Method and device for predicting preparation process of biaxially stretched polytetrafluoroethylene fiber membrane

Abstract

A method and device for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membranes, relating to the field of process prediction. In this method, a data set of multiple groups of fiber membrane samples prepared by preset orthogonal experiments is obtained, the data set including the biaxially oriented process parameter combination, pore structure data, and performance indicators for preparing the target group of fiber membrane samples; an initial machine learning model is constructed, the biaxially oriented process parameter combination in the data set is used as input parameters, the pore structure data and performance indicators are used as output parameters, the initial machine learning model is trained to obtain an intermediate model, and the parameters of the intermediate model are adjusted through cross-validation to obtain a machine learning model; based on the set target performance, the optimal process parameter combination and confidence interval are recommended by the machine learning model. The technical solution provided by this application is implemented to achieve the purpose of efficiently predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membranes.

Description

Technical Field

[0001] The present application relates to the technical field of process prediction, and in particular to a method and apparatus for predicting a process for preparing a biaxially oriented polytetrafluoroethylene fiber membrane. Background Art

[0002] Polytetrafluoroethylene (PTFE) is widely used in filter materials, medical materials, electronic devices, and environmental protection due to its excellent chemical stability, low friction coefficient, high-temperature resistance, and superior dielectric properties. Biaxial stretching is a core technology for producing high-performance PTFE fiber membranes. Through simultaneous or step-by-step stretching in the longitudinal and transverse directions, the membrane's macrostructure, pore structure, and various properties can be manipulated to meet the needs of diverse application scenarios.

[0003] The preparation of biaxially oriented PTFE membranes involves the coupling of multiple parameters, including longitudinal and transverse stretching temperature, stretching rate, stretch ratio, heat setting conditions, and other process parameters. These parameters exhibit complex, nonlinear relationships with the final product's structure (such as porosity and pore size distribution) and properties (such as air permeability, filtration efficiency, and breaking strength). Therefore, achieving a specific combination of structure and properties is a complex and difficult process. Existing biaxially oriented PTFE fiber membrane preparation processes require numerous (often exceeding hundreds of) experiments to determine the required multiple stretching process parameters. This is not only time-consuming and labor-intensive, but also results in significant experimental costs and resource waste. Furthermore, it cannot guarantee that the resulting material will possess the optimal combination of structure and properties.

[0004] Therefore, a method is needed to predict the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane according to target performance. Summary of the Invention

[0005] The present application provides a method and device for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membranes, which achieves the purpose of efficiently predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membranes, significantly improves the accuracy and stability of process parameter selection, and reduces the trial and error cost, providing reliable support for the development of high-performance fiber membranes.

[0006] In a first aspect of the present application, a method for predicting a preparation process of a biaxially oriented polytetrafluoroethylene fiber membrane is provided, which is applied to a preparation platform. The method comprises:

[0007] Obtaining a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment, the data set including a combination of biaxial stretching process parameters, pore structure data, and performance indicators for preparing a target group of fiber membrane samples, wherein the target group of fiber membrane samples is any one group of the multiple groups of fiber membrane samples;

[0008] constructing an initial machine learning model, using the combination of biaxial stretching process parameters in the data set as input parameters, and the pore structure data and performance indicators as output parameters, training the initial machine learning model to obtain an intermediate model, and adjusting the parameters of the intermediate model through cross-validation to obtain a machine learning model;

[0009] Based on the set target performance, the machine learning model is used to recommend the optimal process parameter combination and confidence interval.

[0010] Optionally, constructing the initial machine learning model includes embedding a physical constraint mechanism, specifically including:

[0011] Calculating a critical speed of polytetrafluoroethylene according to preset viscoelastic parameters, setting the critical speed between the input layer and the hidden layer, and triggering parameter correction when the input stretching speed exceeds the critical speed;

[0012] A porosity restriction function is set in the output layer. When the porosity predicted by the model exceeds the interval set by the porosity restriction function, the output result is corrected according to a preset adjustment strategy.

[0013] Optionally, the recommending of the optimal process parameter combination and confidence interval by the machine learning model includes:

[0014] Determining the range and step size of biaxial stretching process parameters, wherein the biaxial stretching process parameters include transverse stretching ratio, transverse speed, transverse temperature, longitudinal stretching ratio, longitudinal speed, and longitudinal temperature;

[0015] Determine a parameter space according to the range, and uniformly generate a plurality of parameter combinations in the parameter space using a Latin hypercube sampling method;

[0016] The crossover probability, mutation probability, elite retention ratio and crowding degree calculation method are set, and the optimal solution is obtained from the multiple parameter combinations through non-dominated sorting and crowding distance calculation.

[0017] Optionally, the recommending of the optimal process parameter combination and confidence interval by the machine learning model further includes:

[0018] introducing random perturbations for each of the biaxial stretching process parameters, running the machine learning model multiple times, and obtaining performance index results under different perturbations;

[0019] The coefficients of variation of air permeability and filtration efficiency are calculated according to the performance indicator results, and a process solution is screened out in which the coefficients of variation of the air permeability and the filtration efficiency are both lower than a first preset threshold.

[0020] Optionally, the method further includes:

[0021] The measured data of the samples prepared according to the optimal process parameter combination are analyzed for deviation from the predicted values. When the absolute error of the porosity or the error of the filtration efficiency exceeds the second preset threshold, the transfer learning strategy is used to freeze the underlying parameters related to the basic feature extraction in the machine learning model, and the preset fully connected layer is adjusted according to the measured data of the samples.

[0022] Optionally, the utilizing of a transfer learning strategy to freeze underlying parameters related to basic feature extraction in the machine learning model and adjusting a preset fully connected layer according to the sample measured data includes:

[0023] Freeze the weight matrix and convolution kernel parameters of the feature extraction layer in the initial machine learning model. When the loss function decrease rate of three consecutive rounds of training is less than a preset percentage, multiply the current learning rate of the fully connected layer by the attenuation coefficient.

[0024] Optionally, the step of obtaining a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment includes:

[0025] The polytetrafluoroethylene dispersion resin and hydrocarbon lubricating oil are mixed in a preset weight ratio and aged at 30-70°C for 24-48 hours; the undeoiled base tape is formed by forming a blank, pushing, extruding and calendering.

[0026] The longitudinal stretching ratio is controlled to be 1 to 16 times, the longitudinal speed is 1 to 5 m / min, and the longitudinal temperature is 120 to 330° C., and the transverse expansion ratio is controlled to be 3 to 25 times, the transverse speed is 1 to 16 m / min, and the transverse temperature is 210 to 320° C. to generate multiple groups of fiber membrane samples;

[0027] The pore structures and properties of multiple groups of the fiber membrane samples are detected to obtain the data set.

[0028] In a second aspect of the present application, a system for predicting a process for preparing a biaxially oriented polytetrafluoroethylene fiber membrane is provided, comprising a data module, a model module, and a prediction module, wherein:

[0029] A data module is configured to obtain a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment, wherein the data set includes a combination of biaxial stretching process parameters, pore structure data, and performance indicators for preparing a target group of fiber membrane samples, wherein the target group of fiber membrane samples is any one group of the multiple groups of fiber membrane samples;

[0030] a model module configured to construct an initial machine learning model, using the combination of biaxial stretching process parameters in the data set as input parameters, using the pore structure data and performance indicators as output parameters, training the initial machine learning model to obtain an intermediate model, and adjusting the parameters of the intermediate model through cross-validation to obtain a machine learning model;

[0031] The prediction module is configured to recommend an optimal process parameter combination and a confidence interval based on the set target performance through the machine learning model.

[0032] In the third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs any of the methods described above.

[0033] In a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions. When the instructions are executed, any one of the methods described above is executed.

[0034] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0035] 1. By acquiring data sets from multiple fiber membrane samples, including biaxial stretching process parameter combinations, pore structure data, and performance indicators, we provide rich training data for the machine learning model. This data covers the fiber membrane preparation under different process conditions, enabling the model to learn the complex relationship between process parameters and pore structure and performance indicators, thereby improving prediction accuracy;

[0036] 2. Using cross-validation to adjust the parameters of the intermediate model can effectively avoid the problem of overfitting or underfitting the model. By dividing the dataset into multiple subsets and training and validating them in turn, we can more comprehensively evaluate the performance of the model, find the optimal parameter combination, and further improve the prediction accuracy of the machine learning model;

[0037] 3. Based on the trained machine learning model, the optimal process parameter combination and confidence interval can be quickly recommended according to the set target performance. Compared with traditional trial-and-error or empirical methods, this method can quickly select the solution that best meets the target performance requirements from a large number of possible process parameter combinations, greatly reducing the time and cost of process optimization and improving production efficiency.

[0038] 4. The model's prediction results can clarify the trend and extent of the impact of process parameters on fiber membrane performance, providing strong guidance for subsequent experimental design. Researchers can conduct targeted experimental verification and further optimization based on the optimal process parameter combination recommended by the model and its confidence interval, avoiding the waste of resources and time delays caused by blind experiments;

[0039] 5. This method can achieve precise control of fiber membrane pore structure data and performance indicators through accurate prediction and optimization of biaxial stretching process parameters. The stability and consistency of key performance indicators such as porosity, air permeability, and filtration efficiency are improved, thereby ensuring the quality and performance stability of fiber membrane products and meeting the stringent performance requirements of fiber membranes in different application scenarios.

[0040] 6. In actual production, various factors may cause fluctuations in process parameters, which in turn affect product quality. This method uses machine learning to model and predict the relationship between process parameters and product performance. It can promptly detect abnormal changes in process parameters and adjust process parameters to compensate for the impact of these changes on product quality, thereby reducing the adverse effects of process fluctuations on product quality and improving product quality stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 1 is a flow chart of a method for predicting a process for preparing a biaxially oriented polytetrafluoroethylene fiber membrane disclosed in an embodiment of the present application;

[0042] Figure 2 This is a schematic diagram of the structure of the core machine learning algorithm disclosed in the embodiments of this application;

[0043] Figure 3 This is a module schematic diagram of a system for predicting a process for preparing a biaxially oriented polytetrafluoroethylene fiber membrane disclosed in an embodiment of the present application;

[0044] Figure 4 This is a structural diagram of an electronic device disclosed in an embodiment of the present application.

[0045] Explanation of the reference numerals: 301, data module; 302, model module; 303, prediction module; 401, processor; 402, communication bus; 403, user interface; 404, network interface; 405, memory. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0047] In the description of the embodiments of this application, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a concrete manner.

[0048] In the description of the embodiments of the present application, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0049] This embodiment discloses a method for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane, which is applied to the preparation platform. Figure 1 Schematic diagram of the process for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane disclosed in the embodiment of the present application, such as Figure 1 As shown, the method includes the following steps:

[0050] S101. Acquire a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment, wherein the data set includes a combination of biaxial stretching process parameters, pore structure data, and performance indicators for preparing a target group of fiber membrane samples, wherein the target group of fiber membrane samples is any one group of the multiple groups of fiber membrane samples;

[0051] S102, constructing an initial machine learning model, using the combination of biaxial stretching process parameters in the data set as input parameters, and the pore structure data and performance indicators as output parameters, training the initial machine learning model to obtain an intermediate model, and adjusting the parameters of the intermediate model through cross-validation to obtain a machine learning model;

[0052] S103. Based on the set target performance, the machine learning model is used to recommend the optimal process parameter combination and confidence interval.

[0053] Prepare multiple sets of fiber membrane samples using pre-defined orthogonal experiments. Orthogonal experiments are a statistical experimental design method that allows for rational experiment arrangement and effective analysis of the impact of multiple factors on experimental results. In this step, the prepared fiber membrane sample dataset contains the following important information:

[0054] Biaxial stretching process parameter combinations: These are key factors influencing the performance of fiber membranes, including transverse stretch ratio, transverse speed, transverse temperature, longitudinal stretch ratio, longitudinal speed, and longitudinal temperature. Different parameter combinations can lead to differences in the orientation of the molecular chains and the formation of the pore structure during the stretching process, thus affecting the performance of the fiber membrane.

[0055] Pore structure data: Pore structure is a key characteristic of fiber membranes, including porosity and pore size distribution. Porosity reflects the proportion of pores in a fiber membrane, while pore size distribution describes the size range and number distribution of pores. These data can be measured using methods such as scanning electron microscopy (SEM) and nitrogen adsorption-desorption isotherms.

[0056] Performance indicators: Performance indicators are used to evaluate the quality and application performance of fiber membranes. Common performance indicators include air permeability, filtration efficiency, tensile strength, and elongation at break. Air permeability reflects the fiber membrane's ability to transmit gas, filtration efficiency indicates how well the fiber membrane filters particulate matter, and tensile strength and elongation at break reflect the fiber membrane's mechanical properties.

[0057] Among these fiber membrane samples, any group can be used as the target group, and their corresponding biaxial stretching process parameter combinations, pore structure data, and performance indicators will serve as the basic data for the subsequent construction of the machine learning model. An appropriate machine learning algorithm, such as a neural network, support vector machine, or decision tree, is selected to build the model. The biaxial stretching process parameter combinations are used as input parameters. These parameters are controllable factors affecting the performance of the fiber membrane, and by adjusting them, the fiber membrane preparation process can be changed. The pore structure data and performance indicators are used as output parameters. They represent the final performance of the fiber membrane and are the targets to be optimized by adjusting the process parameters. The acquired dataset is used to train the initial machine learning model. During training, the model learns the mapping relationship between the corresponding pore structure data and performance indicators based on the input process parameters. By continuously adjusting the model's internal parameters, the model's predictions are made as close as possible to the actual data, thus obtaining an intermediate model. To evaluate the model's performance and avoid overfitting or underfitting, a cross-validation method is used. A common cross-validation method is k-fold cross-validation (k is a positive integer greater than or equal to 2). This involves dividing the dataset into k subsets, selecting one of these subsets as the validation set and the remaining k-1 subsets as the training set. The training and validation process is repeated k times. Based on the performance metrics (such as mean squared error and accuracy) on the validation set, the parameters of the intermediate model, such as the number of neural network layers, number of neurons, and learning rate, are adjusted to improve the model's generalization and prediction accuracy, ultimately resulting in an optimized machine learning model. Based on the set target performance, such as requiring the fiber membrane to have high air permeability and filtration efficiency, or to be within a certain range of mechanical properties, these target performance attributes are input into the trained machine learning model. Based on the previously learned relationships between process parameters and performance metrics, the model predicts the optimal combination of process parameters that achieves the target performance. To assess the reliability of the prediction results, corresponding confidence intervals are also provided. The confidence interval indicates the range of values of the performance indicators corresponding to the predicted optimal process parameter combination under a certain confidence level. It provides a reference basis for process control in actual production and helps operators to reasonably adjust the process parameters within the confidence interval to ensure that the performance of the fiber membrane meets the set requirements.

[0058] Optionally, the step of obtaining a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment includes:

[0059] The polytetrafluoroethylene dispersion resin and hydrocarbon lubricating oil are mixed in a preset weight ratio and aged at 30-70°C for 24-48 hours; the undeoiled base tape is formed by forming a blank, pushing, extruding and calendering.

[0060] The longitudinal stretching ratio is controlled to be 1 to 16 times, the longitudinal speed is 1 to 5 m / min, and the longitudinal temperature is 120 to 330° C., and the transverse expansion ratio is controlled to be 3 to 25 times, the transverse speed is 1 to 16 m / min, and the transverse temperature is 210 to 320° C. to generate multiple groups of fiber membrane samples;

[0061] The pore structures and properties of multiple groups of the fiber membrane samples are detected to obtain the data set.

[0062] A polytetrafluoroethylene (PTFE) dispersion resin (crystallinity ≥97%, number-average molecular weight 2-20 million) is mixed with a hydrocarbon lubricant at a predetermined weight ratio of 1:(0.10-0.35). This ratio is carefully designed to ensure good flow and plasticity during subsequent processing, while the hydrocarbon lubricant provides lubrication and improves processing performance. The mixed material is then aged at 30-70°C for 24-48 hours. This aging process allows the components in the mixture to fully interact with each other, forming a homogeneous and stable system, preparing it for subsequent molding, extrusion, and calendering. The aged material is then subjected to molding, extrusion, and calendering, ultimately forming an undeoiled base tape. Molding is performed at a temperature of 30-50°C, with a holding pressure of 3-10 minutes. This step aims to form a green body with a defined shape and density, providing a foundation for subsequent extrusion. The extrusion process is carried out at a temperature of 30-60°C, with a compression ratio controlled between 50 and 300. This extrusion further homogenizes the material and forms a continuous green body of defined size and shape. The final calendering step ensures a more uniform thickness, resulting in an undeoiled base tape that meets the requirements for subsequent biaxial stretching. In the orthogonal experimental design for biaxial stretching, the longitudinal stretching process parameters were set as follows: the longitudinal stretching ratio was set between 1 and 16 times, the stretching speed was controlled between 1 and 5 m / min, and the stretching temperature was adjusted between 120 and 330°C. Different combinations of stretching ratio, speed, and temperature significantly affect the longitudinal properties of the PTFE fiber membrane. For example, a higher stretching ratio results in a higher degree of longitudinal orientation of the fiber membrane, potentially increasing longitudinal strength but also potentially affecting other properties such as flexibility. Variations in stretching speed and temperature influence the movement and rearrangement of the molecular chains, thus affecting the microstructure and macroscopic properties of the fiber membrane. The transverse expansion process parameters are set as follows: transverse expansion multiple of 3 to 25 times, transverse speed of 1 to 16 m / min, and transverse temperature of 210 to 320°C. The transverse expansion multiple determines the degree of dimensional change of the fiber membrane in the transverse direction, and together with the longitudinal stretching multiple, affects the overall pore structure and performance of the fiber membrane; the transverse speed and temperature also have a significant impact on the transverse performance of the fiber membrane. Together with the longitudinal stretching process parameters, through the cross-combination of transverse and longitudinal parameters, polytetrafluoroethylene fiber membrane samples with different pore structures and properties can be prepared. According to the parameter range in the above-mentioned orthogonal experimental design, multiple groups of fiber membrane samples are generated by controlling parameters such as the longitudinal stretching multiple, longitudinal speed, longitudinal temperature, transverse expansion multiple, transverse speed, and transverse temperature. In the process of preparing each group of fiber membrane samples, the preset process parameters are strictly followed to ensure that the sample preparation conditions are accurate.Advanced testing methods are used to examine the pore structure (including porosity and pore size) and performance (air permeability, filtration efficiency, etc.) of multiple prepared fiber membrane samples. Porosity can be measured using methods such as nitrogen adsorption-desorption isotherms, while pore size can be measured using techniques such as scanning electron microscopy (SEM). Air permeability is typically measured using a specialized air permeability tester according to standard testing methods. Filtration efficiency testing requires filtration experiments using appropriate particles or gases, tailored to the fiber membrane's application scenario, and then measured using specialized testing equipment.

[0063] Mixing polytetrafluoroethylene (PTFE) dispersion resin and hydrocarbon lubricant in a predetermined weight ratio and aging at 30-70°C for 24-48 hours allows the two to fully blend and form a uniform and stable mixture. This helps ensure consistent material flowability and plasticity during subsequent molding, extrusion, and calendering processes, resulting in an undeoiled base tape with uniform quality and stable performance, laying a good foundation for the biaxial stretching process. By controlling parameters such as the molding temperature at 30-50°C, the holding time at 3-10 minutes, the extrusion temperature at 30-60°C, and the compression ratio at 50-300, the undeoiled base tape formation process can be precisely controlled. This improves the density and uniformity of the base tape, reduces internal defects and stress concentration, and enables it to better withstand tensile forces during the subsequent biaxial stretching process, avoiding stretching failure or unstable fiber membrane performance due to base tape quality issues. Setting the parameter ranges of longitudinal stretching ratio of 1 to 16 times, longitudinal speed of 1 to 5 m / min, longitudinal temperature of 120 to 330°C, and transverse expansion ratio of 3 to 25 times, transverse speed of 1 to 16 m / min, and transverse temperature of 210 to 320°C provides a rich range of process options for the preparation of fiber membranes. Different combinations of stretching ratio, speed, and temperature will have a significant impact on the pore structure and properties of polytetrafluoroethylene fiber membranes, and can meet the diverse performance requirements of fiber membranes in different application scenarios. Due to the use of orthogonal experimental design and a wide range of stretching parameters, the generated data set has good diversity and representativeness. This enables the machine learning model to effectively learn and train under different process conditions, avoiding model overfitting or underfitting problems caused by single or limited data, ensuring that the model can accurately predict the performance of fiber membranes under various process parameter combinations, and providing a reliable basis for process optimization and product quality control in actual production.

[0064] Optionally, constructing the initial machine learning model includes embedding a physical constraint mechanism, specifically including:

[0065] Calculating a critical speed of polytetrafluoroethylene according to preset viscoelastic parameters, setting the critical speed between the input layer and the hidden layer, and triggering parameter correction when the input stretching speed exceeds the critical speed;

[0066] A porosity restriction function is set in the output layer. When the porosity predicted by the model exceeds the interval set by the porosity restriction function, the output result is corrected according to a preset adjustment strategy.

[0067] The critical velocity of polytetrafluoroethylene (PTFE) is calculated based on preset viscoelastic parameters. As a viscoelastic polymer, the movement and deformation behavior of its molecular chains during stretching are constrained by its viscoelastic properties. Pre-set viscoelastic parameters, such as relaxation time and modulus, allow the accurate calculation of the critical velocity of PTFE under specific process conditions. The critical velocity refers to the maximum velocity at which a material can maintain stable deformation without breaking or abnormal flow during stretching. The calculated critical velocity serves as an important reference value and is set between the input and hidden layers of the machine learning model. When input stretching velocity data is passed to this layer, the model compares it with the critical velocity and makes a judgment. If the input stretching velocity exceeds the set critical velocity, the model triggers a parameter correction mechanism. This is because, during actual biaxial stretching, if the stretching velocity is too fast, exceeding the critical velocity of PTFE, the material may experience uneven deformation, molecular chain breakage, or structural damage, leading to degraded fiber membrane performance. By triggering parameter correction, the model adjusts the input stretching velocity and other related parameters to maintain them within a reasonable range, ensuring the accuracy and reliability of the prediction results. For example, the stretching speed can be appropriately reduced or other process parameters optimized to simulate measures taken in actual production to avoid exceeding the critical speed. A porosity constraint function is set at the output layer of the machine learning model. Porosity is a key structural parameter of polytetrafluoroethylene (PTFE) fiber membranes, directly affecting their performance, such as air permeability and filtration efficiency. Based on the application requirements and actual production experience of the fiber membrane, a reasonable porosity range can be determined and set as the interval of the porosity constraint function. For example, for certain filtration applications, the porosity of fiber membranes may need to be controlled between 30% and 70% to ensure good filtration performance and mechanical strength. When the model predicts the porosity of the fiber membrane, it compares the predicted result with the interval set by the porosity constraint function. If the predicted porosity exceeds this range, it indicates that the model's prediction may not meet actual production requirements or physical laws. If the predicted porosity exceeds the set range, the model will correct the output according to a preset adjustment strategy. The preset adjustment strategy can be implemented in a variety of ways. For example, the predicted porosity can be adjusted to the boundary of the interval, or it can be reduced or increased according to a certain ratio based on the degree of excess to bring it back within a reasonable range. This ensures that the porosity results output by the model have physical significance and practical application value, avoiding incorrect assessments of fiber membrane performance due to unreasonable porosity predictions. At the same time, by correcting the output results, the model can better learn the reasonable relationship between process parameters and porosity during subsequent training, thereby improving the model's prediction accuracy and stability.

[0068] By calculating the critical velocity of polytetrafluoroethylene (PTFE) based on preset viscoelastic parameters and setting this critical velocity between the input and hidden layers, the physical properties of PTFE can be incorporated into the machine learning model. When the input stretching velocity exceeds the critical velocity, parameter correction is triggered. This prevents the model from predicting results that are inconsistent with the material's physical properties, ensuring that the model's predictions more accurately reflect the actual situation. A porosity constraint function is set at the output layer. When the model-predicted porosity exceeds the set range, the output result is corrected according to a preset adjustment strategy, further ensuring that the model-predicted porosity data meets actual process requirements and material properties. Porosity is a key structural parameter of fiber membranes and significantly impacts their performance. In actual production, due to constraints from process conditions and material properties, the porosity of fiber membranes typically falls within a specific range. By using the porosity constraint function and output correction, the model avoids predicting unreasonable porosity values and improves the reliability of prediction results. Embedding physical constraints effectively prevents anomalous model predictions. Without physical constraints, machine learning models may make predictions based on patterns in the training data, but these patterns may not fully reflect the material's physical properties and actual process requirements. The introduction of physical constraints helps improve the model's generalization capabilities. When faced with varying process parameters and material properties, the model can make reasonable predictions based on physical constraints, rather than relying solely on specific patterns in the training data. This allows the model to better adapt and predict fiber membrane performance when dealing with unseen data or new process conditions, improving its applicability and reliability.

[0069] Optionally, the recommending of the optimal process parameter combination and confidence interval by the machine learning model includes:

[0070] Determining the range and step size of biaxial stretching process parameters, wherein the biaxial stretching process parameters include transverse stretching ratio, transverse speed, transverse temperature, longitudinal stretching ratio, longitudinal speed, and longitudinal temperature;

[0071] Determine a parameter space according to the range, and uniformly generate a plurality of parameter combinations in the parameter space using a Latin hypercube sampling method;

[0072] The crossover probability, mutation probability, elite retention ratio and crowding degree calculation method are set, and the optimal solution is obtained from the multiple parameter combinations through non-dominated sorting and crowding distance calculation.

[0073] Biaxial stretching process parameters include transverse stretch ratio, transverse speed, transverse temperature, longitudinal stretch ratio, longitudinal speed, and longitudinal temperature. The ranges for these parameters are determined based on actual production experience and material properties. For example, the transverse stretch ratio might range from 3 to 25 times, the longitudinal stretch ratio might range from 1 to 16 times, the transverse speed might range from 1 to 16 m / min, the longitudinal speed might range from 1 to 5 m / min, the transverse temperature might range from 210 to 320°C, and the longitudinal temperature might range from 120 to 330°C. Determining these parameter ranges ensures that the parameter combinations generated during the optimization process are within practical limits, avoiding parameter values that exceed the equipment capabilities or material tolerances. The step size refers to the interval between two consecutive parameter values within the parameter range. A reasonable step size ensures a dense sampling of the parameter space, allowing for full exploration of the parameter space during the optimization process. For example, the step size for transverse stretch ratio can be set to 1, for transverse speed to 1 m / min, and for transverse temperature to 10°C. The step size affects the accuracy and efficiency of optimization. A too small step size leads to an excessive number of parameter combinations, increasing the computational effort; a too large step size may result in missing some important parameter combinations. Latin Hypercube Sampling (LHS) is a statistical method for sampling multidimensional parameter spaces. The basic idea is to evenly divide the range of each parameter into several subintervals and then randomly select a sample point within each subinterval. For each parameter, these sample points are evenly distributed within the parameter range, and the sample points for different parameters are independent of each other. This method achieves better coverage of the parameter space with fewer samples, improving sampling efficiency and representativeness. Based on the determined biaxial stretching process parameter range and step size, the dimensions of the parameter space and the value range of each dimension are determined. Then, Latin Hypercube sampling is used to uniformly generate multiple parameter combinations within this parameter space. For example, if the parameter space has six dimensions (i.e., transverse stretch ratio, transverse speed, transverse temperature, longitudinal stretch ratio, longitudinal speed, and longitudinal temperature), and the value ranges for each dimension are [3, 25], [1, 16], [210, 320], [1, 16], [1, 5], and [120, 330], with step sizes of 1, 1, 10, 1, 0.5, and 10, respectively), multiple parameter combinations can be generated, such as (3, 1, 210, 1, 1, 120), (4, 2, 220, 2, 1.5, 130), and (5, 3, 230, 3, 2, 140). These parameter combinations serve as input to the subsequent optimization algorithm to find the optimal process parameter combination. To find the optimal solution from the generated multiple parameter combinations, the optimization algorithm parameters must be set. Common optimization algorithm parameters include crossover probability, mutation probability, elite retention ratio, and crowding calculation method.Crossover probability and mutation probability control the crossover and mutation operations in genetic algorithms. Crossover allows different parameter combinations to exchange information and generate new parameter combinations. Mutation introduces randomness into the parameter combinations, increasing solution diversity. The elite retention ratio is used to retain a certain proportion of excellent individuals during each generation of evolution to ensure the convergence speed and solution quality of the optimization algorithm. The crowding degree calculation method measures the degree of crowding between individuals and is used to calculate the crowding distance in the non-dominated sorting genetic algorithm II (NSGA-II) to maintain solution diversity. Non-dominated sorting and crowding distance calculations are used to obtain the optimal solution from multiple parameter combinations. Non-dominated sorting ranks parameter combinations based on their objective function values, with the best parameter combinations being ranked higher. The crowding distance calculation measures the degree of crowding between individuals after non-dominated sorting. Individuals with larger crowding distances are more dispersed and more likely to be retained. Non-dominated sorting and crowding distance calculations can yield a set of optimal solutions that strike a balance between objective function value and solution diversity. These solutions can be used as recommended optimal process parameter combinations.

[0074] Clearly defining the range and step size of biaxial stretching process parameters, including transverse stretching ratio, transverse speed, transverse temperature, longitudinal stretching ratio, longitudinal speed, and longitudinal temperature, provides clear boundaries and search accuracy for subsequent parameter combination generation. This facilitates a comprehensive and efficient search within the parameter space, avoiding inadequate optimization or ineffective calculations caused by overly broad or narrow parameter ranges. Using the Latin hypercube sampling method to uniformly generate multiple parameter combinations within the parameter space ensures a uniform distribution of parameters across all dimensions, improving the representativeness and diversity of the sampling points. Compared to traditional random sampling methods, Latin hypercube sampling provides better coverage of the parameter space with fewer samples, reduces sampling bias, and enhances the global search capability and efficiency of the optimization algorithm. Fine-tuning the optimization algorithm is achieved by setting parameters such as the crossover probability, mutation probability, elite retention ratio, and crowding calculation method. Properly setting the crossover and mutation probabilities balances the algorithm's global and local search capabilities, preventing premature convergence or becoming trapped in a local optimum. Setting the elite retention ratio preserves the best individuals in the current population, accelerating convergence. The crowding calculation method helps maintain solution diversity within the population. The use of non-dominated sorting and crowding distance calculation to obtain the optimal solution from multiple parameter combinations can simultaneously consider the optimization of multiple objective functions, such as the porosity, air permeability, filtration efficiency and other performance indicators of the fiber membrane. Non-dominated sorting can hierarchically sort parameter combinations according to their performance on each objective function, and crowding distance calculation can sort according to the distribution density of solutions in the same layer, thereby selecting parameter combinations that perform well on the objective function and are evenly distributed, thereby enhancing the diversity and quality of the optimization results. The optimal process parameter combination obtained by the above method can better meet the performance requirements of the fiber membrane and improve product quality and production efficiency. Using these optimized process parameter combinations for the training and verification of machine learning models can enable the model to learn a more accurate relationship between process parameters and fiber membrane performance, thereby improving the model's prediction accuracy.

[0075] Optionally, the recommending of the optimal process parameter combination and confidence interval by the machine learning model further includes:

[0076] introducing random perturbations for each of the biaxial stretching process parameters, running the machine learning model multiple times, and obtaining performance index results under different perturbations;

[0077] The coefficients of variation of air permeability and filtration efficiency are calculated according to the performance indicator results, and a process solution is screened out in which the coefficients of variation of the air permeability and the filtration efficiency are both lower than a first preset threshold.

[0078] Random perturbations are introduced to each biaxial stretching process parameter to simulate the fluctuations and uncertainties that may exist in actual production processes. These fluctuations can be caused by factors such as equipment accuracy, environmental variations, and raw material differences. By introducing random perturbations, the model's performance indicators can be evaluated under different perturbations, thereby identifying process solutions that are insensitive to process parameter fluctuations and exhibit stable performance. During multiple runs of the machine learning model, random perturbations are introduced for each biaxial stretching process parameter (e.g., transverse stretch ratio, transverse speed, transverse temperature, longitudinal stretch ratio, longitudinal speed, and longitudinal temperature). The magnitude of the random perturbations can be determined based on actual production experience or the allowable fluctuation range of the process parameters. Typically, random number generation methods such as normal and uniform distributions are used to generate perturbations, which are then added to the original process parameters to obtain the perturbed process parameter combination. After introducing the random perturbations, the machine learning model is run multiple times, each using the perturbed process parameter combination as input, to obtain the corresponding performance indicator results (e.g., air permeability, filtration efficiency, etc.). Through multiple runs, a set of performance indicator results under different perturbations is obtained, reflecting the performance changes of the fiber membrane under process parameter fluctuations. The performance indicator results from each run are collected and recorded to form a data set containing multiple performance indicator values. This data will be used for subsequent coefficient of variation calculations and process solution screening. The coefficient of variation (CV) is an important indicator of data dispersion. It is equal to the ratio of the standard deviation to the mean and is usually expressed as a percentage. The CV eliminates the influence of data dimension and mean value, more accurately reflecting the relative dispersion of the data. For the two performance indicators, airflow rate and filtration efficiency, their coefficients of variation are calculated under different perturbations. The specific calculation steps are as follows: Calculate the average values of airflow rate and filtration efficiency across multiple runs, denoted as Q and E, respectively. Calculate the standard deviations of airflow rate and filtration efficiency across multiple runs, denoted as SDQ and SDE, respectively. Calculate the coefficient of variation (CVQ) of airflow rate: CVQ = Q / SDQ × 100%. Calculate the coefficient of variation (CVE) of filtration efficiency: CVE = E / SDE × 100%. Screen the coefficients of variation of airflow rate and filtration efficiency based on a set first preset threshold. Only when the coefficient of variation of air permeability and filtration efficiency are both below this threshold, the corresponding process solution is considered to be stable and reliable. The first preset threshold can be determined based on the actual production requirements for fiber membrane performance stability, and can usually be set through experimental verification or empirical judgment.

[0079] By introducing random perturbations to each biaxial stretching process parameter and running the machine learning model multiple times, we can simulate the fluctuations in process parameters caused by various factors during the actual production process. This helps evaluate the model's predictive performance under different perturbations and gain a more comprehensive understanding of the model's stability and reliability. The coefficient of variation of air permeability and filtration efficiency is calculated based on the performance indicators under different perturbations. The smaller the coefficient of variation, the less affected the performance indicator is by process parameter fluctuations, and the more stable the model's prediction results. Selecting process solutions with coefficients of variation below a first preset threshold ensures that the recommended process parameter combinations have high stability and reliability in actual production, reducing the risk of quality fluctuations during production. By calculating the coefficient of variation of air permeability and filtration efficiency, we quantitatively evaluate the performance stability of different process solutions. This not only considers the performance of the process solution under ideal conditions, but also comprehensively considers its performance stability under fluctuations that may be encountered in actual production, thereby selecting process solutions that better meet actual production needs.

[0080] Optionally, the method further includes:

[0081] The measured data of the samples prepared according to the optimal process parameter combination are analyzed for deviation from the predicted values. When the absolute error of the porosity or the error of the filtration efficiency exceeds the second preset threshold, the transfer learning strategy is used to freeze the underlying parameters related to the basic feature extraction in the machine learning model, and the preset fully connected layer is adjusted according to the measured data of the samples.

[0082] Deviation analysis between measured data and predicted values for samples prepared using the optimal process parameter combination is a method for evaluating the predictive performance of machine learning models. Comparing the differences between measured and predicted values provides insight into the accuracy and reliability of the model in real-world applications. Deviation analysis typically involves calculating metrics such as porosity absolute error and filtration efficiency error, which quantify the discrepancy between the model's predicted results and actual results. When the porosity absolute error or filtration efficiency error exceeds a second preset threshold, it indicates a significant deviation between the model's predicted results and actual results, necessitating model adjustment and optimization. In this case, a transfer learning strategy is employed to freeze the underlying parameters related to basic feature extraction in the machine learning model, and then adjust the pre-set fully connected layers based on the measured sample data. These underlying parameters are typically related to basic feature extraction, which is common across different tasks. Freezing these underlying parameters preserves the underlying feature representations learned by the model in the original task, preventing these features from being lost or significantly altered during retraining. Fully connected layers are typically used to map extracted features to specific outputs and may require specific adjustments for different tasks. Adjusting the preset fully connected layer according to the actual sample data can make the model better adapt to the current task and improve the model's prediction accuracy and generalization ability.

[0083] By performing deviation analysis on the measured data of samples prepared according to the optimal process parameter combination and the predicted values, the difference between the model prediction results and the actual situation can be discovered in a timely manner. When the absolute error of porosity or the error of filtration efficiency exceeds the second preset threshold, it means that the prediction accuracy of the model within the current process parameter range needs to be improved and the model needs to be adjusted. Using the transfer learning strategy, the underlying parameters related to basic feature extraction in the machine learning model are frozen, and the preset fully connected layer is adjusted according to the measured data of the sample. This can enable the model to better adapt to new data and process conditions while retaining the original feature extraction capability of the model. This helps to improve the prediction accuracy of the model within a specific process parameter range and enhance the adaptability of the model. Adjusting the preset fully connected layer according to the measured data of the sample can make the model better adapt to the data distribution within the specific process parameter range. The fully connected layer is usually used to map the extracted features to the final output result. By adjusting the parameters of the fully connected layer, the prediction results of the model within a specific process parameter range can be made more accurate, thereby improving the generalization ability of the model.

[0084] Optionally, the utilizing of a transfer learning strategy to freeze underlying parameters related to basic feature extraction in the machine learning model and adjusting a preset fully connected layer according to the sample measured data includes:

[0085] Freeze the weight matrix and convolution kernel parameters of the feature extraction layer in the initial machine learning model. When the loss function decrease rate of three consecutive rounds of training is less than a preset percentage, multiply the current learning rate of the fully connected layer by the attenuation coefficient.

[0086] In the initial machine learning model, the weight matrix and convolution kernel parameters of the feature extraction layer have already learned basic feature representations through extensive data training. These features are important for identifying common patterns and regularities in the biaxially oriented polytetrafluoroethylene fiber membrane preparation process. Freezing these parameters means that they will no longer be updated during subsequent transfer learning, thereby preserving the model's ability to extract basic features and preventing over-adjustment that could cause the model to lose its ability to recognize common features. The preset fully connected layer is the part of the model responsible for mapping features to output results (such as performance indicators such as porosity and filtration efficiency). In this step, the parameters of the fully connected layer are adjusted based on measured data from samples prepared using the optimal process parameter combination to better adapt to the new data distribution and process requirements, thereby improving the model's prediction accuracy within the specified process parameter range. The rate of decrease of the loss function (a metric that measures the difference between the model's predicted and actual values) is monitored over three consecutive training rounds. If the rate of decrease is less than a preset percentage, it indicates that continued training of the fully connected layer at the current learning rate is ineffective, possibly resulting in a local optimum or slow learning. When the rate of decrease in the loss function is detected to be too low, the current learning rate of the fully connected layer is multiplied by the decay coefficient. The learning rate is an important hyperparameter that controls the step size of the model parameter update. Lowering the learning rate can make the model's exploration of the parameter space more refined, helping the fully connected layer further optimize parameters and improve the model's performance and stability.

[0087] Example 1 of the present application: A polytetrafluoroethylene dispersion resin (number average molecular weight 20 million, crystallinity ≥97%) and an isoparaffin lubricant are mixed in a weight ratio of 1:0.25, and the aging condition is 50°C for 36 hours; a non-deoiled base belt with a width of 300 mm is formed through blanking (40°C, holding pressure for 5 minutes), push extrusion (50°C, compression ratio 150) and calendering treatment; an orthogonal experiment of biaxial stretching is designed, with a longitudinal stretching ratio of 1 to 16 times, a speed of 1 to 5 m / min, and a temperature of 120 to 330°C; a transverse expansion ratio of 3 to 25 times, a speed of 1 to 16 m / min, temperature 210-320℃; prepared 600 groups of PTFE fiber membrane samples, recorded the process parameters of each group, and tested their average pore size, porosity, air permeability and filtration efficiency; used the random forest algorithm, with input parameters including longitudinal / transverse stretching temperature, speed, multiple and heat setting temperature, and output targets such as pore size, porosity, air permeability and filtration efficiency. The data set was divided into training set and test set with a ratio of 7:3, and hyperparameters (such as tree depth and number of leaf nodes) were optimized through grid search; the RMSE (root mean square error) was The cross-validation model with an R² value of 1.8 and an R² value of 0.92 (R² is the coefficient of determination) has a strong predictive ability. The target performance is set as follows: porosity ≥ 85%, filtration efficiency ≥ 99.9%. The model reversely recommends the following process parameter combinations: longitudinal stretching temperature 210°C, speed 3m / min, magnification 8 times, transverse expansion temperature 280°C, speed 10m / min, magnification 15 times, and heat setting temperature 360°C. Samples were prepared according to the optimal parameters given by the model, and the measured porosity was 86.2% and the filtration efficiency was 99.95%, achieving the target performance.

[0088] Example 2 of the present application: A polytetrafluoroethylene dispersion resin (number average molecular weight of 20 million, crystallinity ≥97%) and an isoparaffin lubricant were mixed in a weight ratio of 1:0.25 and aged at 50°C for 36 hours. After forming (40°C, holding pressure for 5 minutes), pushing extrusion (50°C, compression ratio of 150) and calendering, a 300 mm wide undeoiled base tape was formed. An orthogonal experiment of biaxial stretching was designed, with a longitudinal stretching ratio of 1 to 16 times, a speed of 1 to 5 m / min, and a temperature of 120 to 330°C; a transverse stretching ratio of 3 to 25 times, a speed of 1 to 16 m / min, and a temperature of 210 to 320°C. 600 groups of PTFE fiber membrane samples were prepared, the process parameters of each group were recorded, and their average pore size, porosity, air permeability, and filtration efficiency were tested. Figure 2 This is a schematic diagram of the core machine learning algorithm structure disclosed in the embodiment of this application, such as Figure 2As shown, an artificial neural network algorithm is used to construct a multi-layer perceptron. The input layer is the longitudinal / transverse temperature, speed, and multiple (a total of 6 input nodes), the hidden layer contains 2 fully connected layers (the number of nodes is 8 and 10 respectively), and the output layer is the performance of pore size, porosity, air permeability and filtration efficiency (12 output nodes). The data set is divided into training set, validation set and test set according to the ratio of 7:2:1. The Adam optimizer is used, the initial learning rate is 0.001, the loss function is the mean square error (MSE), and the training cycle is 300,000 times. The prediction ability of the model is strong through loss curve diagnosis and RMSE of 1.5 and R²=0.984 cross-validation. The target performance is set as follows: porosity ≥88%, air permeability ≥25 L / (m²·s), and filtration efficiency ≥99.95%. The model reversely recommends the following process parameter combinations: longitudinal stretching temperature 230℃, speed 4 m / min, a magnification of 10 times, a lateral expansion temperature of 290°C, a speed of 12m / min, and a magnification of 16 times; the sample was prepared according to the optimal parameters given by the model, and the measured porosity was 89.3%, the air permeability was 26.8L / (m²·s), and the filtration efficiency was 99.98%.

[0089] Pore size test (average pore size, most probable pore size): Use PMI (gas permeation pore size analyzer) equipment for testing. Cut the sample to be tested into a circle or other shape with the diameter specified by the equipment. Note that the inside of the test sample cannot be cut. After the sample is soaked in the surfactant required by the PMI equipment test for 30 minutes, the sample is placed in the test slot of the instrument and the slot cover is tightened to conduct the test. The test result is directly read by the equipment and converted into the distribution of the proportion (%) of each pore size (μm) of the filter material to the entire material (including average pore size, maximum pore size μm, minimum pore size μm, pore size distribution). Sampling is performed at three locations (A, B, C) evenly along the transverse direction of the polytetrafluoroethylene film, and three points (1, 2, 3) are selected at each location.

[0090] Air Filtration Efficiency: Tested in accordance with GB 19083-2010, the sodium chloride (NaCl) aerosol particle size distribution specified in the test conditions should be a number median diameter (CMD) of 0.075 μm ± 0.020 μm, with a geometric standard deviation not exceeding 1.86 (equivalent to a mass median aerodynamic diameter (MMAD) of 0.24 μm ± 0.06 μm). The air flow rate is 85 L / min. Air filtration efficiency testing is performed using the same sampling points as for the aperture test above.

[0091] Air permeability: Based on GB / T5453-1997 standard, the air permeability of polytetrafluoroethylene fiber membrane was tested. The test point was a circle with an area of 100cm2, the pressure was 200Pa, the unit was m3 / m2*min, and 10 test points were evenly spaced in the width direction, and the average value was taken.

[0092] This embodiment also discloses a system for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane. Figure 3 Schematic diagram of a module of a system for predicting a process for preparing a biaxially oriented polytetrafluoroethylene fiber membrane disclosed in an embodiment of the present application. Figure 3 As shown, the system includes a data module 301, a model module 302, and a prediction module 303, wherein:

[0093] A data module 301 is configured to obtain a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment, wherein the data set includes a combination of biaxial stretching process parameters, pore structure data, and performance indicators for preparing a target group of fiber membrane samples, wherein the target group of fiber membrane samples is any group of the multiple groups of fiber membrane samples;

[0094] a model module 302 configured to construct an initial machine learning model, using the combination of biaxial stretching process parameters in the data set as input parameters and the pore structure data and performance indicators as output parameters, training the initial machine learning model to obtain an intermediate model, and adjusting the parameters of the intermediate model through cross-validation to obtain a machine learning model;

[0095] The prediction module 303 is configured to recommend an optimal process parameter combination and confidence interval based on the set target performance through the machine learning model.

[0096] Optionally, the model module 302 is further configured to:

[0097] Calculating a critical speed of polytetrafluoroethylene according to preset viscoelastic parameters, setting the critical speed between the input layer and the hidden layer, and triggering parameter correction when the input stretching speed exceeds the critical speed;

[0098] A porosity restriction function is set in the output layer. When the porosity predicted by the model exceeds the interval set by the porosity restriction function, the output result is corrected according to a preset adjustment strategy.

[0099] Optionally, the prediction module 303 is configured to:

[0100] Determining the range and step size of biaxial stretching process parameters, wherein the biaxial stretching process parameters include transverse stretching ratio, transverse speed, transverse temperature, longitudinal stretching ratio, longitudinal speed, and longitudinal temperature;

[0101] Determine a parameter space according to the range, and uniformly generate a plurality of parameter combinations in the parameter space using a Latin hypercube sampling method;

[0102] The crossover probability, mutation probability, elite retention ratio and crowding degree calculation method are set, and the optimal solution is obtained from the multiple parameter combinations through non-dominated sorting and crowding distance calculation.

[0103] Optionally, the prediction module 303 is configured to:

[0104] introducing random perturbations for each of the biaxial stretching process parameters, running the machine learning model multiple times, and obtaining performance index results under different perturbations;

[0105] The coefficients of variation of air permeability and filtration efficiency are calculated according to the performance indicator results, and a process solution is screened out in which the coefficients of variation of the air permeability and the filtration efficiency are both lower than a first preset threshold.

[0106] Optionally, the system further includes an adjustment module configured to:

[0107] The measured data of the samples prepared according to the optimal process parameter combination are analyzed for deviation from the predicted values. When the absolute error of the porosity or the error of the filtration efficiency exceeds the second preset threshold, the transfer learning strategy is used to freeze the underlying parameters related to the basic feature extraction in the machine learning model, and the preset fully connected layer is adjusted according to the measured data of the samples.

[0108] Optionally, the adjustment module is configured to:

[0109] Freeze the weight matrix and convolution kernel parameters of the feature extraction layer in the initial machine learning model. When the loss function decrease rate of three consecutive rounds of training is less than a preset percentage, multiply the current learning rate of the fully connected layer by the attenuation coefficient.

[0110] Optionally, the data module 301 is configured to:

[0111] The polytetrafluoroethylene dispersion resin and hydrocarbon lubricating oil are mixed in a preset weight ratio and aged at 30-70°C for 24-48 hours; the undeoiled base tape is formed by forming a blank, pushing, extruding and calendering.

[0112] The longitudinal stretching ratio is controlled to be 1 to 16 times, the longitudinal speed is 1 to 5 m / min, and the longitudinal temperature is 120 to 330° C., and the transverse expansion ratio is controlled to be 3 to 25 times, the transverse speed is 1 to 16 m / min, and the transverse temperature is 210 to 320° C. to generate multiple groups of fiber membrane samples;

[0113] The pore structures and properties of multiple groups of the fiber membrane samples are detected to obtain the data set.

[0114] It should be noted that the above embodiments provide devices that implement their functions using only the division of the above functional modules as examples. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0115] This embodiment also discloses an electronic device, referring to Figure 4 The electronic device may include: at least one processor 401 , at least one communication bus 402 , a user interface 403 , a network interface 404 , and at least one memory 405 .

[0116] The communication bus 402 is used to implement the connection and communication between these components.

[0117] The user interface 403 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0118] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0119] Processor 401 may include one or more processing cores. Using various interfaces and circuits, processor 401 connects to various components within the server. It executes instructions, programs, code sets, or instruction sets stored in memory 405, as well as accesses data stored in memory 405, to perform various server functions and process data. Optionally, processor 401 may be implemented using at least one of the following hardware forms: a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). Processor 401 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing content displayed on the display screen; and the modem handles wireless communications. It is understood that the modem may also be implemented independently of the processor 401 and implemented as a separate chip.

[0120] Among them, the memory 405 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). Optionally, the memory 405 includes a non-transitory computer-readable storage medium. The memory 405 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 405 may also be optionally at least one storage device located away from the aforementioned processor 401. As Figure 4 As shown, the memory 405 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program for a method for predicting a process for preparing a biaxially oriented polytetrafluoroethylene fiber membrane.

[0121] exist Figure 4 In the electronic device shown, the user interface 403 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 401 can be used to call the application program stored in the memory 405 for predicting the method for preparing the biaxially oriented polytetrafluoroethylene fiber membrane. When executed by one or more processors 401, the electronic device executes one or more methods as in the above-mentioned embodiments.

[0122] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application.

[0123] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely schematic, such as the division of units, which is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0125] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0126] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0127] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 405 and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of this application. The aforementioned memory 405 includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a magnetic disk, or an optical disk.

[0128] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification, those skilled in the art will easily think of other embodiments of the present disclosure. This application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the technical field that are not recorded in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane, characterized in that: Applied to the preparation platform, the method comprises: Obtain a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment, the data set including a combination of biaxial stretching process parameters, pore structure data, and performance indicators for preparing a target group of fiber membrane samples, the target group of fiber membrane samples being any one of the multiple groups of fiber membrane samples, the biaxial stretching process parameter combination including transverse stretching ratio, transverse speed, transverse temperature, longitudinal stretching ratio, longitudinal speed, and longitudinal temperature; constructing an initial machine learning model, using the combination of biaxial stretching process parameters in the data set as input parameters, and the pore structure data and performance indicators as output parameters, training the initial machine learning model to obtain an intermediate model, and adjusting the parameters of the intermediate model through cross-validation to obtain a machine learning model; According to the set target performance, the machine learning model is used to recommend the optimal process parameter combination and confidence interval. The construction of the initial machine learning model includes embedding a physical constraint mechanism, specifically including: Calculating a critical speed of polytetrafluoroethylene according to preset viscoelastic parameters, setting the critical speed between the input layer and the hidden layer, and triggering parameter correction when the input stretching speed exceeds the critical speed; A porosity restriction function is set in the output layer. When the porosity predicted by the model exceeds the interval set by the porosity restriction function, the output result is corrected according to a preset adjustment strategy.

2. The method for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane according to claim 1, characterized in that: The recommendation of the optimal process parameter combination and confidence interval by the machine learning model includes: Determining the range and step size of biaxial stretching process parameters, wherein the biaxial stretching process parameters include transverse stretching ratio, transverse speed, transverse temperature, longitudinal stretching ratio, longitudinal speed, and longitudinal temperature; Determine a parameter space according to the range, and uniformly generate a plurality of parameter combinations in the parameter space using a Latin hypercube sampling method; The crossover probability, mutation probability, elite retention ratio and crowding degree calculation method are set, and the optimal solution is obtained from the multiple parameter combinations through non-dominated sorting and crowding distance calculation.

3. The method for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane according to claim 2, characterized in that: The method of recommending the optimal process parameter combination and confidence interval by the machine learning model further includes: introducing random perturbations for each of the biaxial stretching process parameters, running the machine learning model multiple times, and obtaining performance index results under different perturbations; The coefficients of variation of air permeability and filtration efficiency are calculated according to the performance indicator results, and a process solution is screened out in which the coefficients of variation of the air permeability and the filtration efficiency are both lower than a first preset threshold.

4. The method for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane according to claim 1, characterized in that: The method further comprises: The measured data of the samples prepared according to the optimal process parameter combination are analyzed for deviation from the predicted values. When the absolute error of the porosity or the error of the filtration efficiency exceeds the second preset threshold, the transfer learning strategy is used to freeze the underlying parameters related to the basic feature extraction in the machine learning model, and the preset fully connected layer is adjusted according to the measured data of the samples.

5. The method for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane according to claim 4, characterized in that: The method of utilizing the transfer learning strategy to freeze the underlying parameters related to basic feature extraction in the machine learning model and adjusting the preset fully connected layer according to the sample measured data includes: Freeze the weight matrix and convolution kernel parameters of the feature extraction layer in the initial machine learning model. When the loss function decrease rate of three consecutive rounds of training is less than a preset percentage, multiply the current learning rate of the fully connected layer by the attenuation coefficient.

6. A system for predicting the preparation process of biaxially oriented polytetrafluoroethylene fiber membrane, characterized in that: It includes data module, model module and prediction module, among which: A data module is configured to obtain a data set of multiple groups of fiber membrane samples prepared by a preset orthogonal experiment, wherein the data set includes a combination of biaxial stretching process parameters, pore structure data, and performance indicators for preparing a target group of fiber membrane samples, wherein the target group of fiber membrane samples is any group of the multiple groups of fiber membrane samples, and the biaxial stretching process parameter combination includes a transverse stretching ratio, a transverse speed, a transverse temperature, a longitudinal stretching ratio, a longitudinal speed, and a longitudinal temperature; a model module configured to construct an initial machine learning model, using the combination of biaxial stretching process parameters in the data set as input parameters, using the pore structure data and performance indicators as output parameters, training the initial machine learning model to obtain an intermediate model, and adjusting the parameters of the intermediate model through cross-validation to obtain a machine learning model; The prediction module is configured to recommend an optimal process parameter combination and a confidence interval based on the set target performance through the machine learning model.

7. An electronic device, characterized in that: It includes a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 5 is executed.

Images