Dry film quality classification model training method, production process parameter setting method and device and medium

Abstract

The invention relates to the field of battery cell manufacturing, in particular to a training method of a dry film quality classification model, a production process parameter setting method, a device and a medium. The training method comprises the following steps: for a production process of a dry film, obtaining production process parameters of each process in the production process of the dry film and a quality index category of the produced dry film; dimension reduction processing is carried out on the production process parameters to screen key process parameters; respectively processing according to the key process parameters under each process to obtain composite process characteristics of the corresponding process; and training a dry film quality classification model by taking the composite process characteristics as input and taking the quality index categories as output. According to the method, the production process parameters are screened and processed to obtain the composite process parameters with higher relevance with the quality indexes, redundant information is removed, each composite process parameter obtained by mining corresponds to each working procedure, and the training efficiency and accuracy of the dry film quality classification model can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of battery cell manufacturing, and particularly to a training method, a production process parameter setting method, a device and a medium for a dry film quality classification model. Background Art

[0002] In the field of battery cell manufacturing, the processes of slurry mixing, wet film and dry film involve various process parameters. The slurry mixing process involves mixing active materials, binders and solvents to form a slurry, and this step has strict requirements on the uniformity and stability of the slurry. The wet film process is to uniformly coat the slurry on the electrode sheet and ensure uniform drying of the coating through control during the drying process. The dry film process is a key step in battery manufacturing, and it is necessary to ensure the final performance of the electrode film through peeling and compaction. Any small deviation in these processes may lead to a significant decline in battery performance.

[0003] The traditional method mainly adjusts the production process parameters in the slurry mixing, wet film and dry film processes manually according to the quality indicators of the products. This method lacks predictability and pertinence, resulting in a large number of unqualified products. There is a technical solution in the prior art that combines the setting of production process parameters with a deep learning model to predict whether the dry film quality is qualified. For example, Chinese Patent Grant Publication No. CN112950071B discloses a training method, a process parameter adjustment method and a device for a process parameter adjustment model. The technical solution records: obtaining training sample data; wherein, the training sample data includes multiple process parameters of the processed sample products and the defect information of the sample products; inputting the training sample data into an initial network model to obtain the predicted adjustment values of multiple process parameters of the sample products based on the defect information; calculating a loss value based on the predicted adjustment values and the true adjustment values obtained in advance, and updating the initial network model by minimizing the loss value to generate a process parameter adjustment model. This training method directly uses the original parameters of the production process as the input features of the model for modeling. For the multi-process dry film production process, this method lacks in-depth excavation of the key features of each process, has a low prediction accuracy, and is difficult to meet the actual requirements. Summary of the Invention

[0004] The purpose of the present invention is to provide a training method, a production process parameter setting method, a device and a medium for a dry film quality classification model. The training method screens out key process parameters based on the original production process parameters, and processes the key process parameters to obtain composite process features under each process to enhance the correlation with the quality indicators, so as to effectively improve the prediction accuracy and adaptively adjust the production parameters.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions to solve: In a first aspect, the present invention provides a method for training a dry film quality classification model, which includes: Obtain the production process parameters of each process in the production of dry film, as well as the quality index categories of the produced dry film; Perform dimensionality reduction processing on the production process parameters to screen out key process parameters; Process the key process parameters under each process to obtain composite process features corresponding to the corresponding process; Based on the composite process features and the corresponding quality index categories, obtain training samples, and use the training samples to train a dry film quality classification model with the composite process features as the input and the quality index categories as the output.

[0006] The present invention performs dimensionality reduction screening on the production process parameters to obtain key process parameters, and further processes the key process parameters to obtain composite process parameters with stronger correlation with quality indicators, so as to be able to remove redundant information, reduce computational complexity and be able to deeply mine key features. In addition, each mined composite process parameter corresponds to each process, which can effectively improve the training efficiency and prediction accuracy of the classification model.

[0007] Optionally, the performing dimensionality reduction processing on the production process parameters includes: For each type of production process parameter, calculate the variance of the collected data; Take the production process parameters with variances higher than the preset threshold as key process parameters.

[0008] The present invention performs dimensionality reduction processing through the low variance filtering method, which is very suitable for processing scalar data such as production process parameters, can effectively reduce irrelevant features, and can retain key process parameters that have a significant impact on quality indicators by screening high variance parameters, thus providing a basis for the construction of subsequent composite process features.

[0009] Optionally, the quality index categories are divided according to a pre-calibrated qualified range: for quality indicators within the qualified range, assign a value of 1; for quality indicators outside the qualified range, assign a value of 0.

[0010] The present invention performs binary (0 / 1) processing on quality indicators through a preset qualified range, provides clear quality index category labels as the actual labels for the training of the dry film quality classification model, thereby improving the reliability of the prediction results of the dry film quality classification model.

[0011] Optionally, the key process parameters include: the slurry viscosity, slurry elasticity, slurry fineness, and slurry surface tension in the slurry mixing process; the coating machine gasket thickness, coating machine gasket flow channel width, coating machine gasket opening angle, and coating oven temperature in the wet film process; the die head lip flatness, die head gap, and back roller circular runout amplitude in the dry film process; and the composite process characteristics include the slurry uniformity index in the slurry mixing process, the coating stability factor in the wet film process, and the die head dynamic balance coefficient in the dry film process.

[0012] The present invention defines the key process parameters for the slurry mixing process, wet film process, and dry film process, and determines the composite process characteristics (the slurry uniformity index in the slurry mixing process, the coating stability factor in the wet film process, and the die head dynamic balance coefficient in the dry film process) based on the key process parameters, enhancing the physical association between the characteristics and quality indicators to improve the training efficiency of the dry film quality classification model.

[0013] Optionally, the calculation formula for the slurry uniformity index is: ; In the formula, is the slurry uniformity index, is the slurry viscosity; is the slurry surface tension; is the slurry elasticity; is the slurry fineness; and are the weight coefficients; The calculation formula for the coating stability factor is: ; In the formula, is the coating stability factor, is the coating machine gasket thickness; is the coating machine gasket flow channel width; is the coating machine gasket opening angle; is the actual oven temperature; is the optimal oven temperature; The calculation formula for the die head dynamic balance coefficient is: ; In the formula, is the die head dynamic balance coefficient, is the die head lip flatness; is the die head gap; is the back roller circular runout amplitude; , , are the standard deviations of the die head lip flatness, die head gap, and back roller circular runout amplitude during continuous production, respectively.

[0014] By defining the calculation formulas for composite process characteristics (slurry uniformity index in the slurry mixing process, coating stability factor in the wet film process, and die head dynamic balance coefficient in the dry film process), the values of the composite process characteristics calculated can accurately represent the process conditions of dry film production, improving the learning effect of the dry film quality classification model.

[0015] Optionally, the dry film quality classification model includes a random forest regression model and a support vector machine classification model connected in series. Training the dry film quality classification model includes: Inputting the composite process characteristics into the random forest regression model to generate a predicted trend value for the quality index category; Combining the predicted trend value of the quality index category and the actual value of the quality index category to obtain a feature vector; Inputting the feature vector into the support vector machine classification model to predict the quality index category; Among them, the random forest regression model is trained using a first loss function to determine the optimal number of decision trees through cross-validation; the feature vector is mapped to a high-dimensional space using a mixed kernel function composed of a Gaussian kernel and a polynomial kernel, and the support vector machine classification model is trained using a second loss function to solve for the optimal decision boundary in the high-dimensional space.

[0016] The present invention captures non-linear relationships through multiple decision trees (usually 100 - 200 trees) of the random forest regression model, outputs the predicted trend value of the quality index to generate a high-order feature vector; subsequently, the support vector machine classification model maps the feature vector to a high-dimensional space using a mixed kernel function to solve for the optimal decision boundary. The sequentially connected random forest regression model and support vector machine classification model can significantly improve the classification accuracy of the dry film quality classification model.

[0017] Optionally, it further includes: Using the trained dry film quality classification model to predict the quality index category during the production process of multiple batches of dry films with known actual values of the quality index category; Determining the prediction error based on the prediction result and the actual value of the quality index category; Generating a calibration curve based on the distribution of prediction errors for multiple batches; the calibration curve is used to map the original prediction value of the dry film quality classification model to a calibrated value as the final prediction result of the dry film quality classification model.

[0018] By correcting the prediction result through the calibration curve generated based on the distribution of prediction errors for multiple batches, the present invention can effectively improve the classification confidence.

[0019] In a second aspect, the present invention provides a training device for a dry film quality classification model, which includes: An acquisition module is used to obtain the production process parameters of each step in the production process of the dry film, as well as the quality index category of the produced dry film; A dimensionality reduction module is used to perform dimensionality reduction processing on the production process parameters to screen out key process parameters; The processing module is used to process the key process parameters of each process to obtain the composite process characteristics of the corresponding process; A training model is used to obtain training samples based on the composite process characteristics and the corresponding quality indicator categories, and use the training samples to train a dry film quality classification model that takes the composite process characteristics as input and the quality indicator categories as output.

[0020] In a third aspect, the present invention provides a method for setting production process parameters for dry film production, comprising: Initial setting of production process parameters for dry film production; Determine composite process characteristics according to the set production process parameters; Inputting the composite process characteristics into a dry film quality classification model to obtain a quality index category of the dry film produced under the set production process parameters; the dry film quality classification model is obtained based on the training method of the dry film quality classification model; If the quality index category corresponds to unqualified quality, the set production process parameters are iteratively adjusted until the quality index category of the dry film corresponds to qualified quality.

[0021] The present invention can adaptively and dynamically adjust the set production process parameters based on the prediction results of the dry film quality classification model, thereby effectively avoiding the production of defective products and improving overall production efficiency and yield rate.

[0022] In a fourth aspect, the present invention provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the training method of the dry film quality classification model, or, when executed by a processor, implement the production process parameter setting method.

[0023] Compared with the prior art, the present invention has the following beneficial effects: This method uses dimensionality reduction and screening of raw production process parameters to obtain key process parameters. These parameters are then processed to derive composite process characteristics corresponding to each process step. This effectively removes redundant information and reduces computational complexity. Furthermore, it can deeply explore the key characteristics of each process step, enhancing the correlation between the input features of the dry film quality classification model and the quality indicator categories, thereby improving the training efficiency and prediction accuracy of the dry film quality classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1Schematic flow chart of the training method in Embodiment 1; Figure 2 Schematic diagram of the training of the support vector machine model in Embodiment 1; Figure 3 Schematic flow chart of the production process parameter setting method in Embodiment 2. Detailed implementation manners

[0025] The technical solution of the present invention will be described in detail below with reference to the drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0026] The term "and / or" only describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " generally represents an "or" relationship between the associated objects before and after.

[0027] In the prior art known to the inventors of the present invention, directly modeling only relying on the original parameters as input features not only makes it difficult for the model to effectively learn the correlation between the quality index and the production parameter index, but also results in a large amount of calculation due to a large amount of data. In addition, due to the complex non-linear relationship between the production process parameters and the indicators, it is difficult for the prior art to directly combine the original parameters with a conventional classification model (such as the SVM support vector machine classification model) to capture the dynamic association, and the model obtained in this way lacks a real-time adaptive mechanism and has weak generalization ability for cross-batch data.

[0028] The present invention will be further described below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and cannot be used to limit the protection scope of the present invention. Embodiment 1

[0029] Combined with Figure 1 , this embodiment provides a training method for a dry film quality classification model, which includes: Step S102, obtaining the production process parameters of each process in the production process of the dry film, and the quality index category of the produced dry film; In some specific embodiments, the production process of the dry film includes a slurry mixing step, a wet film step, and a dry film step. The production process parameters in each step are collected in real time by pre-installed sensors during the production process. To ensure the accuracy and integrity of the data, it is necessary to record the production process parameters and the quality indicators of the dry film at the same time, and collect sufficient data samples during each batch of production. The quality indicators of the dry film are used to characterize the quality of the dry film to screen out unqualified defective dry film surface density. In a specific embodiment, the original production process parameters include: slurry viscosity, slurry elasticity, slurry fineness, slurry surface tension, buffer tank liquid level difference, and screw pump wear in the slurry mixing step; cavity shape / number, coater gasket thickness, coater gasket flow channel width, coater gasket opening angle, coating oven temperature, coating oven quick check in the wet film step; die lip levelness, die gap, back roller circular runout amplitude, back roller pressure, and back roller temperature in the dry film step.

[0030] In some specific embodiments, the quality indicators of the dry film include surface density (rarea), edge thickness (hedge), film width (Wfilm), and post-rolling peel strength (speel). Specifically, surface density (rarea), measured in g / m², is evaluated by mass distribution over the film area. Edge thickness (hedge), measured in μm, is the thickness of the dry film edge and directly affects the film's cutting and post-processing performance. Film width (Wfilm), measured in mm, affects the adaptability of the electrode sheet. Post-rolling peel strength (speel), measured in N / cm, is the adhesion strength between the compacted film and the electrode sheet and determines the dry film's adhesion. Each quality indicator category is pre-calibrated based on experimental data. Quality indicators within the qualified range are assigned a value of 1, while those outside the qualified range are assigned a value of 0. In a specific embodiment, the target values for surface density are 15 g / m², the target value for edge thickness is 100 μm, the target value for film width is 500 mm, and the target value for post-rolling peel strength is 2 N / cm.

[0031] Step S104: performing dimensionality reduction processing on the production process parameters to screen out key process parameters; Since dry film production involves numerous original production process parameters, in order to remove redundant information and deeply mine key features to enhance the correlation between input features and quality indicators, this embodiment performs dimensionality reduction on the original production process parameters, thereby improving the training efficiency of the model by constructing artificial features.

[0032] In some specific embodiments, the dimensionality reduction process is performed using a low variance filtering method. For each type of production process parameter, the variance of the collected data is calculated using the following formula:

[0033] Among them, is the i-th type of production process parameter, represents the variance of the i-th type of production process parameter, m is the number of samples collected for the i-th type of production process parameter, is the mean value of the i-th type of production parameter.

[0034] Compare the calculated variance with the pre-set variance threshold as follows:

[0035] Among them, is the set of key process parameters.

[0036] In some specific embodiments, the variance threshold is determined through pre-experiments. As an optional typical value, the variance threshold can be 3.5. The production process parameters with a variance less than this value are all screened out, and the production process parameters with a variance greater than the variance threshold are retained as key process parameters, thereby effectively reducing irrelevant features and improving the training efficiency of the subsequent model.

[0037] The finally screened key process parameters include: the slurry viscosity, slurry elasticity, slurry fineness, and slurry surface tension in the slurry mixing process, the coating machine gasket thickness, coating machine gasket flow channel width, coating machine gasket opening angle, and coating oven temperature in the wet film process, and the die head lip levelness, die head gap, and back roller circular runout amplitude in the dry film process.

[0038] Step S106: Process the corresponding composite process features for each process according to the key process parameters under each process; Since there is a complex non-linear relationship between production process parameters and index indicators, in order to enhance the correlation between input features and quality indicators to improve the model prediction accuracy, in this embodiment, the key process parameters screened by dimensionality reduction are further processed to obtain the composite process features of each process, so as to accurately reflect the correlation between each process and the quality indicators of the final produced dry film.

[0039] In some specific embodiments, the composite process features include the slurry uniformity index in the slurry mixing process, the coating stability factor in the wet film process, and the die head dynamic balance coefficient in the dry film process.

[0040] Illustratively, the calculation formula of the slurry uniformity index ( ) is: ; In the formula, is the slurry uniformity index, is the slurry viscosity, with the unit of pa·s; is the surface tension of the slurry, in mN / m; is the elasticity of the slurry, in Pa; is the fineness of the slurry, in μm; and are weighting coefficients; and are determined by fitting the dataset after dimensionality reduction. The first term in the formula characterizes the balance effect of measuring the fluidity and spreadability of the slurry. In a specific embodiment, if this value > 50, the slurry has poor fluidity but good spreadability (such as high viscosity, low surface tension), but may form a uniform but too thick coating. If this value < 50: the slurry has good fluidity but poor spreadability (such as low viscosity, high surface tension), which may lead to coating edge accumulation or uneven thickness. The second term in the formula is used to characterize the stability of the microstructure of the slurry. In a specific embodiment, if this value > 30: the slurry has strong elasticity and large particles, which is prone to particle agglomeration and coating defects. If this value < 30: the slurry has moderate elasticity and fine particles, and the dispersion uniformity is better.

[0041] The calculation formula of the coating stability factor is: ; In the formula, is the coating stability factor, is the thickness of the coating machine gasket, in μm; is the width of the flow channel of the coating machine gasket, in mm; is the opening angle of the coating machine gasket, in °; is the actual oven temperature, in °C; is the optimal oven temperature, in °C; The part outside the brackets in the formula is used to characterize the stability of the mechanical structure of the coating machine. The larger the flow channel width and gasket thickness, and the smaller the opening angle, the better the coating uniformity. The part inside the brackets expresses the penalty term for the oven temperature deviating from the optimal value. The closer the temperature is to the optimal value, the higher the stability.

[0042] The calculation formula of the die head dynamic balance coefficient is: ; In the formula, is the die head dynamic balance coefficient, is the horizontality of the die head lip, in μm / m; is the die head gap, in μm; is the circular runout amplitude of the back roller, in μm; 、 、 They are the standard deviations of the die lip flatness, die gap, and back roll roundness runout during continuous production. The denominator part in the formula is used to characterize the negative impact of parameter fluctuations. The smaller the fluctuations, the better the dynamic balance. The numerator part in the formula is used to characterize the synergistic effect between the die and the back roll. The greater the lip flatness and gap, and the smaller the back roll roundness runout, the higher the balance.

[0043] Step S108: Obtain training samples based on the composite process characteristics and the corresponding quality index categories, and use the training samples to train a dry film quality classification model with the composite process characteristics as the input and the quality index categories as the output.

[0044] Based on the composite process characteristics obtained from the foregoing steps, in this embodiment, the composite process characteristics and the quality index categories are used as the input and output to train a pre-constructed dry film quality classification model. Specifically, in this embodiment, based on the production process parameters and quality index categories collected during the production of each batch of dry film, a dataset is constructed using the composite process characteristics and quality index categories obtained from the production process parameters. In some specific embodiments, the dataset is randomly divided into a training set, a validation set, and a test set according to a preset ratio for training and optimizing the dry film quality classification model.

[0045] In some specific embodiments, the dry film quality classification model includes a random forest regression model and a support vector machine classification model connected in series in sequence. The training of the dry film quality classification model includes: Input the composite process characteristics into the random forest regression model to generate a predicted trend value of the quality index category; Combine the predicted trend value of the quality index category and the actual value of the quality index category to obtain a feature vector; Input the feature vector into the support vector machine classification model to predict the quality index category; Among them, the random forest regression model takes the composite process parameters as the input, captures non-linear relationships through multiple decision trees (usually 100 - 200 trees), and outputs the predicted trend value of the quality index category, that is, the intermediate quality prediction value surface density trend value, which is used to extract non-linear features and generate high-order inputs. In this embodiment, the first loss function is used to train the random forest regression model to determine the optimal number of decision trees through cross-validation; in a specific embodiment, the first loss function is the mean squared error (MSE), and the optimal number of trees is determined through cross-validation.

[0046] In addition, in this embodiment, a mixed kernel function composed of a Gaussian kernel and a polynomial kernel is used to map the feature vector to a high-dimensional space, and a second loss function is used to train the support vector machine classification model (SVM) to solve the optimal decision boundary in the high-dimensional space. In a specific embodiment, the quality index categories of the dry film are divided into qualified and unqualified, and the threshold of the binary classification result (qualified / unqualified) is set to 0.5, as Figure 2 shown in the schematic diagram of the training model of the support vector machine classification model.

[0047] In a specific embodiment, the mixed kernel function has the following formula:

[0048] In the formula, RBF represents the Gaussian kernel, Poly represents the polynomial kernel, represents the parameter weight, , controls the contribution ratio of the RBF and Poly kernels, and is optimized by the gradient descent method to adapt to the data distribution of different batches; represents the control parameter, which is used to control the width of the Gaussian kernel and determines the distribution range of the data mapped to the high-dimensional space; represents the square of the Euclidean distance between the input samples xi and xj. xi and xj are input samples, d is the polynomial order, which determines the complexity of the kernel function; c is a constant term, usually set to 1 or 0, and is used to adjust the form of the kernel function.

[0049] In a specific embodiment, the second loss function is the hinge loss function. An adaptive regularization term is set in the hinge loss function in this embodiment, which is expressed as:

[0050] In the formula, represents the adaptively adjusted regularization coefficient, represents that the initially set regularization coefficient is inversely proportional to the variance of the input feature, suppressing the overfitting of high-fluctuation parameters: X represents the input feature, represents the variance, and k represents the preset coefficient.

[0051] In some specific embodiments, the method further includes: using the trained dry film quality classification model to predict the quality index categories of the production processes of multiple batches of dry films with known actual values of quality index categories; determining the prediction error according to the prediction result and the actual value of the quality index category; generating a calibration curve according to the distribution of the prediction errors of multiple batches; the calibration curve is used to map the original prediction value of the dry film quality classification model to a calibrated value as the final prediction result of the dry film quality classification model; in this embodiment, the calibration curve is generated by the kernel density estimation method.

[0052] In addition, if the prediction errors of multiple consecutive batches (e.g., 3 batches) exceed a preset threshold (e.g., 5%), the retraining process is triggered to update the contribution ratios of the Gaussian kernel and the polynomial kernel and retrain the dry film quality classification model. By setting the retraining process in this embodiment, it can ensure that the model continuously adapts to changes in the production environment and has strong generalization ability for cross-batch data.

[0053] In a specific embodiment, this embodiment also uses composite process features as input and quality index categories as output to train a traditional SVM model, and then uses the trained traditional SVM model to predict the quality index of the surface density. The test results show that the performance of the dry film quality classification model is significantly better than that of the traditional SVM model. The specific comparison is as follows in the table: Table 1 Performance comparison table of dry film quality classification model and traditional SVM model

[0054] In another specific embodiment, for the quality index of the surface density, this embodiment verifies the generalization ability of the dry film quality classification model through multi-batch production. The data range used is the data collected from 10 consecutive production batches. The parameter fluctuations are set as the standard deviation of the oven temperature ±5°C and the die gap fluctuation ±3μm. The prediction deviation of the 6th batch reaches 6% to trigger model retraining. After retraining, the average deviation of the dry film quality classification model for the 7th - 10th batches drops to 3%.

[0055] Taking the example of the surface density as the quality index, a set of actual data actually obtained is used for illustration. Taking α = 0.6 and β = 0.4, the calculated = 4.28, coating stability factor: = 0.0812, die dynamic balance coefficient = 13.41, the trend value of the surface density output by the random forest regression = 24.8 g / m 2 ; The classification result of the improved SVM is unqualified, and the production process parameters need to be adjusted; after adjustment, the measured surface density: = 25.2 g / m 2 The pass rate is increased to 95%.

[0056] Embodiment 2 Combined with Figure 3 , a method for setting production process parameters for dry film production in this embodiment includes: Step S202: Initially set the production process parameters for dry film production; The initially set production process parameters are usually set according to experience. In a specific embodiment, the original production process parameters include: the slurry viscosity, slurry elasticity, slurry fineness, slurry surface tension, buffer tank liquid level difference, screw pump wear in the slurry mixing process, the cavity shape / quantity, coater gasket thickness, coater gasket flow channel width, coater gasket opening angle, coating oven temperature, coating oven speed check in the wet film process, the die head lip levelness, die head gap, back roller circular runout amplitude, back roller pressure, and back roller temperature in the dry film process, and set the specific values of each production process parameter.

[0057] Step S204: Determine the composite process characteristics according to the set production process parameters; Illustrated with a specific example, the variances of the production process parameters are as follows in the table: Table 2 Variance Table of the Collected Data of Production Process Parameters

[0058] Table 3 Key Process Characteristics Table Screened by Dimensionality Reduction

[0059] In a specific embodiment, the composite process characteristics are determined by the production process parameters after dimensionality reduction. The composite process characteristics include the slurry uniformity index in the slurry mixing process, the coating stability factor in the wet film process, and the die head dynamic balance coefficient in the dry film process. The composite process characteristics are calculated through the following formula according to the specific values of the set production process parameters.

[0060] The formula for the slurry uniformity index ( ) is: ; In the formula, is the slurry uniformity index, is the slurry viscosity, with the unit of pa·s; is the slurry surface tension, with the unit of mN / m; is the slurry elasticity, with the unit of pa; is the slurry fineness, with the unit of μm; and are the weight coefficients; and are determined by fitting the dataset after dimensionality reduction. The first term in the formula characterizes the balance effect of measuring the fluidity and spreadability of the slurry. In a specific embodiment, if this value > 50, the slurry has poor fluidity but good spreadability (such as high viscosity and low surface tension), but may form a uniform but overly thick coating. If this value < 50: the slurry has good fluidity but poor spreadability (such as low viscosity and high surface tension), which may lead to coating edge accumulation or uneven thickness. The second term Used to characterize the stability of the microstructure of the slurry. In a specific embodiment, if this value > 30: the slurry has strong elasticity and large particles, which easily leads to particle agglomeration and coating defects. If this value < 30: the slurry has moderate elasticity and small particles, and the dispersion uniformity is better.

[0061] The calculation formula for the coating stability factor is: ; In the formula, is the coating stability factor, is the thickness of the coating machine gasket, with the unit of μm; is the width of the flow channel of the coating machine gasket, with the unit of mm; is the opening angle of the coating machine gasket, with the unit of °; is the actual oven temperature, with the unit of °C; is the optimal oven temperature, with the unit of °C; the part outside the brackets in the formula is used to characterize the stability of the mechanical structure of the coating machine. The larger the flow channel width and gasket thickness, and the smaller the opening angle, the better the coating uniformity. The part inside the brackets expresses the penalty term for the oven temperature deviating from the optimal value. The closer the temperature is to the optimal value, the higher the stability.

[0062] The calculation formula for the die head dynamic balance coefficient is: ; In the formula, is the die head dynamic balance coefficient, is the horizontality of the die head lip, with the unit of μm / m; is the die head gap, with the unit of μm; is the circular runout amplitude of the back roll, with the unit of μm; , , are the standard deviations of the horizontality of the die head lip, the die head gap, and the circular runout amplitude of the back roll during continuous production, respectively. The denominator part in the formula is used to characterize the negative impact of parameter fluctuations. The smaller the fluctuations, the better the dynamic balance. The numerator part in the formula is used to characterize the synergistic effect between the die head and the back roll. The larger the horizontality and gap of the lip, and the smaller the circular runout amplitude of the back roll, the higher the balance.

[0063] Step S206: Input the composite process characteristics into the dry film quality classification model to obtain the quality index category of the dry film produced under the set production process parameters; the dry film quality classification model is obtained based on the training method of the dry film quality classification model described in Embodiment 1; After the trained dry film quality classification model can predict the quality index category of the dry film in advance, it can adjust the parameters first before defective products are produced to produce qualified dry films.

[0064] Step S208: If the quality index category corresponds to unqualified quality, iteratively adjust the set production process parameters until the quality index category of the dry film corresponds to qualified quality.

[0065] The specific adjustment method can be adjusted according to manual experience or enumerated by a machine. Input the composite process characteristics calculated from the adjusted production process parameters into the dry film quality classification model until the prediction result is qualified, so as to avoid the appearance of defective products through the pre-adjustment of production process parameters.

[0066] Embodiment 3

[0067] This embodiment provides a training device for a dry film quality classification model, which includes: An acquisition module, configured to acquire the production process parameters of each process in the production process of the dry film, and the quality index category of the produced dry film; A dimensionality reduction module, configured to perform dimensionality reduction processing on the production process parameters to screen out key process parameters; A processing module, configured to process the corresponding composite process characteristics of each process according to the key process parameters of each process; A training model, configured to obtain training samples based on the composite process characteristics and the corresponding quality index categories, and use the training samples to train a dry film quality classification model with the composite process characteristics as the input and the quality index category as the output. Embodiment 4

[0068] This embodiment provides a computer-readable storage medium, on which computer program instructions are stored. When the instructions are executed by a processor, the training method of the dry film quality classification model described in Embodiment 1 is implemented, or when the computer program instructions are executed by a processor, the production process parameter setting method described in Embodiment 2 is implemented.

[0069] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0070] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0071] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0072] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0073] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the claims of the present invention. These all fall within the protection scope of the present invention.

Claims

1. A training method for a dry film quality classification model, characterized in that Including: Obtaining the production process parameters of each process in the production process of the dry film, and the quality index categories of the produced dry film; Performing dimensionality reduction processing on the production process parameters to screen out key process parameters; Processing to obtain the composite process characteristics of the corresponding process according to the key process parameters under each process; Obtaining training samples based on the composite process characteristics and the corresponding quality index categories, and using the training samples to train a dry film quality classification model with the composite process characteristics as the input and the quality index categories as the output.

2. The training method of the dry film quality classification model according to claim 1, wherein The performing dimensionality reduction processing on the production process parameters includes: For each type of production process parameter, calculating the variance of the collected data; Taking the production process parameters with variances higher than the preset threshold as key process parameters.

3. The training method of the dry film quality classification model according to claim 1, characterized in that The quality index categories are divided according to the pre-calibrated qualified range: for the quality indicators within the qualified range, assign a value of 1; for the quality indicators outside the qualified range, assign a value of 0.

4. The training method of the dry film quality classification model according to claim 1, characterized in that The key process parameters include: the slurry viscosity, slurry elasticity, slurry fineness, and slurry surface tension in the sizing process, the coating machine gasket thickness, coating machine gasket flow channel width, coating machine gasket opening angle, and coating oven temperature in the wet film process, and the die head lip levelness, die head gap, and back roll circular runout amplitude in the dry film process; the composite process characteristics include the slurry uniformity index in the sizing process, the coating stability factor in the wet film process, and the die head dynamic balance coefficient in the dry film process.

5. The training method of the dry film quality classification model according to claim 4, characterized in that The calculation formula for the slurry uniformity index is: ； In the formula, is the slurry uniformity index, is the slurry viscosity; is the slurry surface tension; is the slurry elasticity; is the slurry fineness; and are the weight coefficients; The calculation formula for the coating stability factor is: ； In the formula, is the coating stability factor, is the thickness of the coating machine gasket; is the flow channel width of the coating machine gasket; is the opening angle of the coating machine gasket; is the actual oven temperature; is the optimal oven temperature; The calculation formula for the die head dynamic balance coefficient is: ； In the formula, is the dynamic balance coefficient of the die head, is the horizontal degree of the die head lip; is the die head gap; is the circular runout amplitude of the back roll; , , are the standard deviations of the horizontal degree of the die head lip, the die head gap, and the circular runout amplitude of the back roll during continuous production, respectively.

6. The training method of the dry film quality classification model according to claim 1, wherein The dry film quality classification model includes a random forest regression model and a support vector machine classification model connected in series in sequence. Training the dry film quality classification model includes: Inputting the composite process characteristics into the random forest regression model to generate a predicted trend value of the quality index category; Combining the predicted trend value of the quality index category and the actual value of the quality index category to obtain a feature vector; Inputting the feature vector into the support vector machine classification model to predict and obtain the quality index category; Among them, training the random forest regression model using the first loss function to determine the optimal number of decision trees through cross-validation; mapping the feature vector to a high-dimensional space using a mixed kernel function composed of a Gaussian kernel and a polynomial kernel, and training the support vector machine classification model using the second loss function to solve for the optimal decision boundary in the high-dimensional space.

7. The training method of the dry film quality classification model according to claim 6, characterized in that It also includes: Using the trained dry film quality classification model to predict the quality index category of the production process of multiple batches of dry films with known actual values of the quality index category; Determining the prediction error according to the prediction result and the actual value of the quality index category; Generating a calibration curve according to the distribution of the prediction errors of multiple batches; the calibration curve is used to map the original prediction value of the dry film quality classification model to a calibration value as the final prediction result of the dry film quality classification model.

8. A training device for a dry film quality classification model, characterized in that, Including: An acquisition module, configured to acquire the production process parameters of each process in the production process of the dry film, and the quality index categories of the produced dry film; A dimensionality reduction module for performing dimensionality reduction processing on the production process parameters to screen out key process parameters; A processing module for respectively processing according to the key process parameters under each process to obtain composite process characteristics corresponding to the process; A training model for obtaining training samples based on the composite process characteristics and the corresponding quality index categories, and using the training samples to train a dry film quality classification model with the composite process characteristics as the input and the quality index categories as the output.

9. A method for setting production process parameters in dry film production, characterized in that, including: Initial setting of the production process parameters for dry film production; Determining composite process characteristics according to the set production process parameters; Inputting the composite process characteristics into the dry film quality classification model to obtain the quality index category of the dry film produced under the set production process parameters; the dry film quality classification model is obtained based on the training method of the dry film quality classification model according to any one of claims 1-7; If the quality index category corresponds to unqualified quality, iteratively adjust the set production process parameters until the quality index category of the dry film corresponds to qualified quality.

10. A computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the training method of the dry film quality classification model according to any one of claims 1 to 7 is implemented; Or, when the computer program instructions are executed by a processor, the production process parameter setting method according to claim 9 is implemented.

Images