Textile color consistency detection method

Abstract

The invention provides a textile color consistency detection method, which comprises the following steps: acquiring spectral data of a to-be-detected product, comprehensively scanning the surface of the product through multispectral imaging, and performing noise reduction and enhancement processing on scanning data by adopting an image processing algorithm to obtain high-precision color difference detection data; according to the optimal dyeing process parameters and the pretreatment scheme, numerical simulation is adopted to simulate the dyeing process, the diffusion process of the dye in the fiber is calculated through finite element analysis, and a dyeing uniformity prediction result is obtained; and comparing the prediction result with a color fastness test standard, judging whether the dyed fabric can meet the product quality requirement or not, if so, warehousing the batch of products, and if not, feeding back the prediction result to a dyeing process control system, and carrying out the next round of optimization iteration until the product quality requirement is met.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a method for detecting color consistency of textiles. Background Art

[0002] There is a complex technical contradiction in color fastness testing and color management of textiles. Color fastness testing is designed to evaluate the ability of dyed products to maintain color stability under various conditions, while color tolerance setting is to control color differences during the production process. However, there is a potential conflict between the two. Overly strict color tolerance standards may result in a large number of products being judged as unqualified due to slight color differences, even if the color fastness of these products fully meets the requirements. This not only increases production costs, but may also lead to waste of resources. To make it more complicated, different fiber materials have different absorption and retention of dyes, which makes it extremely difficult to formulate unified color fastness and color tolerance standards. For example, natural fibers and synthetic fibers exhibit completely different characteristics during the dyeing process, requiring different evaluation standards. This technical contradiction highlights the challenge of balancing color fastness requirements, color consistency and production efficiency while ensuring product quality. It requires the textile industry to innovate in testing methods, standard setting and production processes to find more flexible and precise color management solutions. Summary of the invention

[0003] The present invention provides a method for detecting color consistency of textiles, which mainly comprises:

[0004] According to the differences in the hygroscopicity, air permeability and dyeing properties of natural and synthetic fibers, the color fastness test standards and color tolerance setting ranges of different fiber materials are established. The historical data are analyzed using a machine learning algorithm, and the dyeing characteristics of different fiber materials are classified through a clustering method to obtain a typical color difference distribution model for various fiber materials.

[0005] According to the obtained color difference distribution model, the support vector machine algorithm is used to establish a color difference judgment model. If the product to be tested is made of natural fiber material, the color difference judgment standard of the product is dynamically relaxed according to the characteristics that its dyeing uniformity is easily affected by the fiber composition and structure. If the product to be tested is made of synthetic fiber material, the color difference judgment standard of the product is dynamically tightened according to the characteristics that its dyeing uniformity is relatively easy to control.

[0006] Obtain the spectral data of the product to be tested, conduct a comprehensive scan of the product surface through multi-spectral imaging, and use image processing algorithms to reduce noise and enhance the scanned data to obtain high-precision color difference detection data;

[0007] Based on high-precision color difference detection data, a deep learning algorithm is used to analyze the color difference distribution of the product surface, and the color difference features are extracted through a convolutional neural network. If the product to be tested is made of natural fiber material, it is determined whether the color difference is caused by uneven fiber composition and structure. If the product to be tested is made of synthetic fiber material, it is determined whether the color difference is caused by improper dyeing process control.

[0008] For products with uneven dyeing, the dyeing process parameters of the batch of products are obtained. If the product to be tested is made of natural fiber, the fiber pretreatment process is optimized and adjusted using a fuzzy control algorithm. If the product to be tested is made of synthetic fiber, the dyeing process parameters are optimized using a genetic algorithm. The optimal dye selection and ratio scheme is obtained through iterative calculation.

[0009] According to the optimal dyeing process parameters and pretreatment scheme, the dyeing process is simulated by numerical simulation, and the diffusion process of the dye in the fiber is calculated by finite element analysis to obtain the prediction result of dyeing uniformity.

[0010] The prediction results are compared with the color fastness test standards to determine whether the dyed fabric can meet the product quality requirements. If so, the batch of products will be put into storage. If not, the prediction results will be fed back to the dyeing process control system for the next round of optimization iterations until the product quality requirements are met.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for detecting color consistency of textiles. First, according to the differences between natural fibers and synthetic fibers in terms of hygroscopicity, air permeability and dyeing performance, targeted color fastness test standards and color difference judgment standards are established, solving the problem that traditional standards cannot adapt to the dyeing characteristics of different materials; secondly, by performing multi-spectral imaging scanning on the product to be tested, a deep learning algorithm is used to analyze the color difference distribution characteristics, and the dyeing characteristics of different fiber materials are classified, thereby reducing manual analysis errors, improving the understanding and prediction capabilities of dyeing characteristics, and improving the precision and accuracy of color difference detection; in addition, according to the fiber type, the judgment standard is dynamically adjusted, and the dyeing process parameters are optimized in a targeted manner, so that according to different fiber materials and the reasons for the uneven dyeing determined, the dyeing process parameters are optimized by using appropriate algorithms, respectively, so as to reduce the defective rate and improve the generation efficiency; finally, the numerical simulation technology is used to predict the dyeing uniformity, solving the problem of lack of visualization and predictability in the actual dyeing process. In summary, the present invention realizes the intelligent detection of the color of dyed products, improves the product qualification rate, and reduces the production cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1The present invention is a flowchart of a method for detecting color consistency of textiles.

[0014] Figure 2 It is a schematic diagram of a method for detecting color consistency of textiles according to the present invention.

[0015] Figure 3 This is another schematic diagram of a method for detecting color consistency of textiles according to the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1 -3. A textile color consistency detection method in this embodiment may specifically include:

[0018] Step S101, according to the differences in the detailed properties of hygroscopicity, air permeability and dyeing performance of natural fibers and synthetic fibers, establish the color fastness test standards and color tolerance setting ranges for different fiber materials, use a machine learning algorithm to analyze historical data, classify the dyeing characteristics of different fiber materials through a clustering method, and obtain a typical color difference distribution model for each type of fiber material.

[0019] Obtain fiber attribute data and corresponding color fastness and color tolerance data, and use multiple regression analysis to establish a relationship model between fiber attributes and color fastness and color tolerance. Calculate the regression coefficient and significance test results of the relationship model to determine the degree of influence of various indicators on the color fastness and the color tolerance. According to the degree of influence, obtain the color fastness test standard and color tolerance setting range of different fiber materials. Use a hierarchical clustering algorithm to analyze historical dyeing data, determine the optimal number of clusters by calculating the silhouette coefficient, and divide the dyeing characteristic categories of different fiber materials. For each type of fiber material, use normal distribution, lognormal distribution and beta distribution to fit the color difference data. Calculate the maximum likelihood estimate and information criterion, select the distribution type with the best fitting effect, and obtain the typical color difference distribution model of each type of fiber material.

[0020] Exemplarily, according to the differences in properties such as hygroscopicity, air permeability, and dyeing performance between natural and synthetic fibers, a multivariate regression analysis is used to establish a relationship model between fiber properties and color fastness and color tolerance. By calculating the regression coefficient and the significance test, the influence of each index on color fastness and color tolerance is determined, thereby obtaining the color fastness test standard and color tolerance setting range of different fiber materials. The hierarchical clustering algorithm is used to analyze the historical dyeing data, clustering is performed according to multidimensional features such as fiber material, dye type, and process parameters, and the optimal number of clusters is determined by calculating the silhouette coefficient, thereby dividing the dyeing characteristic categories of different fiber materials. For each fiber material category, the color difference data is fitted using normal distribution, lognormal distribution, and beta distribution, respectively. By calculating the maximum likelihood estimate and information criteria such as AIC or BIC, the distribution type with the best fitting effect is selected to obtain the typical color difference distribution model of various fiber materials. In the implementation process, representative fiber samples such as cotton, polyester, and wool are first selected to measure their hygroscopicity, air permeability, and dyeing performance. Taking hygroscopicity as an example, the drape method was used to measure the hygroscopicity of cotton fiber to be 7%, polyester to be 0.4%, and wool to be 15%. These data were input into the multiple regression model to obtain the regression equation Y=2.5X1+1.8X2+3.2X3+0.5, where Y is the color fastness rating, and X1, X2, and X3 are hygroscopicity, air permeability, and dyeing performance, respectively. Through calculation, it is known that for cotton fiber, its color fastness rating is 4 levels, and the color tolerance range is ±1.5ΔE. Subsequently, a hierarchical clustering analysis was performed on the historical dyeing data, using Euclidean distance as the similarity metric, and Ward's minimum variance method to construct a cluster tree. By calculating the silhouette coefficient under different cluster numbers, it was found that when the number of clusters was 5, the silhouette coefficient reached a maximum value of 0.68, so the fiber material was divided into 5 dyeing characteristic categories. For each category, normal distribution, lognormal distribution, and beta distribution were tried for fitting. Taking the first category as an example, the AIC value of the normal distribution is 245.3, the lognormal distribution is 238.7, and the beta distribution is 251.2. Therefore, the lognormal distribution is selected as the color difference distribution model for this category.

[0021] Step S102, based on the obtained color difference distribution model, a color difference judgment model is established using a support vector machine algorithm. If the product to be tested is made of natural fiber material, the product color difference judgment standard is dynamically relaxed based on the characteristic that its dyeing uniformity is easily affected by the fiber composition and structure; if the product to be tested is made of synthetic fiber material, the product color difference judgment standard is dynamically tightened based on the characteristic that its dyeing uniformity is relatively easy to control.

[0022] The color difference distribution model is obtained, and a color difference judgment model is constructed by using a support vector machine algorithm; according to the color difference judgment model, the optimal parameters are determined by a grid search method, and a radial basis function is selected as a kernel function; the training data is standardized and then input into the color difference judgment model for training to obtain a decision boundary for color difference judgment. For natural fiber materials, fiber length, fineness and composition information are extracted from a pre-established fiber property database; based on the fiber length, fineness and composition information, a weighted average method is used to calculate the dyeing uniformity influence factor; if the dyeing uniformity influence factor is greater than a preset threshold, the tolerance range of the color difference judgment standard is expanded by a preset proportion. For synthetic fiber materials, fiber length, fineness and composition information are extracted from a pre-established fiber property database; based on the fiber length, fineness and composition information, a weighted average method is used to calculate the dyeing uniformity influence factor; if the dyeing uniformity influence factor is less than a preset threshold, the tolerance range of the color difference judgment standard is reduced by a preset proportion. Traverse different color tolerance setting ranges, calculate the judgment accuracy and error rate for each color tolerance setting range; use the F1 score as an evaluation indicator based on the judgment accuracy and error rate; select the color tolerance setting range with the highest F1 score as the optimal result to determine the final color difference judgment standard for the product. Obtain production batch and seasonal variation factor information; regularly update the color difference judgment model and parameters based on the production batch and seasonal variation factor information; dynamically adjust the color difference judgment standard by updating the color difference judgment model and parameters.

[0023] Exemplarily, according to the obtained color difference distribution model, the support vector machine algorithm is used to construct the color difference judgment model, the optimal parameters are determined by the grid search method, the radial basis function is selected as the kernel function, the training data is standardized and input into the model for training, the decision boundary of color difference judgment is obtained, and the one-to-many strategy is used to handle the multi-category color difference judgment problem. For natural fiber materials, the fiber length, fineness, composition and other information are extracted from the pre-established fiber property database, and the weighted average method is used to calculate the dyeing uniformity influencing factor. Through the statistical analysis of a large amount of historical data, the preset threshold of 0.8 is obtained. If the influencing factor is greater than the threshold, the tolerance range of the color difference judgment standard is expanded by 20%. For synthetic fiber materials, the same method is used to calculate the dyeing uniformity influencing factor, and the preset threshold of 0.5 is obtained through historical data analysis. If the influencing factor is less than the threshold, the tolerance range of the color difference judgment standard is reduced by 15%. According to industry standards and actual production, set a reasonable color tolerance range and step size. By traversing different color tolerance setting ranges, calculate the judgment accuracy and misjudgment rate for each range, use the F1 score as the evaluation index, and select the color tolerance setting range with the highest F1 score as the optimal result, thereby determining the final color difference judgment standard for the product. According to factors such as production batches and seasonal changes, regularly update the judgment model and parameters to achieve dynamic adjustment of the color difference judgment standard. During the implementation process, first preprocess the color difference distribution model and standardize the color difference data of 10,000 samples to the range of 0-1. The support vector machine algorithm is used to construct the color difference judgment model, and the grid search method is used to select the C value [0.1, 1, 10, 100] and the gamma value.

[0024] The parameters were optimized in the range of [0.001, 0.01, 0.1, 1], and the optimal parameters were finally determined to be C=10, gamma=0.1. The model was trained using these parameters, and a one-to-many strategy was used to handle the classification problem of 5 color difference levels, resulting in a judgment model with an accuracy of 92%. For natural fiber materials, information such as fiber length such as 30mm, fineness such as 1.5dtex, and composition such as 100% cotton was extracted from the fiber property database, and the weighted average method was used to calculate the dyeing uniformity influence factor, where the length weight was 0.3, the fineness weight was 0.3, and the composition weight was 0.4, resulting in an influence factor of 0.85. Since this value is greater than the preset threshold of 0.8, the tolerance range of the color difference judgment standard is expanded by 20% from the original ±1.5ΔE to ±1.8ΔE. For synthetic fiber materials, the same method was used to calculate the influence factor of 0.4, which is less than the preset threshold of 0.5, so the tolerance range of the color difference judgment standard is reduced by 15% from ±1.0ΔE to ±0.85ΔE. When determining the final color difference judgment standard, the color tolerance range is set to 0.5ΔE to 2.0ΔE, with a step size of 0.1ΔE, and the F1 score of each range is calculated. For natural fibers, the highest F1 score of 0.88 appears at 1.8ΔE; for synthetic fibers, the highest F1 score of 0.92 appears at 0.9ΔE, and the final color difference judgment standard is determined based on this. The judgment model and parameters are updated once a month, and the influencing factors are recalculated and the support vector machine parameters are optimized based on the production data of the past three months to achieve dynamic adjustment of the color difference judgment standard.

[0025] Step S103, obtaining spectral data of the product to be tested, performing a comprehensive scan of the product surface through multi-spectral imaging, and using an image processing algorithm to perform noise reduction and enhancement processing on the scanned data to obtain high-precision color difference detection data.

[0026] The original multispectral image data of the surface of the product to be tested is obtained by a multispectral imaging device performing a comprehensive scan on the surface of the product to be tested; an adaptive median filtering algorithm is used to perform preprocessing on the original multispectral image data to obtain a preprocessed multispectral image, wherein the adaptive median filtering algorithm dynamically adjusts the filter window size according to the local noise density; a band adaptive enhancement algorithm is used to perform image enhancement on the preprocessed multispectral image to obtain enhanced multispectral image data, wherein the band adaptive enhancement algorithm uses histogram equalization for the visible light band and contrast-limited adaptive histogram equalization for the near infrared band; a principal component analysis algorithm is used to perform dimensionality reduction processing on the enhanced multispectral image data to obtain a feature vector; the Euclidean distance of each pixel point in the feature space is calculated, and the result is converted into a CIELab color space to obtain color difference detection data; if the average color difference error of the color difference detection data compared with the standard color card is less than a preset threshold, it is determined that the color difference detection data meets the high-precision requirement.

[0027] Exemplarily, the surface of the product to be tested is fully scanned by a multispectral imaging device. First, a white balance calibration is performed using a standard white plate to ensure the consistency of the spectral response of the device. Ten narrow-band filters in the wavelength range of 400-700nm are used, and the bandwidth of each band is 10nm. The scanning resolution is set to 1200dpi, and the scanning speed is adjusted to 50mm / s to obtain the original multispectral image data of the product to be tested. The acquired original multispectral image data is preprocessed, and the salt and pepper noise in the image is removed by an adaptive median filtering algorithm. The filter window size is dynamically adjusted according to the local noise density, and the maximum window is set to 5×5 pixels. For the preprocessed multispectral image, the band adaptive enhancement algorithm is used for image enhancement. Histogram equalization is used for visible light bands such as 400-700nm, and contrast-limited adaptive histogram equalization is used for near-infrared bands such as 700-1000nm. By calculating the cumulative distribution function of each band image, the image grayscale value is remapped to improve the image contrast and detail performance. According to the enhanced multispectral image data, the principal component analysis algorithm is used for dimensionality reduction processing. The first three principal components are selected as feature vectors. The Euclidean distance of each pixel in the feature space is calculated and the result is converted to the CIELab color space. The ΔE*ab color difference formula is used to calculate the high-precision color difference detection data. Finally, by comparing with the standard color card, the average color difference error is calculated to evaluate the accuracy of the color difference detection data. If the average color difference error is less than 0.5ΔE, the color difference detection data is judged to meet the high-precision requirements. In the implementation process, the X-Rite Color Checker standard white board is first used for white balance calibration. During the calibration process, 100 frames of images are collected and the average value is taken as the calibration benchmark. The multispectral imaging equipment uses a SPECIMFX10 camera equipped with 10 narrow-band filters with central wavelengths of 420nm, 450nm, 480nm, 510nm, 540nm, 570nm, 600nm, 630nm, 660nm and 690nm. The scanning resolution is set to 1200dpi, and the scanning speed is adjusted to 50mm / s to obtain the original multispectral image data of the product to be tested. The adaptive median filtering algorithm is applied to the original image, with the initial window size of 3×3 pixels and the maximum window size of 5×5 pixels. The noise density is determined by calculating the grayscale standard deviation of the local area, and the filter window size is dynamically adjusted. The filtered image is adaptively enhanced by bands. The visible light band such as 400-700nm uses histogram equalization, and the near-infrared band such as 700-1000nm uses contrast-limited adaptive histogram equalization. The contrast threshold is set to 0.01. The enhanced multispectral image data is subjected to principal component analysis, and the first three principal components with a cumulative contribution rate of 95% are selected as feature vectors.The feature vector is converted to CIELab color space, and the color difference is calculated using the ΔEab color difference formula, where L, a*, and b* represent lightness, redness, greenness, and yellowness, respectively. Finally, the 24-color X-Rite Color Checker color card is used as the standard to calculate the average color difference error. The results show that the average color difference error is 0.42ΔE, which is less than the threshold of 0.5ΔE and meets the high-precision requirements.

[0028] Step S104: Based on the high-precision color difference detection data, a deep learning algorithm is used to analyze the color difference distribution on the product surface, and the color difference features are extracted through a convolutional neural network. If the product to be tested is made of natural fiber material, it is determined whether the color difference is caused by uneven fiber composition and structure; if the product to be tested is made of synthetic fiber material, it is determined whether the color difference is caused by improper dyeing process control.

[0029] The color difference distribution data of the product surface is obtained, and a convolutional neural network is used to extract features of the color difference distribution data to obtain a color difference feature vector; according to the color difference feature vector, the color difference distribution is classified by a K-means clustering algorithm to determine the degree of color difference and spatial distribution characteristics; if the product is made of natural fiber material, a pre-trained decision tree algorithm is used to determine whether the color difference is caused by uneven fiber composition and structure; if the product is made of synthetic fiber material, a pre-trained support vector machine algorithm is used to determine whether the color difference is caused by improper dyeing process control; for natural fiber material, if it is determined that the color difference is caused by fiber unevenness, a raw material selection and pretreatment process optimization plan is determined; for synthetic fiber material, if it is determined that the color difference is caused by dyeing process, a dye ratio and temperature control parameter adjustment plan is determined.

[0030] For example, based on high-precision color difference detection data, a convolutional neural network is used to extract features of the color difference distribution on the product surface, and the color difference data is converted into a three-channel image input. Five convolution layers and three pooling layers are set, with a convolution kernel size of 3×3 and a step size of 1, and a pooling kernel size of 2×2 and a step size of 2. After each convolution layer, a ReLU activation function and a batch normalization layer are used, and finally a 256-dimensional feature vector is output through a fully connected layer. For the extracted color difference features, the K-means clustering algorithm is used to classify the color difference distribution, and the number of clusters is set to 3. According to the Euclidean distance from the cluster center to the origin, slight color difference <0.5, medium color difference 0.5-1.5 and severe color difference >1.5 are defined, and the area proportion and spatial distribution characteristics of each type of color difference are calculated. If the product to be tested is made of natural fiber, the pre-trained decision tree algorithm is used to determine whether the color difference is caused by uneven fiber composition and structure. The input features of the decision tree include the spatial frequency of the color difference distribution calculated by fast Fourier transform, the directionality calculated by Hough transform, and the local coefficient of variation. The training data comes from the natural fiber color difference sample library with known causes. If the product to be tested is made of synthetic fiber, the pre-trained support vector machine algorithm is used to determine whether the color difference is caused by improper dyeing process control. The input features of the support vector machine include the uniformity of the color difference distribution calculated by local entropy, the continuity calculated by the edge detection algorithm, and the edge features extracted by the Canny operator. The training data comes from the synthetic fiber color difference sample library with known causes. According to the judgment results, for natural fiber materials, if the color difference is caused by fiber unevenness, it is recommended to optimize the raw material selection and pretreatment process; for synthetic fiber materials, if the color difference is caused by the dyeing process, it is recommended to adjust the dye ratio and temperature control parameters. In the implementation process, the high-precision color difference detection data is first converted into a 512×512×3 RGB image as the input of the convolutional neural network. The network structure contains 5 convolutional layers with a kernel size of 3×3 and the number of channels of 32, 64, 128, 256, and 512, respectively. Each layer is followed by a ReLU activation function and a batch normalization layer. Three maximum pooling layers are located after the 1st, 2nd, and 4th convolutional layers, with a kernel size of 2×2 and a step size of 2. The fully connected layer flattens the feature map and passes it through 1024 neurons, finally outputting a 256-dimensional feature vector. K-means clustering is applied to the extracted features, with k=3, a maximum number of iterations of 300, and 10 random initializations to select the best result. The Euclidean distance from the cluster center to the origin is less than 0.5, which is defined as a slight color difference, 0.5-1.5 as a moderate color difference, and greater than 1.5 as a severe color difference. The area proportion and centroid coordinates of each type of color difference are calculated. For natural fiber materials, a decision tree based on the CART algorithm is used, with a tree depth limit of 10 and a minimum number of leaf node samples of 5. The input features include the spatial frequency principal components calculated by 2DFFT, the main direction angles obtained by Hough transform, and the local coefficient of variation within a 20×20 grid.The decision tree was trained on a data set containing 1,000 samples with known causes, with an accuracy of 85%. For synthetic fiber materials, a support vector machine with RBF kernel function was used, with penalty parameters C=10 and gamma=0.1. Input features include the local entropy mean and variance of a 9×9 window, gradient continuity calculated by the Sobel operator, and edge density obtained by Canny edge detection. The support vector machine was trained on a data set containing 1,500 samples with known causes, with an accuracy of 90%. Based on the judgment results, optimization suggestions are automatically generated, such as adjusting the selection criteria for natural fiber raw materials or modifying the dyeing temperature curve for synthetic fibers.

[0031] Step S105, for products that are judged to be unevenly dyed, obtain the dyeing process parameters of the batch of products. If the product to be tested is made of natural fiber material, use the fuzzy control algorithm to optimize and adjust the fiber pretreatment process; if the product to be tested is made of synthetic fiber material, use the genetic algorithm to optimize the dyeing process parameters, and obtain the optimal dye selection and ratio scheme through iterative calculation.

[0032] The batch information and dyeing process parameters of the unevenly dyed product are obtained, and the batch information and dyeing process parameters are extracted from a production management database; the fiber material of the unevenly dyed product is determined according to the batch information; if the fiber material is natural fiber, a fuzzy control algorithm is used to optimize the fiber pretreatment process, and the fuzzy control algorithm performs inference calculation on input variables through a fuzzy rule base constructed based on expert experience; if the fiber material is synthetic fiber, a genetic algorithm is used to optimize the dyeing process parameters, and the genetic algorithm encodes the dye type, dye ratio, dyeing temperature and time into chromosomes using real numbers; a process adjustment instruction is generated according to the optimization result, and the process adjustment instruction includes an adjustment range of pretreatment temperature, pH value, dyeing temperature and time; the optimized parameter value is written into a production control system, and the production control system is used to execute the process adjustment instruction; the optimization process and results are recorded in a quality management database, and the quality management database is used for subsequent process improvement and quality traceability.

[0033] Exemplarily, the batch information and dyeing process parameters of the unevenly dyed products are obtained, and the key parameters of the batch, such as fiber type, pretreatment conditions, dye type, dyeing temperature, time and pH value, are extracted from the production management database. If the product to be tested is made of natural fiber material, the fiber pretreatment process is optimized using a fuzzy control algorithm, and the fiber moisture content, alkali solution concentration, pretreatment time, temperature and pH value are set as input variables, and the fiber strength and dyeing uniformity are used as output variables. The input variables are inferred and calculated through a fuzzy rule base built based on expert experience, and the centroid method is used for defuzzification to obtain the optimized pretreatment process parameters. If the product to be tested is made of synthetic fiber, the genetic algorithm is used to optimize the dyeing process parameters. The dye type, dye ratio, dyeing temperature and time are encoded as chromosomes using real numbers. The population size is set to 100, and the roulette selection operation is used. The arithmetic crossover probability is 0.8 and the uniform mutation probability is 0.1. The multi-objective fitness function including dyeing uniformity and dyeing rate is designed, and the dye compatibility and cost performance are considered. The excellent individuals are selected by non-dominated sorting. After 500 generations of iterative calculation, the optimal dyeing process parameter combination and dye formula are obtained. According to the optimization results, the process adjustment instructions are generated. For natural fiber materials, the pretreatment temperature range is adjusted to 60-80℃, the pH range is 10-12, and the step length is 0.5; for synthetic fiber materials, the dyeing temperature range is adjusted to 100-130℃, the time range is 60-120 minutes, and the step length is 5 minutes. Specific dye selection and ratio schemes are given. The optimized parameter values ​​are written into the production control system, and the optimization process and results are recorded in the quality management database for subsequent process improvement and quality traceability. During the implementation process, the uneven dyeing product information of batch number B20240531 was first extracted from the production management database. This batch was made of cotton fiber, and the pretreatment conditions were alkali concentration of 5g / L, temperature of 70℃, time of 60 minutes, reactive dye CIReactiveBlue19, dyeing temperature of 90℃, time of 90 minutes, and pH value of 10.5. For natural fiber materials, the fuzzy control algorithm was used to optimize the pretreatment process, and the input variables were set as fiber moisture content of 50%-80%, alkali concentration of 3-7g / L, pretreatment time of 30-90 minutes, temperature of 60-80℃, and pH value of 9-12. Through 45 fuzzy rules constructed by expert experience, if the moisture content is high and the alkali concentration is low, the pretreatment time is extended. The Mamdani reasoning method and the centroid method were used to defuzzify, and the optimized parameters of alkali concentration of 6.2g / L, temperature of 73℃, time of 75 minutes, and pH value of 11.2 were obtained.For polyester fiber made of synthetic fiber, the dyeing process is optimized by real number coding genetic algorithm. The chromosome coding is [x1, x2, x3, x4, x5], which respectively represent the concentration of disperse dye CIDisperseBlue56 1-3%, the dosage of auxiliary agent 0.5-1.5g / L, the dyeing temperature 110-130℃, the time 60-120 minutes and the pH value 4-5. The fitness function f=0.6 uniformity+0.3 dye uptake+0.1*cost is designed. After 500 generations of iterative optimization, the optimal parameter combination is obtained: dye concentration 2.3%, auxiliary agent dosage 1.1g / L, dyeing temperature 125℃, time 95 minutes, pH value 4.5. According to the optimization results, the process adjustment instructions are generated to adjust the pretreatment temperature of natural fiber to 73℃ and pH value to 11.2; the dyeing temperature of synthetic fiber is adjusted to 125℃ and the time to 95 minutes. The optimization parameters are written into the PLC controller of the production control system, and the optimization process, including initial parameters, optimization algorithm configuration and final results, is recorded in the quality management database to provide data support for future process improvements.

[0034] Step S106, according to the optimal dyeing process parameters and pretreatment scheme, the dyeing process is simulated by numerical simulation, and the diffusion process of the dye in the fiber is calculated by finite element analysis to obtain the dyeing uniformity prediction result.

[0035] The three-dimensional geometric model of the fiber structure is constructed by using AutoCAD software, and the fiber cross-section and axial structure are generated according to the fiber diameter, length and porosity parameters; a mathematical model of dye diffusion is established, and the non-homogeneous diffusion process inside the fiber is described by the Maxwell-Stefan diffusion model, and the relationship between the diffusion coefficient and the temperature, pH value and auxiliary agent concentration is obtained by fitting the experimental data; the flow field and temperature field of the dyeing process are simulated by using FLUENT software, and the low Reynolds number k-ε model or SSTk-ω model is selected according to the dyeing process conditions, and the stirring process is simulated by using a dynamic grid; the dye diffusion process is numerically solved by using ANSYS software, and the time step and iteration accuracy are set to obtain the dye concentration distribution at different times; the concentration variation coefficient of the fiber cross-section and axial direction is calculated, and if the variation coefficient is less than a preset threshold, it is determined to be uniformly dyed.

[0036] For example, according to the optimal dyeing process parameters and pretreatment scheme, the three-dimensional geometric model of the fiber structure is constructed using AutoCAD software, the fiber cross section and axial structure are generated, the fiber diameter, length and porosity and other parameters are set, and the model is imported into ANSYS software for meshing. A mathematical model of dye diffusion is established, and the Maxwell-Stefan diffusion model is used to describe the heterogeneous diffusion process inside the fiber. The influence of temperature, pH value and auxiliary agent concentration on the diffusion coefficient is considered. The relationship between the diffusion coefficient and these factors is obtained by fitting the experimental data, the initial dye concentration distribution and boundary conditions are set, and the partial differential equation is discretized using the finite difference method. The flow field and temperature field of the dyeing process are simulated using FLUENT software, and an appropriate turbulence model such as a low Reynolds number k-ε model or SSTk-ω model is selected according to the dyeing process conditions. The geometric structure of the dyeing kettle is set, including the kettle size, the shape and layout of the agitator, and the position of the inlet and outlet. The dynamic grid technology is used to simulate the stirring process, and the temperature distribution is solved by the energy equation. The flow field velocity field and temperature field data are used as input conditions of the dye diffusion model, and the local diffusion coefficient of each grid node is updated. ANSYS software was used to numerically solve the dye diffusion process, setting the time step to 0.1 seconds and the iteration accuracy to 1e-6. OpenMP was used to implement multi-threaded parallel computing to accelerate the solution process, obtain the dye concentration distribution at different times, and calculate the concentration variation coefficient of the fiber cross section and axial direction. If the coefficient of variation is less than 5%, it is judged to be uniformly dyed. In the implementation process, AutoCAD2022 software was first used to construct a three-dimensional geometric model of the fiber structure, setting the cotton fiber diameter to 15μm, the length to 30mm, and the porosity to 0.35. The model was imported into ANSYS2021R1, and tetrahedral units were used for meshing. The unit size was set to 0.5μm, and the total number of grids was about 5 million. Subsequently, the Maxwell-Stefan diffusion model was established, considering the influence of temperature, pH value and additive concentration on the diffusion coefficient. The relationship between the diffusion coefficient and these factors was obtained by fitting the experimental data, D = D0 * exp (-Ea / RT) * (1 + k1pH + k2C_Auxiliary), where D0 = 1.5e-10m2 / s, Ea = 25kJ / mol, k1 = 0.05, k2 = 0.1. The geometric structure of the dyeing kettle was set in FLUENT2021R1, with a kettle diameter of 1m, a height of 1.5m, a four-blade propeller stirrer with a diameter of 0.3m and a speed of 60rpm. The SSTk-ω turbulence model was selected, the inlet velocity was set to 1m / s, and the outlet pressure was set to normal pressure. The dynamic grid technology was used to simulate the stirring process, with a time step of 0.01s and a total simulation time of 300s. The obtained flow field velocity and temperature field data were imported into ANSYS as input conditions for the dye diffusion model. The dye diffusion process was numerically solved in ANSYS, with the time step set to 0.1s and the iteration accuracy set to 1e-6.OpenMP was used to implement 8-thread parallel computing, and the entire simulation process was completed in about 4 hours on a 16-core CPU. The dye concentration distribution at different times was calculated, and the maximum concentration coefficient of variation in the cross section dropped from the initial 45% to 3.8%, and the axial coefficient of variation dropped from 32% to 4.2%, both of which were less than the 5% threshold, and were judged to be uniform dyeing.

[0037] Step S107, compare the prediction result with the color fastness test standard to determine whether the dyed fabric can meet the product quality requirements. If it meets the requirements, the batch of products will be put into storage; if it does not meet the requirements, the prediction result will be fed back to the dyeing process control system for the next round of optimization iteration until the product quality requirements are met.

[0038] The prediction results of dyeing uniformity and the color fastness test standards are obtained, and the prediction results and test standards are processed by a data processing library to obtain the compliance rate of each indicator and the overall compliance rate. According to the compliance rate, the quality of the dyed fabric is evaluated by using a hierarchical analysis method combined with a neural network model, and the neural network model is trained by a weight matrix to obtain a comprehensive score of the dyed fabric. The quality requirement threshold is obtained from the product database, and the evaluation standard is set according to the quality requirement threshold. If the comprehensive score reaches the preset threshold, the warehousing process of the warehouse management system is triggered, the warehousing document is generated, the warehouse location is allocated, and the inventory information is updated. If the comprehensive score does not reach the preset threshold, the evaluation result is fed back to the dyeing process control system, and the process parameters are optimized by a reinforcement learning algorithm, wherein the state space is set as the current process parameter combination, the action space is the parameter adjustment plan, and the reward function is based on the quality evaluation score; the optimal strategy is learned by interacting with the environment to obtain a new parameter combination; the state, action, and reward data in the optimization process are recorded in the process optimization database.

[0039] Exemplarily, the prediction results of dyeing uniformity and the color fastness test standards are obtained, and the Python pandas library is used for data processing and comparison to calculate the compliance rate of each indicator and the overall compliance rate. According to the comparison results, the quality of dyed fabrics is evaluated by using the hierarchical analysis method combined with the neural network, and the weight matrix of indicators such as color fastness, color difference, and dyeing uniformity is determined by expert scoring, and the neural network model is trained to predict the comprehensive score. The quality requirement thresholds of different product types are obtained from the product database, and the evaluation standards are set dynamically. If the comprehensive score reaches the preset threshold, the warehousing process of the warehouse management system is triggered, the warehousing document is generated, the warehouse location is allocated, the inventory information is updated, and the quality evaluation results and process parameters of the batch of products are recorded. If the comprehensive score does not reach the preset threshold, the evaluation result is fed back to the dyeing process control system, and the Q-learning algorithm is used to optimize the process parameters. The state space is set as the current process parameter combination, the action space is set as the parameter adjustment plan, and the reward function is based on the quality evaluation score. The optimal strategy is learned through interaction with the environment to generate a new parameter combination. The state, action, reward and other data in the optimization process are recorded in the process optimization database for long-term process improvement analysis. The dyeing simulation and quality assessment process are re-executed until the product quality requirements are met or the maximum number of iterations is reached. In the implementation process, Python3.8 and pandas1.2.4 libraries are first used to process the dyeing uniformity prediction results and color fastness test standards, and the color fastness to washing compliance rate is 98%, the color fastness to rubbing compliance rate is 95%, and the overall compliance rate is 96.5%. Subsequently, a three-layer neural network based on TensorFlow2.4.1 is used, with 10 nodes in the input layer, 20 nodes in the hidden layer, and 1 node in the output layer. The quality assessment is combined with the hierarchical analysis method. The weight matrix is ​​obtained through the scoring of 5 experts, in which the color fastness weight is 0.4, the color difference weight is 0.3, and the dyeing uniformity weight is 0.3. The training data contains 1000 sets of historical records. After 5000 iterations, the model prediction accuracy reaches 94%. The quality requirement threshold of the current cotton fabric product obtained from the product database is 0.85. The predicted comprehensive score is 0.88, which exceeds the threshold, triggering the warehousing process of the warehouse management system, generating the warehousing order number IN2023051001, allocating the storage location A3-B2-C5, and updating the inventory information to show that the number of products in this batch is 1000. For the case of failure to meet the standard, the Q-learning algorithm based on PyTorch1.8.1 is used to optimize the process parameters, setting the state space to 5 process parameters such as temperature, time, pH value, additive concentration and dye concentration, each parameter is discretized into 10 levels, the action space is the increase or decrease of each parameter, and the reward function is set to R = score - 0.85. If the score is ≥ 0.85, an additional reward of +1 is given.After 10,000 rounds of iterative learning, the optimal strategy was obtained: the temperature was increased by 2°C, the time was extended by 5 minutes, the pH value was reduced by 0.2, the additive concentration was increased by 0.1 g / L, and the dye concentration remained unchanged. The optimization data was recorded in the MongoDB database for subsequent analysis.

[0040] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for detecting color consistency of textiles, characterized in that: The method comprises: According to the differences in the hygroscopicity, air permeability and dyeing properties of natural and synthetic fibers, the color fastness test standards and color tolerance setting ranges of different fiber materials are established. The historical data are analyzed using a machine learning algorithm, and the dyeing characteristics of different fiber materials are classified through a clustering method to obtain a typical color difference distribution model for various fiber materials. According to the obtained color difference distribution model, the support vector machine algorithm is used to establish a color difference judgment model. If the product to be tested is made of natural fiber material, the color difference judgment standard of the product is dynamically relaxed according to the characteristics that its dyeing uniformity is easily affected by the fiber composition and structure. If the product to be tested is made of synthetic fiber material, the color difference judgment standard of the product is dynamically tightened according to the characteristics that its dyeing uniformity is relatively easy to control. Obtain the spectral data of the product to be tested, conduct a comprehensive scan of the product surface through multi-spectral imaging, and use image processing algorithms to reduce noise and enhance the scanned data to obtain high-precision color difference detection data; Based on high-precision color difference detection data, a deep learning algorithm is used to analyze the color difference distribution of the product surface, and the color difference features are extracted through a convolutional neural network. If the product to be tested is made of natural fiber material, it is determined whether the color difference is caused by uneven fiber composition and structure. If the product to be tested is made of synthetic fiber material, it is determined whether the color difference is caused by improper dyeing process control. For products with uneven dyeing, the dyeing process parameters of the batch of products are obtained. If the product to be tested is made of natural fiber, the fiber pretreatment process is optimized and adjusted using a fuzzy control algorithm. If the product to be tested is made of synthetic fiber, the dyeing process parameters are optimized using a genetic algorithm. The optimal dye selection and ratio scheme is obtained through iterative calculation. According to the optimal dyeing process parameters and pretreatment scheme, the dyeing process is simulated by numerical simulation, and the diffusion process of the dye in the fiber is calculated by finite element analysis to obtain the prediction result of dyeing uniformity. The prediction results are compared with the color fastness test standards to determine whether the dyed fabric can meet the product quality requirements. If so, the batch of products will be put into storage. If not, the prediction results will be fed back to the dyeing process control system for the next round of optimization iterations until the product quality requirements are met.

2. The method according to claim 1, wherein: According to the differences in the detailed properties of hygroscopicity, air permeability and dyeing performance of natural fibers and synthetic fibers, the color fastness test standards and color tolerance setting ranges of different fiber materials are established, the historical data are analyzed by machine learning algorithms, and the dyeing characteristics of different fiber materials are classified by clustering methods to obtain typical color difference distribution models of various fiber materials, including: Obtain fiber attribute data and corresponding color fastness and color tolerance data, and use multiple regression analysis to establish a relationship model between fiber attributes and color fastness and color tolerance; Calculating the regression coefficient and significance test result of the relationship model to determine the influence of various indicators on the color fastness and the color tolerance; According to the influence degree, the color fastness test standard and color tolerance setting range of different fiber materials are obtained; A hierarchical clustering algorithm was used to analyze historical dyeing data, and the optimal number of clusters was determined by calculating the silhouette coefficient to classify the dyeing characteristics of different fiber materials. For each type of fiber material, the color difference data are fitted using normal distribution, lognormal distribution and beta distribution respectively; The maximum likelihood estimation and information criterion were calculated, and the distribution type with the best fitting effect was selected to obtain the typical color difference distribution model of various fiber materials.

3. The method according to claim 1, wherein: The color difference judgment model is established by using a support vector machine algorithm for the obtained color difference distribution model. If the product to be tested is made of natural fiber material, the color difference judgment standard of the product is dynamically relaxed according to the characteristic that its dyeing uniformity is easily affected by the fiber composition and structure; if the product to be tested is made of synthetic fiber material, the color difference judgment standard of the product is dynamically tightened according to the characteristic that its dyeing uniformity is relatively easy to control, including: Acquire the color difference distribution model, and construct a color difference determination model using a support vector machine algorithm; According to the color difference determination model, the optimal parameters are determined by a grid search method, and a radial basis function is selected as a kernel function; After the training data is standardized, the data is input into the color difference judgment model for training to obtain a decision boundary for color difference judgment; For natural fiber materials, fiber length, fineness and composition information are extracted from a pre-established fiber property database; Calculating the dyeing uniformity influencing factor using a weighted average method based on the fiber length, fineness and composition information; If the dyeing uniformity influencing factor is greater than a preset threshold, the tolerance range of the color difference judgment standard is expanded by a preset proportion; For synthetic fiber materials, fiber length, fineness and composition information are extracted from a pre-established fiber property database; Calculating the dyeing uniformity influencing factor using a weighted average method based on the fiber length, fineness and composition information; If the dyeing uniformity influencing factor is less than a preset threshold, the tolerance range of the color difference judgment standard is reduced by a preset ratio; Traversing different color tolerance setting ranges, and calculating the determination accuracy and the misdetermination rate for each color tolerance setting range; According to the determination accuracy and misjudgment rate, the F1 score is used as an evaluation indicator; Select the color tolerance setting range with the highest F1 score as the optimal result and determine the final color difference judgment standard of the product; Obtain information on production batches and seasonal variation factors; Regularly updating the color difference determination model and parameters according to the production batch and seasonal variation factor information; The color difference judgment standard is dynamically adjusted by updating the color difference judgment model and parameters.

4. The method according to claim 1, wherein: The method of obtaining spectral data of the product to be tested, performing a comprehensive scan of the product surface through multi-spectral imaging, and performing noise reduction and enhancement processing on the scanned data using an image processing algorithm to obtain high-precision color difference detection data includes: Acquire original multispectral image data of the surface of the product to be tested, wherein the original multispectral image data is obtained by performing a comprehensive scan of the surface of the product to be tested by a multispectral imaging device; Preprocessing the original multispectral image data using an adaptive median filtering algorithm to obtain a preprocessed multispectral image, wherein the adaptive median filtering algorithm dynamically adjusts the filter window size according to the local noise density; Performing image enhancement using a band adaptive enhancement algorithm on the preprocessed multispectral image to obtain enhanced multispectral image data, wherein the band adaptive enhancement algorithm uses histogram equalization for the visible light band and contrast-limited adaptive histogram equalization for the near infrared band; According to the enhanced multispectral image data, a principal component analysis algorithm is used to perform dimensionality reduction processing to obtain a feature vector; By calculating the Euclidean distance of each pixel in the feature space and converting the result into the CIELab color space, the color difference detection data is obtained; If the average color difference error between the color difference detection data and the standard color card is less than a preset threshold, it is determined that the color difference detection data meets the high-precision requirement.

5. The method according to claim 1, wherein: According to the high-precision color difference detection data, a deep learning algorithm is used to analyze the color difference distribution on the product surface, and the color difference features are extracted through a convolutional neural network. If the product to be tested is made of natural fiber material, it is determined whether the color difference is caused by uneven fiber composition and structure; If the product to be tested is made of synthetic fiber, determine whether the color difference is caused by improper dyeing process control, including: Acquire product surface color difference distribution data, and use a convolutional neural network to perform feature extraction on the color difference distribution data to obtain a color difference feature vector; According to the color difference feature vector, the color difference distribution is classified by using a K-means clustering algorithm to determine the degree of color difference and spatial distribution characteristics; If the product is made of natural fiber, a pre-trained decision tree algorithm is used to determine whether the color difference is caused by uneven fiber composition and structure; If the product is made of synthetic fiber, a pre-trained support vector machine algorithm is used to determine whether the color difference is caused by improper control of the dyeing process; For natural fiber materials, if it is determined that the color difference is caused by fiber unevenness, the raw material selection and pretreatment process optimization plan are determined; For synthetic fiber materials, if it is determined that the color difference is caused by the dyeing process, the dye ratio and temperature control parameter adjustment plan are determined.

6. The method according to claim 1, wherein: For the product with uneven dyeing, the dyeing process parameters of the batch of products are obtained. If the product to be tested is made of natural fiber material, the fiber pretreatment process is optimized and adjusted using a fuzzy control algorithm. If the product to be tested is made of synthetic fiber, the genetic algorithm is used to optimize the dyeing process parameters, and the optimal dye selection and ratio scheme is obtained through iterative calculation, including: Acquiring batch information and dyeing process parameters of the unevenly dyed product, wherein the batch information and dyeing process parameters are extracted from a production management database; Determining the fiber material of the unevenly dyed product according to the batch information; If the fiber material is natural fiber, a fuzzy control algorithm is used to optimize the fiber pretreatment process, and the fuzzy control algorithm performs reasoning calculations on input variables through a fuzzy rule base constructed based on expert experience; If the fiber material is synthetic fiber, a genetic algorithm is used to optimize the dyeing process parameters, and the genetic algorithm encodes the dye type, dye ratio, dyeing temperature and time into chromosomes using real numbers; Generating a process adjustment instruction according to the optimization result, wherein the process adjustment instruction includes an adjustment range of pretreatment temperature, pH value, dyeing temperature and time; Writing the optimized parameter value into a production control system, the production control system is used to execute the process adjustment instruction; The optimization process and results are recorded in the quality management database, which is used for subsequent process improvement and quality traceability.

7. The method according to claim 1, wherein: The dyeing process is simulated by numerical simulation according to the optimal dyeing process parameters and pretreatment scheme, and the diffusion process of the dye in the fiber is calculated by finite element analysis to obtain the dyeing uniformity prediction result, including: The three-dimensional geometric model of the fiber structure is constructed using AutoCAD software, and the fiber cross-section and axial structure are generated according to the fiber diameter, length and porosity parameters; A mathematical model of dye diffusion is established, and the Maxwell-Stefan diffusion model is used to describe the heterogeneous diffusion process inside the fiber. The relationship between the diffusion coefficient and the temperature, pH value and additive concentration is obtained by fitting the experimental data. The flow field and temperature field of the dyeing process are simulated by using FLUENT software, a low Reynolds number k-ε model or a SSTk-ω model is selected according to the dyeing process conditions, and a dynamic grid is used to simulate the stirring process; The dye diffusion process is numerically solved using ANSYS software, the time step and iteration accuracy are set, and the dye concentration distribution at different times is obtained; The concentration variation coefficient of the fiber cross section and the axial direction is calculated. If the variation coefficient is less than a preset threshold, it is determined to be uniformly dyed.

8. The method according to claim 1, wherein: The prediction results are compared with the color fastness test standards to determine whether the dyed fabrics meet the product quality requirements. If they meet the requirements, the batch of products will be put into storage; If the requirements are not met, the prediction results will be fed back to the dyeing process control system for the next round of optimization iterations until the product quality requirements are met, including: Obtain dyeing uniformity prediction results and color fastness test standards, use a data processing library to process the prediction results and test standards, and obtain the compliance rate of each indicator and the overall compliance rate; According to the compliance rate, the quality of the dyed fabric is evaluated by using a hierarchical analysis method combined with a neural network model, and the neural network model is trained by a weight matrix to obtain a comprehensive score of the dyed fabric; Acquire a quality requirement threshold from a product database, and set an evaluation standard according to the quality requirement threshold; If the comprehensive score reaches the preset threshold, the warehouse management system's warehousing process is triggered, a warehousing document is generated, a warehouse location is allocated, and inventory information is updated; If the comprehensive score does not reach the preset threshold, the evaluation result is fed back to the dyeing process control system, and the process parameters are optimized using a reinforcement learning algorithm, wherein the state space is set to the current process parameter combination, the action space is set to the parameter adjustment scheme, and the reward function is based on the quality evaluation score; By learning the optimal strategy through interaction with the environment, a new parameter combination is obtained; The status, actions and reward data during the optimization process are recorded in a process optimization database.

Images