A dye formula optimization method and system for disperse dark one-bath dyeing process

Abstract

The present invention relates to the technical field of dyeing formula data processing, specifically a dye formula optimization method and system for a dispersed dark same-bath dyeing process, comprising: constructing a process formula data set, calculating dye mixture spectral data using the Beer-Lambert law, extracting and fusing features through a CNN and Transformer prediction method, and then calculating color difference. If the color difference exceeds a threshold, a genetic algorithm is used to optimize the dye ratio concentration. During the dyeing process, the dye bath fuzzy set is divided and fuzzy rules are set based on a fuzzy control model, and the real-time process parameter adjustment amount is calculated and optimized in combination with the membership degree. After dyeing is completed, process image data is collected, and the process color difference, maximum local color difference and color fastness are evaluated with the help of a computer vision model. If the evaluation result does not meet the standard, the dye ratio concentration and the real-time process parameter adjustment amount are adjusted to reduce the color difference in the dark sportswear dyeing process and improve the accuracy of the dye formula.

Description

Technical Field

[0001] The present invention relates to the technical field of dyeing formula data processing, and in particular to a dye formula optimization method and system for a disperse dark color one-bath dyeing process. Background Art

[0002] In the textile manufacturing industry, dyeing formula management and process optimization for sportswear are crucial for product quality, production efficiency, and cost control. Sportswear is primarily composed of polyester fibers, and the dyeing process requires disperse dyes at high temperatures. However, achieving the desired color for dark-colored sportswear typically requires higher dye concentrations, but excessive dye use can lead to floating color, color shift, and dyeing uniformity issues. Furthermore, sportswear fabrics often contain elastic fibers, and during the dyeing process, different fibers have varying dye absorption capacities, which can result in uneven dye depth and impact the consistency of batch production.

[0003] Currently, the dyeing process for dark sportswear mainly relies on empirical methods, that is, by repeatedly adjusting the disperse dye ratio, dyeing temperature and auxiliary agent dosage, and then conducting experiments to determine the appropriate dyeing parameters. Although this method is intuitive, the experimental cost is high and there is a lack of standardized management. Clothing produced in different batches may have slight color differences, affecting product consistency. In order to optimize the dye formula, some technicians have tried to use mathematical models such as Lambert-Beer's law and Nernst distribution law to calculate the optimal dye ratio. In theory, this method can reduce the number of experiments and improve prediction accuracy. However, the dyeing process involves multiple complex variables, and traditional mathematical models cannot fully consider the dynamic dye bath environment, resulting in a lot of trial and error in practical applications.

[0004] In order to reduce the color difference in the dyeing process of dark sportswear and improve the accuracy of dye formula in the dyeing process, a dye formula optimization method and system for disperse dark one-bath dyeing process are proposed. Summary of the Invention

[0005] The present invention aims to provide a dye formula optimization method and system for a dispersed dark same-bath dyeing process, thereby reducing color difference during the dyeing process of dark sportswear and improving the accuracy of the dye formula during the dyeing process. A process formula dataset is constructed, the Beer-Lambert law is used to calculate the dye mixture spectrum data, and features are extracted and integrated using a CNN and Transformer prediction method to calculate the color difference. If the color difference exceeds a threshold, a genetic algorithm is used to optimize the dye ratio concentration. During the dyeing process, the dye bath fuzzy set is divided based on a fuzzy control model, fuzzy rules are set, and the membership is combined to calculate and optimize the real-time process parameter adjustment amount. After dyeing is completed, process image data is collected and the process color difference, maximum local color difference, and color fastness are evaluated using a computer vision model. If the evaluation results do not meet the standards, the dye ratio concentration and real-time process parameter adjustment amount are adjusted, thereby reducing color difference during the dyeing process of dark sportswear and improving the accuracy of the dye formula.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for optimizing a dye formula for a disperse dark one-bath dyeing process, comprising:

[0008] Collect dye absorption spectrum data and historical dyeing data to build a process recipe data set;

[0009] Based on the process recipe data set, dye mixture spectral data is calculated using the Beer-Lambert law, and the spectral feature data is input into a convolutional neural network model to extract spectral feature data. The dye ratio concentration is input into a self-attention mechanism model to obtain concentration feature data; the spectral feature data and the concentration feature data are feature fused to calculate the mixed color feature and recipe error value; if the recipe error value exceeds a target threshold, the dye ratio concentration is iteratively calculated using a genetic algorithm;

[0010] detecting real-time process parameter data under the dye ratio concentration, dividing the dye bath fuzzy set and setting the dye bath fuzzy rules, and outputting the real-time process parameter adjustment amount in combination with the membership degree; optimizing the weight of the dye bath fuzzy rules using nonlinear regression analysis, and updating the real-time process parameter adjustment amount;

[0011] The process image data after dyeing is collected, and the analysis results of the process image data are evaluated using a computer vision model; and the dye ratio concentration and the real-time process parameter adjustment amount are adjusted using an adaptive optimization strategy.

[0012] Furthermore, the implementation steps of the genetic algorithm include:

[0013] Step S10: randomly generating N groups of different dye ratio concentrations according to the process recipe data set;

[0014] Step S20: For the dye ratio concentration, calculate the fitness value according to the formula error value, which is expressed as:

[0015] ;

[0016] in, is the fitness value, is the formula error value;

[0017] Step S30: selecting the dye ratio concentration with the fitness value greater than the fitness threshold as a high-quality formula and placing it in a mating pool, randomly selecting two high-quality formulas from the mating pool for arithmetic crossover to obtain a new high-quality formula;

[0018] Step S40: performing Gaussian variation on the new high-quality formula to obtain a new dye ratio concentration;

[0019] Step S50: Repeat steps S10 to S40 until the recipe error value is no greater than the target threshold.

[0020] Furthermore, the setting process of the dye bath fuzzy rule includes:

[0021] The real-time process parameter data includes dye bath temperature, pH value and dye bath time;

[0022] Dividing the dye bath temperature, the pH value and the dye bath time into dye bath fuzzy sets respectively;

[0023] Establishing a fuzzy rule base based on expert experience to describe the relationship between the real-time process parameter data and the real-time process parameter adjustment amount to obtain a dye bath fuzzy output;

[0024] The membership of the real-time process parameter data is calculated using a triangular membership function, and combined with the dye bath fuzzy output, a weighted summation method is used to obtain the real-time process parameter adjustment amount.

[0025] Furthermore, the evaluation process of the computer vision model includes:

[0026] The analysis results include process color difference, maximum local color difference, rubbing color fastness and washing color fastness;

[0027] The process image data is converted into Lab color space, and the process color difference is calculated using the CIELab color difference formula;

[0028] Using a sliding window to divide the process image data into fabric areas, calculating the Lab color value of each fabric area, and calculating the local color difference of adjacent fabric areas for the Lab color values ​​to obtain the maximum local color difference;

[0029] Calculating the rubbing color fastness based on the process image data before and after rubbing;

[0030] The washing color fastness is calculated based on the process image data before and after washing.

[0031] Furthermore, the dye ratio concentration adjustment process includes: if the process color difference does not meet the preset standard, the dye ratio concentration is linearly adjusted using the adaptive optimization strategy to obtain the optimized dye ratio concentration, and the dye bath is re-dyeing according to the optimized dye ratio concentration; otherwise, the optimized dye ratio concentration is saved in the process formula database.

[0032] Furthermore, the adjustment process of the real-time process parameter adjustment amount includes: if the maximum local color difference does not meet the preset standard, the real-time process parameter adjustment amount is linearly adjusted using the adaptive optimization strategy to obtain the optimized real-time process parameter adjustment amount, and the dyeing bath is re-performed according to the optimized real-time process parameter adjustment amount; otherwise, the optimized real-time process parameter adjustment amount is saved to the process formula database.

[0033] A dye formula optimization system for a disperse dark one-bath dyeing process, comprising:

[0034] Recipe data acquisition module, used to build process recipe data set;

[0035] A mixture concentration input module is configured to calculate dye mixture spectral data using the Beer-Lambert law, input the data into a convolutional neural network model to extract spectral feature data, and input the dye mixture concentration into a self-attention mechanism model to obtain concentration feature data; perform feature fusion on the spectral feature data and the concentration feature data to calculate the mixed color feature and the mixture error value; and if the mixture error value exceeds a target threshold, iteratively calculate the dye mixture concentration using a genetic algorithm;

[0036] a process parameter adjustment module for detecting real-time process parameter data under the dye ratio concentration, dividing the dye bath fuzzy set and setting the dye bath fuzzy rules, and outputting real-time process parameter adjustment amounts in combination with the membership degree; optimizing the weights of the dye bath fuzzy rules using nonlinear regression analysis, and updating the real-time process parameter adjustment amounts;

[0037] The finished product color difference evaluation module is used to collect process image data after dyeing, use a computer vision model to evaluate the analysis results of the process image data, and use an adaptive optimization strategy to adjust the dye ratio concentration and the real-time process parameter adjustment amount.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. By constructing a dye formula dataset and combining it with the Beer-Lambert law to calculate dye mixing spectral data, this method can accurately simulate dye mixing effects and provide basic data support for subsequent optimization. Using a CNN and self-attention mechanism model to extract spectral and concentration features, respectively, and then perform feature fusion to calculate mixed color features and formula error values, this method effectively captures the complex relationship between dye ratios and color changes, thereby improving prediction accuracy. If the error value exceeds a preset threshold, a genetic algorithm is used to dynamically optimize the dye ratio concentration, thereby improving the accuracy of the dyeing formula.

[0040] 2. By detecting real-time process parameter data at dye ratio concentrations, dividing the dye bath into fuzzy sets, and setting fuzzy rules, the present invention accurately describes the complex characteristics of real-time process parameter data, providing a scientific basis for subsequent adjustments. The real-time process parameter adjustment value is output based on the membership degree, ensuring the accuracy of the adjustment value. Using nonlinear regression analysis to optimize the weights of fuzzy rules and membership degrees, the control strategy for real-time process parameter data can be continuously optimized, improving the adaptability of the adjustment value. Real-time updating of the real-time process parameter adjustment value dynamically responds to changes in the dye bath environment, effectively enhancing the stability and consistency of the dyeing process.

[0041] 3. By collecting post-dyeing process image data and using computer vision models to evaluate and analyze the results, the present invention can rapidly identify process color difference, maximum local color difference, rubbing color fastness, and washing color fastness, providing data support for subsequent adjustments. If the test results do not meet preset conditions, the system automatically adjusts the dye ratio concentration and process parameter adjustments in real time, achieving closed-loop control of dyeing quality and ensuring the stability of the dyeing effect. This present invention enables intelligent evaluation and dynamic feedback of dyeing quality, thereby improving the reliability and accuracy of dyeing quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic flow chart of a method for optimizing a dye formula for a disperse dark one-bath dyeing process provided by the present invention;

[0043] Figure 2 A schematic diagram of the genetic algorithm flow provided by the present invention;

[0044] Figure 3 A schematic diagram of the structure of a dye formula optimization system for a disperse dark color one-bath dyeing process provided by the present invention;

[0045] Figure 4 Schematic diagram of the feature extraction process of the CNN and Transformer models provided by the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] See also Figures 1 to 4 The present invention provides a method and system for optimizing the dye formula of a dispersed dark one-bath dyeing process, and the technical solution is as follows:

[0048] Example 1:

[0049] Sportswear is typically composed of polyester and elastane fibers. Due to the hydrophobic nature of polyester, disperse dyes are typically used for high-temperature dyeing. However, for dark-colored sportswear (such as black, dark blue, and navy), the dyeing process often encounters significant color variations, severe color floating, poor uniformity, and insufficient color fastness. Currently, dyeing dark sportswear relies primarily on theoretical calculations and process parameter optimization methods, but these methods often overlook environmental factors, resulting in limited color stability.

[0050] To solve these problems, Figure 1 A schematic flow chart of a method for optimizing a dye formula for a disperse dark one-bath dyeing process provided by the present invention is shown in FIG. Figure 1 As shown in the figure, a dye formula optimization method for disperse dark one-bath dyeing process is proposed, including:

[0051] Collect dye absorption spectrum data and historical dyeing data to build a process recipe data set;

[0052] Based on the process recipe data set, dye mixture spectral data is calculated using the Beer-Lambert law, and the spectral feature data is input into a convolutional neural network model to extract spectral feature data. The dye ratio concentration is input into a self-attention mechanism model to obtain concentration feature data; the spectral feature data and the concentration feature data are feature fused to calculate the mixed color feature and recipe error value; if the recipe error value exceeds a target threshold, the dye ratio concentration is iteratively calculated using a genetic algorithm;

[0053] The mixed color feature refers to the color value corresponding to the predicted dye ratio concentration. Both the convolutional neural network model and the self-attention mechanism model are trained using historical dyeing data.

[0054] detecting real-time process parameter data at the dye ratio concentration, dividing the dye bath fuzzy set and setting the dye bath fuzzy rules for outputting real-time process parameter adjustment amounts in combination with the membership; optimizing the weights of the dye bath fuzzy rules and the membership degrees using nonlinear regression analysis, and updating the real-time process parameter adjustment amounts;

[0055] Collect process image data after dyeing, and use a computer vision model to evaluate the analysis results of the process image data; if the analysis results do not meet the preset standards, adjust the dye ratio concentration and the real-time process parameter adjustment amount.

[0056] Before dyeing, a process recipe dataset is constructed and dye mixture spectral data is calculated using the Beer-Lambert law. This allows for more scientific selection of dye ratio concentrations, mitigating dyeing deviations at the source. Spectral features are extracted using a CNN and concentration features are extracted using a Transformer. Feature fusion is then performed to accurately predict mixed color characteristics and calculate recipe error. A genetic algorithm is then used to optimize dye ratio concentrations, enabling even more accurate prediction of mixed color characteristics and ensuring a close match between the target and actual colors.

[0057] During dyeing, the system monitors process parameter data in real time, divides the dye bath into fuzzy sets, and sets fuzzy rules. Membership calculation and fuzzy control are then used to adjust process parameter data in real time, making control more flexible and adaptable to varying dyeing environments. Fuzzy rule weights are then optimized through nonlinear regression, improving the accuracy of parameters for dyeing dark sportswear and ensuring a stable dyeing process.

[0058] After dyeing, computer vision models are used to collect process image data, assess process color difference, local color difference, and color fastness, enabling automated quality inspection. The system also automatically optimizes dye ratio concentration and real-time process parameters, reducing dye waste and adjustment time. By dynamically updating the process recipe dataset and continuously optimizing the dyeing model to ensure the dyeing process meets preset standards, the system reduces color difference in the dyeing process of dark sportswear and improves the accuracy of the dye recipe during the dyeing process.

[0059] Furthermore, the implementation steps of the genetic algorithm include:

[0060] Step S10: randomly generating N groups of different dye ratio concentrations according to the process recipe data set;

[0061] Step S20: For the dye ratio concentration, calculate the fitness value according to the formula error value, which is expressed as:

[0062] ;

[0063] in, is the fitness value, is the formula error value, obtained by combining the Beer-Lambert law, CNN and Transformer models.

[0064] Step S30: sorting by the fitness value, selecting the dye ratio concentration with the fitness value greater than the fitness threshold as a high-quality formula, and placing it into a mating pool, randomly selecting two high-quality formulas from the mating pool for arithmetic crossover to obtain a new high-quality formula;

[0065] Step S40: performing Gaussian variation on the new high-quality formula to obtain a new dye ratio concentration;

[0066] Step S50: Repeat steps S10 to S40 until the recipe error value is no greater than the target threshold.

[0067] Figure 2 This is a schematic diagram of the genetic algorithm process provided by the present invention. By using the genetic algorithm to optimize the dye ratio concentration, efficient and accurate dyeing formula optimization is achieved. Figure 2 As shown, by first randomly generating multiple initial dye recipes and calculating fitness values ​​based on recipe errors, high-quality recipes that meet the target color difference requirements can be quickly screened. These high-quality recipes continuously optimize dye concentrations during the evolutionary process through crossover and mutation operations, avoiding falling into local optimal solutions and thus improving the convergence speed of the genetic algorithm. At the same time, the Gaussian mutation operation maintains the diversity of dye ratios, enhancing the adaptability and stability of the optimization process. Ultimately, when the color difference meets a preset threshold, the algorithm stops iterating, ensuring the stability and consistency of the dyeing effect. Compared with fixed parameter optimization methods, the ability to obtain recipe error values ​​through a predictive model reduces the number of trials, improves color matching accuracy and process standardization, and makes the dyeing process of dark sportswear more efficient.

[0068] Furthermore, the setting process of the dye bath fuzzy rule includes:

[0069] The real-time process parameter data includes dye bath temperature, pH value and dye bath time;

[0070] Dividing the dye bath temperature, the pH value and the dye bath time into dye bath fuzzy sets respectively;

[0071] Establishing a fuzzy rule base based on expert experience to describe the relationship between the real-time process parameter data and the real-time process parameter adjustment amount to obtain a dye bath fuzzy output;

[0072] The membership of the real-time process parameter data is calculated using a triangular membership function, and combined with the dye bath fuzzy output, a weighted summation method is used to obtain the real-time process parameter adjustment amount.

[0073] By setting a multi-level fuzzy set of temperature, pH value, and dye bath time, it is possible to flexibly adapt to different dyeing conditions and ensure the stability of the dye bath state. The triangular membership function is then used to calculate the fuzziness of the parameters, and combined with fuzzy rules for output, the impact of each factor on dye bath adjustment can be effectively quantified. Furthermore, the real-time process parameter adjustment amount is calculated using a weighted summation method, allowing the dye bath state to more accurately approach the target range. Compared with the traditional fixed parameter adjustment method, it can dynamically adapt to nonlinear changes in the dyeing process, improve the accuracy and stability of process control, thereby improving dyeing uniformity, reducing color difference in the dyeing process of dark sportswear, and improving the accuracy of dye formulas during the dyeing process.

[0074] Furthermore, the evaluation process of the computer vision model includes:

[0075] The analysis results include process color difference, maximum local color difference, rubbing color fastness and washing color fastness;

[0076] The process image data is converted into Lab color space, and the process color difference is calculated using the CIELab color difference formula;

[0077] Using a sliding window to divide the process image data into fabric areas, calculating the Lab color value of each fabric area, and calculating the local color difference of adjacent fabric areas for the Lab color values ​​to obtain the maximum local color difference;

[0078] Calculating the rubbing color fastness based on the process image data before and after rubbing;

[0079] The washing color fastness is calculated based on the process image data before and after washing.

[0080] By converting the Lab color space and combining it with CIELab color difference calculation, the overall color difference can be accurately measured. By segmenting the fabric area through a sliding window and calculating the local color difference, the uneven dyeing area can be accurately identified to ensure that the maximum local color difference meets the preset standard. Furthermore, this method can automatically calculate the rubbing color fastness and washing color fastness by comparing the process images before and after friction and before and after washing, and provide color fastness evaluation. The dyeing process results are comprehensively evaluated through a computer vision model. Through multiple key indicators such as process color difference, maximum local color difference, rubbing color fastness and washing color fastness, the stability and consistency of the dyeing quality are ensured. Compared with the single-point measurement method, the accuracy of color difference and color fastness evaluation can be improved, while reducing the color difference fluctuation between batches, making the dyeing process of dark sportswear more standardized.

[0081] Furthermore, the dye ratio concentration adjustment process includes: if the process color difference does not meet the preset standard, the dye ratio concentration is linearly adjusted using the adaptive optimization strategy to obtain the optimized dye ratio concentration, and the dye bath is re-dyeing according to the optimized dye ratio concentration; otherwise, the optimized dye ratio concentration is saved in the process formula database.

[0082] By linearly adjusting the dye ratio concentration, the dyeing formula is accurately optimized, ensuring that the final dyeing effect meets the preset standards. Specifically, if the process color difference exceeds the threshold, the optimized dye ratio concentration will be automatically calculated and applied in the next dye bath, thereby reducing the repeated trials caused by color difference problems. When the dyeing results meet the standards, the optimized formula will be stored in the process formula data set for use in subsequent dyeing batches, which helps to improve the stability and standardization of production. Compared with the method that relies on experience adjustment, it not only reduces the color difference in the dyeing process of dark sportswear, but also improves the accuracy of the dye formula in the dyeing process.

[0083] Furthermore, the adjustment process of the real-time process parameter adjustment amount includes: if the maximum local color difference does not meet the preset standard, the real-time process parameter adjustment amount is linearly adjusted using the adaptive optimization strategy to obtain the optimized real-time process parameter adjustment amount, and the dyeing bath is re-performed according to the optimized real-time process parameter adjustment amount; otherwise, the optimized real-time process parameter adjustment amount is saved to the process formula database.

[0084] Similar to the dye ratio concentration adjustment process, precise control of the dyeing process is achieved by linearly adjusting real-time process parameters (including temperature, pH value, and dye bath time) to ensure dyeing uniformity. If the maximum local color difference exceeds the preset standard, the system will automatically calculate the optimized real-time process parameter adjustment amount and apply it to the next dye bath to reduce dyeing unevenness. If the optimized parameters can ensure that the color difference meets the standard, they are stored in the process recipe dataset for use in future production batches, improving the stability and reusability of the process. This process not only improves dyeing consistency but also increases the accuracy of the dye recipe during the dyeing process, making the dyeing process for dark sportswear more standardized and efficient.

[0085] Example 2:

[0086] A textile company specializes in producing high-quality sportswear, especially dark sportswear. Company A hopes to reduce the color difference during the dyeing process of dark sportswear and improve the accuracy of the dye formula during the dyeing process by optimizing the dyeing process. Based on the dye formula optimization method for the dispersed dark one-bath dyeing process proposed in Example 1, Figure 3 As shown, a dye formulation optimization system using a disperse dark one-bath dyeing process includes:

[0087] refer to Figure 3 The recipe data acquisition module is used to build the process recipe data set.

[0088] Specifically, sample preparation was first performed. Disperse dyes commonly used for dyeing dark sportswear (such as disperse black, disperse blue, and disperse brown) were selected and standard dye solutions were prepared at varying concentrations to ensure comprehensive and accurate data. A UV-visible spectrophotometer was used to scan the 400-700 nm spectrum, and the absorbance A values ​​were recorded and stored in the process recipe database. Table 1 shows the spectral data for disperse black and disperse blue. Next, intelligent sensors were installed on the production line to collect real-time data such as temperature, pH, and dye concentration, which was then stored in the process recipe database. Finally, dyeing data from past production batches was recorded for use in predicting and optimizing future dyeing processes. An SQL database was used to store spectral data, process parameters, and historical dyeing data, facilitating subsequent model training and optimization calculations. This also provided accurate data support for dyeing process optimization and quality optimization.

[0089] Table 1 Example of dye spectral data

[0090]

[0091] refer to Figure 3 The ratio concentration input module is used to calculate the dye mixture spectral data using the Beer-Lambert law, input the data into a convolutional neural network (CNN) model to extract spectral feature data, and input the dye ratio concentration into a self-attention mechanism (Transformer) model to obtain concentration feature data; the spectral feature data and the concentration feature data are feature fused to calculate the mixed color feature and the formula error value; if the formula error value exceeds the target threshold, the dye ratio concentration is iteratively calculated using a genetic algorithm.

[0092] Specifically, the absorbance data of a mixture of multiple dyes is calculated based on the Beer-Lambert law to ensure the physical consistency of the input data of the subsequent deep learning model, which is expressed as:

[0093] ;

[0094] in, is the total absorbance, For dyes The concentration of For dyes The molar absorption coefficient, is the optical path length (set to 1 cm), and N is the total number of dyes in the mixture.

[0095] Figure 4This is a flow chart of feature extraction using CNN and Transformer models, which extract features of dye mixture spectral data and ratio concentration data from different perspectives. Figure 4 As shown, the structure of the CNN model is:

[0096] Input layer, used to input the collected mixed spectral data;

[0097] 1D convolution layer, used to extract local spectral features.

[0098] Pooling layer, used to normalize data and improve stability.

[0099] Fully connected layer, used to calculate the spectral feature vector.

[0100] The output layer is used to output the spectral feature vector, which represents the overall spectral performance after mixing different dyes.

[0101] Then, the Transformer structure is:

[0102] The input layer is used to input the ratio concentration vectors of various dyes (such as [1.2, 0.8, 0.5]), analyze the dye ratio concentration data, and learn the influence relationship of different dye combinations.

[0103] The position encoding layer is used to introduce sequence information to ensure that the order information of the dye ratio data is not lost.

[0104] The multi-head attention layer is used to calculate the weight relationship between different dye ratios.

[0105] A feed-forward neural network layer that learns the nonlinear effect of dye concentration on the target color value.

[0106] The output layer is used to output the concentration feature vector, which represents the contribution of different dye ratios to the final color.

[0107] Then, the spectral feature vector is fused with the concentration feature vector to calculate the mixed color feature (Lab value), which is expressed as:

[0108] ;

[0109] in, is the mixed color feature after fusion, is the spectral feature vector, is the concentration eigenvector, and are learnable weights.

[0110] According to the predicted mixed color features and the initially set target color, the recipe error value is calculated and expressed as:

[0111] ;

[0112] in, is the recipe error value, 、 and Predict the L value, a value, and b value of the mixed color feature for the model, 、 and are the L value, a value, and b value of the target color.

[0113] Among them, if If it is less than 1, it means that the human eye cannot perceive the color difference. Greater than or equal to 1 and less than or equal to 2, indicating a slight color difference, but acceptable. If the color difference is greater than 2, it indicates a significant color difference, and the dye ratio may need to be optimized. Compared with the single CNN model and the single Transformer model, the prediction accuracy is improved by 27% and 24%, respectively.

[0114] Table 2 Dye ratio concentration examples

[0115]

[0116] Table 3 Fitness value calculation results

[0117]

[0118] Furthermore, the implementation steps of the genetic algorithm include:

[0119] Step S10: randomly generating N groups of different dye ratio concentrations according to the process recipe data set;

[0120] Step S20: For the dye ratio concentration, calculate the fitness value according to the formula error value, which is expressed as:

[0121] ;

[0122] in, is the fitness value, is the formula error value;

[0123] Step S30: sorting by the fitness value, selecting the dye ratio concentration with the fitness value greater than the fitness threshold as a high-quality formula, and placing it into a mating pool, randomly selecting two high-quality formulas from the mating pool for arithmetic crossover to obtain a new high-quality formula;

[0124] Step S40: performing Gaussian variation on the new high-quality formula to obtain a new dye ratio concentration;

[0125] Step S50: Repeat steps S10 to S40 until the recipe error value is no greater than the target threshold.

[0126] In this embodiment, as shown in Table 2, the dye concentration range is set to [0.5, 1.5] g / L, and 4 groups of dye ratio concentrations are randomly generated.

[0127] Next, calculate the The results are shown in Table 3. The fitness threshold was set to 0.4 to screen for high-quality formulas. Plans 1 and 2 were then placed in a mating pool for arithmetic crossover and mutation. The resulting new formula (1.08, 0.85, 0.52) ultimately outputted the optimal dye ratio. Using a genetic algorithm to optimize dye ratios accurately predicts the optimal dye concentration without extensive experimentation, improving the success rate of color matching.

[0128] refer to Figure 3 The process parameter adjustment module is used to detect real-time process parameter data under the dye ratio concentration, divide the dye bath fuzzy set and set the dye bath fuzzy rules, and output the real-time process parameter adjustment amount in combination with the membership degree; use nonlinear regression analysis to optimize the weight of the dye bath fuzzy rules and update the real-time process parameter adjustment amount.

[0129] Furthermore, the setting process of the dye bath fuzzy rule includes:

[0130] The real-time process parameter data includes dye bath temperature, pH value and dye bath time;

[0131] Dividing the dye bath temperature, the pH value and the dye bath time into dye bath fuzzy sets, including a first temperature set, a second temperature set, a third temperature set, a first pH value set, a second pH value set, a third pH value set, a first time set, a second time set and a third time set;

[0132] Establishing a fuzzy rule base based on expert experience to describe the relationship between the real-time process parameter data and the real-time process parameter adjustment amount to obtain a dye bath fuzzy output;

[0133] The membership of the real-time process parameter data is calculated using a triangular membership function, and combined with the dye bath fuzzy output, a weighted summation method is used to obtain the real-time process parameter adjustment amount.

[0134] Specifically, as shown in Table 4, fuzzy logic is used to represent the states of variables, dividing temperature, pH, and bath dyeing time into fuzzy sets to provide more flexible control. Fuzzy rules are then set to determine adjustment strategies based on real-time process parameter data. A total of 27 rules can be set, each with different weights and membership levels. Table 5 shows three example rules.

[0135] Then, the triangular membership function is used to calculate the fuzzy membership of the current dye bath state variable, quantifying the degree of belonging of the variable in the fuzzy set. Assuming the current temperature is 124℃, the low temperature membership is 0.6 and the normal temperature membership is 0.4.

[0136] The real-time process parameter adjustment is calculated based on the membership degree to return the dye bath state to stability, which is expressed as:

[0137] ;

[0138] in, is the real-time process parameter adjustment amount, is the rule weight, is the membership degree. Assuming that the weight of rule 1 (heating) is 0.7 and the membership degree is 0.8, and the weight of rule 2 (no heating) is 0.3 and the membership degree is 0.2, then the temperature adjustment ratio is 0.62, that is, the temperature adjustment value is 0.62×(130-120)=6.2℃, and the final target temperature should be 130.2℃. Nonlinear regression analysis is then used to optimize the weights of the dye bath fuzzy rules and update the real-time process parameter adjustment amount, so that the fuzzy control can automatically adapt to changes in the production process, improving the accuracy and stability of the dyeing process. Using a fuzzy control method that does not use nonlinear regression analysis as a comparison scheme, the average color difference of the comparison scheme in the finished product color difference evaluation module is 2.3, and the maximum local color difference is 2.5, while the average color difference of the present invention is 1.7 and the maximum local color difference is 1.6, indicating that dyeing uniformity is improved and color difference is reduced.

[0139] Table 4 Fuzzy sets

[0140]

[0141] Table 5 Fuzzy rule examples

[0142]

[0143] refer to Figure 3 The finished product color difference assessment module is used to collect process image data after dyeing using an industrial camera and perform image preprocessing. The analysis results of the process image data are evaluated using a computer vision model, and the dye ratio concentration and the real-time process parameter adjustment amount are adjusted using an adaptive optimization strategy.

[0144] Furthermore, the evaluation process of the computer vision model includes:

[0145] The analysis results include process color difference, maximum local color difference, rubbing color fastness and washing color fastness;

[0146] Among them, the process color difference is the overall color deviation of the entire sportswear; the maximum local color difference is the color difference between different areas of the sportswear, which is used to judge the local color unevenness; the friction color fastness is used to judge the friction resistance of the dyeing; and the washing color fastness is used to judge the washability of the dyeing.

[0147] The process image data is converted into Lab color space, and the process color difference is calculated using the CIELab color difference formula;

[0148] Using a sliding window, the process image data is divided into fabric regions (e.g., 10×10 pixel blocks), the Lab color value of each fabric region is calculated, and the local color difference of adjacent fabric regions is calculated for the Lab color values, and the window with the largest local color difference is selected to obtain the maximum local color difference;

[0149] The color fastness to rubbing is calculated for the process image data before and after rubbing and is expressed as:

[0150] ;

[0151] in, For color fastness to rubbing, 、 and are the L value, a value and b value before friction, 、 and are the L value, a value and b value after friction.

[0152] The washing color fastness is calculated based on the process image data before and after washing, and is expressed as:

[0153] ;

[0154] in, For washing color fastness, 、 and are the L value, a value and b value before washing, 、 and are the L value, a value and b value after washing.

[0155] Among them, the preset standards for process color difference, maximum local color difference, rubbing color fastness and washing color fastness are the same. If it is greater than 2, the color difference is obvious, and it may be necessary to adjust the dye ratio concentration and the real-time process parameter adjustment amount.

[0156] Furthermore, the dye ratio concentration adjustment process includes: if the process color difference does not meet the preset standard, the dye ratio concentration is linearly adjusted using the adaptive optimization strategy to obtain the optimized dye ratio concentration, and the dye bath is re-dyeing according to the optimized dye ratio concentration; otherwise, the optimized dye ratio concentration is saved in the process formula database.

[0157] Specifically, the linear adjustment process includes:

[0158] First, set the initial values ​​of the dye ratio concentration (for example, 1.0 g / L of disperse black dye and 0.8 g / L of disperse blue dye). Then, based on the difference between the calculated process color difference and the target process color difference (for example, 2.0), calculate the optimized dye ratio concentration, expressed as:

[0159] ;

[0160] in, is the optimized dye ratio concentration, is the original dye ratio concentration, is the target process color difference, It is the actual measured process color difference.

[0161] If the optimized dye ratio concentration meets the target color difference standard, the optimized dye ratio concentration is saved in the process formula database for future reference and use.

[0162] Furthermore, the adjustment process of the real-time process parameter adjustment amount includes: if the maximum local color difference does not meet the preset standard, the real-time process parameter adjustment amount is linearly adjusted using the adaptive optimization strategy to obtain the optimized real-time process parameter adjustment amount, and the dyeing bath is re-performed according to the optimized real-time process parameter adjustment amount; otherwise, the optimized real-time process parameter adjustment amount is saved to the process formula database.

[0163] The process for adjusting the real-time process parameter adjustments is similar to that for adjusting the dye concentration. If the optimized real-time process parameter adjustments meet the target color difference standard, they are saved to the process recipe dataset for future reference and use.

[0164] In addition, if the color fastness to rubbing is low and you need to improve the color fixing effect, you can adjust the dyeing time and use the linear adjustment method to extend the dyeing time. If the color fastness to washing is low, use the linear adjustment method to increase the fixing agent concentration.

[0165] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing dye formula for dispersed dark one-bath dyeing process, characterized in that: include: Collect dye absorption spectrum data and historical dyeing data to build a process recipe data set; Based on the process recipe dataset, dye mixture spectral data is calculated using the Beer-Lambert law, and the spectral feature data is input into a convolutional neural network model to extract spectral feature data. The dye ratio concentration is input into a self-attention mechanism model to obtain concentration feature data; the spectral feature data and the concentration feature data are feature fused to calculate the mixed color feature and recipe error value; If the formula error value exceeds the target threshold, the dye ratio concentration is iteratively calculated using a genetic algorithm; detecting real-time process parameter data at the dye ratio concentration, the real-time process parameter data including dye bath temperature, pH value, and dye bath time; dividing the dye bath temperature, the pH value, and the dye bath time into dye bath fuzzy sets respectively; Divide the dye bath fuzzy set and set the dye bath fuzzy rules to output the real-time process parameter adjustment value in combination with the membership degree; Optimizing the weights of the dye bath fuzzy rules using nonlinear regression analysis and updating the real-time process parameter adjustments; collecting process image data after dyeing, and evaluating analysis results of the process image data using a computer vision model; And an adaptive optimization strategy is used to adjust the dye ratio concentration and the real-time process parameter adjustment amount.

2. The method for optimizing the dye formula of a disperse dark one-bath dyeing process according to claim 1, wherein: The implementation steps of the genetic algorithm include: Step S10: randomly generating N groups of different dye ratio concentrations according to the process recipe data set; Step S20: For the dye ratio concentration, calculate the fitness value according to the formula error value, which is expressed as: in, is the fitness value, is the formula error value; Step S30: selecting the dye ratio concentration with the fitness value greater than the fitness threshold as a high-quality formula and placing it in a mating pool, randomly selecting two high-quality formulas from the mating pool for arithmetic crossover to obtain a new high-quality formula; Step S40: performing Gaussian variation on the new high-quality formula to obtain a new dye ratio concentration; Step S50: Repeat steps S10 to S40 until the recipe error value is no greater than the target threshold.

3. The method for optimizing the dye formula of a disperse dark one-bath dyeing process according to claim 1, wherein: The evaluation process of the computer vision model includes: The analysis results include process color difference, maximum local color difference, rubbing color fastness and washing color fastness; The process image data is converted into Lab color space, and the process color difference is calculated using the CIELab color difference formula; Using a sliding window to divide the process image data into fabric areas, calculating the Lab color value of each fabric area, and calculating the local color difference of adjacent fabric areas for the Lab color values ​​to obtain the maximum local color difference; Calculating the rubbing color fastness based on the process image data before and after rubbing; The washing color fastness is calculated based on the process image data before and after washing.

4. The method for optimizing the dye formula of a disperse dark one-bath dyeing process according to claim 1, wherein: The dye ratio concentration adjustment process includes: if the process color difference does not meet the preset standard, the dye ratio concentration is linearly adjusted using the adaptive optimization strategy to obtain the optimized dye ratio concentration, and the dye bath is re-dyeing according to the optimized dye ratio concentration; otherwise, the optimized dye ratio concentration is saved in the process formula database.

5. The method for optimizing the dye formula of a disperse dark color one-bath dyeing process according to claim 1, wherein: The adjustment process of the real-time process parameter adjustment amount includes: if the maximum local color difference does not meet the preset standard, the real-time process parameter adjustment amount is linearly adjusted using the adaptive optimization strategy to obtain the optimized real-time process parameter adjustment amount, and the dye bath is re-performed according to the optimized real-time process parameter adjustment amount; otherwise, the optimized real-time process parameter adjustment amount is saved in the process formula database.

6. A dye formula optimization system for disperse dark one-bath dyeing process, characterized in that: include: Recipe data acquisition module, used to build process recipe data set; A ratio concentration input module is used to calculate dye mixture spectral data using the Beer-Lambert law, input the data into a convolutional neural network model to extract spectral feature data, and input the dye ratio concentration into a self-attention mechanism model to obtain concentration feature data; the spectral feature data and the concentration feature data are subjected to feature fusion to calculate the mixed color feature and formula error value; If the formula error value exceeds the target threshold, the dye ratio concentration is iteratively calculated using a genetic algorithm; A process parameter adjustment module is used to detect real-time process parameter data under the dye ratio concentration, divide the dye bath fuzzy set and set the dye bath fuzzy rules, and output the real-time process parameter adjustment amount in combination with the membership degree; Optimizing the weights of the dye bath fuzzy rules using nonlinear regression analysis and updating the real-time process parameter adjustments; The finished product color difference evaluation module is used to collect process image data after dyeing, use a computer vision model to evaluate the analysis results of the process image data, and use an adaptive optimization strategy to adjust the dye ratio concentration and the real-time process parameter adjustment amount.

Images