Intelligent imaging generation system and method

Abstract

According to the intelligent imaging generation system and method, parameters of a spraying mechanical arm and imaging equipment are adjusted in a linkage mode in real time through a dynamic parameter adjusting module, and closed-loop control over the spraying quality is achieved in combination with a defect prediction and optimization decision module and a multi-modal process knowledge base. A high-resolution visual sensor, a capacitive thickness detector and a temperature and humidity sensor are integrated in the system, multi-source data are fused through a Kalman filtering algorithm, and the mechanical arm speed, the nozzle pressure and the imaging gain coefficient are dynamically adjusted. The defect prediction module simulates a coating defect form based on a physical constraint generative adversarial network and generates a process optimization instruction set through reinforcement learning. The multi-modal process knowledge base stores a correlation model of material characteristics, environmental data and defect modes, and supports incremental learning and historical data attenuation updating. And through dynamic parameter collaborative optimization, defect prediction preposition and multi-modal data fusion, the spraying quality detection efficiency and the coating consistency are improved, and the rework rate is reduced.

Description

Technical Field

[0001] The present invention relates to the field of spray production technology, and in particular to an intelligent imaging generation system and method, which combines dynamic parameter adjustment, multimodal data fusion and spray process optimization to achieve real-time monitoring and closed-loop control of industrial spray quality, and is applied to spray production lines. Background Art

[0002] In the field of industrial manufacturing, the spray coating production line is a key link in surface treatment. Its spray coating quality directly affects the appearance, durability, and market competitiveness of the product. With the continuous development of intelligent manufacturing technology, the spray coating imaging system, as an important component of the spray coating production line, undertakes the important task of real-time monitoring and feedback control of spray coating quality. However, the existing spray coating imaging system still has many shortcomings in practical application, as follows:

[0003] 1. The spray imaging parameters are fixed and cannot be dynamically adjusted according to the characteristics of the spray material, resulting in missed defects or overexposure. 2. Spray defect assessment requires manual visual inspection or offline analysis, and cannot be fed back to the spray equipment in real time to adjust the parameters. 3. Spray process parameters, material data and imaging data are processed independently, lacking cross-modal optimization capabilities. 4. When adjusting the spray process parameters, the system may over-invest resources in pursuit of high quality. For example, the spray parameters are constantly adjusted to reduce the defect rate, but the cost increases brought about by these adjustments may not be taken into account, such as the increase in coating costs and the extension of production time. This will cause the company to significantly increase costs while improving quality, reducing the company's economic benefits and market competitiveness. The existence of these problems leads to inefficient spray quality inspection, insufficient accuracy, and an inability to balance quality and cost, which is not conducive to the real-time optimization control of the spray process.

[0004] Therefore, how to overcome the above-mentioned defects has become an important issue to be solved urgently by those skilled in the art. Summary of the Invention

[0005] The present invention overcomes the deficiencies of the above-mentioned technologies and provides an intelligent imaging generation system and method.

[0006] The purpose of this application is to provide an intelligent imaging generation system and method and a spray imaging method, which have the advantages of real-time dynamic adjustment of spray imaging parameters, realization of multimodal data collaborative optimization, and improvement of spray quality detection efficiency and accuracy.

[0007] The first aspect of the present application provides an intelligent imaging generation system and method, including:

[0008] The dynamic parameter adjustment module 11 obtains the spray surface status in real time through multi-sensor fusion and adjusts the operating parameters of the imaging device and the spray robot arm in a coordinated manner;

[0009] Defect prediction and optimization decision module 12, based on a physical constraint-based generative adversarial network model, simulates coating defect morphology under different process parameters and generates a process optimization instruction set through reinforcement learning. The process optimization instruction set includes collaborative optimization instructions for coating viscosity adjustment value, spray speed threshold, and spray gun distance parameters;

[0010] The multimodal process knowledge base 13 stores the correlation between the material characteristic data, environmental sensor data, process parameter history data and defect patterns of the sprayed device.

[0011] Preferably, the dynamic parameter adjustment module 11 includes:

[0012] The multi-data acquisition unit 111 includes:

[0013] A high-resolution visual sensor 1111 is used to capture surface texture features and generate a first data stream;

[0014] a capacitive thickness detector 1112, which monitors the coating thickness distribution in real time and generates a second data stream;

[0015] The temperature and humidity sensor 1113 collects environmental parameters and generates a third data stream;

[0016] The data fusion unit 112 uses a Kalman filter algorithm to perform spatiotemporal alignment and noise suppression on the first, second, and third data streams.

[0017] Preferably, the dynamic parameter adjustment module 11 further includes:

[0018] Intelligent perception and decision-making unit 113, which includes:

[0019] The robot movement speed adjustment unit 1131 is used to extract the spray surface roughness characteristic value Ra in real time based on the first data stream, and dynamically adjust the robot movement speed v according to a preset robot movement speed formula; wherein the robot movement speed formula is: v = k1 * Ra + b1; k1 and b1 are roughness-speed response coefficients obtained through experimental calibration; Ra ranges from 0.1 to 10 μm;

[0020] Nozzle pressure optimization and adjustment unit 1132 is configured to optimize nozzle pressure P using a preset nozzle pressure optimization formula based on the coating thickness distribution data monitored by the second data stream and the adhesion level F obtained by the mechanical adhesion tester. The nozzle pressure optimization formula is: P = k2*F+b2, where k2 and b2 are experimentally calibrated pressure compensation coefficients. Adhesion level F is divided into levels 1-5.

[0021] Imaging device gain coefficient adjustment unit 1133: used to combine the third data source and use the imaging device gain coefficient compensation formula to compensate the imaging device gain coefficient G; the imaging device gain coefficient compensation formula is: G = G0*(1+α*RH%+β*T℃); G0 is the reference gain coefficient; RH is the ambient relative humidity; T is the ambient temperature; α and β are temperature and humidity compensation coefficients, α is in 1 / %, and β is in 1 / ℃.

[0022] Preferably, the dynamic adjustment process of each adjustment unit of the dynamic parameter adjustment module 11 satisfies:

[0023] When the surface roughness Ra>5μm, the workpiece surface is judged to be relatively rough and the spiral spraying trajectory is automatically switched;

[0024] When the surface roughness Ra is less than 1μm, the workpiece surface is judged to be relatively smooth and the normal high-speed spraying trajectory is automatically switched;

[0025] When the adhesion level F is less than level 1, the adhesion is judged to be too low and the nozzle pressure is automatically increased to enhance the bite between the coating and the substrate;

[0026] When the adhesion level F>5, the adhesion is judged to be too high and the nozzle pressure is automatically reduced to avoid coating embrittlement;

[0027] When the temperature T>35℃ or the humidity RH>75%, the working environment is judged to be overheated or overhumid, and the gain is dynamically adjusted through the imaging device gain coefficient compensation formula to correct the environmental interference.

[0028] Preferably, the dynamic parameter adjustment module 11 further includes an exception handling mechanism:

[0029] When the instantaneous wind speed is detected to be greater than the preset threshold, the following operations are executed:

[0030] Pause spraying and start 3D laser scanning to reconstruct the surface topography;

[0031] Generate local spraying path based on point cloud data;

[0032] When the coating thickness unevenness is greater than 1%, the adaptive grid spraying mode is triggered.

[0033] Preferably, the defect prediction and optimization decision module 12 includes:

[0034] A defect simulation unit 121 generates a coating defect prediction map under different process parameters based on a physical constraint adversarial network model; the adversarial network model is a GAN model;

[0035] The generator input of the GAN model includes a set of process parameters, material property data, environmental data, and substrate state parameters, so as to output a coating defect prediction map under different process parameters through its discriminator; the discriminator of the GAN model integrates the coating defect physical equation constraints, including:

[0036] Sagging length Where ρ is the density of the coating, g is the acceleration of gravity, h is the coating thickness, η is the viscosity of the coating, v is the spraying speed, "C is an empirical coefficient containing the length dimension, and σ is the surface tension of the coating;

[0037] The reinforcement learning optimization unit 122 uses the real-time spraying quality index as the state space and the process parameter adjustment amount as the action space to optimize the process parameters through the reward function;

[0038] The closed-loop control unit 123 sends the optimized process parameters to the spray robot 301 and the imaging device 302 in real time, and receives new sensor data for iterative optimization;

[0039] Among them, the reward function is defined as:

[0040] R = w1·(1-(|D_{actual}-D_{target}| / D_{target}))+w2·C_{saved} / C_{total}; D_{actual} is the actual defect rate, D_{target} is the target defect rate, C_{saved} is the actual saved paint cost, C_{total} is the total paint cost budget, w1 and w2 are weight coefficients, satisfying w1+w2=1.

[0041] Preferably, it also includes:

[0042] A mode selection module (14) is used to dynamically select a quality priority mode and a cost priority mode according to the type of workpiece production;

[0043] Among them, when the system detects the production of critical workpieces, it automatically selects the quality priority mode and limits w_1>w_2; the system will call the preset process optimization instruction set to reduce defects.

[0044] When the system detects a non-critical workpiece, it automatically selects the cost priority mode and limits w_2>w_1; the system will call the preset process optimization instruction set to ensure cost.

[0045] Preferably, the multimodal process knowledge base 13 includes:

[0046] The material property storage unit 131 is used to store the material property data of the sprayed device, and the material property data at least includes: substrate type and its physical properties; coating properties;

[0047] Environmental data storage unit 132, for storing environmental data, the environmental data including at least: temperature, humidity parameters, wind speed parameters;

[0048] The process parameter storage unit 133 is used to store process parameters, which include at least nozzle pressure and spraying speed parameters;

[0049] An image data storage unit 134 is used to store optical detection images of defective parts, wherein the optical detection images include at least microscopic photos of orange peel and sagging;

[0050] A text record storage unit 135 is used to store operation logs, maintenance reports, and quality inspection notes during the production process;

[0051] The defect case storage unit 136 is used to store the operating condition characteristic data packets of historical defect cases, wherein the operating condition characteristic data includes process parameter combinations and defect information; the defect information includes at least defect type, defect severity, and defect repair measures;

[0052] The defect mode association storage unit 137 is used to store the association between material properties, environmental data and defect modes, including:

[0053] a) Mapping relationship between material type and environmental parameters to defect type;

[0054] b) Quantitative correlation model between environmental parameters and defect morphology;

[0055] c) Correlation model between substrate surface characteristics and defect spatial distribution.

[0056] Preferably, the multimodal process knowledge base 13 is updated in the following manner:

[0057] When the prediction deviation of a new defect case is greater than 15%, incremental learning is triggered to update the associated graph weights;

[0058] Apply a time decay factor to historical data older than 90 days;

[0059] Correlation between porosity distribution of storage coatings and process parameters.

[0060] A second aspect of the present application discloses an intelligent imaging generation method, characterized in that it is applied to the spray imaging system described in the first aspect, comprising the following steps:

[0061] S1. Real-time data acquisition: Use multiple sensors to synchronously acquire spray surface texture characteristics, coating thickness distribution, and ambient temperature and humidity data;

[0062] S2. Dynamic parameter coordination adjustment: Dynamically calculate the robot arm movement speed v based on the surface roughness Ra, optimize the nozzle pressure P based on the adhesion level F, and adjust the imaging device gain coefficient G based on the temperature and humidity compensation formula;

[0063] S3. Defect simulation prediction: Input the process parameter set into the physically constrained GAN model to generate a coating defect morphology prediction map and identify sagging defect risk areas;

[0064] S4. Process optimization decision-making: Select weight coefficients based on the quality / cost model and iteratively generate a collaborative optimization instruction set for paint viscosity, spray speed, and gun distance through reinforcement learning;

[0065] S5. Closed-loop execution and feedback: Optimization instructions are sent to the spray robot and imaging equipment, and the reward function is updated based on the newly collected coating quality data, triggering incremental learning to optimize the knowledge base association model.

[0066] Furthermore, the method of the present application also proposes: S6. Abnormal adaptive processing: when it is detected that the wind speed is greater than 2m / s or the thickness unevenness is greater than 1%, the spraying is suspended and the local spraying path planning is started, and the exposure parameters of the imaging equipment are corrected synchronously.

[0067] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0068] 1. This case realizes dynamic optimization of spray imaging parameters and real-time defect prediction through the real-time linkage of equipment parameters by the dynamic parameter adjustment module, generation of process optimization instruction set by the defect prediction module, and establishment of a cross-modal association model by the multimodal knowledge base, thus achieving the advantages of improving detection efficiency and accuracy.

[0069] 2. This case achieved cross-modal correlation between surface state, coating performance, and environmental parameters by establishing a mathematical model for multi-source data fusion, solving the problems of parameter adjustment lag and single-dimensional optimization. In addition, the robot arm speed of this application is dynamically adjusted according to the surface roughness, avoiding defects caused by paint accumulation or insufficient coverage; the nozzle pressure is optimized based on the adhesion level, improving the coating bonding strength and durability; the imaging gain coefficient is compensated for temperature and humidity, ensuring detection accuracy in different environments; thus, real-time coordinated optimization of spraying parameters and imaging parameters is achieved, effectively solving the problem of missed defects caused by parameter solidification in traditional systems, and at the same time, through multi-dimensional parameter linkage optimization, improving spraying quality and detection efficiency.

[0070] 3. By setting multi-dimensional threshold conditions, this solution establishes a dynamic correlation between surface characteristics, environmental parameters, and device control parameters, enabling adaptive parameter adjustment and improving system response speed. By intelligently matching the spray trajectory with surface roughness, this solution avoids paint accumulation on rough surfaces or insufficient coverage on smooth surfaces. A dynamic balance mechanism between adhesion level and nozzle pressure is established, ensuring both coating bond strength and preventing embrittlement defects. Temperature and humidity compensation formulas are used to automatically correct imaging parameters, ensuring the stability of the detection system in different environments.

[0071] 4. This project addresses the issues of unexpected environmental interference and coating defects that cannot be automatically corrected during the spraying process through the design of a specific exception handling mechanism. Through real-time scanning and dynamic path planning, local repairs can be completed without stopping the machine, effectively reducing paint waste and improving spray quality consistency. Furthermore, the grid-based spray pattern allows for fine-tuning of thickness distribution, avoiding the overspray or underspray associated with traditional uniform spraying. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0073] Figure 1 It is a schematic diagram of the overall system architecture of the first embodiment of the present invention.

[0074] Figure 2 It is a schematic diagram of the operation flow of embodiment 1 of the present invention. DETAILED DESCRIPTION

[0075] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0076] It should be noted that the terms "first," "second," "third," "fourth," etc. in the description and claims of the present invention are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having," as well as any variations thereof, in the embodiments of the present invention, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products, or apparatuses.

[0077] Example 1

[0078] like Figure 1 and Figure 2 As shown, the present application proposes an intelligent imaging generation system and method, including a dynamic parameter adjustment module 11, which obtains the spray surface status in real time through multi-sensor fusion, and adjusts the operating parameters of the imaging equipment and the spray robot arm in a linked manner; a defect prediction and optimization decision module 12, which simulates the coating defect morphology under different process parameters based on a generative adversarial network model with physical constraints, and generates a process optimization instruction set through reinforcement learning; a multimodal process knowledge base 13, which stores the material property data of the sprayed device, environmental sensor data, process parameter history data and the correlation between the defect pattern.

[0079] Among them, the dynamic parameter adjustment module refers to a system component that realizes data acquisition and fusion by integrating vision, thickness and environmental sensors. Specifically, it can be implemented by a combination of high-resolution industrial cameras, capacitive thickness gauges and temperature and humidity sensors to eliminate the deviation of single sensor data. The defect prediction and optimization decision module refers to a defect simulation engine built based on a generative adversarial network. Specifically, it can adopt an adversarial neural network structure containing physical constraints and output a defect prediction map by inputting a set of process parameters. The multimodal process knowledge base refers to a database system that stores the relationship between material properties, environmental parameters and defects. Specifically, it can adopt a hybrid architecture of distributed time series database and graph database to realize efficient retrieval and correlation analysis of cross-modal data.

[0080] Specifically, during the spraying process, visual sensors capture surface texture features in real time, thickness detectors simultaneously monitor coating distribution, and environmental sensors collect temperature and humidity data. This data is then spatially and temporally aligned and fed into a parameter adjustment model to drive the coordinated optimization of the robot arm's speed and nozzle pressure. Simultaneously, a generative adversarial network simulates potential defect morphology based on current process conditions, and a reinforcement learning algorithm dynamically adjusts coating viscosity and spray gun distance based on quality indicators. A multimodal knowledge base continuously accumulates data on the material-environment-defect relationship, providing historical case studies to support process optimization.

[0081] Compared to existing technologies, traditional systems use fixed imaging parameters, resulting in an imbalance between detection sensitivity and spray quality. This solution achieves dynamic parameter matching through multi-source data fusion. Existing technologies rely on manual experience to adjust process parameters, while this solution utilizes physically constrained adversarial networks to enable preemptive defect prediction. Existing systems process process data and inspection data in isolation, while this solution builds a cross-domain data association model using a multimodal knowledge base.

[0082] Through the above technical solution, this application achieves real-time coordinated optimization of spraying and imaging parameters, significantly reducing coating rework rates. The defect prediction model proactively identifies sagging risk areas and guides immediate correction of process parameters. Multimodal data correlation analysis effectively improves spraying quality consistency, reducing material waste and equipment idling time.

[0083] This application further proposes a dynamic parameter adjustment module comprising a multi-data acquisition unit and a data fusion unit. The multi-data acquisition unit includes a high-resolution visual sensor, a capacitive thickness detector, and a temperature and humidity sensor. The high-resolution visual sensor captures surface texture features and generates a first data stream; the capacitive thickness detector monitors coating thickness distribution in real time and generates a second data stream; and the temperature and humidity sensor collects environmental parameters and generates a third data stream. The data fusion unit employs a Kalman filter algorithm to perform spatiotemporal alignment and noise suppression on the first, second, and third data streams.

[0084] The core of this application is to build a technical architecture of dynamic parameter adjustment-defect prediction-knowledge base closed-loop optimization, which specifically includes the following modules:

[0085] 1. Dynamic parameter adjustment module 11

[0086] Multi-data acquisition unit 111: integrates high-resolution visual sensor 1111, capacitive thickness detector 1112, temperature and humidity sensor 1113, collects surface roughness Ra, coating thickness μm, ambient temperature and humidity T / RH in real time; and obtains multi-dimensional information of the spraying surface through the synchronous collection of different physical quantities.

[0087] Data fusion unit 112: uses Kalman filter algorithm to align multi-source data;

[0088] Among them, high-resolution visual sensors refer to image acquisition devices with micron-level resolution capabilities, which can be implemented by industrial line array cameras or area array cameras to capture the microscopic texture features of the sprayed surface. Among them, capacitive thickness detectors refer to instruments that measure the thickness of non-conductive coatings through changes in the electric field, which can be implemented by multi-electrode array sensors to obtain spatial distribution data of coating thickness in real time. Among them, temperature and humidity sensors refer to devices that measure ambient temperature and relative humidity, which can be implemented by digital temperature and humidity composite sensors to collect environmental parameters of the spraying operation area. Among them, data fusion units refer to modules that integrate and process multi-source data, which can be implemented by embedded processors equipped with Kalman filtering algorithms to improve data reliability through spatiotemporal coordinate alignment and noise suppression.

[0089] In specific implementation, a high-resolution vision sensor captures images of the workpiece surface at a fixed frame rate and uses an edge detection algorithm to extract surface roughness features, generating a first data stream containing position information. A capacitive thickness detector continuously scans the coating thickness during the robot arm's motion, converting changes in capacitance into thickness distribution data to generate a second data stream. A temperature and humidity sensor collects ambient temperature and humidity parameters in the spraying area at a preset sampling frequency, generating a third data stream. After receiving these three data streams, the data fusion unit uses a Kalman filter algorithm to eliminate timestamp differences between sensors and suppress measurement noise, ultimately outputting a spatiotemporally synchronized multidimensional dataset. This dataset is transmitted to a subsequent decision-making unit for real-time adjustment of the spray robot arm's operating parameters and the imaging device's detection parameters. In this way, through the collaborative work of multiple sensors, this solution synchronously collects three key data types: surface texture, coating thickness, and environmental parameters. Combined with the Kalman filter algorithm, it eliminates time delays and measurement errors between sensors, providing more complete and accurate input data for subsequent parameter adjustments.

[0090] Intelligent Perception and Decision-Making Unit 113: Dynamically adjusts the robot arm speed v, nozzle pressure P, and imaging gain G using a preset formula. This unit includes a robot arm speed adjustment unit 1131, a nozzle pressure optimization adjustment unit 1132, and an imaging device gain coefficient adjustment unit 1133. The robot arm speed adjustment unit extracts the spray surface roughness characteristic value Ra in real time based on the first data stream and dynamically adjusts the robot arm speed v based on a preset robot arm speed formula. The robot arm speed formula is v = k1Ra + b1, where k1 and b1 are roughness-speed response coefficients obtained through experimental calibration. The nozzle pressure optimization adjustment unit optimizes the nozzle pressure P using a preset nozzle pressure optimization formula based on the coating thickness distribution data monitored by the second data stream and the adhesion level F obtained by a mechanical adhesion tester. The nozzle pressure optimization formula is P = k2F + b2, where k2 and b2 are pressure compensation coefficients obtained through experimental calibration. The imaging device gain coefficient adjustment unit combines the third data stream and uses the imaging device gain coefficient compensation formula to compensate the imaging device gain coefficient G; the imaging device gain coefficient compensation formula is G=G0(1+αRH%+β*T℃), where G0 is the reference gain coefficient, RH is the ambient relative humidity, T is the ambient temperature, and α and β are temperature and humidity compensation coefficients. Among them, the surface roughness characteristic value Ra refers to the arithmetic mean deviation value of the microscopic profile of the sprayed surface, which can be specifically achieved by three-dimensional laser scanning or white light interferometer measurement, and is used to quantitatively characterize the flatness of the workpiece surface. The roughness-speed response coefficients k1 and b1 refer to linear relationship parameters calibrated by the spraying experiment, for example, they can be obtained by spraying different roughness samples and measuring the optimal moving speed for fitting, and are used to establish a quantitative mapping of the surface state and the movement of the robotic arm. The adhesion grade F refers to the grading index of the bonding strength between the coating and the substrate, which can be specifically achieved by the grid method or the pull-off method test, and is used to reflect the mechanical bonding performance of the coating. Pressure compensation coefficients k2 and b2 are parameters determined through adhesion testing and nozzle pressure optimization experiments. For example, they can be calibrated by adjusting pressure at different adhesion levels and measuring coating quality to dynamically match spray pressure to coating requirements. Temperature and humidity compensation coefficients α and β are quantitative parameters that quantify the effect of ambient temperature and humidity on imaging device gain. For example, they can be calibrated by simulating different environmental conditions in a temperature and humidity chamber and testing imaging quality to eliminate environmental interference with imaging detection.

[0091] During specific implementation, the robot arm movement speed adjustment unit obtains the texture features of the sprayed surface in real time through a high-resolution visual sensor, extracts the Ra value through an image processing algorithm, and automatically adjusts the movement speed according to a preset linear formula. When the surface roughness increases, the system reduces local paint accumulation by increasing the movement speed; when the surface is smooth, the speed is reduced to ensure uniform coating coverage. The nozzle pressure optimization adjustment unit dynamically calculates the optimal pressure value based on the coating distribution data obtained by the capacitive thickness detector and the adhesion test results. When the adhesion level is low, the system enhances paint penetration by increasing the pressure; when the adhesion is too high, the pressure is reduced to avoid coating cracking. The imaging device gain coefficient adjustment unit obtains environmental parameters through temperature and humidity sensors, and uses a compensation formula to correct the gain setting of the imaging device in real time. For example, in a high temperature and high humidity environment, the system automatically increases the gain coefficient to compensate for changes in the refractive index of light to ensure imaging clarity.

[0092] As mentioned above, since traditional spray parameter adjustment usually relies on fixed empirical values ​​or single sensor feedback, it is impossible to achieve multi-parameter collaborative optimization. Therefore, this solution realizes cross-modal correlation of surface state, coating performance and environmental parameters by establishing a mathematical model of multi-source data fusion, and solves the problems of parameter adjustment lag and single-dimensional optimization. In addition, the speed of the robotic arm of this application is dynamically adjusted according to the surface roughness, avoiding defects caused by paint accumulation or insufficient coverage; the nozzle pressure is optimized based on the adhesion level, which improves the coating bonding strength and durability; the imaging gain coefficient is compensated by temperature and humidity to ensure detection accuracy in different environments; thereby achieving real-time collaborative optimization of spray parameters and imaging parameters, effectively solving the problem of missed defects caused by parameter solidification in traditional systems, and at the same time improving spray quality and detection efficiency through multi-dimensional parameter linkage optimization.

[0093] As a preferred embodiment, the present application further proposes that the dynamic adjustment process of each adjustment unit of the dynamic parameter adjustment module satisfies:

[0094] When the surface roughness Ra exceeds 5μm, the workpiece surface is judged to be relatively rough and automatically switches to a spiral spray trajectory. When the surface roughness Ra exceeds 1μm, the workpiece surface is judged to be relatively smooth and automatically switches to a normal high-speed spray trajectory. When the adhesion level reaches F_1, the adhesion is judged to be too low and the nozzle pressure is automatically increased to enhance the adhesion between the coating and the substrate. When the adhesion level reaches F_5, the adhesion is judged to be too high and the nozzle pressure is automatically reduced to prevent coating embrittlement. When the temperature is T_35°C or the humidity is RH_75%, the working environment is judged to be overheated or overhumid, and the imaging device gain coefficient compensation formula is used to dynamically adjust the gain to correct for environmental interference. The spiral spray trajectory refers to a spraying method that covers the workpiece surface in a spiral path. It can be implemented using a robotic arm motion control system and is used to form a uniform coating on rough surfaces. Nozzle pressure adjustment refers to the dynamic adjustment of the spray pressure based on the adhesion level. It can be implemented using a proportional valve or a servo motor-driven pressure pump to maintain the coating adhesion within a reasonable range. The gain compensation formula refers to a mathematical model that dynamically corrects the gain of the imaging device based on the temperature and humidity parameters. It can be implemented using a linear regression model to eliminate the impact of environmental interference on imaging quality. Specifically, when the surface roughness Ra exceeds the threshold, the system automatically switches the spray trajectory according to the preset conditions. For example, a spiral trajectory on a rough surface can improve the uniformity of paint coverage, while a high-speed trajectory on a smooth surface can improve work efficiency. When the adhesion level is lower than the preset lower limit, the nozzle pressure is increased to enhance the permeability of the paint. When the adhesion level exceeds the upper limit, the pressure is reduced to avoid cracking of the coating. When it is detected that the temperature and humidity exceed the safe range, the imaging parameters are adjusted using the gain coefficient compensation formula to reduce the interference of environmental factors on the detection accuracy. This adjustment process triggers the corresponding control instructions by comparing the real-time sensor data with the preset threshold, forming a closed-loop feedback mechanism.

[0095] As mentioned above, traditional systems use fixed spray trajectories and pressure parameters, which cannot adapt to different surface conditions and environmental changes, resulting in fluctuations in coating quality. This case establishes a dynamic association between surface characteristics, environmental parameters, and equipment control parameters by setting multi-dimensional threshold conditions, achieving adaptive adjustment of parameters and improving system response speed. This case avoids the problem of paint accumulation on rough surfaces or insufficient coverage on smooth surfaces through intelligent matching of spray trajectories and surface roughness; establishes a dynamic balance mechanism between adhesion level and nozzle pressure, which not only ensures coating bonding strength but also prevents brittle defects; and automatically corrects imaging parameters through temperature and humidity compensation formulas to ensure the stability of the detection system in different environments.

[0096] As a preferred embodiment, the present application further proposes that the dynamic parameter adjustment module also includes an exception handling mechanism: when the instantaneous wind speed is detected to be greater than a preset threshold, such as wind speeds greater than 2m / s, spraying is paused and 3D laser scanning is initiated to reconstruct the surface topography; a localized re-spraying path is generated based on point cloud data; and when the coating thickness unevenness exceeds a set threshold, such as greater than 1%, an adaptive gridded spraying mode is triggered. The adaptive gridded spraying mode divides the spraying area into multiple sub-areas and independently adjusts parameters. For example, dynamic gridding technology can be used to automatically adjust the spraying speed and spray gun angle within each grid based on thickness distribution differences. Instantaneous wind speed detection refers to real-time monitoring of airflow disturbances in the spraying area using a wind pressure sensor. For example, a digital anemometer can be used to collect data at a sampling rate of at least twice per second. This detection provides a trigger condition for exception handling, preventing airflow interference from causing paint dispersion. 3D laser scanning to reconstruct surface topography refers to acquiring 3D coordinate data of the workpiece surface using a line laser scanner. For example, a blue laser can be used in conjunction with a high-speed CMOS sensor to generate a millimeter-level precision point cloud model for precise identification of areas requiring re-spraying. The local re-spraying path refers to the spraying trajectory generated based on the surface morphology characteristics. For example, a spiral or zigzag trajectory can be generated in the re-spraying area through a path planning algorithm to ensure the integrity of the coating coverage. In specific implementation, when the wind speed is detected to be greater than 2m / s, the system in this case immediately suspends the spraying operation to prevent the paint from flying. Then the three-dimensional laser scanning device is started, for example, a rotating line laser scanning head is used to reconstruct the current surface morphology with a resolution of 0.1 mm. Based on the point cloud data obtained by scanning, for example, the coating defect area is identified through a point cloud density analysis algorithm, a local re-spraying path is generated and the robotic arm is controlled to perform precise re-spraying. When it is detected that the coating thickness difference exceeds 1%, the system divides the spraying area into multiple grid units, for example, the octree space segmentation algorithm is used to dynamically generate a grid, and the spraying parameters are independently adjusted according to the measured thickness value in each grid, such as increasing the paint flow rate in areas with insufficient thickness and reducing the spray gun movement speed in areas with excessive thickness.

[0097] As mentioned above, traditional spray systems can only issue alarms or shut down for repairs when abnormal wind speeds or uneven thickness occur. However, this case achieves autonomous repair through three-dimensional scanning reconstruction and dynamic path planning, avoiding production line stagnation caused by manual intervention. At the same time, this case generates an adaptive path based on real-time point cloud data, which can accurately cover the defective area. In addition, this case solves the problem of sudden environmental interference and coating defects that cannot be automatically corrected during the spraying process through the design of a specific exception handling mechanism. Through real-time scanning and dynamic path planning, local repairs can be completed without stopping the machine, effectively reducing paint waste and improving the consistency of spraying quality. At the same time, the grid spraying mode can make fine adjustments based on differences in thickness distribution, avoiding the phenomenon of overspray or underspray of materials caused by traditional uniform spraying.

[0098] As a preferred embodiment, the present application further proposes that the defect prediction and optimization decision module 12 includes a defect simulation unit 121, a reinforcement learning optimization unit 122 and a closed-loop control unit 123. Among them, the defect simulation unit 121 generates a coating defect prediction map under different process parameters based on a physical constraint adversarial network model, and the adversarial network model is a GAN model; the generator input of the GAN model includes a set of process parameters (such as spraying speed v, nozzle pressure P), material property data (such as surface energy γ=35mN / m, ρ coating density, coating viscosity η, coating surface tension σ, coating thickness h), environmental data (such as ambient temperature T, humidity RH), substrate state parameters (such as roughness Ra), and the discriminator outputs a coating defect prediction map under different process parameters.

[0099] The discriminator of the GAN model integrates the physical equation constraints of coating defects, including:

[0100] Sagging defect length Where ρ is the density of the paint, g is the acceleration due to gravity, which is a constant of approximately 9.81 m / s 2, h is the coating thickness, η is the coating viscosity, v is the spraying speed, and C is an empirical coefficient used to adjust the accuracy of the formula; σ is the surface tension of the coating, representing the attraction between molecules on the coating surface. In specific implementation, the discriminator is responsible for determining the difference between the predicted patterns output by the generator and the actual defect patterns. Through continuous adversarial training, the accuracy of the generator's predicted patterns is improved. During the training of the GAN model, physical constraints (i.e., the constraints of the coating defect physics equations mentioned above) are introduced to ensure that the generated coating defect prediction patterns conform to the physical laws of coating defect formation. The generator and discriminator are then continuously optimized through adversarial training. For example, the generator strives to generate more realistic prediction patterns to deceive the discriminator, while the discriminator strives to improve its accuracy in distinguishing between real and generated patterns. After adversarial training, the generator of the GAN model is able to generate accurate coating defect prediction patterns based on multi-source input data. These patterns intuitively demonstrate the types and distribution of coating defects that may occur under different process parameters, providing an important basis for optimizing the spraying process. For example, when the system detects a decrease in paint viscosity, the model can predict the potential for a sag defect and provide the probable sag length. These predictions can guide the system to adjust process parameters in a timely manner, such as reducing spray speed or increasing paint viscosity, to prevent defects. Sag defects are caused by the downward flow of paint due to gravity during the drying process, and their formation mechanism can be explained by fluid dynamics equations. The sag defect length formula describes the quantitative relationship between sag defect length and paint viscosity (η), spray speed (v), paint density (ρ), gravitational acceleration (g), and coating thickness (h). For example, high-viscosity paint (η↑) or high spray speed (v↑) increases intermolecular cohesion and suppresses sag (L↓); whereas thin coatings (h↓) or low-density paints (ρ↓) are more susceptible to sag due to increased gravity. By embedding this physical equation into the GAN model's discriminator, the generator is constrained to output a defect map that conforms to the laws of fluid dynamics, ensuring that the predictions align with actual physical processes.

[0101] The reinforcement learning optimization unit (122) uses the real-time spraying quality index as the state space and the process parameter adjustment amount as the action space to optimize the process parameters through the reward function. In specific implementation, the state space is the set of all environmental state parameters that the system needs to observe during the decision-making process, which is used to characterize the real-time quality state of the spraying process. The state space includes the following real-time spraying quality indicators: thickness uniformity, adhesion level, and surface gloss. The action space is the process parameter adjustment amount (speed change Δv, pressure change ΔP); wherein the speed change Δv is the adjustment range of the spraying robot arm's movement speed (such as ±0.2m / s); the pressure change ΔP is the adjustment range of the nozzle pressure (such as ±5kPa). The reward function is defined as R = w1·(1-(|D_actual-D_target| / D_target))+w2·C_saved / C_total, where D_actual is the actual defect rate, D_target is the target defect rate, C_saved is the actual saved paint cost, C_total is the total paint cost budget, w1 and w2 are weight coefficients, satisfying w1+w2=1. Among them, the state space of the reinforcement learning optimization unit refers to the quantification of the spraying quality indicators into multi-dimensional vectors, including the standard deviation of coating thickness, adhesion level and surface roughness. Specifically, the normalized sensor data can be spliced ​​into a state vector to achieve a comprehensive evaluation of complex quality indicators. The action space refers to the set of discrete operation instructions for the process parameter adjustment amount. Specifically, the adjustment step size of the spraying speed, nozzle pressure and gun distance parameters can be used as the action dimension. The weight coefficient in the reward function refers to the balance factor between quality priority and cost priority. Specifically, the ratio of w1 and w2 can be dynamically adjusted. When the system detects a critical workpiece, the value of w1 is automatically increased to achieve multi-objective optimization decision-making.

[0102] The closed-loop control unit 123 sends the optimized process parameters to the spraying robot and imaging equipment in real time, and receives new sensor data for iterative optimization. Specifically, the defect simulation unit receives the process parameter set and environmental data from multi-source sensors, and inputs it together with the material property data into the generator of the GAN model. The generator outputs a microscopic defect distribution map of the coating surface under physical constraints, and the discriminator identifies the difference between the predicted results and the real defects through adversarial training. The reinforcement learning optimization unit constructs a state vector based on the current spraying quality index, selects the process parameter adjustment instruction in the action space, obtains the reward value after execution and updates it. The closed-loop control unit sends the optimized spraying speed and nozzle pressure parameters to the robot controller, and adjusts the gain coefficient of the imaging device to adapt to environmental changes. After the preset period, the system collects new coating thickness and surface morphology data, recalculates the reward function value and triggers the model parameter update to form an iterative optimization loop.

[0103] As mentioned above, traditional spraying systems can only adjust process parameters through fixed rules and cannot simulate the nonlinear relationship between complex defect morphology and process parameters. Existing defect detection methods rely on offline analysis and lack a real-time feedback mechanism, resulting in a lag in parameter adjustment. In this case, the GAN model with physical constraints is used to accurately predict defect morphology, and the dynamic exploration capability of reinforcement learning is combined to optimize the multi-parameter collaborative adjustment strategy. At the same time, closed-loop control is used to achieve real-time linkage between process parameters and imaging parameters, effectively solving the problems of parameter solidification and feedback delay. In this way, the defect prediction and optimization decision module 12 in this case can accurately predict the distribution of coating defects under different process conditions through the settings of its various units, dynamically generate optimization instructions to balance quality and cost goals, and realize adaptive control of the spraying process. Moreover, the prediction model is continuously corrected through real-time data closed-loop feedback, which improves the collaborative efficiency of defect identification and process adjustment and reduces the need for manual intervention.

[0104] As a preferred embodiment, the present application further proposes a mode selection module 14 for dynamically selecting a quality priority mode and a cost priority mode according to the workpiece production type;

[0105] Among them, when the system detects the production of critical workpieces, it automatically selects the quality priority mode and limits w_1>w_2; the system will call the preset process optimization instruction set to reduce defects.

[0106] When the system detects a non-critical workpiece, it automatically selects the cost priority mode and limits w_2>

[0107] w_1; the system will invoke a preset process optimization instruction set to ensure cost. Quality-first mode refers to an operating state in which the system prioritizes coating quality by adjusting weight coefficients. This can be achieved through preset weight thresholds, for example, setting the quality weight coefficient to 0.7 and the cost weight coefficient to 0.3. This can be achieved through weight adjustment of the reward function within the reinforcement learning unit, which is used to prioritize different optimization objectives within the objective function. The process optimization instruction set refers to a set of collaborative control instructions that includes a paint viscosity adjustment value, a spray speed threshold, and a spray gun distance parameter. This can be achieved through the output of a multi-dimensional parameter combination from a physically constrained generative adversarial network model. In specific implementation, the mode selection module uses the workpiece type recognition unit to determine the attribute label of the currently produced workpiece. When a critical workpiece label is identified, the quality-first mode is activated, forcing the quality weight coefficient to be higher than the cost weight coefficient. In this case, the reinforcement learning optimization unit prioritizes reducing the defect rate when generating the paint viscosity adjustment value, and the spray speed threshold is set to the lower limit of the allowable range to extend the effective spraying time. When a non-critical workpiece label is identified, the cost priority mode is activated and the cost weight coefficient is forced to be higher than the quality weight coefficient. At this time, the nozzle pressure parameter will be calculated based on the paint consumption optimization formula, and the spray speed threshold will be set to the upper limit of the allowable range to shorten the production cycle.

[0108] As described above, this case achieves dynamic weight allocation through a mode selection mechanism. In critical workpiece scenarios, defect suppression is enhanced by increasing the quality weight coefficient, while in non-critical workpiece scenarios, resource conservation is achieved by increasing the cost weight coefficient. This resolves the imbalance between inspection accuracy and cost control caused by parameter rigidity, enabling the automatic matching of optimal optimization strategies based on workpiece attributes, while ensuring the quality of critical workpieces while reducing the production costs of non-critical workpieces. When producing high-value workpieces, the quality-first mode is used to reduce quality risks caused by coating defects; when producing ordinary workpieces, the cost-first mode is used to reduce coating consumption and production time, achieving a dynamic balance between quality and cost.

[0109] As a preferred embodiment, the multimodal process knowledge base 13 includes:

[0110] The material property storage unit 131 is used to store the material property data of the sprayed device. The material property data includes at least: substrate type (metal / plastic / composite material) and its physical properties (density, thermal conductivity, surface energy); coating properties (viscosity, solid content, surface tension coefficient, curing reaction activation energy); substrate physical properties are obtained through material testing equipment (such as surface tension meter, thermal conductivity meter), and coating properties are measured by rheometer and gas chromatograph.

[0111] Environmental data storage unit 132, for storing environmental data, the environmental data including at least: temperature, humidity parameters, wind speed parameters;

[0112] The process parameter storage unit 133 is used to store process parameters, which include at least nozzle pressure and spraying speed parameters;

[0113] An image data storage unit 134 is used to store optical detection images of defective parts, wherein the optical detection images include at least microscopic photos of orange peel and sagging;

[0114] A text record storage unit 135 is used to store operation logs, maintenance reports, and quality inspection notes during the production process;

[0115] The defect case storage unit 136 is used to store the operating condition characteristic data packets of historical defect cases, wherein the operating condition characteristic data includes process parameter combinations and defect information; the defect information includes at least defect type, defect severity, and defect repair measures;

[0116] The defect mode association storage unit 137 is used to store the association between material properties, environmental data and defect modes, including:

[0117] a) Mapping relationship between material type and environmental parameters to defect type;

[0118] b) Quantitative correlation model between environmental parameters and defect morphology;

[0119] c) Correlation model between substrate surface characteristics and defect spatial distribution.

[0120] As described above, in this case, the material property storage unit 131 integrates key material property data, such as substrate type and its physical parameters, and coating properties, providing comprehensive basic material information for spray process optimization. This data is acquired using specialized testing equipment, ensuring its accuracy and reliability. The environmental data storage unit 132 records environmental parameters during the spraying process, such as temperature, humidity, and wind speed. These parameters have a significant impact on spraying quality and facilitate analysis of the impact of environmental factors on spraying results. The process parameter storage unit 133 stores key process parameters during the spraying process, such as nozzle pressure and spray speed, which are directly targeted for spray process optimization. The image data storage unit 134 stores optical inspection images of defective areas, such as microphotographs of orange peel and sagging. These images visually demonstrate defects that occur during the spraying process, helping technicians quickly identify and analyze problems. The text record storage unit 135 records operation logs, maintenance reports, and quality inspection notes during the production process. This textual information provides detailed background information for defect analysis and resolution. The defect case storage unit 136 stores operating condition characteristic data packages for historical defect cases, including process parameter combinations and defect information. These cases provide valuable experience references for technicians, helping to quickly locate and solve similar problems. The defect pattern association storage unit 137 stores the association between material properties, environmental data, and defect patterns, including the mapping relationship between material type and environmental parameters to defect type, the quantitative association model between environmental parameters and defect morphology, and the association model between substrate surface characteristics and defect spatial distribution. These associations help to deeply understand the formation mechanism of defects in the spraying process and provide a scientific basis for process optimization. In this way, comprehensive data integration is achieved, which helps the system automatically adjust the spraying process parameters based on real-time data and historical experience, and achieve continuous optimization of spraying quality.

[0121] As a preferred embodiment, the present application further proposes that the multimodal process knowledge base is updated in the following manner: when the prediction deviation of a new defect case is greater than 15%, incremental learning is triggered to update the associated graph weights; a time decay factor is applied to historical data of more than 90 days; and the correlation between the coating porosity distribution and the process parameters is stored.

[0122] Among them, incremental learning refers to updating the weight parameters of the association map through incremental model training. Specifically, it can be achieved by using an online learning algorithm combined with historical data to fine-tune the model parameters, which is used to dynamically adapt to changes in new defect patterns and solve the problem of prediction deviation of the old model for new working conditions. The time decay factor refers to the coefficient that exponentially decays the importance of historical data. Specifically, the decay weight can be calculated using the formula where the decay factor lambda = 0.98^{t / 30})(t is the number of days) to reduce the interference of outdated data on the current model and maintain the timeliness of the knowledge base. The correlation between the coating porosity distribution and process parameters refers to the mapping relationship between the microstructure data obtained by high-resolution μCT scanning and the spraying parameters. Specifically, three-dimensional image processing technology can be used to extract the porosity distribution characteristic values ​​and establish a regression model to optimize the spraying parameters to reduce porosity defects. In specific implementation, when it is detected that the prediction deviation of a new defect case exceeds a threshold, the system automatically starts the incremental learning process, such as using an online gradient descent algorithm to adjust the association weight between material properties and defect types, and synchronously updates the quantitative association model in the knowledge base. For historical data stored for more than three months, a time decay factor is used to reduce its contribution to model training. For example, the weight of coating thickness data from 90 days ago is reduced to less than 50% of the initial value, thereby prioritizing learning from recent operating data. In addition, by storing the corresponding relationship between high-resolution μCT scan data and spray parameters, for example, associating areas with porosity exceeding 5% with process parameters with spray gun movement speeds below 0.8m / s, this provides a physical basis for subsequent spray parameter optimization.

[0123] As mentioned above, traditional knowledge bases usually adopt a static storage mode, which cannot dynamically adjust the association model according to new defect data. In addition, the fixed weight distribution of historical data limits the generalization ability of the model. This case realizes the continuous evolution of the knowledge base through an incremental learning mechanism, combines the time decay factor to eliminate the interference of outdated data, and introduces microstructure data to enhance the physical interpretability of process parameter optimization, significantly improving the collaborative optimization ability of defect prediction and parameter adjustment. In addition, this application realizes the dynamic optimization of the knowledge base and cross-scale data fusion, solving the problem of process parameter mismatch caused by data update lag in traditional systems.

[0124] Example 2

[0125] This case proposes an intelligent imaging generation method, including the following steps: S1. Real-time data acquisition: synchronously acquiring spray surface texture characteristics, coating thickness distribution, and ambient temperature and humidity data through multiple sensors;

[0126] S2. Dynamic parameter coordination adjustment: Dynamically calculate the robot arm movement speed v based on the surface roughness Ra, optimize the nozzle pressure P based on the adhesion level F, and adjust the imaging device gain coefficient G based on the temperature and humidity compensation formula;

[0127] S3. Defect simulation prediction: Input the process parameter set into the physically constrained GAN model to generate a coating defect morphology prediction map and identify sagging defect risk areas;

[0128] S4. Process optimization decision-making: Select weight coefficients based on the quality / cost model and iteratively generate a collaborative optimization instruction set for paint viscosity, spray speed, and gun distance through reinforcement learning;

[0129] S5. Closed-loop execution and feedback: Optimization instructions are sent to the spray robot and imaging equipment. The reward function is updated based on newly collected coating quality data, triggering incremental learning to optimize the knowledge base association model.

[0130] S6. Abnormal adaptive processing: When the wind speed is detected to be greater than 2m / s or the thickness unevenness is greater than 1%, the spraying is suspended and the local supplementary spraying path planning is started, and the exposure parameters of the imaging equipment are corrected synchronously. Abnormal adaptive processing refers to triggering an adaptive control strategy through real-time monitoring of environmental parameters and coating status. Specifically, three-dimensional laser scanning combined with point cloud data processing algorithms can be used to generate local supplementary spraying paths. For example, the optimal supplementary spraying trajectory is calculated based on a dynamic path planning model. The exposure parameter correction of the imaging equipment can be achieved through an environmental interference compensation algorithm. For example, a multivariate regression model is used to establish a correlation between wind speed and exposure time. The local supplementary spraying path planning includes two stages: surface morphology reconstruction and spraying parameter re-optimization. Specifically, a grid-based spraying parameter adjustment module can be used to achieve differential spraying in different areas. Synchronous correction refers to the coordinated issuance of imaging equipment parameter adjustments and spraying robot arm control instructions. Specifically, real-time communication between devices can be achieved through an industrial bus protocol.

[0131] Specifically, during the implementation process, a visual sensor captures surface texture images, while a capacitive thickness gauge acquires coating distribution data. A temperature and humidity sensor collects environmental parameters and transmits them to the central processing unit. Surface roughness characteristic values ​​are input into a preset speed formula to calculate the robot arm's movement speed, while nozzle pressure parameters are adjusted based on real-time adhesion test results. The imaging device's gain coefficient automatically corrects for environmental interference using a compensation formula, for example, automatically increasing the gain to compensate for thermal noise in high-temperature environments. The process parameter set is input into a generative adversarial network to predict defect morphology. The prediction results trigger the reinforcement learning module to generate optimization instructions. The closed-loop control module simultaneously transmits the optimized spray speed and coating viscosity parameters to the actuator and feeds newly collected coating quality data back into the knowledge base to update the association model. Furthermore, when sudden environmental disturbances are detected, emergency response procedures can be initiated. For example, if wind speed exceeds the standard, spraying can be suspended and 3D scanning can be initiated to reconstruct the surface topography, generating a localized re-spraying path based on point cloud data. The imaging device's exposure parameters can be dynamically adjusted based on environmental changes, for example, automatically switching to an optical filter mode to eliminate water mist interference when humidity suddenly changes. Knowledge base updates can use a time decay mechanism to process historical data, such as applying an exponential decay weight to process data that exceeds the storage period.

[0132] As mentioned above, the method in this case overcomes the technical defect of manual detection lag through real-time data collection and closed-loop feedback mechanism. The multimodal correlation model established effectively integrates process parameters and material property data, improving the prediction accuracy of coating defects and the response speed of process adjustments. The introduction of the exception handling mechanism enhances the robustness of the system under complex working conditions and ensures the stability of quality control in the continuous production process. Through the abnormal adaptive processing steps, it is convenient to establish a closed-loop control mechanism for abnormal response and parameter correction, which effectively solves the problem of uncontrollable coating quality under abnormal spraying conditions, reduces material waste and improves repair efficiency through local re-spraying path planning. The dynamic correction of the exposure parameters of the imaging equipment avoids image distortion caused by environmental interference and provides reliable data support for online quality detection. The coordinated optimization of process parameters and imaging parameters enhances the system's adaptability to complex working conditions and realizes full-process quality control of the spraying process.

[0133] Specifically, in the first embodiment, the car body painting is used as an application. The system performs topcoat spraying on the car door. The substrate is a galvanized steel plate (surface energy γ = 35mN / m), and the paint is a water-based polyurethane (paint viscosity η = 250mPa·s, paint surface tension σ = 30mN / m, paint density ρ = ρ = 1.2g / cm 3 ), ambient temperature T = 28°C, ambient humidity RH = 65%, and instantaneous wind speed < 1m / s. The baseline environment is: T = 25°C, RH = 50%. The system must achieve real-time coordinated optimization of spraying and imaging parameters to ensure coating quality consistency and reduce defect rates.

[0134] The system operation process is as follows:

[0135] 1. System initialization and parameter configuration

[0136] Artifact type identification

[0137] By identifying the current workpiece as a "car door" through RFID or QR code scanning, the system automatically loads the material property data of the galvanized steel sheet substrate and the water-based polyurethane coating.

[0138] The mode selection module (14) determines that the mode is quality priority (because the car door is a key workpiece), and sets the reward function weights: w1 = 0.7 (quality priority), w2 = 0.3 (cost secondary).

[0139] 2. Data collection and fusion:

[0140] The high-resolution visual sensor captures the surface roughness of the galvanized steel sheet, Ra = 2.5 μm, generating the first data stream; the capacitive thickness detector measures the average thickness, h = 85 μm, generating the second data stream; the temperature and humidity sensor collects environmental data, T = 28 ± 0.5 ° C, RH = 65 ± 3%, generating the third data stream. The data fusion unit (112) uses the Kalman filter algorithm to align the timestamps and suppress the noise of the three data streams, and outputs a spatiotemporally synchronized multidimensional data set:

[0141] 3. Dynamic parameter adjustment:

[0142] Using the robot arm movement speed formula v = k1 * Ra + b1, with k1 set to 0.2 and b1 set to 0.5, the robot arm speed is calculated to be v = 0.2 * 2.5 + 0.5 = 1.0 m / s. At this time, a spiral spray trajectory is used (because Ra = 2.5 μm is between 1 and 5 μm).

[0143] The adhesion level F=3 obtained by the mechanical adhesion tester is optimized using the nozzle pressure formula P=k2*F+b2; k2 is 5, b2 is 20, P=5*3+20=35kPa;

[0144] The imaging device gain compensation formula is: G = G0 * (1 + α * RH% + β * T°C); G0 is 1.0, α is 0.5, and β is 0.3, with G0 being the baseline gain coefficient; RH is the ambient relative humidity, and T is the ambient temperature; α and β are the temperature and humidity compensation coefficients. The parameters α = 0.5 ( / %) and β = 0.3 ( / °C) mean that every 1% change in humidity results in a 0.5% change in gain, and every 1°C change in temperature results in a 0.3% change in gain. G = G0 * (1 + 0.5 * 65% * 0.01 + 0.3 * 28°C * 0.01) = 1.409 G0 = 1.409.

[0145] 4. Defect simulation prediction and process optimization

[0146] 4.1 Defect Simulation Unit (121)

[0147] The process parameters (v = 1.0 m / s, P = 35 kPa), material properties (surface roughness Ra = 2.5 μm, coating viscosity η = 250 mPa·s, coating surface tension σ = 30 mN / m, coating density ρ = ρ = 1.2 g / cm 3 , coating thickness h = 85um), and environmental data (T = 28 ° C, RH = 65%) are input into the physical constraint GAN model.

[0148] The generator outputs a coating defect prediction map, and the discriminator identifies potential defects:

[0149] Predict the sagging defect risk area (assuming C = 1. In practical applications, the value of C needs to be determined through experiments; Sagging defect length L = 3.4 mm)

[0150] The GAN model predicts that the probability of flow-on risk under the current parameters is 12%;

[0151] After adjustment, the actual defect rate dropped to 0.4%, and the paint saving was 18%.

[0152] 4.2 Reinforcement Learning Optimization Unit (122)

[0153] Build state space: thickness uniformity = 95%, adhesion level = level 3.

[0154] Select optimization instructions in the action space:

[0155] Reduce the spraying speed to 0.9m / s (Δv=-0.1m / s)

[0156] Increase the coating viscosity to 280mPa·s (by adding thickener)

[0157] 4.3 Closed-loop control unit (123)

[0158] The optimization instructions are sent to the spray robot arm and imaging equipment, and the knowledge base association model is updated synchronously.

[0159] 5. Adaptive exception handling

[0160] 5.1 Wind Speed ​​Anomaly Detection

[0161] Real-time wind speed = 1.2 m / s (not exceeding the threshold of 2 m / s), and the system is operating normally.

[0162] 5.2 Thickness unevenness detection

[0163] Thickness non-uniformity = 4.0% (<1% threshold), no need to trigger grid spray mode.

[0164] 6. Closed-loop feedback and knowledge base updates

[0165] 6.1 New Data Collection and Reward Function Update

[0166] Quality of new coating: thickness uniformity = 96%, adhesion grade = level 4, defect rate = 0.35%.

[0167] Reward function value:

[0168] R=0.7*(1-(∣0.35-0.3∣ / 0.3))+0.3*(5 / 10)=0.7*0.833+0.3*0.05=0.583+0.015=0.598

[0169] 6.2 Incremental Learning of Knowledge Base

[0170] Prediction deviation = |0.35-0.3| / 0.3 = 16.7% > 15%, triggering incremental learning and updating the material-defect association weight.

[0171] 7. Spraying completion and quality verification

[0172] 7.1 Spraying effect evaluation

[0173] Final coating thickness: 85±3μm (uniformity improved)

[0174] Defect rate: 0.2% (below the target of 0.5%)

[0175] 7.2 Data Archiving

[0176] Store process parameters, environmental data and defect cases in a multimodal process knowledge base (13).

[0177] Parameter calibration process description:

[0178] 1. The process of obtaining k1 and b1 in the robot arm speed formula:

[0179] In the robot arm speed formula, k1 and b1 are the roughness-speed response coefficients obtained through experimental calibration: This was achieved by selecting galvanized steel sheets with different surface roughness (Ra values) as the spray substrate in the experiment. Galvanized steel sheets are widely used in industries such as automobiles and construction, and their surface characteristics have a direct impact on spray quality (such as coating adhesion and uniformity). Through mechanical processing or surface treatment, multiple samples with roughness ranging from 1μm to 1μm were prepared to cover common situations in actual production. The optimal movement speed was then measured, and the least squares method was used to fit the linear relationship.

[0180] Table 1: Experimental data for calibration of robotic arm speed formula parameters

[0181]

[0182] As shown in the table above, the optimal robot arm speed was experimentally measured by spraying galvanized steel samples with different roughness. The least squares method was used to fit the linear formula, and the fitting degree R 2 =0.97, the error is controlled within ±3%. 2 =0.97 indicates that the model can explain 97% of the data variation, indicating that the linear relationship between roughness and speed is highly reliable and can effectively guide parameter adjustment in actual production.

[0183] 2. Adjust the nozzle pressure at different adhesion levels (F = 1 to 5), test the coating bonding strength by the cross-hatch method, and determine the optimal pressure value.

[0184] Table 2: Experimental data for nozzle pressure formula parameter calibration

[0185] Adhesion grade The optimal pressure P was measured experimentally. Fitting formula error 1 24.8 25 -0.2 2 30.2 30 +0.2 3 34.9 35 -0.1 4 40.3 40 +0.3 5 44.7 45 -0.3

[0186] As shown in the table above, the maximum error between the measured pressure and the theoretical value is ±0.3 kPa, which meets the industry standard (±0.5 kPa). The coating adhesion level at different pressures was tested using the cross-hatch method to determine the formula parameters (k2 = 5, b2 = 20).

[0187] The above is a detailed introduction to an intelligent imaging generation system and method disclosed in an embodiment of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. An intelligent imaging generation system, characterized in that: include: A dynamic parameter adjustment module (11) acquires the spray surface status in real time through multi-sensor fusion and adjusts the operating parameters of the imaging device and the spray robot arm in a coordinated manner; A defect prediction and optimization decision module (12) simulates coating defect morphologies under different process parameters using a generative adversarial network model based on physical constraints, and generates a process optimization instruction set through reinforcement learning. The process optimization instruction set includes collaborative optimization instructions for a coating viscosity adjustment value, a spraying speed threshold, and a spray gun distance parameter; The multimodal process knowledge base (13) stores the relationship between the material characteristic data, environmental sensing data, process parameter history data and defect patterns of the sprayed device.

2. The system according to claim 1, wherein: The dynamic parameter adjustment module (11) comprises: A multi-data acquisition unit (111), comprising: A high-resolution vision sensor (1111) for capturing surface texture features and generating a first data stream; A capacitive thickness detector (1112) monitors the coating thickness distribution in real time and generates a second data stream; A temperature and humidity sensor (1113) collects environmental parameters and generates a third data stream; The data fusion unit (112) uses a Kalman filter algorithm to perform spatiotemporal alignment and noise suppression on the first, second and third data streams.

3. The system according to claim 2, characterized in that The dynamic parameter adjustment module (11) further includes: Intelligent perception and decision-making unit (113), comprising: A robot arm movement speed adjustment unit (1131) is used to extract the spray surface roughness characteristic value Ra in real time in combination with the first data stream, and dynamically adjust the robot arm movement speed v according to a preset robot arm movement speed formula; wherein the robot arm movement speed formula is: v=k1*Ra+b1; k1 and b1 are roughness-speed response coefficients obtained through experimental calibration; Ra ranges from 0.1 to 10 μm; A nozzle pressure optimization and adjustment unit (1132) is configured to optimize the nozzle pressure P using a preset nozzle pressure optimization formula based on the coating thickness distribution data monitored by the second data stream and the adhesion level F obtained by the mechanical adhesion tester; wherein the nozzle pressure optimization formula is: P=k2*F+b2; k2 and b2 are pressure compensation coefficients calibrated by the experiment; and the adhesion level F is divided into levels 1-5; An imaging device gain coefficient adjustment unit (1133) is used to compensate the imaging device gain coefficient G by combining a third data source and using an imaging device gain coefficient compensation formula; the imaging device gain coefficient compensation formula is: G=G0*(1+α*RH%+β*T℃); G0 is a reference gain coefficient; RH is an ambient relative humidity; T is an ambient temperature; α and β are temperature and humidity compensation coefficients, α is in 1 / %, and β is in 1 / ℃.

4. The system according to claim 3, characterized in that The dynamic adjustment process of each adjustment unit of the dynamic parameter adjustment module (11) satisfies: When the surface roughness Ra>5μm, the workpiece surface is judged to be relatively rough and the spiral spraying trajectory is automatically switched; When the surface roughness Ra is less than 1μm, the workpiece surface is judged to be relatively smooth and the normal high-speed spraying trajectory is automatically switched; When the adhesion level F is less than level 1, the adhesion is judged to be too low and the nozzle pressure is automatically increased to enhance the bite between the coating and the substrate; When the adhesion level F>5, the adhesion is judged to be too high and the nozzle pressure is automatically reduced to avoid coating embrittlement; When the temperature T>35℃ or the humidity RH>75%, the working environment is judged to be overheated or overhumid, and the gain is dynamically adjusted through the imaging device gain coefficient compensation formula to correct the environmental interference.

5. The system according to claim 3, wherein: The dynamic parameter adjustment module (11) also includes an exception handling mechanism: When the instantaneous wind speed is detected to be greater than the preset threshold, the following operations are executed: Pause spraying and start 3D laser scanning to reconstruct the surface topography; Generate local spraying path based on point cloud data; When the coating thickness unevenness is greater than 1%, the adaptive grid spraying mode is triggered.

6. The system according to claim 1, wherein: The defect prediction and optimization decision module (12) includes: A defect simulation unit (121) generates a coating defect prediction map under different process parameters based on a physical constraint adversarial network model; the adversarial network model is a GAN model; The generator input of the GAN model includes a set of process parameters, material property data, environmental data and substrate state parameters, so as to output a coating defect prediction map under different process parameters through its discriminator; the discriminator of the GAN model integrates the physical equation constraints of the coating defect, including: sag length Where ρ is the density of the coating, g is the acceleration of gravity, h is the coating thickness, η is the viscosity of the coating, v is the spraying speed, C is an empirical coefficient, and σ is the surface tension of the coating; A reinforcement learning optimization unit (122) uses the real-time spraying quality index as the state space and the process parameter adjustment amount as the action space to optimize the process parameters through a reward function; The closed-loop control unit (123) sends the optimized process parameters to the spraying robot (301) and the imaging device (302) in real time, and receives new sensor data for iterative optimization; Among them, the reward function is defined as: R = w1·(1-(|D_{actual}-D_{target}| / D_{target}))+w2·C_{saved} / C_{total}; D_{actual} is the actual defect rate, D_{target} is the target defect rate, C_{saved} is the actual saved paint cost, C_{total} is the total paint cost budget, w1 and w2 are weight coefficients, satisfying w1+w2=1.

7. The system according to claim 6, characterized in that Also includes: A mode selection module (14) is used to dynamically select a quality priority mode and a cost priority mode according to the type of workpiece production; Among them, when the system detects the production of critical workpieces, it automatically selects the quality priority mode and limits w_1>w_2; the system will call the preset process optimization instruction set to reduce defects. When the system detects a non-critical workpiece, it automatically selects the cost priority mode and limits w_2>w_1; the system will call the preset process optimization instruction set to ensure cost.

8. The system according to claim 1, wherein: The multimodal process knowledge base (13) includes: A material property storage unit (131) is used to store material property data of the sprayed device, the material property data including at least: substrate type and its physical property parameters, and coating properties; An environmental data storage unit (132) is used to store environmental data, the environmental data including at least: temperature, humidity parameters, and wind speed parameters; A process parameter storage unit (133) is used to store process parameters, which include at least nozzle pressure and spraying speed parameters; An image data storage unit (134) is used to store an optical detection image of a defective part, the optical detection image at least including a microscopic photograph of sagging; A text record storage unit (135) for storing operation logs, maintenance reports, and quality inspection notes during the production process; A defect case storage unit (136) is used to store operating condition characteristic data packets of historical defect cases, wherein the operating condition characteristic data includes a combination of process parameters and defect information; the defect information includes at least defect type, defect severity, and defect repair measures; The defect mode association storage unit (137) is used to store the association relationship between material characteristics, environmental data and defect modes, including: a) Mapping relationship between material type and environmental parameters to defect type; b) Quantitative correlation model between environmental parameters and defect morphology; c) Correlation model between substrate surface characteristics and defect spatial distribution.

9. The system according to claim 8, characterized in that The multimodal process knowledge base (13) is updated in the following manner: When the prediction deviation of a new defect case is greater than 15%, incremental learning is triggered to update the associated graph weights; Apply a time decay factor to historical data older than 90 days; Correlation between porosity distribution of storage coatings and process parameters.

10. An intelligent imaging generation method, characterized in that: The spraying system according to any one of claims 1 to 9 comprises the following steps: S1. Real-time data acquisition: Use multiple sensors to synchronously acquire spray surface texture characteristics, coating thickness distribution, and ambient temperature and humidity data; S2. Dynamic parameter coordination adjustment: Dynamically calculate the robot arm movement speed v based on the surface roughness Ra, optimize the nozzle pressure P based on the adhesion level F, and adjust the imaging device gain coefficient G based on the temperature and humidity compensation formula; S3. Defect simulation prediction: Input the process parameter set into the physically constrained GAN model to generate a coating defect morphology prediction map and identify sagging defect risk areas; S4. Process optimization decision-making: Select weight coefficients based on the quality / cost model and iteratively generate a collaborative optimization instruction set for paint viscosity, spray speed, and gun distance through reinforcement learning; S5. Closed-loop execution and feedback: Optimization instructions are sent to the spray robot and imaging equipment, and the reward function is updated based on the newly collected coating quality data, triggering incremental learning to optimize the knowledge base association model.

Images