Slit coating production process parameter prediction method

Abstract

The application discloses a kind of slit coating production process parameter prediction methods, first by experimental method, gradually find the minimum thickness of critical film and the maximum moving speed of substrate, and collect various parameters in slit coating process;Then the slit spraying speed of slit coating, the height of coating head distance substrate, substrate moving speed, film thickness are obtained, and data are normalized, to obtain model input data;Established with slit spraying speed, the height of coating head distance substrate as input, the maximum substrate moving speed, the minimum film thickness as output limit learning machine slit coating prediction model;Finally, the improved bionic optimization algorithm optimization limit learning machine slit coating prediction is used.The application solves the problem that only slit coating parameters can be predicted by production experience in the prior art, and if the process parameters are not set reasonably, it is very likely to cause film defects.

Description

Technical Field

[0001] This invention belongs to the field of slot coating technology, specifically relating to a method for predicting parameters in the slot coating production process. Background Technology

[0002] The semiconductor display industry is developing rapidly, and screen display technologies, led by OLED, are also advancing at a high speed. The demand for matching liquid crystal display films is also increasing daily. Currently, there are many methods for preparing thin films, including dip coating, air knife coating, blade coating, roller coating, and slot coating. However, in actual preparation, slot coating has become the most widely used method due to its ease of control and high precision.

[0003] Slit coating involves applying pressure above a coating head to force a coating solution through the head and evenly coat the substrate surface, forming a thin film in conjunction with a specific substrate movement speed. In actual production, due to the non-linear characteristics of the coating solution's flow, operators often have to rely on experience to predict the minimum film thickness and maximum substrate movement speed for different process parameters. Inappropriate process parameter settings can easily lead to film defects. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting parameters in the slit coating production process, which solves the problem that in the prior art, slit coating parameters can only be predicted based on production experience. If the process parameters are not set reasonably, defects in the film are very likely to occur.

[0005] The technical solution adopted in this invention is a method for predicting parameters in the slot coating production process, which is implemented according to the following steps:

[0006] Step 1: Through experimental methods, gradually find the minimum thickness of the critical thin film and the maximum moving speed of the substrate, and collect various parameters during the slit coating process;

[0007] Step 2: Obtain the slit spraying speed, nozzle height from substrate, substrate moving speed, and thin film thickness for slit coating, and normalize the data to obtain the model input data;

[0008] Step 3: Establish an extreme learning machine slit coating prediction model with slit spraying speed and nozzle height from substrate as inputs, and maximum substrate moving speed and minimum thin film thickness as outputs.

[0009] Step 4: Optimize the Extreme Learning Machine slit coating prediction model using the improved beetle whisker search optimization algorithm.

[0010] The invention is further characterized in that,

[0011] In step 1, the substrate moving speed, the height of the coating head from the substrate, and the coating head spraying speed are set on the coating experimental platform. By adjusting the changes of these three quantities, the minimum film thickness is obtained, and the substrate moving speed at this time is the maximum substrate moving speed.

[0012] Step 2 is implemented in the following steps:

[0013] Step 2.1: Obtain the data of slit spraying speed and the height of the coating head from the substrate as input to the extreme learning machine slit coating prediction model in Step 3, and the minimum thin film thickness and the maximum substrate moving speed as output to the extreme learning machine slit coating prediction model in Step 3.

[0014] Step 2.2: After data standardization, the raw data is of the same order of magnitude, making it suitable for comprehensive comparison and evaluation. Therefore, the input data is normalized.

[0015] The normalized expression is:

[0016] (1)

[0017] In the formula, It is the maximum value after normalization. It is the normalized minimum value. It is to normalize the maximum value of the array. It is the minimum value of the array to be normalized. for The normalized value.

[0018] Step 3 is implemented in the following steps:

[0019] Step 3.1: Given a training set in Indicates the first Data , Indicates the first Output corresponding to each data point The set refers to all training data. The mathematical expression for the output function of the Extreme Learning Machine is:

[0020] (2)

[0021] in, The weights from the hidden layer to the output layer. The activation function of the hidden layer. The first step from the input layer to the hidden layer The weights of each neuron It is the first input sample One dimension, It is the first hidden layer One bias vector;

[0022] Step 3.2: After simplifying equation (2), we get ,

[0023] in

[0024] (3)

[0025] (4)

[0026] In the formula It is the output matrix of the hidden layer. It is the target output matrix of the training set. It is a weight matrix;

[0027] Step 3.3, in the Extreme Learning Machine We obtain through equation (5)

[0028] (5)

[0029] It is the output matrix of the hidden layer. The generalized inverse matrix is ​​obtained, thus yielding the weights from the hidden layer to the output layer. That is, when the entire training process of the Extreme Learning Machine is completed, Back to Right now

[0030] (6)

[0031] This is the predicted output.

[0032] Step 4 is implemented in the following steps:

[0033] Step 4.1: Optimize the Extreme Learning Machine (ELM) slit coating prediction model by incorporating a beetle whisker search algorithm to optimize the input weights and thresholds. Initialize the current center position of the beetle as... , The direction of the longhorn beetle search is shown in equation (7).

[0034] (7)

[0035] in For directional random functions, For positional dimensions,

[0036] Determine the position of the longhorn beetle's left and right whiskers based on its direction. The left whisker position, For the right-hand position, the mathematical model is shown in equation (8):

[0037] (8)

[0038] Step 4.2: Determine the fitness of the left and right whiskers at their respective positions based on their positions. The fitness calculation is as follows:

[0039] (9)

[0040] in, For the fitness function, The fitness value of the left whisker. The fitness value of the right whisker;

[0041] Step 4.3: The longhorn beetle moves to the next position:

[0042] (10)

[0043] in For symbolic functions, The distance between the two whiskers, The step size function for the search is given by the expression for the beetle position update iteration, as shown in equation (11):

[0044] (11)

[0045] Then determine whether the iteration stopping condition is met. If it is met, stop the iteration and output the optimal solution; otherwise, continue the iteration.

[0046] Step 4.4: The mathematical model for the adaptive strategy of the longhorn beetle is shown in equation (12).

[0047] (12)

[0048] In the formula, Indicates the first During the nth iteration The distance to the longhorn beetle's whiskers, Representing the To ensure the characteristic of varying antennae length, the distance base of the longhorn beetle's antennae can be determined by... In the coefficient array, half of the bases are set to numbers greater than 1, and the other half are set to numbers less than or equal to 1. Representing the Distance coefficient for the longhorn beetle's whiskers.

[0049] Step 4.5: Once the optimization algorithm for the cow whisker adaptive strategy satisfies the iteration stopping condition, output the optimized extreme learning machine parameters and the current center position. These are the output parameters of the optimized extreme learning machine.

[0050] The beneficial effects of this invention are as follows: The method for predicting parameters in the slot coating production process is modeled experimentally. First, various coating parameters are collected. Experiments reveal that the slot spraying speed and the distance between the coating head and the substrate are control variables highly correlated with the minimum film thickness and the maximum substrate moving speed. Therefore, these two sets of variables are selected. Next, the input variables of the model are normalized, and the processed data is input into an Extreme Learning Machine (ELM). Some hyperparameters in the ELM are optimized using an improved beetle whisker search algorithm. The optimized hyperparameters obtained from the improved beetle whisker search algorithm are then fed into the ELM to obtain an ELM slot coating model optimized by the improved beetle whisker search algorithm. Finally, this model is used to predict the minimum film thickness and the maximum substrate moving speed in slot coating. The predicted results are close to the actual situation, making the adjustment of process parameters during slot coating a reliable guideline. Attached Figure Description

[0051] Figure 1 This is a diagram of the slit coating prediction framework structure;

[0052] Figure 2 This is a two-dimensional schematic diagram of slot coating;

[0053] Figure 3 This is a comparison chart of the maximum substrate movement speed predicted by the Extreme Learning Machine and the Improved Longhorn Beetle Optimized Extreme Learning Machine under different conditions.

[0054] Figure 4 This is a comparison chart of the minimum film thickness predicted by the Extreme Learning Machine and the improved beetle-optimized Extreme Learning Machine under different conditions.

[0055] Figure 5 This is the flowchart of the Extreme Learning Machine operation that needs to be optimized to improve the longhorn beetle;

[0056] Figure 6 This is a flowchart of the operation of the improved longhorn beetle optimization algorithm. Detailed Implementation

[0057] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0058] The method for predicting parameters in the slot coating production process is implemented according to the following steps:

[0059] Step 1: Through experimental methods, gradually find the minimum thickness of the critical thin film and the maximum moving speed of the substrate, and collect various parameters during the slit coating process;

[0060] In step 1, the substrate moving speed, the height of the coating head from the substrate, and the coating head spraying speed are set on the coating experimental platform. By adjusting the changes of these three quantities, the minimum film thickness is obtained, and the substrate moving speed at this time is the maximum substrate moving speed.

[0061] Step 2: Obtain the slit spraying speed, nozzle height from substrate, substrate moving speed, and thin film thickness for slit coating, and normalize the data to obtain the model input data;

[0062] Combination Figure 2 Step 2 is implemented in the following steps:

[0063] Step 2.1: During the slit coating process, the slit spraying speed and the distance between the coating head and the substrate are control variables that are highly correlated with the minimum film thickness and the maximum substrate moving speed. Therefore, the data of the slit spraying speed and the distance between the coating head and the substrate are obtained as inputs to the extreme learning machine slit coating prediction model in Step 3, and the minimum film thickness and the maximum substrate moving speed are used as outputs of the extreme learning machine slit coating prediction model in Step 3.

[0064] Step 2.2: To eliminate the influence of different dimensions between indicators, data standardization is required to ensure comparability. After data standardization, the raw data is of the same order of magnitude, making it suitable for comprehensive comparison and evaluation. Therefore, the input data is normalized as follows:

[0065] The normalized expression is:

[0066] (1)

[0067] In the formula, It is the maximum value after normalization. It is the normalized minimum value. It is to normalize the maximum value of the array. It is the minimum value of the array to be normalized. for The normalized value.

[0068] Step 3: Establish an extreme learning machine slit coating prediction model with slit spraying speed and nozzle height from substrate as inputs, and maximum substrate moving speed and minimum thin film thickness as outputs.

[0069] Step 3 is implemented in the following steps:

[0070] Extreme Learning Machine (ELM) is a machine learning method based on a feedforward neural network. It is a shallow neural network with only a single hidden layer, and requires relatively few parameters to be adjusted. The weights from the input layer to the hidden layer, the biases of the hidden layer, and the number of neurons in the hidden layer are all randomly generated. Its structure is as follows: Figure 1 As shown.

[0071] Step 3.1: Given a training set in Indicates the first Data , Indicates the first Output corresponding to each data point The set refers to all training data. The mathematical expression for the output function of the Extreme Learning Machine is:

[0072] (2)

[0073] in, The weights from the hidden layer to the output layer. The activation function of the hidden layer. The first step from the input layer to the hidden layer The weights of each neuron It is the first input sample One dimension, It is the first hidden layer One bias vector;

[0074] Step 3.2: After simplifying equation (2), we get ,

[0075] in

[0076] (3)

[0077] (4)

[0078] In the formula It is the output matrix of the hidden layer. It is the target output matrix of the training set. It is a weight matrix;

[0079] Step 3.3, in the Extreme Learning Machine We obtain through equation (5)

[0080] (5)

[0081] It is the output matrix of the hidden layer. The generalized inverse matrix is ​​obtained, thus yielding the weights from the hidden layer to the output layer. That is, when the entire training process of the Extreme Learning Machine is completed, Back to Right now

[0082] (6)

[0083] This is the predicted output.

[0084] Step 4: Optimize the Extreme Learning Machine slit coating prediction model using the improved beetle whisker search optimization algorithm to obtain the optimized model.

[0085] Combination Figure 5 , Figure 6 Step 4 is implemented in the following steps:

[0086] In the Extreme Learning Machine (ELM) slit coating prediction model, the input weights, threshold, and number of hidden layer neurons affect the accuracy of the prediction results. Therefore, the ELM model was optimized by incorporating a beetle whisker search algorithm to optimize the input weights and threshold. The beetle whisker search algorithm is a relatively efficient intelligent natural algorithm that can achieve efficient optimization without knowing the specific form of the function. Compared to general population algorithms, the beetle whisker search algorithm only requires one individual, significantly reducing the computational load. The position of the beetle in the beetle whisker search algorithm represents a feasible solution. The fitness strength of the beetle's left and right whiskers determines the direction of the next optimization step. The step size and optimization direction are then iteratively updated until the optimal solution is found, at which point the iteration stops. Furthermore, to further improve the search speed and accuracy of the beetle whisker search algorithm, improvements were made.

[0087] Step 4.1: Optimize the Extreme Learning Machine (ELM) slit coating prediction model by incorporating a beetle whisker search algorithm to optimize the input weights and thresholds. Initialize the current center position of the beetle as... , The direction of the longhorn beetle search is shown in equation (7).

[0088] (7)

[0089] in For directional random functions, For positional dimensions,

[0090] Determine the position of the longhorn beetle's left and right whiskers based on its direction. The left whisker position, For the right-hand position, the mathematical model is shown in equation (8):

[0091] (8)

[0092] Step 4.2: Determine the fitness of the left and right whiskers at their respective positions based on their positions. The fitness calculation is as follows:

[0093] (9)

[0094] in, For the fitness function, The fitness value of the left whisker. The fitness value of the right whisker;

[0095] Step 4.3: The longhorn beetle moves to the next position:

[0096] (10)

[0097] in For symbolic functions, The distance between the two whiskers, The step size function for the search is given by the expression for the beetle position update iteration, as shown in equation (11):

[0098] (11)

[0099] Then determine whether the iteration stopping condition is met. If it is met, stop the iteration and output the optimal solution; otherwise, continue the iteration.

[0100] Step 4.4: The selection of the beetle whisker distance is particularly important in the beetle whisker search algorithm. The original strategy for updating the whisker distance was to gradually decrease it with each iteration. Once the number of iterations and the initial step size are fixed, the whisker distance remains constant in each subsequent iteration. If the whisker distance is too large during the optimization process, the convergence speed slows down, but the global search capability becomes stronger. Conversely, if the whisker distance is too small, the convergence speed speeds up, but the global search capability weakens, making it prone to getting trapped in local optima. Furthermore, since the beetle in the beetle whisker search algorithm relies on only two whiskers to sense its environment, this also leads to a slow convergence speed, preventing the finding of the optimal solution.

[0101] To improve the convergence speed of the algorithm and enhance the global search capability, the number of beetle whiskers is increased, the direction of each pair of whiskers is randomly generated, and a multi-whisker adaptive search strategy is proposed based on the changes in fitness.

[0102] The adaptive search strategy of the longhorn beetle is described in detail below. The mathematical model of the adaptive search strategy of the longhorn beetle is shown in Equation (12).

[0103] (12)

[0104] In the formula, Indicates the first During the nth iteration Regarding the distance between the longhorn beetle's whiskers, multiple sets of whiskers can ensure diversity in the longhorn beetle's search for the optimal solution. Representing the To ensure the characteristic of varying antennae length, the distance base of the longhorn beetle's antennae can be determined by... Half of the coefficient array is set to have a base greater than 1, and the other half is set to have a base less than or equal to 1. This ensures that the longhorn beetle can perform both fine-grained searches within a small range and searches within a larger range, guaranteeing diversity of solutions and avoiding getting trapped in local optima. Representing the Distance coefficient for the longhorn beetle's whiskers.

[0105] Step 4.5: Once the optimization algorithm for the cow whisker adaptive strategy satisfies the iteration stopping condition, output the optimized extreme learning machine parameters and the current center position. These are the output parameters of the optimized extreme learning machine.

[0106] To clearly demonstrate the effectiveness of each modeling approach, the following comparison results of evaluation metrics are presented. This invention defines the evaluation metrics shown in Table 1: mean squared error, mean absolute error, and mean absolute percentage error to evaluate model performance.

[0107] Table 1 Evaluation Indicators

[0108]

[0109] Specific errors are shown in Table 2:

[0110] Table 2 Mean Square Error, Mean Absolute Error, Mean Absolute Percentage Error

[0111]

[0112] The table clearly shows that the improved beetle-optimized Extreme Learning Machine (ELM) has the smallest error across all prediction performance metrics. The results indicate that the improved beetle-optimized ELM performs better and significantly reduces error.

[0113] Figure 3 The figures show the experimental true value of the maximum substrate moving speed, the prediction of the improved Extreme Learning Machine (ELM) model, and the prediction of the ELM model. It can be seen from the figures that the prediction of the maximum substrate moving speed by the improved ELM model is significantly better than that by the ELM model, and is closer to the true value.

[0114] Figure 4 The figures show the experimental true value of the minimum film thickness, the prediction of the improved extreme learning machine model, and the prediction of the extreme learning machine model. It can be seen from the figures that the prediction of the minimum film thickness by the improved extreme learning machine model is significantly better than that by the extreme learning machine model, and is closer to the true value.

[0115] This invention provides a method for predicting process parameters in slot coating production. Since multiple process variables are involved in the slot coating process for thin film preparation, these variables are collected and analyzed experimentally. Among these, the slot spraying speed and the distance between the coating head and the substrate are control variables highly correlated with the minimum film thickness and the maximum substrate moving speed. Changes in these two variables directly alter the movement of the coating liquid, thus limiting the final minimum film thickness and the maximum substrate moving speed, ultimately determining film quality. Therefore, a dual-input, dual-output model is established, where the inputs are the slot spraying speed and the distance between the coating head and the substrate, and the outputs are the minimum film thickness and the maximum substrate moving speed. First, the data is normalized to obtain the model's input variables. Simultaneously, to better model the slot coating process, this invention introduces an extreme learning machine (ELM) model. Based on this, the invention improves the beetle whisker search algorithm and uses it to optimize the ELM model, proposing a method for predicting slot coating process parameters. This method can accurately predict the maximum substrate moving speed and the minimum film thickness during slot coating, thereby adjusting process parameters and providing technical support for high-precision slot coating processes.

Claims

1. A method for predicting parameters in the slot coating production process, characterized in that, The specific steps are as follows: Step 1: Through experimental methods, gradually find the minimum thickness of the critical thin film and the maximum moving speed of the substrate, and collect various parameters during the slit coating process; Step 2: Obtain the slit spraying speed, nozzle height from substrate, substrate moving speed, and thin film thickness for slit coating, and normalize the data to obtain the model input data; Step 3: Establish an extreme learning machine slit coating prediction model with slit spraying speed and nozzle height from substrate as inputs, and maximum substrate moving speed and minimum thin film thickness as outputs. Step 4: Optimize the extreme learning machine slit coating prediction model using an improved beetle whisker search optimization algorithm. Step 4 is specifically implemented according to the following steps: Step 4.1: Optimize the Extreme Learning Machine (ELM) slit coating prediction model by incorporating a beetle whisker search algorithm to optimize the input weights and thresholds. Initialize the current center position of the beetle as... , The direction of the longhorn beetle search is shown in equation (7). （7） in For directional random functions, For positional dimensions, Determine the position of the longhorn beetle's left and right whiskers based on its direction. The left whisker position, For the right-hand position, the mathematical model is shown in equation (8): （8） The distance between the two whiskers; Step 4.2: Determine the fitness of the left and right whiskers at their respective positions based on their positions. The fitness calculation is as follows: （9） in, For the fitness function, The fitness value of the left whisker. The fitness value of the right whisker; Step 4.3: The longhorn beetle moves to the next position: （10） in For symbolic functions, The distance between the two whiskers, The step size function for the search is given by the expression for the beetle position update iteration, as shown in equation (11): （11） Then determine whether the iteration stopping condition is met. If it is met, stop the iteration and output the optimal solution; otherwise, continue the iteration. Step 4.4: The mathematical model for the adaptive strategy of the longhorn beetle is shown in equation (12). （12） In the formula, Indicates the first During the nth iteration The distance to the longhorn beetle's whiskers, Representing the To ensure the characteristic of varying antennae length, the distance base of the longhorn beetle's antennae can be determined by... In the coefficient array, half of the bases are set to numbers greater than 1, and the other half are set to numbers less than or equal to 1. Representing the Distance coefficient to the longhorn beetle's whiskers; Step 4.5: Once the optimization algorithm for the cow whisker adaptive strategy satisfies the iteration stopping condition, output the optimized extreme learning machine parameters and the current center position. These are the output parameters of the optimized extreme learning machine.

2. The method for predicting parameters in the slot coating production process according to claim 1, characterized in that, In step 1, the substrate moving speed, the height of the coating head from the substrate, and the coating head spraying speed are set on the coating experimental platform. The minimum film thickness is obtained by adjusting the changes of these three quantities, and the substrate moving speed at this time is the maximum substrate moving speed.

3. The method for predicting parameters in the slot coating production process according to claim 2, characterized in that, Step 2 is implemented in the following steps: Step 2.1: Obtain the data of slit spraying speed and the height of the coating head from the substrate as input to the extreme learning machine slit coating prediction model in Step 3, and the minimum thin film thickness and the maximum substrate moving speed as output to the extreme learning machine slit coating prediction model in Step 3. Step 2.2: After data standardization, the raw data is of the same order of magnitude, making it suitable for comprehensive comparison and evaluation. Therefore, the input data is normalized. The normalized expression is: (1) In the formula, It is the maximum value after normalization. It is the normalized minimum value. It is to normalize the maximum value of the array. It is the minimum value of the array to be normalized. for The normalized value.

4. The method for predicting parameters in the slot coating production process according to claim 3, characterized in that, Step 3 is implemented in the following steps: Step 3.1: Given a training set in Indicates the first Data , Indicates the first Output corresponding to each data point The set refers to all training data. The mathematical expression for the output function of the Extreme Learning Machine is: （2） in, The weights from the hidden layer to the output layer. The activation function of the hidden layer. The first step from the input layer to the hidden layer The weights of each neuron It is the first input sample One dimension, It is the first hidden layer One bias vector; Step 3.2: After simplifying equation (2), we get , in （3） （4） In the formula It is the output matrix of the hidden layer. It is the target output matrix of the training set. It is a weight matrix; Step 3.3, in the Extreme Learning Machine We obtain through equation (5) （5） It is the output matrix of the hidden layer. The generalized inverse matrix is ​​obtained, thus yielding the weights from the hidden layer to the output layer. That is, when the entire training process of the Extreme Learning Machine is completed, Back to Right now （6） This is the predicted output.

Images