Intelligent spraying method, device and equipment and storage medium

Abstract

The invention discloses an intelligent spraying method, device and equipment and a storage medium, and relates to the field of robot control, and the intelligent spraying method comprises the steps that a three-dimensional CAD model of a to-be-sprayed workpiece is acquired, and target geometric features of the to-be-sprayed workpiece are determined based on the three-dimensional CAD model; the target geometric features comprise a surface curvature distribution feature, an edge contour feature and a potential shadow region; a target parameter sequence is obtained, and a predicted coating thickness distribution diagram corresponding to the to-be-sprayed workpiece is obtained based on the target parameter sequence and the target geometric features through the target prediction model; the target parameter sequence is a sequence formed by real-time spraying process parameters; and target thickness deviation is determined based on the predicted coating thickness distribution diagram and the target coating thickness, a target spraying strategy is generated based on the target thickness deviation, and the target robot is controlled to spray the to-be-sprayed workpiece based on the target spraying strategy. According to the method, lasting, stable and high-precision control over the spraying quality of the complex curved surface is achieved.

Description

Technical Field

[0001] This invention relates to the field of robot control, and in particular to an intelligent spraying method, apparatus, equipment, and storage medium. Background Technology

[0002] In the automated spraying of complex curved workpieces in aerospace, automotive, and high-end equipment industries, ensuring coating thickness uniformity is a major challenge. Workpiece geometry (such as sharp edges, deep concave areas, and high-curvature surfaces) can trigger "edge effects" and "shadow effects," leading to systematic and predictable thickness deviations.

[0003] In existing technologies, there are some methods for parameter compensation based on workpiece geometry. These methods improve uniformity to some extent by identifying high- and low-risk areas and applying predefined compensation strategies. However, such static compensation systems have inherent drawbacks: firstly, their core compensation rules (such as the adjustment of speed and flow rate) rely on manual experience or offline experiments, making it difficult to accurately match dynamically changing actual working conditions; secondly, once deployed, the system's performance is fixed and cannot cope with slow time-varying interferences such as spray gun wear and coating characteristic fluctuations, leading to a gradual decline in compensation effectiveness and requiring frequent manual recalibration.

[0004] Therefore, there is an urgent need for an intelligent spraying system that can evolve autonomously and become more accurate with use, in order to overcome the limitations of static compensation systems and achieve sustained, stable and high-precision control of the spraying quality of complex curved surfaces. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide an intelligent spraying method, apparatus, device, and storage medium, capable of achieving persistent, stable, and high-precision control over the spraying quality of complex curved surfaces. The specific solution is as follows: In a first aspect, this application discloses an intelligent spraying method, comprising: A three-dimensional CAD model of the workpiece to be coated is obtained, and the target geometric features of the workpiece to be coated are determined based on the three-dimensional CAD model; the target geometric features include surface curvature distribution features, edge contour features, and potential shadow areas; A target parameter sequence is obtained, and a target prediction model is used to obtain a predicted coating thickness distribution map corresponding to the workpiece to be sprayed based on the target parameter sequence and the target geometric features; the target parameter sequence is a sequence composed of real-time spraying process parameters; the spraying process parameters include robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from spray gun to workpiece surface; The target thickness deviation is determined based on the predicted coating thickness distribution map and the target coating thickness, and a target spraying strategy is generated based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

[0006] Optionally, determining the target geometric features of the workpiece to be coated based on the three-dimensional CAD model includes: Based on the three-dimensional CAD model, the curvature information of each point on the surface of the workpiece to be sprayed is determined, and based on the curvature information, the surface curvature distribution characteristics of the workpiece to be sprayed are determined. The edge contour features of the workpiece to be coated are determined based on the three-dimensional CAD model using a target edge detection algorithm; the edge contour features include physical edges, ridges, and corresponding edge types. Based on the spray cone angle model of the target spray gun, the relative pose of the spray gun and the workpiece surface, and the three-dimensional CAD model, the potential shadow area of ​​the workpiece to be sprayed is determined; the potential shadow area is the area that cannot directly receive paint because it is blocked by the workpiece itself under the preset spray gun pose.

[0007] Optionally, the step of obtaining the predicted coating thickness distribution map corresponding to the workpiece to be coated based on the target parameter sequence and the target geometric features using the target prediction model includes: The target parameter sequence and the target geometric features are input into the target prediction model; The first model branch of the target prediction model is used to fuse the geometric features of each target to obtain a target feature map; the first model branch is a model branch built based on a convolutional neural network. The target dependency is obtained based on the target parameter sequence using the second model branch of the target prediction model; the second model branch is a model branch constructed based on a long short-term memory network; the target dependency is the temporal dependency of the spraying process parameters on the scanning path; The target feature map and the target dependency are fused to obtain fused feature values, and the predicted coating thickness distribution map corresponding to the workpiece to be sprayed is obtained based on the fused feature values ​​using the fully connected layer of the target prediction model.

[0008] Optionally, the step of generating the target coating strategy based on the target thickness deviation includes: Determine whether the target thickness deviation is less than the target deviation threshold; If the target thickness deviation is less than the target deviation threshold, then the preset basic spraying strategy is determined as the target spraying strategy; If the target thickness deviation is greater than the target deviation threshold, the trajectory normal of the basic spraying strategy is corrected and the spraying process parameters are adjusted based on the preset safety constraints, and the adjusted basic spraying strategy is determined as the target spraying strategy.

[0009] Optionally, the step of generating a target spraying strategy based on the target thickness deviation, and controlling the target robot to spray the workpiece to be sprayed based on the target spraying strategy, includes: The geometric complexity of the workpiece to be coated is determined based on the target geometric features. The target spraying responsibility area is determined based on the geometric complexity, the target thickness deviation, and the workspace range corresponding to each target robot. Based on the geometric complexity and the target thickness deviation, a target spraying strategy is generated corresponding to the target spraying responsibility area; Based on the workspace range corresponding to each target robot, the target robot is assigned to the target spraying responsibility area, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

[0010] Optionally, controlling the target robot to spray the workpiece based on the target spraying strategy includes: An overlapping spraying area is set at the boundary of adjacent target spraying responsibility areas; When at least one of the target robots corresponding to the overlapping spraying area sprays the overlapping spraying area based on the target spraying strategy, the data transmission speed of all the target robots corresponding to the overlapping spraying area is synchronized, and the spraying flow command corresponding to each target robot is weighted and fused based on position, so as to spray the workpiece to be sprayed based on the fused spraying flow command.

[0011] Optionally, after controlling the target robot to spray the workpiece based on the target spraying strategy, the method further includes: Obtain the actual coating thickness data corresponding to the target spraying strategy; A target training dataset is constructed based on the actual coating thickness data and the target parameter sequence when spraying the workpiece to be coated. The target prediction model is updated based on the target training dataset using a preset incremental learning mechanism to obtain the updated target prediction model. When performing a spraying task on a new workpiece to be sprayed, the updated target prediction model is used to obtain the predicted coating thickness distribution map corresponding to the new workpiece to be sprayed.

[0012] Secondly, this application discloses an intelligent spraying device, comprising: The geometric feature determination module is used to acquire a three-dimensional CAD model of the workpiece to be sprayed, and to determine the target geometric features of the workpiece to be sprayed based on the three-dimensional CAD model; the target geometric features include surface curvature distribution features, edge contour features, and potential shadow areas; The thickness prediction module is used to acquire a target parameter sequence and use a target prediction model to obtain a predicted coating thickness distribution map corresponding to the workpiece to be coated based on the target parameter sequence and the target geometric features; the target parameter sequence is a sequence of real-time spraying process parameters; the spraying process parameters include robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from the spray gun to the workpiece surface; The workpiece spraying module is used to determine the target thickness deviation based on the predicted coating thickness distribution map and the target coating thickness, and to generate a target spraying strategy based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

[0013] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned intelligent spraying method.

[0014] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned intelligent spraying method.

[0015] In this application, during spraying, a three-dimensional CAD model of the workpiece to be sprayed is acquired, and the target geometric features of the workpiece to be sprayed are determined based on the three-dimensional CAD model. The target geometric features include surface curvature distribution features, edge contour features, and potential shadow areas. A target parameter sequence is acquired, and a target prediction model is used to obtain a predicted coating thickness distribution map corresponding to the workpiece to be sprayed based on the target parameter sequence and the target geometric features. The target parameter sequence is a sequence composed of real-time spraying process parameters. The spraying process parameters include the robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from the spray gun to the workpiece surface. A target thickness deviation is determined based on the predicted coating thickness distribution map and the target coating thickness, and a target spraying strategy is generated based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy. As can be seen, this application utilizes a target prediction model, based on real-time spraying process parameter sequences and the target geometric features of the workpiece, to actively respond to the influence of slow time-varying factors such as spray gun wear, coating characteristic fluctuations, and the target geometric features of the workpiece on the spraying thickness, generating a predicted coating thickness distribution map, thereby achieving more accurate thickness prediction under dynamic working conditions. The deviation between the predicted coating thickness distribution map and the target thickness directly drives the generation of the target spraying strategy, ensuring that spraying actions (such as robot path planning) specifically compensate for local defects. This not only handles the spatial heterogeneity of complex curved surfaces (such as simultaneously dealing with high and low risk areas), but also achieves the consistency of the overall thickness distribution through closed-loop feedback, ultimately improving the spraying quality of complex curved surfaces. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a flowchart of an intelligent spraying method disclosed in this application; Figure 2 This is a schematic diagram of the core architecture of a specific intelligent spraying system disclosed in this application; Figure 3 This is a schematic diagram of a specific intelligent spraying method disclosed in this application; Figure 4 This is a schematic diagram of the architecture of a specific target prediction model disclosed in this application; Figure 5 This is a schematic diagram of the incremental learning process of a specific intelligent spraying system disclosed in this application; Figure 6This is a schematic diagram of the structure of an intelligent spraying device disclosed in this application; Figure 7 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the automated spraying of complex curved workpieces in aerospace, automotive, and high-end equipment industries, ensuring coating thickness uniformity is a major challenge. Workpiece geometry (such as sharp edges, deep recesses, and high-curvature surfaces) can trigger "edge effects" and "shadow effects," leading to systematic and predictable thickness deviations. Existing technologies include methods for parameter compensation based on workpiece geometry. These methods improve uniformity to some extent by identifying high- and low-risk areas and applying predefined compensation strategies. However, these static compensation systems have inherent drawbacks: firstly, their core compensation rules (such as speed and flow rate adjustments) rely on manual experience or offline experiments, making it difficult to accurately match dynamically changing actual working conditions; secondly, once deployed, the system's performance is fixed, unable to cope with slow time-varying disturbances such as spray gun wear and coating characteristic fluctuations, leading to a gradual decline in compensation effectiveness and requiring frequent manual recalibration. To address these technical problems, this application discloses an intelligent spraying method capable of achieving persistent, stable, and high-precision control of the spraying quality of complex curved surfaces.

[0020] See Figure 1 As shown, an embodiment of the present invention discloses an intelligent spraying method, comprising: Step S11: Obtain a three-dimensional CAD model of the workpiece to be coated, and determine the target geometric features of the workpiece based on the three-dimensional CAD model; the target geometric features include surface curvature distribution features, edge contour features, and potential shadow areas.

[0021] In this embodiment, as Figure 2The diagram illustrates the core architecture of a specific intelligent spraying system, comprising a digital domain (i.e., the intelligent decision-making layer) and a physical domain (i.e., the spraying site). The digital domain contains the system's core intelligent module, responsible for information processing, prediction, and decision-making. The physical domain includes the actual spraying equipment, workpieces, and measuring devices, responsible for execution and feedback. During intelligent spraying tasks, the intelligent spraying system's digital twin module extracts geometric features from the input CAD model. In one specific implementation, the target geometric features of the workpiece to be coated are determined based on a 3D CAD model, including: determining the curvature information of each point on the surface of the workpiece based on the 3D CAD model, and determining the surface curvature distribution features corresponding to the workpiece based on the curvature information; determining the edge contour features of the workpiece based on the 3D CAD model using a target edge detection algorithm; the edge contour features include physical edges, ridges, and corresponding edge types (such as convex edges and concave edges); determining the potential shadow areas of the workpiece based on the spraying cone angle model of the target spray gun, the relative pose of the spray gun and the workpiece surface, and the 3D CAD model; the potential shadow areas are areas that cannot directly receive paint because they are obscured by the workpiece itself under a preset spray gun pose. It is understood that these target geometric features are key geometric features that easily affect coating deposition; for example, areas with high curvature are prone to coating accumulation, while areas with low curvature are prone to overspraying. Based on the spraying cone angle model of the spray gun, the relative pose of the spray gun and the workpiece surface, and the 3D model of the workpiece, the areas that may be obscured by the workpiece itself and cannot directly receive paint under a given spray gun pose are determined through spatial ray intersection calculation.

[0022] Step S12: Obtain the target parameter sequence, and use the target prediction model to obtain the predicted coating thickness distribution map corresponding to the workpiece to be sprayed based on the target parameter sequence and the target geometric features; the target parameter sequence is a sequence composed of real-time spraying process parameters; the spraying process parameters include the robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from the spray gun to the workpiece surface.

[0023] In this embodiment, as Figure 3 As shown, the thickness prediction module of the intelligent spraying system uses a built-in target prediction model to calculate and output a predicted thickness distribution map based on the received geometric features (such as image-based feature maps) and real-time process parameters from the robot (i.e., the target parameter sequence). In other words, the target prediction model can map and output a predicted coating thickness distribution map covering the entire workpiece based on the input geometric features and dynamic process parameters. The target parameter sequence is a sequence of real-time spraying process parameters, including the robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and the distance from the spray gun to the workpiece surface. The target prediction model can preferably be a machine learning model.

[0024] In one specific implementation, the target prediction model can preferably be a CNN-LSTM (Convolutional Neural Network, Long Short-Term Memory) fusion model, the architecture of which is illustrated in the diagram below. Figure 4 As shown, the target prediction model obtains the predicted coating thickness distribution map of the workpiece to be sprayed based on the target parameter sequence and target geometric features. This includes: inputting the target parameter sequence and target geometric features into the target prediction model; fusing the target geometric features using the first model branch of the target prediction model to obtain a target feature map; the first model branch is a model branch built on a convolutional neural network (i.e., a CNN branch); obtaining the target dependency relationship based on the target parameter sequence using the second model branch of the target prediction model; the second model branch is a model branch built on a long short-term memory network (i.e., an LSTM branch); the target dependency relationship is the temporal dependency relationship of the spraying process parameters on the scanning path; fusing the target feature map and the target dependency relationship to obtain the fused feature quantity, and using the fully connected layer of the target prediction model to obtain the predicted coating thickness distribution map of the workpiece to be sprayed based on the fused feature quantity. The CNN branch specifically processes the two-dimensional geometric feature map from the digital twin construction module (e.g., fusing curvature, edge, and shadow information into a multi-channel feature map), extracting deep spatial correlation patterns through convolutional layers, such as learning "under what process parameters a high-curvature edge region is prone to thickness peaks". Through multiple convolutional layers (Conv) and activation functions, deep, abstract spatial features are progressively extracted, and finally flattened into a one-dimensional feature vector through a global pooling layer. The LSTM branch receives dynamic spraying process parameters organized in a time series. This branch effectively captures the temporal dependencies of process parameters on the scanning path through multiple LSTM units and outputs its learned temporal features. For example, "After the current high-speed spraying, if the flow rate drops sharply at the next moment, it may cause the coating to be discontinuous." The high-level features extracted by the CNN and LSTM branches are fused in the stitching layer, and then nonlinearly transformed and integrated through a fully connected layer to regress and output the thickness value of each predicted point, finally generating a complete predicted thickness distribution map. The outputs processed by the two branches are fused in the stitching layer, combining the spatial features representing the "shape" of the workpiece with the temporal features representing "how to spray," forming a comprehensive and integrated description of the spraying process. It can be understood that the predicted coating thickness distribution map is the decision-making basis of the entire intelligent spraying system, and its accuracy directly determines the final spraying quality.

[0025] Step S13: Determine the target thickness deviation based on the predicted coating thickness distribution map and the target coating thickness, and generate a target spraying strategy based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

[0026] In this embodiment, the compensation decision module of the intelligent spraying system generates an adaptive compensation strategy based on the predicted deviation. This strategy is then transmitted to the spraying robot via the network, enabling precise spraying of the workpiece. Together with the aforementioned steps, this forms a rapid and intelligent feedforward control loop from "perception" to "execution." Generating a target spraying strategy based on the target thickness deviation specifically includes: determining whether the target thickness deviation is less than a target deviation threshold; if the target thickness deviation is less than the target deviation threshold, then determining the preset basic spraying strategy as the target spraying strategy; if the target thickness deviation is greater than the target deviation threshold, then correcting the trajectory normal and adjusting the spraying process parameters of the basic spraying strategy based on preset safety constraints, and determining the adjusted basic spraying strategy as the target spraying strategy. In other words, the compensation decision module compares the predicted coating thickness distribution map with the preset target thickness, generates a comprehensive compensation strategy based on the thickness deviation at various points in space, and then adjusts the preset basic spraying strategy based on this compensation strategy to obtain the corresponding target spraying strategy. The compensation strategy includes real-time correction instructions for the robot's motion trajectory and / or dynamic adjustment instructions for the spraying process parameters. It is understood that the parameters adjusted by the dynamic adjustment command include, but are not limited to, the adjustment amount of the robot's TCP speed and / or the adjustment amount of the spray gun flow rate, and the adjustment range is limited by preset safety constraints. This compensation strategy enables the intelligent spraying system to automatically and online compensate for slow time-varying disturbances that static systems cannot handle, such as spray gun efficiency decay, paint viscosity changes, and environmental fluctuations. It possesses adaptive capabilities to combat slow time-varying disturbances, greatly reducing the system's reliance on manual calibration and improving the long-term stability and consistency of the production process. In a specific implementation, the compensation strategy generation method includes: 1. Trajectory Correction: Based on the spatial gradient of the thickness deviation, calculate the offset of each point on the robot's original planned trajectory along the normal to the workpiece surface. For areas where the predicted thickness is lower than the target, the trajectory is finely adjusted inward to shorten the spraying distance; for areas where the predicted thickness is higher than the target, the trajectory is finely adjusted outward.

[0027] 2. Parameter Adjustment: Based on the thickness deviation, the adjustment amount of the spraying parameters is calculated using a preset control law (such as a proportional-integral rule). For example, the flow rate adjustment amount. , where e(t) is the thickness deviation at the current point, Kp, Ki are adjustable gains, and ΔFlow is limited by the preset maximum safe variation range.

[0028] 3. Dual-dimensional collaboration: Trajectory correction and parameter adjustment instructions are calculated and packaged and issued synchronously by this module to ensure coordinated changes in spatial motion and process output, avoiding mutual conflicts or oscillations.

[0029] In this embodiment, the number of painting robots can be two or more. For this type of operation, a zoned coordination strategy can be adopted to control multiple robots to perform collaborative painting. Specifically, this can include: determining the geometric complexity of the workpiece to be painted based on the target geometric features; determining the target painting responsibility area based on the geometric complexity, target thickness deviation, and the workspace range of each target robot; generating a target painting strategy corresponding to the target painting responsibility area based on the geometric complexity and target thickness deviation; allocating corresponding target robots to the target painting responsibility area based on the workspace range of each target robot, so as to control the target robots to paint the workpiece to be painted based on the target painting strategy. Simultaneously, an overlapping painting area can be set at the boundary of adjacent target painting responsibility areas; when at least one target robot corresponding to the overlapping painting area paints the overlapping painting area based on the target painting strategy, the data transmission speed of all target robots corresponding to the overlapping painting area is synchronized, and the painting flow commands corresponding to each target robot are weighted and fused based on position, so as to paint the workpiece to be painted based on the fused painting flow commands.

[0030] In other words, in this embodiment, a partitioning coordination strategy can be implemented through intelligent partitioning and overlapping area synchronization control to control multiple painting robots to perform collaborative painting. Specifically, the intelligent partitioning process includes: dividing the workpiece surface into several non-overlapping or partially overlapping painting responsibility areas based on the geometric complexity of the workpiece CAD model and the workspace range of each robot, and assigning them to different robots. The overlapping area synchronization control process includes: pre-setting an adjustable-width overlapping painting area at the boundary of adjacent robot responsibility areas. When two robots paint into this overlapping area, they synchronize their TCP speed with the control module and perform position-based weighted fusion of their painting flow commands (e.g., the farther away from the center of their respective responsibility areas, the lower the weight) to ensure a smooth transition of coating thickness within this area without significant differences. In addition, the actual thickness data measured within the overlapping area can be simultaneously fed back to the online evolution modules corresponding to the two robots responsible for this area for incremental learning of their respective models. When the target robot controls the target spraying strategy to spray the workpiece, the execution control module of the intelligent spraying system will decode the received target spraying strategy into motion commands for each axis of the robot and control signals for the spray gun valve, driving the robot to complete precise spraying.

[0031] In this embodiment, as Figure 5As shown, after the target robot controls the spraying of the workpiece based on the target spraying strategy, the process further includes: acquiring the actual coating thickness data corresponding to the target spraying strategy; constructing a target training dataset based on the actual coating thickness data and the target parameter sequence when spraying the workpiece; updating the target prediction model based on the target training dataset using a preset incremental learning mechanism to obtain an updated target prediction model; and when performing a spraying task on a new workpiece, using the updated target prediction model to obtain the predicted coating thickness distribution map corresponding to the new workpiece. In other words, the intelligent spraying system can collect the actual process parameters and corresponding actual thickness data of the current spraying operation, and use this data to incrementally train the currently used machine learning model to generate an updated and more accurate machine learning model. When spraying the next similar workpiece, the updated machine learning model replaces the previous model, and the aforementioned steps are repeated to achieve continuous evolution of the intelligent spraying system. Even if the target prediction model is not perfect, the system can quickly converge to a high-precision state after a small number of production batches through online learning, reducing the model's dependence on previous data and training, thereby reducing the cost and difficulty of the massive training data required in the early stages and accelerating system deployment.

[0032] For example, after a single workpiece is coated or during interlayer intervals, the online evolution module of the intelligent coating system collects the actual process parameter sequence during the coating process of the workpiece and the corresponding actual coating thickness data obtained by a measuring device (such as an offline thickness gauge or an integrated online thickness sensor), forming a new training data pair; and using these newly collected online data, incremental learning is performed on the current machine learning model in the thickness prediction module to update its model parameters, forming an updated and more accurate machine learning model, which is immediately used for the prediction task of the next workpiece to be coated.

[0033] The online evolution module employs the Online Stochastic Gradient Descent algorithm for incremental model learning. Its specific process includes: 1. Data Buffer: Collect the "actual process parameters - actual thickness" data pairs generated during the current workpiece spraying process to form a mini-batch training dataset.

[0034] 2. Model Loading: Loads the currently used model version in the thickness prediction module (i.e., the current version of the target prediction model).

[0035] 3. Forward and Loss Calculation: Input the micro-batch dataset into the current model and calculate the loss function value (such as mean square error) between the model's predicted thickness and the actual measured thickness.

[0036] 4. Backpropagation and parameter fine-tuning: Based on the loss function value, perform one or a finite number of gradient descent iterations to update some network weights of the model with a low learning rate (preferably updating the weights of fully connected layers) to complete the incremental training of the model.

[0037] 5. Version Update and Security Rollback (Core Security Mechanism): After incremental training is completed, a new version of the candidate model is generated. The candidate model undergoes performance verification (e.g., comparing its performance on a reserved historical verification dataset with the previous version). If verification passes, it is deployed as the updated target prediction model for subsequent workpiece painting; if verification fails, the system automatically rolls back to the stable model version before the update, thereby ensuring the continuity and safety of the production process.

[0038] In one specific implementation, such as Figure 5 As shown, the incremental learning process is triggered by data acquisition signals and includes a readiness check to ensure that the learning process starts when data is available. During incremental learning, the intelligent spraying system first loads and locks the currently used target prediction model as the learning foundation, ensuring the continuity of learning. The core employs incremental learning algorithms such as online stochastic gradient descent, and the process includes single-round iteration, loss calculation, and weight fine-tuning, rather than retraining, ensuring learning speed and efficiency and meeting the real-time requirements of industrial sites. After generating an updated model, the system assigns it a new version number, achieving versioned model management. A unique performance verification and rollback mechanism is key to incremental learning; if model verification fails after an update, the system automatically rolls back to the previous stable version. This mechanism greatly improves the system's robustness and engineering practicality, preventing production interruptions or quality accidents due to a single poor learning outcome. Regardless of learning success or failure, the process eventually returns to a listening state, waiting for the next batch of data, forming a continuous, safe, and reliable learning loop, perfectly embodying the design concept of "lifelong learning." This incremental learning mechanism transforms the model update proposed in this embodiment from a simple "training-deployment" cycle into an advanced model management mechanism with industrial-grade reliability, version control, and self-recovery capabilities.

[0039] In this embodiment, from Figure 3As can be seen, the decision box "Continue spraying the next workpiece?" and the arrow pointing back from the "Online Learning" step to the "Input Model" step intuitively form a self-optimization closed loop of the method. This clearly shows that the updated model will be immediately applied to the prediction of subsequent workpieces, enabling the system performance to continuously evolve with the production batches. This flowchart strongly demonstrates that the intelligent spraying method implemented in this embodiment is no longer a one-time, fixed compensation procedure, but an intelligent method with continuous learning and autonomous optimization capabilities. Through the "single workpiece-incremental learning" closed loop, the intelligent spraying system can continuously learn from each production practice, making the target prediction model increasingly closer to the instantaneous state of the actual production line (including equipment degradation and material fluctuations). The compensation accuracy does not decrease but increases over time, realizing the system's lifelong learning and performance evolution, completely eliminating the dependence on periodic manual calibration. It can be understood that this embodiment, by integrating a dedicated prediction model of CNN-LSTM, a two-dimensional real-time compensation strategy, and a multi-robot collaborative control mechanism, provides a closed-loop, intelligent, and scalable complete solution for the uniform spraying of large, complex curved surface workpieces, with comprehensive technical advantages.

[0040] As can be seen, this application utilizes a target prediction model, based on real-time spraying process parameter sequences and the target geometric features of the workpiece, to actively respond to the influence of slow time-varying factors such as spray gun wear, coating characteristic fluctuations, and the target geometric features of the workpiece on the spraying thickness, generating a predicted coating thickness distribution map, thereby achieving more accurate thickness prediction under dynamic working conditions. The deviation between the predicted coating thickness distribution map and the target thickness directly drives the generation of the target spraying strategy, ensuring that spraying actions (such as robot path planning) specifically compensate for local defects. This not only handles the spatial heterogeneity of complex curved surfaces (such as simultaneously dealing with high and low risk areas), but also achieves the consistency of the overall thickness distribution through closed-loop feedback, ultimately improving the spraying quality of complex curved surfaces.

[0041] As described in the previous embodiment, this application discloses an intelligent spraying method that can achieve persistent, stable, and high-precision control over the spraying quality of complex curved surfaces. Next, this application will provide spraying embodiments in specific application scenarios to further illustrate the intelligent spraying method.

[0042] In the first application scenario, taking drone fuselage painting as an example, when the system is first activated, the thickness prediction module has a built-in CNN-LSTM model (initial model V1.0) pre-trained on a historical dataset. The operator imports the drone fuselage CAD model, and the system automatically extracts features such as surface curvature, edges (e.g., wing leading edges), and potential shadow areas (e.g., the junction of the fuselage belly and tail). The system predicts, compensates for, and paints the first fuselage. After completion, the operator uses a thickness gauge to measure the actual thickness at 50 points on the fuselage surface. The online evolution module combines the new "process parameters - actual thickness" data from these 50 selected thickness measurement points into a micro-batch dataset. Instead of retraining the entire model, it uses an online stochastic gradient descent algorithm with a low learning rate (e.g., 0.001) to iteratively update only a portion of the weights (fully connected layer weights) of the initial model (V1.0) using this new dataset. This process is extremely short (usually within seconds, generating the updated V1.1 model). When painting the second part of the machine body, the system automatically calls the V1.1 model for thickness prediction. Since the V1.1 model has already incorporated the actual information of the first workpiece, its prediction accuracy for the current spray gun and paint is higher than that of the initial model. This process is repeated, with the model version evolving from V1.1 to V1.2, V1.3, and so on. The system performance continuously evolves, eventually achieving a compensation accuracy on the production line that far exceeds that of the static system.

[0043] In the second application scenario, taking the painting of a large automotive door panel as an example, the workpiece is large and has a complex curved surface. Two six-axis painting robots work collaboratively. Partitioning and Planning: The execution and control module divides the door panel into two main painting areas, left and right, based on the door panel's CAD model, assigning them to robots A and B respectively, and setting a 20mm wide overlap zone in the middle area. Prediction and Compensation: The digital twin construction module extracts the overall geometric features of the door panel. The thickness prediction module (using the current model evolved to V2.5) outputs a predicted thickness map of the entire door panel. The adaptive compensation planning module generates a compensation strategy based on this. For the recessed area of ​​the door panel handle (where predictions tend to be thicker), the strategy includes: slightly deflecting the robot trajectory outwards at this location, while slightly reducing the flow rate. Collaborative Execution: The two robots start painting synchronously. When they move to the middle overlap zone, the execution and control module dynamically adjusts to keep their TCP speeds consistent and proportionally merges their calculated flow rate commands to ensure uniform coating in the overlap area. Collaborative Learning: After painting is completed, a full-field thickness scan of the door panel is performed. The measurement data of the overlapping area is simultaneously provided to the online evolution modules corresponding to robots A and B, which are used to update their respective dedicated prediction model sub-modules (or jointly update a global model), thereby allowing both robots to increase their "experience".

[0044] As can be seen from the above embodiments, the intelligent spraying system built based on the intelligent spraying method proposed in this application can not only become more proficient through online learning, but also effectively manage complex operation scenarios involving multiple robots, demonstrating strong industrial applicability.

[0045] See Figure 6 As shown, this application discloses an intelligent spraying device, comprising: The geometric feature determination module 11 is used to acquire a three-dimensional CAD model of the workpiece to be sprayed, and to determine the target geometric features of the workpiece to be sprayed based on the three-dimensional CAD model; the target geometric features include surface curvature distribution features, edge contour features, and potential shadow areas; The thickness prediction module 12 is used to acquire a target parameter sequence and use a target prediction model to acquire a predicted coating thickness distribution map corresponding to the workpiece to be sprayed based on the target parameter sequence and the target geometric features; the target parameter sequence is a sequence of real-time spraying process parameters; the spraying process parameters include robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from the spray gun to the workpiece surface; The workpiece spraying module 13 is used to determine the target thickness deviation based on the predicted coating thickness distribution map and the target coating thickness, and generate a target spraying strategy based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

[0046] As can be seen, this application utilizes a target prediction model, based on real-time spraying process parameter sequences and the target geometric features of the workpiece, to actively respond to the influence of slow time-varying factors such as spray gun wear, coating characteristic fluctuations, and the target geometric features of the workpiece on the spraying thickness, generating a predicted coating thickness distribution map, thereby achieving more accurate thickness prediction under dynamic working conditions. The deviation between the predicted coating thickness distribution map and the target thickness directly drives the generation of the target spraying strategy, ensuring that spraying actions (such as robot path planning) specifically compensate for local defects. This not only handles the spatial heterogeneity of complex curved surfaces (such as simultaneously dealing with high and low risk areas), but also achieves the consistency of the overall thickness distribution through closed-loop feedback, ultimately improving the spraying quality of complex curved surfaces.

[0047] In one specific implementation, the geometric feature determination module 11 may include: The first feature determination unit is used to determine the curvature information of each point on the surface of the workpiece to be sprayed based on the three-dimensional CAD model, and to determine the surface curvature distribution features of the workpiece to be sprayed based on the curvature information. The second feature determination unit is used to determine the edge contour features of the workpiece to be coated based on the three-dimensional CAD model using a target edge detection algorithm; the edge contour features include physical edges, ridges and corresponding edge types. The third feature determination unit is used to determine the potential shadow area of ​​the workpiece to be sprayed based on the spraying cone angle model of the target spray gun, the relative pose of the spray gun and the workpiece surface, and the three-dimensional CAD model; the potential shadow area is the area that cannot directly receive paint because it is blocked by the workpiece itself under the preset spray gun pose.

[0048] In one specific implementation, the predicted thickness determination module 12 may include: A data input unit is used to input the target parameter sequence and the target geometric features into the target prediction model; The feature map acquisition unit is used to fuse the geometric features of each target using the first model branch of the target prediction model to obtain a target feature map; the first model branch is a model branch constructed based on a convolutional neural network; The dependency determination unit is used to obtain the target dependency based on the target parameter sequence using the second model branch of the target prediction model; the second model branch is a model branch constructed based on a long short-term memory network; the target dependency is the temporal dependency of the spraying process parameters on the scanning path; The predicted thickness determination unit is used to fuse the target feature map and the target dependency to obtain the fused feature quantity, and use the fully connected layer of the target prediction model to obtain the predicted coating thickness distribution map corresponding to the workpiece to be sprayed based on the fused feature quantity.

[0049] In one specific embodiment, the workpiece spraying module 13 may include: A deviation determination unit is used to determine whether the target thickness deviation is less than a target deviation threshold. The first strategy determination unit is used to determine the preset basic spraying strategy as the target spraying strategy if the target thickness deviation is less than the target deviation threshold. The second strategy determination unit is used to correct the trajectory normal and adjust the spraying process parameters of the basic spraying strategy based on preset safety constraints if the target thickness deviation is greater than the target deviation threshold, and to determine the adjusted basic spraying strategy as the target spraying strategy.

[0050] In one specific embodiment, the workpiece spraying module 13 may include: A complexity determination unit is used to determine the geometric complexity of the workpiece to be coated based on the target geometric features. The responsibility area division unit is used to determine the target spraying responsibility area based on the geometric complexity, the target thickness deviation, and the workspace range corresponding to each target robot. A spraying strategy generation unit is used to generate a target spraying strategy corresponding to the target spraying responsibility area based on the geometric complexity and the target thickness deviation. The first spraying unit is used to allocate the corresponding target robot to the target spraying responsibility area based on the workspace range of each target robot, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

[0051] In one specific embodiment, the workpiece spraying module 13 may include: An overlapping area determination unit is used to set an overlapping spraying area at the boundary of adjacent target spraying responsibility areas; The second spraying unit is used to synchronize the data transmission speed of all target robots corresponding to the overlapping spraying area when at least one target robot corresponding to the overlapping spraying area sprays the overlapping spraying area based on the target spraying strategy, and to perform position-based weighted fusion of the spraying flow instructions corresponding to each target robot, so as to spray the workpiece to be sprayed based on the fused spraying flow instructions.

[0052] In one specific embodiment, the device may further include: The actual thickness determination unit is used to obtain the actual coating thickness data corresponding to the target spraying strategy. The training set construction unit is used to construct a target training dataset based on the actual coating thickness data and the target parameter sequence when spraying the workpiece to be coated. The model update unit is used to update the target prediction model based on the target training dataset using a preset incremental learning mechanism to obtain the updated target prediction model, and when performing a spraying task on a new workpiece to be sprayed, to use the updated target prediction model to obtain the predicted coating thickness distribution map corresponding to the new workpiece to be sprayed.

[0053] Furthermore, embodiments of this application also disclose an electronic device, Figure 7 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0054] Figure 7This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the intelligent spraying method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0055] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0056] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored thereon can include an operating system 221, computer programs 222, etc., and the storage method can be temporary storage or permanent storage.

[0057] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the intelligent spraying method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0058] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned intelligent spraying method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0059] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0060] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0061] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0062] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0063] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An intelligent spraying method, characterized in that, include: Obtain a 3D CAD model of the workpiece to be coated, and determine the target geometric features of the workpiece based on the 3D CAD model; The target geometric features include surface curvature distribution features, edge contour features, and potential shadow regions; Obtain the target parameter sequence, and use the target prediction model to obtain the predicted coating thickness distribution map corresponding to the workpiece to be coated based on the target parameter sequence and the target geometric features; The target parameter sequence is a sequence of real-time spraying process parameters; the spraying process parameters include robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from the spray gun to the workpiece surface. The target thickness deviation is determined based on the predicted coating thickness distribution map and the target coating thickness, and a target spraying strategy is generated based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

2. The intelligent spraying method according to claim 1, characterized in that, Determining the target geometric features of the workpiece to be coated based on the 3D CAD model includes: Based on the three-dimensional CAD model, the curvature information of each point on the surface of the workpiece to be sprayed is determined, and based on the curvature information, the surface curvature distribution characteristics of the workpiece to be sprayed are determined. The edge contour features of the workpiece to be coated are determined based on the three-dimensional CAD model using a target edge detection algorithm; the edge contour features include physical edges, ridges, and corresponding edge types. Based on the spray cone angle model of the target spray gun, the relative pose of the spray gun and the workpiece surface, and the three-dimensional CAD model, the potential shadow area of ​​the workpiece to be sprayed is determined; the potential shadow area is the area that cannot directly receive paint because it is blocked by the workpiece itself under the preset spray gun pose.

3. The intelligent spraying method according to claim 1, characterized in that, The step of obtaining the predicted coating thickness distribution map corresponding to the workpiece to be coated based on the target parameter sequence and the target geometric features using the target prediction model includes: The target parameter sequence and the target geometric features are input into the target prediction model; The first model branch of the target prediction model is used to fuse the geometric features of each target to obtain a target feature map; the first model branch is a model branch built based on a convolutional neural network. The target dependency is obtained based on the target parameter sequence using the second model branch of the target prediction model; the second model branch is a model branch constructed based on a long short-term memory network; the target dependency is the temporal dependency of the spraying process parameters on the scanning path; The target feature map and the target dependency are fused to obtain fused feature values, and the predicted coating thickness distribution map corresponding to the workpiece to be sprayed is obtained based on the fused feature values ​​using the fully connected layer of the target prediction model.

4. The intelligent spraying method according to claim 1, characterized in that, The target coating strategy based on the target thickness deviation includes: Determine whether the target thickness deviation is less than the target deviation threshold; If the target thickness deviation is less than the target deviation threshold, then the preset basic spraying strategy is determined as the target spraying strategy; If the target thickness deviation is greater than the target deviation threshold, the trajectory normal of the basic spraying strategy is corrected and the spraying process parameters are adjusted based on the preset safety constraints, and the adjusted basic spraying strategy is determined as the target spraying strategy.

5. The intelligent spraying method according to claim 1, characterized in that, The step of generating a target spraying strategy based on the target thickness deviation, and controlling a target robot to spray the workpiece to be sprayed based on the target spraying strategy, includes: The geometric complexity of the workpiece to be coated is determined based on the target geometric features. The target spraying responsibility area is determined based on the geometric complexity, the target thickness deviation, and the workspace range corresponding to each target robot. Based on the geometric complexity and the target thickness deviation, a target spraying strategy is generated corresponding to the target spraying responsibility area; Based on the workspace range corresponding to each target robot, the target robot is assigned to the target spraying responsibility area, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

6. The intelligent spraying method according to claim 5, characterized in that, The step of controlling the target robot to spray the workpiece based on the target spraying strategy includes: An overlapping spraying area is set at the boundary of adjacent target spraying responsibility areas; When at least one of the target robots corresponding to the overlapping spraying area sprays the overlapping spraying area based on the target spraying strategy, the data transmission speed of all the target robots corresponding to the overlapping spraying area is synchronized, and the spraying flow command corresponding to each target robot is weighted and fused based on position, so as to spray the workpiece to be sprayed based on the fused spraying flow command.

7. The intelligent spraying method according to claim 1, characterized in that, After controlling the target robot to spray the workpiece based on the target spraying strategy, the method further includes: Obtain the actual coating thickness data corresponding to the target spraying strategy; A target training dataset is constructed based on the actual coating thickness data and the target parameter sequence when spraying the workpiece to be coated. The target prediction model is updated based on the target training dataset using a preset incremental learning mechanism to obtain the updated target prediction model. When performing a spraying task on a new workpiece to be sprayed, the updated target prediction model is used to obtain the predicted coating thickness distribution map corresponding to the new workpiece to be sprayed.

8. An intelligent spraying device, characterized in that, include: The geometric feature determination module is used to acquire a three-dimensional CAD model of the workpiece to be sprayed, and to determine the target geometric features of the workpiece to be sprayed based on the three-dimensional CAD model. The target geometric features include surface curvature distribution features, edge contour features, and potential shadow regions; The predicted thickness determination module is used to obtain a target parameter sequence and use a target prediction model to obtain a predicted coating thickness distribution map corresponding to the workpiece to be coated based on the target parameter sequence and the target geometric features; The target parameter sequence is a sequence of real-time spraying process parameters; the spraying process parameters include robot tool center point speed, spray gun paint flow rate, atomization pressure, forming air pressure, and distance from the spray gun to the workpiece surface. The workpiece spraying module is used to determine the target thickness deviation based on the predicted coating thickness distribution map and the target coating thickness, and to generate a target spraying strategy based on the target thickness deviation, so as to control the target robot to spray the workpiece to be sprayed based on the target spraying strategy.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the intelligent spraying method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the intelligent spraying method as described in any one of claims 1 to 7.

Images