Coating liquid preparation optimization method and system based on adaptive control and storage medium

Abstract

The invention relates to the technical field of coating liquid configuration optimization control, and discloses a coating liquid configuration optimization method and system based on adaptive control and a storage medium. The method comprises the following steps: collecting coating liquid parameter data by utilizing multiple sensors, and obtaining a feature vector and an environmental influence factor data set; establishing a multi-layer mapping model of parameters and coating quality, and generating a sensitivity weight matrix; inputting the weight matrix and the constraint condition into an annealing algorithm, and calculating an optimal formula instruction; and driving the metering pump and the stirrer to execute configuration through sliding mode control based on the instruction. According to the application, accurate modeling of the relationship between the coating liquid parameters and the coating quality is realized, and accurate regulation and control of the coating liquid preparation process are realized through adaptive control on the basis, so that the consistency and stability of the coating quality are improved.

Description

Technical Field

[0001] The present application relates to the technical field of coating liquid configuration optimization control, and in particular to a coating liquid configuration optimization method, system and storage medium based on adaptive control. Background Art

[0002] In the field of film coating, especially in the production of high-performance films such as nylon materials (such as BOPA), the quality of the coating solution configuration directly determines the uniformity, adhesion and final product performance of the coating. Traditional coating solution configuration technology mainly relies on manual experience and fixed formulas. Operators usually configure the component ratios and mixing parameters of the coating solution based on experience. With the development of industrial automation, some production lines have begun to use basic automatic control systems for coating solution configuration, such as PLC-controlled basic batching systems and metering pump automatic control systems. These systems usually complete the coating solution configuration process through preset formula parameters and simple closed-loop control, which has improved the consistency of coating solution configuration to a certain extent.

[0003] However, the existing coating liquid configuration technology has obvious deficiencies, leading to a series of technical problems that need to be solved urgently: First, the traditional fixed formula is difficult to cope with fluctuations in raw material batches and changes in environmental conditions. Different batches of raw materials may have different chemical properties, and changes in ambient temperature and humidity will also affect the performance of the coating liquid. Fixed formulas are difficult to dynamically adjust to adapt to these changes, which leads to technical problems such as unstable coating liquid performance; second, the existing automatic configuration system usually adopts a simple PID control algorithm with limited control accuracy, which is difficult to meet the strict requirements of high-end thin film coating on coating liquid stability, which causes technical problems such as large fluctuations in coating liquid parameters; third, the coating There is a complex, nonlinear relationship between coating parameters and final coating quality. Existing technologies lack effective models to characterize this relationship, making it impossible to optimize parameter configuration. This leads to technical problems such as insufficient optimization of the coating fluid formulation. Fourth, current technologies lack the ability to monitor and dynamically adjust the coating fluid during configuration. Once the coating fluid parameters deviate from expectations, it is difficult to accurately correct them during production, resulting in technical problems such as inaccurate control of the coating fluid configuration process. Finally, the optimal formulation of multi-component coating fluids under different production conditions is difficult to determine empirically, and there is a lack of systematic formulation optimization methods. This leads to technical problems such as difficulty adapting the coating fluid formulation to changing conditions. These issues collectively lead to problems such as fluctuating coating quality, poor product consistency, and raw material waste, which restrict the production efficiency and quality stability of high-performance film materials.

[0004] In addition, in terms of coating liquid parameter data collection, existing technologies often use a single sensor or low-frequency sampling, and the data processing method is simple, which makes it difficult to capture small changes and short-term fluctuations in the coating liquid configuration process, resulting in technical problems such as incomplete parameter feature extraction. In terms of modeling the relationship between coating liquid parameters and coating quality, traditional models mostly use linear regression or simple neural networks, which make it difficult to characterize the complex coupling relationship and interaction effects between parameters, resulting in technical problems such as insufficient model prediction accuracy. In terms of coating liquid formula optimization algorithms, conventional optimization methods such as gradient descent and genetic algorithms have difficulty handling global optimization problems under multi-objective and multi-constraint conditions, and are prone to falling into local optimality, resulting in technical problems such as limited formula optimization effects. In the coating liquid configuration execution link, traditional control methods have weak anti-interference capabilities and are difficult to cope with various disturbances in the production process, causing technical problems such as insufficient execution accuracy. Summary of the Invention

[0005] The present application provides a coating liquid configuration optimization method, system and storage medium based on adaptive control, which are used to accurately model the relationship between coating liquid parameters and coating quality, and on this basis, achieve precise regulation of the coating liquid configuration process through adaptive control, thereby improving the consistency and stability of the coating quality.

[0006] In the first aspect, the present application provides a coating liquid configuration optimization method based on adaptive control, which includes: using a temperature sensor, a viscosity sensor, a pH sensor and a solid content detector set in a mixing tank to collect and analyze parameter data in the coating liquid configuration process, and obtain a coating liquid parameter characteristic vector and an environmental influencing factor data set; based on the coating liquid parameter characteristic vector and the environmental influencing factor data set, a multi-layer coupling mapping model of coating liquid parameters and coating quality is established, and a parameter sensitivity weight matrix is generated after performing parameter sensitivity quantitative analysis; the parameter sensitivity weight matrix and preset formula constraints are input into an adaptive simulated annealing algorithm, the optimal formula of the coating liquid is iteratively calculated, and a formula instruction including the precise addition amount of the components and the mixing parameters is output; based on the formula instruction, the metering pump control unit and the agitator control unit are driven by the hierarchical integral sliding mode control in the distributed architecture to perform coating liquid configuration.

[0007] In a second aspect, the present application provides a coating liquid configuration optimization system based on adaptive control, the coating liquid configuration optimization system based on adaptive control comprising:

[0008] The acquisition module is used to collect and analyze parameter data during the coating liquid preparation process using the temperature sensor, viscosity sensor, pH sensor and solid content detector set in the mixing tank to obtain the coating liquid parameter feature vector and environmental influencing factor data set;

[0009] An execution module is used to establish a multi-layer coupling mapping model between coating liquid parameters and coating quality based on the coating liquid parameter feature vector and the environmental influencing factor data set, and generate a parameter sensitivity weight matrix after performing parameter sensitivity quantitative analysis;

[0010] An input module is used to input the parameter sensitivity weight matrix and preset formula constraints into an adaptive simulated annealing algorithm, iteratively calculate the optimal formula of the coating liquid, and output a formula instruction including the precise addition amount of the components and the mixing parameters;

[0011] The control module is used for driving the metering pump control unit and the stirrer control unit to execute coating liquid configuration based on the recipe instruction through hierarchical integral sliding mode control in a distributed architecture.

[0012] In a third aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned coating liquid configuration optimization method based on adaptive control.

[0013] The technical solution provided in this application solves a number of technical problems in the coating liquid configuration process by constructing an adaptive control closed loop, from data acquisition, model building, formula optimization to control execution. First, by using a variety of sensors set in the mixing tank to collect parameter data at high frequency, the coating liquid parameter feature vector and environmental influencing factor data set are obtained, realizing refined data capture of the entire process of coating liquid configuration, improving the integrity and accuracy of the data, and solving the problems of low data acquisition frequency and narrow coverage in the traditional coating liquid configuration process. Secondly, based on these high-quality data, a multi-layer coupled mapping model of coating liquid parameters and coating quality is constructed through a self-organizing map network and a Gaussian process regression algorithm. This model can accurately characterize the complex nonlinear relationship between coating liquid parameters and coating quality, providing a reliable theoretical basis for subsequent optimization, and overcoming the shortcomings of low accuracy and weak generalization ability of traditional empirical models. In particular, the application of self-organizing map networks in parameter space dimensionality reduction and clustering enables complex multidimensional parameter relationships to be intuitively visualized, greatly improving the interpretability of the model. Third, a parameter sensitivity quantification analysis is performed to generate a parameter sensitivity weight matrix. This step quantitatively characterizes the impact of each parameter on coating quality, clarifies the optimization focus, and addresses the difficulty in quantifying parameter importance in traditional optimization. Fourth, the parameter sensitivity weight matrix and pre-set recipe constraints are input into an adaptive simulated annealing algorithm, which uses intelligent iterative calculations to determine the optimal coating solution recipe. This algorithm balances global exploration with local refinement by adaptively adjusting the search step size to temperature, effectively avoiding the risk of falling into a local optimum. The resulting optimal recipe satisfies various constraints while achieving the best coating effect, addressing the limited search capabilities of traditional optimization methods. Finally, based on the recipe instructions, a hierarchical integral sliding mode control system in a distributed architecture drives the metering pump control unit and the agitator control unit to execute coating solution configuration. This advanced nonlinear control method exhibits excellent interference resistance and robustness, capable of handling various disturbances such as raw material batch fluctuations and changing environmental conditions, ensuring precise execution of the configuration process and addressing the limited precision and interference resistance of traditional PID control. It is worth emphasizing that the advantages of the self-organizing map network of the present invention in processing high-dimensional nonlinear data enable the complex relationship between coating liquid parameters to be effectively captured and analyzed, greatly improving the accuracy, stability and adaptability of coating liquid configuration. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1Schematic diagram of an embodiment of a coating liquid configuration optimization method based on adaptive control in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of a coating liquid configuration optimization system based on adaptive control in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application provide a method, system and storage medium for optimizing the configuration of a coating liquid based on adaptive control. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the coating liquid configuration optimization method based on adaptive control includes:

[0019] Step S101: using the temperature sensor, viscosity sensor, pH sensor and solid content detector provided in the mixing tank to collect and analyze parameter data during the coating liquid preparation process, and obtain the coating liquid parameter feature vector and environmental influencing factor data set;

[0020] Step S102: establishing a multi-layer coupling mapping model between coating liquid parameters and coating quality based on the coating liquid parameter feature vector and the environmental influencing factor data set, performing parameter sensitivity quantitative analysis, and generating a parameter sensitivity weight matrix;

[0021] Step S103: Input the parameter sensitivity weight matrix and the preset formula constraint conditions into the adaptive simulated annealing algorithm, iteratively calculate the optimal formula of the coating liquid, and output the formula instructions including the precise addition amount of the components and the mixing parameters;

[0022] Step S104 : Based on the recipe instruction, the metering pump control unit and the stirrer control unit are driven to perform coating liquid configuration through the hierarchical integral sliding mode control in the distributed architecture.

[0023] It is understandable that the execution subject of the present application can be a coating liquid configuration optimization system based on adaptive control, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

[0024] Specifically, this is achieved through the strategic placement of multiple sensors within the mixing tank. Specifically, the mixing tank is equipped with temperature sensors, viscosity sensors, pH sensors, and solids content detectors, distributed at various locations within the tank to form a sensor network. The temperature sensor uses a PT100 resistance thermometer with an accuracy of ±0.1°C; the viscosity sensor uses a rotational viscometer with a measurement range of 20-20,000 mPa·s; the pH sensor uses a glass electrode pH meter with an accuracy of ±0.01 pH; and the solids content detector uses a near-infrared reflectance analyzer with an accuracy of ±0.5%. These sensors collect data at a high frequency, set at no less than 10 times per second, to capture subtle changes during the coating solution preparation process. Furthermore, a weighing system is used to record the actual addition amounts and ratios of the resin, solvent, additive, and functional components during the preparation process. The collected time series data is then subjected to curvature-resistant segmented filtering, numerical domain conversion, and wavelet denoising. These three processing methods work synergistically to effectively remove noise from the data. Curvature-resistance segmented filtering processes data in segments, using different filter coefficients for each segment to adapt to the data characteristics at different stages. Numerical domain conversion converts parameters of different dimensions into a unified range. Wavelet denoising effectively removes high-frequency noise through wavelet decomposition and reconstruction. In practical applications, when the pH value data of BOPA film coating fluids show a sudden change, the curvature-resistance segmented filtering automatically adjusts the filter coefficients, preserving the actual change while filtering out noise interference. This processed data forms a characteristic vector of coating fluid parameters, which is then analyzed in correlation with component addition data to construct a database of corresponding relationships between coating fluid parameters and coating quality indicators. Statistical analysis of this database determines the valid ranges and initial configurations for each coating fluid parameter. Furthermore, external factors such as ambient temperature, humidity, and substrate tension are collected and integrated with the initial parameter set to form a dataset of environmental influencing factors.

[0025] The coating solution parameter feature vectors were classified based on component factors, process factors, and environmental factors, and a three-dimensional feature matrix was constructed. The three-dimensional feature matrix has the structure [parameter type × number of parameters × time series], which intuitively reflects the temporal variation of each parameter. The three-dimensional feature matrix was then input into a self-organizing map network, where topology-preserving dimensionality reduction of the feature space was performed using a competitive learning mechanism. The self-organizing map network employs a six-layer structure: the input layer receives the three-dimensional feature matrix, the nodes in the competition layer are arranged in a 20×20 hexagonal topology, the middle layer includes a weight transfer layer, a distance calculation layer, and a winner determination layer, and the output layer is a two-dimensional topological plane. During network training, the Euclidean distance between the input vector and the weight vector of the competition layer neurons is calculated. The neuron with the smallest distance is determined as the winner. The weights of the winner and its neighboring neurons are then updated using a time-varying neighborhood function. After training, a parameter cluster map is generated through U-matrix analysis, which visually demonstrates the topological relationships between the parameters. Based on the parameter cluster mapping diagram, Gaussian process regression with multi-level cross-validation was used to construct the nonlinear functional relationship P = f(C, T, E), where P represents the coating fluid performance parameters, C represents the component parameters, T represents the process parameters, and E represents the environmental parameters. Gaussian process regression captures the nonlinear relationships between parameters by defining a kernel function, while multi-level cross-validation ensures the model's generalization. The resulting multi-layer coupled mapping model accurately characterizes the complex relationship between coating fluid parameters and coating quality. To quantify the impact of each parameter on coating quality, a perturbation input matrix D was introduced into the multi-layer coupled mapping model, and the output response change matrix R was calculated. The parameter gradient was calculated by calculating the ratio of the response change to the perturbation input, quantitatively characterizing the influence of each parameter on the output. Variance decomposition and main effects analysis were performed on the parameter impact quantification results to identify key parameter combinations and their interactions, forming a parameter interaction matrix. The parameter interaction matrix was subjected to eigenvalue decomposition, and the main eigenvectors were extracted and assigned weights to generate a parameter sensitivity weight matrix.

[0026] After obtaining the parameter sensitivity weight matrix, the iterative calculation phase for the optimal coating fluid formula begins. The parameter sensitivity weight matrix and the preset formula constraints are input into the adaptive simulated annealing algorithm. First, a multi-objective optimization function is defined, using the weight values in the parameter sensitivity weight matrix as weight coefficients for each optimization sub-objective to construct a weighted objective function. Preset formula constraints include a component sum of 100%, upper and lower limits for each component content, and solvent-to-solid ratio requirements. These constraints are converted into hard boundary conditions. A real number encoding method is used to construct the coating fluid formula state vector, with the vector dimension equal to the sum of the number of components and the number of mixing parameters. The initial temperature parameter is set to 1000, and the cooling coefficient is set to 0.95, generating an initial formula state that meets the constraints. During the iteration process, perturbations are performed on the current formula state to generate adjacent formula states. The perturbation step size is proportional to the current temperature, forming a large-scale search. The objective function values of the adjacent formula states are calculated and compared with the current state. If the adjacent state's objective function value is better, it is accepted directly. If the adjacent state's objective function value is worse, the worse state is accepted with a certain probability according to the Metropolis criterion. This mechanism helps the algorithm escape from local optimal solutions. Multiple perturbation-evaluation-acceptance cycles are performed at each temperature, followed by decreasing the temperature parameter. Calculations terminate when the temperature drops to the preset termination temperature or when there is no significant change after multiple iterations. The optimal recipe state vector obtained from the final iteration is decoded into the precise component addition amounts and mixing parameter values, forming the recipe instructions.

[0027] Finally, based on the recipe instructions, precise configuration of the coating fluid is achieved through hierarchical integral sliding mode control in a distributed architecture. The recipe instructions are parsed into component addition amount control targets and mixing parameter control targets, which are then transmitted to the top-level decision layer of the distributed control architecture via an industrial communication protocol. A hierarchical integral sliding mode controller is constructed in the top-level decision layer, with weighted sliding surface functions for the state deviation, deviation change rate, and deviation integral. Hierarchical integral sliding mode control has excellent anti-interference performance and can effectively cope with various disturbances in the production process. After calculation, the control law is decomposed into component addition control instructions and mixing control instructions. The component addition control instructions are transmitted to the middle-level process control layer, where they are converted into an execution sequence for the metering pump control unit, including the liquid injection rate, injection time, and injection sequence, generating the metering pump drive signal. The mixing control instructions are converted into an execution sequence for the agitator control unit, including the speed curve, stirring time, and stirring power, generating the agitator drive signal. These drive signals are transmitted via a control bus to the various actuators in the lower-level device control layer to accurately configure the coating fluid. During the execution process, the underlying equipment continuously collects real-time feedback data, calculates the deviation from the control target, and inputs the deviation signal into the hierarchical integral sliding mode controller for online adjustment.

[0028] In the embodiments of this application,

[0029] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0030] The actual addition amount and proportion of resin components, solvent components, additive components and functional components are recorded through the weighing system to form component addition data;

[0031] The data collected by the temperature sensor, viscosity sensor, pH sensor and solid content detector are collected at a frequency of not less than 10 times per second to generate time series data of the entire coating liquid preparation process;

[0032] Perform curvature resistance segmentation filtering, numerical domain conversion processing and wavelet denoising on time series data to generate coating liquid parameter feature vectors;

[0033] Conduct correlation analysis between component addition data and coating liquid parameter feature vectors to build a database of corresponding relationships between coating liquid parameters and coating quality indicators;

[0034] Perform statistical analysis based on the corresponding relationship database to determine the effective range and initial configuration scheme of each parameter of the coating liquid, and form an initial parameter set for the coating liquid configuration;

[0035] The environmental temperature, environmental humidity, and substrate tension factors are collected and integrated with the initial parameter set to generate an environmental influencing factor data set.

[0036] Specifically, a weighing system records the actual addition amount and proportion of each component. This weighing system refers to a high-precision electronic weighing device installed on the mixing tank feed line, with a resolution of 0.01 kg and a measurement range of 0-500 kg. The BOPA film coating solution preparation process requires precise recording of the addition amounts of the resin, solvent, additive, and functional components. The resin component primarily refers to polyamide resin, the primary substance that forms the coating matrix; solvents, such as water and alcohols, dissolve the resin to form the liquid coating solution; additives, such as dispersants and defoamers, adjust the coating solution's properties; and functional components, such as antistatic agents and antioxidants, impart specific functionalities to the coating. The weighing system records the mass change of each component in real time during addition, directly calculating the actual addition amount and proportion of each component to generate component addition data. This data is stored in a structured table, containing information such as component name, addition time, addition amount, cumulative addition amount, and relative proportions, providing essential information for subsequent analysis. As components are added, data collected by various sensors within the mixing tank is sampled at high frequency. The temperature sensor uses a PT100 platinum resistance temperature sensor with a measurement range of -50°C to 150°C; the viscosity sensor uses an online rotational viscometer with a measurement range of 20-20,000 mPa·s; the pH sensor uses a glass electrode pH meter with a measurement range of 0-14 pH; and the solids content detector uses a near-infrared reflectance analyzer with a measurement range of 0-100%. These sensors collect data at a rate of no less than 10 times per second, significantly higher than the traditional sampling rate of 1-2 times per minute. The purpose of high-frequency sampling is to capture transient changes in coating fluid parameters, particularly sudden changes that may occur during component addition. The time series data generated by high-frequency sampling includes three dimensions: sensor type, sampling timestamp, and measurement value, forming a multidimensional time series data matrix. This high-frequency sampling method overcomes the technical problem that traditional low-frequency sampling has difficulty capturing subtle changes and brief fluctuations during the coating fluid preparation process.

[0037] The collected time series data requires three steps of processing: curvature-tolerant segmented filtering, numerical domain conversion, and wavelet denoising. Curvature-tolerant segmented filtering is a filtering method for non-stationary time series. It first segments the time series according to the amplitude of parameter changes. A low damping coefficient is used to preserve the characteristics of the changes in the rapidly changing segments, while a high damping coefficient is used to filter out noise in the relatively stable segments. Specifically, the time series is divided into fast-changing and slow-changing segments based on the rate of change. A low damping coefficient of 0.2-0.4 is used for the fast-changing segments, and a high damping coefficient of 0.6-0.8 is used for the slow-changing segments. Weighted average filtering is performed. Numerical domain conversion unifies parameters of different physical quantities and dimensions onto the same scale. Using the maximum-minimum transformation method, all parameters are mapped to the [0, 1] interval to facilitate subsequent comprehensive analysis. Wavelet denoising leverages the multi-resolution analysis capabilities of the wavelet transform to decompose the signal, set a threshold to eliminate high-frequency noise components, and then reconstruct the denoised signal. In the actual processing, the db4 wavelet was used, with a decomposition layer of four, and the coefficients were processed using the soft thresholding method. After these three steps, the time series data had significantly reduced noise components, a more regular data structure, and a standardized coating solution parameter feature vector.

[0038] Correlation analysis is performed between component addition data and coating fluid parameter feature vectors to construct a database of corresponding relationships between coating fluid parameters and coating quality indicators. This correlation analysis first determines a time alignment strategy, precisely matching component addition times with parameter change times to establish a temporal relationship between component addition and parameter changes. The correlation coefficient between changes in component addition and parameter changes is then calculated to identify strongly correlated component-parameter pairs. Coating quality indicators such as uniformity, adhesion, and surface finish are also included to construct a ternary relationship matrix: component addition amount - coating fluid parameters - coating quality. This ternary relationship matrix is stored in a structured database, supporting multidimensional query and analysis. This correlation analysis overcomes the technical challenge of establishing the complex relationships between component ratios, coating fluid parameters, and final coating quality using traditional methods. Based on this constructed database of corresponding relationships, statistical analysis is used to determine the effective range and initial configuration of each coating fluid parameter. The statistical analysis includes three levels: descriptive statistics, correlation analysis, and regression analysis. Descriptive statistics calculate the distribution characteristics of each parameter when obtaining high-quality coating effects, including minimum, maximum, mean, standard deviation, quartiles, etc., to determine the effective range of the parameters. Correlation analysis calculates the Pearson correlation coefficient and partial correlation coefficient between parameters to identify the dependency between parameters. Regression analysis establishes a functional relationship between coating quality and multiple parameters to identify key parameters that have a significant impact on coating quality. Based on these analysis results, the optimal addition ratio, optimal addition order, optimal mixing parameters, etc. of each component of the coating liquid are determined to form an initial parameter set for the coating liquid configuration. This data-driven parameter optimization method solves the technical problem that traditional empirical formulas are difficult to cope with raw material fluctuations and environmental changes.

[0039] External factors such as ambient temperature, humidity, and substrate tension are collected and integrated with the initial parameter set to generate a dataset of environmental influencing factors. Ambient temperature is collected using a factory temperature sensor, humidity using a humidity sensor, and substrate tension using a tension sensor. These environmental factors directly impact the performance of the coating fluid: for example, temperature affects viscosity, humidity affects solvent evaporation rate, and substrate tension affects coating thickness. By integrating this environmental factor data with the initial parameter set and using timestamp alignment, a mapping between coating fluid parameters and environmental factors is established. This generates a comprehensive dataset encompassing all influencing factors, resolving the problem of traditional coating fluid formulation methods neglecting environmental influences.

[0040] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0041] The coating liquid parameter feature vectors are classified into component factors, process factors and environmental factors to construct a three-dimensional feature matrix;

[0042] The three-dimensional feature matrix is input into the self-organizing map network, and the feature space is topologically preserved and dimensionally reduced through a competitive learning mechanism to generate a parameter clustering map.

[0043] Based on the parameter cluster mapping diagram, a nonlinear functional relationship P = f(C, T, E) was constructed using multi-level cross-validation Gaussian process regression, where P represents the coating fluid performance parameters, C represents the component parameters, T represents the process parameters, and E represents the environmental parameters. This forms a multi-layer coupling mapping model between coating fluid parameters and coating quality.

[0044] The disturbance input matrix D is introduced into the multi-layer coupled mapping model, and the output response change matrix R is calculated. The parameter gradient is solved by R / D, and the influence of each parameter on the output is quantitatively characterized to obtain the quantitative results of parameter influence.

[0045] Perform variance decomposition and main effect analysis on the quantitative results of parameter impact, identify key parameter combinations and their interaction effects, and form a parameter interaction matrix;

[0046] The parameter interaction matrix is subjected to eigenvalue decomposition, the main eigenvectors are extracted and assigned normalized weights to generate the parameter sensitivity weight matrix.

[0047] Specifically, the coating fluid parameter feature vectors are factor-classified and structured. Classifying the coating fluid parameter feature vectors into component factors, process factors, and environmental factors involves categorizing and organizing the parameter feature vectors generated in the previous stage according to different dimensions. Component factors refer to parameters directly related to the coating fluid's composition, including resin content, solvent ratio, additive concentration, and functional component content. Process factors refer to process parameters involved in the coating fluid preparation process, including mixing time, mixing speed, mixing temperature, and addition sequence. Environmental factors include ambient temperature, humidity, and substrate tension. This classification approach addresses the issues of mixed parameters and ambiguous relationships in traditional coating fluid preparation, providing a clear data structure for subsequent modeling. The classified data forms a three-dimensional feature matrix, with the three dimensions corresponding to parameter type (component / process / environment), specific parameter indicator, and time series. Each matrix element represents the parameter value of a specific type and indicator at a specific point in time. This three-dimensional structure intuitively reflects the relationships between different parameter types and the patterns of their changes over time.

[0048] The three-dimensional feature matrix is input into a self-organizing map network to achieve dimensionality reduction and visualization of the feature space. A self-organizing map network is an unsupervised learning neural network that maps high-dimensional data to a low-dimensional space through a competitive learning mechanism while maintaining the topological relationships between data points. In this method, the self-organizing map network consists of an input layer, a competitive layer, and an output layer. The number of nodes in the input layer is equal to the dimensionality of the flattened three-dimensional feature matrix. The nodes in the competitive layer are arranged in a 20×20 hexagonal topology, with each node representing a prototype vector in the feature space. The output layer is a two-dimensional plane used to display the results of the dimensionality reduction. The network training process uses a batch training approach. First, the input data is normalized. Then, the Euclidean distance between each input vector and the weight vectors of all competitive layer neurons is calculated. The neuron with the smallest distance is called the winner neuron and represents the position of the input vector in the output space. The weights of the winner neuron and its neighbors are then updated using the following update formula: the new weight value is equal to the old weight value plus the learning rate multiplied by (the input vector minus the old weight value) multiplied by the neighborhood function. The neighborhood function typically uses a Gaussian function. As training progresses, the neighborhood radius and learning rate gradually decrease, allowing the network to converge. After training is complete, the distances between adjacent neuron weight vectors are analyzed using a U-matrix (unified distance matrix) to generate a parameter cluster map. The color depth on the map represents the degree of similarity between parameters. Areas with similar colors indicate similarities between parameters, while areas with clear boundaries indicate differences between parameters. This visualization method solves the problem of difficult intuitive analysis of high-dimensional data in traditional coating solution parameter analysis.

[0049] Based on the parameter cluster map, Gaussian process regression with multi-level cross-validation was used to construct nonlinear functional relationships. Gaussian process regression is a nonparametric Bayesian method that captures nonlinear relationships in data by defining a kernel function. It is suitable for processing small sample sizes, high-dimensional, and nonlinear data. In this method, Gaussian process regression is used to establish the functional relationship between coating fluid parameters and coating quality: P = f(C, T, E), where P represents coating fluid performance parameters (such as viscosity, solids content, and pH), C represents component parameters, T represents process parameters, and E represents environmental parameters. The construction process first selects an appropriate kernel function, typically a radial basis function (RBF), which effectively captures nonlinear relationships between parameters. Multi-level cross-validation is then used to determine the kernel function's hyperparameters. The dataset is divided into training and validation sets, and the optimal hyperparameter combination is found through grid search or Bayesian optimization. During model training, the covariance matrix of the training data is calculated, and the posterior distribution of the predicted points is derived using the Bayesian formula. This multi-level cross-validation method improves the model's generalization ability and overcomes the overfitting and underfitting issues of traditional modeling methods. The resulting multi-layer coupling mapping model accurately characterizes the complex relationship between coating fluid parameters and coating quality, solving the technical problem that traditional methods have difficulty in establishing accurate models.

[0050] Parameter sensitivity analysis of the multilayer coupled mapping model requires the introduction of a perturbation input matrix D for quantitative evaluation. The perturbation input matrix D represents small changes to the original input parameters. Each element in the matrix corresponds to the perturbation of an input parameter, typically a small fraction of the parameter's standard deviation. The original and perturbated parameters are input into the multilayer coupled mapping model, yielding two sets of outputs. The output response change matrix R is then calculated, with the elements in R representing the changes in the output parameters. By calculating the ratio of R to D, the parameter gradient—the sensitivity of the output to the input—is obtained. The calculation involves applying a perturbation to each input parameter while keeping all other parameters constant and observing the change in the output. This method is known as local sensitivity analysis. The calculated parameter gradients are normalized to quantify the parameter impact. Parameters with large values have a significant impact on the output, while parameters with small values have a weaker impact. This sensitivity analysis method addresses the technical challenge of quantifying parameter importance in traditional coating fluid configuration. The parameter impact quantification results are subjected to variance decomposition and main effects analysis to identify interactions between the parameters. Variance decomposition decomposes the output variance into the contributions of each input parameter and its interaction term, calculating the contribution of each parameter and parameter combination to the output variability. The specific method is to construct parameter combinations for a full factorial design, calculate the variance of the model output under different parameter combinations, and then perform an ANOVA analysis to determine the main and interaction effects of each parameter. The main effect represents the influence of a single parameter on the output, while the interaction effect represents the influence of a combination of parameters on the output. By calculating the interaction effect coefficients between parameters, a parameter interaction matrix is formed, where the elements aij in the matrix represent the interaction strength between parameters i and j. This analytical method solves the technical problem of neglecting parameter interactions in traditional coating solution configuration, providing a more comprehensive basis for formulation optimization.

[0051] The parameter interaction matrix is subjected to eigenvalue decomposition, and the main eigenvectors are extracted and assigned weights. Eigenvalue decomposition is a method in linear algebra that decomposes a matrix into a combination of eigenvalues and eigenvectors. The parameter interaction matrix A is subjected to eigenvalue decomposition to obtain a set of eigenvalues and corresponding eigenvectors. The size of the eigenvalue represents the importance of the corresponding eigenvector. The eigenvalues are sorted by size, and the eigenvectors corresponding to the first k eigenvalues are selected. These eigenvectors represent the main interaction modes between the parameters. The selected eigenvectors are then normalized to ensure that the sum of the weights is 1, forming a parameter sensitivity weight matrix W. The elements wij in W represent the weights of parameter i in eigenvector j. These weights reflect the comprehensive influence of the parameters on the coating quality. The weight calculation method solves the problem of subjective and lack of basis for parameter weight setting in traditional formula optimization.

[0052] In a specific embodiment, the process of inputting the three-dimensional feature matrix into the self-organizing map network may specifically include the following steps:

[0053] A six-layer self-organizing map network was constructed, in which the number of nodes in the input layer matched the dimensions of the three-dimensional feature matrix. The nodes in the competition layer were arranged in a 20×20 hexagonal topology. The middle layer included a weight transfer layer, a distance calculation layer, and a winner determination layer. The output layer was a two-dimensional topological plane.

[0054] Perform Z-Score normalization on the three-dimensional feature matrix to eliminate the scale differences between parameters of different dimensions and generate a standardized feature matrix;

[0055] The standardized feature matrix is input into the input layer of the self-organizing map network in batches, and the Euclidean distance between the input vector and the weight vector of the neuron in each competitive layer is calculated through the weight transfer layer to form a distance matrix;

[0056] Based on the distance matrix, a minimum distance search is performed in the distance calculation layer to determine the position of the winner neuron, and the winner neuron index is output through the winner determination layer;

[0057] According to the winner neuron index, the weight vectors of the winner neuron and its topological neighborhood in the competition layer are updated using a time-varying neighborhood function and an adaptive learning rate. The weight update amount is inversely proportional to the neighborhood distance, and training is iterated until the convergence condition is met.

[0058] The trained self-organizing map network is subjected to U-matrix analysis to calculate the average distance between adjacent neuron weight vectors. The topological relationship between parameters is visualized through a heat map to generate a parameter clustering map.

[0059] Specifically, a six-layer self-organizing map (SOM) network was constructed. This structure is significantly more complex than traditional two-layer SOM networks and is capable of processing high-dimensional nonlinear data. The network consists of six layers: an input layer, a weight transfer layer, a distance calculation layer, a winner determination layer, a competition layer, and an output layer. The number of nodes in the input layer is the same as the dimensionality of the three-dimensional feature matrix—that is, the total number of parameters for the component, process, and environmental factors. Since the configuration of BOPA film coating solutions typically contains 20-30 parameters, the number of nodes in the input layer is set accordingly. The competition layer is the core layer of the network, with nodes arranged in a 20×20 hexagonal topology. This structure provides better continuity and neighborhood consistency than a square grid. Each node in the competition layer is fully connected to the input layer, and the connection weight represents the node's position in parameter space. The three middle layers each perform different functions: the weight transfer layer receives data from the input layer and transmits it to each node in the competition layer; the distance calculation layer calculates the distance between the input vector and the weight vector of each node in the competition layer; and the winner determination layer determines the winning node based on this distance. The output layer is a two-dimensional topological plane used to visualize the dimensionality reduction results of high-dimensional data. This multi-layered network structure effectively solves the problem of complex parameter relationships and difficulty in intuitive expression in traditional coating fluid parameter analysis.

[0060] Before inputting data into the SOM network, the three-dimensional feature matrix needs to be Z-score normalized. Z-score normalization is a commonly used data preprocessing method. It transforms the raw data to a mean of 0 and a standard deviation of 1. This is calculated by subtracting the mean from the raw value and then dividing by the standard deviation. Coating fluid parameters, such as resin content, mixing temperature, and ambient humidity, vary greatly in physical dimensions and numerical ranges, requiring normalization to ensure fair comparison within the same network. Normalization first calculates the mean and standard deviation of each parameter, then applies a Z-score transformation to each data point. For example, for coating fluid viscosity, raw data may range from 1000 to 5000 mPa·s, while pH may range from 5 to 9. These differences can bias network training towards parameters with larger values. Z-score normalization transforms all parameters to the same scale, ensuring that each parameter has an equal chance of influencing the network training results. The resulting normalized feature matrix contains dimensionless, normalized values with a mean of 0 and a standard deviation of 1. This standardization method effectively solves the weight imbalance problem caused by different parameter dimensions in traditional coating fluid parameter analysis.

[0061] The normalized feature matrix is input to the input layer of the self-organizing map network in batches for processing. Batching is used to improve training efficiency. Each batch contains a number of samples, typically set to 16-32. Each input sample is transferred to each neuron in the competitive layer via a weight transfer layer. During the weight transfer process, the Euclidean distance between the input vector and the weight vector of each competitive layer neuron is calculated. Euclidean distance is a common method for measuring the similarity between two vectors; the smaller the Euclidean distance, the more similar the two vectors are. For coating fluid parameter analysis, Euclidean distance calculation reveals the degree of similarity between the current parameter combination and historical parameter combinations. During the calculation, the sum of the squared differences between the input vector X and the weight vector W is calculated, and the square root is taken to obtain the distance value. These distance values form a distance matrix, in which each element represents the distance between the input vector and a neuron in the competitive layer. The dimension of the distance matrix is batch size × number of competitive layer neurons. This Euclidean distance-based similarity calculation solves the technical problem of quantitatively measuring the similarity of parameter combinations in traditional coating fluid parameter analysis.

[0062] Based on the distance matrix, a minimum distance search is performed in the distance calculation layer to determine the location of the winner neuron. Minimum distance search involves finding the neuron index corresponding to the minimum value in the distance matrix. In coating fluid parameter analysis, this step is equivalent to finding the historical parameter combination that is most similar to the current parameter combination. The search process employs a global comparison strategy, comparing the distance between the input vector and all neurons in the competition layer and selecting the neuron with the smallest distance as the winner. For each input sample, a unique winner neuron is determined. This one-to-one mapping is the foundation of dimensionality reduction achieved by self-organizing maps. After the winner neuron is determined, the winner neuron index is output by the winner determination layer. The index is a two-dimensional coordinate (i, j), representing the location of the winner neuron in the competition layer. This minimum distance search method efficiently maps high-dimensional parameter space to two-dimensional topological space, addressing the technical challenge of visualizing high-dimensional data in traditional coating fluid parameter analysis. After the winner neuron is determined, the neuron weights are updated. Weight update is the core of self-organizing map learning and is performed using a time-varying neighborhood function and an adaptive learning rate. A time-varying neighborhood function defines the update range for the winner neuron and its neighboring neurons. As training progresses, the neighborhood range gradually decreases. The neighborhood function is typically a Gaussian function, reaching a maximum of 1 at the winner neuron and decaying with increasing distance from the winner neuron. An adaptive learning rate controls the step size of weight updates, starting with a large value and gradually decreasing as training progresses. The weight update rule is: the new weight equals the old weight plus the learning rate multiplied by the neighborhood function value multiplied by (the input vector minus the old weight). This update method causes the winner and its neighboring neurons to move toward the input sample, with the magnitude of the movement inversely proportional to the neighborhood distance. For coating fluid parameter analysis, this process is equivalent to continuously adjusting the partitioning of the parameter space, mapping similar parameter combinations to adjacent topological regions. The update process is iterative until convergence conditions are met, typically when the number of training rounds reaches a set value or the average quantization error falls below a threshold. This competitive learning-based weight update strategy addresses the problem of irrational parameter space partitioning in traditional coating fluid parameter analysis.

[0063] A U-matrix analysis was performed on the trained self-organizing map network to generate a parameter cluster map. The U-matrix (unified distance matrix) is a method for visualizing the distance relationship between neurons. The distance matrix is constructed by calculating the Euclidean distance between the weight vectors of adjacent neurons. For a 20×20 competitive layer, the size of the U-matrix is 39×39, which is twice as large as the original network minus one, because nodes representing distances need to be inserted between adjacent neurons. In the U-matrix, areas with large distance values represent the class boundaries of the parameter space, and areas with small distance values represent the internal areas of the class. The U-matrix is visualized using a heat map, using different colors to represent different distance values. Usually, dark colors represent high distance values (class boundaries) and light colors represent low distance values (inside the class). This visualization method intuitively presents the topological relationship between parameters and forms a parameter cluster map. In the coating liquid parameter analysis, the cluster map reveals the similarities and differences between different parameter combinations. Similar parameters will cluster in the same area of the map, while parameters that are far apart will be located in different areas. The topology-preserving dimensionality reduction visualization method solves the problem of difficulty in understanding the relationship between high-dimensional data in traditional coating fluid parameter analysis.

[0064] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0065] Define a multi-objective optimization function, use the weight values in the parameter sensitivity weight matrix as the weight coefficients of each optimization sub-objective, construct a weighted objective function, and convert the preset recipe constraints into hard constraint boundary conditions;

[0066] The coating liquid formula state vector is constructed using real number coding, where the dimension of the formula state vector is equal to the sum of the number of components and the number of mixing parameters. The initial temperature parameter and cooling coefficient are set to generate the initial formula state.

[0067] Perform a perturbation operation on the initial recipe state to generate an adjacent recipe state. The perturbation step is proportional to the current temperature. Calculate the objective function value of the adjacent recipe state to form a perturbation state evaluation result.

[0068] Based on the disturbance state evaluation results, the Metropolis criterion is used to determine whether to accept the new state. The acceptance probability is related to the current temperature and the change in the objective function, and the current recipe state is updated.

[0069] According to the Markov chain length setting, the perturbation-evaluation-acceptance cycle is repeated multiple times at each temperature, and then the temperature parameter is reduced. The calculation is stopped when the temperature drops to the preset termination temperature or there is no significant change after multiple consecutive iterations;

[0070] The optimal recipe state vector obtained by the final iteration is decoded into the precise addition amount of components and mixing parameter values to generate recipe instructions.

[0071] Specifically, a multi-objective optimization function is defined, converting the parameter sensitivity weight matrix into specific optimization objectives. A multi-objective optimization function involves simultaneously optimizing multiple, potentially conflicting, objective functions. In the case of coating fluid configuration, these objectives include performance indicators such as viscosity suitability, drying speed, and surface uniformity. The elements in the parameter sensitivity weight matrix represent the degree of influence of each parameter on coating quality, with larger values indicating a more significant impact. These weights are used as weight coefficients for each optimization sub-objective, constructing a weighted objective function. The objective function is equal to the sum of each sub-objective function multiplied by its corresponding weight. For example, for the three sub-objectives of coating fluid viscosity, drying speed, and surface uniformity, assuming their weights are 0.5, 0.3, and 0.2, respectively, the weighted objective function is the sum of these three sub-objectives multiplied by their respective weights. Furthermore, coating fluid configuration is subject to various constraints, such as upper and lower limits on component content, a component sum of 100%, and specific component ratios. These constraints are converted into boundary conditions for the optimization problem. Hard boundary conditions are strict constraints that must be met; any solution that violates these constraints is rejected during the iteration process. This multi-objective weighted optimization method solves the problems of single objective and inflexible constraint processing in traditional coating liquid formulation optimization.

[0072] After defining the optimization function, the coating fluid formulation state vector is constructed using real number encoding. Real number encoding directly represents points in the solution space using real numbers and is more suitable for continuous variable optimization problems than binary encoding. In the coating fluid configuration, each element of the coating fluid formulation state vector represents a specific parameter, such as resin content, solvent ratio, mixing temperature, etc. The vector dimension is equal to the sum of the number of components and the number of mixing parameters. For example, for a coating fluid containing four components and three mixing parameters, the state vector dimension is 7. Next, the initial temperature parameter and cooling coefficient are set. The initial temperature is typically set to a large value, such as 1000, to indicate a wide initial search range. The cooling coefficient is typically set between 0.9 and 0.99, such as 0.95, to control the temperature drop rate. The initial formulation state is then generated by randomly selecting a starting point within the solution space that satisfies the constraints, or by using a known good formulation as the initial solution. The initial formulation state must satisfy all hard constraints, such as ensuring that the component contents are within the permitted range and that their sum totals reach 100%. This real number encoding and random initialization method solves the problems of single initial solution selection and limited search space in traditional coating liquid formula optimization.

[0073] Starting from the initial solution, the current recipe state is perturbed to generate adjacent recipe states. Perturbation is a core step in the simulated annealing algorithm, exploring for better solutions by randomly moving around the current solution. The perturbation is performed by adding or subtracting a random value from each element of the state vector, with the magnitude of the random value proportional to the current temperature. Higher temperatures increase the perturbation step size, broadening the search range; lower temperatures decrease the perturbation step size, resulting in a more refined search. This temperature-dependent adaptive perturbation step size strategy enables broad exploration early in the algorithm and fine-tuning later, balancing global search with local optimization. After perturbation, each element in the coating fluid recipe state vector is checked for constraint compliance. If it exceeds the bounds, a rebound or truncation strategy is employed to ensure constraint compliance. The objective function value corresponding to the perturbed recipe state is calculated and compared with the objective function value of the current state to form an evaluation result for the perturbed state. This evaluation result, including the change in the objective function and constraint satisfaction, is used to determine whether the new state should be accepted. This perturbation search mechanism addresses the problem of traditional coating fluid recipe optimization, which suffers from a single search direction and a tendency to fall into local optima.

[0074] Based on the perturbed state evaluation results, the Metropolis criterion is used to determine whether to accept the new state. The Metropolis criterion is the core decision rule of the simulated annealing algorithm, allowing the algorithm to accept a less favorable solution with a certain probability, thereby escaping the local optimum. The decision rule is: if the new state's objective function value is better than the current state, it is accepted directly; if the new state's objective function value is worse than the current state, it is accepted with a certain probability, equal to the power of the exponential function e, where the exponent is equal to the negative value of the change in the objective function divided by the current temperature. This decision mechanism has a higher probability of accepting a less favorable solution at high temperatures and gradually decreases with decreasing temperature. For coating fluid formulation optimization, this random acceptance mechanism can help avoid falling into a local optimum when exploring different formulation combinations. If the new state is accepted, the current formulation state is updated to the new state; otherwise, the current state remains unchanged. The perturbation-evaluation-acceptance cycle is repeated multiple times at each temperature, depending on the Markov chain length. The Markov chain length refers to the number of searches performed at the current temperature and is typically set as a multiple of the state vector dimension to ensure that the solution space is fully explored at each temperature. For coating fluid formulation optimization, a reasonable Markov chain length setting can help find the relatively optimal formulation at each temperature. After completing the search at the current temperature, the temperature parameter is lowered by multiplying the current temperature by the cooling coefficient. As the temperature gradually decreases, the search range gradually narrows, and the algorithm gradually shifts from global exploration to local, refined search. The calculation terminates when the temperature drops to a preset termination temperature (e.g., 0.01) or when there is no significant change after multiple iterations (e.g., the objective function changes by less than 0.0001 after 10 consecutive iterations).

[0075] The optimal formula state vector obtained by the final iteration needs to be decoded into an actual executable coating liquid configuration instruction. The decoding process converts the numerical results of the optimization algorithm into specific component addition amounts and mixing parameter values. For the component addition amount, it is necessary to ensure that the total amount meets the production requirements, and it is usually scaled up to the actual production batch; for the mixing parameters, the optimization results are directly used as control parameters. The decoded formula instructions include detailed parameters such as the precise addition amount of each component (usually accurate to 0.1kg or higher), addition order, mixing temperature (accurate to 1°C), mixing speed (accurate to 1rpm), and mixing time (accurate to 1 minute). These parameters form a complete formula instruction for subsequent coating liquid configuration execution.

[0076] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0077] Parse the recipe instructions into component addition control targets and mixing parameter control targets, and transmit them to the top-level decision-making layer of the distributed control architecture through industrial communication protocols;

[0078] A hierarchical integral sliding mode controller is constructed in the top decision layer for the control target. The weighted sliding mode surface function of the state deviation, deviation change rate and deviation integral is set. The control law is calculated and decomposed into component addition control instructions and mixed control instructions.

[0079] The component addition control instructions are transmitted to the middle process control layer, converted into the execution sequence of the metering pump control unit, including the liquid injection rate, liquid injection time and liquid injection sequence, and the metering pump drive signal is generated;

[0080] The mixing control instructions are transmitted to the middle process control layer and converted into the execution sequence of the agitator control unit, including the speed curve, mixing time and mixing power, to generate the agitator drive signal;

[0081] Transmit metering pump drive signals and agitator drive signals to the actuators of the underlying equipment control layer through the control bus to execute precise configuration of the coating liquid;

[0082] Real-time feedback data is collected through the underlying equipment, the deviation from the control target is calculated, and the deviation signal is input into the hierarchical integral sliding mode controller for online adjustment.

[0083] Specifically, the recipe instructions are parsed into specific control objectives. Recipe instructions refer to the optimal coating liquid formula calculated by the adaptive simulated annealing algorithm, including precise component addition amounts and mixing parameter values. The parsing process categorizes recipe instructions into two categories: component addition amount control objectives and mixing parameter control objectives. Component addition amount control objectives refer to target values for the addition amounts of individual components, such as resin, solvent, additive, and functional components, typically accurate to 0.1 kg. Mixing parameter control objectives refer to target values for process parameters such as mixing temperature, mixing speed, and mixing time, with accuracies of 1°C, 1 rpm, and 1 minute, respectively. These parsed control objectives are transmitted to the top-level decision layer of the distributed control architecture via industrial communication protocols. Industrial communication protocols are data communication standards specifically designed for industrial control, such as OPC UA, PROFINET, and EtherCAT. These protocols offer high real-time performance and strong anti-interference capabilities, making them suitable for industrial control scenarios. The distributed control architecture is a multi-layered control system structure. The top-level decision layer is responsible for global decision-making and control strategy generation, and receives the parsed control objective data. Within the top-level decision layer, a hierarchical integral sliding mode controller is implemented to achieve high-precision control. Hierarchical integral sliding mode control is an advanced nonlinear control method that combines the advantages of hierarchical control and sliding mode control and is suitable for high-precision tracking control problems. For the coating solution preparation process, the hierarchical integral sliding mode controller can cope with the nonlinear, time-varying characteristics and external disturbances of component addition and mixing. Building the controller first requires setting a weighted sliding surface function for the state deviation, the rate of change of the deviation, and the integral of the deviation. The state deviation refers to the difference between the actual value and the target value, such as the difference between the actual amount of resin added and the target amount. The rate of change of the deviation is the speed of change of the deviation, reflecting the dynamic trend of the deviation. The integral of the deviation is the accumulation of the deviation over time and is used to eliminate static errors. The sliding surface function combines these three parameters in a weighted manner to form the target trajectory of the control system. Based on the sliding surface function, the control law is calculated. The control law is the core calculation formula of the controller and determines the control signals of the actuators. The calculated control law is then decomposed into the component addition control command and the mixing control command, which are used to control the metering pump and the agitator, respectively.

[0084] After component addition control instructions are transmitted to the middle process control layer, they need to be converted into specific execution sequences for the metering pump control unit. The middle process control layer, the middle layer of the distributed architecture, is responsible for translating high-level control instructions into specific instruction sequences executable by lower-level devices. For component addition control instructions, the conversion process considers factors such as the metering pump's execution characteristics, flow range, and response time to generate a detailed execution sequence. This execution sequence includes three key elements: injection rate, injection time, and injection sequence. The injection rate refers to the metering pump's flow setpoint, typically measured in kg / min, and is determined based on the component properties and addition amount. The injection time refers to the duration of each component addition, calculated from the addition amount and the injection rate. The injection sequence refers to the order in which the components are added, which significantly impacts the final coating fluid performance. These execution parameters constitute a complete metering pump control sequence, which further generates metering pump drive signals, including start / stop signals, speed signals, and direction signals. This hierarchical control instruction conversion approach addresses the issues of inaccurate component addition and crude addition process control in traditional coating fluid configuration.

[0085] Similar to component addition control, mixing control instructions are also transmitted to the middle-level process control layer and converted into an execution sequence for the agitator control unit. Mixing control instructions mainly include three parameters: mixing temperature, mixing speed, and mixing time. These parameters determine the uniformity and stability of the coating liquid. In the middle-level process control, these parameters are converted into specific execution sequences, including speed curves, stirring time, and stirring power. The speed curve refers to the change in agitator speed over time. It is not a simple constant speed, but a variable speed curve designed according to the viscosity changes of the coating liquid and the mixing stage; the stirring time refers to the duration of the entire stirring process, usually accurate to the minute level; the stirring power refers to the input power setting of the agitator, which is related to the speed and coating liquid viscosity. These execution parameters constitute a complete agitator control sequence, which further generates the agitator drive signal, including the motor speed signal, temperature control signal, and timing signal.

[0086] The metering pump and agitator drive signals are transmitted via a control bus to the actuators in the underlying device control layer, enabling precise configuration of the coating solution. The control bus is a communication network within an industrial control system, such as a fieldbus or industrial Ethernet, connecting controllers and actuators. The underlying device control layer is the lowest level of the distributed architecture and directly interacts with physical devices, including hardware such as metering pumps, agitators, valves, and sensors. After receiving the drive signals, the actuators precisely execute the coating solution configuration actions according to the instructions, including adding components at specified rates, controlling the agitation speed according to a set curve, and maintaining the required mixing temperature. Throughout the configuration process, sensors in the underlying devices collect real-time feedback data, and deviations from the control targets are calculated. Feedback data includes the actual weight of added components, actual mixing temperature, and actual agitation speed. This data is collected via the sensor network and transmitted back to the control system. The deviation calculation compares the real-time feedback data with the control target to calculate the state deviation, the rate of change of the deviation, and the integral of the deviation. The calculated results serve as deviation signals and are input into a hierarchical integral sliding mode controller, which adjusts the control output in real time based on the current deviation, forming a closed-loop control system. This feedback-based real-time adjustment mechanism can cope with various interferences and changes in the configuration process, such as raw material fluctuations, environmental changes, etc., to ensure that the final configuration results meet the expected goals.

[0087] In a specific embodiment, the process of transmitting the component addition control instruction to the middle process control layer may specifically include the following steps:

[0088] The addition amount parameter in the component addition control instruction is converted into a mass flow control point sequence, and a continuous flow curve is generated by using cubic spline interpolation between the flow control points;

[0089] Based on the continuous flow curve, the flow setting value and operating time of each metering pump are calculated, and feedforward compensation is performed considering the pump response characteristics to form the compensated metering pump control parameters;

[0090] Sort the metering pump control parameters according to the preset addition order and generate a timing control table containing the precise addition time point, flow rate change point and stop point of each component;

[0091] Perform conflict detection on the timing control table, identify and eliminate overlapping areas of addition time, optimize the order of component addition, and generate a conflict-free execution sequence;

[0092] Convert the conflict-free execution sequence into a control instruction format that can be recognized by the metering pump control unit, including start / stop signals, speed signals, and direction signals, to form the underlying control instructions of the metering pump;

[0093] The metering pump's underlying control instructions are transmitted to each metering pump control unit according to the preset communication protocol through the real-time data bus, and a feedback channel is established to receive the metering pump's actual operating status data.

[0094] Specifically, the addition amount parameters in the component addition control instructions are converted into a sequence of mass flow control points. A component addition control instruction refers to the component addition amount data transmitted from the upper-level control system, including the addition amount values for the resin component, solvent component, additive component, and functional component. A mass flow control point sequence is a set of discrete data points describing the flow setpoint values at different time points during the component addition process. The conversion process first identifies key flow control points, which generally include a start point, an end point, and intermediate characteristic points. The start point is set as the initial flow rate at the beginning of addition, typically the pump's minimum stable flow rate or a preset startup flow rate; the end point is set as the flow rate at the end of addition, typically zero flow rate; and the intermediate characteristic points are determined based on the component characteristics, addition amount, and process requirements. For example, high-viscosity components such as resins generally require lower addition flow rates, while low-viscosity components such as solvents can use higher addition flow rates. After determining these control points, cubic spline interpolation is used to generate a continuous flow curve for the region between the control points. Cubic spline interpolation is a mathematical method that generates smooth curves by connecting the control points using piecewise cubic polynomials, ensuring that the curve is continuous and smooth at the control points. This method generates a flow curve that not only passes through all control points but also avoids the sudden changes and oscillations associated with simple linear interpolation, ensuring smooth metering pump operation. The flow curve has a mathematical expression that describes how flow changes over time. Based on this continuous flow curve, the specific flow setpoint and run time for each metering pump are calculated. The flow setpoint refers to the flow control parameter of the metering pump at a specific time point, typically expressed in kg / min. The run time refers to the duration of the metering pump's operation, which determines the duration of the addition process. This calculation requires feedforward compensation, taking into account the pump's response characteristics. The pump response characteristic refers to the dynamic response of the metering pump from receiving the control signal to the actual output reaching the setpoint, including the response delay time and the shape of the response curve. Feedforward compensation is a control strategy that proactively adjusts the control signal to compensate for inherent delays and dynamic response deviations by taking into account the system's dynamic characteristics. This is achieved by applying time advance and amplitude correction to the flow curve. Time advance refers to issuing the control signal earlier to offset pump response delays, while amplitude correction refers to adjusting the flow setpoint amplitude based on the pump's dynamic gain characteristics. The compensated metering pump control parameters are formed through feedforward compensation. These parameters fully consider the actual dynamic characteristics of the pump body and can make the actual output of the metering pump closer to the ideal value.

[0095] The calculated metering pump control parameters are sorted according to a preset addition sequence to generate a timing control table. The preset addition sequence refers to the order in which the components are added, which has a significant impact on the coating fluid's performance. For example, for BOPA film coating fluids, the solvent is typically added first, followed by the resin, then additives and functional components, and finally the remaining solvent. This order of addition facilitates complete dissolution of the resin and uniform mixing of the components. The timing control table is a structured data table that records the key timing nodes and control parameters for each component throughout the coating fluid preparation process. The timing control table includes three key time nodes for each component: the precise addition time point, the flow rate change point, and the stop point. The precise addition time point refers to the moment when component addition begins; the flow rate change point refers to the moment when the flow rate changes, corresponding to a characteristic point on a cubic spline curve; and the stop point refers to the moment when component addition is complete. These time nodes and the corresponding flow rate setpoints constitute a complete timing control table, providing a comprehensive description of the time dimension for precise metering pump control.

[0096] Conflict detection is performed on the generated timing control table to ensure the coordinated addition of each component. Conflict detection is a data analysis method used to identify potential time and resource conflicts within the timing control table. A time conflict occurs when the addition times of different components overlap and the corresponding metering pump resources are insufficient; a resource conflict occurs when multiple components simultaneously use the same metering pump or share equipment. Conflict detection uses a time window scanning algorithm to scan the timing control table in chronological order, examining resource usage within each time window. When a conflict is detected, a conflict resolution strategy is implemented. Conflict resolution methods include time staggering and resource reallocation. Time staggering adjusts the addition times of conflicting components to stagger their execution; resource reallocation, when multiple metering pumps are present, allocates conflicting components to different metering pumps. Through conflict detection and resolution, the component addition sequence is optimized, generating a conflict-free execution sequence. This conflict-free execution sequence ensures the coordinated addition of each component and the rational use of resources during the coating solution preparation process, avoiding configuration anomalies caused by time and resource conflicts.

[0097] The conflict-free execution sequence is converted into a control instruction format recognizable by the metering pump control unit, forming the metering pump's low-level control instructions. The metering pump control unit is the hardware module that directly controls the metering pump's operation. It typically includes a microprocessor and a communication interface, capable of receiving upper-level control instructions and converting them into specific drive signals. The control instruction format refers to the data structure and communication protocol recognized by a specific metering pump control unit. Different brands and models of metering pumps may use different instruction formats. The conversion process requires converting the high-level control parameters in the timing control table into low-level drive signals, which primarily include start / stop signals, speed signals, and direction signals. The start / stop signal controls the metering pump's start and stop; the speed signal controls the metering pump's speed and is directly related to the flow rate; and the direction signal controls the direction of liquid flow and is particularly important for reversible metering pumps. These signals typically use standard industrial control signal formats, such as 4-20mA current signals, 0-10V voltage signals, or industrial fieldbus digital signals. The conversion process must consider the metering pump's characteristic parameters, such as the flow-speed characteristic curve, minimum stable flow rate, and maximum flow rate, to ensure that the converted control instructions accurately represent the original control intent. The metering pump's low-level control instructions are transmitted to each metering pump control unit via a real-time data bus, enabling precise control. A real-time data bus is a high-speed, low-latency industrial communication network, such as EtherCAT and PROFINET, that ensures real-time and reliable data transmission. The preset communication protocol refers to the data transmission rules and format on the bus, ensuring that the data content is correctly understood by both the sender and receiver. During the transmission process, control instructions are sent to each metering pump control unit according to the preset timing and protocol format. Upon receiving the instructions, the control unit drives the metering pump to perform the corresponding action. Simultaneously, a feedback channel is established to receive data on the metering pump's actual operating status. The feedback channel is the data transmission path from the metering pump to the control system, used to transmit the metering pump's actual operating status, such as actual flow rate, pump pressure, and motor current. This feedback data is transmitted back to the upper-level control system via the same real-time data bus, forming a closed-loop control loop. This feedback data is used to monitor the metering pump's operating status in real time, detect abnormalities, and provide a basis for real-time adjustments to the control system. This control method based on real-time communication and feedback solves the technical problems of traditional coating fluid configuration, which lack real-time monitoring and dynamic adjustment of control execution.

[0098] The above describes the coating liquid configuration optimization method based on adaptive control in the embodiment of the present application. The following describes the coating liquid configuration optimization system based on adaptive control in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the coating liquid configuration optimization system based on adaptive control includes:

[0099] The acquisition module 201 is used to collect and analyze parameter data during the coating liquid preparation process using the temperature sensor, viscosity sensor, pH sensor, and solid content detector provided in the mixing tank, and obtain the coating liquid parameter feature vector and environmental influencing factor data set;

[0100] An execution module 202 is used to establish a multi-layer coupling mapping model between coating liquid parameters and coating quality based on the coating liquid parameter feature vector and the environmental influencing factor data set, and generate a parameter sensitivity weight matrix after performing parameter sensitivity quantitative analysis;

[0101] Input module 203, for inputting the parameter sensitivity weight matrix and the preset formula constraint conditions into the adaptive simulated annealing algorithm, iteratively calculating the optimal formula of the coating liquid, and outputting the formula instructions including the precise addition amount of the components and the mixing parameters;

[0102] The control module 204 is configured to drive the metering pump control unit and the agitator control unit to configure the coating liquid based on the recipe instruction through hierarchical integral sliding mode control in a distributed architecture.

[0103] By integrating these components and building an adaptive control closed loop, the system addresses multiple technical challenges in the coating fluid preparation process, from data acquisition, model building, formulation optimization, to control execution. First, by utilizing multiple sensors installed within the mixing tank to collect parameter data at high frequency, a dataset of coating fluid parameter feature vectors and environmental influencing factors is obtained. This enables refined data capture of the entire coating fluid preparation process, improving data integrity and accuracy and addressing the low frequency and limited coverage of data collection in traditional coating fluid preparation processes. Second, based on this high-quality data, a multi-layer coupled mapping model of coating fluid parameters and coating quality is constructed using a self-organizing map network and a Gaussian process regression algorithm. This model accurately characterizes the complex, nonlinear relationship between coating fluid parameters and coating quality, providing a reliable theoretical basis for subsequent optimization and overcoming the shortcomings of traditional empirical models, such as low accuracy and weak generalization. In particular, the application of the self-organizing map network in parameter space dimensionality reduction and clustering enables intuitive visualization of complex multidimensional parameter relationships, significantly improving the model's interpretability. Third, a parameter sensitivity quantification analysis is performed to generate a parameter sensitivity weight matrix. This step quantitatively characterizes the impact of each parameter on coating quality, clarifies the optimization focus, and addresses the difficulty in quantifying parameter importance in traditional optimization. Fourth, the parameter sensitivity weight matrix and pre-set recipe constraints are input into an adaptive simulated annealing algorithm, which uses intelligent iterative calculations to determine the optimal coating solution recipe. This algorithm balances global exploration with local refinement by adaptively adjusting the search step size to temperature, effectively avoiding the risk of falling into a local optimum. The resulting optimal recipe satisfies various constraints while achieving the best coating effect, addressing the limited search capabilities of traditional optimization methods. Finally, based on the recipe instructions, a hierarchical integral sliding mode control system in a distributed architecture drives the metering pump control unit and the agitator control unit to execute coating solution configuration. This advanced nonlinear control method exhibits excellent interference resistance and robustness, capable of handling various disturbances such as raw material batch fluctuations and changing environmental conditions, ensuring precise execution of the configuration process and addressing the limited precision and interference resistance of traditional PID control. It is worth emphasizing that the advantages of the self-organizing map network of the present invention in processing high-dimensional nonlinear data enable the complex relationship between coating liquid parameters to be effectively captured and analyzed, greatly improving the accuracy, stability and adaptability of coating liquid configuration.

[0104] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the coating liquid configuration optimization method based on adaptive control.

[0105] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0106] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a coating liquid configuration optimization device based on adaptive control (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc., various media that can store program code.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A coating liquid configuration optimization method based on adaptive control, characterized in that: The method comprises: The temperature sensor, viscosity sensor, pH sensor and solid content detector installed in the mixing tank are used to collect and analyze parameter data during the coating liquid preparation process to obtain the coating liquid parameter feature vector and environmental influencing factor data set; Establishing a multi-layer coupling mapping model between coating liquid parameters and coating quality based on the coating liquid parameter characteristic vector and the environmental influencing factor data set, performing parameter sensitivity quantitative analysis, and generating a parameter sensitivity weight matrix; Inputting the parameter sensitivity weight matrix and the preset formula constraint conditions into the adaptive simulated annealing algorithm, iteratively calculating the optimal formula of the coating liquid, and outputting the formula instructions including the precise addition amount of the components and the mixing parameters; Based on the recipe instruction, the metering pump control unit and the stirrer control unit are driven to perform coating liquid configuration through hierarchical integral sliding mode control in a distributed architecture.

2. The coating liquid configuration optimization method based on adaptive control according to claim 1, characterized in that: The temperature sensor, viscosity sensor, pH sensor and solid content detector provided in the mixing tank are used to collect and analyze parameter data during the coating liquid preparation process to obtain the coating liquid parameter feature vector and environmental influencing factor data set, including: The actual addition amount and proportion of resin components, solvent components, additive components and functional components are recorded through the weighing system to form component addition data; Setting a collection frequency of not less than 10 times per second for the data collected by the temperature sensor, viscosity sensor, pH sensor, and solid content detector to generate time series data of the entire coating liquid preparation process; Performing curvature resistance segmented filtering, numerical domain conversion processing, and wavelet denoising on the time series data to generate a coating liquid parameter feature vector; Performing correlation analysis on the component addition data and the coating liquid parameter characteristic vectors to construct a database of corresponding relationships between coating liquid parameters and coating quality indicators; Performing statistical analysis based on the corresponding relationship database to determine the effective range and initial configuration scheme of each parameter of the coating liquid, and forming an initial parameter set for the coating liquid configuration; Environmental temperature, environmental humidity, and substrate tension factors are collected and integrated with the initial parameter set to generate an environmental influencing factor data set.

3. The coating liquid configuration optimization method based on adaptive control according to claim 1, characterized in that: The method includes establishing a multi-layer coupling mapping model between coating liquid parameters and coating quality based on the coating liquid parameter characteristic vector and the environmental influencing factor data set, performing parameter sensitivity quantitative analysis, and generating a parameter sensitivity weight matrix, including: Classifying the coating liquid parameter characteristic vector by component factors, process factors and environmental factors to construct a three-dimensional characteristic matrix; Inputting the three-dimensional feature matrix into a self-organizing map network, performing topology-preserving dimensionality reduction on the feature space through a competitive learning mechanism, and generating a parameter clustering map; Based on the parameter cluster mapping diagram, a nonlinear functional relationship P=f(C, T, E) is constructed using multi-level cross-validation Gaussian process regression, where P represents the coating liquid performance parameters, C represents the component parameters, T represents the process parameters, and E represents the environmental parameters, forming a multi-layer coupling mapping model of the coating liquid parameters and the coating quality; Introducing a disturbance input matrix D into the multi-layer coupled mapping model, calculating the output response change matrix R, solving the parameter gradient by R / D, quantitatively characterizing the influence of each parameter on the output, and obtaining the parameter influence quantification result; Perform variance decomposition and main effect analysis on the quantitative results of the parameter impact, identify key parameter combinations and their interaction effects, and form a parameter interaction matrix; The parameter interaction matrix is subjected to eigenvalue decomposition, the main eigenvectors are extracted and assigned normalized weights, and a parameter sensitivity weight matrix is generated.

4. The coating liquid configuration optimization method based on adaptive control according to claim 3, characterized in that: The three-dimensional feature matrix is input into a self-organizing map network, and the feature space is subjected to topology-preserving dimensionality reduction through a competitive learning mechanism to generate a parameter clustering map, including: A six-layer self-organizing map network was constructed, in which the number of nodes in the input layer matched the dimensions of the three-dimensional feature matrix. The nodes in the competition layer were arranged in a 20×20 hexagonal topology. The middle layer included a weight transfer layer, a distance calculation layer, and a winner determination layer. The output layer was a two-dimensional topological plane. Performing Z-Score normalization on the three-dimensional feature matrix to eliminate scale differences between parameters of different dimensions and generate a normalized feature matrix; Input the standardized feature matrix into the input layer of the self-organizing map network in batches, and calculate the Euclidean distance between the input vector and the neuron weight vector of each competitive layer through the weight transfer layer to form a distance matrix; Performing a minimum distance search in a distance calculation layer based on the distance matrix to determine a winner neuron position, and outputting a winner neuron index through a winner determination layer; According to the winner neuron index, a time-varying neighborhood function and an adaptive learning rate are used to update the weight vector of the winner neuron and the neurons in its topological neighborhood in the competition layer. The weight update amount is inversely proportional to the neighborhood distance, and the training is iterated until the convergence condition is met. The trained self-organizing map network is subjected to U-matrix analysis to calculate the average distance between adjacent neuron weight vectors, and the topological relationship between parameters is visualized through a heat map to generate a parameter clustering map.

5. The coating liquid configuration optimization method based on adaptive control according to claim 1, characterized in that: The parameter sensitivity weight matrix and the preset formula constraint conditions are input into the adaptive simulated annealing algorithm, the optimal formula of the coating liquid is iteratively calculated, and the formula instructions including the precise addition amount of the components and the mixing parameters are output, including: A multi-objective optimization function is defined, and the weight values in the parameter sensitivity weight matrix are used as weight coefficients of each optimization sub-objective to construct a weighted objective function, while converting the preset recipe constraints into hard constraint boundary conditions; The coating liquid formula state vector is constructed using real number coding, where the dimension of the formula state vector is equal to the sum of the number of components and the number of mixing parameters. The initial temperature parameter and cooling coefficient are set to generate the initial formula state. Performing a perturbation operation on the initial recipe state to generate an adjacent recipe state, wherein the perturbation step is proportional to the current temperature, calculating the objective function value of the adjacent recipe state, and forming a perturbation state evaluation result; Based on the disturbance state evaluation result, the Metropolis criterion is used to determine whether to accept the new state, the acceptance probability is related to the current temperature and the change in the objective function, and the current recipe state is updated; According to the Markov chain length setting, the perturbation-evaluation-acceptance cycle is repeated multiple times at each temperature, and then the temperature parameter is reduced. The calculation is stopped when the temperature drops to the preset termination temperature or there is no significant change after multiple consecutive iterations; The optimal recipe state vector obtained by the final iteration is decoded into the precise addition amount of components and mixing parameter values to generate recipe instructions.

6. The coating liquid configuration optimization method based on adaptive control according to claim 1, characterized in that: The method of driving the metering pump control unit and the agitator control unit to configure the coating liquid based on the recipe instruction through hierarchical integral sliding mode control in a distributed architecture includes: Parsing the recipe instructions into component addition amount control targets and mixing parameter control targets, and transmitting them to the top decision layer of the distributed control architecture via an industrial communication protocol; A hierarchical integral sliding mode controller is constructed in the top decision layer for the control target, a weighted sliding mode surface function of the state deviation, the deviation change rate and the deviation integral is set, the control law is calculated and decomposed into component addition control instructions and mixed control instructions; The component addition control instruction is transmitted to the middle process control layer, converted into an execution sequence of the metering pump control unit, including the liquid injection rate, liquid injection time and liquid injection sequence, and generates a metering pump drive signal; The mixing control instruction is transmitted to the middle process control layer and converted into an execution sequence of the agitator control unit, including a speed curve, agitation time and agitation power, to generate an agitator drive signal; Transmitting the metering pump drive signal and the stirrer drive signal to the actuators of the bottom device control layer through the control bus to execute the precise configuration of the coating liquid; Real-time feedback data is collected through the underlying equipment, the deviation from the control target is calculated, and the deviation signal is input into the hierarchical integral sliding mode controller for online adjustment.

7. The coating liquid configuration optimization method based on adaptive control according to claim 6, characterized in that: The component addition control instruction is transmitted to the middle process control layer, converted into an execution sequence of the metering pump control unit, including the liquid injection rate, liquid injection time and liquid injection sequence, and generates a metering pump drive signal, including: Converting the addition amount parameter in the component addition control instruction into a mass flow control point sequence, and using cubic spline interpolation between the flow control points to generate a continuous flow curve; Based on the continuous flow curve, the flow setting value and the operating time of each metering pump are calculated, and feedforward compensation is performed considering the response characteristics of the pump body to form compensated metering pump control parameters; Sorting the metering pump control parameters according to a preset addition order to generate a timing control table containing the precise addition time point, flow rate change point, and stop point of each component; Performing conflict detection on the timing control table, identifying and eliminating overlapping addition time areas, optimizing the order of component addition, and generating a conflict-free execution sequence; Converting the conflict-free execution sequence into a control instruction format recognizable by the metering pump control unit, including a start / stop signal, a speed signal, and a direction signal, to form a bottom-level control instruction for the metering pump; The metering pump bottom-level control instructions are transmitted to each metering pump control unit according to a preset communication protocol through a real-time data bus, and a feedback channel is established to receive the actual operating status data of the metering pump.

8. A coating liquid configuration optimization system based on adaptive control, characterized in that: For implementing the coating liquid configuration optimization method based on adaptive control according to any one of claims 1 to 7, the coating liquid configuration optimization system based on adaptive control comprises: The acquisition module is used to collect and analyze parameter data during the coating liquid preparation process using the temperature sensor, viscosity sensor, pH sensor and solid content detector set in the mixing tank to obtain the coating liquid parameter feature vector and environmental influencing factor data set; An execution module is used to establish a multi-layer coupling mapping model between coating liquid parameters and coating quality based on the coating liquid parameter feature vector and the environmental influencing factor data set, and generate a parameter sensitivity weight matrix after performing parameter sensitivity quantitative analysis; An input module is used to input the parameter sensitivity weight matrix and preset formula constraints into an adaptive simulated annealing algorithm, iteratively calculate the optimal formula of the coating liquid, and output a formula instruction including the precise addition amount of the components and the mixing parameters; The control module is used for driving the metering pump control unit and the stirrer control unit to execute coating liquid configuration based on the recipe instruction through hierarchical integral sliding mode control in a distributed architecture.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the coating liquid configuration optimization method based on adaptive control according to any one of claims 1 to 7.

Images