A method for controlling production of an anti-surge resistor

Abstract

The application belongs to the technical field of resistance and provides a surge resistance production control method. The method comprises the following steps: detecting concentration data of a plating solution and production environment fluctuation data of the surge resistance in real time; obtaining a quality influence index according to the concentration data and the production environment fluctuation data; when the quality influence index is higher than a preset value, analyzing the concentration data and the production environment fluctuation data by using a plating parameter optimization model to obtain plating parameters; and controlling a plating equipment to perform electrode plating on the surge resistance in the batch according to the plating parameters. The surge resistance production control method can identify production risks in advance by detecting data in real time and evaluating a quality influence index, and can optimize plating parameters in time when the index exceeds the preset value, thereby effectively overcoming the interference of environmental and equipment factors, ensuring stable plating quality, and significantly improving the electrical performance consistency and production efficiency of the surge resistance.

Description

Technical Field

[0001] This invention relates to the field of resistor technology, and more specifically, to a method for controlling the production of surge-resistant resistors. Background Technology

[0002] In the manufacturing of surge resistors, the electrode coating process is a crucial step in ensuring the electrical performance of the product. The thickness and uniformity of the electrode coating directly affect the contact resistance. If the coating is too thin, it will lead to poor contact between the electrode and the resistive element, causing unstable signal transmission, and even local overheating or open circuits under surge current impact. On the other hand, if the coating is too thick, it will increase parasitic capacitance, which will seriously affect the response speed of the surge resistor in high-frequency circuits, making it unable to meet the application requirements of modern high-speed electronic systems.

[0003] Currently, traditional electrode coating processes rely heavily on operator experience and manual adjustments, lacking precise quantitative control. In actual production, variations in coating solution concentration and environmental temperature and humidity fluctuations make it difficult to consistently maintain coating thickness and uniformity. While some companies have introduced online monitoring equipment, the results often lag behind the production process, hindering real-time feedback and dynamic adjustments to the coating process.

[0004] Therefore, there is an urgent need for a model-based method to improve coating quality, enabling precise control of the electrode coating process to enhance the overall performance and production efficiency of surge resistors. Summary of the Invention

[0005] In response, the present invention provides a surge resistor production control method, electronic device, computer storage medium, and computer program product to solve at least one of the above-mentioned technical problems.

[0006] This invention provides a method for controlling the production of surge-resistant resistors, comprising the following steps: A coating parameter optimization model was built based on Reformer, and the model was pre-trained using training data based on fixed process conditions. Before starting this batch of coating, the concentration data of the coating solution and the production environment fluctuation data of the surge resistor are monitored in real time. The production environment fluctuation data includes temperature fluctuation data and humidity fluctuation data. The quality impact index is evaluated and derived based on the concentration data and the production environment fluctuation data. When the quality impact index is higher than the preset value, the concentration data and the production environment fluctuation data are analyzed using the coating parameter optimization model to obtain the coating parameters, which include sputtering pressure, sputtering power, target-substrate distance, substrate temperature and sputtering gas flow rate. Under the same fixed process conditions, the coating equipment is controlled to perform electrode coating on the surge protection resistors of this batch according to the coating parameters.

[0007] The present invention also provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any of the preceding claims.

[0008] The present invention also provides a computer storage medium storing a computer program that can be executed by a processor to implement the methods described in any of the preceding claims.

[0009] The present invention also provides a computer program product comprising a computer program that can be executed by a processor to implement the method as described in any of the preceding claims.

[0010] The surge resistor production control method of the present invention identifies production risks in advance by real-time detection of data and evaluation of the quality impact index. When the index exceeds the preset value, the coating parameters can be optimized in a timely manner, thereby effectively overcoming the interference of environmental and equipment factors, ensuring stable coating quality, and significantly improving the consistency of electrical performance and production efficiency of surge resistors. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic flowchart of a surge resistor production control method disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the coating parameter optimization model disclosed in the embodiments of the present invention. Detailed Implementation

[0013] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0014] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0015] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0016] In the production of surge-resistant resistors, the quality of the electrode coating directly determines the product's electrical performance. However, dynamic changes in the coating solution concentration and fluctuations in temperature and humidity in the production environment are key factors affecting coating quality. Traditional methods relying on manual experience for control are insufficient to accurately capture the impact of these variables on coating parameters, leading to significant batch-to-batch variations in product quality and failing to meet the stringent requirements of modern electronics industry for product stability and consistency. Therefore, this invention achieves precise control of the coating process through the following specific implementation methods.

[0017] like Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for controlling the production of surge protectors, comprising the following steps: S1. Based on Reformer, a coating parameter optimization model is constructed, and the coating parameter optimization model is pre-trained using training data based on fixed process conditions.

[0018] In actual production environments, there are complex nonlinear relationships between coating parameters (such as sputtering pressure, sputtering power, target-substrate distance, substrate temperature, and sputtering gas flow rate) and coating solution concentration, as well as the temperature and humidity of the production environment. In order to quantify this relationship and achieve accurate prediction of coating parameters, this invention first constructs a coating parameter optimization model, which is built based on Reformer.

[0019] In practice, under fixed process conditions (such as fixed equipment models and operating procedures), a large amount of sample data is collected, including different coating solution concentrations, ambient temperature and humidity data, and corresponding actual coating parameters. After standardization, this sample data is used as a training set and input into the coating parameter optimization model. Through repeated training (e.g., training with more than 500 sets of data), the coating parameter optimization model continuously adjusts its internal parameters, learns the mapping relationship between the input data and the coating parameters, and thus has the ability to make preliminary predictions about the coating parameters.

[0020] S2, Before starting this batch of coating, the concentration data of the coating solution and the production environment fluctuation data of the surge resistor are monitored in real time. The production environment fluctuation data includes temperature fluctuation data and humidity fluctuation data. The quality impact index is evaluated based on the concentration data and the production environment fluctuation data.

[0021] Because the concentration of the coating solution, as well as the temperature and humidity in the production environment, are constantly changing during each batch of coating, these changes directly or indirectly affect the quality of the coating. If these dynamic changes cannot be monitored in a timely manner and their potential impact on coating quality cannot be assessed, it will be difficult to make accurate adjustments during production, potentially leading to unstable coating quality. Therefore, a real-time monitoring mechanism is activated before starting the current batch of coating.

[0022] Specifically, high-precision sensors are used to monitor the concentration of the coating solution in real time. These sensors can accurately measure the content of various components in the coating solution. At the same time, professional environmental monitoring equipment is deployed to continuously monitor the temperature and humidity in the production environment, thereby capturing subtle fluctuations in environmental temperature and humidity in real time, thus obtaining temperature fluctuation data and humidity fluctuation data.

[0023] The acquired real-time data is input into a pre-set evaluation algorithm. By comprehensively considering the influence weights of coating solution concentration, temperature fluctuation data, and humidity fluctuation data on coating quality, these multi-dimensional data are integrated into a single numerical index, namely the quality influence index.

[0024] For example, when the concentration of the coating solution is detected to exceed the standard range by 10%, and the ambient temperature is 0.5°C higher than normal and the ambient humidity is 5% higher than normal, the quality impact index calculated by the algorithm reaches 70 (the preset threshold is 50). This index indicates that the current production conditions pose a significant potential risk to the coating quality, and parameters need to be adjusted in a timely manner to ensure that the coating quality meets the requirements.

[0025] S3. When the quality impact index is higher than the preset value, the coating parameter optimization model is used to analyze the concentration data and the production environment fluctuation data to obtain the coating parameters, which include sputtering pressure, sputtering power, target-substrate distance, substrate temperature and sputtering gas flow rate.

[0026] When the quality impact index exceeds the preset value, it indicates that under the current production conditions, continuing to use conventional coating parameters will make it difficult to guarantee coating quality. At this point, the coating parameter optimization model is activated, using real-time detected coating solution concentration data and production environment fluctuation data as input.

[0027] The pre-trained coating parameter optimization model performs in-depth analysis and processing of the input data, outputting coating parameters adapted to current production conditions. For example, when the model receives data showing low coating solution concentration and high ambient temperature and humidity, it analyzes and outputs adjusted coating parameters: reducing sputtering pressure by 10%, increasing sputtering power by 15%, shortening the target-substrate distance by 5%, keeping the substrate temperature unchanged, and increasing the sputtering gas flow rate by 8%. These optimized parameters effectively compensate for the impact of unfavorable production conditions on coating quality, providing parameter assurance for high-quality coating.

[0028] S4. Under the same fixed process conditions, control the coating equipment to perform electrode coating on the surge protection resistors of this batch according to the coating parameters.

[0029] In actual production, to ensure that the coating parameters predicted by the coating parameter optimization model can play an effective role, it is necessary to precisely control the coating equipment under the same fixed process conditions as when the pre-training data was collected.

[0030] In practice, the optimized coating parameters are input into the control system of the coating equipment, which then performs electrode coating on the surge protectors of that batch according to these parameters. This method achieves precise control of the coating process, ensuring high consistency and stability in the coating quality of each batch of products.

[0031] The surge resistor production control method of the present invention identifies production risks in advance by real-time detection of data and evaluation of the quality impact index. When the index exceeds the preset value, the coating parameters can be optimized in a timely manner, thereby effectively overcoming the interference of environmental and equipment factors, ensuring stable coating quality, and significantly improving the consistency of electrical performance and production efficiency of surge resistors.

[0032] As an example, the construction of the coating parameter optimization model and the pre-training of the coating parameter optimization model using training data based on fixed process conditions include: constructing coating parameter optimization models corresponding to the bottom layer, functional layer, and outer layer respectively, and pre-training each of the coating parameter optimization models using the corresponding training data based on fixed process conditions.

[0033] As an example, before using the coating parameter optimization model to analyze the concentration data and the production environment fluctuation data to derive the coating parameters, the process further includes: selecting the corresponding coating parameter optimization model based on the layer attributes targeted by the current coating; the layer attributes include the bottom layer, functional layer, or outer layer.

[0034] In the process of surge-resistant coating, the bottom layer, functional layer, and outer layer have different functional and performance requirements, and their coating processes are affected by different factors and are more sensitive to coating parameters. The bottom layer mainly enhances the adhesion between the coating and the substrate, and is more sensitive to factors such as the interaction and bonding force between the coating solution and the substrate. The functional layer needs to meet specific electrical performance requirements, such as conductivity and antistatic properties, and its coating process is more complex due to factors such as material characteristics and environmental conditions. The outer layer mainly protects the internal coating and has high requirements for wear resistance and corrosion resistance; its coating process is also affected by environmental factors differently than the other two layers. Therefore, to more accurately control the coating quality of each layer, this invention constructs optimization models for coating parameters for different layers.

[0035] Under fixed process conditions, data related to the coating of the bottom layer, functional layer, and outer layer were collected separately. This data included different coating solution concentrations, ambient temperature and humidity data, and corresponding actual coating parameters (such as sputtering pressure, sputtering power, target-substrate distance, substrate temperature, and sputtering gas flow rate). Since the coating processes of different layers differ, the collected data will also vary to ensure that the model can learn the specific patterns of each layer's coating process. Furthermore, the collected data needs to be standardized to eliminate dimensional differences between different data features and improve the model's training effectiveness.

[0036] In actual coating processes, different coating stages correspond to different layer properties (bottom layer, functional layer, or outer layer). Selecting an appropriate model based on the layer property being coated ensures that the model accurately outputs parameters suitable for that layer. Using an incompatible model without proper selection may result in inaccurate coating parameters, thus affecting the coating quality of that layer. Layer properties can be determined through production plans, process documents, or other identifying information.

[0037] As an example, the assessment of the quality impact index based on the concentration data and the production environment fluctuation data includes: The entropy values ​​of the concentration data, temperature fluctuation data, and humidity fluctuation data are calculated. Based on the entropy values, weights are calculated using the entropy weight method. A preliminary quality impact index is then evaluated using a multiple linear regression model. The multiple linear regression model is constructed based on each entropy value and its corresponding weight. The equipment aging assessment model and several preset assessment indicators are used to evaluate and obtain the equipment aging degree index, and an adjustment coefficient is obtained based on the equipment aging degree index. The initial quality impact index is corrected using the adjustment coefficient to obtain the quality impact index.

[0038] As an example, several preset evaluation metrics include cumulative equipment uptime, number of times key components have been replaced, and number of recent equipment failures.

[0039] To accurately assess the impact of coating quality on coating solution concentration and production environment fluctuations (temperature and humidity fluctuations), this invention employs an algorithm combining entropy weighting and multiple linear regression. This algorithm comprehensively considers the weights of each factor and their interrelationships to derive a scientifically sound quality impact index.

[0040] Let the concentration of the coating solution be C, the temperature fluctuation data be T, and the humidity fluctuation data be H. Max-min normalization is used to normalize these raw data, eliminating the influence of dimensions and making the different data comparable.

[0041] According to the information entropy theory, for the j-th index (C, T, H), its entropy value is... .

[0042] in, , The number of data samples (i.e., concentration data and production environment fluctuation data) was increased. (secondary sampling) , Entropy is the standardized value of the j-th indicator in the i-th sample. This reflects the degree of dispersion of the indicator data. The greater the dispersion, the smaller the entropy value, indicating that the indicator provides more information and has a greater weight.

[0043] The weight of the j-th indicator ,in, For the number of indicators ( The three indicators are: coating solution concentration, temperature, and humidity. The weights are calculated using the entropy weighting method. , , It can objectively reflect the relative importance of each factor influencing the coating quality.

[0044] Constructing a multiple linear regression model .in, For the calculation of the preliminary quality impact index, , , These are the normalized values ​​for the three indicators: coating solution concentration, temperature, and humidity. This represents the random error term. The above multiple linear regression model comprehensively considers the influence of coating solution concentration, temperature fluctuation data, and humidity fluctuation data, and the weights of each factor are determined by the entropy weight method, which can accurately reflect the linear relationship between each factor and the quality influence index. A large amount of historical coating data was collected, including different coating solution concentrations, temperature fluctuation data, humidity fluctuation data, and corresponding actual coating quality assessment results (which can be obtained through subsequent product performance testing, etc.). This data was divided into training and validation sets. The multiple linear regression model was trained using the training set data, and the model parameters were adjusted to better fit the data. Then, the trained multiple linear regression model was validated using the validation set data to evaluate its accuracy and generalization ability. For example, the mean squared error (MSE), coefficient of determination (...), and other parameters were used to evaluate the model. The model performance is measured by metrics such as MSE. If the MSE is small, A value close to 1 indicates that the model fits well and can be used to calculate the actual quality impact index.

[0045] Meanwhile, in actual production, the aging of coating equipment can significantly impact coating quality. As equipment usage time increases, its performance gradually declines, such as reduced sputtering efficiency and decreased temperature control accuracy. This can lead to variations in coating quality even under the same coating solution concentration and production environment conditions. Therefore, after initially calculating the quality impact index, adjustments need to be made to account for the aging factors of the coating equipment.

[0046] Equipment operating parameters closely related to coating quality are selected as indicators for assessing the degree of equipment aging, such as cumulative equipment operating time. Number of replacements of key components (such as sputtering targets) Number of recent equipment failures wait.

[0047] The Analytic Hierarchy Process (AHP) was used to determine the weights of each evaluation indicator. First, a judgment matrix was constructed, and the relative importance of each indicator was determined through methods such as expert scoring. For example, for indicators i and j, if i is considered slightly more important than j, then the elements of the judgment matrix... Conversely Then, the eigenvectors and the largest eigenvalue of the judgment matrix are calculated, and the eigenvectors are normalized to obtain the weights of each indicator. , , .

[0048] Equipment aging .

[0049] Based on the degree of equipment aging The preliminary calculated quality impact index Adjustments are made. Let the adjustment factor be... , With the degree of equipment aging A positive correlation was observed. For example, by fitting experimental data... ,in and Here are the fitting coefficients. The final adjusted quality influence index is... .

[0050] As an example, such as Figure 2 As shown, when constructing the coating parameter optimization model based on Reformer, the multi-head attention mechanism is replaced with a lightweight attention module. The lightweight attention module adopts a sparse attention calculation method, that is, it only performs detailed attention calculations on the feature dimensions related to highly sensitive coating parameters, while only performing linear transformations or skipping some calculation steps on the feature dimensions related to low-sensitive coating parameters.

[0051] As an example, the highly sensitive coating parameters include contact resistance and parasitic capacitance.

[0052] In the production process of surge protectors, the coating parameter optimization model needs to quickly and accurately predict suitable coating parameters (such as sputtering pressure and sputtering power) based on data such as coating solution concentration and ambient temperature and humidity to minimize the impact on production speed. While the multi-head attention mechanism used in traditional Reformer models can effectively capture the complex relationships between data features, it performs comprehensive calculations on all input features during the computation process, resulting in a large computational load and low operating efficiency, making it difficult to meet the needs of rapid prediction in actual production. Furthermore, not all coating parameters have the same impact on the final surge protector performance; only some parameters (such as contact resistance and parasitic capacitance) have a more significant impact on product performance and are considered "highly sensitive coating parameters," while other parameters have a relatively smaller impact.

[0053] To address the aforementioned technical issues, this invention replaces the original multi-head attention mechanism of the Reformer model with a lightweight attention module.

[0054] In the lightweight attention module, highly sensitive coating parameters are first identified based on the sensitivity analysis results between pre-determined surge resistance performance indicators (such as contact resistance and parasitic capacitance) and coating parameters (obtained through experimental data statistics, correlation analysis, etc.). For feature dimensions related to these highly sensitive coating parameters, the lightweight attention module performs detailed attention calculations. For example, experiments show that coating thickness has a significant impact on contact resistance, classifying it as a highly sensitive parameter. Therefore, when the input data contains features related to coating thickness, the lightweight attention module meticulously calculates the association weights between this feature and other features using traditional attention calculation methods, ensuring the model can fully capture the impact of this crucial information on coating parameter prediction.

[0055] For feature dimensions related to low-sensitivity coating parameters, the lightweight attention module no longer performs a complete attention calculation process. Instead, it performs only simple linear transformations, such as weighted summation of features using a weight matrix to quickly complete feature processing; or it skips some complex calculation steps to reduce unnecessary computation. For example, if a gas flow rate parameter has a negligible impact on product performance and is considered a low-sensitivity parameter, the module performs only a simple weighting operation on it during the calculation process and then passes the processed result to the next layer, avoiding excessive computational resources spent on that feature.

[0056] As an example, the coating parameter optimization model is pre-trained using the following loss function: ; The coating parameter vector predicted by the coating parameter optimization model is: The actual coating parameter vector is , The number of coating parameters; , is a weighting coefficient used to adjust the proportion of performance index error in the total loss; This is the actual contact resistance. For actual parasitic capacitance, , These are mapping models used to predict contact resistance and parasitic capacitance, respectively. A set of highly sensitive parameters. The weighting coefficients for highly sensitive parameters, ; The weighting coefficients are non-highly sensitive parameters. .

[0057] The main body of the loss function designed in this invention adopts the form of mean squared error, that is, it calculates the sum of squares of the differences between the predicted value and the actual value; at the same time, the loss function also considers highly sensitive parameters and performance indicators. Specifically:

[0058] Because highly sensitive parameters have a more significant impact on product performance, higher weights are given to highly sensitive coating parameters that are highly correlated with key performance indicators of surge resistance (such as contact resistance and parasitic capacitance) in the loss function. By increasing the weights, the model will pay more attention to the prediction errors of these parameters during training, thus improving its ability to predict highly sensitive parameters.

[0059] The above loss function further considers the performance indicators of surge protection resistors (contact resistance). Parasitic capacitance The relationship between performance indicators and coating parameters. This is achieved by establishing a mapping model between performance indicators and coating parameters. , Furthermore, the prediction error of performance indicators is also included in the loss function. This setting ensures that the model not only focuses on the accuracy of the coating parameters themselves, but also considers the impact of these parameters on the final performance of the product.

[0060] Among them, by adjusting the weighting coefficients , The impact of coating parameter prediction errors and performance index prediction errors on model training can be balanced according to actual needs. For example, if more attention is paid to product performance indicators, then the error can be appropriately increased. , The value of makes the model focus more on reducing the error of performance indicators; if you want the model to be more stable in predicting coating parameters while ensuring performance indicators, you can appropriately reduce . , The value of . Additionally, the weighting coefficient. , The optimal value can also be determined through methods such as cross-validation, which will not be elaborated further.

[0061] The loss function designed in this invention prioritizes improving the prediction accuracy of key parameters by weighting highly sensitive coating parameters, ensuring more accurate prediction of core parameters affecting surge resistance performance. At the same time, it also incorporates the prediction error of performance indicators into the loss function, making the model training directly related to the actual performance of the product and outputting coating parameters that are more in line with production needs, thereby effectively improving product performance and quality and ensuring production results.

[0062] As an example, the pre-training of the coating parameter optimization model using training data based on fixed process conditions includes adjusting the learning rate of the coating parameter optimization model as follows: In the early stages of training, an initial learning rate is used, which is the maximum value of the learning rate. In the later stages of training, the learning rate is dynamically adjusted according to the changing trends of performance indicators, which refer to contact resistance and parasitic capacitance.

[0063] When pre-training a coating parameter optimization model using training data based on fixed process conditions, setting the learning rate appropriately is crucial to the model's training effectiveness and speed. If the learning rate remains constant, the model parameters will update slowly in the early stages of training, leading to slow convergence and wasting a significant amount of training time. In the later stages of training (which can be determined by the ratio of the training data used to the total training data), when the model is close to the optimal solution, a large learning rate may cause excessively large parameter updates, resulting in an "oscillation" phenomenon. This can prevent the model from converging to the optimal state or even lead to overfitting, reducing the model's prediction accuracy in actual production.

[0064] Meanwhile, the contact resistance and parasitic capacitance of the surge protector are key performance indicators, directly reflecting the quality of the coating and closely related to the coating parameters. Therefore, this invention monitors the changing trends of these two performance indicators in real time during the mid-to-late stages of training to determine the accuracy of the model's predictions of the coating parameters. This provides a reliable basis for the dynamic adjustment of the learning rate, ensuring that the model training proceeds in a more optimal direction.

[0065] At the beginning of training, since the model parameters differ significantly from the optimal parameters, using an initial learning rate allows for larger step updates of the model parameters in the parameter space, accelerating the convergence speed towards the optimal solution and shortening the overall training time. For example, setting the initial learning rate to 0.01 allows the model to quickly capture the general patterns in the data within fewer training iterations compared to a smaller learning rate, initially establishing the relationship between coating parameters and influencing factors. The initial learning rate is the maximum learning rate throughout the entire training period.

[0066] As training progresses, in the later stages, the model has learned most of the features and patterns in the data and gradually approaches the optimal solution. At this point, the learning rate is dynamically adjusted based on the changing trends of contact resistance and parasitic capacitance.

[0067] As an example, the dynamic adjustment of the learning rate based on the changing trend of performance metrics includes: During training, if the contact resistance or parasitic capacitance values ​​fluctuate significantly, reduce the learning rate by a coefficient less than 1; when both the contact resistance and parasitic capacitance values ​​stabilize and change in the direction of the target performance value, maintain the current learning rate or make small adjustments.

[0068] When significant fluctuations in contact resistance or parasitic capacitance are detected during training, it indicates that the model is adjusting parameters too drastically at the current learning rate. This prevents the model from accurately learning the relationship between coating parameters and performance metrics, potentially causing it to oscillate around the optimal solution and struggle to converge. In such cases, reducing the learning rate with coefficients less than 1 decreases the step size of parameter updates, allowing for more precise model adjustments. For example, assuming a current learning rate of 0.001 and coefficients less than 1 set to 0.1, when significant fluctuations are detected, the learning rate is adjusted to 0.0001, enabling the model to explore the parameter space in smaller steps, gradually stabilizing predictions and moving closer to the optimal solution.

[0069] When the changes in contact resistance and parasitic capacitance stabilize and trend towards the pre-defined target performance values, it indicates that the current learning rate is appropriate and the model is optimizing parameters in the correct direction. At this point, you can choose to maintain the current learning rate, allowing the model to continue training at a stable pace; or you can make small adjustments to further optimize the learning rate, ensuring the model converges to the optimal solution more accurately. The magnitude of the adjustments can be determined based on actual training conditions and experience, such as adjusting the learning rate by ±0.00001, to improve training effectiveness while ensuring stable model training.

[0070] This invention also provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any of the foregoing embodiments.

[0071] This invention also provides a computer storage medium storing a computer program that can be executed by a processor to implement the methods described in any of the foregoing claims.

[0072] This invention also provides a computer program product comprising a computer program that can be executed by a processor to implement the method as described in any of the foregoing embodiments.

[0073] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0074] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for controlling the production of surge-resistant resistors, characterized in that: The methods and steps include the following: A coating parameter optimization model was built based on Reformer, and the model was pre-trained using training data based on fixed process conditions. Before starting this batch of coating, the concentration data of the coating solution and the production environment fluctuation data of the surge resistor are monitored in real time. The production environment fluctuation data includes temperature fluctuation data and humidity fluctuation data. The quality impact index is evaluated and derived based on the concentration data and the production environment fluctuation data. When the quality impact index is higher than the preset value, the concentration data and the production environment fluctuation data are analyzed using the coating parameter optimization model to obtain the coating parameters, which include sputtering pressure, sputtering power, target-substrate distance, substrate temperature and sputtering gas flow rate. Under the same fixed process conditions, the coating equipment is controlled to perform electrode coating on the surge protectors of this batch according to the coating parameters. The step of constructing a coating parameter optimization model and pre-training the coating parameter optimization model using training data based on fixed process conditions includes: constructing coating parameter optimization models corresponding to the bottom layer, functional layer, and outer layer respectively, and pre-training each of the coating parameter optimization models using the corresponding training data based on fixed process conditions. Before using the coating parameter optimization model to analyze the concentration data and the production environment fluctuation data to derive the coating parameters, the process further includes: selecting and deriving the corresponding coating parameter optimization model based on the layer attributes targeted by the current coating; the layer attributes include the bottom layer, functional layer, or outer layer; The quality impact index, derived from the concentration data and the production environment fluctuation data, includes: The entropy values ​​of the concentration data, temperature fluctuation data, and humidity fluctuation data are calculated. Based on the entropy values, weights are calculated using the entropy weight method. A preliminary quality impact index is then evaluated using a multiple linear regression model. The multiple linear regression model is constructed based on each entropy value and its corresponding weight. The equipment aging assessment model and several preset assessment indicators are used to evaluate and obtain the equipment aging degree index, and an adjustment coefficient is obtained based on the equipment aging degree index. The initial quality impact index is corrected using the adjustment coefficient to obtain the quality impact index. Several pre-set evaluation indicators include cumulative equipment operating time, number of times key components have been replaced, and number of recent equipment failures; When constructing the coating parameter optimization model based on Reformer, the multi-head attention mechanism is replaced with a lightweight attention module. The lightweight attention module adopts a sparse attention calculation method, that is, it performs detailed attention calculation only on the feature dimensions related to highly sensitive coating parameters, while performing only linear transformation or skipping some calculation steps on the feature dimensions related to low-sensitive coating parameters.

2. The surge resistor production control method according to claim 1, characterized in that: The highly sensitive coating parameters include contact resistance and parasitic capacitance.

3. The surge resistor production control method according to claim 2, characterized in that: The coating parameter optimization model is pre-trained using the following loss function: ； The coating parameter vector predicted by the coating parameter optimization model is: The actual coating parameter vector is , The number of coating parameters; , is a weighting coefficient used to adjust the proportion of performance index error in the total loss; This is the actual contact resistance. For actual parasitic capacitance, , These are mapping models used to predict contact resistance and parasitic capacitance, respectively. A set of highly sensitive parameters. The weighting coefficients for highly sensitive parameters, ; The weighting coefficients are non-highly sensitive parameters. .

4. The surge resistor production control method according to claim 3, characterized in that: The pre-training of the coating parameter optimization model using training data based on fixed process conditions includes adjusting the learning rate of the coating parameter optimization model as follows: In the early stages of training, an initial learning rate is used, which is the maximum value of the learning rate. In the later stages of training, the learning rate is dynamically adjusted according to the changing trends of performance indicators, which refer to contact resistance and parasitic capacitance.

5. The surge resistor production control method according to claim 4, characterized in that: The method of dynamically adjusting the learning rate based on the changing trend of performance indicators includes: During training, if the contact resistance or parasitic capacitance values ​​fluctuate significantly, reduce the learning rate by a coefficient less than 1; when both the contact resistance and parasitic capacitance values ​​stabilize and change in the direction of the target performance value, maintain the current learning rate or make small adjustments.

Images