CN-121999927-A - Deep learning-based polylactic acid fiber membrane hydrophobic performance technological parameter optimization method

CN121999927ACN 121999927 ACN121999927 ACN 121999927ACN-121999927-A

Abstract

The invention discloses a deep learning-based polylactic acid fiber membrane hydrophobic performance process parameter optimization method, which belongs to the technical field of material science and engineering, wherein a 5-layer neural network model is constructed on the basis of acquired process parameters and contact angle data, each layer of network adopts a fully connected network structure, a ReLU function is used for nonlinear processing after linear transformation so as to improve the fitting capacity of the model to a nonlinear function, and a batch gradient descent algorithm is adopted to optimize model parameters on the basis of constructing the model, so that the model can fit the mapping relation between the polylactic acid process parameter input value and the contact angle hydrophobic performance. The trained neural network model is utilized to evaluate the selected various process parameter combinations within the process parameter value range, so that the process flow of adjusting and optimizing the process parameters of the polylactic acid fiber membrane is simplified, the experiment times and cost in the process of adjusting and optimizing the process parameters are reduced, and various polylactic acid products meeting the hydrophobicity requirement can be produced.

Inventors

LIU XIN
Gan Genxin

Assignees

江苏海洋大学

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (6)

1. The deep learning-based polylactic acid fiber membrane hydrophobic performance technological parameter optimization method is characterized by comprising the following specific steps of: s1, acquiring hydrophobic performance data of a polylactic acid fiber membrane, and measuring contact angle data under current process parameters on electrostatic spinning equipment by using uniformly distributed sampling process parameter data; S2, constructing a 5-layer large-width fully-connected neural network model to fit the relation between the technological parameters and the hydrophobicity of the polylactic acid fiber membrane, wherein the dimension of an input layer of the model is the characteristic quantity of the technological parameters, and an output layer of the model is output as a contact angle value; S3, inputting the acquired process parameters into a model, outputting the model as a contact angle predicted value, evaluating the predicted error of the model by using a mean square error function, and optimizing the model parameters by using a gradient descent algorithm according to the model predicted error so that the mean square error function reaches the minimum value on a sample; S4, uniformly sampling the fine particles in the process parameter value range of the electrostatic spinning equipment, inputting the sampling value into a trained neural network model to predict the contact angle, and obtaining the process parameters with different hydrophobic performance requirements.
2. The deep learning-based polylactic acid fiber membrane hydrophobic property process parameter optimization method according to claim 1, wherein the polylactic acid fiber membrane process parameter and contact angle data acquisition in S1 further comprises: S1-1, constructing a variety of polylactic acid fiber membrane process parameters and contact angle data sets, adopting a plurality of rounds of random value taking from each process parameter value range sequentially by using uniform distribution according to the sequence of inputting data by a neural network in the process parameter value ranges of spinning temperature, material pushing rate, electric field voltage, receiving distance and the like, forming an input value of electrostatic spinning equipment by using four process parameter random values in each round, then preparing the polylactic acid fiber membrane by using the equipment, dripping water on the polylactic acid fiber membrane to measure the contact angle, detecting the hydrophobicity of the polylactic acid fiber membrane, and obtaining a polylactic acid fiber membrane process parameter and contact angle corresponding relation sample.
3. The deep learning-based polylactic acid fiber membrane hydrophobic property process parameter optimization method according to claim 1, wherein the deep learning-based polylactic acid fiber membrane hydrophobic property process parameter optimization model in S2 further comprises: s2-1, constructing a polylactic acid fiber membrane hydrophobic performance optimization model based on deep learning, respectively calculating the mean mu and the variance sigma of different parameters for a polylactic acid fiber membrane process parameter sample, and completing the normalization operation of input parameters by using the following formula: A relation between technological parameters and contact angles of a polylactic acid fiber membrane is fitted by using a multi-layer large-width neural network, a 5-layer fully-connected neural network model is constructed, the model comprises 1 input layer, 3 hidden layers and 1 output layer, wherein the input layer comprises 4 neurons which are respectively used for inputting four parameters of spinning temperature, pushing rate, electric field voltage and receiving distance, each hidden layer comprises 60 neurons, the output of each hidden layer is nonlinear by using Relu functions, the output of the last hidden layer is linearly transformed to one neuron of the output layer, and the output value is a predicted value of the contact angle of the model under the technological parameters of the current polylactic acid fiber membrane.
4. The deep learning-based polylactic acid fiber membrane hydrophobic property process parameter optimization method according to claim 1, wherein optimizing the prediction model parameters in S3 using a mean square error function and a small batch random gradient algorithm comprises: S3-1, inputting the normalized process parameters such as spinning temperature, pushing rate, electric field voltage, receiving distance and the like into a deep learning network model to obtain a contact angle predicted value, and calculating model estimation errors according to the contact angle predicted value and a measured contact angle true value by the following formula: where n is the number of samples per batch of the training model, And Setting the contact angle model predictive value and the measured true value error for the related process parameter value; And S3-2, in the model training process, a small batch of random gradient algorithm is used for optimizing model parameters, in each round of training process of the model, a plurality of samples are randomly extracted to form a training batch of the training batch, training sample data of the training batch are input into a deep learning network model, the error between the contact angle predicted value and the true value of the training batch is calculated by using a mean square error function, the gradient of the mean square error function relative to the model parameters is calculated by using a chain rule derived by a hidden function, and the model parameters are updated by using the gradient of the model parameters and the learning rate, so that the mean square error function of the model on the whole data set is minimum.
5. The deep learning-based polylactic acid fiber membrane hydrophobic property process parameter optimization method according to claim 1, wherein the technique for preparing the polylactic acid fiber membranes with different hydrophobic properties by using the trained deep learning network model in S4 comprises the following steps: S4-1, optimizing the technological parameters of the polylactic acid fiber membrane by using a trained deep learning network model, dividing the value ranges n of the technological parameters such as spinning temperature, pushing rate, electric field voltage, receiving distance and the like in equal parts, using the equal parts to carry out full combination, traversing each sample of the full combination, inputting the sample into the deep learning network model to calculate the corresponding contact angle, judging the hydrophobicity of the polylactic acid fiber membrane under the technological parameters according to the contact angle, and obtaining technological parameter values with different hydrophobic properties of the polylactic acid fiber membrane so as to meet different requirements of different products on the hydrophobic properties of the polylactic acid fiber membrane made of high polymer materials.
6. A method for optimizing the hydrophobic performance technological parameters of polylactic acid fiber membrane based on deep learning is characterized in that a deep learning model is used for fitting the relation between the technological parameters of polylactic acid fiber membrane such as spinning temperature, pushing rate, electric field voltage, receiving distance and the like and the hydrophobic performance contact angle, and different technological parameter values which need to be set are reversely pushed according to different requirements of the hydrophobic performance of the polylactic acid fiber membrane made of high polymer materials by utilizing the learned relation model.

Description

Deep learning-based polylactic acid fiber membrane hydrophobic performance technological parameter optimization method Technical Field The invention relates to the technical field of high polymer materials, in particular to a deep learning-based optimization method for the hydrophobic performance technological parameters of a polylactic acid fiber membrane. Background Polylactic acid (PLA) is thermoplastic aliphatic polyester from renewable biological resources such as corn, cassava and the like, has the advantages of complete biodegradability, good biocompatibility, processability and the like, and is one of important environment-friendly materials for replacing petroleum plastics and relieving white pollution. The electrostatic spinning can process PLA into a nanofiber membrane with large surface area, multiple pore structures and excellent controllable fiber morphology. The structural characteristics enable the PLA nanofiber membrane to have wide application space in the fields of drug controlled release, tissue engineering brackets, high-efficiency air filtering materials, oil-water emulsion separation membranes and the like. In the application process of the fields, the water wettability, namely the hydrophobicity, of the surface of the material is a key characteristic affecting the function and stability of the material, and the static contact angle of water on the surface of the material is often adopted for measurement. The contact angle is related to the diffusion and permeation of liquid drops on the surface of the membrane, and the pollution resistance, the separation selectivity and the long-term usability of the membrane, so that the research on the hydrophobic property of the PLA fiber membrane has important significance for improving the material property. However, there are many difficulties in producing PLA fiber films meeting the hydrophobic requirements by using electrospinning, which is an important means for preparing polylactic acid fiber films, and is a complex process involving multiple physical field strength coupling such as electric field mechanics, jet dynamics, polymer solution rheology, and solvent evaporation phase change. The morphology, diameter distribution, porosity and surface chemistry of the final fiber film affect its hydrophobic properties, which are affected by a range of process parameters. The key parameters include the temperature and humidity of the spinning environment, the advancing rate of the polymer solution, the high-voltage direct-current voltage applied between the spinning needle and the receiving device, the distance between the needle and the receiver, and the like, and the parameters are strongly coupled and nonlinear, so that the design and preparation of the fiber with any specification have certain difficulty, the secondary structure of the surface of the fiber is controlled finely, and the technological parameters are required to be regulated finely. The existing control of the hydrophobic property of the polylactic acid fiber membrane is mainly to study the influence of different factors on the fiber morphology and the different hydrophobic property of the fiber membrane with different morphological characteristics by setting different gradients on control variables such as the relative molecular weight of the polymer, the concentration of the solution, the type and the proportion of the solvent, the type of the polymer, the humidity and the like. The method adopts an experimental method to analyze the influence of single process parameter factors on the morphology and the hydrophobicity of the fiber membrane, but the method lacks the analysis of the multi-process parameter combination on the hydrophobicity of the fiber membrane, does not establish the number mapping relation between the multi-parameter combination and the hydrophobicity, has large multi-process parameter combination value space of the polylactic acid fiber membrane, and has difficulty in optimizing the process parameter combination. Disclosure of Invention In order to solve the technical problems mentioned in the background art, the invention provides a polylactic acid fiber membrane hydrophobic performance technological parameter optimization method based on deep learning, which adopts the following technical scheme: S1, collecting technological parameters and corresponding contact angle data of a polylactic acid fiber membrane, respectively collecting various technological parameter combinations according to uniform distribution in the range of technological parameter values such as spinning temperature, material pushing rate, electric field voltage, receiving distance and the like, setting each technological parameter on electrostatic spinning experimental equipment, measuring the contact angle data of the prepared polylactic acid fiber membrane, and collecting multiple groups of measurement data which are training data of a deep learning model; s2, constructing a deep learni