A Deep Learning-Based Method for Predicting the Processing Effect of Pure Cotton Bubble Fabric

Abstract

This invention relates to the field of intelligent textile technology, specifically to a method for predicting the treatment effect of pure cotton bubble fabric based on deep learning. The steps include: collecting multi-physics coupling data, including temperature field distribution, tension and pressure, fluid diffusion rate, and fluid viscosity data during fabric treatment; combining the temperature field distribution with the tension and pressure data to construct a temperature-pressure model, which monitors temperature and tension and pressure in real time, collects multiple sets of data, and fuses them to obtain a temperature-pressure feature vector; fusing the fluid diffusion rate and fluid viscosity data with the temperature-pressure feature vector to construct a joint model, which performs real-time prediction of the data, and constructs a treatment effect prediction function based on the real-time prediction data; constructing an optimization algorithm based on the obtained treatment effect prediction function, uploading it to an interactive prediction optimization platform, and dynamically adjusting the joint model through interactive sharing based on different needs and set parameters.

Description

Technical Field

[0001] This invention proposes a deep learning-based method for predicting the processing effect of pure cotton bubble fabric, which relates to the field of intelligent textile technology. Background Technology

[0002] In the textile industry, it is quite difficult to predict the processing effects of new materials such as pure cotton bubble fabric. The nonlinear mapping relationship between its process parameters and effects is more complex, and traditional models are prone to overfitting due to insufficient data, making it difficult to meet industrial needs in terms of prediction stability.

[0003] Current intelligent cotton blending models driven by deep learning require more than two years of production data for training. However, bubble fabric, as an emerging product category, has scarce related data, and directly transferring the training model can easily lead to dimensionality mismatch. Existing models are mostly designed for steady-state production environments, while the effect of bubble fabric is sensitive to instantaneous changes in parameters such as pressure and temperature, requiring the development of a prediction architecture with time resolution. Combining various physical data of pure cotton bubble fabric and processing them in multiple ways can better meet production needs and improve production efficiency.

[0004] To address this, a deep learning-based method for predicting the processing effect of pure cotton bubble fabric is proposed. Summary of the Invention

[0005] The purpose of this invention is to propose a deep learning-based method for predicting the treatment effect of pure cotton bubble fabric. The specific implementation steps include: collecting multi-physics coupled data, including temperature field distribution, tension and pressure, fluid diffusion rate, and fluid viscosity data during fabric treatment; combining the temperature field distribution with the tension and pressure data to construct a temperature-pressure model, which monitors temperature and tension / pressure in real time, collects multiple sets of data, and fuses them to obtain a temperature-pressure feature vector; fusing the fluid diffusion rate and fluid viscosity data with the temperature-pressure feature vector to construct a joint model, which performs real-time prediction of the data, and constructs a treatment effect prediction function based on the real-time prediction data; constructing an optimization algorithm based on the obtained treatment effect prediction function, uploading it to an interactive prediction optimization platform, and dynamically adjusting the joint model according to different needs and set parameters through interactive sharing.

[0006] A deep learning-based method for predicting the processing effect of pure cotton bubble fabric includes:

[0007] Data involving multi-physics coupling is collected, including temperature field distribution, tension and pressure, fluid diffusion rate, and fluid viscosity data of the fabric during processing. The temperature field distribution is combined with the tension and pressure data to construct a temperature-pressure model, which is used for real-time monitoring of temperature and tension / pressure. Multiple sets of data are collected and fused to obtain a temperature-pressure feature vector. The fluid diffusion rate and fluid viscosity data are fused with the temperature-pressure feature vector to construct a joint model, which is used for real-time prediction of pure cotton bubble fabric data. A processing effect prediction function is constructed based on the real-time prediction data. An optimization algorithm is constructed based on the obtained processing effect prediction function, uploaded to an interactive prediction and optimization platform, and the joint model is dynamically adjusted through interactive sharing according to different needs and set parameters.

[0008] Preferably, the process of collecting process parameter data involving multi-physics coupling and corresponding processing effect data, and performing preprocessing, includes:

[0009] The collected multi-physics coupled data were cleaned and normalized to obtain local detail features and global trend features at different scales. The data of different physical modes were aligned at multiple scales on the time axis and spatial axis.

[0010] Preferably, the process of constructing a joint model by extracting features from the process parameter data and the treatment effect data, and combining the temperature field distribution with the applied tension and pressure data to construct a temperature-pressure model includes:

[0011] The temperature-pressure model employs an improved hybrid neural network structure combining a long short-term memory network (LSTM) and a convolutional neural network (CNN). The CNN extracts spatial features of the fabric, including the spatial distribution of the microstructure of the pure cotton bubble fabric, the arrangement of fibers, and the spatial layout of bubbles within the fabric. The LSTM network extracts dynamic temperature and pressure changes over time, extracting key features from the data to construct the temperature-pressure model. Within the temperature-pressure model and between feature interaction layer modules, skip connections are introduced to form multiple interconnected layers, preventing gradient vanishing during deep network training and enabling the handling of the dynamic evolution of bubble size and distribution caused by changes in pressure and temperature parameters over time during the processing of pure cotton bubble fabric.

[0012] Preferably, the temperature-pressure model monitors temperature and applied tension and pressure in real time based on data, and the process of collecting multiple sets of data and fusing them to obtain a temperature-pressure feature vector includes:

[0013] Based on the constructed temperature and pressure model, key features and interrelationships in the data are automatically identified and weighted and fused. The weight coefficients are determined based on the correlation between each data point and the processing effect, generating a temperature and pressure feature vector. The temperature and pressure feature vector contains information from multiple dimensions such as temperature gradient, pressure intensity, and tension distribution.

[0014] Preferably, the process of fusing fluid diffusion rate and fluid viscosity data with temperature and pressure feature vectors to construct a joint model, wherein the joint model is capable of real-time data prediction, and the process of constructing a processing effect prediction function based on the real-time predicted data includes:

[0015] Based on the results of fusing fluid diffusion rate and fluid viscosity data with temperature and pressure feature vectors, a joint model is constructed, capable of real-time prediction and dynamic adjustment. An incremental learning mechanism is introduced into the model, enabling it to automatically update internal parameters with the input of new data without retraining the entire model. The new data comes from real-time monitoring equipment on the production line and new process data. When new pure cotton bubble fabric processing data is input, the model automatically adjusts its internal weights and parameters, records each set of data at fixed intervals, and fuses each set of data to construct a processing effect prediction function.

[0016] Preferably, the process of constructing the optimization algorithm based on the constructed processing effect prediction function includes:

[0017] During model training, a processing effect prediction function is defined by combining mean squared error and physical constraint terms. An adaptive regularization term is added, and the model parameters are optimized using an optimization algorithm. As training progresses, the weight of the physical constraint terms is gradually increased. The physical constraint terms are based on the physical laws in the processing of pure cotton bubble fabric, including the relationship between the fabric volume and various data during bubble formation.

[0018] Preferably, the process of uploading to an interactive prediction and optimization platform and dynamically adjusting the joint model based on different needs and set parameters includes:

[0019] The prediction results are displayed in real time through a visual interface. Based on the results, the detection method of pure cotton bubble fabric can be controlled in real time, the process parameter settings for collecting pure cotton bubble fabric can be dynamically adjusted and fed back to the joint model, and multiple users can use and collaborate online at the same time. They can access this interactive prediction platform from their respective work terminals to conduct predictive analysis and discussion on the same pure cotton bubble fabric project.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. This invention uses an improved dual-neural network structure to construct a temperature-pressure model, obtaining temperature-pressure feature vectors. In the scenario of processing pure cotton bubble fabric, the collected process parameter data such as pressure, temperature, and fluid, as well as processing effect data such as bubble size and fabric strength, come from different physical modes. By aligning and integrating these feature vectors on the time and space axes, skip connections are introduced within the two neural networks and between feature interaction layer modules to form multi-layer connections, preventing gradient vanishing during training. The temperature-pressure model is then used for fusion processing to uncover the deep information hidden behind the complex data, deriving temperature-pressure feature vectors. This lays the data foundation for subsequent construction of a joint model, making production decisions more scientific.

[0022] 2. This invention obtains a joint model through temperature and pressure feature vectors. By fusing temperature, pressure, and fluid data, the joint model can comprehensively characterize various physical phenomena and effects during the processing of pure cotton bubble fabric. The fusion process considers the correlation and weight between different features, enabling the joint model to respond to changes in the processing in real time. This provides a rich information foundation for the model, allowing it to more accurately identify process parameters that significantly affect the processing effect, thereby improving the accuracy of predictions.

[0023] 3. This invention introduces an effect prediction function and constructs an optimization algorithm during model training. Combining mean squared error and physical constraint terms, an adaptive regularization term is added. The mean squared error effectively measures the deviation between the predicted and actual values, while the physical constraint term ensures that the prediction conforms to the physical laws governing the processing of pure cotton bubble fabric. Based on this function, the model parameters are dynamically adjusted and optimized, enabling the model to accurately learn data features during the training phase. In practical applications, the model can be dynamically optimized based on the real-time validated effect prediction function, promptly correcting deviations and ensuring the production process remains at its optimal state, achieving efficient and energy-saving production goals and reducing production costs. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a deep learning-based method for predicting the processing effect of pure cotton bubble fabric, as proposed in an embodiment of this invention application.

[0025] Figure 2 This is a flowchart illustrating the process of obtaining the joint model and the processing effect prediction function as proposed in an embodiment of this invention.

[0026] Figure 3 This is a flowchart illustrating the construction of a temperature-pressure model according to an embodiment of this invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] In the textile industry, predicting the treatment effects on novel materials such as pure cotton bubble fabric is quite challenging. The nonlinear mapping relationship between process parameters and effects is more complex, and traditional models are prone to overfitting due to insufficient data, making it difficult to meet industrial requirements in terms of predictive stability. Therefore, it is particularly important to invent an efficient and adaptive method for treating pure cotton bubble fabric.

[0029] This invention proposes a deep learning-based method for predicting the processing effect of pure cotton bubble fabric, belonging to the field of intelligent textile technology. It is suitable for predicting the processing effect of pure cotton bubble fabric, providing more comprehensive and specific prediction results, reducing production costs, and ensuring the production process is always in optimal condition. To illustrate the effectiveness of the method and system of this invention, detailed descriptions will be provided in conjunction with the accompanying drawings and the following embodiments.

[0030] Example 1

[0031] Factory A is a manufacturer of pure cotton bubble fabric. They want to predict the fabric treatment effect in real time during production to adjust process parameters promptly and improve product quality. This invention introduces a deep learning-based method for predicting the treatment effect of pure cotton bubble fabric, as shown in the flowchart below. Figure 1 As shown.

[0032] This invention proposes a deep learning-based method for predicting the treatment effect of pure cotton bubble fabric, belonging to the field of intelligent textile technology. The specific implementation steps include: collecting multi-physics coupling data, including temperature field distribution, tension and pressure, fluid diffusion rate, and fluid viscosity data of the fabric during treatment; combining the temperature field distribution with the tension and pressure data to construct a temperature-pressure model, which is used for real-time monitoring of temperature and tension and pressure, collecting multiple sets of data and fusing them to obtain a temperature-pressure feature vector; fusing the fluid diffusion rate and fluid viscosity data with the temperature-pressure feature vector to construct a joint model, which is used for real-time prediction of pure cotton bubble fabric data, and constructing a treatment effect prediction function based on the real-time prediction data; constructing an optimization algorithm based on the obtained treatment effect prediction function, uploading it to an interactive prediction optimization platform, and dynamically adjusting the joint model through interactive sharing according to different needs and set parameters.

[0033] Furthermore, data on pure cotton bubble fabric is collected. Corresponding to the steps above, the specific implementation includes:

[0034] Before acquiring data, various high-precision sensors, including pressure, temperature, and humidity sensors, need to be installed at key locations on the production line to monitor and record parameter data in real time, collecting data on temperature field distribution, applied tension and pressure, fluid diffusion rate, and fluid viscosity. Simultaneously, image acquisition equipment, including high-speed cameras, is used to capture data on bubble size, distribution, fabric strength, fabric elasticity, and bubble stability during the processing of pure cotton bubble fabric. The resolution and frame rate of these high-precision devices clearly capture changes in the fabric's microstructure and processing effects. The acquired data is synchronized through a data acquisition system to ensure a one-to-one correspondence between process parameter data and processing effect data over time, and is stored in real time in a local or cloud database. The acquired process parameter and processing effect data are first cleaned to remove noise and outliers; then, the cleaned data is normalized to unify data of different dimensions and magnitudes into a relatively standard range to facilitate subsequent analysis and modeling.

[0035] This application embodiment acquires various process parameter data of pure cotton bubble fabric through high-precision instruments, which can obtain various process parameters in the production process in real time and accurately. This provides a solid data foundation for subsequent analysis and prediction. Accurate data helps to detect abnormalities in production in a timely manner, realize refined control of the production process, and provide supervision information for subsequent data processing and analysis, as well as for training deep learning models.

[0036] Furthermore, a joint model is constructed to extract features from the process parameter data and the treatment effect data to obtain feature vectors. The construction process, corresponding to the steps described above, is as follows: Figure 3 The specific implementation process includes:

[0037] A joint model is constructed, combining LSTM and CNN neural network structures and introducing skip connections to form a multi-layered interconnected structure. The CNN part focuses on extracting spatial features from the data, deeply analyzing the microstructure of pure cotton bubble fabric, accurately capturing key spatial information such as fiber arrangement and the spatial distribution of bubbles within the fabric. The LSTM part focuses on extracting dynamic features from the time series, accurately handling the evolution of bubble size and distribution caused by subtle changes in key process parameters such as pressure and temperature over time during the complex processing of pure cotton bubble fabric. It fully grasps the patterns in these dynamic changes. An attention mechanism module is embedded into the neural network structure of the temperature and pressure model to identify and focus on key features, automatically assigning attention weights. When the CNN extracts spatial features, the attention mechanism highlights densely populated areas of bubbles or areas with unique fiber arrangements, focusing on whether bubbles are evenly distributed and whether there are bubble clusters. This allows the model to extract more representative and discriminative features, providing higher-quality input for subsequent prediction tasks and improving the model's prediction accuracy for the processing effect of pure cotton bubble fabric. Feature selection was performed, and the data was decomposed and reconstructed at multiple levels to obtain feature representations at different levels. Key features closely related to the processing effect of pure cotton bubble fabric were screened out, and redundant information and noise were removed. Potential information in the data was mined from different angles and levels to more comprehensively characterize the complex features of the pure cotton bubble fabric processing process. Finally, data from different physical modalities were aligned and integrated on the time and spatial axes to ensure consistency and correlation between the data from each modality, providing high-quality temperature and pressure feature vectors for subsequent joint model construction.

[0038] This application's embodiments, through data cleaning and combining Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), fully leverage the advantages of both network structures. The introduction of skip connections prevents the vanishing gradient problem during deep network training, allowing gradients to propagate more smoothly backpropagate, thus contributing to model convergence and training stability. Simultaneously, feature vectors are extracted and categorized into three main types: temporal features, frequency domain features, and spatial features. This process fully considers the complex characteristics of multi-physics coupling during the processing of pure cotton bubble fabric, further improving the accuracy and stability of the model's prediction of processing effects. This provides a solid data foundation and technical support for optimizing production processes and improving product quality.

[0039] Furthermore, based on the joint model, the feature vectors are fused into a multi-dimensional feature vector. Corresponding to the above steps, the specific process includes:

[0040] Based on the temperature gradient, pressure intensity, and tension extracted from the temperature-pressure model, normalization is performed to eliminate the influence of numerical differences and ensure a uniform numerical range. These values ​​are then weighted and fused using weights of 0.4, 0.3, and 0.3 to obtain a temperature-pressure feature vector. The weight coefficients are determined based on the correlation between each feature vector and the processing effect. When processing a new fabric, the model transfers previously learned weight adjustment experience as prior knowledge and inputs the temperature-pressure feature vector into the database. This allows the weight optimization mechanism to quickly adapt to plain, twill, and satin cotton bubble fabrics and new production conditions. Only a small amount of new data is needed to adapt to new scenarios, greatly improving the model's versatility and rapid deployment capability.

[0041] This application embodiment fuses feature vectors into a multi-dimensional feature vector. By weighting and fusing feature vectors from different physical modes, it fully considers the varying degrees of influence of different process parameters and processing effect data on the prediction target. This allows the fused multi-dimensional feature vector to more comprehensively and accurately reflect the complex relationships in the processing of pure cotton bubble fabric, improving the model's data utilization efficiency. It enables the model to simultaneously capture spatial and temporal features, providing richer information for the deep learning model, enhancing its learning ability and prediction accuracy on complex data, and improving the generalization ability of the adaptive prediction model to processing effects, thus maintaining stable and reliable prediction performance under different production conditions and data changes.

[0042] Furthermore, based on the input requirements and set parameters, the model makes predictions in real time and dynamically adjusts based on feedback. Upon inputting specific process parameters, the model immediately initiates the prediction process without waiting for all data to be collected. The model has a built-in feedback mechanism that feeds back actual processing effect data. Upon receiving this feedback, the model corrects its prediction results in real time. Modifications to input parameters during the prediction process allow the model to respond instantly to these changes and update its predictions. This dynamic adjustment mechanism continuously optimizes input parameters during the prediction process to achieve more ideal prediction results. The model possesses adaptive learning capabilities, automatically updating its internal parameters as new data is input. New data comes from real-time monitoring equipment on the production line or from new process data. When new data on pure cotton bubble fabric processing is input, the model automatically adjusts its internal weights and parameters to adapt to changes in data distribution. This allows the model to maintain high prediction accuracy and stability over long-term use, eliminating the need for frequent manual intervention to update the model, reducing maintenance costs, and improving production efficiency.

[0043] This application's embodiments obtain an adaptive prediction model by fusing data. This adaptive prediction model possesses adaptive learning capabilities, enabling it to automatically adapt to changes in data characteristics and production processes. When a change in data distribution or production process is detected, the model automatically triggers a retraining process without frequent manual intervention. This adaptive capability ensures that the model maintains high prediction accuracy and stability over long-term use, reduces model maintenance costs, and minimizes production risks caused by model obsolescence.

[0044] During model training, the generated performance prediction function is defined by combining the mean squared error (MSE) with physical constraints. The process of constructing the performance prediction function is as follows: Figure 2 As shown. Specifically, MSE is used to measure the difference between the model's predicted value and the true value, while the physical constraint term is based on the physical laws of the pure cotton bubble fabric processing process to ensure that the model output conforms to the physical laws. During model training, a gradient descent-based optimization algorithm is used to optimize the model parameters, including the model's generalization ability, training time, and number of iterations. For example, the mean squared error and mean absolute error of the model on the test set are calculated. The optimization algorithm calculates the gradient of the loss function with respect to the model parameters and updates the model parameters according to the gradient to minimize the effect prediction function, thereby gradually approaching the optimal solution. During training, the model is validated in real time based on the value of the effect prediction function, and the model parameters are adaptively and dynamically adjusted and optimized according to the validation results to improve the model's predictive performance.

[0045] This application's embodiments, by generating and optimizing the effect prediction function, combining the mean squared error with physical constraint terms, and using optimization algorithms, ensure that the model can accurately predict the processing effect of pure cotton bubble fabric. By introducing attention mechanisms and data drift detection mechanisms, the model can not only improve prediction accuracy but also adapt to new data environments, enhancing its robustness and generalization ability.

[0046] The process of building an interactive prediction platform begins with establishing its basic architecture, which typically consists of a front-end and a back-end. The front-end is primarily responsible for the user interface display and interaction, while the back-end handles business logic, data storage, and model computation. In front-end development, a visual interface is built to display the predicted effects of processing pure cotton bubble fabric in real time through charts, graphs, and other formats. The back-end receives this data and transmits it to an adaptive prediction model. The adaptive prediction model, based on deep learning algorithms, analyzes and calculates the input data, outputting the corresponding predicted processing effects. When a user inputs the process parameters to be predicted into the front-end interface, the back-end receives this data and transmits it to the adaptive prediction model. The adaptive prediction model, based on deep learning algorithms, analyzes and calculates the input data, outputting the corresponding predicted processing effects. During operation, the model continuously monitors whether data characteristics and production processes have changed. This is achieved through a built-in data drift detection mechanism. The data drift detection mechanism is based on a distance metric and divides the data stream into multiple windows in chronological order. With 100 windows set, 20 samples are slid in at a time to extract data windows sequentially. When the data drift detection mechanism detects that the distribution difference between the new data and the training data exceeds the set significance level (p-value less than 0.05), it is considered that data drift has occurred, and the model will automatically trigger a retraining process to ensure the accuracy and adaptability of the prediction results.

[0047] This application embodiment constructs an interactive prediction platform, enabling data visualization and real-time control of the detection method and adjustment of process parameters for pure cotton bubble fabric based on prediction results. This interactive prediction platform, as an intelligent production aid, closely integrates deep learning technology with the pure cotton bubble fabric production process, injecting intelligent elements into enterprise production. Through real-time prediction, dynamic adjustment, and automatic optimization functions, it achieves refined control and intelligent management of the production process, improving the enterprise's level of production intelligence.

[0048] This invention proposes a deep learning-based method for predicting the treatment effect of pure cotton bubble fabric, belonging to the field of intelligent textile technology. The specific implementation steps include: collecting multi-physics coupling data, including temperature field distribution, tension and pressure, fluid diffusion rate, and fluid viscosity data of the fabric during treatment; combining the temperature field distribution with the tension and pressure data to construct a temperature-pressure model, which is used for real-time monitoring of temperature and tension and pressure, collecting multiple sets of data and fusing them to obtain a temperature-pressure feature vector; fusing the fluid diffusion rate and fluid viscosity data with the temperature-pressure feature vector to construct a joint model, which is used for real-time prediction of pure cotton bubble fabric data, and constructing a treatment effect prediction function based on the real-time prediction data; constructing an optimization algorithm based on the obtained treatment effect prediction function, uploading it to an interactive prediction optimization platform, and dynamically adjusting the joint model through interactive sharing according to different needs and set parameters.

[0049] This invention proposes a temperature-pressure model, a joint model, an effect prediction function, and an optimization algorithm for the above process. The temperature-pressure model is composed of an improved dual neural network structure, which obtains temperature-pressure feature vectors by aligning feature vectors of each data point. The effect prediction function provides a dynamic optimization model and an index for timely correction of deviations. The adaptive dynamic adjustment learning ability enables the model to automatically adjust its learning ability, ensuring long-term effectiveness, reducing maintenance costs, and improving production efficiency.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0051] Example 2

[0052] Company B is a research and development company. When developing a new type of pure cotton bubble fabric, it needs to predict the processing effects under different process parameters in order to quickly select the optimal combination of process parameters and shorten the development cycle. The process involves collecting multi-physics coupling data, including temperature field distribution, tension and pressure, fluid diffusion rate, and fluid viscosity data during fabric processing. The temperature field distribution is combined with the tension and pressure data to construct a temperature-pressure model. This model is used to monitor temperature and tension / pressure in real time, collecting multiple sets of data and fusing them to obtain a temperature-pressure feature vector. The fluid diffusion rate and fluid viscosity data are then fused with the temperature-pressure feature vector to construct a joint model. This joint model is used to predict the pure cotton bubble fabric data in real time. A processing effect prediction function is constructed based on the real-time prediction data. An optimization algorithm is built based on the obtained processing effect prediction function, uploaded to an interactive prediction and optimization platform, and the joint model is dynamically adjusted through interactive sharing according to different needs and set parameters.

[0053] Furthermore, data on pure cotton bubble fabric were collected, and the production process of pure cotton bubble fabric was simulated in the laboratory, using infrared detection data in particular. Different process parameters were set, including pressure, temperature, humidity, processing time, fluid flow rate, and fiber stress. Detailed tests were conducted on the pure cotton bubble fabric produced under each set of process parameters, and data on the treatment effect were collected, including bubble size, bubble distribution, fabric strength, fabric elasticity, and bubble stability, followed by pretreatment.

[0054] Furthermore, a temperature and pressure model is constructed to extract features from the process parameter data and the treatment effect data, resulting in feature vectors. A convolutional neural network is used to extract spatial features of the data, including the spatial distribution characteristics of the microstructure of the pure cotton bubble fabric, the fiber arrangement, and the spatial layout of the bubbles in the fabric. A long short-term memory network is used to extract dynamic change features in the time series, extracting key features from the data, including time-domain features, frequency-domain features, and spatial features. Then, the data from different physical modalities are aligned and integrated along the time and spatial axes to ensure consistency and correlation between the data from each modality, providing high-quality temperature and pressure feature data for subsequent joint model construction.

[0055] Furthermore, the extracted temperature field distribution and the applied tension and pressure data are normalized, and the feature vectors are fused into a temperature and pressure feature vector through weighted fusion to establish an adaptive prediction model. During the model training phase, an effect prediction function is constructed by combining mean square error and physical constraint terms. The physical constraint terms are set according to the physical laws in the processing of pure cotton bubble fabric, and the gradient descent optimization algorithm is used to optimize the model parameters.

[0056] Inputting specific process parameters into the joint model immediately initiates the prediction process, without waiting for all data to be collected. The model has a built-in feedback mechanism that feeds back actual processing effect data. Upon receiving this feedback, the model corrects its predictions in real time, obtaining the corresponding processing effect prediction results. During model training, a combined mean squared error and physical constraint terms are used to define the effect prediction function. An adaptive regularization term is added, and optimization algorithms are used to optimize the model parameters. As training progresses, the weight of the physical constraint terms is gradually increased. These physical constraint terms are based on the physical laws governing the processing of pure cotton bubble fabric, including the pressure-volume relationship during bubble formation.

[0057] The results are sent to an interactive prediction platform, where they are presented in an intuitive, visual interface. Researchers can quickly adjust the detection methods for pure cotton bubble fabric and modify process parameter settings based on the prediction results. The platform allows multiple users to use and collaborate online simultaneously. R&D team members can access the interactive prediction platform from their respective workstations to conduct predictive analysis and discussions on the same pure cotton bubble fabric project.

[0058] Using this predictive method, research institutions can quickly predict the processing effects under different process parameters during the development of new pure cotton bubble fabrics, significantly reducing the number of experiments and R&D costs. Ultimately, the optimal combination of process parameters was successfully selected, enabling the developed pure cotton bubble fabric to meet the expected goals in terms of bubble size uniformity, reasonable distribution, fabric strength, and elasticity, while significantly improving bubble stability and shortening the R&D cycle.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the treatment effect of pure cotton bubble fabric based on deep learning, characterized in that, include: Collect and preprocess process parameter data and corresponding treatment effect data involving multi-physics coupling. The multi-physics coupling data includes the temperature field distribution, tension and pressure, fluid diffusion rate and fluid viscosity data of the fabric during the treatment process. By combining the temperature field distribution with the applied tension and pressure data, a temperature-pressure model is constructed. This model employs an improved hybrid neural network structure combining a long short-term memory network (LSTM) and a convolutional neural network (CNN). The CNN extracts the spatial features of the fabric, including the spatial distribution features of the microstructure of the pure cotton bubble fabric, the fiber arrangement, and the spatial layout of the bubbles within the fabric. The LSTM network extracts the dynamic changes in temperature and pressure over time, extracting key features from the data to construct the temperature-pressure model. Skip connections are introduced within the temperature-pressure model and between feature interaction layers to form multi-layer connections, preventing gradient vanishing during deep network training and enabling the dynamic evolution of bubble size and distribution caused by changes in pressure and temperature parameters over time during the processing of pure cotton bubble fabric. The temperature-pressure model is used to monitor temperature and applied tension and pressure in real time, collect multiple sets of data and fuse them to obtain a temperature-pressure feature vector; The fluid diffusion rate and fluid viscosity data are fused with temperature and pressure feature vectors to construct a joint model. The joint model is used to make real-time predictions on pure cotton bubble fabric data. A processing effect prediction function is constructed based on the real-time prediction data. An optimization algorithm is built based on the obtained processing effect prediction function, uploaded to an interactive prediction and optimization platform, and the joint model is dynamically adjusted through sharing and interaction according to different needs and set parameters.

2. The method for predicting the treatment effect of pure cotton bubble fabric based on deep learning according to claim 1, characterized in that, The process of collecting and preprocessing process parameter data involving multiphysics coupling and corresponding processing effect data includes: The collected multi-physics coupled data were cleaned and normalized to obtain local detail features and global trend features at different scales. The data of different physical modes were aligned at multiple scales on the time axis and spatial axis.

3. The method for predicting the treatment effect of pure cotton bubble fabric based on deep learning according to claim 1, characterized in that, The temperature-pressure model is used for real-time monitoring of temperature and applied tension and pressure. The process of collecting multiple sets of data and fusing them to obtain a temperature-pressure feature vector includes: Based on the constructed temperature and pressure model, key features and interrelationships in the data are automatically identified and weighted fusion is performed. The weight coefficients are determined based on the correlation between each data point and the processing effect, generating a temperature and pressure feature vector. The temperature and pressure feature vector contains information from multiple dimensions, including temperature gradient, pressure intensity, and tension distribution.

4. The method for predicting the treatment effect of pure cotton bubble fabric based on deep learning according to claim 1, characterized in that, The process of fusing fluid diffusion rate and fluid viscosity data with temperature and pressure feature vectors to construct a joint model, which is used to make real-time predictions on pure cotton bubble fabric data, includes the following steps: Based on the results of fusing fluid diffusion rate and fluid viscosity data with temperature and pressure feature vectors, a joint model is constructed, which is capable of real-time prediction and dynamic adjustment. An incremental learning mechanism is introduced into the model, enabling it to automatically update its internal parameters with the input of new data without retraining the entire model. The new data comes from real-time monitoring equipment on the production line and new process data. When new pure cotton bubble fabric processing data is input, the model automatically adjusts its internal weights and parameters, records each set of data at fixed intervals, and fuses each set of data to construct a processing effect prediction function.

5. The method for predicting the treatment effect of pure cotton bubble fabric based on deep learning according to claim 1, characterized in that, The process of constructing the optimization algorithm based on the obtained processing effect prediction function includes: During model training, a processing effect prediction function is defined by combining mean squared error and physical constraint terms. An adaptive regularization term is added, and the model parameters are optimized using an optimization algorithm. As training progresses, the weight of the physical constraint terms is gradually increased. The physical constraint terms are based on the physical laws in the processing of pure cotton bubble fabric, including the relationship between the fabric volume and various data during bubble formation.

6. The method for predicting the treatment effect of pure cotton bubble fabric based on deep learning according to claim 1, characterized in that, The process of dynamically adjusting the joint model through sharing and interaction based on different needs and set parameters, as described in the upload interactive prediction optimization platform, includes: The prediction results are displayed in real time through a visual interface. Based on the results, the detection method of pure cotton bubble fabric can be controlled in real time, the process parameter settings for collecting pure cotton bubble fabric can be dynamically adjusted and fed back to the joint model, and multiple users can use and collaborate online at the same time. They can access this interactive prediction and optimization platform from their respective work terminals to conduct predictive analysis and discussion on the same pure cotton bubble fabric project.

Images