Textile wear resistance optimization method and device based on machine learning, equipment and storage medium

Abstract

The invention provides a textile wear resistance optimization method and device based on machine learning, equipment and a storage medium, and is characterized in that scattered experience knowledge is converted into structured data assets by constructing a database containing fiber raw materials, fabric weaves, after-finishing processes and performance test results; data are processed by calling a data preprocessing and feature engineering algorithm, a wear resistance prediction model is trained, an accurate mapping relation from design parameters to performance results is established, the prediction model is combined with an intelligent optimization algorithm, an optimal combination can be automatically searched in a huge parameter space, and finally, the wear resistance prediction model is optimized. Through a closed-loop mechanism for feeding back new experimental data to a database and retraining the model, the model can continuously learn and iteratively optimize, the prediction precision and the application range are continuously improved, and the problems that the research and development period is long, the cost is high, and the research and development efficiency is high due to the fact that existing wear-resistant textile research and development depend on trial and error experiments are solved. And an optimal parameter combination is difficult to find under a multi-factor complex coupling effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of textile wear resistance optimization, in particular to a textile wear resistance optimization method, device and equipment based on machine learning and a storage medium. BACKGROUND

[0002] Wear resistance is one of the core indicators of textile durability, and plays a key role in the fields of workwear, outdoor equipment, military textiles, furniture fabrics and industrial textiles. Improving the wear resistance of textiles can significantly prolong product life, reduce resource consumption and environmental waste.

[0003] Existing wear-resistant textile research and development mainly relies on the experience accumulation and repeated trial-and-error experiments of researchers, including selecting fiber raw materials, designing yarn and fabric structures, applying finishing processes, and verifying performance through physical tests (such as Martin Dale or Taber wear resistance tests). However, this method has significant limitations: long development cycle, numerous experimental combinations and time-consuming tests; high cost, consuming a large amount of raw materials, auxiliaries, energy and manpower for each round of experiments; difficult optimization, wear resistance is affected by nonlinear coupling of multiple factors, and experience-based methods are prone to local optimization; knowledge transfer challenges, it is difficult to quantify and transfer tacit experience.

[0004] Although existing technologies have made preliminary attempts to introduce data analysis or simulation, they lack a systematic multi-dimensional database, accurate prediction model and intelligent optimization closed loop, and cannot effectively solve the above problems.

[0005] Therefore, the present application is proposed. SUMMARY

[0006] The present application discloses a textile wear resistance optimization method, device and equipment based on machine learning and a storage medium, aiming to solve the problem that existing wear-resistant textile research and development relies on trial-and-error experiments, resulting in long development cycle, high cost, and difficulty in finding optimal parameter combinations under complex coupling of multiple factors.

[0007] The first embodiment of the present application provides a textile wear resistance optimization method based on machine learning, comprising: obtaining a textile wear resistance database, the database including fiber raw material parameters, fabric organization parameters and finishing process parameters and corresponding wear resistance test results; calling a data preprocessing and feature engineering algorithm to process the data in the database, and using the processed data to train and generate a wear resistance prediction model; According to the wear resistance prediction model, and combining an intelligent optimization algorithm, the optimal material, structure and process parameter combination is searched and generated in the parameter space with the set wear resistance performance target as the guide; Obtaining new experimental data of the sample prepared according to the optimal parameter combination, and feeding the new experimental data back into the database to retrain the wear resistance performance prediction model to form a closed loop regulation.

[0008] Preferably, the fiber raw material parameters include fiber type, linear density, length, strength, yarn linear density, twist and blending ratio. The fabric organization parameters include weaving method, warp density, weft density and fabric gram weight; The finishing process parameters include finishing agent type, concentration, padding pick-up, curing temperature and time; The wear resistance test results include wear resistance revolutions measured according to ISO 12947-2 standard or ASTM D4966 standard as the core index, and breaking strength, tear strength, anti-pilling grade, fabric thickness, air permeability and cost as auxiliary indexes.

[0009] Preferably, the calling data preprocessing and feature engineering algorithm process the data in the database, specifically: Cleaning the data, processing missing values and removing outliers; One-hot encoding for category features and standardizing numerical features; Using feature selection methods to screen feature subsets related to wear resistance revolutions, and constructing new features based on textile theory, including fabric coverage factor, the calculation formula of which is:

[0010] Wherein, CF is the fabric coverage factor, EWD is the warp yarn linear density, and EPI is the warp density, EWT is the weft yarn linear density, and PPI is the weft density.

[0011] Preferably, the processed data is used to train and generate a wear resistance performance prediction model, specifically: Gradient boosting decision tree is selected as the prediction model, and the data set is divided into training set and test set; The training set is used to train the prediction model to learn the mapping relationship from design parameters and process parameters to wear resistance revolutions; The prediction model is evaluated by using the determination coefficient R 2 , root mean square error RMSE and mean absolute error MAE, wherein the calculation formula of RMSE is:

[0012] Wherein, n is the sample number, is the true wear resistance revolution, is the predicted wear resistance revolution.

[0013] Preferably, the searching and generating the optimal material, structure and process parameter combination in the parameter space are specifically: An optimization target is set as maximizing the wear-resistant revolutions by using a genetic algorithm, and a constraint condition is set; A candidate parameter combination is generated in the parameter space, and is input into the prediction model to obtain a predicted wear-resistant revolution; The next generation of candidate combinations is generated according to the prediction result by selection, crossover and mutation operations, and iteration is performed until an optimal combination meeting the constraint condition is found.

[0014] Preferably, the constraint condition includes that the fabric gram weight is less than a preset threshold, the air permeability is greater than a preset threshold, and the cost is less than a preset threshold.

[0015] The second embodiment of the present application provides a textile wear-resistant performance optimization device based on machine learning, comprising: A database query unit is configured to obtain a textile wear-resistant performance database, wherein the database includes fiber raw material parameters, fabric organization parameters, finishing process parameters and corresponding wear-resistant performance test results; A model construction unit is configured to call a data preprocessing and feature engineering algorithm to process data in the database, and train and generate a wear-resistant performance prediction model by using the processed data; An optimal combination searching unit is configured to search and generate an optimal material, structure and process parameter combination in a parameter space according to the wear-resistant performance prediction model and in combination with an intelligent optimization algorithm, and with a set wear-resistant performance target as a guide; A feedback unit is configured to obtain new experimental data of a sample prepared according to the optimal parameter combination, and feed the new experimental data back to the database, so as to retrain the wear-resistant performance prediction model to form a closed-loop regulation.

[0016] The third embodiment of the present application provides a textile wear-resistant performance optimization device based on machine learning, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the textile wear-resistant performance optimization method based on machine learning according to any one of the above.

[0017] The fourth embodiment of the present application provides a computer readable storage medium, wherein a computer program is stored, and the computer program can be executed by a processor of a device where the computer readable storage medium is located to implement the textile wear-resistant performance optimization method based on machine learning according to any one of the above.

[0018] Based on the machine learning-based method, apparatus, equipment, and storage medium provided by this invention, a database containing fiber raw materials, fabric structure, finishing processes, and performance test results is constructed to transform scattered experiential knowledge into structured data assets. By calling data preprocessing and feature engineering algorithms to process data and train an abrasion resistance prediction model, a precise mapping relationship from design parameters to performance results is established. Combining the prediction model with intelligent optimization algorithms enables automatic searching for optimal combinations within a vast parameter space. Finally, a closed-loop mechanism, through feeding new experimental data back to the database and retraining the model, allows the model to continuously learn and iteratively optimize, constantly improving prediction accuracy and applicability. This solves the problem that existing abrasion-resistant textile R&D relies on trial-and-error experiments, resulting in long development cycles, high costs, and difficulty in finding optimal parameter combinations under complex coupling effects of multiple factors. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for optimizing the abrasion resistance of textiles based on machine learning, provided in the first embodiment of the present invention. Figure 2 This is a schematic diagram of a machine learning-based textile abrasion resistance optimization device provided in the second embodiment of the present invention. Detailed Implementation

[0020] 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.

[0021] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] This invention discloses a method, apparatus, equipment, and storage medium for optimizing the abrasion resistance of textiles based on machine learning. It aims to solve the problem that the current research and development of abrasion-resistant textiles relies on trial-and-error experiments, resulting in long development cycles, high costs, and difficulty in finding the optimal parameter combination under the complex coupling of multiple factors.

[0023] The first embodiment of the present invention provides a method for optimizing the abrasion resistance of textiles based on machine learning, which can be executed by a machine learning-based textile abrasion resistance optimization device (hereinafter referred to as optimization device or system), specifically, executed by one or more processors within the optimization device or system, to at least achieve the following steps: S101, Obtain a textile abrasion resistance database, the database including fiber raw material parameters, fabric structure parameters and finishing process parameters and corresponding abrasion resistance test results; In this embodiment, the optimization device or system can be a desktop computer, laptop computer, server, workstation, or other terminal with data processing capabilities. The optimization device can be equipped with a corresponding operating system and application software, and the functions required in this embodiment can be realized through the combination of the operating system and application software.

[0024] Specifically, in this embodiment, the basic architecture of the database is first established using a relational database management system, such as MySQL or PostgreSQL, and a data model containing multiple interrelated data tables is designed. The core structure of the database revolves around four main dimensions: fiber raw material parameters, fabric structure parameters, finishing process parameters, and abrasion resistance test results. These dimensions are linked together using unique identifiers. In the implementation of fiber raw material parameter collection, the system obtains raw material information through multiple channels. Fiber type parameters cover natural fibers such as cotton, linen, silk, and wool; chemical fibers such as polyester, nylon, acrylic, and spandex; and various high-performance fibers such as aramid, carbon fiber, and ultra-high molecular weight polyethylene fiber. Each fiber type is uniquely identified through standardized coding. Linear density parameters use decitex (dtex) or denier (D) as the unit of measurement, obtaining precise values ​​through fiber fineness analyzers or microscopy. The measurement range covers 0.5 to 100 dtex, meeting the full range of requirements from microfiber to coarse denier fibers. Fiber length parameters use different measurement methods depending on the fiber type; short fibers are measured using a comb-type length analyzer, while filaments have their continuous length characteristics recorded. Fiber strength is obtained through single-fiber strength testing, recording breaking strength (cN / dtex) and breaking elongation (%) data. Yarn linear density is expressed in English count (Ne) or metric count (Nm), determined using a yarn length measuring machine and a precision balance. Twist parameters are measured using a twist meter, recording the number of twists per meter or inch, and the twist direction (S-twist or Z-twist). The blending ratio is accurate to one decimal place, and the consistency between the actual blending ratio and the design ratio is verified by chemical analysis or microscopy.

[0025] During the acquisition of fabric structure parameters, the weaving method parameters are recorded in detail, including basic weaves such as plain weave, twill weave, and satin weave, as well as complex weaves such as jacquard, double-layer, and multi-layer weaves. Each weave is digitally stored using weave diagrams and heddle charts. Warp and weft density parameters are measured using a fabric density microscope under standard atmospheric conditions. Multiple locations on the fabric are selected for measurement, and the average value is taken, with units of ends / 10cm, achieving a measurement accuracy of 0.5 ends. Fabric weight is measured using a circular sampler to cut standard area samples, weighed using a precision balance, and the result is converted to grams per square meter. At least five different locations are tested for each sample, and the coefficient of variation is calculated to ensure data reliability. For knitted fabrics, additional characteristic parameters such as loop density and loop length are recorded; for nonwoven fabrics, process parameters such as web formation and consolidation methods are recorded.

[0026] In the data collection and implementation of post-finishing process parameters, a complete classification system was established for finishing agent type parameters, including resin finishing agents such as DMDHEU (dimethyloldihydroxyethylene urea), modified DMDHEU, polyurethane finishing agents, and functional finishing agents such as fluorocarbon water-repellent finishing agents, silicone softening finishing agents, and antibacterial finishing agents. For each finishing agent, key characteristics such as chemical composition, solid content, pH value, and ionicity were recorded. Concentration parameters accurately recorded the content of effective components in the working solution, in g / L or percentage concentration, while also recording the dosage of auxiliaries such as catalysts, penetrants, and pH adjusters. Padding residue was controlled through a two-dip, two-padding process, recording the liquid carry-over rate under different pressures, typically ranging from 60% to 120%, accurately determined by weighing. Baking temperature parameters recorded the pre-baking temperature (80-120°C) and baking temperature (140-180°C), with temperature sensors monitoring and recording temperature fluctuations in real time. Baking time was accurate to the second, recording the actual residence time of the fabric in the drying chamber, verified by calculations based on speed and drying chamber length.

[0027] The abrasion resistance test results were obtained strictly in accordance with international standardized procedures. For the Martindale abrasion test according to ISO 12947-2, the specimen was mounted on an abrasive table, and a standard pressure (9 kPa for lightweight fabrics, 12 kPa for heavyweight fabrics) was applied. Friction was performed along a Lissajous figure, and the number of friction cycles at which the fabric broke was recorded as the abrasion resistance revolutions. For tests using ASTM D4966, a Wyzenbeek abrasion tester was used, and reciprocating friction was performed under specified tension. The number of cycles at which breakage occurred was also recorded. At least three specimens were tested for each sample; the number of tests was increased when the coefficient of variation exceeded 10%. Breaking strength was tested according to ISO 13934-1 or ASTM D5034 standards using an electronic tensile testing machine under standard atmospheric conditions. The breaking strength (N) and elongation at break (%) in the warp and weft directions were recorded. Tear strength was tested using the pendulum method or single-tongue method, performed according to ISO 13937-2 standards. The tear strength (N) in the warp and weft directions was recorded. Pilling resistance is tested using the pilling box method or the Martindale method, and rated according to ISO 12945-2 standards, employing a 1-5 rating system, where level 5 indicates no pilling and level 1 indicates severe pilling. Rating is conducted using standard sample photos for comparison, and image analysis systems are used for objective evaluation when necessary. Fabric thickness is measured using a fabric thickness gauge under specified pressure, typically 20 cN / cm², with a measurement accuracy of 0.01 mm. Ten different locations are tested for each sample. Air permeability is tested according to ISO 9237 standards, measuring airflow under specified pressure differentials (typically 100 Pa or 200 Pa), in mm / s. Cost data includes a detailed breakdown of raw material costs, processing costs, and finishing costs for cost-effectiveness analysis.

[0028] S102, call the data preprocessing and feature engineering algorithm to process the data in the database, and use the processed data to train and generate a wear resistance prediction model; It should be noted that in the data preprocessing stage, the raw data extracted from the database is first thoroughly cleaned. Multiple imputation methods can be used to handle missing values. For numerical missing data, an iterative imputation algorithm based on random forest is used to predict missing values ​​based on other complete features. For categorical missing data, a strategy of mode imputation or creating a "missing" category is adopted. Outlier detection employs a combined approach: box plots are used to identify numerical anomalies exceeding 1.5 times the interquartile range, while the isolated forest algorithm is used to detect anomalous samples in the multidimensional feature space. For confirmed outliers, a processing strategy of removal, correction, or retention and labeling is selected based on the specific circumstances.

[0029] During feature encoding, one-heat encoding transformation is performed on categorical features such as fiber type, weaving method, and finishing agent type, converting each category into a binary vector representation to ensure the model can correctly handle non-numerical inputs. For numerical features such as linear density, intensity, warp density, weft density, concentration, and temperature, the Z-score standardization method is used to convert the feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of different dimensions and numerical ranges on model training. The standardization formula is (x-μ) / σ, where μ is the feature mean and σ is the standard deviation.

[0030] In the feature selection process, multiple methods are used to screen the optimal feature subset. First, the Pearson correlation coefficient between each feature and the wear resistance revolution is calculated to initially screen features with a correlation greater than 0.3. Then, the recursive feature elimination method combined with the feature importance score of random forest is applied to gradually eliminate features with low contribution. Finally, the feature space is further compressed by L1 regularization of LASSO regression to retain the core feature set that is most explanatory to the prediction target.

[0031] Based on textile theory, feature engineering innovatively constructs a comprehensive feature called fabric coverage coefficient, which is expressed by the formula... Where CF is the fabric coverage factor. EPI is the warp yarn density, and EPI is the warp density. PPI represents weft yarn linear density. It effectively reflects the fabric's tightness and yarn coverage, and is strongly correlated with abrasion resistance. In addition, derived variables such as the interaction characteristics of yarn strength and twist, and the ratio of fabric density to weight were constructed, enriching the feature dimensions.

[0032] During the model training phase, Gradient Boosting Decision Tree (GBDT) was selected as the core prediction algorithm. GBDT integrates multiple weak learners to form a strong prediction model, making it particularly suitable for handling the nonlinear and complex relationship between textile parameters and performance. The dataset can be randomly divided into training and test sets in an 8:2 ratio, while stratified sampling is used to ensure the consistency of the distribution of samples with different abrasion resistance levels in the training and test sets.

[0033] The model training process employs grid search combined with 5-fold cross-validation to optimize hyperparameters, including key parameters such as tree depth (3-10 layers), learning rate (0.01-0.3), subsampling ratio (0.6-1.0), and feature sampling ratio (0.6-1.0). An early stopping strategy is implemented during training; iteration stops when the validation set error no longer decreases after 20 consecutive rounds to prevent overfitting. The model learns the complex mapping relationship from fiber raw material parameters, fabric structure parameters, finishing process parameters to abrasion resistance revolutions by minimizing the mean squared error loss function.

[0034] Model evaluation employs multiple metrics to comprehensively measure predictive performance. The coefficient of determination reflects the model's explanatory power for data variation; a coefficient closer to 1 indicates a better model fit. The root mean square error (RMSE) is calculated using the formula... The calculation intuitively reflects the average deviation between the predicted and actual values. Mean absolute error (MSE) provides another perspective on error measurement. Implementation results show that the optimized model achieves a prediction accuracy of >0.85, <150 rpm, and <100 rpm on the test set, meeting the requirements for industrial applications.

[0035] S103, based on the wear resistance performance prediction model and combined with the intelligent optimization algorithm, guided by the set wear resistance performance target, the optimal combination of material, structure and process parameters is searched and generated in the parameter space. In this embodiment, a genetic algorithm can be used as the core optimization engine. This algorithm simulates the biological evolution process and finds the global optimal solution through population iterative evolution. First, the optimization objective is to maximize the abrasion resistance revolutions, which is set as the fitness function, i.e., fitness = f(X), where X represents a vector containing multi-dimensional design parameters such as fiber type, linear density, twist, blend ratio, warp density, weft density, fabric structure, finishing agent type, concentration, and baking temperature, and f() is the trained abrasion resistance prediction model.

[0036] The parameter space is defined based on the feasible domain of the actual production process. Each design variable has a set reasonable value range, such as fiber linear density (50-300 dtex), warp density (100-500 threads / 10cm), weft density (80-400 threads / 10cm), finishing agent concentration (20-150 g / L), and baking temperature (140-180°C). This ensures the integrity of the search space while avoiding unrealistic parameter combinations. For discrete variables such as fiber type and fabric structure, a coding mapping table is established to convert them into integer codes for easier algorithm processing.

[0037] The constraints are set with full consideration of practical application needs and production limitations. Fabric weight constraints ensure the lightweight requirements of the product, and the upper limit is usually set at 200-350 g / m². 2 Specific values ​​are adjusted based on product positioning. Breathability constraints ensure wearing comfort, with lower limits set according to intended use; for example, sportswear requires a breathability greater than 100L / m². 2 / s, while the standards for protective clothing can be appropriately lowered. Cost constraints are achieved by establishing a cost calculation model, comprehensively considering raw material costs, processing costs, and finishing costs, and setting a total cost ceiling to ensure the economic feasibility of the solution. It should be noted that the constraints are integrated into the optimization process using the penalty function method. When a candidate solution violates the constraints, its fitness value will be penalized, and the degree of penalty is proportional to the degree of violation.

[0038] During the algorithm initialization phase, an initial population of 100-200 individuals is randomly generated. Each individual represents a complete set of design parameters. Real number encoding and hybrid encoding strategies are employed, with continuous variables directly represented by real numbers and discrete variables encoded by integers. To ensure the diversity of the initial population, Latin hypercube sampling is used to uniformly distribute the initial solutions in the parameter space, avoiding local clustering.

[0039] The fitness evaluation process inputs each candidate parameter combination into the prediction model to obtain the predicted wear resistance revolutions, while simultaneously calculating the degree of constraint violation, and comprehensively forming the fitness value. To improve evaluation efficiency, a parallel computing strategy is implemented, with multiple candidate schemes being predicted and evaluated simultaneously, significantly shortening the evaluation time per generation.

[0040] The selection process employs a tournament selection mechanism, randomly selecting 3-5 individuals from the population each time, with the individual with the highest fitness being chosen as the parent. This method ensures the inheritance of superior individuals while maintaining population diversity. The elite preservation strategy ensures that the best 5-10% of individuals in each generation directly enter the next generation, preventing the loss of optimal solutions.

[0041] For continuous variables, simulated binary crossover (SBX) is used to generate offspring through probability distribution, maintaining the continuity of the search and local search capability. For discrete variables, uniform crossover is used, exchanging parent genes with a probability of 0.6-0.8. An adaptive mechanism is introduced during the crossover process, dynamically adjusting the crossover probability according to the population convergence. A higher crossover rate is used initially to promote exploration, and the crossover rate is reduced in the later stages to enhance local search.

[0042] The mutation operation employs a polynomial mutation operator to handle continuous variables. The mutation intensity is adaptively adjusted with the iteration process. Initially, the mutation rate is set to 0.1-0.2 to promote global exploration, and gradually reduced to 0.01-0.05 as iterations progress to enhance local fine-grained search. For discrete variables, random reset mutation is used to replace gene values ​​with random values ​​within the domain with a relatively low probability (0.05-0.1).

[0043] The algorithm iteration process is set with multiple termination conditions, including reaching the maximum number of iterations (usually 500-1000 generations), no improvement in the optimal solution for 50 consecutive generations, or the fitness value reaching a preset target value. After each iteration, the best individual, average fitness, and population diversity indicators are recorded to monitor the algorithm's convergence process and adjust strategy parameters.

[0044] To avoid premature convergence, a multi-population parallel evolution strategy is implemented, dividing the total population into 3-5 subpopulations. Each subpopulation evolves independently, and inter-population migration is performed periodically to enhance global search capabilities. Simultaneously, local search operators are introduced to perform gradient descent or pattern search on the best individual in each generation, accelerating convergence and improving solution accuracy.

[0045] The validation of the optimization results includes running the algorithm independently multiple times to obtain the optimal result, performing sensitivity analysis on the obtained optimal parameter combination, evaluating the impact of small parameter changes on performance, and ensuring the robustness of the solution. The final output of the optimal parameter combination not only includes specific design parameter values, but also provides predicted wear resistance speeds, various constraint index values, cost estimates, and confidence intervals for parameter selection to support decision-making.

[0046] S104, acquire new experimental data of the sample prepared according to the optimal parameter combination, and feed the new experimental data back to the database to retrain the wear resistance prediction model to form a closed-loop adjustment.

[0047] It should be noted that during the sample preparation stage, actual production is carried out according to the optimal parameter combination output by the intelligent optimization algorithm, including the selected fiber raw material ratio, yarn specifications, weaving process conditions, and finishing process parameters. Detailed process tracking records are established during the preparation process, and the actual parameter values ​​of each step are collected and recorded in real time using a Manufacturing Execution System (MES), such as the actual material input ratio, machine operating parameters, and ambient temperature and humidity. These actual production parameters may have slight deviations from the theoretical design values ​​and need to be accurately recorded for subsequent analysis. Multiple parallel samples are prepared for each batch, typically 5-10.

[0048] The acquisition of new experimental data followed standardized testing procedures, and comprehensive performance testing was conducted on the prepared samples. Abrasion resistance testing was strictly performed according to ISO 12947-2 or ASTM D4966 standards. After 24 hours of pre-conditioning under constant temperature and humidity conditions (temperature 20±2°C, relative humidity 65±4%), the actual abrasion resistance revolutions for each sample were recorded, and the average value, standard deviation, and coefficient of variation were calculated. Other key performance indicators, including actual basis weight, thickness, air permeability, breaking strength, and tear strength, were also tested simultaneously.

[0049] The data validation phase involves quality checks on newly acquired experimental data. First, it verifies whether the experimental conditions meet standard requirements and whether the testing equipment has been calibrated. Grubbs' criterion is used to identify potential outliers, and data points exceeding three standard deviations are retested. The deviation between the experimental data and the model's predicted values ​​is calculated using the formula (measured value - predicted value) / predicted value × 100%. Under normal circumstances, the deviation rate should be controlled within ±15%; deviations exceeding this range require further analysis to determine the cause.

[0050] The data feedback mechanism employs an automated data transmission interface, transmitting validated new experimental data to the central database in real time via API. Upon data entry, a timestamp, batch number, and data source identifier are automatically generated. New data is correlated with historical data to identify innovative parameter combinations and performance improvements. The database management system automatically updates relevant statistical indicators, such as parameter distribution range, mean, and standard deviation.

[0051] The model retraining strategy employs an incremental learning approach, avoiding retraining from scratch and instead updating the existing model with new data. Retraining is triggered when a certain amount of new data accumulates (typically 5-10% of the original training set) or periodically (e.g., monthly). During retraining, new and historical data are mixed with time-weighted data, with more recent data given higher weights to reflect the impact of technological advancements and process improvements. Online learning algorithms, such as variants of stochastic gradient descent, are used to enable the model to quickly adapt to new data patterns.

[0052] Closed-loop regulation is achieved by establishing a feedback control mechanism, using the deviation between measured and predicted performance as a feedback signal to automatically adjust model parameters and optimization strategies. When the system detects that the prediction error exceeds a set threshold, it triggers model structure adjustments, such as increasing the decision tree depth, adjusting regularization parameters, or introducing new features. Based on accumulated deviation patterns, the system identifies systematic biases in the model, such as overestimating or underestimating predictions for specific fiber types or process conditions, and compensates for these biases using a deviation correction algorithm.

[0053] The continuous improvement mechanism tracks the model's evolution through a version management system, generating a new version each time it is retrained and recording the trends in model performance metrics. An A / B testing strategy is employed, running the new version alongside the old version for a period of time, comparing actual prediction results to determine whether a full switch is necessary. A performance monitoring dashboard is established to display key metrics such as model prediction accuracy, bias distribution, and confidence intervals in real time, providing timely alerts when performance declines.

[0054] The second embodiment of the present invention provides a machine learning-based device for optimizing the abrasion resistance of textiles, comprising: The database query unit is used to obtain a database of textile abrasion resistance properties, which includes fiber raw material parameters, fabric structure parameters, finishing process parameters, and corresponding abrasion resistance test results. The model building unit is used to call data preprocessing and feature engineering algorithms to process the data in the database, and use the processed data to train and generate a wear resistance prediction model. The optimal combination search unit is used to search and generate the optimal combination of material, structure and process parameters in the parameter space based on the wear resistance performance prediction model and combined with the intelligent optimization algorithm, guided by the set wear resistance performance target. The feedback unit is used to acquire new experimental data of the sample prepared according to the optimal parameter combination, and feed the new experimental data back to the database to retrain the wear resistance prediction model to form a closed-loop adjustment.

[0055] The third embodiment of the present invention provides a textile abrasion resistance optimization device based on machine learning, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the textile abrasion resistance optimization method based on machine learning as described in any of the above embodiments.

[0056] The fourth embodiment of the present invention provides a computer-readable storage medium, characterized in that it stores a computer program, which can be executed by the processor of the device where the computer-readable storage medium is located, to implement the machine learning-based textile abrasion resistance optimization method as described in any of the above embodiments.

[0057] Based on the machine learning-based method, apparatus, equipment, and storage medium provided by this invention, a database containing fiber raw materials, fabric structure, finishing processes, and performance test results is constructed to transform scattered experiential knowledge into structured data assets. By calling data preprocessing and feature engineering algorithms to process data and train an abrasion resistance prediction model, a precise mapping relationship from design parameters to performance results is established. Combining the prediction model with intelligent optimization algorithms enables automatic searching for optimal combinations within a vast parameter space. Finally, a closed-loop mechanism, through feeding new experimental data back to the database and retraining the model, allows the model to continuously learn and iteratively optimize, constantly improving prediction accuracy and applicability. This solves the problem that existing abrasion-resistant textile R&D relies on trial-and-error experiments, resulting in long development cycles, high costs, and difficulty in finding optimal parameter combinations under complex coupling effects of multiple factors.

[0058] Exemplary examples show that the computer program described in the third and fourth embodiments of the present invention can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in implementing a machine learning-based textile abrasion resistance optimization device. For example, the apparatus described in the second embodiment of the present invention.

[0059] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the aforementioned machine learning-based method for optimizing the abrasion resistance of textiles, connecting various parts of the entire system through various interfaces and circuits.

[0060] The memory can be used to store the computer program and / or modules. The processor, by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory, implements various functions of a machine learning-based method for optimizing the abrasion resistance of textiles. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, text conversion function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0061] If the implemented module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0062] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0063] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine learning based method for textile wear performance optimization, characterized in that, The method comprises the following steps: obtaining a textile wear resistance database, the database comprising fiber raw material parameters, fabric structure parameters and finishing process parameters and corresponding wear resistance test results; calling a data preprocessing and feature engineering algorithm to process the data in the database, and using the processed data to train and generate a wear resistance prediction model; according to the wear resistance prediction model and combining an intelligent optimization algorithm, searching and generating an optimal material, structure and process parameter combination in the parameter space with a set wear resistance target as a guide; obtaining new experimental data of a sample prepared according to the optimal parameter combination, and feeding the new experimental data back to the database to retrain the wear resistance prediction model to form a closed-loop regulation.

2. A machine learning based textile abrasion resistance optimization method according to claim 1, characterized in that, The fiber raw material parameters include fiber type, linear density, length, strength, yarn linear density, twist and blending ratio; The fabric structure parameters include weaving method, warp density, weft density and fabric weight; The finishing process parameters include finishing agent type, concentration, padding pick-up, curing temperature and time; The wear resistance test results include wear resistance revolutions measured according to ISO 12947-2 standard or ASTM D4966 standard as a core index, and breaking strength, tearing strength, anti-pilling grade, fabric thickness, air permeability and cost as auxiliary indexes.

3. A machine learning based textile abrasion resistance optimization method according to claim 1, characterized in that, The calling of the data preprocessing and feature engineering algorithm to process the data in the database is specifically: cleaning the data, processing missing values and removing abnormal values; one-hot encoding the category features and standardizing the numerical features; using a feature selection method to screen a feature subset related to wear resistance revolutions, and constructing new features based on textile theory, including fabric coverage coefficient, the calculation formula of the fabric coverage coefficient being: where CF is the fabric cover factor, is the warp yarn density, EPI is the warp ends per inch, is the weft yarn density, PPI is the weft ends per inch.

4. A machine learning based textile abrasion resistance optimization method according to claim 1, wherein, The use of the processed data to train and generate a wear resistance prediction model is specifically: selecting a gradient boosting decision tree as a prediction model, and dividing the data set into a training set and a test set; using the training set to train the prediction model to learn the mapping relationship from design parameters and process parameters to wear resistance revolutions; Using the determination coefficient R 2 the root mean square error RMSE and the mean absolute error MAE, wherein the RMSE is calculated as: where n is the number of samples, is the true wear resistant revolutions, is the predicted wear resistant revolutions.

5. A machine learning based textile abrasion resistance optimization method according to claim 4, characterized in that, The searching and generation of the optimal material, structure and process parameter combination in the parameter space are specifically: adopting a genetic algorithm, setting the optimization target as maximizing wear resistance revolutions, and setting constraint conditions; generating candidate parameter combinations in the parameter space, inputting the candidate parameter combinations into the prediction model to obtain predicted wear resistance revolutions; generating next-generation candidate combinations according to the prediction results through selection, crossover and mutation operations, and iterating until the optimal combination satisfying the constraint conditions is found.

6. A machine learning based textile abrasion resistance optimization method according to claim 4, wherein, The constraint conditions include: fabric weight less than a preset threshold, air permeability greater than a preset threshold, and cost less than a preset threshold.

7. A machine learning based textile abrasion resistance performance optimization apparatus, characterized by, The method comprises the following steps: a database query unit for obtaining a textile wear resistance database, the database comprising fiber raw material parameters, fabric structure parameters and finishing process parameters and corresponding wear resistance test results; a model construction unit for calling a data preprocessing and feature engineering algorithm to process the data in the database, and using the processed data to train and generate a wear resistance prediction model; An optimal combination searching unit is configured to search and generate an optimal combination of material, structure and process parameters in a parameter space according to the wear resistance performance prediction model and in combination with an intelligent optimization algorithm, and to guide the searching according to a set wear resistance performance target; A feedback unit is configured to obtain new experimental data of a sample prepared according to the optimal parameter combination, and to feed the new experimental data back to the database, so as to retrain the wear resistance performance prediction model to form a closed-loop regulation.

8. A machine learning based textile abrasion resistance performance optimization apparatus, characterized by, A computer readable storage medium stores a computer program, and the computer program can be executed by a processor to implement the machine learning-based textile wear resistance performance optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer readable storage medium stores a computer program, and the computer program can be executed by a processor to implement the machine learning-based textile wear resistance performance optimization method according to any one of claims 1 to 6.

Images