Artificial intelligence optimization method for screen evaporation process

Abstract

The invention provides an artificial intelligence optimization method for a screen evaporation process. The artificial intelligence optimization method comprises the following steps: acquiring multi-dimensional data of a screen in an evaporation process; inputting the multi-dimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features; the evaporation state characteristics are used for representing error distribution and / or state information existing in the evaporation process; the evaporation state characteristics are input into a reinforcement learning module in an evaporation error prediction model, and optimization parameters of the screen in the evaporation process are obtained; the optimization parameters are used for adjusting equipment parameters of the evaporation process equipment so as to optimize the evaporation process of the screen. And by utilizing the trained feature extraction module and reinforcement learning module, according to the multi-dimensional data of the screen in the evaporation process, obtaining the equipment parameters of the evaporation process equipment, so as to optimize the evaporation process of the screen, realize quick response, dynamically adjust the equipment parameters, reduce the influence of personal error and subjective judgment, and improve the consistency of the evaporation quality.

Description

Technical Field

[0001] The present application relates to the field of screen evaporation technology, and in particular to an artificial intelligence optimization method for a screen evaporation process. Background Art

[0002] In the production of display screens, the vacuum evaporation process plays a vital role, directly determining the luminous efficiency, color reproduction and service life of the screen. However, due to the complex environmental and material factors involved in the vacuum evaporation process, error control in the production process has always been a problem that has plagued the industry. At present, traditional vacuum evaporation error control relies on manual control, which results in low optimization efficiency and accuracy of the evaporation process. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide an artificial intelligence optimization method, program product, electronic device and storage medium for a screen evaporation process, which are used to improve the problem of low optimization efficiency and accuracy of the above-mentioned evaporation process.

[0004] In the first aspect, an embodiment of the present application provides an artificial intelligence optimization method for a screen evaporation process, including: obtaining multidimensional data of the screen during the evaporation process; inputting the multidimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state characteristics; the evaporation state characteristics are used to characterize the error distribution and / or state information existing in the evaporation process; the evaporation state characteristics are input into a reinforcement learning module in the evaporation error prediction model to obtain optimization parameters of the screen during the evaporation process; the optimization parameters are used to adjust equipment parameters used for the evaporation process equipment to optimize the evaporation process of the screen.

[0005] In the above implementation process, the trained feature extraction module and reinforcement learning module are used to obtain the equipment parameters of the evaporation process equipment according to the multi-dimensional data of the screen during the evaporation process to optimize the evaporation process of the screen. Rapid response is achieved, and equipment parameters are adjusted dynamically to reduce the impact of human errors and subjective judgments, and improve the consistency of evaporation quality. The reinforcement learning module can find the optimal equipment parameter settings based on historical data and complex algorithms to improve the evaporation process of the screen, while manual control often relies on experience and may not be able to achieve the global optimal solution. The above-mentioned automated control can be executed uninterruptedly without waiting for manual operation or supervision, thereby improving efficiency.

[0006] Optionally, in an embodiment of the present application, before inputting multidimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state characteristics, the method also includes: obtaining historical multidimensional training data, the historical multidimensional training data including at least one of temperature distribution data, film thickness distribution data or mask deformation data; inputting the historical multidimensional training data into a pre-set convolutional neural network to obtain a predicted output; calculating the difference between the predicted output and the true value corresponding to the historical multidimensional training data, and updating the parameters of the convolutional neural network based on the difference using a back propagation algorithm to obtain a trained feature extraction module.

[0007] In the above implementation process, the accuracy and consistency of the evaporation process are improved through automated feature extraction and parameter optimization. The feature extraction module can quickly process data and provide real-time feedback, improving production efficiency. It also generates evaporation state features to provide accurate data feature support for the reinforcement learning module, which helps the reinforcement learning module make more accurate decisions.

[0008] Optionally, in an embodiment of the present application, after inputting multidimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state characteristics, the method also includes: during the evaporation process of the screen, using a network sensor to collect multidimensional data in real time; using the multidimensional data and the evaporation state characteristics to perform error calculations to obtain a feedback signal; calculating the parameter gradient of the feature extraction module based on the feedback signal, and adjusting the weight of the feature extraction module in real time to obtain an adjusted feature extraction module, so that subsequent multidimensional data uses the adjusted feature extraction module to obtain corresponding evaporation state characteristics.

[0009] In the above implementation process, in the actual production process, the real-time collected data provides immediate feedback and adjustment opportunities, effectively realizing the dynamic optimization of the feature extraction module and the precise control of the production process. In this way, the performance of the feature extraction module is improved and changes in the production process are responded to in a timely manner.

[0010] Optionally, in an embodiment of the present application, the parameter gradient of the feature extraction module is calculated based on the feedback signal, and the weight of the feature extraction module is adjusted in real time to obtain an adjusted feature extraction module, including: using the feedback signal to calculate the parameter gradient of the feature extraction module through a back propagation algorithm; using an optimization algorithm to update the weight and bias of the feature extraction module according to the parameter gradient to obtain an adjusted feature extraction module.

[0011] In the above implementation process, by utilizing the feedback signal, the model can adjust its parameters in time to better adapt to the current evaporation process, improve the accuracy of the feature extraction module in extracting the evaporation state features, and thus improve the quality of the evaporation process.

[0012] Optionally, in an embodiment of the present application, before inputting the evaporation state characteristics into the reinforcement learning module in the evaporation error prediction model to obtain the optimized parameters of the screen during the evaporation process, the method also includes: obtaining historical multi-dimensional training data and a pre-defined action space and strategy network; the action space is used to adjust the equipment parameters of the evaporation process equipment; the evaporation equipment parameters include at least one of evaporation source power, heating current, vacuum degree, mask position or mask angle; according to the strategy network, determining the target action from the action space, and executing it on the historical multi-dimensional training data to obtain observation results; the observation results are used to represent the film quality corresponding to the historical multi-dimensional training data; calculating reward feedback based on the observation results; the reward feedback is used to evaluate the effect of the action space; updating the strategy network according to the reward feedback to obtain a reinforcement learning module.

[0013] In the above implementation process, through the combination of historical data and strategy network, the optimal equipment parameter adjustment strategy can be selected more accurately to improve the accuracy of decision-making; through real-time feedback and adjustment, the evaporation process is dynamically optimized to improve the film quality and consistency, thereby significantly improving the film accuracy, material utilization and product yield. Reduce manual intervention, improve the degree of automation of the production process, and reduce the risk of human error.

[0014] Optionally, in an embodiment of the present application, the multidimensional data includes at least one of temperature distribution data, film thickness distribution data or mask deformation data; obtaining the multidimensional data of the screen during the evaporation process includes: during the evaporation process, using a thin film thickness detector to scan and sample the mask of the screen to generate film thickness distribution data; and / or, using a thermocouple array and an infrared thermal imager to monitor the temperature distribution of a preset part in real time to generate a three-dimensional temperature distribution map; and / or, using a laser displacement sensor and a strain sensor to scan and monitor the thermal expansion state of the mask at high temperature to obtain mask deformation data.

[0015] In the above implementation process, real-time data acquisition provides data support for the feature extraction module and reinforcement learning module in the evaporation error prediction model, thereby obtaining the optimized parameters of the screen during the evaporation process, reducing manual monitoring and intervention, improving the degree of production automation, and reducing the risk of human errors.

[0016] Optionally, in an embodiment of the present application, after obtaining the optimized parameters of the screen during the evaporation process, the method also includes: obtaining an optimization target, the optimization target including at least one of thickness uniformity, material utilization rate or yield rate; establishing an objective function for each optimization target, the objective function being an evaluation criterion for calculating fitness; calculating the fitness of the optimization parameters under each objective function; the fitness is used to characterize the performance quantification value of the optimization parameters; according to the fitness, using a multi-objective optimization method based on a genetic algorithm to determine a combination of process parameters; the combination of process parameters is used to achieve a multi-objective balance.

[0017] In the above implementation process, through this multi-objective optimization method based on genetic algorithm, a dynamic balance can be achieved among multiple production objectives to optimize objectives such as thickness uniformity, material utilization and yield rate, thereby improving material utilization and production efficiency while ensuring product quality.

[0018] In the second aspect, the embodiment of the present application also provides an artificial intelligence optimization device for the screen evaporation process, including: a data acquisition module, used to obtain multi-dimensional data of the screen during the evaporation process; a feature extraction module, used to input the multi-dimensional data into the feature extraction module in the pre-trained evaporation error prediction model to obtain evaporation state characteristics; the evaporation state characteristics are used to characterize the error distribution and / or state information existing in the evaporation process. A reinforcement learning module, used to input the evaporation state characteristics into the reinforcement learning module in the evaporation error prediction model to obtain the optimization parameters of the screen during the evaporation process; the optimization parameters are used to adjust the equipment parameters used for the evaporation process equipment to optimize the evaporation process of the screen.

[0019] In a third aspect, an embodiment of the present application further provides a computer program product, including computer program instructions, which, when executed by a processor, execute the method provided by the first aspect or any one of the implementations of the first aspect.

[0020] In a fourth aspect, an embodiment of the present application further provides an electronic device, comprising: a processor and a memory, the memory storing computer program instructions, and the computer program instructions, when executed by the processor, execute the method provided by the first aspect or any one of the implementations of the first aspect.

[0021] In a fifth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method provided by the first aspect or any one of the implementations of the first aspect is executed.

[0022] The artificial intelligence optimization method, program product, electronic device and storage medium of the screen evaporation process provided by the present application are adopted, and the trained feature extraction module and reinforcement learning module are used to obtain the equipment parameters of the evaporation process equipment according to the multi-dimensional data of the screen during the evaporation process to optimize the evaporation process of the screen. Rapid response is achieved, and equipment parameters are adjusted dynamically to reduce the impact of human errors and subjective judgments, and improve the consistency of evaporation quality. The reinforcement learning module can find the optimal equipment parameter settings based on historical data and complex algorithms to improve the evaporation process of the screen, while manual control often relies on experience and may not be able to achieve the global optimal solution. The above-mentioned automated control can be executed uninterruptedly without waiting for manual operation or supervision, thereby improving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 A schematic diagram of a process flow of an artificial intelligence optimization method for a screen evaporation process provided in an embodiment of the present application;

[0025] Figure 2 This is a schematic diagram of the evaporation error prediction model provided in the embodiment of the present application;

[0026] Figure 3 A schematic diagram of the structure of an artificial intelligence optimization device for a screen evaporation process provided in an embodiment of the present application;

[0027] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The following embodiments of the technical solution of the present application are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0030] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0031] The vacuum evaporation process in the production of display screens plays a decisive role in the luminous efficiency, color reproduction and service life of the screen. However, there are many complex factors in the vacuum evaporation process, such as environmental and material factors, which makes error control in the production process a major challenge. The traditional error control method relies on manual control or preset parameter thresholds for error detection. This method has the following problems: it is impossible to feedback and correct errors in real time, making it difficult to ensure product consistency. Parameters such as temperature and vacuum fluctuate rapidly due to the environment and equipment conditions, and traditional manual adjustments are difficult to respond quickly to nonlinear changes. Mask wear and thermal expansion effects are another major source of errors. Traditional methods are difficult to dynamically compensate for nonlinear changes in masks. In addition, this method has low efficiency and accuracy, and is greatly affected by human errors and subjective judgments, which is not conducive to the consistency of evaporation quality. Manual control may result in inaccurate parameter adjustments due to the operator's experience and emotional fluctuations, and therefore cannot meet the needs of optimizing the evaporation process.

[0032] The embodiment of the present application provides an artificial intelligence optimization method for the screen evaporation process, which uses a trained feature extraction module and a reinforcement learning module to obtain the equipment parameters of the evaporation process equipment according to the multi-dimensional data of the screen during the evaporation process, so as to optimize the evaporation process of the screen, achieve rapid response, dynamically adjust the equipment parameters, reduce the impact of human errors and subjective judgments, and improve the consistency of evaporation quality. The reinforcement learning module can find the optimal equipment parameter settings based on historical data and complex algorithms to improve the evaporation process of the screen, while manual control often relies on experience and may not be able to achieve the global optimal solution. The above-mentioned automated control can be executed uninterruptedly without waiting for manual operation or supervision, thereby improving efficiency.

[0033] See also Figure 1 A schematic flow chart of an artificial intelligence optimization method for a screen evaporation process provided by an embodiment of the present application is shown. The artificial intelligence optimization method for a screen evaporation process provided by an embodiment of the present application can be applied to an electronic device, which may include a physical device such as a server, a PC, a tablet computer, or a smart phone, or a virtual device such as a virtual machine or a container. The electronic device may be a single device, a combination of multiple devices, or a cluster of a large number of devices. The artificial intelligence optimization method for the screen evaporation process may include:

[0034] Step S110: Acquire multi-dimensional data of the screen during the evaporation process.

[0035] Step S120: input the multi-dimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features; the evaporation state features are used to characterize the error distribution and / or state information existing in the evaporation process.

[0036] Step S130: inputting the evaporation state characteristics into the reinforcement learning module in the evaporation error prediction model to obtain the optimized parameters of the screen during the evaporation process; the optimized parameters are used to adjust the equipment parameters used for the evaporation process equipment to optimize the evaporation process of the screen.

[0037] In step S110, the screen may be an OLED screen, an LCD screen or other electronic display screen, etc. The evaporation process of the screen refers to the use of evaporation technology in the screen manufacturing process, that is, evaporating the material by heating or other means in a vacuum environment, and then allowing the evaporated atoms or molecules to condense on the substrate to form a thin film.

[0038] The screen evaporation process can include the following steps: 1. Preparation stage: The substrate needs to be cleaned and dried to remove dirt and impurities on the surface. Then, the substrate is placed in a vacuum chamber and the vacuum degree is adjusted to the required level. 2. Heating evaporation: Next, the evaporation source (i.e., the material to be evaporated) is heated to the evaporation point. This can be achieved by resistance heating, electron beam heating, or laser heating. As the temperature rises, the material gradually evaporates and turns into a gaseous state. 3. Atomic or molecular movement: The evaporated atoms or molecules move in a straight line with a large free path in a vacuum. These particles may collide with other particles during movement, but generally they move toward the substrate. 4. Condensation into film: When atoms or molecules collide with the surface of the substrate, they lose energy and gradually condense into a thin film. In this process, the thickness, composition, uniformity, film forming accuracy, material utilization rate, and product yield of the film are all affected by factors such as the temperature of the evaporation source, the vacuum degree, the temperature of the substrate, and the evaporation time. These factors can be adjusted through the equipment parameters of the process equipment used for evaporation, thereby controlling and optimizing the evaporation process of the screen.

[0039] Taking the screen as an OLED screen as an example, multi-dimensional data refers to various parameter data collected in real time during the OLED evaporation process, including temperature distribution data, film thickness distribution data, mask deformation data, and pressure state and / or mask state data. Among them, the temperature distribution data includes the temperature fluctuation of the evaporation source, the temperature distribution of the substrate support platform and / or the temperature fluctuation of the mask. The collection method can use special sensors and measuring equipment (such as film thickness detectors, thermocouple arrays, infrared thermal imagers, laser displacement sensors and strain sensors) to monitor the evaporation process in real time to obtain multi-dimensional data.

[0040] In step S120, the evaporation error prediction model includes a feature extraction module and a reinforcement learning module. The feature extraction module may be a convolutional neural network (CNN) module, and the network structure of the feature extraction module may include a convolution layer, a pooling layer, a fully connected layer, etc. The feature extraction module can extract evaporation state features useful for predicting evaporation errors from the original multi-dimensional data, and these evaporation state features can represent the error distribution and / or state information existing in the evaporation process.

[0041] Exemplarily, the multidimensional data can be input into the feature extraction module in the pre-trained evaporation error prediction model. The feature extraction module processes the input data through the convolution layer, slides the convolution kernel on the data, and extracts local features. The convolution operation can capture the spatial hierarchy of the data, such as edges, textures, etc. After the convolution operation, a pooling operation, such as maximum pooling or average pooling, can be performed to reduce the spatial size of the feature map, reduce the computational complexity, and retain important features. After multiple layers of convolution and pooling, the multidimensional data is converted (or mapped) into one or more groups of evaporation state features, which can represent the important features of the input data while filtering out noise. Among them, the error distribution is such as the possible thickness error distribution map during the evaporation process; the state information is such as the pattern position offset (displacement vector Δx, Δy) caused by the mask. The evaporation state feature provides the reinforcement learning module with a detailed description of the current evaporation state, which helps the reinforcement learning module make more accurate decisions.

[0042] In step S130, the reinforcement learning module is used to optimize the equipment parameters by learning the historical multi-dimensional data of the evaporation process. The reinforcement learning module can determine the best or better equipment parameter adjustment strategy according to the current evaporation state characteristics to optimize the evaporation process of the screen.

[0043] The evaporation state characteristics are input into the reinforcement learning module in the evaporation error prediction model. The reinforcement learning module calculates a set of optimization parameters based on these characteristics and the strategies it has learned through historical data, namely, the equipment parameters of the evaporation process equipment, such as evaporation source power, heating current, vacuum degree, mask position and angle, etc. These equipment parameters are transmitted to the evaporation equipment and the equipment is adjusted in real time to optimize the evaporation process. For example, the equipment parameters can be automatically adjusted according to the above optimization parameters through the control system interface of the equipment.

[0044] In the implementation process of the above embodiment: using the trained feature extraction module and reinforcement learning module, according to the multi-dimensional data of the screen during the evaporation process, the equipment parameters of the evaporation process equipment are obtained to optimize the evaporation process of the screen, achieve rapid response, dynamically adjust the equipment parameters, reduce the impact of human errors and subjective judgments, and improve the consistency of evaporation quality. The reinforcement learning module can find the optimal equipment parameter settings based on historical data and complex algorithms to improve the evaporation process of the screen, while manual control often relies on experience and may not be able to achieve the global optimal solution. The above-mentioned automated control can be executed uninterruptedly without waiting for manual operation or supervision, thereby improving efficiency.

[0045] Optionally, in an embodiment of the present application, before inputting the multidimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features, the feature extraction module is first trained, and the method further includes:

[0046] Obtain historical multidimensional training data, the historical multidimensional training data including at least one of temperature distribution data, film thickness distribution data or mask deformation data. For example, the historical multidimensional training data can be collected during the historical evaporation process, and the historical multidimensional training data is used to form a training set.

[0047] The historical multi-dimensional training data is input into a pre-set convolutional neural network to obtain the predicted output. The pre-set convolutional neural network is an early-stage CNN model with multiple convolutional layers, pooling layers, and / or fully connected layers for extracting features from multi-dimensional data. The historical multi-dimensional training data is input into the convolutional neural network, and the model extracts features from the data through its convolutional layers and learns the associations in the historical multi-dimensional training data.

[0048] Calculate the difference between the predicted output and the true value corresponding to the historical multi-dimensional training data, and use the back propagation algorithm based on the difference to update the parameters of the convolutional neural network to obtain a trained feature extraction module. The true value is the actual measurement value corresponding to the historical multi-dimensional training data. The true value represents the actual result of the error distribution after the evaporation process is completed, such as the actual film thickness, temperature distribution, etc. The difference calculation can use a certain loss function (such as mean square error MSE) to calculate the difference between the predicted output and the true value. This difference is an indicator to measure the accuracy of the model prediction.

[0049] The back propagation algorithm is a method for training neural networks, which updates the parameters by calculating the gradient of the loss function with respect to the network parameters. Parameter update refers to the use of the calculated gradient to update the weights and biases of the CNN using optimization algorithms (such as Adam, SGD, etc.), with the goal of reducing the difference between the predicted output and the true value. After sufficient iterative training, the parameters of the CNN model will be optimized so that it can accurately extract the evaporation state features that are useful for predicting evaporation errors from multidimensional data, complete the training, and obtain a trained feature extraction module.

[0050] In the implementation process of the above embodiment: through automated feature extraction and parameter optimization, the accuracy and consistency of the evaporation process are improved. The feature extraction module can quickly process data and provide real-time feedback, thereby improving production efficiency. In addition, the evaporation state features are generated to provide accurate data feature support for the reinforcement learning module, which helps the reinforcement learning module make more accurate decisions.

[0051] Optionally, in the embodiment of the present application, after inputting the multi-dimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features, the method further includes:

[0052] During the evaporation process of the screen, network sensors are used to collect multi-dimensional data in real time. For example, during the evaporation process of the OLED screen, a networked sensor system is deployed. These sensors can monitor and record key parameters of the evaporation process in real time, such as temperature, pressure, mask position and shape, etc. The network sensors transmit the collected data to the central processing system through wired (such as industrial Ethernet) or wireless (such as Wi-Fi or Internet of Things protocol) networks.

[0053] The multi-dimensional data and the characteristics of the evaporation state are used to calculate the error and obtain the feedback signal. The feedback signal can be calculated in the form of a simple difference or a more complex error metric such as mean square error (MSE).

[0054] Considering that in a dynamic production process such as OLED evaporation, the environment and equipment status may change at any time. The real-time collected data represents the actual situation at the current moment, while the prediction output of the CNN module is based on the immediate response of the current model to these real-time data. By comparing the two, the deviation between the model prediction and the actual observation can be immediately found, and the model parameters can be adjusted accordingly to achieve rapid response.

[0055] Because in actual operation, obtaining the true value (such as the final film quality index) may require waiting until the production process is completed, which is time-delayed. The multi-dimensional data collected in real time provides instant feedback, allowing the model to be fine-tuned during the production process instead of waiting until the production is completed. This instant feedback (also known as online learning) is crucial to improving production efficiency and quality control.

[0056] The parameter gradient of the feature extraction module is calculated based on the feedback signal, and the weight of the feature extraction module is adjusted in real time to obtain an adjusted feature extraction module, so that subsequent multi-dimensional data can use the adjusted feature extraction module to obtain corresponding evaporation state features.

[0057] In the implementation process of the above embodiment: in the actual production process, the real-time collected data provides immediate feedback and adjustment opportunities, effectively realizing the dynamic optimization of the feature extraction module and the precise control of the production process. In this way, the performance of the feature extraction module is improved and changes in the production process are responded to in a timely manner.

[0058] Optionally, in an embodiment of the present application, a parameter gradient of a feature extraction module is calculated based on a feedback signal, and a weight of the feature extraction module is adjusted in real time to obtain an adjusted feature extraction module, including:

[0059] The feedback signal is used to calculate the parameter gradient of the feature extraction module through the back propagation algorithm; the optimization algorithm is used to update the weight and bias of the feature extraction module according to the parameter gradient to obtain the adjusted feature extraction module.

[0060] For example, the gradient of the error to the model weight can be calculated through backpropagation, the weights of the convolutional layer and the fully connected layer of the CNN module can be updated, and the weights in the policy network of the reinforcement learning module can be updated to optimize the action selection strategy. The formula for calculating the parameter gradient is as follows:

[0061]

[0062] in, is the parameter gradient, is the weight of the convolutional layer or the fully connected layer, is the loss function.

[0063] Through the gradient of the weights, using the Adam optimizer, the model parameters are updated using the gradient descent method:

[0064]

[0065] Where η is the learning rate, is the gradient of the weight, W new is the updated weight, Wold is the weight before update. Use a smaller learning rate (generally η = 10e-4) to ensure that each update is small but can gradually improve model performance.

[0066] After updating the weights and biases, the feature extraction module (CNN) is adjusted to better adapt to the current evaporation process. The adjusted feature extraction module can more accurately extract the evaporation state features from the multidimensional data. This adjustment process is dynamic and can be continuously carried out with the real-time data of the evaporation process, so that the model is always in the best state.

[0067] As an implementation method, online learning can use regularization technology to limit the amplitude of weight updates, regularly mix multi-dimensional data collected in real time with historical data, perform batch training, and maintain the long-term stability of the model; through sliding window technology, dynamically select data from the most recent period for updating to ensure that the model adapts to the current situation and is more robust to short-term anomalies. Finally, when the verification set error has not decreased significantly for many consecutive times, stop the current online learning update to avoid overfitting.

[0068] In the Bayesian optimization module, after fine-tuning the model weights, the model's hyperparameters (such as learning rate, regularization coefficient, and loss function weight) will be efficiently optimized to further improve the model performance. The optimization goal is defined as minimizing the model error:

[0069]

[0070] Where L(θ; D) is the prediction error, R(θ) is the regularization term, and λ is the regularization coefficient. The Gaussian process (GP) is used to model the objective function in the hyperparameter space. Based on the sampled historical hyperparameters and their corresponding model errors, the target value of the unsampled point is predicted. Then, the expected improvement (EI) or upper confidence bound (UCB) strategy is used to select the next set of hyperparameters from the unexplored parameter space. The model is updated with the selected hyperparameters, and the loss or accuracy on the validation set after each hyperparameter update is recorded to observe whether the performance tends to stabilize. The performance improvement threshold ε (generally ε<10) is set. -4 ), if the model performance has not been significantly improved for many consecutive times, the Bayesian optimization is stopped in advance; thus, the Bayesian optimization module can help the feature extraction model quickly converge to the optimal parameters at a low computational cost.

[0071] In the implementation process of the above embodiment: by utilizing the feedback signal, the model can adjust its parameters in time to better adapt to the current evaporation process, improve the accuracy of the feature extraction module in extracting the evaporation state features, and thus improve the quality of the evaporation process.

[0072] Optionally, in the embodiment of the present application, before inputting the evaporation state characteristics into the reinforcement learning module in the evaporation error prediction model to obtain the optimized parameters of the screen during the evaporation process, the method further includes:

[0073] Obtain historical multidimensional training data and predefined action space and strategy network; the action space is used to adjust the equipment parameters of the evaporation process equipment; the evaporation equipment parameters include at least one of evaporation source power, heating current, vacuum degree, mask position or mask angle. Historical multidimensional training data refers to multidimensional data collected during past evaporation processes. The strategy network is a trained neural network used to select the optimal action based on the current state. The strategy network outputs a probability distribution of an action based on the input evaporation state characteristics.

[0074] According to the policy network, the target action is determined from the action space and executed on the historical multidimensional training data to obtain the observation results; the observation results are used to represent the film quality corresponding to the historical multidimensional training data. According to the current evaporation state characteristics, the policy network will select an optimal target action from the predefined action space. The selected target action is applied to the evaporation equipment to adjust the equipment parameters. After executing the target action, the results of the evaporation process are recorded. These results are used to represent the film quality corresponding to the historical multidimensional training data, such as film thickness, uniformity, etc.

[0075] Reward feedback is calculated based on the observations; reward feedback is used to evaluate the effect of the action space. Reward feedback is calculated based on the difference between the observations and the expected goals (such as film thickness, uniformity, etc.). Reward feedback is a quantitative indicator that can be calculated in the following way: If the observations meet expectations, a positive reward is given. If the observations do not meet expectations, a negative reward or zero reward is given. Reward feedback is used to evaluate the effect of the actions taken and help the model understand which actions are effective and which are ineffective.

[0076] Update the policy network according to the reward feedback to obtain a reinforcement learning module. Use reward feedback to adjust the parameters of the policy network. Common methods include policy gradient methods (such as REINFORCE) or Q-learning. By calculating the impact of reward feedback on the policy network parameters, use optimization algorithms (such as Adam, SGD, etc.) to update the weights of the policy network to improve the accuracy of future action selection. The updated policy network forms a trained reinforcement learning module that can better select the optimal action based on the current state.

[0077] In the implementation process of the above embodiment: through the combination of historical data and strategy network, the optimal equipment parameter adjustment strategy can be selected more accurately to improve the accuracy of decision-making; through real-time feedback and adjustment, the evaporation process is dynamically optimized to improve the film quality and consistency, thereby significantly improving the film accuracy, material utilization and product yield. Reduce manual intervention, improve the degree of automation of the production process, and reduce the risk of human error.

[0078] Optionally, in an embodiment of the present application, the multidimensional data includes at least one of temperature distribution data, film thickness distribution data, or mask deformation data; acquiring the multidimensional data of the screen during the evaporation process includes:

[0079] During the evaporation process, a thin film thickness detector is used to scan and sample the mask of the screen to generate film thickness distribution data. For example, non-contact measurement technology can be used to obtain film thickness distribution data in real time. The film thickness distribution data reflects the uniformity and thickness of the deposition of the evaporation material on the substrate. The film thickness distribution data can be a three-dimensional thickness distribution graph.

[0080] And / or, using a thermocouple array and an infrared thermal imager to monitor the temperature distribution of a preset part in real time and generate a three-dimensional temperature distribution map. Using a thermocouple array and an infrared thermal imager to monitor the temperature distribution of a preset part in real time. The thermocouple array can provide accurate temperature point measurement, while the infrared thermal imager can generate a three-dimensional temperature distribution map of the entire area.

[0081] And / or, use laser displacement sensors and strain sensors to scan, monitor the thermal expansion state of the mask at high temperature, and obtain mask deformation data. Use laser displacement sensors and strain sensors to perform regular scans to monitor the thermal expansion state of the mask at high temperature. Laser displacement sensors and strain sensors can detect small deformations of the mask caused by temperature changes.

[0082] The collected data (film thickness distribution, temperature distribution, mask deformation data) can be integrated to form multi-dimensional data of the screen during the evaporation process. Of course, it is also possible not to integrate and directly use the collected data as multi-dimensional data.

[0083] In the implementation process of the above embodiment: real-time data acquisition provides data support for the feature extraction module and reinforcement learning module in the evaporation error prediction model, thereby obtaining the optimized parameters of the screen during the evaporation process, reducing manual monitoring and intervention, improving the degree of production automation, and reducing the risk of human error.

[0084] Optionally, in the embodiment of the present application, after obtaining the optimized parameters of the screen during the evaporation process, the method further includes:

[0085] Obtain optimization targets, which include at least one of thickness uniformity, material utilization, or yield rate. Thickness uniformity refers to the consistency of film thickness across the entire screen, which affects display effects and product quality. Material utilization refers to the proportion of material effectively deposited during the evaporation process, reflecting the efficiency of material use. Yield rate refers to the ratio of qualified products to the total production quantity, which directly affects production efficiency and cost.

[0086] Different optimization objectives may compete or influence each other, and improving one objective may have a negative impact on other objectives. Therefore, it is necessary to define functions that quantify these objectives and find a set of better solutions that balance multiple objectives. This better solution represents the optimal process parameter combination under the trade-off of different optimization objectives.

[0087] An objective function is established for each optimization goal, and the objective function is an evaluation criterion for calculating fitness.

[0088] Exemplarily, the objective function of thickness uniformity is: f1(x)=V(T), where V(T) represents the thickness variation amplitude;

[0089] Objective function of material utilization:

[0090] The objective function of yield rate is:

[0091] Calculate the fitness of the optimization parameters under each objective function; the fitness is used to characterize the performance quantification value of the optimization parameters; the higher the fitness, the better the performance of the optimization parameters under the objective. The calculation method can input the obtained optimization parameters into each objective function and calculate the corresponding fitness value. The fitness value will be used for subsequent optimization decisions.

[0092] According to the fitness, the process parameter combination is determined using a multi-objective optimization method based on a genetic algorithm; the process parameter combination is used to achieve multi-objective balance. Genetic algorithm is an optimization algorithm based on the principles of natural selection and genetics, suitable for solving complex multi-objective optimization problems. According to the fitness value, the genetic algorithm will select a parameter combination with high fitness, perform crossover and mutation operations, and generate a new process parameter combination.

[0093] For example, based on the fitness value, tournament selection is adopted to retain individuals with better performance (optimized parameters), and then crossover is performed on the selected individuals to generate a new process parameter combination: new =αx1+(1-α)x2, where x new represents the new process parameters, x represents the specific solution in the solution space (x={(power), (vacuum degree), (mask position)}), α is the random weight, and a new process parameter combination is obtained.

[0094] At the same time, the mutation operation randomly changes the values ​​of certain parameters of individuals through mutation processes such as site mutation, uniform mutation and Gaussian mutation to explore local or unknown process parameter areas, increase the genetic diversity of the population, and help the algorithm avoid premature convergence to the local optimum. Then the algorithm selects a new generation of populations according to non-dominated sorting, gradually approaches the Pareto optimal solution set, and through multiple rounds of iterations, the overall fitness of the population continues to improve, and the Pareto frontier (a set of optimal solutions, in which any improvement in a certain goal of any solution must be at the expense of some losses in other goals) gradually converges. In this way, the combination of process parameters that best suits the current production conditions is obtained.

[0095] In the implementation process of the above embodiment: through this multi-objective optimization method based on genetic algorithm, a dynamic balance can be achieved between multiple production objectives to optimize objectives such as thickness uniformity, material utilization and yield rate, thereby improving material utilization and production efficiency while ensuring product quality.

[0096] See also Figure 2 The structure diagram of the evaporation error prediction model provided in the embodiment of the present application is shown, wherein the evaporation error prediction model is also a mask vacuum evaporation error prediction model.

[0097] In an optional embodiment, a thin film thickness detector is used to scan and sample the mask using a non-contact sensor to generate film thickness distribution data; a thermocouple array and an infrared thermal imager are used to arrange multiple temperature sensors at each key position (such as the evaporation source and different areas of the mask) to monitor the temperature distribution of the evaporation source, the mask and the substrate in real time to generate a three-dimensional temperature distribution map; a laser displacement sensor and a strain sensor are used to scan and monitor the thermal expansion state of the mask at high temperature to obtain mask deformation data, which can be a mask deformation image.

[0098] The feature extraction module (convolutional neural network (CNN) module) uses the local perception and spatial invariance characteristics of CNN, through convolution operations and pooling mechanisms, to extract key features from the multi-dimensional data collected during the evaporation process (such as temperature distribution, mask deformation image, film thickness distribution, etc.) and filter out noise to ensure the efficiency and accuracy of feature extraction.

[0099] The feature extraction module defines multi-channel data input (such as thickness distribution map, temperature distribution map) based on the extracted key features, sends the noise-filtered data into the input channel, and gradually adjusts the model parameters until the error prediction reaches the optimal value. The mean square error (MSE) is used to calculate the difference between the predicted value and the true value, and then the prediction performance of the model is evaluated on the validation set.

[0100] In the reinforcement learning module, the multidimensional parameters extracted from the historical data and the feature information output by the feature extraction module (such as temperature, thickness, and mask deformation distribution) are used to correspond to different strategies for adjusting the evaporation parameters (such as adjusting the power, current, vacuum degree, etc.), and then a reward value is assigned to each action based on the film quality in the historical data (thickness uniformity, yield rate, etc.).

[0101] The BCQ (Batch-Constrained Q-learning) algorithm is used to learn strategies from existing data. Strategies are learned based on historical data to prevent strategies from deviating from the distribution of training data.

[0102] See also Figure 3 The schematic diagram of the structure of the artificial intelligence optimization device for the screen evaporation process provided by the embodiment of the present application is shown; the embodiment of the present application provides an artificial intelligence optimization device 200 for the screen evaporation process, including:

[0103] The data acquisition module 210 is used to acquire multi-dimensional data of the screen during the evaporation process;

[0104] The feature extraction module 220 is used to input the multi-dimensional data into the feature extraction module in the pre-trained evaporation error prediction model to obtain the evaporation state features; the evaporation state features are used to characterize the error distribution and / or state information existing in the evaporation process.

[0105] The reinforcement learning module 230 is used to input the evaporation state characteristics into the reinforcement learning module in the evaporation error prediction model to obtain the optimization parameters of the screen during the evaporation process; the optimization parameters are used to adjust the equipment parameters used for the evaporation process equipment to optimize the evaporation process of the screen.

[0106] Optionally, in an embodiment of the present application, the artificial intelligence optimization device 200 for the screen evaporation process also includes a training feature extraction module for obtaining historical multidimensional training data, the historical multidimensional training data including at least one of temperature distribution data, film thickness distribution data or mask deformation data; inputting the historical multidimensional training data into a pre-set convolutional neural network to obtain a predicted output; calculating the difference between the predicted output and the true value corresponding to the historical multidimensional training data, and updating the parameters of the convolutional neural network based on the difference using a back propagation algorithm to obtain a trained feature extraction module.

[0107] Optionally, in an embodiment of the present application, the artificial intelligence optimization device 200 for the screen evaporation process also includes an adjustment module, which is used to use a network sensor to collect multidimensional data in real time during the evaporation process of the screen; use the multidimensional data and the evaporation state characteristics to perform error calculations to obtain a feedback signal; calculate the parameter gradient of the feature extraction module based on the feedback signal, and adjust the weight of the feature extraction module in real time to obtain an adjusted feature extraction module, so that subsequent multidimensional data uses the adjusted feature extraction module to obtain corresponding evaporation state characteristics.

[0108] Optionally, in an embodiment of the present application, the artificial intelligence optimization device 200 for the screen evaporation process, the adjustment module, is also used to use the feedback signal to calculate the parameter gradient of the feature extraction module through the back propagation algorithm; use the optimization algorithm to update the weight and bias of the feature extraction module according to the parameter gradient to obtain the adjusted feature extraction module.

[0109] Optionally, in an embodiment of the present application, the artificial intelligence optimization device 200 for the screen evaporation process also includes a training reinforcement learning module for acquiring historical multi-dimensional training data and a pre-defined action space and strategy network; the action space is used to adjust the equipment parameters of the evaporation process equipment; the evaporation equipment parameters include at least one of the evaporation source power, heating current, vacuum degree, mask position or mask angle; according to the strategy network, the target action is determined from the action space, and is executed on the historical multi-dimensional training data to obtain observation results; the observation results are used to represent the film quality corresponding to the historical multi-dimensional training data; reward feedback is calculated based on the observation results; the reward feedback is used to evaluate the effect of the action space; the strategy network is updated according to the reward feedback to obtain a reinforcement learning module.

[0110] Optionally, in an embodiment of the present application, the artificial intelligence optimization device 200 for the screen evaporation process, the multidimensional data includes at least one of temperature distribution data, film thickness distribution data or mask deformation data; the data acquisition module 210 is specifically used to scan and sample the mask of the screen during the evaporation process using a thin film thickness detector to generate film thickness distribution data; and / or, use a thermocouple array and an infrared thermal imager to monitor the temperature distribution of a preset part in real time to generate a three-dimensional temperature distribution map; and / or, use a laser displacement sensor and a strain sensor to scan, monitor the thermal expansion state of the mask at high temperature, and obtain mask deformation data.

[0111] Optionally, in an embodiment of the present application, the artificial intelligence optimization device 200 for the screen evaporation process also includes a multi-objective optimization module for obtaining optimization targets, the optimization targets including at least one of thickness uniformity, material utilization or yield rate; establishing an objective function for each optimization target, the objective function being an evaluation criterion for calculating fitness; calculating the fitness of the optimization parameters under each objective function; the fitness is used to characterize the performance quantification value of the optimization parameters; according to the fitness, a multi-objective optimization method based on a genetic algorithm is used to determine a combination of process parameters; the combination of process parameters is used to achieve a multi-objective balance.

[0112] It should be understood that the device corresponds to the above-mentioned embodiment of the artificial intelligence optimization method for the screen evaporation process, and can perform the various steps involved in the above-mentioned method embodiment. The specific functions of the device can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here. The device includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the operating system (OS) of the device.

[0113] See also Figure 4 The electronic device 300 provided in the embodiment of the present application includes: a processor 310 and a memory 320, wherein the memory 320 stores machine-readable instructions executable by the processor 310, and when the machine-readable instructions are executed by the processor 310, the above method is executed.

[0114] Figure 4 Each component shown in can be implemented by hardware, software or a combination thereof. The electronic device 300 may be a physical device, such as a server, a PC, etc., or a virtual device, such as a virtual machine, a virtualized container, etc. Moreover, the electronic device 300 is not limited to a single device, but may also be a combination of multiple devices or a cluster consisting of a large number of devices.

[0115] An embodiment of the present application further provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above method is executed.

[0116] Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0117] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0118] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0119] The above description is only an optional implementation manner of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the embodiments of the present application, which should be covered within the protection scope of the embodiments of the present application.

Claims

1. An artificial intelligence optimization method for screen evaporation process, characterized in that: include: Obtain multi-dimensional data of the screen during the evaporation process; Inputting the multi-dimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features; The evaporation state characteristics are used to characterize the error distribution and / or state information existing in the evaporation process; The evaporation state characteristics are input into the reinforcement learning module in the evaporation error prediction model to obtain the optimization parameters of the screen during the evaporation process; the optimization parameters are used to adjust the equipment parameters of the evaporation process equipment to optimize the evaporation process of the screen.

2. The method according to claim 1, characterized in that Before inputting the multi-dimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features, the method further includes: Acquire historical multidimensional training data, wherein the historical multidimensional training data includes at least one of temperature distribution data, film thickness distribution data, or mask deformation data; Inputting the historical multi-dimensional training data into a pre-set convolutional neural network to obtain a prediction output; The difference between the predicted output and the true value corresponding to the historical multidimensional training data is calculated, and based on the difference, the parameters of the convolutional neural network are updated using a back propagation algorithm to obtain the trained feature extraction module.

3. The method according to claim 1, characterized in that After inputting the multi-dimensional data into a feature extraction module in a pre-trained evaporation error prediction model to obtain evaporation state features, the method further includes: During the evaporation process of the screen, the multi-dimensional data is collected in real time using a network sensor; Performing error calculation using the multi-dimensional data and the evaporation state characteristics to obtain a feedback signal; The parameter gradient of the feature extraction module is calculated based on the feedback signal, and the weight of the feature extraction module is adjusted in real time to obtain an adjusted feature extraction module, so that the subsequent multidimensional data uses the adjusted feature extraction module to obtain corresponding evaporation state features.

4. The method according to claim 3, characterized in that The step of calculating the parameter gradient of the feature extraction module based on the feedback signal and adjusting the weight of the feature extraction module in real time to obtain the adjusted feature extraction module comprises: Utilizing the feedback signal, calculating the parameter gradient of the feature extraction module through a back propagation algorithm; An optimization algorithm is used to update the weights and biases of the feature extraction module according to the parameter gradient to obtain an adjusted feature extraction module.

5. The method according to claim 1, characterized in that Before inputting the evaporation state characteristics into the reinforcement learning module in the evaporation error prediction model to obtain the optimized parameters of the screen during the evaporation process, the method further includes: Acquire historical multi-dimensional training data and a predefined action space and strategy network; the action space is used to adjust the equipment parameters of the evaporation process equipment; the evaporation equipment parameters include at least one of evaporation source power, heating current, vacuum degree, mask position or mask angle; According to the strategy network, a target action is determined from the action space, and is executed on the historical multi-dimensional training data to obtain an observation result; the observation result is used to represent the film quality corresponding to the historical multi-dimensional training data; Calculating reward feedback based on the observation result; the reward feedback is used to evaluate the effect of the action space; The strategy network is updated according to the reward feedback to obtain the reinforcement learning module.

6. The method according to claim 1, characterized in that The multidimensional data includes at least one of temperature distribution data, film thickness distribution data or mask deformation data; the multidimensional data of the screen during the evaporation process is obtained, including: During the evaporation process, a thin film thickness detector is used to scan and sample the mask of the screen to generate the film thickness distribution data; and / or, using a thermocouple array and an infrared thermal imager to monitor the temperature distribution of a preset part in real time and generate a three-dimensional temperature distribution map; And / or, a laser displacement sensor and a strain sensor are used to perform scanning to monitor the thermal expansion state of the mask at high temperature and obtain the deformation data of the mask.

7. The method according to any one of claims 1 to 6, characterized in that: After obtaining the optimized parameters of the screen during the evaporation process, the method further includes: Acquiring an optimization target, wherein the optimization target includes at least one of thickness uniformity, material utilization rate, or yield rate; Establishing an objective function for each of the optimization objectives, wherein the objective function is an evaluation criterion for calculating fitness; Calculating the fitness of the optimization parameter under each of the objective functions; the fitness is used to characterize the performance quantification value of the optimization parameter; According to the fitness, a multi-objective optimization method based on a genetic algorithm is used to determine a combination of process parameters; the combination of process parameters is used to achieve a multi-objective balance.

8. A computer program product, characterized in that The method comprises computer program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is executed.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is executed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is executed.

Images