A furniture spraying processing control method, system and device

Abstract

This invention discloses a furniture spraying process control method, system, and device, belonging to the field of intelligent manufacturing technology. The method comprises four deeply coupled core steps: multi-source visual fusion modeling, adaptive trajectory dynamic planning, real-time quality monitoring and feedback, and multi-station collaborative optimization. It constructs a high-precision 3D model and extracts surface features through multi-source data fusion from a 3D laser scanner and a depth camera; generates spraying trajectories based on hierarchical surface features using an adaptive algorithm, and dynamically adjusts spraying parameters according to environmental parameters; extracts coating quality features in real time using a deep learning network, triggering closed-loop feedback control when quality deviations are detected; and establishes a multi-station global scheduling model to maximize production efficiency. This invention solves problems such as inaccurate shape recognition, lagging quality monitoring, and insufficient parameter adjustment capabilities in existing technologies, significantly improving spraying quality uniformity, production efficiency, and system intelligence.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and industrial control technology, specifically to a furniture spraying process control method, system, and device based on multi-source vision fusion and adaptive closed-loop control. Background Technology

[0002] As the furniture manufacturing industry moves towards intelligent manufacturing, spraying, as a key process, directly impacts product competitiveness in terms of quality and efficiency. However, existing furniture spraying systems suffer from numerous technical bottlenecks.

[0003] Chinese patent CN114879621B discloses a method, system, and device for controlling furniture spraying. This technology monitors target steel furniture using multiple image acquisition devices, obtains monitoring images, determines the furniture's outline information based on these images, and then judges whether repainting is necessary. While this technology achieves post-processing quality inspection, it still has the following technical shortcomings: First, it only uses two-dimensional images to recognize the workpiece outline, limiting its ability to recognize the three-dimensional shape of complex curved furniture surfaces. It cannot accurately obtain surface feature parameters, resulting in a lack of precise geometric information to support the spraying trajectory planning. Second, this technology only performs quality inspection after spraying is completed, a post-processing inspection mode that cannot achieve real-time quality monitoring and dynamic adjustment during the spraying process. Once quality problems are discovered, complete rework is required, resulting in material waste and time loss. Third, this technology lacks a closed-loop feedback mechanism between the spraying trajectory and quality inspection; the inspection results cannot influence the spraying parameters in reverse, and the system cannot adaptively adjust the spraying strategy based on quality deviations. Fourth, this technology does not consider the influence of workpiece surface curvature characteristics on the spraying process, and uses uniform spraying parameters to treat all areas, resulting in significant differences in coating thickness uniformity between planar areas and complex curved areas. Fifth, this technology lacks a correlation model between environmental parameters and spraying quality, and cannot adaptively compensate for environmental factors such as temperature and humidity changes, resulting in insufficient system robustness.

[0004] Furthermore, most existing spraying control technologies employ fixed trajectory control, making it impossible to dynamically adjust spraying parameters based on the actual shape of the workpiece. Traditional PID control algorithms have fixed parameters, making it difficult to adapt to changes in coating properties and fluctuations in environmental factors. Most systems only consider the processing control of a single workpiece, lacking multi-station collaborative optimization capabilities, thus limiting production efficiency.

[0005] Therefore, there is an urgent need to develop a new furniture spraying processing control technology that integrates three-dimensional visual perception, adaptive trajectory planning, real-time quality monitoring and closed-loop feedback control, in order to solve key technical problems such as inaccurate shape recognition, lagging quality monitoring and insufficient parameter adjustment capability in existing technologies. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a furniture spraying process control method, system and device. By constructing four deeply coupled core modules, namely multi-source vision fusion modeling, adaptive trajectory dynamic planning, real-time quality monitoring and feedback and multi-station collaborative optimization, the invention achieves intelligent control and quality assurance of the entire spraying process.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A furniture spraying process control method includes: multi-source visual fusion modeling, which uses a 3D laser scanner and a depth camera to scan the furniture workpiece to be sprayed, acquires 3D point cloud data and depth images, corrects and fuses the multi-source data based on an adaptive weight fusion algorithm, constructs a high-precision 3D model, and extracts surface curvature features; adaptive trajectory dynamic planning, which divides the workpiece surface into different feature regions according to the surface curvature features, generates a spraying trajectory for each region using a hierarchical adaptive algorithm, and establishes a dynamic compensation model based on environmental parameters and workpiece material to adjust spraying parameters in real time; real-time quality monitoring and feedback, which uses a high-speed industrial camera to acquire coating images during the spraying process, extracts coating quality features through a deep learning network, calculates quality evaluation indicators, triggers feedback control when a quality deviation is detected, transmits the quality deviation information to the trajectory planning module, automatically generates a compensation trajectory, and adjusts the spraying parameters; and multi-station collaborative optimization, which establishes a global scheduling model for a multi-station production line, comprehensively considers workpiece characteristics, equipment status, and quality requirements, and uses an optimization algorithm to solve for the optimal workpiece delivery sequence and station start-up time to maximize production efficiency.

[0009] A furniture spraying processing control system includes: a multi-source vision fusion modeling module for acquiring three-dimensional shape data and depth images of the workpiece, constructing a high-precision three-dimensional model and extracting surface curvature features through an adaptive weight fusion algorithm, including a three-dimensional laser scanner, a depth camera, a point cloud processing unit, and a feature extraction unit; an adaptive trajectory dynamic planning module for dividing the workpiece surface into regions based on its features, generating spraying trajectories for different feature regions using a hierarchical adaptive algorithm, and establishing a dynamic compensation model based on environmental parameters to adjust spraying parameters in real time, including a curvature analysis unit, a trajectory generation unit, an environmental perception unit, and a parameter adjustment unit; a real-time quality monitoring and feedback module for acquiring coating images during the spraying process, extracting quality features and calculating evaluation indicators through a deep learning network, and generating feedback signals to trigger trajectory compensation and parameter adjustment when quality deviations are detected, including a high-speed industrial camera, an image processing unit, a quality evaluation unit, and a feedback control unit; and a multi-station collaborative optimization module for establishing a global scheduling model for a multi-station production line, comprehensively considering workpiece characteristics and equipment status, and using optimization algorithms to solve for the optimal scheduling scheme, including a scheduling calculation unit, a status monitoring unit, and an instruction distribution unit.

[0010] A furniture spraying processing control device includes: a spraying workshop, which is a closed structure with a spraying operation area and an equipment area inside, equipped with a ventilation and filtration system and a temperature and humidity control system; a six-degree-of-freedom spraying robot, fixedly installed in the workshop, including the robot body, a servo drive system and an end effector; an intelligent spray gun assembly, installed at the end of the robot, including an atomizing nozzle, a flow control valve and a pressure sensor; a three-dimensional scanning device, located at the entrance of the workshop, including a three-dimensional laser scanner and a depth camera; a vision inspection device, including multiple high-speed industrial cameras distributed at different angles, equipped with an LED ring light source; a workpiece conveying device, which adopts a chain conveyor mechanism to realize automatic loading, positioning and unloading of workpieces; and a control cabinet, located outside the workshop, which houses a motion controller, an image processing unit and a communication module, and the control cabinet is connected to each device via an industrial Ethernet.

[0011] Compared with the prior art, the present invention has the following significant advantages:

[0012] First, a multi-source vision fusion scheme using a 3D laser scanner and a depth camera is adopted. The multi-source data is corrected and fused through an adaptive weight fusion algorithm. Compared with the scheme of CN114879621B which only uses 2D images to recognize contours, this scheme can accurately obtain the 3D shape and surface feature parameters of the workpiece, providing high-precision geometric information for trajectory planning and solving the problem of insufficient recognition capability of existing technologies for complex curved furniture.

[0013] Second, an adaptive trajectory dynamic planning mechanism based on surface feature hierarchy was constructed. Different spraying strategies were adopted for planar areas, curved areas and complex curved areas. The trajectory density and spraying parameters were dynamically adjusted according to the local curvature. Compared with the existing technology that uses uniform parameters, the uniformity of coating thickness was significantly improved and the defects of missed coating and overcoating at the corners were reduced.

[0014] Third, a closed-loop feedback control system of quality prediction, real-time detection, and intelligent compensation was established. During the spraying process, the coating quality characteristics are extracted in real time through a deep learning network. When a quality deviation is detected, feedback control is immediately triggered to automatically generate a compensation trajectory and adjust the spraying parameters. Compared with the post-detection mode of CN114879621B, dynamic quality control during the process is realized, which significantly reduces the rework rate and improves the first-pass yield.

[0015] Fourth, a deep coupling relationship has been established among the four core modules. The output of the multi-source vision fusion modeling module is directly used as the input of the adaptive trajectory dynamic planning module. The evaluation results of the real-time quality monitoring and feedback module have a reverse influence on trajectory planning and parameter adjustment. The global decision of the multi-station collaborative optimization module affects the working parameters of each module, forming a technical system that promotes each other and enhances synergy, achieving a synergistic effect of 1+1 greater than 2.

[0016] Fifth, a dynamic correlation model between environmental parameters and spraying quality was established. Environmental parameters such as temperature, humidity and air pressure were collected in real time. The spraying pressure and moving speed were adaptively adjusted through a dynamic compensation algorithm. Compared with the lack of environmental adaptability in existing technologies, this model enhances the robustness of the system to environmental changes and ensures the stability of processing quality under different environmental conditions.

[0017] Sixth, through multi-station collaborative scheduling optimization, a global scheduling model that considers workpiece characteristics, equipment status and quality requirements was established. Compared with single-station independent control, global optimization of the production line was achieved, production efficiency was increased by 15% to 25%, and the overall utilization rate of equipment was significantly improved. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the furniture spraying process control method of the present invention.

[0019] Figure 2 A schematic diagram of the structure of the multi-source visual fusion modeling module.

[0020] Figure 3 This is a block diagram illustrating the working principle of the real-time quality monitoring and feedback module.

[0021] Figure 4 This is a diagram illustrating the overall architecture of a furniture spraying and coating control system. Detailed Implementation

[0022] Please refer to the attached document. Figures 1-4 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, the furniture spraying process control method provided in this embodiment includes four core steps: multi-source vision fusion modeling, adaptive trajectory dynamic planning, real-time quality monitoring and feedback, and multi-station collaborative optimization. The steps are deeply coupled through data flow and control flow.

[0024] The solid wood dining tabletop to be painted is placed on the conveyor chain, and the workpiece moves with the conveyor chain to the scanning station. For example... Figure 2 As shown, the multi-source visual fusion modeling module 1 includes a 3D laser scanner, a depth camera, a point cloud processing unit, and a feature extraction unit.

[0025] The 3D laser scanner employs the principle of line laser ranging, scanning the workpiece surface at a sampling frequency of 50,000 points per second, achieving a scanning accuracy of 0.05 mm. Mounted on a gantry, the scanner can move along the X, Y, and Z directions to ensure full coverage scanning of the workpiece. During the scanning process, a point cloud dataset containing approximately 500,000 3D coordinate points is obtained. ,in This represents the number of point clouds obtained by laser scanning.

[0026] Meanwhile, the depth camera acquires depth images of the workpiece based on structured light principles. The depth images have a resolution of 1920×1200 pixels, a depth measurement range of 0.5m to 3.0m, and an accuracy of 1mm. The depth data obtained by the depth camera is then transformed into a point cloud dataset. ,in The number of point clouds generated by the depth camera.

[0027] Statistical filtering algorithms are used to reduce noise in the original point cloud data. The number of neighboring points is set. Standard deviation multiple For any point in the point cloud Calculate its to The average distance between the nearest neighbors ,like If it is an outlier, it is identified and removed. The mean distance of all points. The standard deviation of the average distance is given. The denoised point cloud data retains approximately 95% of the valid points.

[0028] Because 3D laser scanners and depth cameras operate on different measurement principles, the point cloud data they acquire differ in accuracy, density, and noise characteristics. This invention proposes an adaptive weighted fusion algorithm based on measurement uncertainty to fuse multi-source point cloud data.

[0029] For any point in space , respectively in and Search for it in the middle Calculate the local point cloud density using the nearest neighbor points. and Point cloud density is defined as... The nearest neighbor point to the point The inverse of the average distance is calculated. Simultaneously, the eigenvalues ​​of the covariance matrix of the local point cloud are calculated to assess the complexity of the local geometry.

[0030] The adaptive weight calculation formula proposed in this invention is as follows:

[0031] ,

[0032] ,

[0033] in, For laser point cloud at point The fusion weight at the point, For the fusion weights of depth camera point clouds, and These are two types of point clouds at points Local density at that location and These are the normalized values ​​of the largest eigenvalues ​​of the local covariance matrix, reflecting the complexity of the local geometric structure. and Let be the weighting coefficient, satisfying In this embodiment, The value is 0.6. The value is set to 0.4. The core idea of ​​this formula is that regions with higher point cloud density and simpler geometric structures have higher measurement accuracy and correspondingly larger fusion weights.

[0034] The coordinates of the merged point cloud are calculated as follows:

[0035] ,

[0036] in, and These are the mid-range points of the laser point cloud and the depth camera point cloud, respectively. The coordinates of the nearest point The coordinates are those of the fused point cloud. Through this adaptive weighted fusion algorithm, the fused point cloud data combines the high precision of laser scanning with the high density of depth cameras, improving the overall accuracy of the point cloud to 0.03mm.

[0037] The fused point cloud is triangulated, and a smooth triangular mesh surface is generated using the Poisson reconstruction algorithm. The triangular mesh generated in this embodiment contains approximately 100,000 triangular elements, and the mesh quality is good, with no obvious cracks or overlaps.

[0038] For each vertex in the triangular mesh, its principal curvature is calculated. A discrete differential geometry method is used to estimate the curvature based on the vertex's single-ring neighborhood. Its mean curvature and Gaussian curvature Principal curvature is calculated using the rate of change of the normal vectors of neighboring vertices. and This is obtained by solving the following equation:

[0039] ,

[0040] ,

[0041] This embodiment sets a curvature classification threshold: First threshold Second threshold Traverse all vertices and classify them according to their average curvature: If It is classified as a planar region; if It is classified as a curved surface region; if These are classified as complex curved surface areas. Statistical analysis shows that planar areas account for 75% of the total area, curved surface areas account for 20%, and complex curved surface areas are mainly distributed at the edges and corners of the workpiece and in carved decorative parts, accounting for 5%.

[0042] The adaptive trajectory dynamic planning module 2 divides the workpiece surface into regions based on its features and generates spraying trajectories using a hierarchical adaptive algorithm for different feature regions. This module receives the 3D model and curvature feature data output from the multi-source vision fusion modeling module 1, forming a deeply coupled data stream.

[0043] For planar areas, a bidirectional S-shaped scanning path is used. The effective spray width is set. Considering a 30% overlap rate, the trajectory spacing The spray gun movement speed is set to... Paint flow rate set to The spraying pressure is set to .

[0044] For curved regions, the trajectory spacing is dynamically adjusted based on the local curvature. The adaptive trajectory density adjustment formula proposed in this invention is as follows:

[0045] ,

[0046] in, For local curvature The corresponding trajectory spacing, in mm; The basic trajectory spacing is set to 105mm; This is the linear adjustment factor, with a value of 0.4. This is a non-linear adjustment factor with a value of 0.15. The local average curvature is expressed in units of . The innovation of this formula lies in the introduction of a sinusoidal nonlinear term. This allows the trajectory density to change smoothly in the curvature transition region, avoiding the trajectory abrupt change problem at the region boundary caused by traditional linear adjustment methods.

[0047] For mean curvature The adaptive trajectory spacing is calculated by substituting the curved surface region into the formula. As can be seen, the reduced trajectory spacing in areas with greater curvature increases the spraying density and ensures the uniformity of the coating thickness.

[0048] For complex curved surface areas, such as the rounded chamfers at the edges of workpieces and the carved grooves on the surface, a reinforcement learning-based trajectory optimization algorithm is used to generate the optimal spraying trajectory. First, the spraying trajectory is evenly distributed across the complex curved surface area. One quality assessment sampling point. Set the target coating thickness. .

[0049] Establish a coating thickness prediction model. The spray gun is positioned... With speed When moving, spatial point Coating thickness increment at the location This can be described using a Gaussian distribution model:

[0050] ,

[0051] in, Paint flow rate, unit: ; This represents the increment of spraying time, in units of... ; The spray diffusion radius is expressed in units of 1 / 2. The value is 50mm; The speed of the spray gun movement, in units of ; The angle between the spraying direction and the surface normal vector is given. This model comprehensively considers the effects of spraying flow rate, moving speed, diffusion characteristics, and incident angle.

[0052] A reinforcement learning optimization framework is constructed. The state space is defined as the coordinates of the sampling points on the workpiece surface and the current cumulative thickness, the action space is the position and orientation of the spray gun, and the reward function is the negative value of the coating thickness uniformity. A deep Q-network algorithm is used for training. The network contains three hidden layers, each with 128 neurons, and uses the ReLU activation function. During training, initialization... - Exploration rate of greedy strategy The value decreases to 0.1 with each training round. After 5000 rounds of iterative training, the algorithm converges, yielding the optimal trajectory scheme.

[0053] Real-time collection of environmental parameters in the spraying workshop: temperature ,humidity air pressure Establish a dynamic correlation model between coating atomization quality and environmental parameters.

[0054] The formula for calculating the atomization quality evaluation index proposed in this invention is as follows:

[0055] ,

[0056] in, This is a dimensionless index for evaluating atomization quality, with a value ranging from 0 to 1. This refers to the actual spraying pressure, expressed in MPa. For reference spraying pressure, a value of 0.25 MPa is used; Ambient temperature, in °C; The optimal temperature is set at 25℃. The temperature tolerance is set at 5℃. Ambient humidity, expressed as a percentage (%). The optimal humidity level is set at 50%. The humidity tolerance is set at 10%. , , Here are the weighting coefficients, with values ​​of 0.4, 0.3, and 0.3 respectively, satisfying the following conditions: The innovation of this formula lies in its use of an exponential function. This model describes the impact of temperature deviation on atomization quality and, compared to traditional linear models, can more accurately reflect the sensitivity of environmental parameters near their optimal values.

[0057] Substituting the current environmental parameters, the atomization quality evaluation index is calculated. Set the atomization quality threshold. Since the actual value is greater than the threshold, the current environmental conditions are deemed good, and no parameter compensation is required.

[0058] like This triggers the dynamic compensation mechanism. The pressure compensation amount is calculated as follows: ,in This is the proportional gain coefficient. The speed adjustment factor is calculated as follows: The adjusted spray gun movement speed is ,in For planning speed.

[0059] The joint configuration of the six-DOF painting robot is as follows: joint 1 is responsible for the base rotation; joints 2 and 3 control the pitch movement of the upper and lower arms; joints 4, 5, and 6 constitute the wrist mechanism, controlling the flipping, deflection, and rotation of the spray gun. An improved PID position servo controller is configured for each joint.

[0060] For joints The position control law is:

[0061] ,

[0062] in, For joints At any moment Position error, in rad; For reference angle, From a practical perspective; , , These are proportional gain, integral gain, and derivative gain, respectively. This is the feedforward compensation term, in units of ; To control torque, the unit is... .

[0063] The feedforward compensation term is calculated based on the robot's dynamics model:

[0064] ,

[0065] in, For joints The equivalent moment of inertia, in units of ; Reference angular acceleration, in units of ; These are the coefficients for the Coriolis force and centrifugal force terms; Reference angular velocity, unit: ; This is the gravity compensation term, in units of... .

[0066] Taking joint 2 as an example, set the control parameters: proportional gain. Integral gain Differential gain Through coordinated control of six joints, the robot's end effector can track the spraying trajectory with an average positional accuracy of 0.02 mm and an attitude accuracy of 0.1°, meeting the requirements for high-quality spraying.

[0067] like Figure 3 As shown, the real-time quality monitoring and feedback module 3 acquires coating images through a high-speed industrial camera during the spraying process, extracts quality features using a deep learning network, and triggers feedback control when a quality deviation is detected, transmitting the deviation information to the adaptive trajectory dynamic planning module 2 to form a closed-loop feedback loop, thereby achieving deep coupling between modules.

[0068] A high-speed industrial camera captures images of the sprayed area at 30fps. The camera has a resolution of 1920×1200 pixels, a 50mm fixed-focus lens, and a ring-shaped LED light source. The light source brightness can adaptively adjust according to the ambient light intensity to ensure uniform illumination. The camera is mounted near the robot's end effector, maintaining a fixed relative position to the spray gun, and captures real-time images of the area immediately after coating.

[0069] An improved convolutional neural network is used to extract coating quality features. The network architecture consists of 5 convolutional layers, 3 pooling layers, and 2 fully connected layers. The first convolutional layer uses 64... The convolutional kernels are strided by 2, followed by batch normalization layers and ReLU activation functions. The second to fifth convolutional layers use 128, 256, 512, and 512 kernels respectively. The convolution kernel. The pooling layer uses... Max pooling. The fully connected layer contains 1024 and 512 neurons, and the output layer is a 128-dimensional quality feature vector.

[0070] The network was trained using a dataset containing 50,000 coating images. Image annotations included quality metrics such as coating thickness uniformity, surface smoothness, and defect type. The cross-entropy loss function and Adam optimizer were employed, with an initial learning rate of 0.001 and a batch size of 32. The network converged after 100 epochs, achieving a validation set accuracy of 96.5%.

[0071] Based on the extracted 128-dimensional quality feature vector, a coating quality evaluation index is calculated. This invention proposes a comprehensive quality index. The calculation formula is:

[0072] ,

[0073] in, This is an index for coating thickness uniformity, with a value ranging from 0 to 1; This is a surface smoothness index, with a value ranging from 0 to 1; This is a defect severity index, with a value ranging from 0 to 1; , , These are weighting coefficients, with values ​​of 0.5, 0.3, and 0.2 respectively.

[0074] The coating thickness uniformity index is calculated by analyzing the grayscale distribution of the image. After converting the image to grayscale, the mean grayscale value is calculated. and grayscale standard deviation The thickness uniformity index is:

[0075] ,

[0076] in, This represents the maximum allowable coefficient of variation. The closer this value is to 1, the more uniform the thickness.

[0077] Surface smoothness is calculated using texture features. Texture features are extracted using the gray-level co-occurrence matrix method, and contrast is calculated. and homogeneity The smoothness index is:

[0078] ,

[0079] in, This is a weighting factor. The closer this value is to 1, the smoother the surface.

[0080] The defect severity index is calculated based on the defect identification results of a deep learning network. The network output includes the defect type (sagging, orange peel, pinholes, missed coating, etc.) and the defect location. Severity weights are assigned according to the defect type, and the proportion is calculated based on the defect area to obtain the overall defect severity index.

[0081] Set quality assessment thresholds Each captured image frame is evaluated for quality, and a comprehensive quality index is calculated. In this embodiment, the quality index for most areas is between 0.87 and 0.93, indicating that the coating quality is acceptable.

[0082] When detected At that time, the closed-loop feedback control mechanism is triggered. The system automatically locates the quality deviation area, extracts the deviation type and severity information, and transmits the feedback signal to the adaptive trajectory dynamic planning module 2.

[0083] A corresponding compensation strategy is generated based on the type of quality deviation. If insufficient thickness is detected, a denser touch-up spray trajectory is generated in that area, increasing the number of sprays. If sagging is detected, the paint flow rate in that area is reduced and the spray gun movement speed is increased. If orange peel texture is detected, the spray pressure and atomization parameters are adjusted.

[0084] The respray trajectory uses a spiral path, centered on the defect. With the origin as the axis, the helix radius increases linearly with the angle:

[0085] ,

[0086] in, The helix angle is expressed in rad. The initial radius is set to 5 mm. The final radius is determined based on the size of the defect area; This represents the number of spiral rotations, with a value of 3. During respraying, the spray gun flow rate is reduced to one-third of the normal flow rate. times, that is Movement speed reduced to .

[0087] After the touch-up spraying is completed, the visual inspection device photographs the area again and recalculates the quality indicators. If If the re-spraying is successful, the workpiece proceeds to the next process. If it is still unqualified, the area is recorded as a quality anomaly, triggering a second re-spraying or manual re-inspection process.

[0088] The multi-station collaborative optimization module 4 establishes a global scheduling model for the multi-station production line, receives equipment status and quality feedback information from each station, comprehensively optimizes the workpiece delivery sequence and station start-up time, and its optimization results inversely affect the working parameters of each module, forming a global collaborative control.

[0089] This embodiment implements global collaborative scheduling for an automated production line with 5 spraying stations. Production line layout: Station 1 is responsible for primer spraying, station 2 is responsible for primer drying, station 3 is responsible for topcoat spraying, station 4 is responsible for topcoat drying, and station 5 is responsible for quality inspection and packaging.

[0090] The average processing time for each workstation was obtained through historical data statistics: primer spraying time Primer drying time Topcoat spraying time Topcoat drying time Quality inspection and packaging time Transfer time between workstations This includes the time required for a workpiece to move from one workstation to the next.

[0091] The current queue of workpieces awaiting processing contains 10 workpieces, numbered as follows: The geometric complexity and surface area of ​​different workpieces vary, resulting in different actual processing times. Using the workpiece feature information obtained from the multi-source vision fusion modeling module 1, the actual processing time of each workpiece at each workstation is predicted. ,in For workstation number, Number the workpiece.

[0092] Establish a multi-workstation collaborative scheduling optimization model. Decision variables include the workpiece delivery sequence. ,in Indicates the first The workpiece number assigned to each workpiece; and the start time of each workpiece at each workstation. , indicating workpiece At workstation The time when processing begins.

[0093] The optimization objective is to minimize the total production cycle. :

[0094] ,

[0095] The constraints include:

[0096] Timing constraints: the timing relationship of the same workpiece at adjacent workstations.

[0097] ,

[0098] Resource constraints: Multiple workpieces cannot be processed simultaneously at the same workstation.

[0099] ,

[0100] Initial constraint: The start time of the first workpiece at station 1 is 0.

[0101] ,

[0102] A genetic algorithm is used to solve this optimization problem. The population size is set to 100, chromosome encoding is based on the arrangement of the workpiece sequence, the crossover probability is 0.8, the mutation probability is 0.1, and the maximum number of generations is 200. The fitness function is defined as the reciprocal of the total production cycle.

[0103] The genetic algorithm uses tournament selection as its selection operator, randomly selecting three individuals from the population each time, and choosing the individual with the highest fitness to advance to the next generation. The crossover operator uses sequential crossover, preserving some sequence information of the parent chromosomes. The mutation operator uses swap mutation, randomly selecting two positions for exchange.

[0104] After 147 generations of evolution, the algorithm converged to the optimal solution. The optimal workpiece placement sequence is: The optimal start-up time for each workpiece at each workstation is obtained through backtracking calculations to ensure that all constraints are met.

[0105] By carefully scheduling the production process, idle workstations and workpiece waiting times were avoided, maximizing equipment utilization. The optimized total production cycle was... Compared to the simple first-in-first-out (FIFO) scheduling of 1920 seconds, this method improves efficiency by 12.5% ​​and saves 240 seconds.

[0106] During production, the multi-station collaborative optimization module 4 monitors the execution status of each station in real time. When the real-time quality monitoring and feedback module 3 detects that a workpiece needs re-spraying, the actual processing time of that workpiece at the current station is extended. The system automatically updates the remaining processing time of that workpiece, recalculates the start time of subsequent workpieces, and dynamically adjusts the scheduling plan.

[0107] For example, when the workpiece During topcoat spraying at station 3, quality inspection revealed insufficient thickness in some areas, triggering a recoating process. This extended the actual processing time for the workpiece at station 3 from 200 seconds to 250 seconds. The system immediately recalculated the workpiece's processing time. The start time for workstation 3 was originally planned. The timing was delayed to 460 seconds, and the timing of subsequent tasks was adjusted accordingly to minimize the impact of scheduling disturbances.

[0108] Store the complete data of this painting operation in the knowledge base: workpiece number W3, workpiece type solid wood dining table top, dimensions Surface curvature distribution data, generated spray trajectory file, environmental parameters The control parameters used, the location of the detected defects and the results of the respray, the total spraying time was 185s, the paint consumption was 580mL, and the quality inspection score was 92 points.

[0109] like Figure 4 As shown, the furniture spraying processing control system in this embodiment adopts a modular architecture design. Each module is deeply coupled through data flow and control flow to form a closed-loop collaborative control system.

[0110] The multi-source vision fusion modeling module 1 includes a 3D laser scanner, a depth camera, a point cloud processing unit, and a feature extraction unit. The 3D laser scanner is a SICK LMS4000 series, with a scanning frequency of 600Hz and a measurement range of 0.7m to 6.0m. The depth camera is an Intel RealSense D435 series, with a depth resolution of 1280×720 and a frame rate of 90fps. The point cloud processing unit uses the PCL point cloud library to perform noise reduction, registration, and meshing functions. The feature extraction unit uses the CGAL computational geometry library to calculate curvature features.

[0111] The adaptive trajectory dynamic planning module 2 includes a curvature analysis unit, a trajectory generation unit, an environmental perception unit, and a parameter adjustment unit. The curvature analysis unit calculates the curvature and divides the input 3D model into regions. The trajectory generation unit calls appropriate trajectory planning algorithms based on region characteristics, including S-shaped scanning, adaptive density adjustment, and reinforcement learning optimization. The environmental perception unit collects environmental parameters in real time through temperature, humidity, and air pressure sensors. The parameter adjustment unit dynamically adjusts parameters such as spraying pressure, flow rate, and speed based on environmental parameters and quality feedback information.

[0112] The real-time quality monitoring and feedback module 3 includes a high-speed industrial camera, an image processing unit, a quality assessment unit, and a feedback control unit. The high-speed industrial camera uses Basler's Ace series, with a resolution of 1920×1200, a frame rate of 60fps, and supports the GigE interface. The image processing unit implements image preprocessing functions based on the OpenCV library and extracts quality features using a deep learning network based on the PyTorch framework. The quality assessment unit calculates a comprehensive quality index and determines whether to trigger feedback control. The feedback control unit generates a compensation strategy and sends control commands to the adaptive trajectory dynamic programming module 2.

[0113] The multi-station collaborative optimization module 4 includes a scheduling calculation unit, a status monitoring unit, and an instruction distribution unit. The scheduling calculation unit runs a genetic algorithm to solve for the optimal scheduling scheme. The algorithm runs on an industrial control computer, and a single optimization calculation takes approximately 2 seconds. The status monitoring unit collects real-time information on the equipment status, workpiece position, and processing progress of each station. The instruction distribution unit sends scheduling instructions to the controllers of each station via industrial Ethernet.

[0114] The system adopts a hierarchical control architecture. The upper layer is an industrial control computer running Windows, responsible for computationally intensive tasks such as trajectory planning, quality assessment, and scheduling optimization. The middle layer is a motion controller, using Advantech's EtherCAT bus controller, responsible for the robot's real-time motion control. The lower layer consists of servo drives and sensors, communicating with the motion controller via the EtherCAT bus.

[0115] The system's real-time performance is ensured through reasonable task scheduling. The industrial control computer has a task cycle of 100ms, including image acquisition, quality assessment, and trajectory update. The motion controller has a control cycle of 1ms, enabling position servo control of the robot joints. The servo driver's current loop control cycle is 125μs, ensuring rapid motor response.

[0116] The deep coupling between modules is reflected in the following aspects:

[0117] The output of the multi-source vision fusion modeling module 1 (3D model and curvature features) is directly used as the input of the adaptive trajectory dynamic planning module 2, and the spraying trajectory generated by module 2 serves as a reference for robot motion control. This forward data flow ensures that trajectory planning is based on accurate geometric information.

[0118] The output of the real-time quality monitoring and feedback module 3 (quality assessment results and deviation information) is fed back to the adaptive trajectory dynamic planning module 2, triggering trajectory compensation and parameter adjustment. This reverse feedback flow achieves closed-loop quality control, enabling the system to dynamically adjust its strategy based on the actual spraying effect.

[0119] The output (global scheduling scheme) of the multi-station collaborative optimization module 4 affects the working parameters of each module, such as the target processing time and quality requirement level of each station. This parameter flow realizes the guiding role of global optimization on local control.

[0120] Through deep coupling between modules, the system achieves a synergistic effect greater than the sum of its parts. The high-precision geometric information provided by the multi-source visual fusion modeling module 1 enables the adaptive trajectory dynamic planning module 2 to generate better trajectories, reducing quality deviations caused by blind spraying. The timely feedback from the real-time quality monitoring and feedback module 3 allows module 2 to quickly adjust its strategy, preventing the spread of large-scale quality problems. The global scheduling of the multi-station collaborative optimization module 4 ensures that all modules are coordinated in terms of time and resources, avoiding global suboptimal results caused by local optima.

[0121] The furniture spraying process control device in this embodiment adopts an integrated design, integrating scanning, spraying, detection and control functions into a compact system.

[0122] The spray painting workshop adopts a closed structure with external dimensions of [missing information]. The workshop consists of a steel frame and is enclosed by color steel panels, with an anti-corrosion coating on the inner surface. The interior of the workshop is divided into a spraying operation area and an equipment area, separated by transparent protective panels for easy observation of the spraying process.

[0123] The ventilation and filtration system includes a supply air system and an exhaust air system. The supply air system is located at the top of the work area and includes a fan, a pre-filter, and a medium-efficiency filter, with a supply air volume of 3000. The supply air temperature can be adjusted via an electric heater. The exhaust system, located at the bottom of the work area, includes an exhaust fan and a three-stage filtration system, with an exhaust volume of 3500 cubic meters per second. To ensure a slight negative pressure is maintained in the work area and prevent paint mist leakage, the three-stage filtration system includes an inertial separator, a fiber filter, and an activated carbon adsorber, achieving a paint mist removal rate of over 98%.

[0124] The temperature and humidity control system employs a precision air conditioner, with a temperature control accuracy of ±1℃ and a humidity control accuracy of ±5%. Four temperature and humidity sensors are installed in different locations within the work area to monitor environmental parameters in real time. The precision air conditioner uses PID control based on the average values ​​fed back from the sensors to maintain the set temperature and humidity conditions.

[0125] The six-DOF painting robot is the FANUC P-40iA model, with a maximum working radius of 1813mm, a payload of 40kg, and a repeatability of ±0.08mm. The robot body is made of cast iron with an anti-corrosion treatment. The servo drive system uses FANUC's αi series servo motors with rated torques of 11.8N·m, 29.4N·m, 19.6N·m, 1.96N·m, 1.96N·m, and 0.98N·m, corresponding to the six joints.

[0126] The intelligent spray gun assembly uses the W-101 automatic spray gun from Iwata Corporation of Japan, with an atomizing nozzle diameter of 1.3mm and an adjustable spray fan width ranging from 150mm to 300mm. The flow control valve is a proportional solenoid valve from FESTO, with a response time of less than 50ms, a flow rate adjustment range of 0mL / min to 600mL / min, and an adjustment accuracy of ±2%. The pressure sensor is the A-10 series from WIKA Corporation, with a measurement range of 0MPa to 1.0MPa and an accuracy of 0.25%FS. The spray gun assembly is mounted on the robot's end effector via a quick-connect coupling, with a changeover time of less than 5 minutes.

[0127] A 3D scanning device, comprising a 3D laser scanner and a depth camera, is installed at the entrance of the painting workshop. The scanning device is mounted on a liftable gantry, which consists of two columns and a crossbeam. The columns are constructed of welded square tubing, and the crossbeam is made of I-beams to ensure sufficient rigidity. The gantry has a lifting stroke of 1500mm, driven by a stepper motor and ball screw, with a positioning accuracy of 0.1mm.

[0128] The vision inspection device includes four high-speed industrial cameras, installed at the front, back, left, and right of the work area to achieve omnidirectional monitoring of the sprayed surface. Each camera is equipped with an LED ring light source, the brightness of which can be adaptively adjusted according to the ambient light intensity. The cameras are connected to an industrial control computer via a GigE network switch with a network bandwidth of 1Gbps, supporting simultaneous acquisition from all four cameras.

[0129] The workpiece conveying device adopts a chain conveyor mechanism with a double-row roller chain and a chain pitch of 50.8mm. The conveying speed is adjustable from 5m / min to 15m / min. Ten workpiece clamps are installed on the conveyor chain, with a clamp spacing of 600mm. Each clamp consists of a base, a locating pin, and a clamping cylinder, accommodating workpieces of different sizes. The clamp base is bolted to the chain pitch block. The locating pin is adjustable, and the clamping cylinder is from SMC's MY1B series, with a stroke of 50mm and adjustable clamping force.

[0130] The control cabinet is located outside the spray painting workshop, and its external dimensions are... The control cabinet features a sheet metal structure and a transparent window on the front door for easy observation of the internal equipment status. Inside, it houses an industrial control computer, motion controllers, servo drives, power modules, network switches, and terminal blocks. The industrial control computer is an Advantech IPC-610 model, configured with an Intel Core i7-9700 processor, 16GB of RAM, a 512GB solid-state drive, and an NVIDIA GeForce GTX 1660 dedicated graphics card. The motion controller is an Advantech ECAT-2610C model, supporting 8-axis EtherCAT servo control. Six FANUC αi series servo drives are used, corresponding to the robot's six joints. The power modules include 24V DC and 220V AC power supplies, providing power to the control system and actuators respectively. The network switch is a MOXA EDS-408A model, an 8-port Gigabit Ethernet switch supporting an industrial-grade wide temperature range.

[0131] The control cabinet panel features a 10.4-inch touchscreen human-machine interface, using Siemens' SIMATICHMI KTP1000 model, with a resolution of 800×600 pixels and support for True Color display. The touchscreen interface includes a main interface, parameter setting interface, operation monitoring interface, fault diagnosis interface, and data query interface. The main interface displays the system's operating status, current workpiece information, and real-time video feed. The parameter setting interface is used to set spraying parameters, environmental parameters, and scheduling parameters. The operation monitoring interface displays the robot's position, spraying progress, and quality indicators in real time. The fault diagnosis interface displays fault codes and descriptions, providing troubleshooting guidance. The data query interface allows users to view historical spraying records and statistical analysis results.

[0132] The above embodiments illustrate the technical solution and implementation effects of the present invention in detail, but the scope of protection of the present invention is not limited thereto. Without departing from the concept of the present invention, those skilled in the art can make various modifications and improvements to the technical solution, and all such modifications and improvements should fall within the scope of protection of the present invention.

Claims

1. A method for controlling furniture spraying processes, characterized in that, include: Multi-source visual fusion modeling utilizes a 3D laser scanner and a depth camera to scan the furniture workpiece to be painted, acquiring 3D point cloud data and depth images. Based on an adaptive weight fusion algorithm, the multi-source data is corrected and fused to construct a high-precision 3D model and extract surface curvature features. Adaptive trajectory dynamic planning divides the workpiece surface into planar regions, curved regions, and complex curved regions based on surface curvature characteristics. For planar regions, a grid scanning path is used. For curved regions, the trajectory spacing is dynamically adjusted according to local curvature. For complex curved regions, a reinforcement learning algorithm is used to generate the optimal trajectory. A dynamic compensation model is established based on environmental parameters to adjust the spraying pressure and moving speed in real time. Real-time quality monitoring and feedback: During the spraying process, high-speed industrial cameras are used to capture coating images, and quality features are extracted through deep learning networks to calculate comprehensive quality evaluation indicators. When the quality indicators are lower than the threshold, feedback control is triggered, and the quality deviation information is transmitted to the trajectory planning step to automatically generate a compensation trajectory and adjust the spraying parameters. Multi-station collaborative optimization: A global scheduling model for multi-station production lines is established, which comprehensively considers workpiece characteristics, equipment status, and quality requirements. A genetic algorithm is used to solve for the optimal workpiece delivery sequence and station start time.

2. The furniture spraying process control method according to claim 1, characterized in that, The adaptive weighted fusion algorithm calculates the fusion weight based on the local density and geometric complexity of multi-source point clouds. Regions with higher point cloud density and simpler geometric structures have greater fusion weights. The fused point cloud data combines the high precision of laser scanning with the high density of depth cameras.

3. The furniture spraying process control method according to claim 1, characterized in that, In the adaptive trajectory dynamic programming, a trajectory spacing calculation method including linear and nonlinear adjustment terms is adopted for the curved surface region. The nonlinear adjustment term uses a sine function to describe the curvature change, so that the trajectory density changes smoothly in the curvature transition region.

4. The furniture spraying process control method according to claim 1, characterized in that, The dynamic compensation model uses an exponential function to describe the impact of temperature and humidity deviations on atomization quality, calculates atomization quality evaluation index, and increases spraying pressure and reduces spray gun movement speed when the atomization quality evaluation index is below the threshold.

5. The furniture spraying process control method according to claim 1, characterized in that, The deep learning network includes convolutional layers, pooling layers, and fully connected layers, and outputs a quality feature vector. Based on the quality feature vector, the coating thickness uniformity index, surface smoothness index, and defect severity index are calculated, and the weighted summation yields a comprehensive quality evaluation index.

6. The furniture spraying process control method according to claim 1, characterized in that, In the real-time quality monitoring and feedback, corresponding compensation strategies are generated according to the type of quality deviation. For insufficient thickness, a spiral-shaped densified replenishment spray trajectory is generated. For sagging, the paint flow rate is reduced and the spray gun movement speed is increased.

7. The furniture spraying process control method according to claim 1, characterized in that, In the multi-station collaborative optimization, the optimization objective is to minimize the total production cycle, and the constraints include the timing constraints of the same workpiece in adjacent stations and the resource constraints that the same station cannot process multiple workpieces at the same time.

8. The furniture spraying process control method according to claim 1, characterized in that, The output information of the real-time quality monitoring and feedback step is transmitted in reverse to the adaptive trajectory dynamic planning step, triggering trajectory compensation and parameter adjustment. The scheduling scheme of the multi-station collaborative optimization step affects the working parameters of each step, forming a deeply coupled closed-loop control system.

9. A furniture spraying process control system, used to implement the furniture spraying process control method according to any one of claims 1-8, characterized in that, include: The multi-source vision fusion modeling module is used to acquire the three-dimensional shape data and depth image of the workpiece, construct a high-precision three-dimensional model and extract surface curvature features through an adaptive weight fusion algorithm, including a three-dimensional laser scanner, a depth camera, a point cloud processing unit and a feature extraction unit. The adaptive trajectory dynamic planning module is used to divide the workpiece surface into regions, generate spraying trajectories using a hierarchical adaptive algorithm for different feature regions, and adjust the spraying parameters in real time based on a dynamic compensation model established according to environmental parameters. It includes a curvature analysis unit, trajectory generation unit, environmental perception unit, and parameter adjustment unit. The real-time quality monitoring and feedback module is used to acquire coating images during the spraying process, extract quality features and calculate evaluation indicators through a deep learning network, and generate feedback signals to trigger trajectory compensation and parameter adjustment when quality deviations are detected. It includes a high-speed industrial camera, an image processing unit, a quality evaluation unit, and a feedback control unit. The multi-station collaborative optimization module is used to establish a global scheduling model for a multi-station production line. It comprehensively considers workpiece characteristics and equipment status, and uses a genetic algorithm to solve the optimal scheduling scheme. It includes a scheduling calculation unit, a status monitoring unit, and an instruction distribution unit.

10. A furniture spraying process control device, used to execute the furniture spraying process control method according to any one of claims 1-8, characterized in that, include: The spraying workshop is a closed structure, with a spraying operation area and an equipment area inside, equipped with a ventilation and filtration system and a temperature and humidity control system; A six-degree-of-freedom painting robot is fixedly installed in the workshop and includes the robot body, servo drive system and end effector; an intelligent spray gun assembly is installed at the end of the robot and includes atomizing nozzle, flow control valve and pressure sensor; a three-dimensional scanning device is set at the entrance of the workshop and includes a three-dimensional laser scanner and depth camera. The visual inspection device includes multiple high-speed industrial cameras distributed at different angles and equipped with an LED ring light source; The workpiece conveying device adopts a chain conveyor mechanism to realize automatic workpiece loading, positioning and unloading; the control cabinet is located outside the work area and has a built-in motion controller, image processing unit and communication module. The control cabinet is connected to each device through industrial Ethernet.

Images