Spray coating workpiece mounting position coating precision removal system and method

Abstract

The application provides a spraying workpiece installation site coating precise removal system and method, relates to the surface treatment technical field in mechanical manufacturing, and the system comprises a visual scanning module, a positioning module, a control system, an executing mechanism, a real-time monitoring module and an auxiliary device, and the method comprises the following steps: system initialization and calibration steps, visual scanning and hole site three-dimensional positioning steps, path planning and laser parameter adaptation steps, multi-mechanical arm collaborative work deployment steps, coating removal execution and real-time regulation steps and quality detection and correction steps. The application realizes a dynamic coupling adjustment mechanism of the track point density and the laser parameter, improves the scanning rate to improve the efficiency, achieves the double optimization goals of efficiency and quality, and provides a basis for realizing efficient and intelligent laser coating removal.

Description

Technical Field

[0001] This invention relates to the field of surface treatment technology in mechanical manufacturing, specifically to a system and method for precise removal of coatings at the mounting position of sprayed workpieces, and more particularly to a method for precise removal of coatings at the mounting position of sprayed workpieces based on visual scanning positioning. Background Technology

[0002] In traditional techniques, coating removal from mounting holes primarily relies on manual operation. The process involves: before spraying, workers manually cover the mounting holes with tape to ensure they are not covered by coating during spraying; after spraying, the tape is manually removed from the hole surface, and then tools such as knives and sandpaper are used to trim any remaining coating at the edges until a clean substrate surface is exposed. In some scenarios, laser equipment with fixed parameters is used to assist in removal to improve efficiency. However, parameters such as laser power and scanning speed need to be preset manually and remain fixed throughout the process, making it impossible to adjust in real time according to the surface shape of the hole or the coating thickness.

[0003] When manually removing tape and residual coating from mounting holes using a knife, the process is inefficient (taking 3-5 minutes per workpiece) due to poor operational stability and reliance on experience, as well as insufficient removal accuracy (a defect rate of over 15%). Furthermore, it cannot meet the processing requirements of complex curved holes. In addition, the traditional process of covering and removing tape is cumbersome, easily leaving adhesive residue that affects subsequent installation accuracy. Moreover, laser equipment with fixed parameters is difficult to adapt to the removal requirements of multi-material composite coatings such as modified potassium silicate resin and epoxy resin, resulting in a high rework rate (up to 20%) and high labor and time costs.

[0004] The existing technology has the following technical defects: First, manual operation is limited by the stability and experience of the human hand, making it difficult to control the accuracy of coating removal at the edge of the hole, resulting in a defect rate of over 15%; Second, manual processing of a single workpiece takes 3-5 minutes and requires dedicated quality inspection, with labor costs accounting for 40%; Third, the traditional process includes steps such as tape covering, tearing, and manual trimming, which is not only cumbersome (up to 4 steps), but may also affect assembly accuracy due to residual adhesive residue. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a system and method for precise removal of coatings at the mounting position of sprayed workpieces.

[0006] According to the present invention, a system and method for precise removal of coatings at the mounting position of a sprayed workpiece are provided, the solution of which is as follows:

[0007] In a first aspect, a system for precise removal of coatings from the mounting position of a sprayed workpiece is provided, the system comprising:

[0008] Visual scanning module: By combining an industrial camera with a linear laser, it acquires two-dimensional images and three-dimensional point cloud data of the mounting holes on the workpiece. After image preprocessing, feature extraction (U-Net network with improved dual cross attention) and three-dimensional modeling (ICP algorithm and CAD model registration), it achieves high-precision positioning of the hole edge and outputs the data to the positioning module. It also extracts three-dimensional geometric features, including the normal change rate and principal curvature, and transmits them to the control system.

[0009] Positioning module: Receives positioning data from the vision scanning module, establishes coordinate transformation relationship with the actuator through the calibration plate, calibrates workpiece placement deviation, and transmits the corrected hole position coordinates to the control system to ensure the operating reference accuracy of the actuator (coordinate mapping accuracy ≤ ±0.05mm, registration error ≤ 0.03mm).

[0010] Control system: Receives 3D geometric features from the visual scanning module, coordinate transformation data from the positioning module, and coating state vector data from the real-time monitoring module. It quantifies the region complexity through a coupled response function, generates laser trajectory and parameter commands through a reinforcement learning strategy network, sends the optimized parameter commands to the actuator, and simultaneously receives debris concentration data from the auxiliary device for environmental adaptation.

[0011] Actuator: Receives trajectory and parameter instructions from the control system, performs coating removal operation through the ultraviolet laser head, and adaptively adjusts trajectory density, power, speed and incident angle according to parameter instructions in different complexity areas. During operation, it receives feedback from the real-time monitoring module to coordinate parameter adjustments.

[0012] Real-time monitoring module: Collects data such as coating thickness, surface temperature, and residue rate in real time (updated every 50ms), updates periodically, and integrates them into coating state vector data to feed back to the control system;

[0013] Auxiliary device: The negative pressure adsorption device simultaneously adsorbs the coating debris removed by the actuator, the optical sensor monitors the debris concentration and transmits the data to the control system. When the concentration exceeds the threshold, the cleaning program is automatically started, providing a clean processing environment for the actuator and the real-time monitoring module.

[0014] Preferably, in the actuator, the laser parameter range is adjusted to within a set range value;

[0015] In regions with abrupt changes in curvature, the path sampling frequency is automatically increased while the scanning speed and power are simultaneously reduced. In flat regions, the distance between trajectory nodes is increased to improve the scanning rate. The laser incident angle is dynamically adjusted, and the power and frequency are matched synchronously to ensure uniformity of removal.

[0016] Preferably, the real-time monitoring module includes: calculating the error by comparing the real-time collected coating thickness with a preset value; and triggering the control system to synchronously adjust the trajectory density and laser parameters when the thickness error exceeds a set threshold range or the coating surface temperature exceeds a set temperature value.

[0017] Preferably, the negative pressure value of the auxiliary device is maintained within a set range, and the distance between the adsorption port and the workpiece surface is controlled to be less than a preset value; an optical sensor is integrated in the adsorption airflow to detect the concentration of debris in real time, and the cleaning program is automatically started when the concentration exceeds the threshold.

[0018] Secondly, a method for precise removal of coating at the mounting position of a sprayed workpiece is provided, the method comprising:

[0019] System initialization and calibration steps: Perform parameter calibration on the coordinate system of the vision scanning module, laser thickness gauge, negative pressure device and robotic arm. Establish the coordinate transformation relationship between the vision scanning module and the actuator through the calibration board. Preset the negative pressure value and suction port distance. This step provides the equipment reference and coordinate association basis for all subsequent operations. The calibration data will support the accuracy of subsequent hole positioning and trajectory planning.

[0020] Visual scanning and 3D hole location steps: acquire images and denoise them, extract hole features through an improved U-Net network; combine linear laser to acquire 3D point cloud, construct a stereo model and register it with the CAD model, output the spatial coordinates of the hole, and transmit the positioning data to the control system as the initial geometric reference for path planning and parameter adaptation.

[0021] Path planning and laser parameter adaptation steps: Receive the 3D point cloud from the previous step, generate normal, curvature, and thickness mapping functions, quantify the region complexity through the coupled response function, link the reinforcement learning network to output the laser trajectory and parameter instructions (power, speed, etc.), and transmit the optimized trajectory and parameter instructions to the actuator to provide the operational basis for coating removal;

[0022] Multi-robotic arm collaborative operation deployment steps: Based on trajectory and parameter instructions, distributed AI is used to allocate robotic arm tasks, combined with dynamic obstacle avoidance planning of collaborative paths; the optical sensors of the auxiliary device are activated simultaneously to monitor debris concentration in real time and preset cleaning thresholds;

[0023] Coating removal execution and real-time control steps: Multiple robotic arms equipped with ultraviolet laser heads perform laser removal operations according to a planned trajectory. The real-time monitoring module collects thickness and temperature data at intervals and feeds it back to the control system. When the data exceeds the threshold, the control system adjusts the parameters within a set time. The negative pressure device simultaneously adsorbs debris. This step is the execution phase. The parameters are dynamically corrected through real-time data closed-loop feedback to ensure removal accuracy.

[0024] Quality inspection and correction steps: After the removal is completed, the vision scanning module scans the hole a second time to verify the coating removal effect, ensure the removal quality, and form a complete processing closed loop.

[0025] Preferably, the path planning and laser parameter adaptation steps include: generating a surface normal mapping tensor N(x,y,z) and a principal curvature field K(x,y,z) based on the 3D point cloud, and constructing a coupled response function L(x,y,z) from geometry to physical properties to energy demand by combining the coating thickness mapping function T(x,y,z):

[0026]

[0027] Where L(x,y,z) represents the complexity response weight of the current laser operation point; L represents the rate of change of normal; K is the curvature value; T is the local coating thickness; ω1, ω2, and ω3 are empirical adjustment weights, and the larger L is, the more careful the laser parameters at that point need to be and the denser the path needs to be.

[0028] Meanwhile, a reinforcement learning network is introduced into the laser parameter configuration. The appropriate laser operation action 'a' is selected through the policy network π(a|s). The state 's' is composed of the current path point's geometric state and sensor feedback. The reward / penalty function R(s,a) is defined as follows:

[0029] R(s,a)=-|E actual -E target |-λ·Δv-μ·Q residue

[0030] Among them, E actual E represents the actual unit energy density. target The coating removes the threshold target; Δv represents the velocity disturbance of the trajectory segment; Q residue This represents the detection residual rate in this area; λ and μ are adjustment parameters; the strategy network is trained on a large number of workpiece datasets and provides the optimal laser parameters in real time during execution to minimize energy deviation and residue.

[0031] Preferably, the path planning and laser parameter adaptation step further includes:

[0032] Step 1: Quantize local complexity using L(x,y,z);

[0033] The complexity of each laser operation point is calculated using the coupling response function L(x,y,z). The more complex the geometry and the thicker the coating, the larger the value of L.

[0034] Step 2: Convert L into target energy E target ;

[0035] Based on E target(L)=E0·(1+β·ln(1+L)) sets a high target energy density for high complexity regions and a relatively low target energy density for low complexity regions;

[0036] Step 3: Use R(s,a) to evaluate the match between the action and the target;

[0037] In the reward / punishment function R(s,a), E target It is no longer a fixed value, but a variable dynamically determined by L.

[0038] Preferably, in the third step, the penalty term of the reward / penalty function is directly related to the local complexity:

[0039] In the high L region, the actual energy E actual Not reached E target (L), then |E actual -E target | Increasing the R value decreases the penalty, forcing the policy network to output actions with higher power / lower speed;

[0040] If in the low L region, E actual More than E target (L) is also penalized for increased deviation, causing the network to reduce its energy output.

[0041] Preferably, the coating removal execution and real-time control step includes:

[0042] The laser thickness gauge updates the coating thickness data every set time interval and calculates the error by comparing it with the preset value. When the thickness error exceeds the set threshold range, the PID controller adjusts the laser power and scanning speed within the set time. If the coating surface temperature exceeds the set temperature value, the laser power is automatically reduced. The negative pressure adsorption device works synchronously to adsorb coating debris in real time.

[0043] Preferably, the quality inspection and correction includes:

[0044] If the defect is found, the system will automatically call the correction path and readjust the laser parameters based on real-time monitoring data for rework; once the defect is found to be acceptable, the auxiliary device cleaning program will be started to remove residual debris.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. In terms of technical effectiveness, existing technologies rely on manual operation, which is limited by the stability and experience of the human hand. The accuracy of coating removal at the hole edges is difficult to control, resulting in a defect rate of over 15%, and it cannot meet the processing requirements of complex curved holes. In this invention, the visual scanning module uses a U-Net network with improved dual cross-attention combined with 3D point cloud registration to control the hole edge positioning accuracy within 0.1 pixels. Combined with the ±0.05mm coordinate mapping accuracy of the positioning module, it provides a sub-millimeter-level reference for subsequent processing. The control system introduces a coupled response of geometric-physical-energy requirements. The combined strategy of function and deep reinforcement learning breaks the limitations of traditional trajectory and parameter separation, enabling laser parameters (power, speed, etc.) to dynamically adapt to the curvature of the hole and the coating thickness. In areas with steep curvature changes, the trajectory is automatically encrypted and the energy density is reduced, while in flat areas, the scanning rate is increased. This real-time collaborative adjustment mechanism keeps the coating removal depth error within ±2μm, with no overburning or residue at the edges, and the defect rate is reduced to below 0.3%. It can also adapt to multi-material composite coatings such as modified potassium silicate resin and epoxy resin, which is something that existing laser equipment with fixed single parameters or manual operation cannot achieve.

[0047] 2. In terms of efficiency and cost, existing technologies require an average of 3-5 minutes for manual removal of coating from holes in a single workpiece, and necessitate dedicated quality inspectors, with labor costs accounting for 40% of the total cost. This invention achieves a leap in efficiency and cost optimization through multi-dimensional technological innovation: the actuator employs B-spline interpolation trajectory and dynamic incident angle adjustment, coupled with distributed AI task allocation in the control system, increasing the single-arm operation efficiency to 15-20 seconds per piece. When multiple robotic arms work collaboratively, parallel processing is achieved through a dynamic obstacle avoidance response of ≤50ms, resulting in an overall efficiency improvement of 8-10 times. The real-time monitoring module updates coating data and triggers PID adjustments every 50ms, avoiding rework caused by parameter mismatch in traditional processing, reducing the rework rate from 20% to below 1%. The auxiliary device's negative pressure adsorption and debris concentration monitoring are linked, reducing coating debris contamination of the workpiece surface and lowering subsequent cleaning process costs by 60%. Furthermore, the reinforcement learning strategy network achieves parameter self-optimization through workpiece dataset training, eliminating the need for repeated manual adjustments. The equipment debugging cycle is shortened from 2-3 days for traditional laser equipment to 4 hours, significantly reducing the technical threshold and maintenance costs.

[0048] 3. Traditional methods typically rely on empirically defined laser parameter tables, lacking robustness to various coating types, substrate differences, and surface variations. They suffer from low efficiency and high residue rates when handling complex conditions. This invention introduces a deep reinforcement learning policy network to construct an intelligent laser operation decision model. This model uses real-time 3D geometric state and sensor feedback as state inputs, and power, scanning speed, and incident angle as the action space. The model is continuously trained to converge to the optimal energy control strategy by setting a comprehensive reward and penalty function that includes unit energy density deviation, trajectory perturbation penalty, and regional residue rate. Unlike existing technologies, which primarily use reinforcement learning for path optimization rather than energy control, this invention is the first to apply reinforcement learning to fine-tuning scenarios of local energy density and coordinate it with a complexity response function. This allows the system to automatically provide the most suitable laser output strategy for regions of different complexity, significantly reducing the probability of overheating and coating residue, achieving a dual improvement in technical effectiveness and control granularity.

[0049] Other beneficial effects of the present invention will be explained in detail through the introduction of specific technical features and technical solutions in specific embodiments. Those skilled in the art should be able to understand the beneficial technical effects brought about by these technical features and technical solutions through the introduction of these technical features and technical solutions. Attached Figure Description

[0050] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0051] Figure 1 This is a module framework diagram of the present invention;

[0052] Figure 2 This is a flowchart illustrating the collaborative optimization process of paths and parameters in this invention.

[0053] Figure 3 This is a diagram illustrating the path and parameter adaptation strategy driven by the complexity of this invention. Detailed Implementation

[0054] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0055] This invention provides a system for precise removal of coatings at the mounting position of a sprayed workpiece, referring to... Figure 1 As shown, it specifically includes:

[0056] The visual scanning module uses an industrial camera and a linear laser to acquire 2D images and 3D point cloud data of the workpiece mounting holes. It extracts edge features of the holes using the U-Net semantic segmentation algorithm and calculates their spatial coordinates using laser triangulation. The original images are binarized, and noise is removed through morphological operations. An improved U-Net network with a dual-cross attention module (DCA) is used to extract multi-scale features of the holes, achieving an edge positioning accuracy of ≤0.1 pixels. A 3D model of the holes is constructed using the 3D point cloud data, and the ICP algorithm is used to register it with the CAD model to correct positioning errors. Simultaneously, the local normal variation rate of the hole area is extracted. Three-dimensional geometric features such as principal curvature K.

[0057] Positioning module: Establishes the coordinate transformation relationship between the industrial camera and the actuator through the calibration board, dynamically calibrates the workpiece placement deviation, and ensures that the accuracy of the hole position coordinates mapped to the actuator coordinate system is ≤ ±0.05mm, and the registration error between the 3D point cloud and the CAD model is ≤0.03mm.

[0058] Control system: Taking local 3D geometric features and coating property information from the vision scanning module as core inputs, it receives coordinate data from the positioning module, integrates the laser action-material response model and deep reinforcement learning algorithm, and constructs a coupled response function L(x,y,z) from geometry to physical properties to energy requirements; Based on this function, it dynamically generates the laser removal trajectory, and simultaneously optimizes parameters such as laser power, scanning speed, and incident angle to achieve coordinated convergence of trajectory and parameters; The optimal operation action is output in real time through the strategy network π(a|s) to ensure that both adapt to the workpiece state based on real-time feedback.

[0059] Actuator: Equipped with an ultraviolet laser head, it performs cold processing to remove the coating at the mounting holes according to the instructions of the control system. The laser parameters are: wavelength 355nm, power 20-50W, scanning speed 200-800mm / s, pulse frequency 50-150kHz. In areas with steep curvature (high L value), it automatically increases the path sampling frequency (refines the trajectory) while simultaneously reducing the scanning speed and power. In flat areas (low L value), it lengthens the trajectory node spacing to increase the scanning rate. When dynamically adjusting the laser incident angle (±15°), it simultaneously matches the power and frequency to ensure uniform removal.

[0060] Real-time monitoring module: Provides real-time feedback of coating thickness T via a laser thickness gauge, simultaneously integrates an infrared temperature sensor to monitor surface temperature, and detects coating residue rate Q via image recognition. residueEvery 50ms, data such as thickness, temperature, and residual rate are integrated into a coating property state vector, which is then input into the coupled response function and reinforcement learning strategy network of the control system. The error is calculated by comparing it with the preset value. When the thickness error exceeds ±5μm or the temperature exceeds 80℃, the control system is triggered to synchronously adjust the trajectory density and laser parameters, with a response time ≤100ms.

[0061] Auxiliary device: An integrated negative pressure adsorption device adsorbs coating debris during the removal process to prevent contamination of other areas; the negative pressure value is maintained between -0.05MPa and -0.1MPa, and the distance between the adsorption port and the workpiece surface is ≤2mm; an optical sensor is integrated in the adsorption airflow to detect the debris concentration in real time. When the concentration exceeds the threshold, the cleaning program is automatically started, and the debris concentration data is input into the control system as an environmental state variable.

[0062] This invention also provides a method for precise removal of coating at the mounting position of a sprayed workpiece, referring to... Figure 2 and Figure 3 As shown, it specifically includes:

[0063] 1. System initialization and calibration steps: Perform parameter calibration on the vision scanning module (industrial camera and linear laser combination), and establish the coordinate transformation relationship with the actuator through the calibration board; synchronously calibrate the laser thickness gauge, negative pressure adsorption device and multi-robotic arm coordinate system to ensure that the control accuracy of the robotic arm spacing is ≤±2cm, the negative pressure value is preset to -0.05MPa to -0.1MPa, and the distance between the adsorption port and the workpiece surface is adjusted to ≤2mm.

[0064] 2. Visual scanning and 3D hole positioning steps: An industrial camera acquires images of the workpiece surface, which are first binarized and subjected to morphological operations to remove noise. Then, a U-Net network with improved dual cross-attention is used to extract multi-scale features of the mounting holes, achieving an edge positioning accuracy of ≤0.1 pixels. Simultaneously, a linear laser is used to acquire 3D point cloud data, constructing a 3D model of the hole. The ICP algorithm is used to register the model with the CAD model to correct positioning errors, and finally, the spatial coordinates of the hole are output.

[0065] 3. Path planning and laser parameter adaptation steps: Driven by local three-dimensional geometric features and coating property information, by introducing a laser action-material response model and a deep reinforcement learning joint optimization strategy, the trajectory and parameters are no longer two independent processes, but converge to the optimal operation solution based on real-time workpiece feedback information.

[0066] In specific implementation, the system first generates a surface normal mapping tensor N(x,y,z) and principal curvature field K(x,y,z) based on the 3D point cloud, and then constructs a coupled response function L(x,y,z) from geometry to physical properties to energy demand by combining the coating thickness mapping function T(x,y,z):

[0067]

[0068] Where L(x,y,z) represents the complexity response weight of the current laser operation point; L represents the rate of change of normal (corresponding to the need for trajectory densification at the point of curvature abrupt change); K is the curvature value (reflecting the surface unevenness); T is the local coating thickness; ω1, ω2, and ω3 are empirical adjustment weights, and the larger L is, the more careful the laser parameters at that point need to be and the denser the path needs to be; this function can dynamically drive the adjustment of trajectory point density and fine matching of parameters.

[0069] Meanwhile, a reinforcement learning network is introduced into the laser parameter configuration. The appropriate laser operation action a (including power, scanning speed, incident angle, etc.) is selected through the policy network π(a|s). The state s is composed of the geometric state of the current path point and sensor feedback. The reward and penalty function R(s,a) is defined as follows:

[0070] R(s,a)=-|E actual -E target |-λ·Δv-μ·Q residue

[0071] Among them, E actual E represents the actual unit energy density. target The coating removes the threshold target; Δv represents the velocity disturbance of the trajectory segment; Q residue This represents the detection residual rate in this area; λ and μ are adjustment parameters; the strategy network is trained on a large number of workpiece datasets and provides the optimal laser parameters in real time during execution to minimize energy deviation and residue.

[0072] Specifically:

[0073] Step 1: Quantize local complexity using L(x,y,z);

[0074] The complexity of each laser operation point is calculated using the coupling response function L(x,y,z). (The more complex the geometry and the thicker the coating, the larger the value of L.)

[0075] Step 2: Convert L into target energy E target ;

[0076] Based on E target (L) = E0·(1+β·ln(1+L)) sets a high target energy density for high complexity regions (large L) (requiring stronger energy to remove thick coatings or complex curved surface coatings), and sets a relatively low target energy for low complexity regions (small L) (to avoid excessive energy leading to substrate damage).

[0077] Step 3: Use R(s,a) to evaluate the match between the action and the target;

[0078] In the reward / punishment function R(s,a), E target It is no longer a fixed value, but a variable dynamically determined by L.

[0079] At this point, the penalty term of the reward / penalty function is directly related to the local complexity: if in the high L region, the actual energy E actual Not reached E target (L), then |E actual -E target Increasing the R value decreases the penalty, forcing the policy network to output actions with higher power / lower speed; in the low L region, E... actual More than E target (L) is also penalized for increased deviation, causing the network to reduce its energy output.

[0080] During the path execution phase, this invention implements a dynamic coupling adjustment mechanism between the density of trajectory points and laser parameters. For example, in regions with steep curvature changes, the system automatically increases the path sampling frequency, reduces the laser scanning speed, and decreases the pulse energy density to prevent overheating or uneven energy distribution. In flat regions, the system automatically lengthens the spacing between trajectory control nodes and increases the scanning rate to improve efficiency, achieving the dual optimization goals of efficiency and quality.

[0081] This invention offers significant advantages over traditional trajectory-parameter separation strategies: the path trajectory more closely matches real-world geometric complexity, and the laser parameter configuration provides real-time feedback and adaptive learning, demonstrating greater adaptability in multi-material composite coating removal tasks. This joint optimization scheme provides precise and coordinated trajectory and energy input for subsequent robotic arm collaborative operations, laying the foundation for efficient and intelligent laser coating removal.

[0082] 4. Deployment steps for multi-robotic arm collaborative operation: Use distributed AI algorithm to allocate tasks to each robotic arm, and combine dynamic obstacle avoidance algorithm (response time ≤ 50ms) to plan collaborative path; synchronously start the optical sensor of the auxiliary device to monitor the debris concentration in real time and preset the cleaning threshold.

[0083] 5. Coating Removal Execution and Real-time Control Steps: Multiple robotic arms equipped with ultraviolet laser heads perform the removal operation according to the planned trajectory. The laser thickness gauge updates the coating thickness data every 50ms and calculates the error by comparing it with the preset value. When the error exceeds ±5μm, the PID controller adjusts the laser power and scanning speed within 100ms. If the coating surface temperature exceeds 80℃, the laser power is automatically reduced by 10%-20%. The negative pressure adsorption device works synchronously to adsorb coating debris in real time.

[0084] 6. Quality Inspection and Correction Steps: After removal is completed, the vision scanning module scans the hole positions a second time to verify the coating removal effect; if there are defects (such as edge residue or substrate damage), the system automatically calls the correction path and readjusts the laser parameters (power 20-50W, scanning speed 200-800mm / s) based on real-time monitoring data for rework; after passing the test, the auxiliary device cleaning program is started to remove residual debris.

[0085] Using the method of this invention, the single-hole removal time is ≤10 seconds, which is 80% more efficient than manual methods; the defect rate is reduced from 5% in manual operations to below 0.3%. The coating material compatibility of this method includes: applicability to organic coatings such as modified potassium silicate resin and epoxy resin; no heat-affected zone on the metal substrate; and a surface roughness Ra ≤1.6μm.

[0086] The present invention will now be described in more detail.

[0087] Taking an automobile engine block (made of gray cast iron, with a surface coated with a composite coating of modified potassium silicate resin and bisphenol A epoxy resin, containing 8 M12 mounting holes, with a 5mm radius transition surface at the edge of the holes) as an example, the following 6 steps are used to achieve precise removal of the coating at the mounting points:

[0088] Step 1: The visual scanning module collects and processes data.

[0089] A 20-megapixel industrial camera combined with a 650nm linear laser was used to scan the cylinder mounting hole area, acquiring a 2D image with a resolution of 2048×1536 and a point cloud density of 100 points / mm. 2 The three-dimensional data was used. Adaptive threshold binarization was applied to the original image, and noise points caused by bubbles on the coating surface were removed using opening operations. A U-Net network improved with a dual cross-attention module (DCA) was used to extract multi-scale features of the hole edges, controlling the edge localization accuracy of the eight holes to within 0.08 pixels. A three-dimensional model of the holes was constructed using the 3D point cloud, and registered with the cylinder block CAD model using the ICP algorithm (registration error ≤ 0.03 mm). Simultaneously, the local normal variation rate of the transition surface of the holes was calculated. (Maximum 0.8 rad / mm) and principal curvature K (range -0.05 to 0.1 mm) -1 ).

[0090] Step 2: The positioning module establishes coordinate association.

[0091] A calibration plate is fixed to the cylinder's reference surface (flatness ≤ 0.02 mm). Images of the calibration plate are captured by a camera to obtain internal and external parameters, and a transformation matrix is ​​established between the camera coordinate system and the six-axis robotic arm's execution coordinate system. The placement deviation of the cylinder caused by the tooling clamping is dynamically detected (maximum deviation ≤ ±0.5 mm). The three-dimensional coordinates of the eight hole positions (error ≤ ±0.05 mm) are mapped to the robotic arm coordinate system through coordinate transformation to ensure the matching accuracy between the laser head's movement trajectory and the hole positions.

[0092] Step 3: The control system generates a joint optimization strategy.

[0093] Receive geometric features output by the visual scanning module The system constructs a coupled response function L(x,y,z) based on the preset coating thickness distribution (80μm for the planar region of the cylinder block and 120μm for the transition surface region). When the L value of the transition surface at the hole reaches 1.2 (0.5 for the planar region), the system automatically calls the deep reinforcement learning policy network π(a|s) to generate the corresponding laser parameters: 35W power, 500mm / s scanning speed, and 100kHz pulse frequency for the planar region; and a density of trajectory points of 5 points / mm for the transition surface region, while simultaneously reducing the power to 28W and adjusting the scanning speed to 300mm / s to ensure a stable energy density of 8J / mm². 2 .

[0094] Step 4: The actuator performs laser removal.

[0095] A robotic arm equipped with a 355nm ultraviolet laser head initiates operation according to commands from the control system. In the central region of the hole, the laser head moves along a B-spline interpolation trajectory, with speed fluctuations controlled within ±3%. Upon reaching the transition surface, the laser incident angle is dynamically adjusted (range -12° to +15°) to ensure the laser spot remains perpendicular to the surface normal. For the circumferential distribution of eight holes, the robotic arm employs distributed motion planning, with adjacent hole switching time ≤0.5s and single-hole removal time controlled within 12 seconds.

[0096] Step 5: The real-time monitoring module provides dynamic feedback and adjustments.

[0097] A laser thickness gauge collects coating thickness data every 50ms (accuracy ±1μm), while an infrared temperature sensor simultaneously monitors the coating surface temperature (initial temperature 25℃). When a coating thickness error of +6μm (actual 126μm) is detected in the transition curved area, the PID controller is triggered to increase the laser power to 30W within 80ms; when the temperature rises to 82℃, the power is automatically reduced by 15% to 23.8W to prevent oxidation and discoloration of the gray cast iron substrate due to overheating (temperature controlled below 75℃). The coating residue rate Q is detected in real time through image recognition. residueWhen the residual rate at the edge of a certain hole reaches 3%, the strategy network automatically extends the scanning time of that area by 0.8 seconds.

[0098] Step 6: Auxiliary devices ensure the processing environment.

[0099] A ring-shaped negative pressure adsorption port is set 2 mm below the laser removal area, maintaining a negative pressure of -0.08 MPa. Coating debris (particle size range 5-50 μm) is adsorbed by an airflow velocity of 5 m / s. An integrated optical sensor in the airflow monitors the debris concentration in real time; when the concentration exceeds 50 mg / m³... 3 When the high-pressure air gun cleaning program is activated (air pressure 0.5MPa, lasting 2 seconds), it prevents debris from adhering to the cylinder sealing surface and affecting subsequent assembly.

[0100] This invention provides a system and method for precise removal of coatings from the mounting position of sprayed workpieces, covering a complete process from system initialization and multi-device calibration, 3D visual positioning and point cloud matching, to path planning and joint optimization of laser parameters. The key lies in constructing a geometry-property-energy coupled response function and a complexity-driven mechanism to achieve dynamic linkage adjustment of trajectory density and laser parameters, and introducing a reinforcement learning strategy network to minimize energy deviation and residue rate. During the execution phase, the system achieves high-precision and high-efficiency laser coating removal through multi-robotic arm collaborative operation, combined with real-time coating thickness detection and temperature control feedback. Finally, a quality inspection and adaptive correction mechanism ensures excellent removal results without thermal damage to the substrate. This method is compatible with various organic coating materials and significantly improves operational efficiency and quality stability compared to traditional manual operations, demonstrating promising prospects for industrial application.

[0101] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0102] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for precise removal of coating at the mounting position of a sprayed workpiece, characterized in that, include: System initialization and calibration steps: Perform parameter calibration on the coordinate system of the vision scanning module, laser thickness gauge, negative pressure device and robotic arm; establish the coordinate transformation relationship between the vision scanning module and the actuator through the calibration board; preset the negative pressure value and suction port distance. Visual scanning and 3D hole location steps: acquire images and denoise them, and extract hole features through an improved U-Net network; By combining linear laser to acquire three-dimensional point clouds, a stereo model is constructed and registered with the CAD model. The spatial coordinates of the hole positions are output, and the positioning data is transmitted to the control system as the initial geometric reference for path planning and parameter adaptation. Path planning and laser parameter adaptation steps: Receive the 3D point cloud from the previous step, generate normal, curvature, and thickness mapping functions, quantify the region complexity through the coupling response function, link the reinforcement learning network to output the laser trajectory and parameter instructions, and transmit the optimized trajectory and parameter instructions to the actuator. Multi-robotic arm collaborative operation deployment steps: Based on trajectory and parameter instructions, distributed AI is used to allocate robotic arm tasks, combined with dynamic obstacle avoidance planning of collaborative paths; The optical sensor of the auxiliary device is activated simultaneously to monitor the debris concentration in real time and preset the cleaning threshold. Coating removal execution and real-time control steps: Multiple robotic arms equipped with ultraviolet laser heads perform laser removal operations according to a planned trajectory. The real-time monitoring module collects thickness and temperature data at intervals and feeds it back to the control system. When the data exceeds the threshold, the control system adjusts the parameters within a set time, and the negative pressure device simultaneously adsorbs the debris. Quality inspection and correction steps: After the removal is completed, the vision scanning module scans the hole a second time to verify the coating removal effect, ensure the removal quality, and form a complete processing closed loop; The path planning and laser parameter adaptation steps include: generating a surface normal mapping tensor based on a 3D point cloud. and the main song And combined with coating thickness mapping function Construct a coupled response function from geometry to physical properties to energy demand. : in, This represents the complexity response weight of the current laser operation point; Indicates the rate of change of the normal direction; It is the curvature value; It refers to the local coating thickness; , , Adjusting weights based on experience, The larger the value, the more careful the laser parameters at that point need to be, and the denser the laser path needs to be. Simultaneously, a reinforcement learning network is introduced into the laser parameter configuration, through a policy network. Select appropriate laser operation actions ,state The reward / penalty function is composed of the current path point's geometric state and sensor feedback. The definition is as follows: in, Represents actual unit energy density, Remove threshold targets for coating; Indicates velocity disturbance in the trajectory segment; This indicates the residual detection rate in the area; , To adjust the parameters, the strategy network is trained on a large dataset of workpieces and provides the optimal laser parameters in real time during execution to minimize energy deviation and residue.

2. The method for precise removal of coating at the mounting position of a sprayed workpiece according to claim 1, characterized in that, The path planning and laser parameter adaptation steps also include: Step 1: Use Quantify local complexity; By coupling response function The complexity of each laser operation point was calculated; the more complex the geometry and the thicker the coating, the better. The larger the value; Step 2: Convert into target energy ; based on Set a high target energy density for high-complexity regions and a relatively low target energy for low-complexity regions; Step 3: Use Assess the alignment between actions and objectives; Reward and punishment functions middle, It is no longer a fixed value, but rather determined by... Variables that are dynamically determined.

3. The method for precise removal of coating at the mounting position of a sprayed workpiece according to claim 2, characterized in that, In the third step, the penalty term of the reward / penalty function is directly related to the local complexity: If in high Region, actual energy Not achieved ,but Increase The lower the value, the stronger the penalty, forcing the policy network to output actions with higher power or lower speed. If at low area, Exceed Similarly, the increased deviation results in a penalty, causing the network to reduce its energy output.

4. The method for precise removal of coating at the mounting position of a sprayed workpiece according to claim 1, characterized in that, The coating removal execution and real-time control steps include: The laser thickness gauge updates the coating thickness data every set time interval and compares it with the preset value to calculate the error. When the thickness error exceeds the set threshold range, the PID controller adjusts the laser power and scanning speed within the set time. If the coating surface temperature exceeds the set temperature value, the laser power is automatically reduced. The negative pressure adsorption device works synchronously to adsorb coating debris in real time.

5. The method for precise removal of coating at the mounting position of a sprayed workpiece according to claim 1, characterized in that, The quality inspection and correction include: If the defect is found, the system will automatically call the correction path and readjust the laser parameters based on real-time monitoring data for rework; once the defect is found to be acceptable, the auxiliary device cleaning program will be started to remove residual debris.

6. A system for precise removal of coatings from the mounting position of a sprayed workpiece, characterized in that, For performing the method for precise removal of coating at the mounting position of a sprayed workpiece as described in any one of claims 1-5, the removal system comprises: Visual scanning module: By combining an industrial camera with a linear laser, it acquires two-dimensional images and three-dimensional point cloud data of the workpiece mounting hole positions. After image preprocessing, feature extraction and three-dimensional modeling, it achieves high-precision positioning of the hole position edges and outputs the data to the positioning module. It also extracts three-dimensional geometric features, including the normal change rate and principal curvature, and transmits them to the control system. Positioning module: Receives positioning data from the vision scanning module, establishes a coordinate transformation relationship with the actuator through a calibration plate, calibrates the workpiece placement deviation, and transmits the corrected hole position coordinates to the control system to ensure the operating reference accuracy of the actuator; Control system: Receives 3D geometric features from the visual scanning module, coordinate transformation data from the positioning module, and coating state vector data from the real-time monitoring module. It quantifies the region complexity through a coupled response function, generates laser trajectory and parameter commands through a reinforcement learning strategy network, sends the optimized parameter commands to the actuator, and simultaneously receives debris concentration data from the auxiliary device for environmental adaptation. Actuator: Receives trajectory and parameter instructions from the control system, performs coating removal operation through the ultraviolet laser head, and adaptively adjusts trajectory density, power, speed and incident angle according to parameter instructions in different complexity areas. During operation, it receives feedback from the real-time monitoring module to coordinate parameter adjustments. Real-time monitoring module: Collects data including coating thickness, surface temperature, and residue rate in real time, updates it periodically, and integrates it into coating state vector data to feed back to the control system; Auxiliary device: The negative pressure adsorption device simultaneously adsorbs the coating debris removed by the actuator, the optical sensor monitors the debris concentration and transmits the data to the control system. When the concentration exceeds the threshold, the cleaning program is automatically started, providing a clean processing environment for the actuator and the real-time monitoring module.

7. The precise coating removal system for the sprayed workpiece mounting position according to claim 6, characterized in that, In the actuator, the laser parameter range is adjusted to within a set range value; In regions with abrupt changes in curvature, the path sampling frequency is automatically increased while the scanning speed and power are simultaneously reduced. In flat regions, the distance between trajectory nodes is increased to improve the scanning rate. The laser incident angle is dynamically adjusted, and the power and frequency are matched synchronously to ensure uniformity of removal.

8. The precise coating removal system for the sprayed workpiece mounting position according to claim 6, characterized in that, The real-time monitoring module includes: calculating the error by comparing the real-time collected coating thickness with the preset value; and triggering the control system to synchronously adjust the trajectory density and laser parameters when the thickness error exceeds the set threshold range or the coating surface temperature exceeds the set temperature value.

9. The precise coating removal system for the sprayed workpiece mounting position according to claim 6, characterized in that, The auxiliary device maintains the negative pressure value within a set range, controlling the distance between the adsorption port and the workpiece surface to be less than a preset value; an optical sensor is integrated in the adsorption airflow to detect the debris concentration in real time, and automatically starts the cleaning program when the concentration exceeds the threshold.

Images