Textile design optimization method based on artificial intelligence

Abstract

The invention discloses a textile design optimization method based on artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: constructing a textile processing digital twinborn model; multi-modal data acquisition and feature fusion: providing data input by deploying a sensor array and fusing features; simulation and deduction are carried out based on a digital twinborn model, a diagnosis result of the state sensing module is used as model input, the predictive control module carries out advanced simulation and optimization calculation by using a prediction model, and the collaborative scheduling module realizes global optimization based on the model and outputs a precise control instruction; and executing the precise control instruction, returning the executed real data as feedback, and updating and correcting the S1 digital twinborn model to form a closed loop. According to the method, refinement, real-time adaptive optimization and efficient cooperative control of the textile processing process can be achieved, the processing quality and consistency are remarkably improved, the waste and defective rate is reduced, and the method has the self-evolution capacity.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a textile design optimization method based on artificial intelligence. BACKGROUND

[0002] With the increasing demand for intelligent, efficient and flexible production modes in global manufacturing, the textile industry, as an important traditional pillar industry, is also continuously promoting its technological innovation and transformation. Under this background, the textile design optimization method based on artificial intelligence has gradually become a hot field of industry research and application. This innovative method aims to optimize the whole life cycle processing flow of textiles from raw material processing to finished product output by integrating advanced intelligent algorithms, thereby significantly improving production efficiency, product quality, and giving the production line stronger adaptability and market competitiveness.

[0003] To address the above challenges, the prior art has made many explorations. The Chinese patent document with publication number CN113638156B explores the intelligent application of textile processing equipment. This patent focuses on the printing and dyeing link, and its scheme includes conveying equipment, printing and dyeing equipment and auxiliary equipment, aiming to ensure the smoothness of the textile surface through the conveying equipment, then perform dye replenishment operation by the printing and dyeing equipment, and assist with the auxiliary equipment to accurately replenish the defective parts, so as to improve the printing and dyeing quality and efficiency. However, the printing and dyeing process parameters of CN113638156B rely on preset values, even if some printing and dyeing defects are solved, it is difficult to make fine adjustments according to the real-time feedback of the textile surface to the dye absorption, the printing and dyeing ink drop diffusion characteristics or the changes of environmental temperature and humidity. This not only may cause local accumulation of dye, uneven coloring or color difference, resulting in material waste and product defect rate rising, but also deeper problem is that it deprives the system of the ability to adaptively correct unknown or unexpected conditions. In other words, the automation of the existing scheme stops at the repeated execution of mechanical actions, rather than the intelligent understanding and response to complex and dynamic variables.

[0004] Therefore, how to break through the limitations of traditional fixed parameter control, introduce advanced artificial intelligence algorithms, and build a textile processing design method that can real-time perceive the textile processing state, dynamically optimize process parameters, and realize multi-module intelligent collaboration, has become a key challenge and technical problem to be solved for those skilled in the art. SUMMARY

[0005] The present application provides a textile design optimization method based on artificial intelligence to solve the problems of relying on preset static process parameters, lacking real-time dynamic adaptive ability and insufficient collaborative optimization among multiple modules in the prior art.

[0006] To solve the above technical problems, the present application provides the following technical solutions: a textile design optimization method based on artificial intelligence, comprising the following steps: Step S1, constructing a textile processing digital twin model; S1-1, a textile material characteristic parameterization model for representing the basic physical, chemical and structural characteristics of the textile to be processed, the textile material characteristic parameterization model being initialized and parameterized by a pre-established material database, physical property test data and spectral analysis data; S1-2, a real-time state model of the processing environment for reflecting the environmental parameters in the processing area in real time, the data input of the real-time state model of the processing environment being derived from environmental sensors and transmitted to the central processing unit through an industrial Ethernet protocol; S1-3, a processing equipment operation parameter model for real-time recording and updating the current operation parameters of each processing unit, the operation parameters being collected and acquired in real time through the feedback signals of the equipment controller; S1-4, a real-time physical state model of the textile for dynamically representing the physical properties of the textile at the current time in the processing process, the data update of the real-time physical state model of the textile being derived from real-time data collection and fusion of multi-modal sensors; S1-5, a processing process physical-chemical behavior prediction model for predicting the physical changes and chemical reactions of the textile in a specific processing step based on the current textile material characteristics, processing environment and equipment operation parameters, the processing process physical-chemical behavior prediction model being constructed by coupling physical equations and machine learning algorithms; S1-6, a processing quality evaluation model for quantitatively evaluating the quality indicators of the textile at each processing stage, the processing quality evaluation model comparing the measurement results of the standard quality detection instrument with the preset quality standard and combining the expert experience knowledge base to generate quantitative quality evaluation scores or indicators.

[0007] The data structure of the textile material characteristic parameterization model is a multi-dimensional tensor, which can store and represent the basic physical, chemical, structural and performance parameters of the textile, as well as their change trends under different temperatures, humidities and stress states in a unified manner, forming a digital image of the textile material characteristics.

[0008] The processing process physical-chemical behavior prediction model is constructed by coupling physical equations and machine learning algorithms, wherein: The physical equations include mass conservation equation, energy conservation equation, momentum conservation equation, diffusion model based on Fick's law and heat transfer model based on Fourier's law; The machine learning algorithms include physical information neural network PINN and long short-term memory network LSTM; The physical-chemical behavior prediction model of the processing process can output the prediction of the state change of the textile based on the input parameters.

[0009] Step S2, multi-modal data acquisition and feature fusion, by deploying a sensor array and fusing features, provides data input for the S1 digital twin model; In step S2, the following sub-steps are further included: S2-1, multi-modal sensor deployment: deploying sensors at key links of the textile processing pipeline; S2-2, heterogeneous data unified interface and synchronization: various sensors are connected to data acquisition cards and industrial computers through communication protocols, data format conversion and standardization are carried out through a unified software interface, and all sensor data are synchronized through time stamps; S2-3, multi-modal feature fusion: inputting the preprocessed multi-modal sensor data into a deep learning fusion network, extracting spatial features from image data, extracting time series dynamic features from time series data, and combining features of different modalities by weighting.

[0010] Step S3, simulation and deduction based on the digital twin model, the diagnosis results of the state perception module are used as model input, the predictive control module uses the prediction model for advanced simulation and optimization calculation, and the collaborative scheduling module realizes global optimization based on the model, and finally outputs precise control instructions; In step S3, the following sub-steps are further included: S3-1, textile state perception and abnormal diagnosis module: receiving the fused multi-modal feature vector, perceiving the real-time state of the textile in the processing process, identifying and quantifying various physical parameters of the textile and identifying potential processing abnormalities, and outputting quantitative state indicators and abnormal alarm signals; S3-2, predictive control and process parameter optimization module: using model predictive control strategy, based on the prediction ability of the digital twin model, simulating the state evolution of the textile under different process parameters, and integrating deep reinforcement learning algorithm to process optimization problems, realizing adaptive strategy learning, outputting the current textile state and target quality requirements; S3-3, multi-module collaborative optimization and scheduling module: receiving local optimization results, and through a reinforcement learning agent or a rule-based expert system, macro-scheduling and cross-module collaboration of the processing flow.

[0011] Step S4, execute the final output of the precise control instruction, and return the real data after execution as feedback, which is used for continuous updating and correction of the S1 digital twin model, forming a closed loop of "perception-decision-execution-learning".

[0012] In step S4, the following sub-steps are further included: S4-1, instruction analysis and execution: send dynamic process parameters to the local controller of each processing equipment, analyze the instructions into voltage, current, pulse signals that can be recognized by the underlying execution mechanism, and adjust the physical parameters through the execution mechanism according to the signals; S4-2, real-time state monitoring and data back: while the instructions are being executed, the actual running state of the equipment and the real-time physical state of the textile are continuously monitored, and the data is returned; S4-3, closed-loop feedback and model updating: update the textile real-time physical state model and the processing equipment running parameter model through the returned real-time state data, and evaluate the actual effect of artificial intelligence decision-making to drive online learning and iterative optimization of the model.

[0013] Multi-modal feature fusion, including: For image data, multiple independent CNN branches are used to extract spatial features in parallel; for time series data, an independent RNN branch is used to extract time series dynamic features, capturing the trend and correlation of parameters over time; then, an attention fusion layer is used to weight and combine the features of different modalities.

[0014] Prediction based on digital twin model, including: An advanced model predictive control strategy MPC is adopted to simulate the state evolution of the textile under different process parameter combinations in the future; the MPC controller calculates a series of optimized control sequences at future time points by solving a real-time optimization problem, and the optimization goal is to make the final quality indicators of the textile optimal while the energy and material consumption are minimized under the premise of meeting all key process constraints.

[0015] The real-time physical state model of the textile and the running parameter model of the processing equipment are updated based on the returned real-time state data, including: When there is a significant deviation between the actual output and the digital twin prediction, the deviation will be captured by the system and used to correct the prediction model of the physical-chemical behavior of the processing process, through online parameter adjustment or local model retraining, to ensure the continuous accuracy and matching degree of the digital twin model to the real world; in addition, the long-term accumulated processing data containing input parameters, real-time state and final quality results are stored in a large-scale big data platform for regular offline retraining and optimization of all AI models.

[0016] Compared with the prior art, the beneficial effects of the present application are: The present application realizes comprehensive, real-time, and high-precision virtual mapping of the physical, chemical, structural, environmental, and equipment states of textiles during the processing by constructing and dynamically maintaining a multi-level textile processing digital twin model, not only records historical data, but more importantly, can make forward-looking predictions, effectively overcoming the fundamental defects of relying on static preset parameters and being unable to perceive and predict dynamic changes in the prior art, enabling the processing to change from "blind processing" to "predictive processing", and significantly improving the accuracy and foresight of decision-making.

[0017] The present application introduces multi-modal sensor data acquisition and deep fusion technology, breaking through the problem of limited data dimension of traditional single sensor, through the fusion of multi-source heterogeneous data, the system can obtain more comprehensive and detailed real-time state information of textiles.

[0018] The present application proposes a hierarchical artificial intelligence decision and optimization architecture, which deeply combines model predictive control and deep reinforcement learning, and realizes fine and adaptive optimization of complex and dynamic processing. This combination enables the system not only to cope with known variables, but also to intelligently adapt to unknown or unexpected situations, and to achieve millisecond-level dynamic adjustment of process parameters, significantly improving processing precision and reducing waste and defective product rate.

[0019] The present application realizes intelligent collaborative optimization and scheduling between multiple processing modules, breaking through the bottleneck of relatively independent module functions and lack of global linkage control in the prior art. Through the high-level AI decision module, the system can consider the mutual influence between different processing stages, realize global optimization of resources, and ensure seamless connection and efficient operation of the entire production process, thereby minimizing production cost and environmental load.

[0020] The closed-loop feedback mechanism constructed by the present application enables the digital twin model and all artificial intelligence models to continuously learn and iteratively optimize through actual processing data, which means that the method has self-evolution ability, and its decision-making ability and optimization effect will continuously improve with the accumulation of processing data, so as to cope with more complex and diversified textile processing demands, and realize truly flexible intelligent manufacturing. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. The following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0022] Figure 1 The method flowchart provided by the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0023] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of the present application. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only to be a selected embodiment of the present application.

[0024] Please refer to Figure 1 , Figure 1 is a method flowchart of a textile design optimization method based on artificial intelligence provided by an embodiment of the present application, comprising the following steps: Step S1, constructing a textile processing digital twin model, through high-fidelity virtual mapping of all key elements in the real textile processing process, including: S1-1, a textile material characteristic parameterization model, for comprehensively characterizing the inherent properties of the textile to be processed, covering the basic physical, chemical and structural characteristics of the textile, including the type of constituent fiber, the count of yarn, the density of fabric, the thickness, the porosity, the initial moisture content, the chemical composition and the dyeing performance curve.

[0025] In actual operation, the initialization and parameterization setting of the model depend on multiple data sources, and a pre-established material database stores detailed attribute data of various typical textile fibers; for cotton fibers, the database not only contains the average density, Young's modulus, thermal conductivity, specific heat capacity, moisture expansion coefficient, and adsorption isotherm and fixation rate curve of common dyes, but also contains the microstructure change characteristics of cotton of different maturity and different origin in the processing process.

[0026] The physical property test data is obtained through standardized laboratory test methods, ensuring the accuracy and comparability of the data; the fabric thickness is measured under a certain pressure using a lever or pressure disc thickness gauge with a precision of up to 0.001 mm; the grammage is weighed by a standard area sampler and an electronic balance with a precision of 0.001 g; the air permeability is measured according to the GB / T5453 or ASTM D737 standard using an air permeameter under a given pressure difference; the tensile strength and tear strength are tested by a universal material testing machine under constant extension rate or constant load to obtain key mechanical performance indicators such as breaking strength and elongation at break.

[0027] Spectroscopy data is obtained by non-destructive analysis of textile samples using Fourier Transform Infrared Spectrometer or Near Infrared Spectrometer; these spectroscopy data can reveal the molecular structure, chemical bond information of the fiber, and can accurately calculate the moisture content, residual chemical auxiliary content and dye concentration of the textile by establishing a spectroscopy-component quantitative calibration model.

[0028] The data structure of this parameterized model is designed in the form of a multi-dimensional tensor, which can store and represent the above-mentioned physical, chemical, structural and performance parameters, as well as their change trends under different temperature, humidity and stress states in a unified and efficient manner, thus forming a comprehensive digital portrait of the textile material properties from macro to micro.

[0029] S1-2, Real-time Processing Environment State Model, used to accurately capture and update the environmental parameters in the processing area in real time, data input from a series of high-precision environmental sensors, data transmitted through industrial Ethernet protocol with low delay and high reliability.

[0030] Specific sensor configuration includes: high-precision digital temperature sensors are deployed at different key points in the processing area to form an accurate temperature field distribution map; capacitive humidity sensors are used to monitor relative humidity, and dew point sensors are also equipped for specific processing environments; thermal air speed sensors are used to monitor air flow rate in real time, especially in drying and cooling areas, which is crucial for moisture evaporation and heat transfer efficiency.

[0031] S1-3, Processing Equipment Operating Parameter Model, used to record and dynamically update the current operating parameters of each processing unit on the textile processing line, including water scraping unit, water absorbing unit, ironing unit, drying unit, printing and dyeing unit, drying unit, finishing unit.

[0032] The parameters contained in the model are extremely detailed, including: scraper pressure, water absorbing roller linear speed, ironing temperature, drying zone heating power, printing and dyeing nozzle flow, dye concentration, drying air volume, conveyor belt speed, tension setting value and roller gap.

[0033] All parameters are collected and updated in real time through the feedback signals of the equipment controllers of each processing unit at a predefined extremely high frequency, ensuring the accurate reflection of the digital twin model on the equipment state.

[0034] S1-4, Real-time physical state model of textiles, used to dynamically characterize the changes in physical properties of textiles at every moment throughout the entire processing process, through the integration of real-time data acquisition and fusion technology of multi-modal sensors, the key states of textiles are accurately perceived, including: its surface temperature distribution; internal moisture content distribution; dimensional deformation; surface flatness; color absorption uniformity; dye penetration depth; and surface finish; the data update frequency of this model is synchronized with the sensor acquisition frequency, ensuring that it can accurately capture the small changes of textiles in the high-speed processing process.

[0035] S1-5, Physical-chemical behavior prediction model of the processing process, used to predict the physical changes and chemical reactions that will occur in textiles in a specific processing step based on the current textile material properties, processing environment and equipment operating parameters.

[0036] The physical-chemical behavior prediction model of the processing process adopts an innovative construction method that couples physical equations and machine learning algorithms; among them, physical equations provide a solid theoretical foundation and global constraints, including mass conservation equations, energy conservation equations, momentum conservation equations, diffusion models based on Fick's law, and heat transfer models based on Fourier's law; these physical equations are solved by numerical methods, building a prediction framework for temperature field, humidity field, stress field, and dye concentration field in the textile processing process.

[0037] On this basis, machine learning algorithms, including physical information neural networks and long short-term memory networks, are integrated to fit complex nonlinear relationships, compensate for the simplification errors of physical models, and handle complex interactions that are difficult to describe through explicit physical equations; PINN introduces physical constraints into the loss function, so that the neural network learns data while also enforcing physical laws, thereby improving the generalization ability and prediction accuracy of the model; LSTM is particularly suitable for processing process data with strong time-dependent, and can learn and predict the dynamic evolution of textile state over time.

[0038] The physical-chemical behavior prediction model of the processing process is highly optimized and can output the prediction of textile state changes in future time steps based on given input parameters within sub-second time, providing key forward-looking information for subsequent artificial intelligence decision-making, so that the system can predict and avoid potential quality problems in advance.

[0039] S1-6, Processing quality evaluation model, used to quantitatively evaluate the quality indicators of textiles at each processing stage, by integrating the measurement results of various quality detection instruments and comparing them with the preset strict quality standards, finally generating a quantitative quality evaluation score or indicator combined with expert experience knowledge base.

[0040] Specific quality indicators include: color difference; surface flatness level; and whether there are defects; the expert experience knowledge base contains a large amount of historical processing data and expert identification experience of specific defects, which is used to correct or optimize the evaluation results, improve the accuracy and robustness of the evaluation.

[0041] Step S2, multi-modal data acquisition and feature fusion, deploy multi-modal sensors, synchronize data through a unified interface, and perform multi-modal feature fusion to provide data input for the S1 digital twin model; S2-1, multi-modal sensor deployment, strategically deploy a series of high-precision, high-frequency sensors at each key link of the textile processing pipeline; before the textile enters the pretreatment stage, deploy a line-scan visible light camera and a hyperspectral camera to obtain fabric surface texture, initial defects, and initial color information; in the drying unit, deploy a non-contact infrared thermal imager and a microwave moisture sensor to ensure uniform drying of the fabric to the target moisture content.

[0042] S2-2, heterogeneous data unified interface and synchronization: various sensors are connected to multiple distributed data acquisition cards and high-performance industrial computers through their respective communication protocols. To achieve unified management and efficient transmission of data, the system uses a unified software interface for data format conversion, standardization, and semantic encapsulation. Raw data from different manufacturers and different protocols are unified into a standard format that can be understood and processed by the central processing unit.

[0043] All sensor data are precisely synchronized by high-precision timestamps to ensure strict alignment of different modal data in the time dimension, which is crucial for subsequent feature fusion and causal analysis.

[0044] S3-3, multi-modal feature fusion: input the pre-processed and time-synchronized multi-modal sensor data into a carefully designed deep learning fusion network. This network uses a convolutional neural network based on attention mechanism and a recurrent neural network fusion architecture.

[0045] The specific design includes: first, for image data, use multiple independent convolutional neural network branches to extract spatial features in parallel. These features can capture visual information such as fabric texture, defect areas, color blocks, and temperature distribution patterns; at the same time, for time series data, use an independent RNN branch to extract time series dynamic features, capturing the trends and correlations of these parameters over time; then, through an attention fusion layer, the features of different modalities are weighted and combined.

[0046] Through multi-source sensor data acquisition and fusion, a high-dimensional feature vector is formed, which will serve as the core asset, mainly used to drive and update the digital twin model constructed in step S1, ensuring real-time synchronization with the physical factory.

[0047] Step S3, simulation and deduction based on digital twin model, the diagnosis result of state perception module as model input, predictive control module uses prediction model for advance simulation and optimization calculation, and collaborative scheduling module realizes global optimization based on model, and finally outputs precise control instruction; S3-1, textile state perception and abnormal diagnosis module: the module receives the multi-modal feature vector fused in step S2, and accurately perceives and quantifies the real-time state of the textile in the processing process through the pre-trained deep learning model.

[0048] The model not only simply identifies surface features, but also can identify and quantify various deep physical parameters of the textile in real time, and can also identify and classify potential processing abnormalities.

[0049] The output of the diagnosis module is a series of accurately quantified state indicators and real-time abnormal alarm signals.

[0050] S3-2, predictive control and process parameter optimization module, which is the core of intelligent decision, highly depends on the physical-chemical behavior prediction model of processing process constructed in step S1-5, the system uses model predictive control framework, takes S1-5 model as internal prediction model, and rolls to optimize process parameters in future time domain to make the output score of S1-6 processing quality evaluation model optimal.

[0051] Firstly, the system adopts advanced model predictive control strategy, which has unique foresight advantage, based on the powerful prediction ability of digital twin model, simulates the state evolution of textile under different process parameter combinations in a period of time in the future.

[0052] The model predictive controller calculates a series of optimized control sequences at future time points by solving a real-time optimization problem, and the optimization goal is to make the final quality index of the textile optimal and the energy and material consumption minimum under the premise of meeting all key process constraints.

[0053] Secondly, in order to deal with the inherent complex, high-dimensional and nonlinear dynamic characteristics of textile processing, especially in the face of unknown or rapidly changing processing conditions, the invention deeply integrates deep reinforcement learning algorithm to realize more robust and adaptive strategy learning; at the same time, the adjustable parameters of the processing equipment constitute its continuous action space, including conveyor belt speed, temperature of each heating zone, roller pressure, nozzle flow, inkjet frequency, dye concentration, pH adjustment amount, drying air volume, steam pressure, cooling air speed, tension setting value and roller gap.

[0054] The final output of the module is the accurate process parameter values that each processing unit should adjust in real time for the current textile state and target quality requirements. Specific instructions include: "adjust the squeegee pressure from 1.5 bar to 1.3 bar", "adjust the ironing temperature from 180°C to 175°C", "adjust the dye flow rate from 5 L / min to 4.8 L / min", "adjust the drying fan speed from 1200 RPM to 1150 RPM"; the adjustment granularity of these parameters matches the minimum adjustable step of the equipment control system, and the adjustment frequency is highly matched with the sensor data acquisition frequency, ensuring the refinement and real-time nature of the control.

[0055] S3-3, multi-module collaborative optimization and scheduling module, receives local optimization results from the predictive control and process parameter optimization module, and through a high-level reinforcement learning agent or rule-based expert system, makes global macro-scheduling and cross-module collaboration for the entire textile processing flow; this collaboration considers the complex mutual influence between different processing stages, thereby avoiding the problem of local optimum but global suboptimal caused by traditional isolated optimization.

[0056] This module performs global situation awareness based on the shared digital twin model, making cross-unit collaborative scheduling decisions, realizing real-time and efficient data exchange and control instruction transmission between processing units, and the decision goal is to achieve the global optimization of the overall efficiency, product quality and resource utilization of textile processing.

[0057] Step S4, execute the final output of the precise control instructions, and return the real data after execution as feedback for continuous updating and correction of the S1 digital twin model, forming a "perception-decision-execution-learning" closed loop; S4-1, instruction analysis and execution: the generated optimization instructions are sent to the local controllers of each processing equipment through a high-reliability industrial communication network with extremely low delay; the local controller is responsible for parsing high-level instructions into precise control signals recognizable by the underlying execution mechanism; S4-2, real-time state monitoring and data feedback: while the instructions are being executed, various sensors and feedback units of the execution mechanism built into each processing equipment continuously and at a high frequency monitor the actual running state of the equipment and the real-time physical state of the textile; these monitoring data are collected at a very high frequency and fed back to the data acquisition and preprocessing subsystem through the industrial network; this high-frequency data feedback ensures immediate perception of the production line state, providing the latest and most accurate basis for subsequent decision-making.

[0058] S4-3, closed-loop feedback and model updating: the real-time state data fed back constitute the key feedback loop of the intelligent system of the present application; firstly, these data are immediately used to update the real-time physical state model of the textile and the operating parameter model of the processing equipment, so as to provide the latest and most accurate input for the next round of artificial intelligence decision-making; secondly, these data are also used to evaluate the actual effect of artificial intelligence decision-making: compare the actual output quality indicators with the target quality predicted by the digital twin, and use it as a reward signal for the reinforcement learning agent, driving its online learning and iterative optimization.

[0059] When there is a significant deviation between the actual output and the digital twin prediction, the deviation will be captured by the system and used to correct the prediction model of the physical-chemical behavior of the processing process, through online parameter adjustment or local model retraining, to ensure the continuous accuracy and matching degree of the digital twin model to the real world; in addition, the long-term accumulated processing data containing input parameters, real-time state and final quality results are stored in a large-scale big data platform for regular offline retraining and optimization of all AI models; this periodic offline learning can further improve the generalization ability, robustness and decision-making accuracy of the model, so that it can cope with more complex and diversified textile processing demands, and realize truly flexible intelligent manufacturing and self-evolution.

[0060] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application has various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A textile design optimization method based on artificial intelligence, characterized in that, Includes the following steps: Step S1: Construct a digital twin model of textile processing; Step S2, multimodal data acquisition and feature fusion, provides data input for the digital twin model in S1 by deploying a sensor array and fusing features; Step S3: Simulation and deduction are performed based on the digital twin model. The diagnostic results of the state perception module are used as the model input. The predictive control module uses the predictive model to perform advanced simulation and optimization calculation. The collaborative scheduling module achieves global optimization based on the model and finally outputs precise control commands. Step S4: Execute the final output of precise control commands and send back the actual data after execution as feedback to continuously update and correct the digital twin model S1, forming a closed loop of "perception-decision-execution-learning".

2. The textile design optimization method based on artificial intelligence according to claim 1, characterized in that, The steps of constructing a digital twin model of textile processing include constructing the following model: S1-1, Textile Material Property Parameterization Model, used to characterize the basic physical, chemical and structural properties of textiles to be processed. The textile material property parameterization model is initialized and parameterized through a pre-established material database, physical property test data and spectral analysis data. S1-2, Real-time processing environment status model, used to reflect environmental parameters within the processing area in real time. The data input of the real-time processing environment status model comes from environmental sensors and is transmitted to the central processing unit via the industrial Ethernet protocol. S1-3, Processing equipment operating parameter model, used to record and update the current operating parameters of each processing unit in real time. The operating parameters are collected and obtained in real time through feedback signals from the equipment controller. S1-4, Real-time physical state model of textiles, used to dynamically characterize the physical properties of textiles at the current moment during processing. The data update of the real-time physical state model of textiles comes from the real-time data acquisition and fusion of multimodal sensors. S1-5, Processing physical-chemical behavior prediction model, used to predict the physical changes and chemical reactions of the textile in a specific processing step based on the current textile material properties, processing environment and equipment operating parameters. The processing physical-chemical behavior prediction model is constructed by coupling physical equations and machine learning algorithms. S1-6, Processing quality assessment model, used to quantitatively assess the quality indicators of the textile at each processing stage. The processing quality assessment model compares the measurement results of standard quality testing instruments with preset quality standards and combines them with an expert experience knowledge base to generate quantitative quality assessment scores or indicators.

3. The textile design optimization method based on artificial intelligence according to claim 2, characterized in that, The data structure of the parameterized model of textile material properties is a multidimensional tensor. The multidimensional tensor can store and represent the basic physical, chemical, structural and performance parameters of textiles, along with their variation trends under different temperatures, humidity and stress states, in a unified manner, forming a digital profile of the textile material properties.

4. The textile design optimization method based on artificial intelligence according to claim 2, characterized in that, The physical-chemical behavior prediction model for the processing procedure is constructed using coupled physical equations and machine learning algorithms, wherein: The physical equations include the mass conservation equation, the energy conservation equation, the momentum conservation equation, the diffusion model based on Fick's law, and the heat transfer model based on Fourier's law. The machine learning algorithms include the Physical Information Neural Network (PINN) and the Long Short-Term Memory Network (LSTM). The physical-chemical behavior prediction model for the processing can output a prediction of the state changes of textiles based on the input parameters.

5. The textile design optimization method based on artificial intelligence according to claim 1, characterized in that: Step S2 further includes the following sub-steps: S2-1, Multimodal sensor deployment: Deploying sensors in key stages of textile processing lines; S2-2, Unified Interface and Synchronization of Heterogeneous Data: Various sensors are connected to the data acquisition card and industrial computer through communication protocols. Data format conversion and standardization are performed through a unified software interface, and data from all sensors is synchronized through timestamps. S2-3, Multimodal Feature Fusion: The preprocessed multimodal sensor data is input into a deep learning fusion network to extract spatial features from image data, extract temporal dynamic features from time-series data, and weighted combination of features from different modalities.

6. The textile design optimization method based on artificial intelligence according to claim 1, characterized in that: Step S3 further includes the following sub-steps: S3-1, Textile State Perception and Anomaly Diagnosis Module: Receives the fused multimodal feature vector, perceives the real-time state of textiles during processing, identifies and quantifies various physical parameters of textiles and identifies potential processing anomalies, and outputs quantified state indicators and anomaly alarm signals. S3-2, Predictive Control and Process Parameter Optimization Module: Utilizing model predictive control strategies and the predictive capabilities of digital twin models, this module simulates the state evolution of textiles under different process parameters and integrates deep reinforcement learning algorithms to handle optimization problems, achieving adaptive strategy learning and outputting the current state of the textiles and target quality requirements. S3-3, Multi-module Collaborative Optimization and Scheduling Module: Receives local optimization results and performs macro-level scheduling and cross-module collaboration of the processing flow through a reinforcement learning agent or rule-based expert system.

7. The textile design optimization method based on artificial intelligence according to claim 1, characterized in that: Step S4 further includes the following sub-steps: S4-1, Command parsing and execution: Sends dynamic process parameters to the local controller of each processing equipment, parses the commands into voltage, current and pulse signals that can be recognized by the underlying actuators, and adjusts the physical parameters according to the signals through the actuators; S4-2, Real-time Status Monitoring and Data Feedback: While executing instructions, continuously monitor the actual operating status of the equipment and the real-time physical status of the textiles, and feed back the data. S4-3, Closed-loop feedback and model update: The real-time physical state model of textiles and the operating parameter model of processing equipment are updated by the real-time status data transmitted back, and the actual effect of artificial intelligence decision-making is evaluated, driving the online learning and iterative optimization of the model.

8. The textile design optimization method based on artificial intelligence according to claim 5, characterized in that, The multimodal feature fusion includes: For image data, multiple independent CNN branches are used to extract spatial features in parallel; for temporal data, independent RNN branches are used to extract temporal dynamic features, capturing the changing trends and interrelationships of parameters over time; subsequently, an attention fusion layer is used to weight and combine the features of different modalities.

9. The artificial intelligence-based textile design optimization method according to claim 6, characterized in that, The predictive capabilities based on the digital twin model include: An advanced model predictive control (MPC) strategy is adopted to simulate the state evolution of textiles under different combinations of process parameters over a period of time in the future. The MPC controller calculates a series of optimal control sequences for future moments by solving a real-time optimization problem. The optimization objective is to achieve the best final quality index of the textiles while minimizing energy and material consumption, under the premise of satisfying all key process constraints.

10. The artificial intelligence-based textile design optimization method according to claim 7, characterized in that, The process of updating the real-time physical state model of textiles and the operating parameter model of processing equipment through the returned real-time status data includes: When there is a significant deviation between the actual output and the digital twin prediction, the deviation will be captured by the system and used to correct the physical-chemical behavior prediction model of the processing process. Through online parameter adjustment or local model retraining, the continuous accuracy of the digital twin model and its matching degree to the real world are ensured. In addition, the long-term accumulated processing data, which includes input parameters, real-time status and final quality results, is stored in a large-scale big data platform for regular offline retraining and optimization of all AI models.

Images