A robotically closed loop controlled sizing method

Abstract

The present application relates to the technical field of intelligent control, and particularly relates to a robot closed-loop control sizing method, which constructs a robot sizing integration scheme of "multi-source data acquisition-intelligent double-branch identification-real-time deviation quantification-dynamic closed-loop regulation and control", cooperatively collects the industrial camera and the structured light scanner through the pressure sensor trigger, obtains the two-dimensional image information of the workpiece and captures the three-dimensional point cloud data, and provides comprehensive basis for sizing path planning; synchronously collects the glue bead data in the sizing process, real-time calculates three types of deviations of width, height and continuity, and ensures the dynamic perception of the sizing quality; generates adjustment control quantity based on the deviation, realizes real-time optimization of the robot control parameter, effectively compensates the glue path deviation and uneven glue amount problems caused by the flexible workpiece deformation and positioning deviation, significantly improves the sizing precision and consistency, guarantees the workpiece bonding strength, reduces the production scrap rate, and is suitable for the large-scale production scene of flexible workpieces such as shoemaking.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a robot closed-loop control method for applying adhesive. Background Technology

[0002] In the production of flexible workpieces such as footwear, robotic gluing technology is a core element in ensuring product bonding strength and assembly quality.

[0003] Current mainstream gluing solutions have significant technical shortcomings: the gluing trajectory relies entirely on pre-set CAD pattern data planning, without fully considering the elastic deformation and positioning deviations that may easily occur in flexible workpieces such as shoe uppers and soles during handling, positioning, and production. This leads to deviations between the actual gluing path and the pre-set trajectory, resulting in quality hazards such as uneven glue distribution and misaligned glue paths, which directly affect the bonding strength of the workpiece and the product qualification rate.

[0004] In addition, although some improvement solutions introduce image recognition technology to assist in positioning, they are limited to a single static calibration before gluing. They cannot monitor and provide feedback on the morphological parameters and continuity of the glue beads in real time during the gluing process, making it difficult to dynamically correct gluing deviations and fundamentally solve the industry pain point of insufficient gluing accuracy for flexible workpieces.

[0005] Therefore, there is an urgent need for a sizing technology solution that combines dynamic adaptability and real-time control capabilities, enabling more accurate and effective sizing. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a robot closed-loop control method for applying glue, which can perform more accurate and effective glue application.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A robot-controlled closed-loop glue application method includes the following steps: S1. Triggered by a pressure sensor at a predetermined position on the tooling table, workpiece images and workpiece point cloud data are acquired from the tooling table using an industrial camera and a structured light scanner, respectively. S2. Input the grayscale image of the workpiece and the point cloud data of the workpiece into the pre-trained glue application area recognition model, and output the glue application path data in the form of discrete coordinate point set by the glue application area recognition model; S3. Control the robot to apply glue according to the glue application path data, and collect glue bead images and glue bead point cloud data through the industrial camera and the structured light scanner. S4. Based on the glue bead image and the glue bead point cloud data, obtain the current glue bead width, glue bead height and the gap distance between adjacent glue beads in real time, and calculate the glue bead width deviation, glue bead height deviation and glue bead continuity deviation in combination with the preset standard width and standard height. S5. Based on the deviation of the glue bead width, the deviation of the glue bead height, and the deviation of the glue bead continuity, an adjustment control quantity is generated to adjust the control parameters of the robot.

[0008] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A storage medium storing a computer program thereon, which, when executed, implements the steps of the above-described robot closed-loop control glue application method.

[0009] The beneficial effects of this invention are as follows: This invention provides a robot closed-loop control method for applying adhesive, constructing an integrated robot adhesive application solution of "multi-source data acquisition - intelligent dual-branch recognition - real-time deviation quantification - dynamic closed-loop control." Through pressure sensors triggering the collaborative acquisition of industrial cameras and structured light scanners, it obtains both two-dimensional image information of the workpiece and captures three-dimensional point cloud data, providing a comprehensive basis for adhesive path planning. During the adhesive application process, adhesive bead data is simultaneously acquired, and three types of deviations—width, height, and continuity—are calculated in real time to ensure dynamic perception of adhesive application quality. Based on the deviation, control quantities are generated and adjusted to achieve real-time optimization of robot control parameters, effectively compensating for adhesive path offset and uneven adhesive amount caused by deformation and positioning deviations of flexible workpieces. This significantly improves adhesive application accuracy and consistency, ensures workpiece bonding strength, reduces production scrap rates, and is suitable for large-scale production scenarios of flexible workpieces such as footwear. Attached Figure Description

[0010] Figure 1 This is a flowchart of a robot closed-loop control glue application method according to an embodiment of the present invention. Detailed Implementation

[0011] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0012] The above-described robot closed-loop control method for applying glue according to the present invention is applicable to glue application control in the field of flexible workpiece production, and is particularly suitable for glue application control in footwear production. The following is a detailed description of specific embodiments: Please refer to Figure 1 One embodiment of the present invention is as follows: A robot-controlled closed-loop glue application method includes the following steps: S1. Triggered by a pressure sensor at a predetermined position on the tooling table, workpiece images and workpiece point cloud data are acquired from the tooling table using an industrial camera and a structured light scanner, respectively.

[0013] In this embodiment, a 2D industrial camera and a 3D structured light scanner are integrated into the robot's end effector for gluing application, with the camera's field of view covering the gluing area. A calibration plate is set on the worktable for calibrating the camera's intrinsic and extrinsic parameters and the robot's hand-eye calibration. The robot drives the vision module to perform 3D structured light scanning on the shoe upper / sole placed on the tooling table to acquire workpiece point cloud data; simultaneously, workpiece images (2D grayscale images) are acquired for edge contour recognition of the gluing area.

[0014] Noise is removed by Gaussian filtering, and the glue beads are separated from the workpiece background by a threshold segmentation algorithm. The 3D point cloud data is denoised and registered to generate a 3D model of the workpiece, which is then compared with a preset CAD template model.

[0015] S2. Input the grayscale image of the workpiece and the point cloud data of the workpiece into the pre-trained glue application area recognition model, and output the glue application path data in the form of discrete coordinate point set by the glue application area recognition model; The adhesive application area recognition model includes a 2D recognition branch, a 3D recognition branch, and an attention fusion layer; The 2D recognition branch extracts features from the workpiece image and generates a contour feature vector. The 2D recognition branch is based on an improved YOLOv8 model, which includes an input layer, a backbone network, a neck structure, a detection head, and an output layer. In this embodiment, the standard convolutional layer in the C2f module of the backbone network is replaced with a depthwise separable convolutional layer, and the fourth C2f module of the backbone network is removed. The backbone network uses ReLU6 as the activation function. The neck structure removes two convolutional layers, retains the lightweight PAN-FPN fusion structure, and uses nearest neighbor difference instead of transposed convolution as the upsampling method.

[0016] In this embodiment, the core objective of the 2D recognition branch is to achieve real-time recognition of the edge contour of the adhesive application area with low computational power consumption. Unlike the traditional YOLOv8's full-featured "classification + detection" design, it only retains contour detection and feature extraction capabilities. The specific hierarchical structure is shown in the table below: hierarchy Module Name Core parameters Function Description Input layer Image input module Input dimensions: 640×640×3. Data augmentation: Mosaic (small scale) + brightness adjustment. Receives shoe upper / sole images from 2D industrial cameras, standardizes dimensions, and enhances robustness. Limits the stitching scale of Mosaic enhancements (≤2x) to avoid distortion of flexible workpiece contours. backbone network C2f lightweight module + depthwise separable convolution Convolution kernel: 3×3 (depth separable) Stride: 2 Activation function: ReLU6 Number of channels: [32, 64, 128, 256] Reduce the number of Bottlenecks in the C2f module (from 6 to 3) to balance accuracy and speed. Extract multi-scale texture and edge features from the image and output feature maps at three scales (80×80, 40×40, 20×20). Neck structure SPPF+ Simplified PAN SPPF pooling kernels: 5, 9, 13 PAN: Remove the two 1×1 convolutional layers after upsampling By fusing multi-scale feature maps, the expressive power of contour features is enhanced. Detection head Lightweight profile detection head Regression branch: Predicts bounding box coordinates. Output branch: 1×1 convolution. 1. Remove the classification branch (no need to identify shoe type), only retain the regression branch. 2. Add a feature reduction layer: compress the feature map from 256×20×20 to a 64×1 one-dimensional vector through a 1×1 convolution. 1. Output the bounding box (x, y, w, h) of the adhesive application area; 2. Extract the contour feature map. Output layer Contour feature vector output Output dimension: 64×1. <![CDATA[Output the contour feature vector of the sizing area F 2, for subsequent attention fusion. The feature vector contains key information such as the edge gradient, position offset, and texture complexity of the contour]]>

[0017] The total number of parameters has been reduced from 11.2M in the traditional YOLOv8 to 2.3M, and the inference time is ≤1ms / frame, making it compatible with robot embedded controllers.

[0018] The 3D recognition branch extracts features from the workpiece point cloud data and generates a deformation feature vector. The 3D recognition branch is based on an improved PointNet model, including an input layer, an input transformation layer, a feature extraction layer, a feature transformation layer, a deformable feature extraction layer, a global feature pooling layer, an encoding layer, and an output layer. The deformation feature extraction layer introduces a multi-scale convolution kernel to extract point cloud change features within 10, 20, and 50 neighborhoods. It also adds a deformation factor calculation submodule to compare the coordinate deviation between the actual point cloud and the CAD point cloud and output three types of deformation parameters: stretching deformation, bending deformation, and local warping.

[0019] In this embodiment, the core objective of the 3D recognition branch is to accurately extract the deformation parameters (stretching, bending, warping) of flexible footwear workpieces. Unlike the traditional PointNet's "general point cloud classification / segmentation" design, a new layer for extracting the deformation features of flexible workpieces is added. The specific hierarchical structure is as follows: hierarchy Module Name Core parameters Function Description Input layer Point cloud input module Input dimensions: 1024×3 Receive preprocessed point cloud data. Input Transformation Layer (T-Net) Convolution + Fully Connected Convolutional kernel: 1×1, number of channels = 64; Fully connected layer: 64→128→1024; Activation function: ReLU Learn the spatial transformation matrix of point clouds and align point clouds with different poses; add a regularization term to limit the rotation angle of the transformation matrix (≤5°) to avoid over-transformation. Feature extraction layer Point cloud feature convolution Convolution kernel: 1×1, number of channels = 64 → 128 Extract local features of the point cloud (such as point density and neighborhood distribution). Feature Transformation Layer (T-Net) Convolution + Fully Connected Convolutional kernel: 1×1, number of channels = 128; Fully connected layer: 128 → 256 → 1024 Learn the transformation matrix of the feature space to improve the robustness of feature representation. Deformation Feature Extraction Layer Flexible workpiece deformation feature extraction layer Multi-scale convolution kernel: 1×1 (3 scales) deformation factor calculation module. 1. Introducing multi-scale convolution kernel: extracting point cloud variation features within 10, 20, and 50 neighborhoods respectively; 2. Adding a deformation factor calculation submodule: comparing the coordinate deviation between the actual point cloud and the CAD point cloud, and outputting deformation parameters; 3. Outputting a 128-dimensional deformation feature vector, containing quantized values ​​of 3 types of deformation parameters. Extract the tensile deformation, bending deformation, and local warpage of the workpiece. Global feature pooling layer Max pooling Pooling dimension: 1024×128 → 1×128 Aggregate local features into global deformable features coding layer Fully connected layer Input dimension = 128, output dimension = 64. Weight initialization: Xavier <![CDATA[Map the 128-dimensional deformed feature vector to a 64-dimensional vector F 3, align with the 2D branch feature dimension]]> Output layer Deformation feature vector output Output dimension: 64×1 <![CDATA[Output the deformation feature vector of the flexible workpiece F 3, for attention fusion. The feature vector contains key information such as the maximum deformation amount, the position of the deformation area, and the deformation type]]>

[0020] The attention fusion layer performs attention fusion on the contour feature vector and the deformation to obtain fused features, and generates the adhesive application path data based on the fused features; The attention fusion layer includes a feature input layer, an attention weight calculation layer, a dynamic feature fusion layer, and a feature fusion layer; The attention weight calculation layer introduces a deformation factor, and the contour feature weight and deformation feature weight are calculated by scaling the dot product attention based on the deformation factor.

[0021] In this embodiment, the deformation factor λ is the deformation feature vector output by the 3D PointNet branch. F The maximum deformation calculated in step 3 reflects the degree of deformation of the workpiece. The formula is: ; in, λ Indicates the deformation factor. △L max The actual maximum deformation of the workpiece is indicated by comparing the workpiece point cloud data with the CAD model. △L threshold This indicates the preset deformation threshold.

[0022] The improved scaling dot product attention formula is: ; ; ; ; in, W Q , W K , W V This represents the learnable weight matrix, all with dimensions 64×64. d k This represents the dimension of the key vector, with a value of 64. M denoted as the deformation priority matrix, which is a diagonal matrix with diagonal elements having a value of 1 (indicating that the deformation feature has a higher priority in high deformation scenarios). (1-λ) represents the weight coefficient of the 2D contour feature. The larger λ is, the lower the weight of the contour feature. λ·M represents the weight adjustment term of the deformation feature. The larger λ is, the higher the weight of the deformation feature.

[0023] The calculation process implicitly incorporates the differences in modal importance under different deformation states: When the workpiece deformation is small (λ→0): (1−λ)→1 in the formula. Attention calculation is mainly based on the semantic similarity between contour features and deformation features. If the matching degree between contour features and deformation features is high, it indicates that the reference value of contour features is high. When the workpiece deformation is large (λ→1): the λ·M term in the formula dominates the calculation, forcibly increasing the weight of the deformation feature, because the deformation feature is more critical for path compensation when the deformation is severe.

[0024] The above calculation logic is quantified and refined (simplified and put into practical use) to obtain the contour feature weight and deformation feature weight.

[0025] The calculation of the contour feature weights and the deformation feature weights is expressed as follows: ; ; in, α Indicates the contour feature weights, β Represents the deformation feature weights. F 2 represents the contour feature vector. F 3 represents the deformable feature vector. λ This represents the deformation factor.

[0026] norm ratio correspond The feature semantic matching degree of the reaction—the larger the feature norm, the higher the feature information entropy of the mode and the greater its contribution to the localization of the application path.

[0027] The dynamic feature fusion layer performs weighted fusion of the contour feature vector and the deformation feature vector according to the contour feature weight and the deformation feature weight to generate a fused feature vector.

[0028] F fusion = α×F 2 +β×F 3; It is an engineering-simplified version of fine-grained attention computation, reducing computational complexity from O(d) while maintaining accuracy. k 2 ) decreased to O(d k This better suits the real-time requirements of embedded robot platforms. Fine-grained attention calculation is represented as: F fusion ≈ Attention ( Q , K , V ).

[0029] The feature decoding layer includes a first fully connected layer and a second fully connected layer, through which the fused feature vector is decoded into the sizing path data.

[0030] In this embodiment, the feature decoding layer uses a two-layer fully connected network to fuse the 64-dimensional feature vector. F fusion Decode the application path coordinates in the robot coordinate system. The specific steps are as follows: First fully connected layer: input dimension 64, output dimension 32, activation function is ReLU; The second fully connected layer has an input dimension of 32 and an output dimension of 3, corresponding to the X, Y, and Z axes of the robot coordinate system. Output: A set of discrete coordinate points along the adhesive application path {(x1, y1, z1), (x2, y2, z2), ..., (x...} n y n , z n )}.

[0031] S3. Control the robot to apply glue according to the glue application path data, and collect glue bead images and glue bead point cloud data through the industrial camera and the structured light scanner.

[0032] S4. Based on the glue bead image and the glue bead point cloud data, obtain the current glue bead width, glue bead height and the gap distance between adjacent glue beads in real time, and calculate the glue bead width deviation, glue bead height deviation and glue bead continuity deviation in combination with the preset standard width and standard height.

[0033] In this embodiment, the width of the glue bead is monitored based on the horizontal pixel span (calibrated as physical width) of the glue bead's connected domain. w ( tThe height of the glue beads was monitored by the maximum difference along the Z-axis based on the point cloud data of the glue beads. h ( t The spacing between bead segments is measured based on the minimum edge distance of the connected domains of adjacent glue beads. d i ( t ); Check whether the spacing between the glue beads is greater than the preset standard distance between the glue beads. d threshold Determine the continuity label of the glue beads (Boolean type: 0 if it is greater than the standard distance, otherwise 1).

[0034] The calculation of the bead width deviation, the bead height deviation, and the bead continuity deviation is expressed as follows: e w ( t )= w ( t )- w std ; e h ( t )= h ( t )- h std ; e d ( t )= max ( d i ( t )- d threshold ,0); in, e w ( t The ) indicates the deviation in the width of the glue bead. w ( t () indicates the width of the glue bead. w std Indicates the standard width of the glue bead. e h ( t This indicates the deviation in the height of the glue beads. h ( t The number ) indicates the height of the glue bead. h std Indicates the standard height of the glue beads. e d ( t The ) indicates a deviation in the continuity of the glue beads. d i ( t ) represents the distance between the i-th glue bead and the previous glue bead.d threshold Indicates standard distance.

[0035] S5. Based on the deviation of the glue bead width, the deviation of the glue bead height, and the deviation of the glue bead continuity, an adjustment control quantity is generated to adjust the control parameters of the robot.

[0036] In this embodiment, the three types of deviation values ​​are weighted and fused to obtain the comprehensive input deviation of the PID controller.

[0037] Based on the bead width deviation, the bead height deviation, and the bead continuity deviation, an adjustment control value is generated, including the following steps: The comprehensive input deviation of the PID controller is obtained by weighting and fusing the deviations in width, height, and continuity of the adhesive beads. e total ( t ): e total ( t )= k d × e d ( t )+ k w × e w ( t )+ k h × e h ( t ); in, k d , k w as well as k h These represent the weighting coefficients for the continuity deviation, width deviation, and height deviation of the adhesive beads, respectively. k d > k w > k h , k d + k w + k h =1. For example, k d =0.6, k w =0.3, kh =0.1, ensuring that the continuity deviation takes precedence over the control output.

[0038] The adjustment amount of the PID controller is calculated based on the comprehensive input deviation: ; ; Among them, △ Q PID Indicates the increase in glue dispensing volume from the glue gun, △ v PID Indicates the robot's speed increment. K p This indicates the proportional coefficient for adjusting the amount of adhesive. K i Represents the integral coefficient. K d Represents the differential coefficient. The integral term representing the overall deviation, The differential term representing the overall deviation, - K p This represents the proportional coefficient for speed adjustment.

[0039] Control logic: When e d (t)>0 (small breakpoint exists / spacing exceeds standard): PID output increases glue dispensing and reduces robot movement speed to quickly fill the gap; When e d (t)=0: The PID only fine-tunes the parameters based on the width / height deviation to maintain the stability of the bead shape.

[0040] Embodiment 2 of the present invention is as follows: A storage medium storing a computer program thereon, which, when executed, implements the steps in the robot closed-loop control glue application method described in Embodiment 1 above.

[0041] This invention presents a robot-based closed-loop control method for glue application, constructing an integrated robot glue application solution encompassing "multi-source data acquisition, intelligent dual-branch recognition, real-time deviation quantification, and dynamic closed-loop control." This comprehensively addresses the industry pain points of traditional glue application technologies, such as reliance on preset CAD data, inability to adapt to flexible workpiece deformation, and lack of real-time control capabilities, demonstrating significant technical advantages and application value. This invention utilizes the collaborative acquisition of industrial cameras and structured light scanners to simultaneously acquire 2D images and 3D point cloud data of the workpiece. Combining an improved lightweight 2D branch of YOLOv8 with an optimized 3D branch of PointNet, it accurately extracts contour features and multi-scale deformation features. Dynamic weighted fusion is then achieved through an attention fusion layer incorporating deformation factors, ensuring that the glue application path planning both conforms to the actual contour of the workpiece and adapts to elastic deformation and positioning deviations. During the glue application process, glue bead data is acquired in real-time. Width, height, and continuity deviations are accurately calculated using quantification formulas. Based on weighted fusion PID control logic, the glue application amount and robot speed are dynamically adjusted, forming a closed-loop feedback throughout the entire process. This effectively avoids glue path deviation, uneven glue application, and glue breakage issues. Meanwhile, the lightweight improvements to the model reduce the hardware computing power requirements, and the adaptive design of the storage medium facilitates industrial deployment. The overall solution significantly improves the accuracy and consistency of adhesive application on flexible workpieces, significantly increases product bonding strength and pass rate, reduces scrap rate and production costs, and is suitable for large-scale production scenarios of flexible workpieces such as footwear, combining technological innovation with practical application feasibility.

[0042] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A robot closed-loop control method for applying adhesive, characterized in that, Including the following steps: S1. Triggered by a pressure sensor at a predetermined position on the tooling table, workpiece images and workpiece point cloud data are acquired from the tooling table using an industrial camera and a structured light scanner, respectively. S2. Input the grayscale image of the workpiece and the point cloud data of the workpiece into the pre-trained glue application area recognition model, and output the glue application path data in the form of discrete coordinate point set by the glue application area recognition model; S3. Control the robot to apply glue according to the glue application path data, and collect glue bead images and glue bead point cloud data through the industrial camera and the structured light scanner. S4. Based on the glue bead image and the glue bead point cloud data, obtain the current glue bead width, glue bead height and the gap distance between adjacent glue beads in real time, and calculate the glue bead width deviation, glue bead height deviation and glue bead continuity deviation in combination with the preset standard width and standard height. S5. Based on the deviation of the glue bead width, the deviation of the glue bead height, and the deviation of the glue bead continuity, an adjustment control quantity is generated to adjust the control parameters of the robot.

2. The robot closed-loop control glue application method according to claim 1, characterized in that, The adhesive application area recognition model includes a 2D recognition branch, a 3D recognition branch, and an attention fusion layer; The 2D recognition branch extracts features from the workpiece image and generates a contour feature vector. The 3D recognition branch extracts features from the workpiece point cloud data and generates a deformation feature vector. The attention fusion layer performs attention fusion on the contour feature vector and the deformation to obtain fused features, and generates the adhesive application path data based on the fused features.

3. The robot closed-loop control glue application method according to claim 2, characterized in that, The 2D recognition branch is based on an improved YOLOv8 model, which includes an input layer, a backbone network, a neck structure, a detection head, and an output layer. In this embodiment, the standard convolutional layer in the C2f module of the backbone network is replaced with a depthwise separable convolutional layer, and the fourth C2f module of the backbone network is removed. The backbone network uses ReLU6 as the activation function. The neck structure removes two convolutional layers, retains the lightweight PAN-FPN fusion structure, and uses nearest neighbor difference instead of transposed convolution as the upsampling method.

4. The robot closed-loop control glue application method according to claim 2, characterized in that, The 3D recognition branch is based on an improved PointNet model, including an input layer, an input transformation layer, a feature extraction layer, a feature transformation layer, a deformable feature extraction layer, a global feature pooling layer, an encoding layer, and an output layer. The deformation feature extraction layer introduces a multi-scale convolution kernel to extract point cloud change features within 10, 20, and 50 neighborhoods. It also adds a deformation factor calculation submodule to compare the coordinate deviation between the actual point cloud and the CAD point cloud and output three types of deformation parameters: stretching deformation, bending deformation, and local warping.

5. The robot closed-loop control glue application method according to claim 2, characterized in that, The attention fusion layer includes a feature input layer, an attention weight calculation layer, a dynamic feature fusion layer, and a feature fusion layer; The attention weight calculation layer introduces a deformation factor, and the contour feature weight and deformation feature weight are calculated by scaling the dot product attention in combination with the deformation factor. The dynamic feature fusion layer performs weighted fusion of the contour feature vector and the deformation feature vector according to the contour feature weight and the deformation feature weight to generate a fused feature vector; The feature decoding layer includes a first fully connected layer and a second fully connected layer, through which the fused feature vector is decoded into the sizing path data.

6. The robot closed-loop control glue application method according to claim 5, characterized in that, The deformation factor is expressed as: λ = △L max / △L threshold ； in, λ Indicates the deformation factor. △L max The actual maximum deformation of the workpiece is indicated by comparing the workpiece point cloud data with the CAD model. △L threshold This indicates the preset deformation threshold.

7. The robot closed-loop control glue application method according to claim 5, characterized in that, The calculation of the contour feature weights and the deformation feature weights is expressed as follows: ; ; in, α Indicates the contour feature weights, β Represents the deformation feature weights. F 2 represents the contour feature vector. F 3 represents the deformable feature vector. λ This represents the deformation factor.

8. The robot closed-loop control glue application method according to claim 1, characterized in that, The calculation of the bead width deviation, the bead height deviation, and the bead continuity deviation is expressed as follows: e w ( t )= w ( t )- w std ； e h ( t )= h ( t )- h std ； e d ( t )= max ( d i ( t )- d threshold ，0)； in, e w ( t The ) indicates the deviation in the width of the glue bead. w ( t () indicates the width of the glue bead. w std Indicates the standard width of the glue bead. e h ( t The ) indicates the deviation in the height of the glue beads. h ( t () indicates the height of the glue bead. h std Indicates the standard height of the glue beads. e d ( t The ) indicates a deviation in the continuity of the glue beads. d i ( t () represents the distance between the i-th glue bead and the previous glue bead. d threshold Indicates standard distance.

9. The robot closed-loop control glue application method according to claim 1, characterized in that, Based on the bead width deviation, the bead height deviation, and the bead continuity deviation, an adjustment control value is generated, including the following steps: The comprehensive input deviation of the PID controller is obtained by weighting and fusing the deviations in bead width, bead height, and bead continuity. e total ( t ): e total ( t )= k d × e d ( t )+ k w × e w ( t )+ k h × e h ( t )； in, k d , k w as well as k h These represent the weighting coefficients for the continuity deviation, width deviation, and height deviation of the adhesive beads, respectively. k d > k w > k h , k d + k w + k h =1; The adjustment amount of the PID controller is calculated based on the comprehensive input deviation: ; ; Among them, △ Q PID Indicates the increase in glue dispensing volume from the glue gun, △ v PID Indicates the robot's speed increment. K p This indicates the proportional coefficient for adjusting the amount of adhesive. K i Represents the integral coefficient. K d Denotes the differential coefficient. The integral term representing the overall deviation, The differential term representing the overall deviation, - K p This represents the proportional coefficient for speed adjustment.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the steps in the robot closed-loop control glue application method according to any one of claims 1-9.

Images