Spraying process adaptive adjustment method based on temperature measurement

Abstract

The present application relates to the technical field of spray process control, and discloses a spray process self-adaptive adjustment method based on temperature measurement. The method comprises the following steps: obtaining target coating parameters and substrate physical characteristic data, combining historical process data and real-time environmental monitoring data, and generating initial spraying control parameters through a first intelligent calculation model; when spraying is performed, real-time collection of thermal distribution data of a spraying area is performed through a temperature sensing device, and a first self-adaptive control algorithm is used to dynamically correct process parameters according to the difference between the thermal distribution data and a preset target temperature interval; meanwhile, an actual coating morphology feature is captured by using a machine vision system, and after comparison with target parameters, an excitation parameter is adjusted in a partitioned manner to optimize uniformity through a second self-adaptive control algorithm; whole-process data is collected, a process optimization instruction is generated through comprehensive evaluation of a second intelligent calculation model, and the first intelligent calculation model and the first self-adaptive control algorithm parameters are updated accordingly.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of spray process control, in particular to a spray process adaptive adjustment method based on temperature measurement. BACKGROUND

[0002] In the field of industrial spraying, coating quality directly affects the appearance, protection performance and service life of products, and the precise control of spraying process parameters is the core factor determining coating quality. Currently, traditional spraying processes rely on manual experience to set fixed parameters, or use simple closed-loop control methods to adjust equipment operating conditions, which is difficult to cope with complex and variable working environments and substrate characteristics, resulting in insufficient coating quality stability.

[0003] Spraying operations are significantly affected by environmental factors. Fluctuations in environmental temperature, humidity, air flow speed and other parameters can change the film formation process of the sprayed material on the substrate surface. For example, when the environmental temperature is too low, the solidification speed of the sprayed material slows down, making it prone to defects such as sagging and pinholes; when the environmental temperature is too high, the sprayed material may solidify prematurely before reaching the substrate surface, resulting in a decrease in coating adhesion. In traditional processes, operators often adjust environmental parameters through regular inspections, but this approach has a lag and cannot respond to environmental changes in real time, making it difficult to ensure the consistency of coating quality.

[0004] Different substrates have different physical properties (such as thermal conductivity, surface roughness, thermal expansion coefficient, etc.), and their requirements for spraying process parameters are also different. For example, metal substrates with high thermal conductivity lose heat quickly during spraying, so the spraying temperature needs to be increased or the solidification time needs to be extended to ensure that the coating is fully solidified; non-metallic substrates with low thermal conductivity tend to accumulate heat, and if the spraying temperature is too high, the substrate may deform. Traditional processes usually preset fixed process parameters for a single substrate type, and when the substrate is changed, a large number of tests need to be conducted to adjust the parameters, which not only increases production costs but also prolongs the production cycle.

[0005] Traditional spraying processes rely on manual visual inspection or sampling for quality detection, which cannot fully and timely obtain the morphological characteristics of the coating (such as thickness, uniformity, flatness, etc.). When defects are found in the coating, a large number of unqualified products have already been produced, making it difficult to make timely process adjustments, resulting in low production efficiency and serious material waste. Although some spraying equipment has introduced machine vision systems for quality detection in recent years, most of them can only identify defects and cannot form a closed-loop linkage with process parameter adjustment, making it difficult to fundamentally solve the problem of unstable coating quality. SUMMARY

[0006] The present application aims to provide a spray process adaptive adjustment method based on temperature measurement to solve the problems raised in the background.

[0007] To achieve the above object, the present application provides a spraying process self-adaptive adjustment method based on temperature measurement, which comprises the following steps:

[0008] acquiring target coating parameters of a spraying operation and physical property data of a substrate, combining historical process data and real-time environmental monitoring data, and generating initial spraying control parameters through a first intelligent calculation model;

[0009] performing a spraying operation based on the initial spraying control parameters and collecting thermal distribution data of a spraying area in real time through a temperature sensing device;

[0010] dynamically correcting process parameters of a spraying device through a first self-adaptive control algorithm according to the difference between the thermal distribution data and a preset target temperature interval;

[0011] capturing actual coating morphological features using a machine vision system and comparing them with target coating parameters, and adjusting excitation parameters in different zones through a second self-adaptive control algorithm to optimize coating uniformity;

[0012] collecting process execution data and temperature monitoring data throughout the whole process, comprehensively evaluating them through a second intelligent calculation model, and generating process optimization instructions;

[0013] updating parameters of the first intelligent calculation model and the first self-adaptive control algorithm according to the process optimization instructions.

[0014] Preferably, the specific steps of generating initial spraying control parameters by the first intelligent calculation model comprise:

[0015] processing target coating parameters, substrate physical property data, historical process data and environmental monitoring data using a random forest regression model to output preliminary control parameters;

[0016] performing process feasibility verification on the preliminary control parameters, and outputting them as initial spraying control parameters if they meet the constraint conditions;

[0017] if the verification fails, performing iterative search using a multi-objective particle swarm optimization algorithm to finally output initial spraying control parameters that meet multiple process constraints.

[0018] Preferably, the execution process of the multi-objective particle swarm optimization algorithm comprises:

[0019] initializing position and velocity vectors of a particle swarm according to spraying process constraint conditions;

[0020] designing a double-target fitness function that takes into account coating quality and energy consumption;

[0021] screening a Pareto optimal solution set through non-dominated sorting and crowding degree calculation;

[0022] According to the gradient guidance strategy, the particle search direction and step are updated;

[0023] Output the final control parameter combination satisfying the Pareto optimality.

[0024] Preferably, the specific steps of dynamically correcting the process parameters of the spraying equipment by the first adaptive control algorithm include:

[0025] A dynamic response model of temperature distribution and spraying rate and atomization pressure is established;

[0026] According to the deviation of real-time thermal distribution data and target temperature interval, the parameter adjustment amount is calculated;

[0027] A fuzzy inference mechanism is used to solve the multi-parameter coordinated adjustment strategy;

[0028] The process parameters of the spraying equipment are synchronously adjusted by a distributed actuator.

[0029] Preferably, the process of calculating the parameter adjustment amount includes:

[0030] Based on historical parameter adjustment records and real-time temperature response data;

[0031] The time-varying coefficients of the dynamic response model are identified online by a recursive least squares algorithm;

[0032] According to the model coefficients, the gain parameters of a proportional-integral-derivative controller are adaptively tuned;

[0033] Parameter adjustment instructions are generated to ensure the asymptotic convergence of temperature.

[0034] Preferably, the process of capturing the actual coating morphology characteristics by the machine vision system includes:

[0035] High-resolution spectral imaging devices are used to collect multi-band image data of the coating surface;

[0036] The thickness distribution and texture features of each partition are extracted by an image segmentation algorithm;

[0037] Quantitative calculation of coating uniformity index and defect density index;

[0038] The feature data is converted into a digital format that can be processed by the second adaptive control algorithm.

[0039] Preferably, the specific steps of partitioning and adjusting excitation parameters by the second adaptive control algorithm include:

[0040] For each partition, the deviation measure of the current coating characteristics and the target parameters is calculated;

[0041] The partition excitation parameter correction amount is generated in combination with the material solidification kinetics model;

[0042] Each sub-zone energy input is independently adjusted by a distributed temperature control device;

[0043] Realize the closed-loop optimization of coating morphology characteristics in each sub-zone.

[0044] Preferably, the specific steps of the second intelligent computing model for comprehensive evaluation include:

[0045] A multi-dimensional evaluation index system considering coating quality, energy consumption efficiency and equipment loss is constructed;

[0046] Deep features in process data are extracted by a deep belief network;

[0047] Output process robustness score and optimization priority identifier;

[0048] According to the priority generation model parameter update instruction.

[0049] Preferably, the process of generating process optimization instructions includes:

[0050] Based on the process robustness score and the historical optimization case library, the optimization strategy under similar working conditions is matched through a case reasoning mechanism, the strength parameters of the optimization strategy are adjusted by fusing real-time monitoring data, and executable optimization instructions suitable for the current equipment state are output.

[0051] Preferably, the process of updating the parameters of the first intelligent computing model and the first adaptive control algorithm according to the process optimization instructions includes:

[0052] According to the process optimization instructions, the key feature parameters are extracted, the node segmentation rules of the random forest regression model are updated through an incremental learning mechanism, the inertia weight and learning factor of the multi-objective particle swarm optimization algorithm are adjusted, and the synchronization optimization of algorithm parameters and process requirements is realized.

[0053] Compared with the prior art, the beneficial effects of the present application are:

[0054] In terms of process adaptability, the method does not rely on fixed preset parameters, but first acquires the target coating parameters of the spraying operation and the physical characteristic data of the substrate, and combines historical process data and real-time environmental monitoring data to generate initial spraying control parameters through a first intelligent calculation model. This process fully considers the differences in physical characteristics of different substrates and the real-time changes in environmental parameters, so that the initial process parameters can accurately match the current operation scenario. For example, for metal substrates and non-metal substrates with large differences in thermal conductivity, the model can automatically adjust the initial spraying temperature, spraying distance and other parameters based on the thermal conductivity data of the substrate; for fluctuations in temperature and humidity in the real-time environment, the model can also predict the influence of environmental changes on the spraying process in advance by combining historical data and real-time monitoring data, generate more adaptive initial parameters, and avoid the tedious process of retesting and adjusting parameters when changing substrates or environmental changes, greatly improving the process adaptation ability to different operation scenarios.

[0055] In terms of coating quality control, the method collects thermal distribution data of the spraying area in real time through a temperature sensing device, and dynamically corrects the process parameters of the spraying equipment according to the difference between the thermal distribution data and the preset target temperature interval through a first adaptive control algorithm. Thermal distribution data can directly reflect the temperature changes in the spraying area, and temperature is a key factor affecting the film formation quality of sprayed materials. Through real-time monitoring and dynamic correction, defects such as coating sagging, pinholes, and insufficient adhesion caused by excessive or insufficient local temperature can be effectively avoided. At the same time, the actual coating morphology characteristics are captured using a machine vision system and compared with the target coating parameters, and the excitation parameters are adjusted by a second adaptive control algorithm, which can accurately optimize problems such as uneven coating thickness and insufficient flatness. Compared with the traditional process of lagging manual detection and adjustment, this dual control method of real-time temperature monitoring and morphology feature comparison can realize all-round and fine control of the coating formation process, ensure the stability and consistency of the coating quality, and reduce the number of unqualified products caused by coating defects.

[0056] In terms of production efficiency improvement, the method reduces the frequency of manual intervention and the influence of human operation error on the process through automatic data collection and intelligent control in the whole process. In the traditional process, the operator needs to regularly inspect the environmental parameters, manually adjust the equipment parameters, and conduct manual quality detection, which not only consumes time and effort, but also easily affects the production progress due to improper operation or judgment errors. In the method, the temperature sensing device and the machine vision system can collect data in real time, and the intelligent calculation model and the self-adaptive control algorithm automatically complete parameter generation and correction without frequent manual intervention, significantly shortening the process adjustment time and improving the continuous operation efficiency of the equipment. At the same time, due to the improvement of coating quality stability, the rework rate of unqualified products is greatly reduced, reducing material waste and production cycle delay, and further improving the overall production efficiency.

[0057] In terms of process optimization iteration, the method collects process execution data and temperature monitoring data throughout the whole process, conducts comprehensive evaluation through a second intelligent calculation model to generate process optimization instructions, and then updates the parameters of the first intelligent calculation model and the first self-adaptive control algorithm. This process forms a closed-loop optimization mechanism for process parameters. As the production process continues to advance, the model can continuously accumulate more actual operation data, find potential problems in the current process parameters through comprehensive evaluation, and make targeted optimization adjustments. For example, when the model finds that adjusting the spraying pressure can further improve the coating uniformity for a certain type of substrate under certain environmental conditions, it will generate corresponding optimization instructions to update the relevant model and algorithm parameters, so that the subsequent spraying process under the same scenario can be more accurate and efficient. This continuous optimization capability enables the process system to continuously adapt to new production demands and scenario changes, and maintain a high process level and production efficiency in the long term. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 A working principle diagram of the spraying process self-adaptive adjustment method based on temperature measurement according to the present application;

[0059] Figure 2 A flowchart for the multi-objective particle swarm optimization algorithm;

[0060] Figure 3 A flowchart for calculating the parameter adjustment amount. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0062] Referring to Figure 1 The present application provides a temperature measurement-based spray process adaptive adjustment method, which comprises: obtaining target coating parameters of a spray operation and physical property data of a substrate, the target coating parameters including design thickness, surface finish and adhesion strength requirements, and the substrate physical property data covering thermal conductivity, specific heat capacity and surface roughness; combining past successful process records stored in a historical process database with real-time environmental monitoring data including environmental temperature and humidity and air flow rate; generating initial spray control parameters through a first intelligent calculation model, the initial spray control parameters including spray rate, atomization pressure and spray gun movement trajectory; driving a spray equipment to perform a spray operation based on the initial spray control parameters, and collecting thermal distribution data of a spray area in real time at a sampling frequency of 100 times per second via a high-precision infrared temperature sensing device, the thermal distribution data being expressed in the form of a temperature matrix to represent spatial temperature changes; dynamically correcting process parameters of the spray equipment through a first adaptive control algorithm according to differences between the thermal distribution data and a preset target temperature interval, the correction objects including coating flow and nozzle angle; capturing actual coating morphological features including thickness distribution and surface texture using a multispectral camera carried by a machine vision system, and comparing the morphological features with the target coating parameters; and adjusting excitation parameters by zones through a second adaptive control algorithm to optimize coating uniformity, the excitation parameters referring to output power of each independent temperature control unit. Process execution data including parameter adjustment records and temperature monitoring data are collected, comprehensive evaluation is performed through a second intelligent calculation model, and process optimization instructions are generated, the process optimization instructions including model update strategies and parameter adjustment thresholds. The parameters of the first intelligent calculation model and the first adaptive control algorithm are updated according to the process optimization instructions, so as to realize continuous self-optimization of process parameters.

[0063] Example 1: Referring to Figure 2, the random forest regression model receives target coating parameters as the main input feature vector, which includes coating design thickness, surface finish grade, and adhesion strength requirements in three dimensions. The substrate physical property data is preprocessed by normalization to eliminate dimensional differences. The preprocessed data includes thermal conductivity, specific heat capacity, and surface roughness measurements. The historical process database calls the last thirty successful process records, which include the mapping relationship between environmental parameters and final coating quality. The environmental monitoring data is matched with the current job time window through the time stamp alignment module. The real-time collected environmental temperature and humidity and air flow rate data are updated at a frequency of five times per second. The model has one hundred decision trees to form an ensemble learning architecture. The maximum depth of each decision tree is limited to twenty layers to balance fitting accuracy and computational efficiency. Feature selection uses the Gini impurity criterion for node splitting. When processing discrete variables in the target coating parameters, automatic one-hot encoding conversion is enabled. The output preliminary control parameters include the theoretical spraying rate interval, the recommended atomization pressure value, and the spray gun movement trajectory coordinate sequence.

[0064] The process feasibility verification module performs multi-dimensional constraint detection on the preliminary control parameters. The equipment power limit detection verifies whether the spraying rate exceeds the maximum motor speed. The material flowability constraint detection determines whether the atomization pressure is within the effective atomization interval through the viscosity-pressure relationship curve. The substrate thermal carrying capacity detection calculates the maximum allowed temperature rise based on the material glass transition temperature. When all detection items simultaneously satisfy the threshold conditions, the verification pass flag is triggered and the parameters are transmitted to the spray equipment main controller. If any detection item exceeds the safety range, the multi-objective particle swarm optimization algorithm is immediately activated for parameter space search. In the algorithm initialization stage, fifty particles are generated to form the search population. Each particle's position vector is defined as a seven-dimensional space point, with dimensions corresponding to spraying rate, atomization pressure, spray gun height, and other key parameters. The velocity vector is initialized as a random value within the range of [-0.1, 0.1]. The position boundary is set as a hard constraint according to the equipment technical specifications.

[0065] The dual-objective fitness function is designed by weighted combination. The coating quality objective item calculates the thickness uniformity index, which is obtained by simulating the coating deposition process to get the standard deviation of thickness distribution. The energy consumption objective item collects historical data of unit area power consumption to establish a regression model. The two objective items form a single fitness value after linear weighting. The weight coefficient is dynamically adjusted according to the current production mode. The non-dominated sorting process uses the quicksort algorithm to stratify the particle swarm. The first front contains all the individuals that are not dominated by other particles. The crowding distance calculation is based on the Euclidean distance in the objective function space. The interval distance between particles is calculated separately for each target dimension. The particles with higher crowding distance are retained to maintain the diversity of the solution set. The gradient guide strategy approximates the gradient of the objective function by finite difference method. A small perturbation is applied at each particle position to calculate the fitness change rate. The cognitive component in the particle velocity vector update formula is adjusted according to the gradient direction. The step size adaptive mechanism dynamically scales according to the evolution effect of the last three generations.

[0066] The Pareto optimal solution set screening process lasts for twenty generations of evolutionary cycles, and the particles in the first front of non-dominated sorting are retained in each generation. The final solution set is selected by the weighted summation method, and the weight distribution is adjusted according to the priority of the real-time monitored coating quality requirements. The output final control parameter combination includes the spray rate setting value, the atomization pressure calibration value and the optimized spray gun motion trajectory, which is converted into a G code instruction sequence executable by the equipment. At the same time, a parameter confidence report is generated for the operator to review. The parameter transmission adopts an industrial Ethernet protocol to ensure real-time performance. After the control command reaches the equipment end, it triggers a safety interlock verification. After verification, the spraying robot performs the initialization motion according to the predetermined trajectory to prepare for the formal spraying operation. In the particle swarm optimization process, the historical process database provides reference for the generation of the initial population. The control parameters under similar working conditions are extracted from past successful cases as seed particles. This warm start mechanism significantly speeds up the convergence process. The elite solution set generated in each iteration is automatically archived in the historical database, forming a continuously enhanced knowledge accumulation cycle. When encountering new substrates or special coating requirements, the algorithm automatically expands the search space boundary and explores unknown parameter regions by relaxing the constraint conditions. This exploration mechanism has successfully solved the problem of carbon fiber composite material spraying parameter optimization. After receiving the final parameter combination, the spraying equipment main controller performs a three-level safety verification process. The first level verifies whether the electrical parameters are within the allowable range of the servo driver. The second level verifies whether the mechanical parameters will cause a collision risk. The third level verifies whether the thermal parameters exceed the substrate limit. After passing all levels, the parameters are loaded into the real-time control system. Before the spraying operation starts, the equipment automatically performs a five-minute preheating program to make the key execution components reach a stable working temperature. During the preheating process, the environmental parameters are continuously monitored. If the temperature and humidity change exceeds the set threshold, the parameter optimization process is triggered again. The parameter adjustment log system records the decision-making process in detail, including the importance ranking of input features of the random forest model, the population distribution state in each generation of particle swarm optimization, and the Pareto solution set screening basis. These logs are stored through time stamp binding with the current spraying operation, providing a traceable analysis basis for subsequent process optimization. When the same parameter combination is used for three consecutive operations, the system automatically generates a parameter stability evaluation report, prompting the process engineer to perform manual review and confirmation.

[0067] In terms of special working conditions, when the environmental temperature changes by more than five degrees Celsius, the system automatically interrupts the current optimization process and starts an emergency parameter adjustment mode, which calls the successful parameter records under similar environmental conditions in the past 24 hours, combines the current substrate characteristics, and performs interpolation calculation to generate temporary control parameters. This mechanism effectively solves the problem of coating sagging caused by sudden heavy rain. For spraying large-size substrates, the system starts a partition parameter optimization mode, which divides the substrate surface into several grid areas for independent optimization of parameters, and uses a gradual parameter transition algorithm at the junction of each area to avoid sudden changes in coating thickness. The model self-updating mechanism is automatically activated at the end of each day's work, and the random forest regression model receives all successful job data on the same day as training samples, dynamically adjusts the feature weight distribution strategy, and the particle swarm optimization algorithm recalibrates the inertia weight coefficient based on recent convergence efficiency data. This continuous learning feature allows the system to maintain high accuracy when dealing with seasonal environmental changes. A comprehensive model evaluation test is performed at the end of each month, using the leave-one-out method of historical data sets to test the accuracy of parameter prediction, and when the accuracy continuously decreases, the expert intervention process is triggered, and the process engineer audits and corrects the model parameters. The equipment calibration link is deeply integrated with the parameter generation process, and the latest calibration data, including key parameters such as spray gun flowmeter calibration coefficient and temperature sensor offset compensation value, are automatically called before generating control parameters each time. These compensation values are directly included in the parameter calculation process to ensure consistency between theoretical parameters and physical execution. When the nozzle wear coefficient exceeds the threshold value, the system automatically adds a compensation amount to the atomization pressure parameter. This real-time compensation mechanism extends the nozzle life by 40%. The abnormal processing module monitors the parameter execution effect in real time, and when the actual coating temperature distribution deviates from the predicted value for more than ten minutes, the system automatically freezes the current parameters and starts an emergency diagnosis process. This process first checks the reliability of sensor data, eliminates faults, and then re-executes local parameter optimization, while marking the current working condition as a special case for further analysis. All abnormal events and processing measures generate detailed reports, which are synchronized to the cloud knowledge base through the enterprise-level Internet of Things platform, realizing experience sharing across factory areas.

[0068] Example 2: see Figure 3, the dynamic response model is constructed by adopting a second-order transfer function form to describe the relationship between temperature distribution and process parameters, the transfer function contains a time delay link to accurately reflect the thermal inertia characteristics of the spraying system, the model coefficients are determined through step response test and system identification technology, and the temperature field evolution data at the time of sudden change of spraying rate are recorded during the test process. Real-time thermal distribution data is collected by a sixteen-channel infrared thermal imager, the spatial resolution of the thermal imager reaches 0.5 millimeters, the temperature sampling frequency is set to one hundred frames per second, and the target temperature interval is dynamically set according to the substrate type and is displayed in real time on the human-computer interface. The temperature deviation amount calculation module performs partition processing on the collected matrix, divides the substrate surface into a twenty-by-twenty grid unit, calculates the algebraic difference between the current temperature and the target value for each unit, and simultaneously counts the coordinates of the abnormal area exceeding the allowed deviation threshold. The parameter adjustment amount generation module adopts a double closed-loop control architecture, the outer loop calculates the overall temperature distribution uniformity index, the inner loop focuses on local hot spot area suppression, the deviation integral term uses a sliding time window accumulator to realize, and the window length is adaptively adjusted according to the material thermal response speed. The fuzzy reasoning mechanism is deployed in the edge computing unit, the input variables include the absolute value of the temperature deviation and its first-order derivative, the membership function adopts a triangular distribution, the rule base integrates the material expert's twenty years of process experience, and each rule is like "if the deviation is large and the change is fast, then greatly reduce the rate". The defuzzification process uses the barycentric method to calculate the accurate output value, and the output variable is mapped to the spraying rate correction percentage and the atomizing pressure adjustment range. The distributed execution system includes twelve groups of independently controlled servo actuators, each group of actuator drives a specific spray gun unit, the actuator receives the adjustment instruction and starts the cooperative motion protocol, and the time stamp synchronization ensures the spatio-temporal consistency of parameter adjustment. The servo motor adjusts the coating gear pump speed, the control accuracy reaches 0.005%, the piezoelectric ceramic driven pressure regulating valve realizes millisecond level response, and the states of all actuators are fed back to the main control system in real time through the industrial bus. When the parameter adjustment amount calculation process is started, the historical database automatically retrieves the adjustment records of the last ten similar working conditions, the records include the corresponding relationship between the parameter adjustment amplitude and the temperature response curve, and the retrieval results are visualized in the form of a three-dimensional surface. Real-time temperature response data is obtained through a high-speed data acquisition card, the sampling interval is accurate to 5 milliseconds, and the data stream passes through a digital filter to eliminate sensor noise. The recursive least squares algorithm runs on a dedicated digital signal processor, the initial value of the model coefficient matrix comes from the offline identification result, the forgetting factor is dynamically adjusted according to the model convergence state, and the model is considered to be converged when the coefficient change is less than the set threshold for five consecutive iterations.

[0069] The proportional-integral-derivative controller gain tuning module receives the updated model coefficients, establishes the analytical relationship between the controller parameters and the process model, the proportional term gain is inversely proportional to the system steady-state gain, the integral time constant matches the process time constant, and the derivative term coefficient is calculated based on the system damping ratio. The online tuning process continuously monitors the control effect, and when the temperature deviation integral value exceeds the warning line, the gain recalculation process is triggered. The parameter adjustment command generator synthesizes the controller output and the fuzzy reasoning result, generates the final execution command using a weighted fusion strategy, and the command value is subjected to a rate limiter to prevent the actuator from overshooting. The final output includes the change in the spray flow set value and the adjustment step size of the atomizing pressure. The temperature asymptotic convergence guarantee mechanism designs a special monitoring program to continuously track the temperature deviation integral area trend, and when it is detected that the integral area has not decreased for three consecutive control periods, the integral action weight is automatically increased. The actuator is equipped with an over-temperature protection interlock, which immediately cuts off the paint supply to the corresponding area and starts the cooling air curtain when the local area temperature exceeds the material safety threshold. The control cycle is strictly synchronized with the temperature collection frequency, and the full-process calculation from data collection to command output is completed within each control cycle. In the base material corner area processing, the system automatically enhances the temperature monitoring density, increases the grid division accuracy to five by five millimeters, and relaxes the parameter adjustment amplitude limit to cope with the rapid heat accumulation. For large curvature surface spraying, the dynamic response model introduces a curvature compensation factor, which is calculated by real-time acquisition of surface geometric features through a laser range finder. When processing heat-sensitive materials, the control algorithm automatically switches to a conservative mode, reduces the proportional gain, and sets a more stringent temperature safety margin. The model online updating system evaluates the model accuracy every ten minutes, compares the actual temperature response with the model prediction value, and when the root mean square error continues to exceed the standard, the model re-identification process is started. The re-identification process uses a pseudo-random binary sequence to excite the system, applies small parameter perturbations within the process allowable range, and quickly updates the model parameters through correlation analysis. All model change records are timestamped and marked with process characteristics to form a traceable model version management archive.

[0070] The actuator calibration data is deeply integrated into the control process. The latest calibration coefficients of each execution unit, including flow meter linearity compensation table and valve dead zone compensation value, are automatically loaded before each parameter adjustment. When the actuator response delay is detected to be abnormally increased, the system automatically increases the feedforward compensation in the adjustment command, which is calculated based on the actuator characteristic curve. The actuator health monitoring module continuously analyzes the driving current waveform and provides early warning of potential mechanical wear and tear. The sudden disturbance processing mechanism has a special environmental disturbance observer. When the environmental wind speed is detected to suddenly change by more than two levels, the system automatically adds a wind resistance compensation term to the atomization pressure adjustment. When the material batch is switched, the system starts the parameter smoothing transition program, which gradually switches to the new parameter set within three control cycles to avoid coating quality fluctuations. All parameter adjustment operations generate detailed audit logs, recording adjustment decision basis, execution effect and subsequent temperature response curve. These data are uploaded to the factory data center through the manufacturing execution system. The fault safety mode is automatically activated when the system is abnormal. When the core processor detects that the calculation timeout of five consecutive control cycles, it immediately switches to the preset safety parameter set and runs, and sends a maintenance request alarm. The communication interruption protection mechanism ensures that each actuator maintains the last valid parameter when the bus fails, and the temperature monitoring system switches to independent operation mode to continuously record the heat distribution data. After the daily operation is completed, the system performs a self-checking program to automatically check the operation accuracy of all control modules and the actuator positioning error, and generates a device health status report for the maintenance team to refer to.

[0071] The multi-parameter collaborative optimization algorithm specially handles the coupling effect of spraying rate and atomization pressure, establishes an interaction matrix between the two, and simultaneously calculates the compensation amount of atomization pressure when adjusting the spraying rate. For high-viscosity coatings, the system automatically increases the pressure adjustment weight to maintain the atomization quality, and in the fast-moving area, it prioritizes to ensure the stability of the spraying rate. All parameter adjustment strategies are simulated in advance through a virtual spraying system. The simulation results are displayed in the form of a thermal map to predict the temperature distribution. The operator confirms that there is no error before issuing the execution command.

[0072] Example 3: The high-resolution spectral imaging device adopts a multi-band collaborative acquisition mode. The visible light band (400-700 nm) sensor captures surface texture and macro defects, and the near-infrared band (700-1100 nm) sensor detects coating thickness distribution. Each sensor is equipped with an independent optical filter wheel to achieve rapid switching of the waveband. The image acquisition resolution is set to 2048x2048 pixels, with a single image covering an area of 30 cm x 30 cm and a pixel size of 0.15 mm. The imaging system is equipped with a ring-shaped LED array illumination device, and the illumination intensity is automatically adjusted according to the surface reflectivity to ensure the consistency of image gray scale in different areas. Multi-band image data is transmitted to the image processing unit through Gigabit Ethernet, and a lossless compression algorithm is used to maintain data integrity during transmission. The image segmentation algorithm uses an improved watershed algorithm to process coating area division. First, the original image is subjected to Gaussian filtering to eliminate noise interference, and the filter kernel size is dynamically adjusted according to the image contrast. The gradient calculation uses the Sobel operator to detect edge information, and after the initial watershed segmentation, a region merging algorithm is applied to eliminate over-segmentation, and the merging threshold is calculated based on the gray scale similarity between regions.

[0073]

[0074] where T ij represents the thickness estimate (μm) at pixel point (i,j), G ij is the gray value of the point (0-255), and a, β, γ are calibration coefficients related to the material, which are fitted from standard sample data by the least squares method. The texture feature extraction uses a Gabor filter bank, with six directions (0°, 30°, 60°, 90°, 120°, 150°) and four scales (2px, 4px, 8px, 16px) of filter kernels. The response amplitude of each filter constitutes the texture feature vector.

[0075] The uniformity index calculation is based on the thickness distribution matrix. First, the edge 5% area is excluded to avoid boundary effects, and the ratio of the standard deviation to the average value of the remaining area thickness values is calculated as the uniformity index. The adaptive threshold binarization algorithm is used for defect density detection, and the threshold is dynamically determined according to the local gray scale distribution. Connected component analysis is used to identify defects such as pores and cracks, and the defect density is defined as the ratio of the total number of defect pixels to the total number of effective area pixels. The feature data conversion module maps the image features to a standardized digital format. The thickness distribution matrix is downsampled to 100x100 grid data, and the texture feature vector is reduced to 20 dimensions through principal component analysis. All data are normalized to the range [0, 1].

[0076] The partition deviation metric calculates the substrate surface into 10 cm x 10 cm independent control regions, each region contains about 400 thickness measurement points. The deviation metric uses a modified Euclidean distance formula, taking into account the thickness deviation, texture difference and defect distribution:

[0077]

[0078] wherein: D k is the integrated deviation metric of the kth partition, N k is the number of measurement points in the partition, T ik is the thickness value of the measurement point, T target is the target thickness, F texture，k is the texture feature vector of the partition, F target is the target texture feature, and λ is the weight coefficient (usually 0.3). The material solidification kinetics model integrates the Arrhenius equation to describe the relationship between temperature and solidification rate, and the model parameters are obtained by differential scanning calorimetry test, including key parameters such as activation energy Ea and pre-exponential factor A.

[0079] The partition excitation parameter correction amount generation module receives the deviation metric and the solidification model output, and calculates the required energy input adjustment amount. For under-thickness areas, increase the infrared heating power and extend the action time; for over-solidification areas, reduce the power and start the air cooling device. The distributed temperature control system contains 64 independently controlled infrared heating units, each unit covers an area of 5 cm x 5 cm, the power adjustment range is 0-500 W, and the response time is less than 100 ms. The temperature closed-loop control adopts the PID algorithm, which monitors the surface temperature of each partition in real time and adjusts the heating power through PWM. The zone-by-zone optimization process adopts a sequential scanning strategy, starting from one end of the substrate and processing each partition in turn, with a 2 cm overlap between adjacent partitions to ensure smooth transition. The optimization period of each partition is set to 30 seconds, including image acquisition (5s), feature extraction (10s), parameter calculation (5s), and temperature adjustment (10s). During the optimization process, the coating state changes are continuously monitored, and when abnormal rapid solidification is detected, the current period is immediately interrupted and re-evaluated. The image processing unit is equipped with a GPU acceleration card to improve computing efficiency, and the watershed algorithm realizes parallel processing to segment the image into several blocks for simultaneous processing. The thickness calibration curve is updated regularly, recalibrated once a month using a standard template, and a new calibration is performed immediately when the coating type is changed. The lighting system is equipped with a photometric feedback closed loop to monitor and compensate for environmental light changes in real time, ensuring image acquisition stability. The defect detection algorithm integrates a machine learning classifier, with a training set containing thousands of typical defect images, capable of distinguishing between common defect types such as pores, cracks, and orange peel. Defect areas are automatically marked when the classification result confidence exceeds 90%, and manual review is recommended when the threshold is below this value. The texture feature database stores feature templates of various qualified coatings, supporting automatic calling of corresponding templates according to product specifications. The solidification model considers the influence of environmental temperature, with a built-in temperature compensation algorithm to correct the reaction rate constant. When the environmental humidity exceeds 60%, automatically increase the drying stage power to offset the influence of water evaporation. The heating unit layout is optimally designed to ensure uniform temperature distribution in each partition, with auxiliary heaters in the edge area to compensate for heat loss. The data acquisition and processing process is timestamped, with a deviation of less than 10 ms between the image acquisition time and the temperature recording time. All processing results generate detailed reports, including thickness distribution cloud map, defect location labeling, texture feature spectrum, and other visual data. The report is transmitted in real time to the quality management system through the factory network, providing complete data support for product traceability. The system self-diagnosis function is executed once a day for comprehensive detection, including camera focal length calibration, lighting uniformity check, heating unit power test, etc. The maintenance warning system predicts potential failures based on equipment running time and schedules preventive maintenance in advance. The operation interface provides real-time monitoring views, showing the deviation trend of the current coating quality indicators and target values, supporting historical data backtracking analysis.

[0080] Example 4: The multi-dimensional evaluation index system is constructed to contain three categories of core index items. The coating quality dimension sets three sub-items of thickness uniformity coefficient, surface defect density, and gloss deviation. The energy efficiency dimension records the unit area power consumption, compressed air consumption, and heat energy utilization rate. The equipment wear dimension tracks the nozzle wear index, motion mechanism fatigue, and heating element aging coefficient. The index data collection covers the entire spraying operation cycle, starting from the parameter initialization stage to the completion of coating solidification. The data storage uses a time series database to record millisecond-level changes. The deep belief network architecture contains five hidden layers, each layer configured with 128 neuron nodes. The input layer receives the standardized multi-dimensional index data stream, and the output layer generates a process robustness score value ranging from 0 to 100. Refer to Table 1.

[0081] Table 1: Process evaluation index calculation parameter table

[0082]

[0083] The network pre-training stage uses process data stored for nearly three years, totaling more than 100,000 valid records, to initialize network weights through a contrastive divergence algorithm. The forward propagation process uses a piecewise linear activation function, and the feature extraction automatically identifies the nonlinear correlation between indicators, such as discovering the positive correlation between nozzle wear and compressed air consumption. The process robustness score calculation introduces a time decay factor, with recent data having a higher weight than historical data. The score formula includes a fluctuation amplitude penalty term, and when consecutive indicators are abnormal, the score value decreases rapidly. The optimization priority identifier generation uses a density clustering algorithm to map the current process parameters to the historical case space, and identifies high-priority areas as working conditions with a distance from the historical optimal case of more than three standard deviations. The case reasoning engine performs similarity matching in the historical optimization case library, which is indexed by substrate type, environmental conditions, and coating requirements. The similarity calculation uses an improved cosine similarity algorithm, and the feature vector includes twelve key process parameters, each with an independent weight coefficient. Taking automobile bumper spraying as an example, when the system detects that the environmental humidity suddenly rises to 75%, it automatically matches the optimization scheme numbered CT2023-087 in the case library, which has successfully solved the orange peel problem of coatings in high-humidity environments. The optimization strategy adjustment module receives the matched case and adjusts the original scheme's infrared power increase amplitude from 15% to 22% based on real-time monitoring of temperature gradient data, while extending the drying stage time parameter. The executable optimization instruction generation uses a modular design, including device parameter adjustment instructions, control algorithm update instructions, and monitoring strategy change instructions. Device parameter instruction examples include "spraying rate +5%, region 7 heating power -10%", control algorithm instruction examples include "enable humidity compensation PID mode", and monitoring instruction examples include "increase partition 9 temperature sampling frequency". The instruction intensity parameter is calibrated in real time by the environmental sensor network, and when the spraying workshop air flow speed is detected to be greater than 2m / s, the wind resistance compensation coefficient is automatically added to the atomization pressure adjustment instruction. The final instruction set is packaged in JSON-LD format, including timestamp, target device ID, and execution validity period metadata fields.

[0084] The process optimization instruction triggers a double verification mechanism when transmitted to the equipment layer. First, the edge computing node checks the feasibility of the instruction and checks whether the parameters exceed the physical limits of the equipment. Second, the safety interlocking system verifies the execution timing to prevent conflicts with ongoing emergency braking. The instruction execution process is monitored throughout, and when an abnormal temperature rise in the nozzle is detected, a cooling interval period is automatically inserted. The execution result data flows back to the evaluation system, forming a closed-loop feedback link. New data is automatically imported into the case library and the similarity matching model is updated. In mixed material substrate processing scenarios, the system automatically segments the evaluation area. High-power rapid solidification strategies are used for metal areas, while low-temperature slow solidification modes are used for plastic areas. Special evaluation procedures are started at the junction of multiple materials, with additional adhesion strength test items. The optimization instruction includes a gradient temperature setting scheme. When processing complex curved workpieces, the evaluation index dynamically increases the curvature adaptability factor, which is calculated based on laser scanning point cloud data. The model version management system records the generation logic of each optimization instruction and preserves complete snapshots of the decision path. Model health detection is performed monthly, and the evaluation accuracy is verified by injecting historical test cases. When the score deviation exceeds 5%, the model retraining process is triggered. The engineer intervention interface allows manual correction of priority identifiers, and all modification records are labeled with the modification reason and updated with the case library feature weights. The real-time dashboard visualizes the evaluation process, presenting equipment wear distribution with a heat map and tracking energy efficiency trends with a line chart. When the process robustness score falls below the yellow warning line, the interface automatically pops up an optimization suggestion preview window, and the operator can manually adjust the intensity parameters before issuing the execution instruction. The data traceability function supports playback of the evaluation index state at any time point, and the associated display of the optimization instruction content executed at that time. The equipment maintenance subsystem is deeply integrated with the evaluation system. When the nozzle wear index exceeds the threshold value for three consecutive times, a spare parts replacement work order is automatically generated and the subsequent process parameters are adjusted to compensate for the wear effect. The heating element aging detection uses resistance change analysis, and the result data is directly input into the equipment wear evaluation item. All maintenance records are associated with the corresponding job batch to form a device lifecycle history library. The abnormal condition handling mechanism has twelve preset scene plans. When a sudden voltage fluctuation is detected, plan P-07 is automatically enabled, which suspends the precise heating control and switches to a safety mode. After power restoration, the system executes a compensation solidification program, recalculating the total energy input based on the power outage duration. An special evaluation report is automatically generated after each plan execution, analyzing the impact of the abnormal event on the final coating quality. The cross-line knowledge migration module periodically synchronizes optimization cases from each production line, and through feature matching, the applicable solutions are extended to similar equipment. When a new spraying line is installed and put into operation, the system automatically loads the baseline case set and starts the transfer learning process, significantly shortening the parameter tuning period of the new line. The execution effect of all optimization instructions is included in the factory KPI evaluation system, forming a continuous improvement positive cycle mechanism.

[0085] In Example 5, the process optimization instruction parsing module extracts key feature parameters using a multi-level filtering strategy. The first level identifies the temperature response constant by analyzing the thermal distribution data from the last ten spraying operations. The time constant required for the substrate surface temperature to reach 63.2% of the set value is calculated, which reflects the thermal inertia characteristics of the material. The second level extracts the coating solidification rate feature based on the solidification front movement speed data recorded by the infrared thermal imager. The solidification kinetics parameters are calculated by combining the specific heat capacity and thermal conductivity of the material. The third level obtains the environmental coupling factor, which quantifies the influence of environmental temperature and humidity on the coating formation process. This factor is obtained through multivariate regression analysis of historical operation data. All feature parameters are standardized and stored in the feature library, with data sources and timestamps labeled. When the incremental learning process of the random forest regression model starts, the node partitioning rule set of the current model is first loaded, and the rules are stored in memory in a binary tree structure. When new process data flows in, the model calculates the prediction error of each decision tree for the new data. When the average error exceeds the threshold, the rule updating process is triggered. The node partitioning rule adjustment uses a gradient descent algorithm to fine-tune the partitioning threshold in the direction of error reduction. For continuous numerical features such as spraying rate, the partition point precision is maintained to three decimal places. The classification feature processing uses one-hot encoding conversion to ensure the consistency of the partitioning rules. During the model updating process, statistical summaries of historical data, including feature means and variances, are preserved for the standardization of new data.

[0086] The parameter adjustment of the multi-objective particle swarm optimization algorithm focuses on the inertia weight and learning factor, two core parameters. The inertia weight is initially set to 0.9 and dynamically adjusted based on the convergence speed of the last twenty generations of the population. When the Pareto front is detected to be moving slowly, the inertia weight is gradually reduced to 0.4 to enhance the local search ability; when the population diversity is found to be decreasing, the inertia weight is appropriately increased to promote global exploration. The learning factor includes individual learning factor and social learning factor, which control the step size of particles moving towards their historical optimum and the group optimum, respectively. The individual learning factor is adaptively adjusted based on the historical performance of the particle, with superior particles obtaining greater autonomous exploration rights; the social learning factor is calculated based on the population distribution density, increasing the social learning weight in sparse regions of the solution space. The synchronous optimization of algorithm parameters and process requirements is achieved through a double-loop mechanism, with the inner loop executing the standard particle swarm optimization process and the outer loop monitoring the satisfaction of process constraints. When a solution that violates the process constraints is detected, the outer loop immediately adjusts the algorithm parameters and reinitializes the population. Process constraints include coating thickness tolerance, drying time limit, energy consumption upper limit, and other hard requirements, which are encoded as constraint functions and embedded in the fitness calculation process. During the synchronous optimization process, the feasibility of the solution is continuously evaluated, and for marginal feasible solutions, a fine tuning program is started to search in a small range near the constraint boundary.

[0087] In the case of spraying aluminum alloy wheel hubs for automobiles, the system detected that the curing characteristics of a new type of transparent varnish were significantly different from the original material, with a temperature response constant 30% higher than the standard value. The feature parameter extraction module immediately marked this abnormal situation and triggered the model updating process. The random forest regression model first adjusted the feature weight related to the curing temperature, increasing the decision weight of temperature-sensitive features, and added a material type identification feature. After updating the node segmentation rule, the prediction error of the spraying rate for the new material decreased from 15% to within 5%. The particle swarm optimization algorithm responded to the material change by automatically increasing the weight coefficient of the temperature-related target when it identified the thermal sensitivity of the new material. The inertia weight adjustment module found that the population was oscillating near the temperature constraint boundary, so it gradually reduced the inertia weight from 0.7 to 0.5, enhancing the local search accuracy. The learning factor coordination module noticed that the high social learning factor led to premature convergence of the population, so it appropriately reduced the social learning factor and increased the individual learning factor, promoting diversity preservation. The model updating verification system cross-validated the data before making any parameter modifications, dividing the historical data into training and test sets to ensure that the updated model maintained good performance on unseen data. For parameter adjustment of the particle swarm algorithm, a virtual simulation environment was used to test the effect of new parameters, and the simulation model included a complete simulation of the spraying physical process. After verification, the new parameters were deployed to the actual production system, and the deployment process used a gray release strategy, first running for 24 hours in a single spraying station, and then promoting to the entire line after confirming the effect.

[0088] The version control system records all parameter change history, generates a unique version identifier for each update, and associates the corresponding process optimization instruction number. The rollback mechanism allows quick recovery to the previous stable version when an exception occurs, and the rollback decision is based on the real-time monitored coating quality index change trend. The parameter performance evaluation module continuously tracks the updated effect and calculates the improvement degree of key performance indicators as a reference for subsequent optimization. The exception handling mechanism is specifically designed for the parameter update process. When a sudden increase in model prediction error is detected, the incremental learning process is automatically suspended and an alarm is issued. If the population performance continues to decline during algorithm parameter adjustment, a reset program is triggered to restore to the last known good parameter configuration. All abnormal events are recorded in the system log, including timestamp, exception type, handling measures, and final results. The knowledge migration module supports the application of optimized parameter configurations to similar production scenarios. When a newly installed spraying line has similar configurations to the existing system, the verified parameter set is automatically transferred. Device differences are considered during the migration process, and parameter values are adjusted according to machine specifications, such as scaling the spray gun moving speed according to the track length. Regular parameter consistency checks are performed to ensure that parameter settings between production lines meet standardized requirements. The human-machine interface provides visual monitoring of parameter adjustment, displaying the current model accuracy and algorithm convergence state curve. Operators can manually adjust hyperparameters such as learning rate, and all manual interventions are recorded and included in the effectiveness evaluation system. The system generates a parameter optimization report every month, summarizing this period's adjustment content, effectiveness evaluation, and next phase optimization suggestions. The report is automatically distributed to process engineers and quality management departments.

[0089] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one from another entity or action without necessarily requiring or implying that the entities or actions are in any way mutually exclusive or directional. Moreover, the terms "include", "have", or any variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0090] While the embodiments of the present application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, alternatives, and variations can be made therein without departing from the spirit and scope of the application as defined in the following claims and its equivalents.

Claims

1. A method for adaptive adjustment of a spray process based on temperature metrology, characterized in that, The method comprises the following steps: acquiring target coating parameters of a spraying operation and physical property data of a substrate, combining historical process data and real-time environmental monitoring data, and generating initial spraying control parameters through a first intelligent calculation model; performing a spraying operation based on the initial spraying control parameters and collecting thermal distribution data of a spraying area in real time through a temperature sensing device; dynamically correcting process parameters of a spraying device through a first adaptive control algorithm according to differences between the thermal distribution data and a preset target temperature interval; capturing actual coating morphological characteristics using a machine vision system and comparing them with target coating parameters, and adjusting excitation parameters in different zones through a second adaptive control algorithm to optimize coating uniformity; collecting process execution data and temperature monitoring data throughout the whole process, comprehensively evaluating them through a second intelligent calculation model, and generating process optimization instructions; updating parameters of the first intelligent calculation model and the first adaptive control algorithm according to the process optimization instructions; the specific steps of generating initial spraying control parameters by the first intelligent calculation model include: processing target coating parameters, substrate physical property data, historical process data and environmental monitoring data using a random forest regression model to output preliminary control parameters; performing process feasibility verification on the preliminary control parameters, and outputting the preliminary control parameters as initial spraying control parameters if they meet the constraint conditions; if the preliminary control parameters do not pass the verification, performing iterative search using a multi-objective particle swarm optimization algorithm to finally output initial spraying control parameters that meet multiple process constraints; the specific steps of dynamically correcting process parameters of a spraying device by the first adaptive control algorithm include: establishing a dynamic response model of temperature distribution and spraying rate and atomization pressure; calculating parameter adjustment amounts according to deviations of real-time thermal distribution data from a target temperature interval; solving a multi-parameter coordinated adjustment strategy using a fuzzy reasoning mechanism; synchronously adjusting process parameters of a spraying device through a distributed actuator; the specific steps of adjusting excitation parameters in different zones by the second adaptive control algorithm include: calculating deviation amounts of current coating characteristics from target parameters for each zone; generating zone excitation parameter correction amounts in combination with a material solidification kinetics model; independently adjusting energy input of each zone using a distributed temperature control device; realizing closed-loop optimization of coating morphological characteristics in different zones; the specific steps of comprehensive evaluation by the second intelligent calculation model include: building a multi-dimensional evaluation index system considering coating quality, energy consumption efficiency and equipment loss; extracting deep features in process data through a deep belief network; outputting process robustness scores and optimization priority identifiers; generating model parameter update instructions according to the priority.

2. The method of claim 1, wherein, the execution process of the multi-objective particle swarm optimization algorithm includes: initializing position and velocity vectors of a particle swarm according to spraying process constraint conditions; designing a double-target fitness function considering coating quality and energy consumption; screening a Pareto optimal solution set through non-dominated sorting and congestion calculation; updating particle search direction and step length according to a gradient guide strategy; outputting a final control parameter combination that meets Pareto optimality.

3. The method of claim 1, wherein, the process of calculating parameter adjustment amounts includes: based on historical parameter adjustment records and real-time temperature response data; online identifying time-varying coefficients of a dynamic response model through a recursive least squares algorithm; Adapting gain parameters of a proportional-integral-derivative controller according to model coefficients Generating parameter adjustment instructions ensuring asymptotic convergence of temperature 4. The method of claim 1, wherein, The process of capturing actual coating morphology features by using a machine vision system comprises: Collecting multi-band image data of the coating surface by using a high-resolution spectral imaging device; Extracting thickness distribution and texture features of each partition by using an image segmentation algorithm; Quantitatively calculating coating uniformity indicators and defect density indicators; Converting feature data into a digital format that can be processed by a second adaptive control algorithm.

5. The method of claim 1, wherein, The process of generating process optimization instructions comprises: Based on process robustness scores and a historical optimization case library, matching optimization strategies under similar working conditions by using a case reasoning mechanism, adjusting strength parameters of the optimization strategies by fusing real-time monitoring data, and outputting executable optimization instructions suitable for the current equipment state.

6. The method of claim 1, wherein, The process of updating parameters of the first intelligent calculation model and the first adaptive control algorithm according to the process optimization instructions comprises: Extracting key feature parameters according to the process optimization instructions, updating node partition rules of a random forest regression model by using an incremental learning mechanism, adjusting inertia weights and learning factors of a multi-objective particle swarm optimization algorithm, and realizing synchronous optimization of algorithm parameters and process requirements.

Images