location results, the process of generating reverse motion path planning data for spraying control, and adjusting path parameters according to the dynamic characteristics of the robot arm's rotation process to match the path planning with the robot arm's current operating state, and determining the execution sequence of the re-spraying task includes: Obtain accurate defect location results, and calculate the inverse kinematics solution set based on the accurate defect location results; Initial inverse motion path planning data is generated based on the inverse kinematics solution set; Extract parameters from the initial reverse motion path planning data to calculate joint torque load. If the joint torque load exceeds the threshold, generate corrected path parameters. The path parameters are corrected and matched with the current operating state of the robot arm to determine the execution order of the respraying tasks that match the current operating state of the robot arm.

4. The spray control method of the wave-shaped guardrail spraying robot according to claim 3, characterized in that, It also includes driving the spraying robot's movement through path planning data, generating control commands for adjusting the respraying posture, adjusting the spray gun angle and the distance between the spray gun and the corrugated guardrail surface according to the defect location, so that the changes in the spraying angle and the dynamic adjustment of the distance are consistent, and obtaining a spraying control execution scheme for the respraying action, specifically including: Obtain the discrete trajectory point sequence from the path planning data, and calculate the surface normal vector and tangential curvature feature data at the defect location based on the discrete trajectory point sequence; The six-degree-of-freedom target pose matrix and target spraying distance are calculated based on the surface normal vector and tangential curvature characteristic data. Calculate the attitude error quaternion between the actual end pose and the six-DOF target pose matrix, and generate supplementary spray attitude adjustment control commands based on the attitude error quaternion. The control command for adjusting the attitude of the retardation spraying is converted into target torque and speed control signals to obtain the spraying control execution scheme for the retardation spraying action.

5. The spray control method of the wave-shaped guardrail spraying robot according to claim 4, characterized in that, It also includes real-time monitoring of the robot's posture data during the re-spraying process according to the spraying control execution plan, and combining the coating quality consistency information fed back by the dynamic detection system to make spraying control judgments. When the difference between the re-sprayed area and the surrounding coating is detected to exceed a preset threshold, a secondary re-spraying control command is triggered to determine whether the re-spraying effect meets the standard. Specifically, this includes: Acquire the spatial pose data of the end effector and the surface optical reflectivity distribution map during the robot arm's re-spraying process. Perform spatiotemporal registration between the surface optical reflectivity distribution map and the spatial pose data of the end effector to calculate the texture feature similarity coefficient. If the texture feature similarity coefficient is lower than the preset standard, the secondary spraying entry angle and spraying distance are determined based on the coordinates of local areas with significant differences. The secondary spraying control command is generated based on the secondary spraying entry angle and spraying distance. After the robot arm is driven to perform the repair action according to the secondary spraying control command, the image is collected and the feature fusion degree is calculated to obtain the verification status of whether the spraying effect meets the standard.

6. The spray control method of the wave-shaped guardrail spraying robot according to claim 5, wherein It also includes obtaining the final coating quality consistency data based on the execution results of the secondary respray control command, and recording the time consumption information of the integrated detection and respray spraying control process to improve equipment utilization efficiency, serving as the basis for optimizing the automation level of the spraying control method, specifically including: Obtain the execution status feedback sequence and process timestamp after the secondary spraying control command is executed, and generate coating quality consistency data based on the execution status feedback sequence; The coating quality consistency data is mapped to the process timestamps, and the time consumed in each stage is extracted to calculate the equipment utilization efficiency index.

7. The method of claim 6, wherein: The process of obtaining final coating quality consistency data based on the execution results of the secondary re-spraying control command, and recording the time consumption information of the integrated detection and re-spraying control process to improve equipment utilization efficiency, also serves as a basis for optimizing the automation level of the spraying control method. Based on equipment utilization efficiency indicators, the automation bottleneck nodes are located and an optimized weight matrix is ​​generated.

8. The spraying control method for the corrugated guardrail spraying robot according to claim 7, characterized in that: The process of obtaining final coating quality consistency data based on the execution results of the secondary re-spraying control command, and recording the time consumption information of the integrated detection and re-spraying control process to improve equipment utilization efficiency, also serves as a basis for optimizing the automation level of the spraying control method. An automation level score is calculated by combining the optimized weight matrix with coating quality consistency data, which serves as the basis for optimizing the automation level of the spraying control method.

Images