Coating formula calculation mode and system based on self-learning feedback parameter correction

Abstract

The invention relates to the field of coating material production, and discloses a coating formula calculation method and system based on self-learning feedback parameter correction, and the method comprises the steps: obtaining raw material basic parameters, target performance requirements and process feedback data of historical production batches, combining a material screening mechanism and an environment working condition correction strategy, and calculating a coating formula; constructing a coating formula initial input matrix; performing multi-dimensional variable normalization processing on the initial input matrix of the coating formula, introducing a feature sensitivity analysis model, and extracting a key variable group influencing the coating performance in the feature sensitivity analysis model; based on the performance response relation mapping model, analyzing error distribution between model prediction output and actual detection data by using a prediction deviation recognition mechanism, and extracting learning error features; a self-learning feedback updating mechanism is introduced, and dynamic weight optimization is conducted on the parameter correction factor set; and performing sample test and performance verification on the corrected candidate formula set. The method has the advantage that the coating formula precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of coating material production, and in particular to a coating formula calculation method and system based on self-learning feedback parameter correction. Background Art

[0002] In the coating material production process, the precise calculation of the formula directly impacts the performance and consistency of the finished product. Existing technologies typically rely on empirical formulas, manually set parameters, or static databases to calculate and generate coating formulas. However, even with the same initial formula, variations in the final coating result can still occur due to factors such as raw material batch differences, fluctuating environmental conditions, or unstable equipment operating conditions. Therefore, in actual production, frequent manual intervention and fine-tuning of formula parameters are often required to achieve the target performance, increasing the operational burden and reducing production efficiency. Currently, most mainstream coating formula calculation systems are based on fixed-weight models and lack an effective self-feedback correction mechanism. When the actual coating result deviates from the theoretical prediction, the system cannot automatically correct the formula parameters or update the calculation logic based on historical deviation data, resulting in poor formula adaptability, especially with new material combinations or non-standard environmental conditions. This problem is particularly prominent in production scenarios with high performance requirements or rapid batch switching. Therefore, it is necessary to design a coating formula calculation method and system based on self-learning feedback parameter correction to improve coating formula accuracy. Summary of the Invention

[0003] In response to the deficiencies of the prior art, the present invention provides a coating formula calculation method and system based on self-learning feedback parameter correction, which has the advantage of improving the accuracy of the coating formula and solves the problems in the above-mentioned background technology.

[0004] To achieve the above-mentioned purpose of improving the coating formula accuracy, the present invention provides the following technical solution: a coating formula calculation method based on self-learning feedback parameter correction, comprising the following steps:

[0005] Obtain basic raw material parameters, target performance requirements, and process feedback data from historical production batches. Combined with material screening mechanisms and environmental condition correction strategies, construct an initial input matrix for coating formulations.

[0006] Perform multi-dimensional variable normalization on the initial input matrix of the coating formula, introduce a feature sensitivity analysis model, extract the key variable groups that affect coating performance in the feature sensitivity analysis model, and establish a formula performance response relationship mapping model;

[0007] Based on the performance-response relationship mapping model, the prediction deviation identification mechanism is used to analyze the error distribution between the model prediction output and the actual detection data, extract the learning error characteristics, and construct a parameter correction factor set;

[0008] A self-learning feedback update mechanism is introduced to dynamically optimize the weight of the parameter correction factor set, and the recipe coefficients of the key variable group are corrected through an incremental iterative algorithm to generate a corrected candidate recipe set;

[0009] The revised candidate recipe set is subjected to small-sample testing and performance verification, and the mapping model is reversely trained based on the test data and the parameter correction factor set is updated to form an iterative process for recipe optimization.

[0010] Preferably, the process of constructing the initial input matrix of the coating recipe by combining the material screening mechanism and the environmental condition correction strategy is as follows:

[0011] Collect the chemical composition, physical properties and particle size distribution of each batch of raw materials, combined with the ambient temperature, humidity and air flow rate;

[0012] According to the coating process standards, the material screening mechanism is used to screen out batches of raw materials that do not meet the performance indicators, and the environmental data is corrected in real time to form the environmental condition weight coefficient;

[0013] The screened raw material parameters are combined with the environmental working condition weight coefficients, and a matrix integration algorithm is used to construct a multi-dimensional coating formula initial input matrix.

[0014] Preferably, the process of extracting the key variable group that affects the coating performance in the feature sensitivity analysis model is:

[0015] Using the normalized initial input matrix and sensitivity analysis algorithm, the correlation between each variable and coating thickness, adhesion and surface uniformity was quantified;

[0016] Through principal component analysis and multivariate regression methods, key variable groups whose contribution to performance indicator fluctuations exceeds the set threshold are screened out.

[0017] Preferably, the process of establishing the recipe performance response relationship mapping model is:

[0018] Based on the historical data of key variable groups, nonlinear regression is used to fit the response curve between variables and coating performance indicators;

[0019] Combined with Bayesian optimization method, the model parameters were adjusted;

[0020] Output recipe performance response relationship mapping model.

[0021] Preferably, the error distribution process between the prediction output of the model and the actual detection data is analyzed using the prediction deviation identification mechanism as follows:

[0022] Collect performance prediction values ​​output by the formula calculation model and performance data from actual production tests;

[0023] Through error statistical analysis, calculate the contribution distribution of each key variable to the forecast error and identify systematic deviations and random error components;

[0024] Feature extraction is performed on the error data to form a learning error feature vector.

[0025] Preferably, the process of constructing the parameter correction factor set is:

[0026] Based on the error eigenvector, the corresponding parameter correction factor is calculated using the weighted least squares method;

[0027] Introducing historical feedback data weights to dynamically adjust the initial value and update step size of the correction factor;

[0028] The parameter correction factor set is mapped to the recipe coefficient adjustment range of the key variable group to form a correction factor set applied to recipe optimization.

[0029] Preferably, the process of correcting the recipe coefficients of the key variable group by the incremental iterative algorithm to generate the corrected candidate recipe set is as follows:

[0030] Using the parameter correction factor set, incrementally adjust the recipe coefficients corresponding to the key variable group to generate a series of candidate recipe solutions;

[0031] Adopting iterative optimization algorithms, combined with preset performance objective functions and constraints, multiple rounds of screening and adjustment of candidate formulations are carried out;

[0032] Output the final revised candidate recipe set.

[0033] Preferably, the process of reverse training the mapping model based on the test data and updating the parameter correction factor set to form a recipe optimization iterative process is as follows:

[0034] Conduct small sample tests on the candidate formulation set and collect corresponding coating performance test data;

[0035] Using the test data and the prediction results of the mapping model for error feedback, reverse training is performed to update the parameters of the recipe performance response relationship mapping model;

[0036] Based on the updated mapping model, the prediction error is recalculated and the parameter correction factor set is corrected.

[0037] The coating formula calculation system based on self-learning feedback parameter correction includes:

[0038] Matrix building module: Integrates raw material parameters, performance requirements and historical process feedback, combines material screening and environmental correction, and generates the initial input data matrix for the coating formula;

[0039] Response modeling module: normalizes the input matrix, extracts key variable groups, and establishes a mapping relationship model between coating formula performance and variables;

[0040] Deviation identification module: Analyzes the error between the predicted output of the mapping relationship model and the actual detection data, identifies the deviation pattern, and constructs an error factor set for parameter correction;

[0041] Iterative correction module: By dynamically optimizing the error factor weights and using the incremental iterative algorithm to correct the recipe coefficients of key variables, an optimized candidate recipe set is generated;

[0042] Recipe optimization module: Experimentally verify candidate recipes, and reversely adjust the mapping model and parameter correction factors based on the test results to form an iterative process for recipe optimization.

[0043] Compared with the prior art, the present invention provides a coating formula calculation method and system based on self-learning feedback parameter correction, which has the following beneficial effects:

[0044] 1. By acquiring the basic parameters of raw materials, target performance requirements, and historical process feedback data, and introducing a material screening mechanism and environmental condition correction strategy, the compatibility of selected materials and the accuracy of operating condition data can be ensured at the source. By eliminating raw material batches that do not meet quality or performance requirements and correcting performance deviations caused by environmental changes, the quality and representativeness of input data are significantly improved, providing highly reliable basic data support for subsequent modeling, thereby reducing formula calculation errors and enhancing the system's initial decision-making capabilities.

[0045] 2. By performing multidimensional normalization on the initial input matrix, the numerical scales of different parameters are effectively unified, avoiding the uneven impact of large-scale variables on model training. The introduction of the feature sensitivity analysis model further achieves accurate quantification of the impact relationship between each input variable and coating performance indicators. Through principal component analysis and regression screening methods, it focuses on key variable groups with high correlation and high contribution. This not only simplifies the model structure and reduces computational complexity, but also enhances the model's ability to capture major influencing factors, thereby improving the pertinence and accuracy of performance prediction.

[0046] 3. By comparing the performance response mapping model with actual production test data, the prediction deviation identification mechanism is used to systematically extract the characteristics of the model prediction error, achieving accurate tracing of the cause of the model error. Combined with error statistical analysis, it can not only effectively distinguish between systematic errors and random errors, but also identify the specific contribution of key variables to the error, thereby forming a representative learning error feature vector. On this basis, a parameter correction factor set is constructed to provide a basis for the model's adaptive parameter adjustment, significantly improving the correction capability and robustness of the recipe prediction model.

[0047] 4. By introducing a self-learning feedback update mechanism and a dynamic weight adjustment strategy, the update process of the parameter correction factor can be continuously optimized based on actual error performance, enhancing the system's adaptability to different process environments and material batches. Incremental iterative algorithms are combined to fine-grainedly adjust the recipe coefficients of key variable groups to construct a set of candidate recipes with balanced performance. By setting performance objective functions and physical process constraints, candidate solutions are screened and optimized in multiple rounds. The final output recipe achieves an optimal balance between performance stability and feasibility, providing highly reliable recipe recommendations for actual production.

[0048] 5. By conducting small-scale test validation on the revised candidate formulations and collecting actual coating performance data, we can achieve an accurate comparison between model predictions and measured performance, providing first-hand data support for model revision. We also use experimental errors for reverse training to further update the performance mapping model parameters, enabling the model to continuously learn and evolve. At the same time, we modify the parameter correction factor set to implement a closed-loop self-optimization mechanism. This process achieves efficient linkage between models, data, and practical applications, significantly improving the intelligence, prediction accuracy, and optimization efficiency of coating formulation calculations, and providing reliable support for formulation development under complex process conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the method of the present invention;

[0050] Figure 2 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Example 1: Please refer to Figure 1 As shown, the coating formula calculation method based on self-learning feedback parameter correction described in the embodiment of the present invention includes the following steps:

[0053] S1: Obtain the basic parameters of raw materials, target performance requirements and process feedback data of historical production batches, combine the material screening mechanism with the environmental condition correction strategy, and build the initial input matrix of the coating formula.

[0054] The process of constructing the initial input matrix of the coating formula by combining the material screening mechanism and the environmental condition correction strategy in S1 is as follows:

[0055] Collect basic parameters such as chemical composition, physical properties and particle size distribution of each batch of raw materials, and combine them with working condition data such as ambient temperature, humidity, and air flow rate; collect data on each batch of raw materials and record their chemical composition, physical properties and particle size distribution; data sources include material test reports, supplier technical documents and laboratory measured values; collect key working condition parameters such as temperature, humidity, and air flow rate of the coating environment in real time; if there is multi-point data collection, perform regional weighted averaging or select representative points; use time synchronization mechanism to ensure that the material and working condition data match the current batch

[0056] Based on coating process standards, the material screening mechanism is used to screen out batches of raw materials that do not meet performance indicators, and environmental data is corrected in real time to form environmental condition weight coefficients. Based on the target performance requirements of the coated product, the raw material parameters are screened for index matching. The allowable error range is set to eliminate batches of raw materials that do not meet performance thresholds or have poor quality stability. A rule engine or machine learning model is used for automated screening. Based on historical experience models or statistical regression models, the impact of environmental factors on coating performance is analyzed. Parameters such as temperature, humidity, and wind speed are converted into environmental weight factors. Sensitive parameters of raw materials are corrected.

[0057] The screened raw material parameters are combined with the environmental operating condition weight coefficients, and a matrix integration algorithm is used to construct a multi-dimensional initial input matrix for the coating formula. The screened raw material parameters are dimensionally matched with the corrected environmental weight factors and matrixed in a unified format. Rows represent different raw material batches or samples, and columns represent material parameters, operating condition parameters, and their interaction characteristics. Normalization or standardization is used to unify the scales of each dimension to ensure data homogeneity. The output initial input matrix serves as the input data set for subsequent sensitivity analysis and model training.

[0058] S2: Perform multi-dimensional variable normalization on the initial input matrix of the coating formula, introduce a feature sensitivity analysis model, extract the key variable group that affects the coating performance in the feature sensitivity analysis model, and establish a formula performance response relationship mapping model.

[0059] The process of extracting the key variable group that affects the coating performance in the feature sensitivity analysis model in S2 is:

[0060] Using the normalized initial input matrix and combining it with a sensitivity analysis algorithm, the correlation between each variable and performance indicators such as coating thickness, adhesion, and surface uniformity is quantified. All variables in the initial input matrix are normalized to ensure that variables under different dimensions are on the same scale. Actual performance test data corresponding to the same batch of coating formulas is obtained, including indicators such as coating thickness, adhesion, and surface uniformity. A one-to-one mapping data set between input and output is constructed. A sensitivity analysis algorithm is used to calculate the contribution of each input variable to different performance indicators. For each performance indicator, the linear correlation coefficient and nonlinear correlation between the variable and the indicator are calculated. The results of various analyses are combined into a sensitivity scoring matrix, recording the sensitivity score of each variable under different performance dimensions.

[0061] Through principal component analysis and multivariate regression methods, the key variable groups whose contribution to the fluctuation of performance indicators exceeds the set threshold are screened out; principal component analysis is performed on the normalized variable set: the covariance matrix is ​​calculated and the principal components are extracted; the number of principal components whose cumulative explained variance exceeds the threshold is selected; the projection coefficients of the original variables in the principal components are used as a reference to determine which variables dominate in explaining the system variance; variables with extremely low variance contribution rates are eliminated, and the feature dimensions are preliminarily compressed to reduce redundancy; a multivariate linear regression algorithm is used to establish a mapping model between performance indicators and variables; a significance test is performed on the coefficient of each independent variable in the regression model; the final key variable group is selected based on the model determination coefficient, variable contribution and collinearity test; a performance contribution rate threshold is set, and a list of key variables is output.

[0062] The process of establishing the recipe performance response relationship mapping model in S2 is as follows:

[0063] Based on historical data from key variable groups, nonlinear regression is used to fit the response curves between variables and coating performance indicators. A nonlinear modeling algorithm is selected. The training set is input into the key variables, and the model is trained to fit the nonlinear relationship between the variables and various performance indicators. For each performance indicator, an independent regression model is constructed or a unified model is modeled using multi-output regression. The predictive performance of the model is tested in the validation set, and the error indicators are recorded.

[0064] Combined with Bayesian optimization methods, model parameters are adjusted to improve prediction accuracy and generalization ability; model hyperparameters that need to be optimized are determined; the objective function is set; and the Bayesian optimization algorithm is used to iteratively search for the optimal hyperparameter combination in the parameter space. In each iteration, the surrogate model predicts the optimal point, and then the model performance is actually evaluated and the surrogate is updated. When the optimization meets the early stopping condition or reaches the maximum number of iterations, the optimal parameter model is output.

[0065] Output the formulation-performance-response relationship mapping model; retrain the final model using the optimal hyperparameters and complete training data; encapsulate the trained model into a mapping module, accept the key variable group as input, and output the predicted coating performance indicators; save the model structure, weight file, input feature format, and response function; output a performance evaluation report, including training error, validation error, residual distribution, performance curve visualization, etc.

[0066] The technical solution of this embodiment is: performing multi-dimensional variable normalization processing on the constructed initial input matrix of the coating formula to unify the numerical scales of various raw material parameters and environmental working condition data; then introducing a characteristic sensitivity analysis model, and extracting the key variable group that has a significant impact on the coating performance indicators through correlation analysis, principal component extraction and regression evaluation; combining historical data with a nonlinear fitting method to construct a formula performance response relationship mapping model to achieve an effective mapping between formula parameters and performance outputs; through normalization processing, the comparability and modeling stability between multi-source heterogeneous data are improved, and the interference of data scale on model accuracy is effectively reduced; the sensitivity analysis link realizes noise reduction and screening of variable dimensions, highlights the key factors that significantly affect coating performance, and reduces model complexity; the performance response mapping model finally established has good prediction accuracy and generalization ability, which can provide reliable performance prediction support for subsequent formula optimization and parameter correction, and significantly improve the scientificity and efficiency of coating formula design.

[0067] Example 2: Figure 1 As shown, the coating formula calculation method based on self-learning feedback parameter correction also includes the following steps:

[0068] S3: Based on the performance-response relationship mapping model, the prediction deviation identification mechanism is used to analyze the error distribution between the model prediction output and the actual detection data, extract the learning error characteristics, and construct a parameter correction factor set.

[0069] The error distribution process between the prediction output and the actual detection data analyzed by the prediction deviation identification mechanism in S3 is:

[0070] Collect performance predictions from the formula calculation model and performance data from actual production tests. For each batch of production, obtain the corresponding key variable inputs. Use the established formula-performance-response relationship mapping model to predict performance and obtain predicted values. Extract the actual performance values ​​from the actual testing of the batch from the quality inspection data system. Compare the model prediction output with the actual test results, and record the predicted value-actual value pairing data by batch number. Establish an error sample database and form a unified error record table.

[0071] Through error statistical analysis, calculate the contribution distribution of each key variable to the prediction error and identify systematic bias and random error components; calculate the error value Δ = predicted value - actual value for each performance indicator; draw error histograms, box plots, and residual plots to observe the distribution; determine whether the error is symmetrical around 0 and whether there is any offset; if most errors show unilateral deviation, it indicates systematic bias; calculate the standard deviation and variance of the error; determine whether the error fluctuates with environmental variables or recipe variables; construct a regression model of the error and key variables, or use SHAP value analysis; output the contribution rate ranking of each key variable to the error change to identify error-sensitive characteristics;

[0072] Extract features from the error data to form a learning error feature vector. For each error sample, record the corresponding key variable input, environmental parameters, and prediction error. An example of the row structure for each sample is: [particle size, viscosity, humidity, temperature, flow rate, predicted thickness error, predicted adhesion error, ...]. Use the AutoEncoder method to reduce the dimension of the error features. Extract high-correlation error patterns, such as the consistent low thickness prediction under high humidity and low temperature conditions. The error feature vector of each sample includes: error value, main influencing variable value, and error cluster category. The vector format is: E i =[Δthickness, Δadhesion, principal factor 1 value, principal factor 2 value, deviation direction label]; All error feature vectors are aggregated into a training data set.

[0073] The process of constructing the parameter correction factor set in S3 is:

[0074] Based on the error eigenvector, the weighted least square method is used to calculate the corresponding parameter correction factor; let the key variable group be X=[x1,x2,...,x n ], the error eigenvector is E=[e1,e2,...,e n ]; construct an error fitting function, take the error characteristic vector as the target value and the key variable group as the input; use the weighted least squares method to build a linear regression model and solve the parameter correction factor;

[0075] The weight of historical feedback data is introduced to dynamically adjust the initial value and update step of the correction factor to improve the stability and convergence speed of the correction process. The feedback data of historical batches are introduced and weighted according to their error stability and sample credibility. The weight factor of historical data is expressed as:

[0076]

[0077] For each key variable, set the initial value of the correction factor; dynamically set the update step size based on the convergence speed and volatility of the historical correction iteration;

[0078] Map the parameter correction factor set to the recipe coefficient adjustment range of the key variable group to form a correction factor set for recipe optimization; for each key variable x j , set the allowed correction range The correction factor Δθ j Limit to the allowed range:

[0079]

[0080] S4: Introduce a self-learning feedback update mechanism to dynamically optimize the weights of the parameter correction factor set, and correct the recipe coefficients of the key variable group through an incremental iterative algorithm to generate a corrected candidate recipe set.

[0081] The process of correcting the recipe coefficients of the key variable group by the incremental iterative algorithm in S4 to generate the corrected candidate recipe set is as follows:

[0082] Using parameter correction factor sets, incremental adjustments are made to the recipe coefficients corresponding to the key variable groups to generate a series of candidate recipe solutions. Key variable groups and their basic recipe coefficients, such as solvent ratio, binder content, and additive concentration, are read. The parameter correction factor set, output by the error learning model in the previous stage, is extracted, including the correction direction and correction amplitude weight for each variable. The recipe adjustment interval and adjustment granularity are set, that is, each variable is assigned an adjustable range and step interval. Based on the correction factor direction and amplitude, an incremental adjustment algorithm is used to generate multiple candidate variable combinations. All combined variable groups are converted into complete recipe parameter sets, and verification is performed to ensure that basic process boundary conditions are met, such as that the material content does not exceed the limit, the total does not exceed 100%, and that safety or processing requirements are not violated.

[0083] Adopting an iterative optimization algorithm, combined with preset performance objective functions and constraints, candidate formulations are screened and adjusted multiple times to ensure that the formulation achieves the optimal balance between performance and stability. Constraints are set, including total raw material quantity constraints, upper and lower limits for single raw materials, and environmental adaptability conditions. An optimization algorithm, such as genetic algorithm (GA), particle swarm optimization (PSO), or simulated annealing (SA), is selected to initialize the candidate formulation population. Multiple rounds of optimization iterations are performed, each of which includes: using a performance prediction model to simulate and predict the performance indicators of each candidate formulation; calculating the objective function value and constraint penalty terms; applying crossover, mutation, or search strategies to generate a new round of candidates; retaining formulations with excellent performance indicators and high prediction stability for the next round; monitoring convergence and diversity, terminating if the objective function change is less than a threshold or the number of iterations reaches an upper limit; and outputting a set of converged formulations, which have achieved local or near-optimal results in terms of prediction performance, stability, and feasibility.

[0084] Output the final revised candidate formula set; sort and screen the final optimized formula solutions based on: minimum performance target error; reasonable parameter change range; high stability score; record the detailed parameters of each candidate formula and the corresponding performance prediction values, such as: raw material ratio; predicted thickness, adhesion, uniformity values; corresponding correction factors, number of iterations and convergence status.

[0085] S5: Conduct sample tests and performance verification on the revised candidate recipe set, combine the test data to reversely train the mapping model and update the parameter correction factor set to form an iterative process for recipe optimization.

[0086] In S5, the mapping model is reversely trained based on the test data and the parameter correction factor set is updated to form a recipe optimization iterative process:

[0087] Conduct sample tests on the candidate formulation set and collect corresponding coating performance test data; select representative candidate formulation samples, giving priority to formulations with high performance scores and balanced variable distribution in the prediction for testing; prepare coating samples according to process standards, and control production conditions to be consistent with the batch process; carry out standardized testing projects and perform the following performance tests on each sample:

[0088] Coating thickness: measured by non-contact thickness measuring instrument;

[0089] Adhesion: evaluated by pull / peel test;

[0090] Surface uniformity scoring: dual-channel acquisition combining image recognition and human inspection scoring;

[0091] Record the correspondence between test data and formula numbers to ensure data traceability and backtracking;

[0092] The experimental data and the prediction results of the mapping model are used for error feedback, and reverse training is used to update the parameters of the recipe performance response relationship mapping model. The actual test value of each candidate recipe is compared with the predicted value of the mapping model to calculate the performance deviation index. An error sample set is constructed to evaluate whether there is nonlinear offset or variable mismatch in the model error structure. An appropriate reverse training algorithm is selected to perform incremental learning training on the original mapping model. A sample confidence weight mechanism is introduced to assign higher training weights to samples with smaller errors but higher stability to avoid overfitting. A verification mechanism is introduced into the training process, using some retained samples for cross-validation to monitor whether the model improves the overall prediction accuracy after reverse correction.

[0093] Based on the updated mapping model, the prediction error is recalculated and the parameter correction factor set is corrected; the performance of the same batch of candidate recipes is re-predicted using the updated mapping model to obtain the corrected prediction value; the performance deviation of each recipe is recalculated and compared with the previous error to identify whether there is a new error structure or convergence trend; the corrected error feature vector is extracted, including error direction, amplitude, variable influence weight, etc.; time decay weight is introduced to fuse historical feedback, so that the recent feedback data has a higher weight for factor update and improves dynamic adaptability; and the updated parameter correction factor set is output.

[0094] Example 3: Please refer to Figure 2 As shown, the coating formula calculation system based on self-learning feedback parameter correction includes:

[0095] Matrix building module: Integrates raw material parameters, performance requirements and historical process feedback, combines material screening and environmental correction, and generates the initial input data matrix for the coating formula;

[0096] Response modeling module: normalizes the input matrix, extracts key variable groups, and establishes a mapping relationship model between coating formula performance and variables;

[0097] Deviation identification module: Analyzes the error between the predicted output of the mapping relationship model and the actual detection data, identifies the deviation pattern, and constructs an error factor set for parameter correction;

[0098] Iterative correction module: By dynamically optimizing the error factor weights and using the incremental iterative algorithm to correct the recipe coefficients of key variables, an optimized candidate recipe set is generated;

[0099] Recipe optimization module: Experimentally verify candidate recipes, and reversely adjust the mapping model and parameter correction factors based on the test results to form an iterative process for recipe optimization.

[0100] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0101] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A coating formula calculation method based on self-learning feedback parameter correction, characterized in that: The following steps are involved: Obtain basic raw material parameters, target performance requirements, and process feedback data from historical production batches. Combined with material screening mechanisms and environmental condition correction strategies, construct an initial input matrix for coating formulations. Perform multi-dimensional variable normalization on the initial input matrix of the coating formula, introduce a feature sensitivity analysis model, extract the key variable groups that affect coating performance in the feature sensitivity analysis model, and establish a formula performance response relationship mapping model; Based on the performance-response relationship mapping model, the prediction deviation identification mechanism is used to analyze the error distribution between the model prediction output and the actual detection data, extract the learning error characteristics, and construct a parameter correction factor set; A self-learning feedback update mechanism is introduced to dynamically optimize the weight of the parameter correction factor set, and the recipe coefficients of the key variable group are corrected through an incremental iterative algorithm to generate a corrected candidate recipe set; The revised candidate recipe set is subjected to small-sample testing and performance verification, and the mapping model is reversely trained based on the test data and the parameter correction factor set is updated to form an iterative process for recipe optimization.

2. The coating formula calculation method based on self-learning feedback parameter correction according to claim 1 is characterized in that: The process of constructing the initial input matrix of coating formula by combining material screening mechanism and environmental condition correction strategy is as follows: Collect the chemical composition, physical properties and particle size distribution of each batch of raw materials, combined with the ambient temperature, humidity and air flow rate; According to the coating process standards, the material screening mechanism is used to screen out batches of raw materials that do not meet the performance indicators, and the environmental data is corrected in real time to form the environmental condition weight coefficient; The screened raw material parameters are combined with the environmental working condition weight coefficients, and a matrix integration algorithm is used to construct a multi-dimensional coating formula initial input matrix.

3. The coating formula calculation method based on self-learning feedback parameter correction according to claim 2 is characterized in that: The process of extracting the key variable group that affects coating performance in the feature sensitivity analysis model is as follows: Using the normalized initial input matrix and sensitivity analysis algorithm, the correlation between each variable and coating thickness, adhesion and surface uniformity was quantified; Through principal component analysis and multivariate regression methods, key variable groups whose contribution to performance indicator fluctuations exceeds the set threshold are screened out.

4. The coating formula calculation method based on self-learning feedback parameter correction according to claim 3 is characterized in that: The process of establishing the recipe performance response relationship mapping model is as follows: Based on the historical data of key variable groups, nonlinear regression is used to fit the response curve between variables and coating performance indicators; Combined with Bayesian optimization method, the model parameters were adjusted; Output recipe performance response relationship mapping model.

5. The coating formula calculation method based on self-learning feedback parameter correction according to claim 4 is characterized in that: The error distribution process between the prediction output and the actual detection data is analyzed using the prediction deviation identification mechanism: Collect performance prediction values ​​output by the formula calculation model and performance data from actual production tests; Through error statistical analysis, calculate the contribution distribution of each key variable to the forecast error and identify systematic deviations and random error components; Feature extraction is performed on the error data to form a learning error feature vector.

6. The coating formula calculation method based on self-learning feedback parameter correction according to claim 5 is characterized in that: The process of constructing the parameter correction factor set is: Based on the error eigenvector, the corresponding parameter correction factor is calculated using the weighted least squares method; Introducing historical feedback data weights to dynamically adjust the initial value and update step size of the correction factor; The parameter correction factor set is mapped to the recipe coefficient adjustment range of the key variable group to form a correction factor set applied to recipe optimization.

7. The coating formula calculation method based on self-learning feedback parameter correction according to claim 6 is characterized in that: The process of correcting the recipe coefficients of the key variable group through the incremental iterative algorithm to generate the corrected candidate recipe set is as follows: Using the parameter correction factor set, incrementally adjust the recipe coefficients corresponding to the key variable group to generate a series of candidate recipe solutions; Adopting iterative optimization algorithms, combined with preset performance objective functions and constraints, multiple rounds of screening and adjustment of candidate formulations are carried out; Output the final revised candidate recipe set.

8. The coating formula calculation method based on self-learning feedback parameter correction according to claim 7 is characterized in that: The process of reverse training the mapping model based on the experimental data and updating the parameter correction factor set to form the recipe optimization iterative process is as follows: Conduct small sample tests on the candidate formulation set and collect corresponding coating performance test data; Using the test data and the prediction results of the mapping model for error feedback, reverse training is performed to update the parameters of the recipe performance response relationship mapping model; Based on the updated mapping model, the prediction error is recalculated and the parameter correction factor set is corrected.

9. A coating formula calculation system based on self-learning feedback parameter correction, applied to the method according to any one of claims 1 to 8, characterized in that: include: Matrix building module: Integrates raw material parameters, performance requirements and historical process feedback, combines material screening and environmental correction, and generates the initial input data matrix for the coating formula; Response modeling module: normalizes the input matrix, extracts key variable groups, and establishes a mapping relationship model between coating formula performance and variables; Deviation identification module: Analyzes the error between the predicted output of the mapping relationship model and the actual detection data, identifies the deviation pattern, and constructs an error factor set for parameter correction; Iterative correction module: By dynamically optimizing the error factor weights and using the incremental iterative algorithm to correct the recipe coefficients of key variables, an optimized candidate recipe set is generated; Recipe optimization module: Experimentally verify candidate recipes, and reversely adjust the mapping model and parameter correction factors based on the test results to form an iterative process for recipe optimization.

Images