Fiber dyeing formula forecasting method based on BP neural network

Abstract

The invention discloses a BP neural network-based fiber dyeing formula forecasting method. The method comprises the following steps: preparing a sample; selecting yellow, red and blue dyes for dyeing, and dyeing the wool fibers with different concentration gradients for each dye to obtain K / S values corresponding to different concentration gradients of various single-color dyes; testing the sample by adopting a spectrophotometer, and recording a reflectivity value; establishing a monochromatic basic database based on a K / S value formula; generating a plurality of groups of training data based on the relationship between different dye liquor concentrations of the single dye liquor and the K / S value; after the model training is completed, performing dyeing formula prediction on the sample; the dyeing formula prediction precision of the pre-training model based on the actual dyeing sample optimization is improved to a certain extent, and the factory initial proofing color difference acceptance range can be met. And optimizing the pre-training model in combination with real proofing data. And the accuracy of formula prediction is high.

Description

Technical Field

[0001] The invention relates to a fiber dyeing formula prediction method based on BP neural network, belonging to the technical field of fiber dyeing formula prediction. Background Art

[0002] Color is an important part of color science. With the increasing development of color science, people's requirements for color are also increasing. Especially in the textile field, color is an extremely important factor in textile quality control. Colored yarn is a yarn made by first dyeing the fiber and then spinning fibers of different colors in a certain proportion, which has a unique color mixing effect. In this process, both fiber dyeing and colored yarn spinning processes have high requirements for color. For colored yarn companies, fiber dyeing and colored yarn color design were previously operated based on experience, with long proofing time, difficult cost control, low color matching accuracy, and poor reproducibility, which hindered the development of the colored yarn industry. Therefore, companies urgently need a precise color matching technology to reduce the number of proofing times, save labor costs and raw materials, shorten the production cycle, improve corporate production efficiency, and maximize benefits.

[0003] Computer color matching technology uses color measuring instruments to obtain the spectral reflectance data of the target color sample and inputs it into the computer. Then the spectrum basic database and related color matching software in the computer are used for calculation and correction, and finally a formula that meets the requirements of the target color sample is obtained. At present, factories generally use manual experience proofing, which requires 3-5 or even more times of dyeing proofing. The entire sample cycle is very long, which affects the delivery time of sample processing and product development for rapid market changes.

[0004] Accurate prediction of dye formula has always been a difficult problem for enterprises. At present, computer color matching technology is based on the Kubelka-Munk theory (KM), which calculates the color-related quantities measured on the single-color base sample to achieve the purpose of color matching. However, the dyeing process involves the influence of many factors, and the KM theory is difficult to find a suitable formula to obtain the target color, resulting in a large waste of manpower and cost. Summary of the invention

[0005] The purpose of the present invention is to provide a fiber dyeing formula prediction method based on BP neural network, which uses the BP neural network model to train theoretical data to establish a pre-training model, and optimizes the pre-training model through real proofing data to achieve the effect of accurately predicting the dye formula.

[0006] To solve the above technical problems, the purpose of the present invention is achieved as follows:

[0007] The present invention relates to a fiber dyeing formula prediction method based on BP neural network, comprising the following steps:

[0008] S1. Preparation of samples: According to the following process, wool dyeing → opening → carding → drawing → roving → spun yarn → weaving, wool knitted fabrics are made as samples;

[0009] Select yellow, red and blue dyes for dyeing, and dye wool fibers with different concentration gradients of each dye to obtain K / S values ​​corresponding to different concentration gradients of various single-color dyes; the amount of the single-color dye is 0%-2.5% owf;

[0010] S2. Sample color measurement: Use Datacolor 600 spectrophotometer to test the samples prepared in the previous step, and record the reflectance value of each sample in the visible spectrum of 400-700nm;

[0011] S3. Establishment of a monochrome basic database: Based on the K / S value formula, calculate the K / S values ​​corresponding to different concentration gradients of a single dye solution at a wavelength of 400nm-700nm to establish a monochrome basic database;

[0012] S4. Construction of pre-training model: Based on the relationship between different dye concentrations and K / S values ​​of a single dye solution, a dye concentration is randomly selected from each color of dye solution for mixing, and its corresponding K / S value, reflectivity, and L*, a*, and b* values ​​are obtained to generate several sets of training data, wherein each set of data contains the concentrations and L*, a*, and b* values ​​of three colors of dye solution; the BP neural network is trained using the constructed training data to construct a pre-training model; a mapping relationship between CIE L*, a*, and b* values ​​and dye concentration is established to achieve the effect of dyeing formula prediction, wherein the input layer is the L*, a*, and b* values, and the output layer is the three dye concentrations;

[0013] The BP neural network is designed according to the training data type as follows:

[0014] (1) Data processing: Use data processing software to normalize the training data so that the overall distribution of the training data is between 0 and 1;

[0015] (2) Dataset division: The training dataset is randomly divided into three subsets: training set, test set, and validation set, with the proportions of 70%, 15%, and 15% respectively.

[0016] (3) Number of neuron nodes: The BP neural network structure includes input layer, hidden layer and output layer. The number of nodes in the input layer and output layer is consistent with the data. According to actual needs, the L*, a* and b* color values ​​are used as the input layer, and the number of nodes is 3; the output layer is the dye concentration formula, including yellow, red and blue dyes, and the number of nodes is also 3;

[0017] The number of hidden layer nodes is usually determined roughly by empirical formula (1);

[0018]

[0019] Where: m is the number of input layer nodes; n is the number of output layer nodes; S is the number of hidden layer nodes; k is an integer between 1 and 10;

[0020] (4) Other parameters of BP neural network: the transfer function from input layer to hidden layer is tansig function, the transfer function from hidden layer to output layer is purelin linear function, the model training function is traingdm function, and the gradient descent training function is used to accelerate the convergence of the learning process and avoid falling into the local minimum to a certain extent;

[0021] S5. Dyeing formula prediction: After the model training is completed, the dyeing formula of the sample is predicted;

[0022] S6. Optimization of pre-trained models: Optimize the pre-trained models based on real proofing data.

[0023] On the basis of the above scheme and as a preferred scheme of the above scheme: the dosage of the monochromatic dye is 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.2, 1.4, 1.6, 1.8, 2.0, and 2.5% owf respectively.

[0024] On the basis of the above scheme and as a preferred scheme of the above scheme: in the sample preparation, the dyeing of wool as follows:

[0025] Use Datacolor automatic drip machine to drip dye solution into the dye cup, add 6g of washed 80S Australian wool, the bath ratio is 1:20;

[0026] For low concentration dye solution (<2% owf), add appropriate amount of glacial acetic acid to adjust the pH value of the dye bath (5.5-6.5), increase the temperature from room temperature to 100°C at a heating rate of 2°C / min, and maintain this temperature for 40min;

[0027] For high concentration dye solution (>2% owf), add appropriate amount of formic acid to adjust the pH value of the dye bath (4.5-5.5), increase the temperature from room temperature to 100°C at a heating rate of 2°C / min, and maintain this temperature for 60min;

[0028] The dyed wool fibers are then washed and dried.

[0029] On the basis of the above scheme and as the preferred scheme of the above scheme: in the sample color measurement, Datacolor600 spectrophotometer is used for testing; after the instrument is preheated and calibrated, the sample color is measured after setting the measurement aperture to 20mm, D65 standard light source, specular gloss inclusion, 100% UV, 10° observation angle and other parameters, each sample is measured 4 times at different positions, the average value is taken, and the reflectance value of each sample in the visible spectrum of 400-700nm is recorded, and the band interval is set to 10nm; during the color measurement process, ensure that the color difference obtained by repeated measurements is less than 0.1.

[0030] On the basis of the above scheme and as a preferred scheme of the above scheme: in the establishment of the monochrome basic database, the cubic spline fitting interpolation method is used to perform curve fitting on the dye concentration and the K / S value.

[0031] On the basis of the above scheme and as a preferred scheme of the above scheme: the number of implicit nodes is 9.

[0032] On the basis of the above scheme and as a preferred scheme of the above scheme: in dividing the data set, the three subsets of training set, test set and validation set are divided according to the ratio of 70%, 15% and 15% respectively.

[0033] The beneficial effects of the present invention are as follows: the fiber dyeing formula prediction method based on BP neural network involved in the present invention has a certain improvement in the prediction accuracy of the dyeing formula based on the pre-trained model optimized based on the actual dyeing samples. The color difference of 17% of the prediction results of the optimized model is less than 1, and the average color difference is 1.679, which can meet the factory's initial proofing color difference acceptance range. The pre-trained model is optimized in combination with real proofing data. The accuracy of formula prediction is high. The value of data-driven color matching models in practical applications is verified, providing a new and efficient color matching technology route for the textile industry. DETAILED DESCRIPTION

[0034] The present invention is further described below in conjunction with specific embodiments.

[0035] The present invention relates to a fiber dyeing formula prediction method based on BP neural network, comprising the following steps: S1 sample preparation, S2 sample color measurement, S3 monochrome basic database establishment, S4 pre-training model construction, S5 dyeing formula prediction and S6 pre-training model optimization.

[0036] The experimental materials used in the present invention are shown in Table 1:

[0037] Table 1 Experimental materials

[0038]

[0039]

[0040] The experimental instruments used are shown in Table 2:

[0041] Table 2 Experimental instruments

[0042] Instrument Name model Manufacturer Automatic dripping machine Datacolor AutoLABTM Datacolor Corporation Infrared dyeing machine KM1-HT-12 Suzhou Qile Electronic Technology Co., Ltd. Carding Machine B305 Qingdao Jingjia Textile Machinery Manufacturing Co., Ltd. Roving frame FB441 Qingdao Jingjia Textile Machinery Manufacturing Co., Ltd. Spinning Machine FB583A Qingdao Jingjia Textile Machinery Manufacturing Co., Ltd. Hand Crank Knitting Machine MR-1 Jiangsu Changshu Huashida Machinery Manufacturing Color measurement and matching instrument Datacolor600 Datacolor Corporation

[0043] Establishing a monochrome basic database is the basis of computer color matching, and the accuracy of basic color samples has a direct impact on the accuracy of formula prediction. Each dye is dyed with different concentration gradients to dye wool fibers, and the K / S values ​​corresponding to the different concentration gradients of each monochrome dye are obtained. Usually, 8-10 extreme differences in the dyeing concentration gradient level can meet the requirements, in order to provide sufficient dye concentration and corresponding K / S value data. According to the dye liquor concentration range in the actual production of the enterprise, it is divided into 17 concentration gradients in the present invention, which are 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.2, 1.4, 1.6, 1.8, 2.0, 2.5% owf. And establish blank wool fiber data (0% owf), that is, prepare blank wool fiber dyed fabrics under the same dyeing conditions without adding dye.

[0044] In the sample preparation process, wool fabrics are made as samples according to the following steps: wool dyeing → opening → carding → drawing → roving → spun yarn → weaving.

[0045] The dyeing process of wool fiber is as follows: use Datacolor automatic drip machine to drip dye solution into the dye cup, put in 6g 80S Australian wool, the bath ratio is 1:20. For low concentration dye solution (<2% owf), add appropriate amount of glacial acetic acid to adjust the pH value of the dye bath (5.5-6.5), heat up from room temperature to 100℃ at a heating rate of 2℃ / min, and maintain the temperature for 40min; for high concentration dye solution (>2% owf), add appropriate amount of formic acid to adjust the pH value of the dye bath (4.5-5.5), heat up from room temperature to 100℃ at a heating rate of 2℃ / min, and maintain the temperature for 60min. After that, the dyed wool fiber is processed by washing, drying and other steps.

[0046] The specific operation process of sample preparation is as follows:

[0047] Loosening: Place the fiber ball on the needle plate, hold the upper needle plate and the lower needle plate with the needle tips staggered, move back to back, and simply loosen the fiber ball.

[0048] Combing: Feed the mixed fibers into the needle combing machine and comb them 5 to 7 times to make the fibers more evenly combed. After each combing, the wool fiber web is manually divided into fixed lengths and fed in to control the thickness of the combed wool fiber web.

[0049] Drawing: The carded wool fiber web is fed into the drawing frame and made into wool tops by a one-pass drafting and two-pass drawing process.

[0050] Roving: The carded and drawn wool tops are fed into the roving frame and made into roving yarn through drawing and twisting.

[0051] Spinning: The thick yarn is fed into the spinning frame for further drafting and twisting to produce yarn with smaller diameter, higher uniformity and strength.

[0052] Woven fabric: The spun yarn is woven through a 24-needle hand-cranked flat knitting machine, and the structure is a weft plain knitted fabric.

[0053] In the sample color measurement process, Datacolor600 spectrophotometer (Datacolor, USA) was used for testing. After the instrument was preheated and calibrated, the sample color was measured after setting the measurement aperture to 20mm, D65 standard light source, specular gloss inclusion, 100% UV, 10° observation angle and other parameters. Each sample was measured 4 times at different positions, and the average value was taken. The reflectance value of each sample in the visible spectrum of 400-700nm was recorded, and the band interval was set to 10nm. During the color measurement process, ensure that the color difference obtained by repeated measurements is less than 0.1.

[0054] K / S value is an index for evaluating fabric coloring effect and dye covering power. When the dye concentration is low, the K / S value is approximately linear with the dye concentration, but when the dye concentration reaches a certain level, the K / S value is curvilinear with the dye concentration. In the past, when scholars studied the relationship between the K / S value and the unit concentration of the dye solution, they mainly used linear fitting to explore the relationship between the two, which was difficult to well show the relationship between the dye concentration and the K / S value at high concentrations. In order to better study the relationship between the K / S value and the unit concentration of the dye solution, based on the K / S value formula, the K / S value corresponding to different concentration gradients of a single dye solution at a wavelength of 400nm-700nm (interval 10nm) was calculated by spectral reflectance, and the relationship between the concentration and the K / S value was fitted by the cubic spline interpolation method to obtain the K / S value corresponding to the single dye solution concentration gradient at 400nm-700nm.

[0055] Cubic spline interpolation involves constructing a special piecewise polynomial, which is usually used for data interpolation and smoothing. Given a set of data points {(x0,y0),(x1,y1),…(x n ,y n )} The cubic spline function is a series of cubic polynomials S i (x) where each S i (x) is defined at adjacent data points (x i ,x i+1 ). The function satisfies the following conditions:

[0056] (1) Segment definition

[0057] In each interval [x i ,x i+1 ], the function is represented by the cubic polynomial (2):

[0058] S i (x) = a i +b i (xx i )+c i (xx i ) 2 +d i (xx i ) 3 (2)

[0059] where i = 0, 1, ..., n-1, and a i , b i , c i , d i are the polynomial coefficients.

[0060] (2) Continuity

[0061] The spline function is continuous at all interval points, that is, for all i: S i (x i )=y i and S i-1 (x i )=S i (x i ). And the first and second order derivatives of the spline function at all internal nodes are also continuous.

[0062] Based on the xalglib library in the Python platform, cubic spline fitting was performed on the K / S values ​​corresponding to different concentrations of three dyes, lanner yellow 4G, lanner red 6G and une blue 3GS, at a wavelength of 400-700nm (interval of 10nm), where each dye included 17 fitting curves corresponding to concentrations, and each curve contained 13 continuous cubic functions. Based on the curve fitted by cubic spline interpolation, the K / S values ​​corresponding to the concentrations of each dye at different wavelengths in the concentration range of 0-2.5 with a step size of 0.001 were generated.

[0063] Based on the K / S values ​​corresponding to the wavelength of 400-700nm of the three dyes at different concentrations and the K / S values ​​corresponding to the wavelength of 400-700nm of the Australian wool substrate used, it is assumed that the dyeing effect of the three dye solutions is the sum of the K / S values. The three dye solutions are randomly selected and mixed to obtain their corresponding K / S values, reflectance, and L*, a*, and b* values, and 100 sets of data are generated, each of which contains the concentration of the three primary colors of dye solutions and the corresponding L*, a*, and b* values. Dyeing samples were made for the 100 sets of dye solution concentration combinations. After the proofing was completed, the color was measured by the Datacolor600 spectrophotometer and the color difference was calculated with the color value obtained based on the cubic spline fitting to verify the accuracy of the relationship between the dye solution concentration and the K / S value established by the cubic spline fitting interpolation method. The color difference calculation results are shown in Table 3.

[0064] Table 3100 color difference values ​​of samples

[0065]

[0066]

[0067] It can be seen from Table 3 that the color values ​​calculated based on the cubic spline fitting interpolation method are relatively close to the color values ​​of the real dyed samples, with an average color difference of 1.032, of which the number of samples with a color difference less than 0.5 accounts for 17% of the total, the number of samples with a color difference less than 1 accounts for 55% of the total, the number of samples with a color difference less than 2 accounts for 96% of the total, and the number of samples with a color difference greater than 2 accounts for 4% of the total. It is proved that the relationship between the dye concentration and the K / S value established by the cubic spline fitting interpolation method has a high accuracy, and the next step of research can be carried out based on this method.

[0068] Based on the relationship between dye concentration and K / S value established by cubic spline fitting interpolation method, three dye concentrations are randomly selected and mixed to obtain their corresponding K / S value, reflectivity, and L*, a*, b* values, generating 20,000 sets of data, each of which contains the concentration of three primary colors and L*, a*, b* values. The BP neural network is trained using 20,000 sets of theoretical data to build a pre-training model, and the mapping relationship between CIE L*a*b* values ​​and dye concentration is established to achieve the effect of dyeing formula prediction, where the input layer is L*, a*, b* values, and the output layer is three dye concentrations.

[0069] BP (Back Propagation) neural network is a neural network trained according to the error back propagation algorithm. It has strong nonlinear mapping ability. By learning a large amount of data, it automatically learns the nonlinear mapping rules between input data and output data, and continuously adjusts its weights and deviations through the back propagation algorithm to minimize the prediction error. In the present invention, the BP neural network is designed as follows according to the training data type:

[0070] (1) Data preprocessing

[0071] Before training the network, the data must be preprocessed first. Reasonable data preprocessing can improve the accuracy and generalization ability of the model. This paper uses the mapminmax function in Matlab to normalize the data so that the overall distribution of the data is between 0 and 1.

[0072] (2) Dividing the Dataset

[0073] The data set is usually divided into three subsets: training set, test set, and validation set, with a ratio of 70%, 15%, and 15%, respectively.

[0074] (3) Number of neuron nodes

[0075] This paper designs a three-layer BP neural network structure, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer and the output layer is consistent with the data. According to actual needs, the L*, a*, and b* color values ​​are used as the input layer, and the number of nodes is 3; the output layer is the dye concentration formula, including three dyes: yellow, red, and blue, and the number of nodes is also 3. The number of hidden layer nodes determines the ability of the BP neural network to fit complex relationships. In theory, the more hidden layer nodes there are, the more nonlinear features can be learned. However, too many hidden layer nodes will lead to overfitting of the network. The appropriate number of hidden layer nodes can balance the model complexity and generalization ability. At present, the number of hidden layer nodes is usually determined by the empirical formula (1).

[0076]

[0077] Where: m is the number of input layer nodes; n is the number of output layer nodes; S is the number of hidden layer nodes; k is an integer between 1 and 10. The number of hidden layer nodes calculated by formula (1) ranges from 4 to 12. Simulation experiments are carried out one by one to observe the corresponding correlation coefficient R. The result shows that when the number of hidden layer nodes is 9, the R value is closest to 1, and the fitting degree is the highest at this time. Therefore, the number of hidden layer nodes is 9.

[0078] (4) Other parameters of BP neural network

[0079] The transfer function from the input layer to the hidden layer is the tansig function, which can perform nonlinear mapping on the input data and enhance the network's ability to fit nonlinear patterns. The transfer function from the hidden layer to the output layer is the purelin linear function. The activation of the network output layer is directly mapped to the output value, which directly adjusts and optimizes the prediction error and improves the accuracy of the prediction. The model training function is the traingdm function, which performs gradient descent training to accelerate the convergence of the learning process and can avoid falling into the local minimum to a certain extent. The number of iterations is 1000, the target error value is 10-6, and the learning efficiency is 0.01.

[0080] In the dyeing formula prediction, the measured data (L*a*b* value and the concentration of yellow, red and blue dye solutions) are input into the BP neural network model for learning and training, of which 70% of the data is used for training the network, 15% of the data is used for network verification, and 15% of the data is used for testing. After the model training is completed, the dyeing formula prediction is carried out on 100 samples, and the predicted formula, the true formula, and the absolute error and relative error between them are shown in Table 4.

[0081] Table 4 Comparison of recipe prediction based on pre-trained model and actual recipe

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] It can be seen from Table 4 that most of the absolute error values ​​are within the percentile, and a few appear in the deciles; while the relative error appears in some data with large errors, and the overall fluctuation is large, indicating that there are outliers in the theoretical data, which affect the training of the network; in terms of the average absolute error, the errors between the predicted results of the yellow, red and blue dye formulas and the actual formulas are 0.073, 0.052 and 0.081 respectively, and there are still certain errors.

[0088] The pre-trained model was analyzed and verified. The difference in the formula can only be seen from the comparison between the predicted formula and the actual formula. The analysis using the color difference value has a more intuitive effect. Therefore, 100 model-predicted formulas were actually dyed and proofed. After dyeing, the wool was prepared through the steps of opening → combing → drawing → roving → spun yarn → woven pieces to measure the color. The color difference between the wool and the original sample was calculated to determine the accuracy of the predicted formula and verify the accuracy of the model. The color difference calculation results are shown in Table 5.

[0089] Table 5 Pre-trained model predicts color difference values

[0090]

[0091]

[0092] As shown in Table 5, the color difference of all samples is calculated by using the formula given by the pre-trained model for dyeing proofing. The average color difference is 2.48, 7% of the formulas have a color difference less than 1, 37% of the data are less than 2, and 68% of the data are less than 3. This shows that the formula given by the pre-trained model has a large deviation from the actual color of the sample, making it difficult to accurately predict the dyeing formula and cannot be used in production practice. This is mainly because the model is trained based on theoretical data, and some data may have a certain deviation from the real data, affecting the prediction accuracy of the model.

[0093] The pre-trained model was optimized using actual dyeing sample data. The pre-trained model was optimized with real proofing data in order to improve the model prediction accuracy. Therefore, 400 groups of yellow, red and blue dye solutions with different concentrations were selected for combined dyeing, and the actual dyeing sample data were used to optimize the pre-trained model to make the dye color matching model more realistic and accurate. The L*a*b* values ​​and dye solution concentration data sets of some actual samples are shown in Table 6.

[0094] Table 3 Actual sample L*a*b* values ​​and dye concentration data set

[0095]

[0096]

[0097] Based on the Matlab (2022b) platform, the pre-trained model is directly loaded, keeping the original structure and initial weights of the model unchanged, and then retrained based on 400 sets of real data. The main code is shown in Table 7.

[0098] Table 7 Pre-training model optimization code

[0099]

[0100]

[0101] After optimizing the pre-trained model with real dyeing sample data, in order to compare the prediction effect with the pre-trained model under the same conditions, the same 100 standard samples were selected for dyeing formula prediction, and compared with the actual dyeing formula for analysis. The data of the absolute error and relative error between the real formula, the predicted formula, and the predicted formula and the real formula are shown in Table 8.

[0102]

[0103]

[0104]

[0105]

[0106] It can be seen from Table 8 that the formula prediction results with the addition of real dyeing data are closer to the original formula. The average absolute error of the dye formula for yellow is 0.037, for red it is 0.031, and for blue it is 0.044. Compared with the pre-trained model, the formula prediction accuracy has been improved to a certain extent.

[0107] In order to further prove the accuracy of the model in predicting dyeing formula, the optimized pre-trained model output formula is actually dyed and proofed, and the color difference between it and the original sample is calculated to determine whether the prediction is accurate or not, and verify the accuracy of the model. The color difference calculation results are shown in Table 9.

[0108] Table 2 Real data optimization pre-training model predicts color difference value

[0109]

[0110]

[0111] Table 9 shows that the pre-trained model trained with real data has improved in formula prediction. 17% of the samples have a color difference less than 1, 72% of the samples have a color difference less than 2, the average color difference is 1.679, the maximum color difference is 3.665, and the minimum color difference is 0.222. Overall, the pre-trained model optimized with real data can achieve good prediction accuracy.

[0112] In order to deeply analyze and evaluate the accuracy and practicality of the dyeing color matching model in formula prediction, it is compared with the Datacolor Match (Datacolor, USA) dyeing color matching software, and the formulas given by the two color matching models under the same standard sample are observed, and dyeing proofing is performed, and the color difference value is calculated for analysis. The basic dyeing sample data produced in Section 3.1.3 is entered into the Datacolor Match color matching software as the basic database for dyeing color matching, ensuring that the basic databases used by the two color matching software are the same, and the accuracy of the two color matching models is verified under the same conditions.

[0113] 40 colors were selected from samples sent by factory customers as standard samples. The color was measured by Datacolor 600 spectrophotometer and the color matching software Datacolor Match was used to match the color and give the formula. The L*a*b* values ​​of the 40 standard samples were input into the optimized pre-trained model for formula prediction. The formula was given according to the two color matching models for dyeing proofing. The L*a*b* values ​​of the standard samples and the formula given by the two color matching models are shown in Table 10.

[0114] Table 10 Two color matching model prediction formula table

[0115]

[0116]

[0117]

[0118] According to the formula given by Datacolor Match and the formula given by the optimized pre-trained model, dyeing proofing is carried out, and the color value is measured to calculate the color difference. The distribution of the color difference values ​​between the sample and the standard sample is shown in Table 11.

[0119] Table 11 Color difference value distribution

[0120]

[0121] As shown in Table 11, among the 40 samples, the average color difference of the formula proofing predicted by the optimized pre-trained model is 1.536, the color difference value distribution is within 1 for 27.5%, and the color difference value distribution is within 2 for 77.5%. The average color difference of the formula proofing predicted by DatacolorMatch is 1.624, the color difference value distribution is within 1 for 20%, and the color difference value distribution is within 2 for 72.5%. Overall, the optimized pre-trained model is more accurate than Datacolor Match in giving the formula, with a 27.5% probability of color difference value less than 1, which can meet the initial proofing needs, and a 77.5% probability of color difference value within 2. The proofing master can make adjustments on this basis to shorten the number of proofing times.

[0122] The prediction of fiber dyeing formula was studied. Three basic dyeing samples with different concentration gradients of three dye solutions, namely, Lanasu Yellow 4G, Lanasu Red 6G and Unasu Blue 3GS, were prepared. The relationship between dye concentration and K / S value was determined by cubic spline fitting interpolation method. According to the relationship between dye concentration and K / S value, 20,000 sets of theoretical data were generated. Each set of data contained dye formula and corresponding L*a*b* value. Based on BP neural network, 20,000 sets of theoretical data were trained to establish a pre-training model to verify the accuracy of its prediction formula. 400 sets of real dyeing sample data were used to optimize the pre-training model to improve the accuracy of the model prediction formula. The results show that the dyeing formula prediction accuracy of the pre-training model optimized based on actual dyeing samples has been improved to a certain extent. After optimization, 17% of the prediction results of the model have a color difference less than 1, and the average color difference is 1.679, which can meet the color difference acceptance range of the factory's initial proofing.

[0123] Through actual dyeing comparison tests, the application effects of two different color matching models in textile dyeing formula prediction were explored. 40 color standard samples were randomly selected from factory samples, and the corresponding dye formulas were predicted using Datacolor Match color matching software and the optimized pre-trained model. Dyeing proofing and color difference evaluation were performed under the same experimental conditions to accurately measure and compare the prediction accuracy and practical value of the two models. The experimental results show that compared with Datacolor Match, the optimized pre-trained model shows higher accuracy in formula prediction. The average color difference value of its proofing is smaller than 1.536, and the proportion of color difference values ​​distributed within 1 is more, which is 27.5%, and the accuracy of formula prediction is higher. The value of data-driven color matching models in practical applications has been verified, providing a new and efficient color matching technology route for the textile industry. In the future, by continuously learning more dyeing data, it is expected that the performance of this data-driven model can be further improved, thus bringing new solutions to the color design and quality control of textiles.

[0124] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A fiber dyeing formula prediction method based on BP neural network, characterized in that: The steps include: S1. Preparation of samples: According to the following process, wool dyeing → opening → carding → drawing → roving → spun yarn → weaving, wool knitted fabrics are made as samples; Select yellow, red and blue dyes for dyeing, and dye wool fibers with different concentration gradients of each dye to obtain K / S values ​​corresponding to different concentration gradients of various single-color dyes; the amount of the single-color dye is 0%-2.5% owf; S2. Sample color measurement: Use Datacolor 600 spectrophotometer to test the samples prepared in the previous step, and record the reflectance value of each sample in the visible spectrum of 400-700nm; S3. Establishment of a monochrome basic database: Based on the K / S value formula, calculate the K / S values ​​corresponding to different concentration gradients of a single dye solution at a wavelength of 400nm-700nm, and establish a monochrome basic database; S4. Construction of pre-training model: Based on the relationship between different dye concentrations and K / S values ​​of a single dye solution, a dye concentration is randomly selected from each color of dye solution for mixing, and its corresponding K / S value, reflectivity, and L*, a*, and b* values ​​are obtained to generate several sets of training data, wherein each set of data contains the concentrations and L*, a*, and b* values ​​of three colors of dye solution; the BP neural network is trained using the constructed training data to construct a pre-training model; a mapping relationship between CIE L*, a*, and b* values ​​and dye concentration is established to achieve the effect of dyeing formula prediction, wherein the input layer is the L*, a*, and b* values, and the output layer is the three dye concentrations; The BP neural network is designed according to the training data type as follows: (1) Data processing: Use data processing software to normalize the training data so that the overall distribution of the training data is between 0 and 1; (2) Dataset division: The training dataset is randomly divided into three subsets: training set, test set, and validation set, with the proportions of 70%, 15%, and 15% respectively. (3) Number of neuron nodes: The BP neural network structure includes input layer, hidden layer and output layer. The number of nodes in the input layer and output layer is consistent with the training data. According to actual needs, the L*, a* and b* color values ​​are used as the input layer, and the number of nodes is 3; the output layer is the dye concentration formula, including yellow, red and blue dyes, and the number of nodes is also 3; The number of hidden layer nodes is usually determined roughly by empirical formula (1); Where: m is the number of input layer nodes; n is the number of output layer nodes; S is the number of hidden layer nodes; k is an integer between 1 and 10; (4) Other parameters of BP neural network: the transfer function from input layer to hidden layer is tansig function, the transfer function from hidden layer to output layer is purelin linear function, the model training function is traingdm function, the gradient descent training function is used, and the acceleration Convergence during learning, and being able to avoid falling into local minima to a certain extent; S5. Dyeing formula prediction: After the model training is completed, the dyeing formula of the sample is predicted; S6. Optimization of pre-trained models: Optimize the pre-trained models based on real proofing data.

2. A fiber dyeing formula prediction method based on BP neural network according to claim 1, characterized in that: The amounts of the monochromatic dyes are 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.2, 1.4, 1.6, 1.8, 2.0, and 2.5% owf, respectively.

3. The fiber dyeing formula prediction method based on BP neural network according to claim 1 is characterized in that: In the sample preparation, the wool was dyed as follows: Use Datacolor automatic drip machine to drip dye solution into the dye cup, add 6g of washed 80S Australian wool, the bath ratio is 1:20; For low concentration dye solution (<2% owf), add appropriate amount of glacial acetic acid to adjust the pH value of the dye bath (5.5-6.5), increase the temperature from room temperature to 100°C at a heating rate of 2°C / min, and maintain this temperature for 40min; For high concentration dye solution (>2% owf), add appropriate amount of formic acid to adjust the pH value of the dye bath (4.5-5.5), increase the temperature from room temperature to 100°C at a heating rate of 2°C / min, and maintain this temperature for 60min; The dyed wool fibers are then washed and dried.

4. The fiber dyeing formula prediction method based on BP neural network according to claim 1 is characterized in that: In the sample color measurement, Datacolor 600 spectrophotometer was used for testing; after the instrument was preheated and calibrated, the measurement aperture was set to 20mm, D65 standard light source, specular gloss included, 100% UV, 10° observation angle and other parameters to measure the sample color. Each sample was measured 4 times at different positions, and the average value was taken. The reflectance value of each sample in the visible spectrum of 400-700nm was recorded, and the band interval was set to 10nm; during the color measurement process, it was ensured that the color difference obtained by repeated measurements was less than 0.

1.

5. The fiber dyeing formula prediction method based on BP neural network according to claim 1 is characterized in that: In the establishment of the single-color basic database, the cubic spline fitting interpolation method is used to fit the curve between the dye concentration and the K / S value.

6. The fiber dyeing formula prediction method based on BP neural network according to claim 1 is characterized in that: The number of hidden nodes is 9.

7. The fiber dyeing formula prediction method based on BP neural network according to claim 1 is characterized in that: In the partitioned data set, the three subsets of training set, test set, and validation set are divided according to the ratio of 70%, 15%, and 15% respectively.

Images