Control method and system for precise barrel plating process

Abstract

The invention discloses a control method and system for a precise barrel plating process. A deep learning quality prediction model for performing quality analysis through mutual linkage of a graph neural network, a self-attention mechanism, a multi-task learning structure and a generative adversarial network is established; a technological parameter relation network for describing and representing the mutual relation between technological parameters is constructed, and a graph neural network, a self-attention mechanism, a multi-task learning structure and a generative adversarial network which are combined and linked with each other are analyzed based on the technological parameter relation network. According to the method, the implicit relation of mutual influence among the process parameters and the mutual relation between the process parameters and the product quality can be mined, so that the accuracy of quality prediction is improved. And on the basis, the current process parameters are adjusted in real time according to the plating layer prediction quality parameters, so that real-time precise control is realized, the quality of produced products is improved, the production efficiency is improved, and the cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of roll plating, and in particular to a control method and system for a precision roll plating process. Background Art

[0002] The barrel plating process is an efficient metal surface treatment method, which is widely used in the electronics, automobile, aerospace and other industries. Its main function is to deposit a layer of metal or alloy coating on the surface of the workpiece to improve the corrosion resistance, wear resistance and aesthetics of the workpiece. The barrel plating process has the advantages of small equipment investment, uniform coating and high production efficiency.

[0003] Although the barrel plating process is widely used in many industries, the traditional process control method still has the following shortcomings:

[0004] Reliance on experience: The current roller plating process mostly relies on experienced operators to adjust parameters, lacks scientific and systematic control procedures, and easily leads to unstable product quality.

[0005] Insufficient real-time monitoring: In traditional processes, the monitoring of process parameters is mostly manual, which makes it difficult to achieve real-time feedback and adjust process conditions in a timely manner.

[0006] Quality fluctuations: Due to environmental factors, material differences, etc., the quality of the coating often fluctuates, affecting the market competitiveness of the product.

[0007] Low data utilization: The large amount of data generated in the process has not been effectively utilized, and there is a lack of in-depth analysis of historical data to provide guidance for subsequent production. Summary of the invention

[0008] The object of the present invention is to provide a control method and system for a precision roller plating process to solve the above-mentioned problems existing in the prior art.

[0009] An embodiment of the present invention provides a control method for a precision barrel plating process, comprising:

[0010] Data collection: During the roller plating process, the current process parameters are collected in real time, including the current current, current voltage, current temperature, current flow rate and current plating solution composition;

[0011] Data integration: build a process parameter relationship network based on current process parameters;

[0012] Quality prediction: Input the process parameter relationship network into the pre-trained deep learning quality prediction model in real time to obtain the predicted quality parameters of the coating.

[0013] Among them, the deep learning quality prediction model includes graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network. The input of the graph convolutional neural network is the process parameter relationship network, the input of the self-attention mechanism is the output of the graph neural network, the input of the multi-task learning structure is the output of the graph neural network, and the input of the generative adversarial network includes the output of the self-attention mechanism and the output of the multi-task learning structure; the output of the generative adversarial network is used to feedback and adjust the graph neural network, and the graph neural network outputs the coating prediction quality parameters after adjustment.

[0014] Optional, precision control: real-time adjustment of current process parameters based on predicted coating quality parameters to achieve real-time precision control;

[0015] Optionally, a process parameter relationship network is constructed based on the current process parameters, including:

[0016] Obtaining the correlation index between current process parameters;

[0017] The current process parameters are regarded as nodes, any two nodes are connected, the line between the two nodes is defined as an edge, the association relationship index is used as the weight of the edge, and an initial parameter relationship network is obtained;

[0018] The initial relationship network is optimized to obtain the current process parameter relationship network.

[0019] Optionally, obtain the correlation index between the current process parameters, including:

[0020] Correlation measurement: Use the correlation coefficient to quantify the relationship between two process parameters; the absolute value of the correlation coefficient is used as the correlation index between the two process parameters.

[0021] Optionally, the initial relationship network is optimized to obtain the current process parameter relationship network, including:

[0022] Influence analysis: If the current process parameters have a significant impact on the results during the experiment or production process, multiply the correlation index of the current process parameters by 1.1 to obtain the influence correlation index;

[0023] The influence correlation index is used as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

[0024] Optionally, the initial relationship network is optimized to obtain the current process parameter relationship network, which also includes:

[0025] Entropy measurement: obtain the information entropy of current process parameters;

[0026] The influence correlation index and information entropy are weighted and summed to obtain the optimized weight;

[0027] The optimization weight is used as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

[0028] Optionally, the initial relationship network is optimized to obtain the current process parameter relationship network, which also includes:

[0029] Entropy measurement: obtain the information entropy of current process parameters;

[0030] The correlation index and information entropy between the current process parameters are weighted and summed to obtain the optimization weight;

[0031] The optimization weight is used as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

[0032] Optional deep learning quality prediction model training methods include:

[0033] Obtaining a training sample, the training sample including a plurality of historical process parameters, the historical process parameters including historical current, historical voltage, historical temperature, historical flow rate and historical plating solution composition;

[0034] Building a training parameter relationship network based on historical process parameters; multiple historical process parameters include multiple training parameter relationship networks;

[0035] Input multiple training parameter relationship networks into a pre-trained deep learning quality prediction model, calculate the loss function of the deep learning quality prediction model, and when the loss function of the deep learning quality prediction model converges, determine that the training of the deep learning quality prediction model is completed, and obtain the trained deep learning quality prediction model.

[0036] Optionally, the loss function of the deep learning quality prediction model is calculated using the following formula:

[0037] L total =αL GNN +βL Transformer +γL MTL +δL GAN

[0038] Among them, L total Represents the loss function of the deep learning quality prediction model, L GNN represents the loss function of the graph neural network, L Transformer represents the loss function of the self-attention mechanism, L MTL Represents the loss function of the multi-task learning structure, L GAN Represents the loss function of the generative adversarial network; α, β, γ, δ are hyperparameters; the initial values ​​of the hyperparameters are α=0.6, β=0.1, γ=0.2, δ=0.1.

[0039] Optionally, each hyperparameter is associated with a model. Specifically, α corresponds to a graph neural network, β corresponds to a self-attention mechanism, γ corresponds to a multi-task learning structure, and δ corresponds to a generative adversarial network. The graph neural network, the self-attention mechanism, the multi-task learning structure, and the generative adversarial network are the first model, the second model, the third model, and the fourth model, respectively. For each hyperparameter, update it in the following way:

[0040] Calculate the rate of change of loss: ΔL k,t =L k,t -L k,t-1 , where ΔL k is the rate of change of the loss function of the k-th model at the t-th iteration; L k,t is the loss function of the k-th model at the t-th iteration, L k,t-1 is the loss function of the k-th model at the t-1th iteration; when t = 1, L k,t-1 =0; the value of t is an integer between 1 and T, and T is a positive integer greater than 5;

[0041] Dynamically adjust hyperparameters:

[0042] Calculate the hyperparameter adjustment weights. The hyperparameter adjustment weights are calculated as follows: w k,t+1 The hyperparameter adjustment weights for the kth model at the t+1th iteration, ∈ is a small constant, ∈=0.001;

[0043] Tuning Hyperparameters: θ k,t+1 =θ k,t +μ k (θ k,t +w k,t+1 ), where θ k,t+1 is the hyperparameter of the k-th model at the t+1th iteration, θ k,t The hyperparameters of the kth model at iteration t, μ k is the super constant of the kth model, μ k＝1 =0.9,μ k＝2 =0.99, μ k＝3 =0.5, μ k＝4 =0.5.

[0044] The embodiment of the present invention further provides a control system for a precision barrel plating process, which is used to execute any one of the control methods for the precision barrel plating process described above, and the control system for the precision barrel plating process includes:

[0045] The data acquisition module is used to collect the current process parameters in real time during the roller plating process. The current process parameters include the current current, the current voltage, the current temperature, the current flow rate and the current plating solution composition;

[0046] Data integration module, used to build a process parameter relationship network based on current process parameters;

[0047] The quality prediction module is used to input the process parameter relationship network into the pre-trained deep learning quality prediction model in real time to obtain the predicted quality parameters of the coating;

[0048] Precision control module, used to adjust the current process parameters in real time according to the predicted quality parameters of the coating to achieve real-time precision control;

[0049] Among them, the deep learning quality prediction model includes graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network. The input of the graph convolutional neural network is the process parameter relationship network, the input of the self-attention mechanism is the output of the graph neural network, the input of the multi-task learning structure is the output of the graph neural network, and the input of the generative adversarial network includes the output of the self-attention mechanism and the output of the multi-task learning structure; the output of the generative adversarial network is used to feedback and adjust the graph neural network, and the graph neural network outputs the coating prediction quality parameters after adjustment.

[0050] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0051] The embodiment of the present invention provides a control method and system for a precision barrel plating process, by establishing a deep learning quality prediction model for quality analysis by interconnecting a graph neural network, a self-attention mechanism, a multi-task learning structure and a generative adversarial network, and constructing a process parameter relationship network for characterizing and representing the relationship between process parameters. The two are combined, and the interconnected graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network are analyzed based on the process parameter relationship network, which can mine the implicit relationship of mutual influence between process parameters and the relationship between process parameters and product quality, thereby improving the accuracy of quality prediction. On this basis, the current process parameters are adjusted in real time according to the predicted quality parameters of the coating to achieve real-time precision control, thereby improving the quality of production products, improving production efficiency and reducing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of a control method of a precision roll plating process provided by an embodiment of the present invention.

[0053] Figure 2 It is a schematic diagram of the structure of a deep learning quality prediction model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0055] Example

[0056] The embodiment of the present invention provides a control method for a precision barrel plating process, such as Figure 1 As shown, the control method of the precision roller plating process includes:

[0057] S101: Data collection: During the roll plating process, current process parameters are collected in real time.

[0058] The current process parameters include the current current, the current voltage, the current temperature, the current flow rate, the current plating solution composition, and the content of the plating solution composition. The current plating solution composition may include a nickel plating solution, such as a nickel salt: such as nickel sulfate (NiSO 4 ) or nickel chloride (NiCl 2 Chlorides: such as sodium chloride (NaCl) or ammonium chloride (NH 4 Cl) to increase current density and deposition uniformity. Acid or alkali: adjust pH, commonly sulfuric acid or ammonia. Additives: such as surfactants and brighteners to improve the appearance and performance of the coating. Current plating bath ingredients may also include zinc plating solution, such as zinc salts: such as zinc chloride (ZnCl 2 ) or zinc sulfate (ZnSO 4 ). Acid: dilute sulfuric acid or hydrofluoric acid is usually used to clean and pre-treat the substrate. Additives: such as organic matter and surfactants, which help improve the quality of the coating. Current plating bath ingredients can also include chrome plating solutions, such as chromates: such as sodium chromate (Na 2 Cr 2 O 7 ) or ammonium chromate. Acid: generally sulfuric acid or hydrochloric acid, used to adjust the pH value. Additives: used to improve the gloss and corrosion resistance of the coating. And the current plating solution composition may also include other plating solution components, such as electrolyte: ionic solution to ensure current conduction. Buffer: maintain the stability and pH value of the plating solution. Surfactant: helps improve the uniformity and adhesion of the coating.

[0059] S102: Data integration: construct a process parameter relationship network based on current process parameters.

[0060] S103: Quality prediction: Input the process parameter relationship network into the pre-trained deep learning quality prediction model in real time to obtain the predicted quality parameters of the coating.

[0061] The predicted coating quality parameters include information such as predicted coating thickness and predicted spectral image of the coating surface.

[0062] S104: Precision control: Adjust the current process parameters in real time according to the predicted quality parameters of the coating to achieve real-time precision control.

[0063] Among them, deep learning quality prediction models include Graph Neural Network (GNN), self-attention mechanism (Transformer), multi-task learning structure (MTL) and generative adversarial network (GAN). Please refer to Figure 2 The input of the graph convolutional neural network is the process parameter relationship network, the input of the self-attention mechanism is the output of the graph neural network, the input of the multi-task learning structure is the output of the graph neural network, and the input of the generative adversarial network includes the output of the self-attention mechanism and the output of the multi-task learning structure; the output of the generative adversarial network is used to feedback and adjust the graph neural network, and the graph neural network outputs the predicted quality parameters of the coating after adjustment.

[0064] By constructing a process parameter relationship network, the mutual relationship between process parameters is characterized and represented. However, since the process parameter relationship network is based on experience, its representation of the mutual relationship between parameters and the relationship of their impact on product quality is not accurate. Therefore, through the interconnected graph neural network (GNN), self-attention mechanism (Transformer), multi-task learning (MTL) and generative adversarial network (GAN) based on the process parameter relationship network, the implicit relationship of mutual influence between process parameters and the relationship between process parameters and product quality can be explored, thereby improving the accuracy of quality prediction.

[0065] By adopting the above scheme, the deep learning quality prediction model combining graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network provides an efficient and accurate method for coating quality prediction. Through multi-level feature extraction and complex relationship modeling, the deep learning quality prediction model can significantly improve the performance of coating quality prediction. The graph neural network (GNN), self-attention mechanism (Transformer), multi-task learning (MTL) and generative adversarial network (GAN) interact and interact with each other, which improves the accuracy and reliability of coating quality prediction. Specifically:

[0066] GNN provides rich structural features for other modules. As the input of Transformer, by combining the local features extracted by GNN, Transformer can effectively combine local information with global context, thereby enhancing the accuracy of coating quality prediction. When GNN and Transformer modules learn together through MTL, GNN can provide specific local features for each task, while Transformer can integrate these features to capture the correlation between tasks, thereby improving the overall prediction performance. The generated samples of GAN can be used as input for GNN and Transformer. The generated data not only enriches the training set, but also can be used to train multiple tasks through the design of MTL, further improving the adaptability and performance of the model. The generated data can be used to train different types of coating quality prediction tasks, so that each task can obtain better learning effects under different feature conditions. The self-attention mechanism enables the model to capture the coating characteristics. In coating quality prediction, the combination of graph neural network (GNN), self-attention mechanism (Transformer), multi-task learning (MTL) and generative adversarial network (GAN) can effectively handle complex relationships and improve prediction performance. Therefore, the deep learning quality prediction model combining GNN, Transformer, MTL and GAN can accurately predict the coating quality, and users can make more informed decisions in material selection, process parameter optimization and quality control, thereby improving production efficiency and reducing costs.

[0067] Optionally, a process parameter relationship network is constructed based on the current process parameters, including:

[0068] Obtain the correlation index between the current process parameters. Take the current process parameters as nodes, connect any two nodes, define the line between the two nodes as an edge, use the correlation index as the weight of the edge, and obtain the initial parameter relationship network. Optimize the initial relationship network to obtain the current process parameter relationship network.

[0069] The correlation index between the current process parameters is obtained, including:

[0070] Correlation measurement: Use the correlation coefficient to quantify the relationship between two process parameters; the absolute value of the correlation coefficient is used as the correlation index between the two process parameters.

[0071] In order to improve the accuracy of mixed plating quality prediction, the process parameter relationship network is optimized. Specifically, the initial relationship network is optimized to obtain the current process parameter relationship network, which also includes:

[0072] Entropy measurement: obtain the information entropy of the current process parameters; perform weighted summation on the correlation index and information entropy between the current process parameters to obtain the optimization weight; use the optimization weight as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

[0073] Optionally, the initial relationship network is optimized to obtain the current process parameter relationship network, including:

[0074] Influence analysis: If the current process parameters have a significant impact on the results during the experiment or production process, multiply the correlation index of the current process parameters by 1.1 to obtain the influence correlation index; use the influence correlation index as the weight of the edge of the process parameter to obtain the current process parameter relationship network. This part is set based on experience. For example, if the temperature is higher than 1000 degrees Celsius, it has a greater impact on the roll plating product, so the temperature correlation index is increased by 1.1 times.

[0075] In order to improve the accuracy of the current process parameter relationship network in characterizing the characteristics of the current process parameters, the initial relationship network is optimized to obtain the current process parameter relationship network, which also includes:

[0076] Entropy measurement: obtain the information entropy of the current process parameters; perform weighted summation of the influence correlation index and information entropy to obtain the optimization weight; use the optimization weight as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

[0077] In an embodiment of the present invention, a training method for a deep learning quality prediction model includes:

[0078] Obtaining a training sample, the training sample including a plurality of historical process parameters, the historical process parameters including historical current, historical voltage, historical temperature, historical flow rate and historical plating solution composition;

[0079] Building a training parameter relationship network based on historical process parameters; multiple historical process parameters include multiple training parameter relationship networks;

[0080] Input multiple training parameter relationship networks into a pre-trained deep learning quality prediction model, calculate the loss function of the deep learning quality prediction model, and when the loss function of the deep learning quality prediction model converges, determine that the training of the deep learning quality prediction model is completed, and obtain the trained deep learning quality prediction model.

[0081] The loss function of the deep learning quality prediction model is calculated using the following formula:

[0082] L total =αL GNN +βL Transformer +γL MTL +δL GAN

[0083] Among them, L total Represents the loss function of the deep learning quality prediction model, L GNN represents the loss function of the graph neural network, L Transformer represents the loss function of the self-attention mechanism, L MTL Represents the loss function of the multi-task learning structure, L GAN Represents the loss function of the generative adversarial network; α, β, γ, δ are hyperparameters; the initial values ​​of the hyperparameters are α=0.6, β=0.1, γ=0.2, δ=0.1.

[0084] Among them, the loss function of the graph neural network is equal to the weighted sum of the mean squared error (MSE) loss function and the loss of the discriminator of the generative adversarial network. The loss function of the self-attention mechanism adopts the cross-entropy loss function, the loss function of the multi-task learning structure adopts the binary cross-entropy loss function, and the loss function of the generative adversarial network adopts the loss of the generator.

[0085] In this embodiment of the present invention, each hyperparameter corresponds to a model. Specifically, α corresponds to a graph neural network, β corresponds to a self-attention mechanism, γ corresponds to a multi-task learning structure, and δ corresponds to a generative adversarial network; the graph neural network, the self-attention mechanism, the multi-task learning structure, and the generative adversarial network are the first model, the second model, the third model, and the fourth model, respectively.

[0086] For each hyperparameter, update it as follows:

[0087] Calculate the rate of change of loss: ΔL k,t =L k,t -L k,t-1 , where ΔL k is the rate of change of the loss function of the k-th model at the t-th iteration; L k,t is the loss function of the k-th model at the t-th iteration, L k,t-1 is the loss function of the k-th model at the t-1th iteration; when t = 1, L k,t-1 =0; the value of t is an integer between 1 and T, and T is a positive integer greater than 5;

[0088] Dynamically adjust hyperparameters:

[0089] Calculate the hyperparameter adjustment weights. The hyperparameter adjustment weights are calculated as follows: w k,t+1 The hyperparameter adjustment weights for the k-th model at the t+1th iteration, ∈ is a small constant, ∈=0.001.

[0090] Tuning Hyperparameters: θ k,t+1=θ k,t +μ k (θ k,t +w k,t+1 ), where θ k,t+1 is the hyperparameter of the kth model at the t+1th iteration, θ k,t The hyperparameters of the kth model at iteration t, μ k is the super constant of the kth model, μ k＝1 =0.9,μ k＝2 =0.99, μ k＝3 =0.5, μ k＝4 =0.5.

[0091] In summary, during the training process, a loss function of the deep learning quality prediction model is calculated for each iteration. That is, the loss function of the deep learning quality prediction model is calculated by the following formula:

[0092]

[0093] Among them, L t is the loss function of the deep learning quality prediction model at the tth iteration, when L t When convergence or t>threshold (for example, 1000 times), stop training the deep learning quality prediction model to obtain a trained deep learning quality prediction model. k,t represents the loss function of the k-th model at the t-th iteration.

[0094] As an optional implementation, the loss function of the graph neural network at the tth iteration is L 1,t ,

[0095]

[0096] Where N is the number of samples, that is, the number of historical process parameters, and the value of N is a positive integer. i,t represents the i-th sample in the t-th iteration, is the output of the graph neural network for the i-th sample in the t-th iteration. t is the loss of the discriminator of the generative adversarial network at the tth iteration. λ is the weighting parameter, λ = 0.3.

[0097] By adopting the above solution, the graph neural network is jointly constrained by MTL, Transformer and GAN. When extracting features, the graph neural network jointly considers the influence of the features of MTL, Transformer and GAN, thereby achieving the following beneficial effects:

[0098] Multi-level feature extraction: GNN combines MTL, Transformer and GAN, which can effectively capture local structural features and capture the microscopic interactions of materials; while Transformer enhances the understanding of global features, combines local and global features, and provides a more comprehensive feature representation.

[0099] Complex relationship modeling: Through the self-attention mechanism, LayerFusionNet can handle the complex nonlinear relationship between coating characteristics and process parameters. The combination of GNN and Transformer enables the model to focus on local details and global trends at the same time, thereby improving the accuracy of prediction.

[0100] Enhanced learning ability: Through multi-task learning, the model can share knowledge among different tasks and improve overall performance. In addition, the synthetic data generated by GAN helps the model enhance its learning ability in the case of data scarcity and improves the robustness of the model.

[0101] Real-time monitoring and decision support: The application of this model is not limited to prediction, but can also achieve real-time monitoring and feedback to help dynamic adjustments in the production process, thereby optimizing coating quality and reducing production costs.

[0102] Continuous optimization and adaptive capabilities: By introducing online learning mechanisms, the model can adjust itself based on real-time feedback to maintain high prediction accuracy. This adaptive capability is critical to responding to changes in production.

[0103] Improve production efficiency: By accurately predicting the quality of the coating, companies can make process adjustments more quickly and reduce the production of unqualified products, thereby improving production efficiency and yield rate.

[0104] Reduce costs: Reduce rework and losses caused by quality problems, reduce material waste and production costs, and help companies gain an advantage in competition.

[0105] In an embodiment of the present invention, the current process parameters are adjusted in real time according to the predicted quality parameters of the coating to achieve real-time precision control, including: obtaining the adjusted process parameters through a PID control algorithm. Specifically, the following steps are included:

[0106] 1. Determine the control target: Determine the target value (set value) that needs to be controlled, such as the ideal thickness, temperature, pressure, etc. of the coating.

[0107] 2. Collect real-time data and monitor the current status: Use sensors to monitor the current process parameters in real time and compare them with the set values.

[0108] 3. Calculate the deviation: Use the formula to calculate the deviation between the current value and the set value:

[0109] e(t)=SP-PV

[0110] Here, e(t) represents the deviation at the current moment, SP represents the actual measured value (current value), and PV represents the set value (target value).

[0111] 4.PID controller design

[0112] Define PID parameters: Set the proportional (P), integral (I), and derivative (D) gains, which determine the response characteristics of the controller.

[0113] Proportional (P): The controller's response to the current deviation.

[0114] Integral (I): The controller accumulates responses to past deviations, eliminating static errors.

[0115] Derivative (D): The controller responds to the rate of change of the deviation, predicting future trends.

[0116] 5. Calculate control output

[0117] PID control formula:

[0118]

[0119] Where u(t) is the control output (adjusted process parameters), K p is the proportional gain, K i is the integral gain, K d is the differential gain.

[0120] 6. Execute control actions

[0121] Adjust process parameters: According to the calculated control output u(t), adjust the corresponding process parameters (such as temperature, plating speed, time, etc.) to make the system close to the set value.

[0122] 7. Feedback and loops

[0123] Real-time feedback: The system continuously monitors the current process parameters and makes calculations.

[0124] Real-time feedback: The system continuously monitors the current process parameters, obtains real-time data through sensors, and continuously calculates the deviation e(t).

[0125] Cyclic execution: After each sampling, the above steps (calculation of deviation, PID control, adjustment of parameters) are repeated to form a closed-loop control. This ensures that the system can dynamically respond to any disturbance or change, thereby maintaining the process parameters within the set range.

[0126] 8. Adjust PID parameters

[0127] PID parameter tuning: To ensure the best performance of the control system, it may be necessary to tune the proportional (Kp), integral (Ki) and differential (Kd) gains. p , K i and K d Perform tuning.

[0128] Rules of thumb: Some rules of thumb, such as the Ziegler-Nichols rule, can be used to preliminarily set PID parameters.

[0129] Numerical optimization: Optimize PID parameters by minimizing errors (such as integrated squared error) using simulation tools or experimental data.

[0130] 9. Monitoring and Recording

[0131] Data logging: Real-time recording of process parameters and control signals for subsequent analysis and optimization.

[0132] Performance monitoring: Regularly check the performance indicators of the control system, such as steady-state error, response time, overshoot, etc., to evaluate the control effect.

[0133] 10. Application and Improvement

[0134] Application feedback: Based on the control results and performance analysis, make necessary adjustments and optimizations to improve process quality and production efficiency.

[0135] System improvement: Consider introducing more advanced control strategies (such as fuzzy control, predictive control, etc.) to cope with more complex process requirements and changes.

[0136] In summary, the core idea of ​​adjusting process parameters through PID control algorithm is real-time monitoring and dynamic adjustment. According to the deviation between the set value and the actual value, the PID controller calculates the appropriate control output through the three-part effects of proportion, integration and differentiation, thereby achieving precise control of process parameters. This method can effectively improve the stability of the production process and product quality.

[0137] Based on the above-mentioned control method of the precision barrel plating process, an embodiment of the present invention provides a control system for the precision barrel plating process, which is used to execute the above-mentioned control method of the precision barrel plating process. The control system for the precision barrel plating process includes:

[0138] The data acquisition module is used to collect the current process parameters in real time during the roller plating process. The current process parameters include the current current, the current voltage, the current temperature, the current flow rate and the current plating solution composition;

[0139] Data integration module, used to build a process parameter relationship network based on current process parameters;

[0140] The quality prediction module is used to input the process parameter relationship network into the pre-trained deep learning quality prediction model in real time to obtain the coating prediction quality parameters; the coating prediction quality parameters include the predicted coating thickness and the predicted spectral image of the coating surface;

[0141] Precision control module, used to adjust the current process parameters in real time according to the predicted quality parameters of the coating to achieve real-time precision control;

[0142] Among them, the deep learning quality prediction model includes graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network. The input of the graph convolutional neural network is the process parameter relationship network, the input of the self-attention mechanism is the output of the graph neural network, the input of the multi-task learning structure is the output of the graph neural network, and the input of the generative adversarial network includes the output of the self-attention mechanism and the output of the multi-task learning structure; the output of the generative adversarial network is used to feedback and adjust the graph neural network, and the graph neural network outputs the coating prediction quality parameters after adjustment.

[0143] Regarding the model in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0144] In summary, the control method and system of the precision roll plating process provided by the embodiments of the present invention, in the field of coating quality prediction, combines the models of graph neural network (GNN), self-attention mechanism (Transformer), multi-task learning (MTL) and generative adversarial network (GAN) to achieve more efficient feature extraction and more accurate quality prediction. The following is a detailed analysis of the linkage between these technologies and their beneficial effects in coating prediction.

[0145] Processing of graph structure data: Coating materials usually have complex structural characteristics, and GNN is good at information propagation and feature learning on graph structures. Through the representation of nodes (atoms, molecules) and their edges (interactions), GNN can capture local features and neighborhood information.

[0146] Linkage 1: GNN provides rich structural features for other modules as input to Transformer. This feature representation can enhance Transformer's understanding of coating characteristics and behaviors.

[0147] Global feature capture: Transformer’s self-attention mechanism can process global context information, enabling it to capture long-range dependencies between coating features. This is critical for understanding complex material properties.

[0148] Linkage 2: By combining the local features extracted by GNN, Transformer can effectively combine local information with the global context, thereby enhancing the accuracy of coating quality prediction.

[0149] Knowledge sharing between tasks: MTL allows the model to learn multiple related tasks simultaneously (e.g., quality prediction under different coating types or processing conditions), thereby facilitating information sharing between different tasks. This can improve the learning effect of each task.

[0150] Linkage 3: When GNN and Transformer modules learn together through MTL, GNN can provide specific local features for each task, while Transformer can integrate these features and capture the correlation between tasks, thereby improving the overall prediction performance.

[0151] Data generation and enhancement: GAN generates samples similar to real data through adversarial training of generators and discriminators. In coating quality prediction, real samples may be scarce due to experimental limitations. GAN can generate diverse synthetic data to help the model learn coating characteristics more comprehensively.

[0152] Improved robustness: By introducing synthetic data, the model can learn more feature variations and reduce its dependence on training data, thereby improving its predictive ability for new samples, especially when facing noisy or inconsistent data.

[0153] Linkage 4: The samples generated by GAN can be used as inputs of GNN and Transformer. The generated data not only enriches the training set, but also can be used to train multiple tasks through the design of MTL, further improving the adaptability and performance of the model. For example, the generated data can be used to train different types of coating quality prediction tasks, so that each task can achieve better learning effects under different feature conditions.

[0154] Comprehensive linkage effect: Through the linkage of the above modules, the beneficial effects of the model in coating quality prediction are mainly reflected in the following aspects:

[0155] Multi-level feature extraction: GNN extracts local structural features, Transformer integrates global context information, and MTL enhances knowledge sharing between tasks, making feature extraction more comprehensive and in-depth.

[0156] Improve prediction accuracy: By combining local features with global features, the model can more accurately grasp the physical and chemical properties of the coating, thereby improving the accuracy of the prediction.

[0157] Enhance the generalization ability of the model: The design of MTL enables the model to share knowledge on multiple related tasks, thereby avoiding the degradation of other tasks due to overfitting to a specific task. The diverse data generated by GAN further improves the robustness and generalization ability of the model, especially when facing new materials or new processes.

[0158] Handling complex relationships: Modeling nonlinear relationships: The quality of the coating is affected by many factors, such as material composition, process parameters, and environmental conditions. GNN is able to handle nonlinear relationships in the microstructure of the material and capture local interactions through the representation of graph nodes and edges. Transformer, on the other hand, models long-distance dependencies through the self-attention mechanism to fully understand the complex interactions between different factors. Deep mining of feature interactions: Through the combination of GNN and Transformer, the model can deeply explore the interactions between features. For example, the performance of a certain material component under specific process conditions may be related to the proportion, temperature, or pressure of other components. GNN can effectively capture these local interactions, while Transformer integrates global information to ensure that the model fully understands these complex relationships.

[0159] Data enhancement and model training:

[0160] GAN’s generative power: GAN is able to generate synthetic data that is similar to real coating samples, which is particularly important in situations where data is scarce or difficult to obtain. By introducing this synthetic data into GNN and Transformer training, the model is able to enhance its learning ability, especially in the face of uncertainty.

[0161] Synergistic effect of multi-task learning: MTL allows the model to train multiple related tasks simultaneously (e.g., quality prediction under different coating types and different process conditions), enabling the model to utilize shared knowledge between different tasks. By performing multi-task training on GAN-generated data, the model can improve its adaptability to each task and further enhance the overall prediction performance.

[0162] In summary, combining the above technologies, a real-time coating quality monitoring system can be built, which can continuously predict the quality during the production process and adjust the process parameters according to the prediction results. Through the structural features extracted by GNN and the global features of Transformer, the system can respond to environmental changes in real time, thereby achieving adaptive control. GNN can effectively capture local structural features and capture the microscopic interactions of materials; while Transformer enhances the understanding of global features, combines local and global features, and provides a more comprehensive feature representation. Through the self-attention mechanism, the model can handle the complex nonlinear relationship between coating characteristics and process parameters. The combination of GNN and Transformer enables the model to pay attention to local details and global trends at the same time, thereby improving the accuracy of prediction. Through multi-task learning, the model can share knowledge in different tasks and improve overall performance. In addition, the synthetic data generated by GAN helps the model enhance its learning ability when data is scarce, improving the robustness of the model. By introducing an online learning mechanism, the model can self-adjust according to real-time feedback to maintain high prediction accuracy. This adaptability is crucial to cope with changes in production. By accurately predicting the quality of the coating, companies can make process adjustments more quickly and reduce the production of unqualified products, thereby improving production efficiency and yield rate. Reducing rework and losses caused by quality problems, reducing material waste and production costs can help companies gain an advantage in the competition.

[0163] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0164] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0165] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0166] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0167] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0168] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to an embodiment of the present invention. The present invention may also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may have the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0169] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

Claims

1. A control method for a precision barrel plating process, characterized in that: include: Data collection: During the roller plating process, the current process parameters are collected in real time, including the current current, current voltage, current temperature, current flow rate and current plating solution composition; Data integration: build a process parameter relationship network based on current process parameters; Quality prediction: Input the process parameter relationship network into the pre-trained deep learning quality prediction model in real time to obtain the coating prediction quality parameters; Among them, the deep learning quality prediction model includes graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network. The input of the graph convolutional neural network is the process parameter relationship network, the input of the self-attention mechanism is the output of the graph neural network, the input of the multi-task learning structure is the output of the graph neural network, and the input of the generative adversarial network includes the output of the self-attention mechanism and the output of the multi-task learning structure; the output of the generative adversarial network is used to feedback and adjust the graph neural network, and the graph neural network outputs the coating prediction quality parameters after adjustment.

2. The control method of the precision barrel plating process according to claim 1, characterized in that: The method further comprises: Precision control: Adjust the current process parameters in real time according to the predicted quality parameters of the coating to achieve real-time precision control.

3. The control method of the precision barrel plating process according to claim 1, characterized in that: Build a process parameter relationship network based on current process parameters, including: Obtaining the correlation index between current process parameters; The current process parameters are regarded as nodes, any two nodes are connected, the line between the two nodes is defined as an edge, the association relationship index is used as the weight of the edge, and an initial parameter relationship network is obtained; The initial relationship network is optimized to obtain the current process parameter relationship network.

4. The control method of the precision barrel plating process according to claim 3, characterized in that: Optimize the initial relationship network to obtain the current process parameter relationship network, including: Influence analysis: If the current process parameters have a significant impact on the results during the experiment or production process, multiply the correlation index of the current process parameters by 1.1 to obtain the influence correlation index; The influence correlation index is used as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

5. The control method of the precision barrel plating process according to claim 4, characterized in that: Optimize the initial relationship network to obtain the current process parameter relationship network, which also includes: Entropy measurement: obtain the information entropy of current process parameters; The influence correlation index and information entropy are weighted and summed to obtain the optimized weight; The optimization weight is used as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

6. The control method of the precision barrel plating process according to claim 3, characterized in that: Optimize the initial relationship network to obtain the current process parameter relationship network, which also includes: Entropy measurement: obtain the information entropy of current process parameters; The correlation index and information entropy between the current process parameters are weighted and summed to obtain the optimization weight; The optimization weight is used as the weight of the edge of the process parameter to obtain the current process parameter relationship network.

7. The control method of the precision barrel plating process according to claim 1, characterized in that: The training method of the deep learning quality prediction model includes: Obtaining a training sample, the training sample including a plurality of historical process parameters, the historical process parameters including historical current, historical voltage, historical temperature, historical flow rate and historical plating solution composition; Building a training parameter relationship network based on historical process parameters; multiple historical process parameters include multiple training parameter relationship networks; Input multiple training parameter relationship networks into a pre-trained deep learning quality prediction model, calculate the loss function of the deep learning quality prediction model, and when the loss function of the deep learning quality prediction model converges, determine that the training of the deep learning quality prediction model is completed, and obtain the trained deep learning quality prediction model.

8. The control method of the precision barrel plating process according to claim 7, characterized in that: The loss function of the deep learning quality prediction model is calculated using the following formula: L total =αL GNN +βL Transformer +γL MTL +δL GAN Among them, L total Represents the loss function of the deep learning quality prediction model, L GNN represents the loss function of the graph neural network, L Transformer represents the loss function of the self-attention mechanism, L MTL Represents the loss function of the multi-task learning structure, L GAN Represents the loss function of the generative adversarial network; α, β, γ, δ are hyperparameters; the initial values ​​of the hyperparameters are α=0.6, β=0.1, γ=0.2, δ=0.

1.

9. The control method of the precision barrel plating process according to claim 8, characterized in that: Each hyperparameter is associated with a model. Specifically, α corresponds to the graph neural network, β corresponds to the self-attention mechanism, γ corresponds to the multi-task learning structure, and δ corresponds to the generative adversarial network. The graph neural network, self-attention mechanism, multi-task learning structure, and generative adversarial network are the first model, the second model, the third model, and the fourth model, respectively. For each hyperparameter, update it in the following way: Calculate the rate of change of loss: ΔL k,t =L k，t -L k,t-1 , where ΔL k is the rate of change of the loss function of the k-th model at the t-th iteration; L k，t is the loss function of the k-th model at the t-th iteration, L k，t-1 is the loss function of the k-th model at the t-1th iteration; when t = 1, L k，t-1 =0; the value of t is an integer between 1 and T, and T is a positive integer greater than 5; Dynamically adjust hyperparameters: Calculate the hyperparameter adjustment weights. The hyperparameter adjustment weights are calculated as follows: w k,t+1 The hyperparameter adjustment weights for the kth model at the t+1th iteration, ∈ is a small constant, ∈=0.001; Tuning Hyperparameters: θ k,t+1 =θ k,t +μ k (θ k,t +w k,t+1 ), where θ k,t+1 is the hyperparameter of the k-th model at the t+1th iteration, θ k,t The hyperparameters of the kth model at iteration t, μ k is the super constant of the kth model, μ k＝1 =0.9,μ k＝2 =0.99, μ k＝3 =0.5, μ k＝4 =0.

5.

10. A control system for a precision barrel plating process, characterized in that: A control method for executing the precision barrel plating process according to any one of claims 1 to 9, wherein the control system of the precision barrel plating process comprises: The data acquisition module is used to collect the current process parameters in real time during the roller plating process. The current process parameters include the current current, the current voltage, the current temperature, the current flow rate and the current plating solution composition; Data integration module, used to build a process parameter relationship network based on current process parameters; The quality prediction module is used to input the process parameter relationship network into the pre-trained deep learning quality prediction model in real time to obtain the predicted quality parameters of the coating; Precision control module, used to adjust the current process parameters in real time according to the predicted quality parameters of the coating to achieve real-time precision control; Among them, the deep learning quality prediction model includes graph neural network, self-attention mechanism, multi-task learning structure and generative adversarial network. The input of the graph convolutional neural network is the process parameter relationship network, the input of the self-attention mechanism is the output of the graph neural network, the input of the multi-task learning structure is the output of the graph neural network, and the input of the generative adversarial network includes the output of the self-attention mechanism and the output of the multi-task learning structure; the output of the generative adversarial network is used to feedback and adjust the graph neural network, and the graph neural network outputs the coating prediction quality parameters after adjustment.

Images